SlideShare uma empresa Scribd logo
Joint Multisided Exposure Fairness
for Search and Recommendation
Bhaskar Mitra
Microsoft Research, Canada
bmitra@microsoft.com
Pre-print: https://arxiv.org/pdf/2205.00048.pdf
(Paper accepted @ SIGIR’22)
Joint work with Haolun Wu, Chen Ma,
Fernando Diaz, and Xue Liu
Digital information
access and exposure
Traditional IR is concerned with ranking
of items according to relevance
These information access systems
deployed at web-scale mediate what
information gets exposure
The exposure-framing of IR raises several
fairness concerns, new opportunities for
ranking optimization, and can be
relevant to other FATE considerations
(e.g., privacy and transparency)
Sweeney. Discrimination in online ad delivery. Commun. ACM. (2013)
Crawford. The Trouble with Bias. NeurIPS. (2017)
Singh and Joachims. Fairness of Exposure in Rankings. In KDD, ACM. (2018)
Harms of disparate exposure
Several past studies have pointed out representational
and allocative harms from disparate exposure
Concerns of fairness in the context of IR/ML systems are
inherently interdisciplinary and sociotechnical; and these
concerns span beyond just questions of system design
The role of IR/ML in this process is to deconstruct their
own measures and models in ways that allows a broad
range of researchers and stakeholders to critically
analyze and shape these technologies
In traditional IR, we have made progress in
modeling, measuring, and optimizing for
individual user satisfaction; a key challenge ahead
is to model, measure, and optimize IR systems with
respect to impact on populations of users and
consider disparate impact across subpopulations
Exposure fairness is a multisided problem
It is important to ask not just whether specific content receives
exposure, but who it is exposed to and in what context
Haolun, Mitra, Ma, and Liu. Joint Multisided Exposure Fairness for Recommendation. Under review for SIGIR, ACM. (2022)
Exposure fairness is a multisided problem
Take the example of a job recommendation system
Group-of-users-to-group-of-items fairness (GG-F)
Are groups of items under/over-exposed to groups
of users?
E.g., men being disproportionately recommended
high-paying jobs and women low-paying jobs.
Individual-user-to-Individual-item fairness (II-F)
Are Individual items under/over-exposed to
Individual users?
Individual-user-to-group-of-items fairness (IG-F)
Are groups of items under/over-exposed to
individual users?
E.g., a specific user being disproportionately
recommended low-paying jobs.
Group-of-users-to-Individual-item fairness (GI-F)
Are Individual items under/over-exposed to groups
of users?
E.g., a specific job being disproportionately
recommended to men and not to women and
non-binary people.
All-users-to-Individual-item fairness (AI-F)
Are Individual items under/over-exposed to all users
overall?
E.g., a specific job being disproportionately under-
exposed to all users.
All-users-to-group-of-items fairness (AG-F)
Are groups of items under/over-exposed to all users
overall?
E.g., jobs at Black-owned businesses being
disproportionately under-exposed to all users.
Exposure fairness is a multisided problem
Take the example of a job recommendation system
Group-of-users-to-group-of-items fairness (GG-F)
Are groups of items under/over-exposed to groups
of users?
E.g., men being disproportionately recommended
high-paying jobs and women low-paying jobs.
Individual-user-to-Individual-item fairness (II-F)
Are Individual items under/over-exposed to
Individual users?
Individual-user-to-group-of-items fairness (IG-F)
Are groups of items under/over-exposed to
individual users?
E.g., a specific user being disproportionately
recommended low-paying jobs.
Group-of-users-to-Individual-item fairness (GI-F)
Are Individual items under/over-exposed to groups
of users?
E.g., a specific job being disproportionately
recommended to men and not to women and
non-binary people.
All-users-to-Individual-item fairness (AI-F)
Are Individual items under/over-exposed to all users
overall?
E.g., a specific job being disproportionately under-
exposed to all users.
All-users-to-group-of-items fairness (AG-F)
Are groups of items under/over-exposed to all users
overall?
E.g., jobs at Black-owned businesses being
disproportionately under-exposed to all users.
Exposure fairness is a multisided problem
Take the example of a job recommendation system
Group-of-users-to-group-of-items fairness (GG-F)
Are groups of items under/over-exposed to groups
of users?
E.g., men being disproportionately recommended
high-paying jobs and women low-paying jobs.
Individual-user-to-Individual-item fairness (II-F)
Are Individual items under/over-exposed to
Individual users?
Individual-user-to-group-of-items fairness (IG-F)
Are groups of items under/over-exposed to
individual users?
E.g., a specific user being disproportionately
recommended low-paying jobs.
Group-of-users-to-Individual-item fairness (GI-F)
Are Individual items under/over-exposed to groups
of users?
E.g., a specific job being disproportionately
recommended to men and not to women and
non-binary people.
All-users-to-Individual-item fairness (AI-F)
Are Individual items under/over-exposed to all users
overall?
E.g., a specific job being disproportionately under-
exposed to all users.
All-users-to-group-of-items fairness (AG-F)
Are groups of items under/over-exposed to all users
overall?
E.g., jobs at Black-owned businesses being
disproportionately under-exposed to all users.
Exposure fairness is a multisided problem
Take the example of a job recommendation system
Group-of-users-to-group-of-items fairness (GG-F)
Are groups of items under/over-exposed to groups
of users?
E.g., men being disproportionately recommended
high-paying jobs and women low-paying jobs.
Individual-user-to-Individual-item fairness (II-F)
Are Individual items under/over-exposed to
Individual users?
Individual-user-to-group-of-items fairness (IG-F)
Are groups of items under/over-exposed to
individual users?
E.g., a specific user being disproportionately
recommended low-paying jobs.
Group-of-users-to-Individual-item fairness (GI-F)
Are Individual items under/over-exposed to groups
of users?
E.g., a specific job being disproportionately
recommended to men and not to women and
non-binary people.
All-users-to-Individual-item fairness (AI-F)
Are Individual items under/over-exposed to all users
overall?
E.g., a specific job being disproportionately under-
exposed to all users.
All-users-to-group-of-items fairness (AG-F)
Are groups of items under/over-exposed to all users
overall?
E.g., jobs at Black-owned businesses being
disproportionately under-exposed to all users.
Exposure fairness is a multisided problem
Take the example of a job recommendation system
Group-of-users-to-group-of-items fairness (GG-F)
Are groups of items under/over-exposed to groups
of users?
E.g., men being disproportionately recommended
high-paying jobs and women low-paying jobs.
Individual-user-to-Individual-item fairness (II-F)
Are Individual items under/over-exposed to
Individual users?
Individual-user-to-group-of-items fairness (IG-F)
Are groups of items under/over-exposed to
individual users?
E.g., a specific user being disproportionately
recommended low-paying jobs.
Group-of-users-to-Individual-item fairness (GI-F)
Are Individual items under/over-exposed to groups
of users?
E.g., a specific job being disproportionately
recommended to men and not to women and
non-binary people.
All-users-to-Individual-item fairness (AI-F)
Are Individual items under/over-exposed to all users
overall?
E.g., a specific job being disproportionately under-
exposed to all users.
All-users-to-group-of-items fairness (AG-F)
Are groups of items under/over-exposed to all users
overall?
E.g., jobs at Black-owned businesses being
disproportionately under-exposed to all users.
Exposure fairness is a multisided problem
Take the example of a job recommendation system
Group-of-users-to-group-of-items fairness (GG-F)
Are groups of items under/over-exposed to groups
of users?
E.g., men being disproportionately recommended
high-paying jobs and women low-paying jobs.
Individual-user-to-Individual-item fairness (II-F)
Are Individual items under/over-exposed to
Individual users?
Individual-user-to-group-of-items fairness (IG-F)
Are groups of items under/over-exposed to
individual users?
E.g., a specific user being disproportionately
recommended low-paying jobs.
Group-of-users-to-Individual-item fairness (GI-F)
Are Individual items under/over-exposed to groups
of users?
E.g., a specific job being disproportionately
recommended to men and not to women and
non-binary people.
All-users-to-Individual-item fairness (AI-F)
Are Individual items under/over-exposed to all users
overall?
E.g., a specific job being disproportionately under-
exposed to all users.
All-users-to-group-of-items fairness (AG-F)
Are groups of items under/over-exposed to all users
overall?
E.g., jobs at Black-owned businesses being
disproportionately under-exposed to all users.
Exposure fairness is a multisided problem
Take the example of a job recommendation system
Group-of-users-to-group-of-items fairness (GG-F)
Are groups of items under/over-exposed to groups
of users?
E.g., men being disproportionately recommended
high-paying jobs and women low-paying jobs.
Individual-user-to-Individual-item fairness (II-F)
Are Individual items under/over-exposed to
Individual users?
Individual-user-to-group-of-items fairness (IG-F)
Are groups of items under/over-exposed to
individual users?
E.g., a specific user being disproportionately
recommended low-paying jobs.
Group-of-users-to-Individual-item fairness (GI-F)
Are Individual items under/over-exposed to groups
of users?
E.g., a specific job being disproportionately
recommended to men and not to women and
non-binary people.
All-users-to-Individual-item fairness (AI-F)
Are Individual items under/over-exposed to all users
overall?
E.g., a specific job being disproportionately under-
exposed to all users.
All-users-to-group-of-items fairness (AG-F)
Are groups of items under/over-exposed to all users
overall?
E.g., jobs at Black-owned businesses being
disproportionately under-exposed to all users.
User browsing models and exposure
User browsing models are simplified models of how users inspect
and interact with retrieved results
It estimates the probability that the user inspects a particular item
in a ranked list of items—i.e., the item is exposed to the user
In IR, user models have been implicitly and explicitly employed in
metric definitions and for estimating relevance from historical
logs of user behavior data
For example, let’s consider the RBP user model…
NDCG
RBP
Probability of exposure at different ranks according
to NDCG and RBP user browsing models
exposure event
an item
a ranked list of items
rank of the item in the ranked list
patience factor
Stochastic ranking and expected exposure
In recommendation, Diaz et al. (2020) define a stochastic ranking policy 𝜋𝑢, conditioned on user
𝑢 ∈ U, as a probability distribution over all permutations of items in the collection
The expected exposure of an item 𝑑 for user 𝑢 can then be computed as follows:
Here, 𝑝(𝜖|𝑑,𝜎) can be computed using a user browsing model like RBP as discussed previously
Note: The above formulation can also be applied to search by replacing user with query
Diaz, Mitra, Ekstrand, Biega, and Carterette. Evaluating stochastic rankings with expected exposure. In CIKM, ACM. (2020)
System, target, and random exposure
System exposure. The user-item expected exposure distribution corresponding to a stochastic
ranking policy 𝜋. Correspondingly, we can define a |U|×|D| matrix E, such that E𝑖𝑗 = 𝑝(𝜖|D𝑗 ,𝜋U𝑖
).
Target exposure. The user-item expected exposure distribution corresponding to an ideal
stochastic ranking policy 𝜋*, as defined by some desirable principle (e.g., the equal expected
exposure principle). We denote the corresponding expected exposure matrix as E*.
Random exposure. The user-item expected exposure distribution corresponding to a stochastic
ranking policy 𝜋~ that samples rankings from a uniform distribution over all item permutations.
We denote the corresponding expected exposure matrix as E~.
The deviation of E from E* gives us a quantitative measure of the suboptimality of the retrieval
system under consideration.
Joint multisided exposure (JME) fairness metrics
Haolun, Mitra, Ma, and Liu. Joint Multisided Exposure Fairness for Recommendation. Under review for SIGIR, ACM. (2022)
Toy example
Let, there be 4 candidates (𝑢𝑎1
, 𝑢𝑎2
, 𝑢𝑏1
, 𝑢𝑏2
) and
4 jobs (𝑑𝑥1
, 𝑑𝑥2
, 𝑑𝑦1
, 𝑑𝑦2
)
All 4 jobs are relevant to each of the 4
candidates
The candidates belong to 2 groups 𝑎 (𝑢𝑎1
, 𝑢𝑎2
)
and 𝑏 (𝑢𝑏1
, 𝑢𝑏2
)—e.g., based on gender—and
similarly the jobs belong to 2 groups 𝑥 (𝑑𝑥1
, 𝑑𝑥2
)
and 𝑦 (𝑑𝑦1
, 𝑑𝑦2
)—say based on whether they
pay high or low salaries
Let’s assume that the recommender system
displays only one result at a time and our simple
user model assumes that the user always
inspects the displayed result—i.e., the
probability of exposure is 1 for the displayed
item and 0 for all other items for a given
impression
In this setting, an ideal recommender should
expose each of the four jobs to each candidate
with a probability of 0.25
Toy example
Let, there be 4 candidates (𝑢𝑎1
, 𝑢𝑎2
, 𝑢𝑏1
, 𝑢𝑏2
) and
4 jobs (𝑑𝑥1
, 𝑑𝑥2
, 𝑑𝑦1
, 𝑑𝑦2
)
All 4 jobs are relevant to each of the 4
candidates
The candidates belong to 2 groups 𝑎 (𝑢𝑎1
, 𝑢𝑎2
)
and 𝑏 (𝑢𝑏1
, 𝑢𝑏2
)—e.g., based on gender—and
similarly the jobs belong to 2 groups 𝑥 (𝑑𝑥1
, 𝑑𝑥2
)
and 𝑦 (𝑑𝑦1
, 𝑑𝑦2
)—say based on whether they
pay high or low salaries
Let’s assume that the recommender system
displays only one result at a time and our simple
user model assumes that the user always
inspects the displayed result—i.e., the
probability of exposure is 1 for the displayed
item and 0 for all other items for a given
impression
In this setting, an ideal recommender should
expose each of the four jobs to each candidate
with a probability of 0.25
All of them are equally II-Unfair
Let, there be 4 candidates (𝑢𝑎1
, 𝑢𝑎2
, 𝑢𝑏1
, 𝑢𝑏2
) and
4 jobs (𝑑𝑥1
, 𝑑𝑥2
, 𝑑𝑦1
, 𝑑𝑦2
)
All 4 jobs are relevant to each of the 4
candidates
The candidates belong to 2 groups 𝑎 (𝑢𝑎1
, 𝑢𝑎2
)
and 𝑏 (𝑢𝑏1
, 𝑢𝑏2
)—e.g., based on gender—and
similarly the jobs belong to 2 groups 𝑥 (𝑑𝑥1
, 𝑑𝑥2
)
and 𝑦 (𝑑𝑦1
, 𝑑𝑦2
)—say based on whether they
pay high or low salaries
Let’s assume that the recommender system
displays only one result at a time and our simple
user model assumes that the user always
inspects the displayed result—i.e., the
probability of exposure is 1 for the displayed
item and 0 for all other items for a given
impression
In this setting, an ideal recommender should
expose each of the four jobs to each candidate
with a probability of 0.25
Only (b), (e), and (f) are IG-Unfair
Toy example
Toy example
Let, there be 4 candidates (𝑢𝑎1
, 𝑢𝑎2
, 𝑢𝑏1
, 𝑢𝑏2
) and
4 jobs (𝑑𝑥1
, 𝑑𝑥2
, 𝑑𝑦1
, 𝑑𝑦2
)
All 4 jobs are relevant to each of the 4
candidates
The candidates belong to 2 groups 𝑎 (𝑢𝑎1
, 𝑢𝑎2
)
and 𝑏 (𝑢𝑏1
, 𝑢𝑏2
)—e.g., based on gender—and
similarly the jobs belong to 2 groups 𝑥 (𝑑𝑥1
, 𝑑𝑥2
)
and 𝑦 (𝑑𝑦1
, 𝑑𝑦2
)—say based on whether they
pay high or low salaries
Let’s assume that the recommender system
displays only one result at a time and our simple
user model assumes that the user always
inspects the displayed result—i.e., the
probability of exposure is 1 for the displayed
item and 0 for all other items for a given
impression
In this setting, an ideal recommender should
expose each of the four jobs to each candidate
with a probability of 0.25
Only (c), (d), (e), and (f) are GI-Unfair
Toy example
Let, there be 4 candidates (𝑢𝑎1
, 𝑢𝑎2
, 𝑢𝑏1
, 𝑢𝑏2
) and
4 jobs (𝑑𝑥1
, 𝑑𝑥2
, 𝑑𝑦1
, 𝑑𝑦2
)
All 4 jobs are relevant to each of the 4
candidates
The candidates belong to 2 groups 𝑎 (𝑢𝑎1
, 𝑢𝑎2
)
and 𝑏 (𝑢𝑏1
, 𝑢𝑏2
)—e.g., based on gender—and
similarly the jobs belong to 2 groups 𝑥 (𝑑𝑥1
, 𝑑𝑥2
)
and 𝑦 (𝑑𝑦1
, 𝑑𝑦2
)—say based on whether they
pay high or low salaries
Let’s assume that the recommender system
displays only one result at a time and our simple
user model assumes that the user always
inspects the displayed result—i.e., the
probability of exposure is 1 for the displayed
item and 0 for all other items for a given
impression
In this setting, an ideal recommender should
expose each of the four jobs to each candidate
with a probability of 0.25
Only (e) and (f) are GG-Unfair
Let, there be 4 candidates (𝑢𝑎1
, 𝑢𝑎2
, 𝑢𝑏1
, 𝑢𝑏2
) and
4 jobs (𝑑𝑥1
, 𝑑𝑥2
, 𝑑𝑦1
, 𝑑𝑦2
)
All 4 jobs are relevant to each of the 4
candidates
The candidates belong to 2 groups 𝑎 (𝑢𝑎1
, 𝑢𝑎2
)
and 𝑏 (𝑢𝑏1
, 𝑢𝑏2
)—e.g., based on gender—and
similarly the jobs belong to 2 groups 𝑥 (𝑑𝑥1
, 𝑑𝑥2
)
and 𝑦 (𝑑𝑦1
, 𝑑𝑦2
)—say based on whether they
pay high or low salaries
Let’s assume that the recommender system
displays only one result at a time and our simple
user model assumes that the user always
inspects the displayed result—i.e., the
probability of exposure is 1 for the displayed
item and 0 for all other items for a given
impression
In this setting, an ideal recommender should
expose each of the four jobs to each candidate
with a probability of 0.25
Only (d) and (f) are AI-Unfair
Toy example
Toy example
Let, there be 4 candidates (𝑢𝑎1
, 𝑢𝑎2
, 𝑢𝑏1
, 𝑢𝑏2
) and
4 jobs (𝑑𝑥1
, 𝑑𝑥2
, 𝑑𝑦1
, 𝑑𝑦2
)
All 4 jobs are relevant to each of the 4
candidates
The candidates belong to 2 groups 𝑎 (𝑢𝑎1
, 𝑢𝑎2
)
and 𝑏 (𝑢𝑏1
, 𝑢𝑏2
)—e.g., based on gender—and
similarly the jobs belong to 2 groups 𝑥 (𝑑𝑥1
, 𝑑𝑥2
)
and 𝑦 (𝑑𝑦1
, 𝑑𝑦2
)—say based on whether they
pay high or low salaries
Let’s assume that the recommender system
displays only one result at a time and our simple
user model assumes that the user always
inspects the displayed result—i.e., the
probability of exposure is 1 for the displayed
item and 0 for all other items for a given
impression
In this setting, an ideal recommender should
expose each of the four jobs to each candidate
with a probability of 0.25
Only (f) is AG-Unfair
Relationship between
different JME metrics
Based on the metric definitions, we can
show that a system that is II-Fair (i.e., II-
F=0) will also be fair along the other
five JME-fairness dimensions
Similarly, IG-Fair and GI-Fair
independently implies GG-Fair, and
GG-Fair and AI-Fair implies AG-Fair
Finally, all the other metrics can be
viewed as specific instances of GG-F,
with different (extreme) definitions of
groups on user and item side
II-F=0
IG-F=0 GI-F=0
GG-F=0 AI-F=0
AG-F=0
Disparity and
relevance
Each of our proposed JME-fairness metrics can be
decomposed into a disparity and a relevance component,
such that increasing randomness in the model would
decrease disparity (good!) but also decrease relevance (bad!)
Disparity and
relevance
Each of our proposed JME-fairness metrics can be
decomposed into a disparity and a relevance component,
such that increasing randomness in the model would
decrease disparity (good!) but also decrease relevance (bad!)
Disparity and
relevance
Each of our proposed JME-fairness metrics can be
decomposed into a disparity and a relevance component,
such that increasing randomness in the model would
decrease disparity (good!) but also decrease relevance (bad!)
Different models have
different disparity-relevance
trade-off for each of the
different JME-fairness metrics
How correlated are different
JME-fairness dimensions?
Recall that all six JME-Fairness metrics can be seen as
specific instances of GG-F
For this analysis using MovieLens we had 2 groups
by gender* and 7 groups by age on the user side
and 18 genres on the item side
When we have small number of large groups
“Individual” and “Group” analysis will diverge, and
vice versa
* The gender attribute is available in the MovieLens dataset as a binary annotation. We recognize that
this does not reflect the full spectrum of gender identities, and this is a short-coming of our work.
New metrics, new optimization opportunity!
How can we optimize ranking models for target exposure?
Diaz, Mitra, Ekstrand, Biega, and Carterette. Evaluating stochastic rankings with expected exposure. In CIKM, ACM. (2020)
Stochastic ranking
A stochastic ranking model samples a ranking from a probability distribution over all possible permutations
of items in the collection—i.e., for the same intent it returns a slightly different ranking on each impression
Given a static ranking policy, we can generate a stochastic equivalent using Plackett-Luce sampling—for
example, given items 𝑑1, 𝑑2, 𝑑3, 𝑑4 the probability of sampling a particular ranking 𝑑2, 𝑑1, 𝑑4, 𝑑3 is:
𝜋: a ranking, 𝜙: a transformation, e.g., exponential over score 𝑠𝑖 for document 𝑑𝑖
Equivalent to sequentially sampling documents without replacement with probability 𝜙 𝑠𝑖
restaurants in montreal restaurants in montreal
restaurants in montreal
restaurants in montreal
Luce. Individual Choice Behavior. (1959)
Plackett. The Analysis of Permutations. Journal of the Royal Statistical Society: Series C (Applied Statistics). (1975)
Gradient-based optimization for target exposure
Approach
1. Use the target model to score the items
2. Compute PL sampling probability as a
function of the item scores
3. Sample multiple rankings
4. Compute expected system exposure
across sampled rankings
5. Compute the loss as a difference between
system and target exposure
6. Backpropagate!
Challenges and solutions
The key challenge is the proposed approach is
that both the sampling and the ranking steps
are non-differentiable!
For sampling, we can use Gumbel sampling
as a differentiable approximation
For ranking, we can employ SmoothRank /
ApproxRank as differentiable approximations
of the ranking step
Wu, Chang, Zheng, and Zha. Smoothing DCG for learning to rank: A novel approach using smoothed hinge functions. In Proc. CIKM, ACM. (2009)
Qin, Liu, and Li. A general approximation framework for direct optimization of information retrieval measures. Information retrieval. (2010)
Bruch, Han, Bendersky, and Najork. A stochastic treatment of learning to rank scoring functions. In Proc. WSDM, ACM. (2020)
,
Gradient-based optimization for target exposure
add independently
sampled Gumbel noise
neural scoring
function
compute smooth
rank value
compute exposure
using user model
compute loss with
target exposure
compute average
exposure
items target
exposure
Diaz, Mitra, Ekstrand, Biega, and Carterette. Evaluating stochastic rankings with expected exposure. In CIKM, ACM. (2020)
Trading-off different JME-fairness metrics
We can simultaneously optimize for multiple exposure metrics by
combining them linearly
For example,
Preliminary experiments indicate that we can significantly
minimize GG-F with minimal degradation to II-F and relevance
Discussion
True vs. observed relevance labels. The computation of target exposure itself raises fairness questions. E.g., the equal expected
exposure principle assumes we have access to true relevance labels, but in fact the observed labels reflect huge historical social
biases. E.g., In the job recommendation scenario, it may be more appropriate to define GG-F target exposure for high and low
paying jobs to be uniform across user groups, irrespective of historical disparities reflected in the data.
Choice of group attributes. The choice of group attributes necessitates reflecting on historical and socioeconomic contexts. We
note that our formulation can also be extended to handling multiple group attributes on each side. However, that raises questions
of intersectional fairness that we haven’t yet studied in our work.
Beyond two-sided exposure fairness. While we have primarily focused on two-sided exposure fairness so far, we envision that
extending that to additional stakeholder may also be important. E.g., in product search exposure fairness may concern with being
fair to consumers, manufacturers, and retailers.
Incorporating model uncertainty. The stochastic ranking policies we have considered so far involves randomizing a static policy
with model-independent sampling of noise. In contrast, the stochasticity could also be informed by the model’s own uncertainty
in its prediction. This is an area for potential future work.

Mais conteúdo relacionado

Mais procurados

Find it! Nail it! Boosting e-commerce search conversions with machine learnin...
Find it! Nail it!Boosting e-commerce search conversions with machine learnin...Find it! Nail it!Boosting e-commerce search conversions with machine learnin...
Find it! Nail it! Boosting e-commerce search conversions with machine learnin...
Rakuten Group, Inc.
 
Polyglot Processing - An Introduction 1.0
Polyglot Processing - An Introduction 1.0 Polyglot Processing - An Introduction 1.0
Polyglot Processing - An Introduction 1.0
Dr. Mohan K. Bavirisetty
 
Anatomy of an eCommerce Search Engine by Mayur Datar
Anatomy of an eCommerce Search Engine by Mayur DatarAnatomy of an eCommerce Search Engine by Mayur Datar
Anatomy of an eCommerce Search Engine by Mayur Datar
Naresh Jain
 
Dual Embedding Space Model (DESM)
Dual Embedding Space Model (DESM)Dual Embedding Space Model (DESM)
Dual Embedding Space Model (DESM)
Bhaskar Mitra
 
Technologies pour le Big Data
Technologies pour le Big DataTechnologies pour le Big Data
Technologies pour le Big Data
Minyar Sassi Hidri
 
Neural Learning to Rank
Neural Learning to RankNeural Learning to Rank
Neural Learning to Rank
Bhaskar Mitra
 
Big Data : concepts, cas d'usage et tendances
Big Data : concepts, cas d'usage et tendancesBig Data : concepts, cas d'usage et tendances
Big Data : concepts, cas d'usage et tendances
Jean-Michel Franco
 
A Multi-Armed Bandit Framework For Recommendations at Netflix
A Multi-Armed Bandit Framework For Recommendations at NetflixA Multi-Armed Bandit Framework For Recommendations at Netflix
A Multi-Armed Bandit Framework For Recommendations at Netflix
Jaya Kawale
 
Machine Learning for retail and ecommerce
Machine Learning for retail and ecommerceMachine Learning for retail and ecommerce
Machine Learning for retail and ecommerce
Andrei Lopatenko
 
Introduction au Machine Learning
Introduction au Machine Learning Introduction au Machine Learning
Introduction au Machine Learning
Novagen Conseil
 
Försäkringskassan: Neo4j as an Information Hub (GraphSummit Stockholm 2023)
Försäkringskassan: Neo4j as an Information Hub (GraphSummit Stockholm 2023)Försäkringskassan: Neo4j as an Information Hub (GraphSummit Stockholm 2023)
Försäkringskassan: Neo4j as an Information Hub (GraphSummit Stockholm 2023)
Neo4j
 
A Simple Introduction to Neural Information Retrieval
A Simple Introduction to Neural Information RetrievalA Simple Introduction to Neural Information Retrieval
A Simple Introduction to Neural Information Retrieval
Bhaskar Mitra
 
Unified Approach to Interpret Machine Learning Model: SHAP + LIME
Unified Approach to Interpret Machine Learning Model: SHAP + LIMEUnified Approach to Interpret Machine Learning Model: SHAP + LIME
Unified Approach to Interpret Machine Learning Model: SHAP + LIME
Databricks
 
Large scale-lm-part1
Large scale-lm-part1Large scale-lm-part1
Large scale-lm-part1
gohyunwoong
 
Building an Enterprise Knowledge Graph @Uber: Lessons from Reality
Building an Enterprise Knowledge Graph @Uber: Lessons from RealityBuilding an Enterprise Knowledge Graph @Uber: Lessons from Reality
Building an Enterprise Knowledge Graph @Uber: Lessons from Reality
Joshua Shinavier
 
Deep learning for NLP and Transformer
 Deep learning for NLP  and Transformer Deep learning for NLP  and Transformer
Deep learning for NLP and Transformer
Arvind Devaraj
 
Layout lm paper review
Layout lm paper review Layout lm paper review
Layout lm paper review
taeseon ryu
 
Big data
Big dataBig data
Neo4j - Cas d'usages pour votre métier
Neo4j - Cas d'usages pour votre métierNeo4j - Cas d'usages pour votre métier
Neo4j - Cas d'usages pour votre métier
Neo4j
 
تلخيص تكنولوجيا لطلاب التوجيهي
تلخيص تكنولوجيا لطلاب التوجيهيتلخيص تكنولوجيا لطلاب التوجيهي
تلخيص تكنولوجيا لطلاب التوجيهيSaif mubaslat El Tubasi
 

Mais procurados (20)

Find it! Nail it! Boosting e-commerce search conversions with machine learnin...
Find it! Nail it!Boosting e-commerce search conversions with machine learnin...Find it! Nail it!Boosting e-commerce search conversions with machine learnin...
Find it! Nail it! Boosting e-commerce search conversions with machine learnin...
 
Polyglot Processing - An Introduction 1.0
Polyglot Processing - An Introduction 1.0 Polyglot Processing - An Introduction 1.0
Polyglot Processing - An Introduction 1.0
 
Anatomy of an eCommerce Search Engine by Mayur Datar
Anatomy of an eCommerce Search Engine by Mayur DatarAnatomy of an eCommerce Search Engine by Mayur Datar
Anatomy of an eCommerce Search Engine by Mayur Datar
 
Dual Embedding Space Model (DESM)
Dual Embedding Space Model (DESM)Dual Embedding Space Model (DESM)
Dual Embedding Space Model (DESM)
 
Technologies pour le Big Data
Technologies pour le Big DataTechnologies pour le Big Data
Technologies pour le Big Data
 
Neural Learning to Rank
Neural Learning to RankNeural Learning to Rank
Neural Learning to Rank
 
Big Data : concepts, cas d'usage et tendances
Big Data : concepts, cas d'usage et tendancesBig Data : concepts, cas d'usage et tendances
Big Data : concepts, cas d'usage et tendances
 
A Multi-Armed Bandit Framework For Recommendations at Netflix
A Multi-Armed Bandit Framework For Recommendations at NetflixA Multi-Armed Bandit Framework For Recommendations at Netflix
A Multi-Armed Bandit Framework For Recommendations at Netflix
 
Machine Learning for retail and ecommerce
Machine Learning for retail and ecommerceMachine Learning for retail and ecommerce
Machine Learning for retail and ecommerce
 
Introduction au Machine Learning
Introduction au Machine Learning Introduction au Machine Learning
Introduction au Machine Learning
 
Försäkringskassan: Neo4j as an Information Hub (GraphSummit Stockholm 2023)
Försäkringskassan: Neo4j as an Information Hub (GraphSummit Stockholm 2023)Försäkringskassan: Neo4j as an Information Hub (GraphSummit Stockholm 2023)
Försäkringskassan: Neo4j as an Information Hub (GraphSummit Stockholm 2023)
 
A Simple Introduction to Neural Information Retrieval
A Simple Introduction to Neural Information RetrievalA Simple Introduction to Neural Information Retrieval
A Simple Introduction to Neural Information Retrieval
 
Unified Approach to Interpret Machine Learning Model: SHAP + LIME
Unified Approach to Interpret Machine Learning Model: SHAP + LIMEUnified Approach to Interpret Machine Learning Model: SHAP + LIME
Unified Approach to Interpret Machine Learning Model: SHAP + LIME
 
Large scale-lm-part1
Large scale-lm-part1Large scale-lm-part1
Large scale-lm-part1
 
Building an Enterprise Knowledge Graph @Uber: Lessons from Reality
Building an Enterprise Knowledge Graph @Uber: Lessons from RealityBuilding an Enterprise Knowledge Graph @Uber: Lessons from Reality
Building an Enterprise Knowledge Graph @Uber: Lessons from Reality
 
Deep learning for NLP and Transformer
 Deep learning for NLP  and Transformer Deep learning for NLP  and Transformer
Deep learning for NLP and Transformer
 
Layout lm paper review
Layout lm paper review Layout lm paper review
Layout lm paper review
 
Big data
Big dataBig data
Big data
 
Neo4j - Cas d'usages pour votre métier
Neo4j - Cas d'usages pour votre métierNeo4j - Cas d'usages pour votre métier
Neo4j - Cas d'usages pour votre métier
 
تلخيص تكنولوجيا لطلاب التوجيهي
تلخيص تكنولوجيا لطلاب التوجيهيتلخيص تكنولوجيا لطلاب التوجيهي
تلخيص تكنولوجيا لطلاب التوجيهي
 

Semelhante a Multisided Exposure Fairness for Search and Recommendation

Handout ch7
Handout ch7Handout ch7
Handout ch7
Elaine Cruz
 
Work Deviant Behaviour and Team Cooperation in Selected Manufacturing Compani...
Work Deviant Behaviour and Team Cooperation in Selected Manufacturing Compani...Work Deviant Behaviour and Team Cooperation in Selected Manufacturing Compani...
Work Deviant Behaviour and Team Cooperation in Selected Manufacturing Compani...
ijtsrd
 
Diversity management in apple inc.
Diversity management in apple inc.Diversity management in apple inc.
Diversity management in apple inc.
Service_supportAssignment
 
Practice of International Trade EIMSO2 Lecture V3
Practice of International Trade EIMSO2 Lecture V3Practice of International Trade EIMSO2 Lecture V3
Practice of International Trade EIMSO2 Lecture V3
Fan DiFu, Ph.D. (Steve)
 
20120140506003
2012014050600320120140506003
20120140506003
IAEME Publication
 
Disability-Inclusion-Report-Business-Imperative.pdf
Disability-Inclusion-Report-Business-Imperative.pdfDisability-Inclusion-Report-Business-Imperative.pdf
Disability-Inclusion-Report-Business-Imperative.pdf
vccstr1
 
Gender board diversity spillovers and the public eye
Gender board diversity spillovers and the public eyeGender board diversity spillovers and the public eye
Gender board diversity spillovers and the public eye
GRAPE
 
Fairness-aware Machine Learning: Practical Challenges and Lessons Learned (WS...
Fairness-aware Machine Learning: Practical Challenges and Lessons Learned (WS...Fairness-aware Machine Learning: Practical Challenges and Lessons Learned (WS...
Fairness-aware Machine Learning: Practical Challenges and Lessons Learned (WS...
Krishnaram Kenthapadi
 
Labor Market Effects of Mandatory Benefit Regulations for Maids in Ecuador
Labor Market Effects of Mandatory Benefit Regulations for Maids in EcuadorLabor Market Effects of Mandatory Benefit Regulations for Maids in Ecuador
Labor Market Effects of Mandatory Benefit Regulations for Maids in Ecuador
GDNet - Global Development Network, Cairo Office
 
Excelsior College Business External Environment Report.docx
Excelsior College Business External Environment Report.docxExcelsior College Business External Environment Report.docx
Excelsior College Business External Environment Report.docx
bkbk37
 
From Hype to Reality: AI in Market Research
From Hype to Reality: AI in Market Research From Hype to Reality: AI in Market Research
From Hype to Reality: AI in Market Research
Tom De Ruyck
 
Alibabas Internal( just internal) EnvironmentTimothy .docx
Alibabas Internal( just internal)  EnvironmentTimothy .docxAlibabas Internal( just internal)  EnvironmentTimothy .docx
Alibabas Internal( just internal) EnvironmentTimothy .docx
galerussel59292
 
IRJET- Survey on Facial Feature Detection Methods in Offline Recommendation b...
IRJET- Survey on Facial Feature Detection Methods in Offline Recommendation b...IRJET- Survey on Facial Feature Detection Methods in Offline Recommendation b...
IRJET- Survey on Facial Feature Detection Methods in Offline Recommendation b...
IRJET Journal
 
AI & DEI: With Great Opportunities Comes Great HR Responsibility
AI & DEI: With Great Opportunities Comes Great HR ResponsibilityAI & DEI: With Great Opportunities Comes Great HR Responsibility
AI & DEI: With Great Opportunities Comes Great HR Responsibility
Aggregage
 
Wiegel, Vincent; Lean contingency framework extended abstract
Wiegel, Vincent; Lean contingency framework extended abstractWiegel, Vincent; Lean contingency framework extended abstract
Wiegel, Vincent; Lean contingency framework extended abstract
HAN Lean-QRM Centrum / HAN Lectoraat Lean
 
Mba2216 business research week 3 research methodology 0613
Mba2216 business research week 3 research methodology 0613Mba2216 business research week 3 research methodology 0613
Mba2216 business research week 3 research methodology 0613
Stephen Ong
 
1. IntroductionIn the modern society, all the enterprises in the.docx
1. IntroductionIn the modern society, all the enterprises in the.docx1. IntroductionIn the modern society, all the enterprises in the.docx
1. IntroductionIn the modern society, all the enterprises in the.docx
SONU61709
 
Sports art Eco-Powr _ Whitepaper
Sports art Eco-Powr _ WhitepaperSports art Eco-Powr _ Whitepaper
Sports art Eco-Powr _ Whitepaper
Gary Oleinik
 
ai_and_you_slide_template.pptx
ai_and_you_slide_template.pptxai_and_you_slide_template.pptx
ai_and_you_slide_template.pptx
ganeshjilo
 
Fair Recommender Systems
Fair Recommender Systems Fair Recommender Systems
Fair Recommender Systems
Sharmistha Chatterjee
 

Semelhante a Multisided Exposure Fairness for Search and Recommendation (20)

Handout ch7
Handout ch7Handout ch7
Handout ch7
 
Work Deviant Behaviour and Team Cooperation in Selected Manufacturing Compani...
Work Deviant Behaviour and Team Cooperation in Selected Manufacturing Compani...Work Deviant Behaviour and Team Cooperation in Selected Manufacturing Compani...
Work Deviant Behaviour and Team Cooperation in Selected Manufacturing Compani...
 
Diversity management in apple inc.
Diversity management in apple inc.Diversity management in apple inc.
Diversity management in apple inc.
 
Practice of International Trade EIMSO2 Lecture V3
Practice of International Trade EIMSO2 Lecture V3Practice of International Trade EIMSO2 Lecture V3
Practice of International Trade EIMSO2 Lecture V3
 
20120140506003
2012014050600320120140506003
20120140506003
 
Disability-Inclusion-Report-Business-Imperative.pdf
Disability-Inclusion-Report-Business-Imperative.pdfDisability-Inclusion-Report-Business-Imperative.pdf
Disability-Inclusion-Report-Business-Imperative.pdf
 
Gender board diversity spillovers and the public eye
Gender board diversity spillovers and the public eyeGender board diversity spillovers and the public eye
Gender board diversity spillovers and the public eye
 
Fairness-aware Machine Learning: Practical Challenges and Lessons Learned (WS...
Fairness-aware Machine Learning: Practical Challenges and Lessons Learned (WS...Fairness-aware Machine Learning: Practical Challenges and Lessons Learned (WS...
Fairness-aware Machine Learning: Practical Challenges and Lessons Learned (WS...
 
Labor Market Effects of Mandatory Benefit Regulations for Maids in Ecuador
Labor Market Effects of Mandatory Benefit Regulations for Maids in EcuadorLabor Market Effects of Mandatory Benefit Regulations for Maids in Ecuador
Labor Market Effects of Mandatory Benefit Regulations for Maids in Ecuador
 
Excelsior College Business External Environment Report.docx
Excelsior College Business External Environment Report.docxExcelsior College Business External Environment Report.docx
Excelsior College Business External Environment Report.docx
 
From Hype to Reality: AI in Market Research
From Hype to Reality: AI in Market Research From Hype to Reality: AI in Market Research
From Hype to Reality: AI in Market Research
 
Alibabas Internal( just internal) EnvironmentTimothy .docx
Alibabas Internal( just internal)  EnvironmentTimothy .docxAlibabas Internal( just internal)  EnvironmentTimothy .docx
Alibabas Internal( just internal) EnvironmentTimothy .docx
 
IRJET- Survey on Facial Feature Detection Methods in Offline Recommendation b...
IRJET- Survey on Facial Feature Detection Methods in Offline Recommendation b...IRJET- Survey on Facial Feature Detection Methods in Offline Recommendation b...
IRJET- Survey on Facial Feature Detection Methods in Offline Recommendation b...
 
AI & DEI: With Great Opportunities Comes Great HR Responsibility
AI & DEI: With Great Opportunities Comes Great HR ResponsibilityAI & DEI: With Great Opportunities Comes Great HR Responsibility
AI & DEI: With Great Opportunities Comes Great HR Responsibility
 
Wiegel, Vincent; Lean contingency framework extended abstract
Wiegel, Vincent; Lean contingency framework extended abstractWiegel, Vincent; Lean contingency framework extended abstract
Wiegel, Vincent; Lean contingency framework extended abstract
 
Mba2216 business research week 3 research methodology 0613
Mba2216 business research week 3 research methodology 0613Mba2216 business research week 3 research methodology 0613
Mba2216 business research week 3 research methodology 0613
 
1. IntroductionIn the modern society, all the enterprises in the.docx
1. IntroductionIn the modern society, all the enterprises in the.docx1. IntroductionIn the modern society, all the enterprises in the.docx
1. IntroductionIn the modern society, all the enterprises in the.docx
 
Sports art Eco-Powr _ Whitepaper
Sports art Eco-Powr _ WhitepaperSports art Eco-Powr _ Whitepaper
Sports art Eco-Powr _ Whitepaper
 
ai_and_you_slide_template.pptx
ai_and_you_slide_template.pptxai_and_you_slide_template.pptx
ai_and_you_slide_template.pptx
 
Fair Recommender Systems
Fair Recommender Systems Fair Recommender Systems
Fair Recommender Systems
 

Mais de Bhaskar Mitra

Search and Society: Reimagining Information Access for Radical Futures
Search and Society: Reimagining Information Access for Radical FuturesSearch and Society: Reimagining Information Access for Radical Futures
Search and Society: Reimagining Information Access for Radical Futures
Bhaskar Mitra
 
Neural Information Retrieval: In search of meaningful progress
Neural Information Retrieval: In search of meaningful progressNeural Information Retrieval: In search of meaningful progress
Neural Information Retrieval: In search of meaningful progress
Bhaskar Mitra
 
Conformer-Kernel with Query Term Independence @ TREC 2020 Deep Learning Track
Conformer-Kernel with Query Term Independence @ TREC 2020 Deep Learning TrackConformer-Kernel with Query Term Independence @ TREC 2020 Deep Learning Track
Conformer-Kernel with Query Term Independence @ TREC 2020 Deep Learning Track
Bhaskar Mitra
 
Neural Learning to Rank
Neural Learning to RankNeural Learning to Rank
Neural Learning to Rank
Bhaskar Mitra
 
Duet @ TREC 2019 Deep Learning Track
Duet @ TREC 2019 Deep Learning TrackDuet @ TREC 2019 Deep Learning Track
Duet @ TREC 2019 Deep Learning Track
Bhaskar Mitra
 
Benchmarking for Neural Information Retrieval: MS MARCO, TREC, and Beyond
Benchmarking for Neural Information Retrieval: MS MARCO, TREC, and BeyondBenchmarking for Neural Information Retrieval: MS MARCO, TREC, and Beyond
Benchmarking for Neural Information Retrieval: MS MARCO, TREC, and Beyond
Bhaskar Mitra
 
Deep Learning for Search
Deep Learning for SearchDeep Learning for Search
Deep Learning for Search
Bhaskar Mitra
 
Neural Learning to Rank
Neural Learning to RankNeural Learning to Rank
Neural Learning to Rank
Bhaskar Mitra
 
Deep Learning for Search
Deep Learning for SearchDeep Learning for Search
Deep Learning for Search
Bhaskar Mitra
 
Adversarial and reinforcement learning-based approaches to information retrieval
Adversarial and reinforcement learning-based approaches to information retrievalAdversarial and reinforcement learning-based approaches to information retrieval
Adversarial and reinforcement learning-based approaches to information retrieval
Bhaskar Mitra
 
5 Lessons Learned from Designing Neural Models for Information Retrieval
5 Lessons Learned from Designing Neural Models for Information Retrieval5 Lessons Learned from Designing Neural Models for Information Retrieval
5 Lessons Learned from Designing Neural Models for Information Retrieval
Bhaskar Mitra
 
Neural Models for Information Retrieval
Neural Models for Information RetrievalNeural Models for Information Retrieval
Neural Models for Information Retrieval
Bhaskar Mitra
 
Neural Models for Document Ranking
Neural Models for Document RankingNeural Models for Document Ranking
Neural Models for Document Ranking
Bhaskar Mitra
 
Neural Models for Information Retrieval
Neural Models for Information RetrievalNeural Models for Information Retrieval
Neural Models for Information Retrieval
Bhaskar Mitra
 
Neu-IR 2017: welcome
Neu-IR 2017: welcomeNeu-IR 2017: welcome
Neu-IR 2017: welcome
Bhaskar Mitra
 
The Duet model
The Duet modelThe Duet model
The Duet model
Bhaskar Mitra
 
Neural Text Embeddings for Information Retrieval (WSDM 2017)
Neural Text Embeddings for Information Retrieval (WSDM 2017)Neural Text Embeddings for Information Retrieval (WSDM 2017)
Neural Text Embeddings for Information Retrieval (WSDM 2017)
Bhaskar Mitra
 
Query Expansion with Locally-Trained Word Embeddings (ACL 2016)
Query Expansion with Locally-Trained Word Embeddings (ACL 2016)Query Expansion with Locally-Trained Word Embeddings (ACL 2016)
Query Expansion with Locally-Trained Word Embeddings (ACL 2016)
Bhaskar Mitra
 
Query Expansion with Locally-Trained Word Embeddings (Neu-IR 2016)
Query Expansion with Locally-Trained Word Embeddings (Neu-IR 2016)Query Expansion with Locally-Trained Word Embeddings (Neu-IR 2016)
Query Expansion with Locally-Trained Word Embeddings (Neu-IR 2016)
Bhaskar Mitra
 

Mais de Bhaskar Mitra (19)

Search and Society: Reimagining Information Access for Radical Futures
Search and Society: Reimagining Information Access for Radical FuturesSearch and Society: Reimagining Information Access for Radical Futures
Search and Society: Reimagining Information Access for Radical Futures
 
Neural Information Retrieval: In search of meaningful progress
Neural Information Retrieval: In search of meaningful progressNeural Information Retrieval: In search of meaningful progress
Neural Information Retrieval: In search of meaningful progress
 
Conformer-Kernel with Query Term Independence @ TREC 2020 Deep Learning Track
Conformer-Kernel with Query Term Independence @ TREC 2020 Deep Learning TrackConformer-Kernel with Query Term Independence @ TREC 2020 Deep Learning Track
Conformer-Kernel with Query Term Independence @ TREC 2020 Deep Learning Track
 
Neural Learning to Rank
Neural Learning to RankNeural Learning to Rank
Neural Learning to Rank
 
Duet @ TREC 2019 Deep Learning Track
Duet @ TREC 2019 Deep Learning TrackDuet @ TREC 2019 Deep Learning Track
Duet @ TREC 2019 Deep Learning Track
 
Benchmarking for Neural Information Retrieval: MS MARCO, TREC, and Beyond
Benchmarking for Neural Information Retrieval: MS MARCO, TREC, and BeyondBenchmarking for Neural Information Retrieval: MS MARCO, TREC, and Beyond
Benchmarking for Neural Information Retrieval: MS MARCO, TREC, and Beyond
 
Deep Learning for Search
Deep Learning for SearchDeep Learning for Search
Deep Learning for Search
 
Neural Learning to Rank
Neural Learning to RankNeural Learning to Rank
Neural Learning to Rank
 
Deep Learning for Search
Deep Learning for SearchDeep Learning for Search
Deep Learning for Search
 
Adversarial and reinforcement learning-based approaches to information retrieval
Adversarial and reinforcement learning-based approaches to information retrievalAdversarial and reinforcement learning-based approaches to information retrieval
Adversarial and reinforcement learning-based approaches to information retrieval
 
5 Lessons Learned from Designing Neural Models for Information Retrieval
5 Lessons Learned from Designing Neural Models for Information Retrieval5 Lessons Learned from Designing Neural Models for Information Retrieval
5 Lessons Learned from Designing Neural Models for Information Retrieval
 
Neural Models for Information Retrieval
Neural Models for Information RetrievalNeural Models for Information Retrieval
Neural Models for Information Retrieval
 
Neural Models for Document Ranking
Neural Models for Document RankingNeural Models for Document Ranking
Neural Models for Document Ranking
 
Neural Models for Information Retrieval
Neural Models for Information RetrievalNeural Models for Information Retrieval
Neural Models for Information Retrieval
 
Neu-IR 2017: welcome
Neu-IR 2017: welcomeNeu-IR 2017: welcome
Neu-IR 2017: welcome
 
The Duet model
The Duet modelThe Duet model
The Duet model
 
Neural Text Embeddings for Information Retrieval (WSDM 2017)
Neural Text Embeddings for Information Retrieval (WSDM 2017)Neural Text Embeddings for Information Retrieval (WSDM 2017)
Neural Text Embeddings for Information Retrieval (WSDM 2017)
 
Query Expansion with Locally-Trained Word Embeddings (ACL 2016)
Query Expansion with Locally-Trained Word Embeddings (ACL 2016)Query Expansion with Locally-Trained Word Embeddings (ACL 2016)
Query Expansion with Locally-Trained Word Embeddings (ACL 2016)
 
Query Expansion with Locally-Trained Word Embeddings (Neu-IR 2016)
Query Expansion with Locally-Trained Word Embeddings (Neu-IR 2016)Query Expansion with Locally-Trained Word Embeddings (Neu-IR 2016)
Query Expansion with Locally-Trained Word Embeddings (Neu-IR 2016)
 

Último

Crafting Excellence: A Comprehensive Guide to iOS Mobile App Development Serv...
Crafting Excellence: A Comprehensive Guide to iOS Mobile App Development Serv...Crafting Excellence: A Comprehensive Guide to iOS Mobile App Development Serv...
Crafting Excellence: A Comprehensive Guide to iOS Mobile App Development Serv...
Pitangent Analytics & Technology Solutions Pvt. Ltd
 
PRODUCT LISTING OPTIMIZATION PRESENTATION.pptx
PRODUCT LISTING OPTIMIZATION PRESENTATION.pptxPRODUCT LISTING OPTIMIZATION PRESENTATION.pptx
PRODUCT LISTING OPTIMIZATION PRESENTATION.pptx
christinelarrosa
 
Demystifying Knowledge Management through Storytelling
Demystifying Knowledge Management through StorytellingDemystifying Knowledge Management through Storytelling
Demystifying Knowledge Management through Storytelling
Enterprise Knowledge
 
Christine's Product Research Presentation.pptx
Christine's Product Research Presentation.pptxChristine's Product Research Presentation.pptx
Christine's Product Research Presentation.pptx
christinelarrosa
 
"$10 thousand per minute of downtime: architecture, queues, streaming and fin...
"$10 thousand per minute of downtime: architecture, queues, streaming and fin..."$10 thousand per minute of downtime: architecture, queues, streaming and fin...
"$10 thousand per minute of downtime: architecture, queues, streaming and fin...
Fwdays
 
Taking AI to the Next Level in Manufacturing.pdf
Taking AI to the Next Level in Manufacturing.pdfTaking AI to the Next Level in Manufacturing.pdf
Taking AI to the Next Level in Manufacturing.pdf
ssuserfac0301
 
Astute Business Solutions | Oracle Cloud Partner |
Astute Business Solutions | Oracle Cloud Partner |Astute Business Solutions | Oracle Cloud Partner |
Astute Business Solutions | Oracle Cloud Partner |
AstuteBusiness
 
High performance Serverless Java on AWS- GoTo Amsterdam 2024
High performance Serverless Java on AWS- GoTo Amsterdam 2024High performance Serverless Java on AWS- GoTo Amsterdam 2024
High performance Serverless Java on AWS- GoTo Amsterdam 2024
Vadym Kazulkin
 
Mutation Testing for Task-Oriented Chatbots
Mutation Testing for Task-Oriented ChatbotsMutation Testing for Task-Oriented Chatbots
Mutation Testing for Task-Oriented Chatbots
Pablo Gómez Abajo
 
[OReilly Superstream] Occupy the Space: A grassroots guide to engineering (an...
[OReilly Superstream] Occupy the Space: A grassroots guide to engineering (an...[OReilly Superstream] Occupy the Space: A grassroots guide to engineering (an...
[OReilly Superstream] Occupy the Space: A grassroots guide to engineering (an...
Jason Yip
 
GraphRAG for LifeSciences Hands-On with the Clinical Knowledge Graph
GraphRAG for LifeSciences Hands-On with the Clinical Knowledge GraphGraphRAG for LifeSciences Hands-On with the Clinical Knowledge Graph
GraphRAG for LifeSciences Hands-On with the Clinical Knowledge Graph
Neo4j
 
Connector Corner: Seamlessly power UiPath Apps, GenAI with prebuilt connectors
Connector Corner: Seamlessly power UiPath Apps, GenAI with prebuilt connectorsConnector Corner: Seamlessly power UiPath Apps, GenAI with prebuilt connectors
Connector Corner: Seamlessly power UiPath Apps, GenAI with prebuilt connectors
DianaGray10
 
Session 1 - Intro to Robotic Process Automation.pdf
Session 1 - Intro to Robotic Process Automation.pdfSession 1 - Intro to Robotic Process Automation.pdf
Session 1 - Intro to Robotic Process Automation.pdf
UiPathCommunity
 
zkStudyClub - LatticeFold: A Lattice-based Folding Scheme and its Application...
zkStudyClub - LatticeFold: A Lattice-based Folding Scheme and its Application...zkStudyClub - LatticeFold: A Lattice-based Folding Scheme and its Application...
zkStudyClub - LatticeFold: A Lattice-based Folding Scheme and its Application...
Alex Pruden
 
Apps Break Data
Apps Break DataApps Break Data
Apps Break Data
Ivo Velitchkov
 
inQuba Webinar Mastering Customer Journey Management with Dr Graham Hill
inQuba Webinar Mastering Customer Journey Management with Dr Graham HillinQuba Webinar Mastering Customer Journey Management with Dr Graham Hill
inQuba Webinar Mastering Customer Journey Management with Dr Graham Hill
LizaNolte
 
Main news related to the CCS TSI 2023 (2023/1695)
Main news related to the CCS TSI 2023 (2023/1695)Main news related to the CCS TSI 2023 (2023/1695)
Main news related to the CCS TSI 2023 (2023/1695)
Jakub Marek
 
Skybuffer SAM4U tool for SAP license adoption
Skybuffer SAM4U tool for SAP license adoptionSkybuffer SAM4U tool for SAP license adoption
Skybuffer SAM4U tool for SAP license adoption
Tatiana Kojar
 
Leveraging the Graph for Clinical Trials and Standards
Leveraging the Graph for Clinical Trials and StandardsLeveraging the Graph for Clinical Trials and Standards
Leveraging the Graph for Clinical Trials and Standards
Neo4j
 
Must Know Postgres Extension for DBA and Developer during Migration
Must Know Postgres Extension for DBA and Developer during MigrationMust Know Postgres Extension for DBA and Developer during Migration
Must Know Postgres Extension for DBA and Developer during Migration
Mydbops
 

Último (20)

Crafting Excellence: A Comprehensive Guide to iOS Mobile App Development Serv...
Crafting Excellence: A Comprehensive Guide to iOS Mobile App Development Serv...Crafting Excellence: A Comprehensive Guide to iOS Mobile App Development Serv...
Crafting Excellence: A Comprehensive Guide to iOS Mobile App Development Serv...
 
PRODUCT LISTING OPTIMIZATION PRESENTATION.pptx
PRODUCT LISTING OPTIMIZATION PRESENTATION.pptxPRODUCT LISTING OPTIMIZATION PRESENTATION.pptx
PRODUCT LISTING OPTIMIZATION PRESENTATION.pptx
 
Demystifying Knowledge Management through Storytelling
Demystifying Knowledge Management through StorytellingDemystifying Knowledge Management through Storytelling
Demystifying Knowledge Management through Storytelling
 
Christine's Product Research Presentation.pptx
Christine's Product Research Presentation.pptxChristine's Product Research Presentation.pptx
Christine's Product Research Presentation.pptx
 
"$10 thousand per minute of downtime: architecture, queues, streaming and fin...
"$10 thousand per minute of downtime: architecture, queues, streaming and fin..."$10 thousand per minute of downtime: architecture, queues, streaming and fin...
"$10 thousand per minute of downtime: architecture, queues, streaming and fin...
 
Taking AI to the Next Level in Manufacturing.pdf
Taking AI to the Next Level in Manufacturing.pdfTaking AI to the Next Level in Manufacturing.pdf
Taking AI to the Next Level in Manufacturing.pdf
 
Astute Business Solutions | Oracle Cloud Partner |
Astute Business Solutions | Oracle Cloud Partner |Astute Business Solutions | Oracle Cloud Partner |
Astute Business Solutions | Oracle Cloud Partner |
 
High performance Serverless Java on AWS- GoTo Amsterdam 2024
High performance Serverless Java on AWS- GoTo Amsterdam 2024High performance Serverless Java on AWS- GoTo Amsterdam 2024
High performance Serverless Java on AWS- GoTo Amsterdam 2024
 
Mutation Testing for Task-Oriented Chatbots
Mutation Testing for Task-Oriented ChatbotsMutation Testing for Task-Oriented Chatbots
Mutation Testing for Task-Oriented Chatbots
 
[OReilly Superstream] Occupy the Space: A grassroots guide to engineering (an...
[OReilly Superstream] Occupy the Space: A grassroots guide to engineering (an...[OReilly Superstream] Occupy the Space: A grassroots guide to engineering (an...
[OReilly Superstream] Occupy the Space: A grassroots guide to engineering (an...
 
GraphRAG for LifeSciences Hands-On with the Clinical Knowledge Graph
GraphRAG for LifeSciences Hands-On with the Clinical Knowledge GraphGraphRAG for LifeSciences Hands-On with the Clinical Knowledge Graph
GraphRAG for LifeSciences Hands-On with the Clinical Knowledge Graph
 
Connector Corner: Seamlessly power UiPath Apps, GenAI with prebuilt connectors
Connector Corner: Seamlessly power UiPath Apps, GenAI with prebuilt connectorsConnector Corner: Seamlessly power UiPath Apps, GenAI with prebuilt connectors
Connector Corner: Seamlessly power UiPath Apps, GenAI with prebuilt connectors
 
Session 1 - Intro to Robotic Process Automation.pdf
Session 1 - Intro to Robotic Process Automation.pdfSession 1 - Intro to Robotic Process Automation.pdf
Session 1 - Intro to Robotic Process Automation.pdf
 
zkStudyClub - LatticeFold: A Lattice-based Folding Scheme and its Application...
zkStudyClub - LatticeFold: A Lattice-based Folding Scheme and its Application...zkStudyClub - LatticeFold: A Lattice-based Folding Scheme and its Application...
zkStudyClub - LatticeFold: A Lattice-based Folding Scheme and its Application...
 
Apps Break Data
Apps Break DataApps Break Data
Apps Break Data
 
inQuba Webinar Mastering Customer Journey Management with Dr Graham Hill
inQuba Webinar Mastering Customer Journey Management with Dr Graham HillinQuba Webinar Mastering Customer Journey Management with Dr Graham Hill
inQuba Webinar Mastering Customer Journey Management with Dr Graham Hill
 
Main news related to the CCS TSI 2023 (2023/1695)
Main news related to the CCS TSI 2023 (2023/1695)Main news related to the CCS TSI 2023 (2023/1695)
Main news related to the CCS TSI 2023 (2023/1695)
 
Skybuffer SAM4U tool for SAP license adoption
Skybuffer SAM4U tool for SAP license adoptionSkybuffer SAM4U tool for SAP license adoption
Skybuffer SAM4U tool for SAP license adoption
 
Leveraging the Graph for Clinical Trials and Standards
Leveraging the Graph for Clinical Trials and StandardsLeveraging the Graph for Clinical Trials and Standards
Leveraging the Graph for Clinical Trials and Standards
 
Must Know Postgres Extension for DBA and Developer during Migration
Must Know Postgres Extension for DBA and Developer during MigrationMust Know Postgres Extension for DBA and Developer during Migration
Must Know Postgres Extension for DBA and Developer during Migration
 

Multisided Exposure Fairness for Search and Recommendation

  • 1. Joint Multisided Exposure Fairness for Search and Recommendation Bhaskar Mitra Microsoft Research, Canada bmitra@microsoft.com Pre-print: https://arxiv.org/pdf/2205.00048.pdf (Paper accepted @ SIGIR’22) Joint work with Haolun Wu, Chen Ma, Fernando Diaz, and Xue Liu
  • 2. Digital information access and exposure Traditional IR is concerned with ranking of items according to relevance These information access systems deployed at web-scale mediate what information gets exposure The exposure-framing of IR raises several fairness concerns, new opportunities for ranking optimization, and can be relevant to other FATE considerations (e.g., privacy and transparency)
  • 3. Sweeney. Discrimination in online ad delivery. Commun. ACM. (2013) Crawford. The Trouble with Bias. NeurIPS. (2017) Singh and Joachims. Fairness of Exposure in Rankings. In KDD, ACM. (2018) Harms of disparate exposure Several past studies have pointed out representational and allocative harms from disparate exposure Concerns of fairness in the context of IR/ML systems are inherently interdisciplinary and sociotechnical; and these concerns span beyond just questions of system design The role of IR/ML in this process is to deconstruct their own measures and models in ways that allows a broad range of researchers and stakeholders to critically analyze and shape these technologies In traditional IR, we have made progress in modeling, measuring, and optimizing for individual user satisfaction; a key challenge ahead is to model, measure, and optimize IR systems with respect to impact on populations of users and consider disparate impact across subpopulations
  • 4. Exposure fairness is a multisided problem It is important to ask not just whether specific content receives exposure, but who it is exposed to and in what context Haolun, Mitra, Ma, and Liu. Joint Multisided Exposure Fairness for Recommendation. Under review for SIGIR, ACM. (2022)
  • 5. Exposure fairness is a multisided problem Take the example of a job recommendation system Group-of-users-to-group-of-items fairness (GG-F) Are groups of items under/over-exposed to groups of users? E.g., men being disproportionately recommended high-paying jobs and women low-paying jobs. Individual-user-to-Individual-item fairness (II-F) Are Individual items under/over-exposed to Individual users? Individual-user-to-group-of-items fairness (IG-F) Are groups of items under/over-exposed to individual users? E.g., a specific user being disproportionately recommended low-paying jobs. Group-of-users-to-Individual-item fairness (GI-F) Are Individual items under/over-exposed to groups of users? E.g., a specific job being disproportionately recommended to men and not to women and non-binary people. All-users-to-Individual-item fairness (AI-F) Are Individual items under/over-exposed to all users overall? E.g., a specific job being disproportionately under- exposed to all users. All-users-to-group-of-items fairness (AG-F) Are groups of items under/over-exposed to all users overall? E.g., jobs at Black-owned businesses being disproportionately under-exposed to all users.
  • 6. Exposure fairness is a multisided problem Take the example of a job recommendation system Group-of-users-to-group-of-items fairness (GG-F) Are groups of items under/over-exposed to groups of users? E.g., men being disproportionately recommended high-paying jobs and women low-paying jobs. Individual-user-to-Individual-item fairness (II-F) Are Individual items under/over-exposed to Individual users? Individual-user-to-group-of-items fairness (IG-F) Are groups of items under/over-exposed to individual users? E.g., a specific user being disproportionately recommended low-paying jobs. Group-of-users-to-Individual-item fairness (GI-F) Are Individual items under/over-exposed to groups of users? E.g., a specific job being disproportionately recommended to men and not to women and non-binary people. All-users-to-Individual-item fairness (AI-F) Are Individual items under/over-exposed to all users overall? E.g., a specific job being disproportionately under- exposed to all users. All-users-to-group-of-items fairness (AG-F) Are groups of items under/over-exposed to all users overall? E.g., jobs at Black-owned businesses being disproportionately under-exposed to all users.
  • 7. Exposure fairness is a multisided problem Take the example of a job recommendation system Group-of-users-to-group-of-items fairness (GG-F) Are groups of items under/over-exposed to groups of users? E.g., men being disproportionately recommended high-paying jobs and women low-paying jobs. Individual-user-to-Individual-item fairness (II-F) Are Individual items under/over-exposed to Individual users? Individual-user-to-group-of-items fairness (IG-F) Are groups of items under/over-exposed to individual users? E.g., a specific user being disproportionately recommended low-paying jobs. Group-of-users-to-Individual-item fairness (GI-F) Are Individual items under/over-exposed to groups of users? E.g., a specific job being disproportionately recommended to men and not to women and non-binary people. All-users-to-Individual-item fairness (AI-F) Are Individual items under/over-exposed to all users overall? E.g., a specific job being disproportionately under- exposed to all users. All-users-to-group-of-items fairness (AG-F) Are groups of items under/over-exposed to all users overall? E.g., jobs at Black-owned businesses being disproportionately under-exposed to all users.
  • 8. Exposure fairness is a multisided problem Take the example of a job recommendation system Group-of-users-to-group-of-items fairness (GG-F) Are groups of items under/over-exposed to groups of users? E.g., men being disproportionately recommended high-paying jobs and women low-paying jobs. Individual-user-to-Individual-item fairness (II-F) Are Individual items under/over-exposed to Individual users? Individual-user-to-group-of-items fairness (IG-F) Are groups of items under/over-exposed to individual users? E.g., a specific user being disproportionately recommended low-paying jobs. Group-of-users-to-Individual-item fairness (GI-F) Are Individual items under/over-exposed to groups of users? E.g., a specific job being disproportionately recommended to men and not to women and non-binary people. All-users-to-Individual-item fairness (AI-F) Are Individual items under/over-exposed to all users overall? E.g., a specific job being disproportionately under- exposed to all users. All-users-to-group-of-items fairness (AG-F) Are groups of items under/over-exposed to all users overall? E.g., jobs at Black-owned businesses being disproportionately under-exposed to all users.
  • 9. Exposure fairness is a multisided problem Take the example of a job recommendation system Group-of-users-to-group-of-items fairness (GG-F) Are groups of items under/over-exposed to groups of users? E.g., men being disproportionately recommended high-paying jobs and women low-paying jobs. Individual-user-to-Individual-item fairness (II-F) Are Individual items under/over-exposed to Individual users? Individual-user-to-group-of-items fairness (IG-F) Are groups of items under/over-exposed to individual users? E.g., a specific user being disproportionately recommended low-paying jobs. Group-of-users-to-Individual-item fairness (GI-F) Are Individual items under/over-exposed to groups of users? E.g., a specific job being disproportionately recommended to men and not to women and non-binary people. All-users-to-Individual-item fairness (AI-F) Are Individual items under/over-exposed to all users overall? E.g., a specific job being disproportionately under- exposed to all users. All-users-to-group-of-items fairness (AG-F) Are groups of items under/over-exposed to all users overall? E.g., jobs at Black-owned businesses being disproportionately under-exposed to all users.
  • 10. Exposure fairness is a multisided problem Take the example of a job recommendation system Group-of-users-to-group-of-items fairness (GG-F) Are groups of items under/over-exposed to groups of users? E.g., men being disproportionately recommended high-paying jobs and women low-paying jobs. Individual-user-to-Individual-item fairness (II-F) Are Individual items under/over-exposed to Individual users? Individual-user-to-group-of-items fairness (IG-F) Are groups of items under/over-exposed to individual users? E.g., a specific user being disproportionately recommended low-paying jobs. Group-of-users-to-Individual-item fairness (GI-F) Are Individual items under/over-exposed to groups of users? E.g., a specific job being disproportionately recommended to men and not to women and non-binary people. All-users-to-Individual-item fairness (AI-F) Are Individual items under/over-exposed to all users overall? E.g., a specific job being disproportionately under- exposed to all users. All-users-to-group-of-items fairness (AG-F) Are groups of items under/over-exposed to all users overall? E.g., jobs at Black-owned businesses being disproportionately under-exposed to all users.
  • 11. Exposure fairness is a multisided problem Take the example of a job recommendation system Group-of-users-to-group-of-items fairness (GG-F) Are groups of items under/over-exposed to groups of users? E.g., men being disproportionately recommended high-paying jobs and women low-paying jobs. Individual-user-to-Individual-item fairness (II-F) Are Individual items under/over-exposed to Individual users? Individual-user-to-group-of-items fairness (IG-F) Are groups of items under/over-exposed to individual users? E.g., a specific user being disproportionately recommended low-paying jobs. Group-of-users-to-Individual-item fairness (GI-F) Are Individual items under/over-exposed to groups of users? E.g., a specific job being disproportionately recommended to men and not to women and non-binary people. All-users-to-Individual-item fairness (AI-F) Are Individual items under/over-exposed to all users overall? E.g., a specific job being disproportionately under- exposed to all users. All-users-to-group-of-items fairness (AG-F) Are groups of items under/over-exposed to all users overall? E.g., jobs at Black-owned businesses being disproportionately under-exposed to all users.
  • 12. User browsing models and exposure User browsing models are simplified models of how users inspect and interact with retrieved results It estimates the probability that the user inspects a particular item in a ranked list of items—i.e., the item is exposed to the user In IR, user models have been implicitly and explicitly employed in metric definitions and for estimating relevance from historical logs of user behavior data For example, let’s consider the RBP user model… NDCG RBP Probability of exposure at different ranks according to NDCG and RBP user browsing models exposure event an item a ranked list of items rank of the item in the ranked list patience factor
  • 13. Stochastic ranking and expected exposure In recommendation, Diaz et al. (2020) define a stochastic ranking policy 𝜋𝑢, conditioned on user 𝑢 ∈ U, as a probability distribution over all permutations of items in the collection The expected exposure of an item 𝑑 for user 𝑢 can then be computed as follows: Here, 𝑝(𝜖|𝑑,𝜎) can be computed using a user browsing model like RBP as discussed previously Note: The above formulation can also be applied to search by replacing user with query Diaz, Mitra, Ekstrand, Biega, and Carterette. Evaluating stochastic rankings with expected exposure. In CIKM, ACM. (2020)
  • 14. System, target, and random exposure System exposure. The user-item expected exposure distribution corresponding to a stochastic ranking policy 𝜋. Correspondingly, we can define a |U|×|D| matrix E, such that E𝑖𝑗 = 𝑝(𝜖|D𝑗 ,𝜋U𝑖 ). Target exposure. The user-item expected exposure distribution corresponding to an ideal stochastic ranking policy 𝜋*, as defined by some desirable principle (e.g., the equal expected exposure principle). We denote the corresponding expected exposure matrix as E*. Random exposure. The user-item expected exposure distribution corresponding to a stochastic ranking policy 𝜋~ that samples rankings from a uniform distribution over all item permutations. We denote the corresponding expected exposure matrix as E~. The deviation of E from E* gives us a quantitative measure of the suboptimality of the retrieval system under consideration.
  • 15. Joint multisided exposure (JME) fairness metrics Haolun, Mitra, Ma, and Liu. Joint Multisided Exposure Fairness for Recommendation. Under review for SIGIR, ACM. (2022)
  • 16. Toy example Let, there be 4 candidates (𝑢𝑎1 , 𝑢𝑎2 , 𝑢𝑏1 , 𝑢𝑏2 ) and 4 jobs (𝑑𝑥1 , 𝑑𝑥2 , 𝑑𝑦1 , 𝑑𝑦2 ) All 4 jobs are relevant to each of the 4 candidates The candidates belong to 2 groups 𝑎 (𝑢𝑎1 , 𝑢𝑎2 ) and 𝑏 (𝑢𝑏1 , 𝑢𝑏2 )—e.g., based on gender—and similarly the jobs belong to 2 groups 𝑥 (𝑑𝑥1 , 𝑑𝑥2 ) and 𝑦 (𝑑𝑦1 , 𝑑𝑦2 )—say based on whether they pay high or low salaries Let’s assume that the recommender system displays only one result at a time and our simple user model assumes that the user always inspects the displayed result—i.e., the probability of exposure is 1 for the displayed item and 0 for all other items for a given impression In this setting, an ideal recommender should expose each of the four jobs to each candidate with a probability of 0.25
  • 17. Toy example Let, there be 4 candidates (𝑢𝑎1 , 𝑢𝑎2 , 𝑢𝑏1 , 𝑢𝑏2 ) and 4 jobs (𝑑𝑥1 , 𝑑𝑥2 , 𝑑𝑦1 , 𝑑𝑦2 ) All 4 jobs are relevant to each of the 4 candidates The candidates belong to 2 groups 𝑎 (𝑢𝑎1 , 𝑢𝑎2 ) and 𝑏 (𝑢𝑏1 , 𝑢𝑏2 )—e.g., based on gender—and similarly the jobs belong to 2 groups 𝑥 (𝑑𝑥1 , 𝑑𝑥2 ) and 𝑦 (𝑑𝑦1 , 𝑑𝑦2 )—say based on whether they pay high or low salaries Let’s assume that the recommender system displays only one result at a time and our simple user model assumes that the user always inspects the displayed result—i.e., the probability of exposure is 1 for the displayed item and 0 for all other items for a given impression In this setting, an ideal recommender should expose each of the four jobs to each candidate with a probability of 0.25 All of them are equally II-Unfair
  • 18. Let, there be 4 candidates (𝑢𝑎1 , 𝑢𝑎2 , 𝑢𝑏1 , 𝑢𝑏2 ) and 4 jobs (𝑑𝑥1 , 𝑑𝑥2 , 𝑑𝑦1 , 𝑑𝑦2 ) All 4 jobs are relevant to each of the 4 candidates The candidates belong to 2 groups 𝑎 (𝑢𝑎1 , 𝑢𝑎2 ) and 𝑏 (𝑢𝑏1 , 𝑢𝑏2 )—e.g., based on gender—and similarly the jobs belong to 2 groups 𝑥 (𝑑𝑥1 , 𝑑𝑥2 ) and 𝑦 (𝑑𝑦1 , 𝑑𝑦2 )—say based on whether they pay high or low salaries Let’s assume that the recommender system displays only one result at a time and our simple user model assumes that the user always inspects the displayed result—i.e., the probability of exposure is 1 for the displayed item and 0 for all other items for a given impression In this setting, an ideal recommender should expose each of the four jobs to each candidate with a probability of 0.25 Only (b), (e), and (f) are IG-Unfair Toy example
  • 19. Toy example Let, there be 4 candidates (𝑢𝑎1 , 𝑢𝑎2 , 𝑢𝑏1 , 𝑢𝑏2 ) and 4 jobs (𝑑𝑥1 , 𝑑𝑥2 , 𝑑𝑦1 , 𝑑𝑦2 ) All 4 jobs are relevant to each of the 4 candidates The candidates belong to 2 groups 𝑎 (𝑢𝑎1 , 𝑢𝑎2 ) and 𝑏 (𝑢𝑏1 , 𝑢𝑏2 )—e.g., based on gender—and similarly the jobs belong to 2 groups 𝑥 (𝑑𝑥1 , 𝑑𝑥2 ) and 𝑦 (𝑑𝑦1 , 𝑑𝑦2 )—say based on whether they pay high or low salaries Let’s assume that the recommender system displays only one result at a time and our simple user model assumes that the user always inspects the displayed result—i.e., the probability of exposure is 1 for the displayed item and 0 for all other items for a given impression In this setting, an ideal recommender should expose each of the four jobs to each candidate with a probability of 0.25 Only (c), (d), (e), and (f) are GI-Unfair
  • 20. Toy example Let, there be 4 candidates (𝑢𝑎1 , 𝑢𝑎2 , 𝑢𝑏1 , 𝑢𝑏2 ) and 4 jobs (𝑑𝑥1 , 𝑑𝑥2 , 𝑑𝑦1 , 𝑑𝑦2 ) All 4 jobs are relevant to each of the 4 candidates The candidates belong to 2 groups 𝑎 (𝑢𝑎1 , 𝑢𝑎2 ) and 𝑏 (𝑢𝑏1 , 𝑢𝑏2 )—e.g., based on gender—and similarly the jobs belong to 2 groups 𝑥 (𝑑𝑥1 , 𝑑𝑥2 ) and 𝑦 (𝑑𝑦1 , 𝑑𝑦2 )—say based on whether they pay high or low salaries Let’s assume that the recommender system displays only one result at a time and our simple user model assumes that the user always inspects the displayed result—i.e., the probability of exposure is 1 for the displayed item and 0 for all other items for a given impression In this setting, an ideal recommender should expose each of the four jobs to each candidate with a probability of 0.25 Only (e) and (f) are GG-Unfair
  • 21. Let, there be 4 candidates (𝑢𝑎1 , 𝑢𝑎2 , 𝑢𝑏1 , 𝑢𝑏2 ) and 4 jobs (𝑑𝑥1 , 𝑑𝑥2 , 𝑑𝑦1 , 𝑑𝑦2 ) All 4 jobs are relevant to each of the 4 candidates The candidates belong to 2 groups 𝑎 (𝑢𝑎1 , 𝑢𝑎2 ) and 𝑏 (𝑢𝑏1 , 𝑢𝑏2 )—e.g., based on gender—and similarly the jobs belong to 2 groups 𝑥 (𝑑𝑥1 , 𝑑𝑥2 ) and 𝑦 (𝑑𝑦1 , 𝑑𝑦2 )—say based on whether they pay high or low salaries Let’s assume that the recommender system displays only one result at a time and our simple user model assumes that the user always inspects the displayed result—i.e., the probability of exposure is 1 for the displayed item and 0 for all other items for a given impression In this setting, an ideal recommender should expose each of the four jobs to each candidate with a probability of 0.25 Only (d) and (f) are AI-Unfair Toy example
  • 22. Toy example Let, there be 4 candidates (𝑢𝑎1 , 𝑢𝑎2 , 𝑢𝑏1 , 𝑢𝑏2 ) and 4 jobs (𝑑𝑥1 , 𝑑𝑥2 , 𝑑𝑦1 , 𝑑𝑦2 ) All 4 jobs are relevant to each of the 4 candidates The candidates belong to 2 groups 𝑎 (𝑢𝑎1 , 𝑢𝑎2 ) and 𝑏 (𝑢𝑏1 , 𝑢𝑏2 )—e.g., based on gender—and similarly the jobs belong to 2 groups 𝑥 (𝑑𝑥1 , 𝑑𝑥2 ) and 𝑦 (𝑑𝑦1 , 𝑑𝑦2 )—say based on whether they pay high or low salaries Let’s assume that the recommender system displays only one result at a time and our simple user model assumes that the user always inspects the displayed result—i.e., the probability of exposure is 1 for the displayed item and 0 for all other items for a given impression In this setting, an ideal recommender should expose each of the four jobs to each candidate with a probability of 0.25 Only (f) is AG-Unfair
  • 23. Relationship between different JME metrics Based on the metric definitions, we can show that a system that is II-Fair (i.e., II- F=0) will also be fair along the other five JME-fairness dimensions Similarly, IG-Fair and GI-Fair independently implies GG-Fair, and GG-Fair and AI-Fair implies AG-Fair Finally, all the other metrics can be viewed as specific instances of GG-F, with different (extreme) definitions of groups on user and item side II-F=0 IG-F=0 GI-F=0 GG-F=0 AI-F=0 AG-F=0
  • 24. Disparity and relevance Each of our proposed JME-fairness metrics can be decomposed into a disparity and a relevance component, such that increasing randomness in the model would decrease disparity (good!) but also decrease relevance (bad!)
  • 25. Disparity and relevance Each of our proposed JME-fairness metrics can be decomposed into a disparity and a relevance component, such that increasing randomness in the model would decrease disparity (good!) but also decrease relevance (bad!)
  • 26. Disparity and relevance Each of our proposed JME-fairness metrics can be decomposed into a disparity and a relevance component, such that increasing randomness in the model would decrease disparity (good!) but also decrease relevance (bad!) Different models have different disparity-relevance trade-off for each of the different JME-fairness metrics
  • 27. How correlated are different JME-fairness dimensions? Recall that all six JME-Fairness metrics can be seen as specific instances of GG-F For this analysis using MovieLens we had 2 groups by gender* and 7 groups by age on the user side and 18 genres on the item side When we have small number of large groups “Individual” and “Group” analysis will diverge, and vice versa * The gender attribute is available in the MovieLens dataset as a binary annotation. We recognize that this does not reflect the full spectrum of gender identities, and this is a short-coming of our work.
  • 28. New metrics, new optimization opportunity! How can we optimize ranking models for target exposure? Diaz, Mitra, Ekstrand, Biega, and Carterette. Evaluating stochastic rankings with expected exposure. In CIKM, ACM. (2020)
  • 29. Stochastic ranking A stochastic ranking model samples a ranking from a probability distribution over all possible permutations of items in the collection—i.e., for the same intent it returns a slightly different ranking on each impression Given a static ranking policy, we can generate a stochastic equivalent using Plackett-Luce sampling—for example, given items 𝑑1, 𝑑2, 𝑑3, 𝑑4 the probability of sampling a particular ranking 𝑑2, 𝑑1, 𝑑4, 𝑑3 is: 𝜋: a ranking, 𝜙: a transformation, e.g., exponential over score 𝑠𝑖 for document 𝑑𝑖 Equivalent to sequentially sampling documents without replacement with probability 𝜙 𝑠𝑖 restaurants in montreal restaurants in montreal restaurants in montreal restaurants in montreal Luce. Individual Choice Behavior. (1959) Plackett. The Analysis of Permutations. Journal of the Royal Statistical Society: Series C (Applied Statistics). (1975)
  • 30. Gradient-based optimization for target exposure Approach 1. Use the target model to score the items 2. Compute PL sampling probability as a function of the item scores 3. Sample multiple rankings 4. Compute expected system exposure across sampled rankings 5. Compute the loss as a difference between system and target exposure 6. Backpropagate! Challenges and solutions The key challenge is the proposed approach is that both the sampling and the ranking steps are non-differentiable! For sampling, we can use Gumbel sampling as a differentiable approximation For ranking, we can employ SmoothRank / ApproxRank as differentiable approximations of the ranking step Wu, Chang, Zheng, and Zha. Smoothing DCG for learning to rank: A novel approach using smoothed hinge functions. In Proc. CIKM, ACM. (2009) Qin, Liu, and Li. A general approximation framework for direct optimization of information retrieval measures. Information retrieval. (2010) Bruch, Han, Bendersky, and Najork. A stochastic treatment of learning to rank scoring functions. In Proc. WSDM, ACM. (2020) ,
  • 31. Gradient-based optimization for target exposure add independently sampled Gumbel noise neural scoring function compute smooth rank value compute exposure using user model compute loss with target exposure compute average exposure items target exposure Diaz, Mitra, Ekstrand, Biega, and Carterette. Evaluating stochastic rankings with expected exposure. In CIKM, ACM. (2020)
  • 32. Trading-off different JME-fairness metrics We can simultaneously optimize for multiple exposure metrics by combining them linearly For example, Preliminary experiments indicate that we can significantly minimize GG-F with minimal degradation to II-F and relevance
  • 33. Discussion True vs. observed relevance labels. The computation of target exposure itself raises fairness questions. E.g., the equal expected exposure principle assumes we have access to true relevance labels, but in fact the observed labels reflect huge historical social biases. E.g., In the job recommendation scenario, it may be more appropriate to define GG-F target exposure for high and low paying jobs to be uniform across user groups, irrespective of historical disparities reflected in the data. Choice of group attributes. The choice of group attributes necessitates reflecting on historical and socioeconomic contexts. We note that our formulation can also be extended to handling multiple group attributes on each side. However, that raises questions of intersectional fairness that we haven’t yet studied in our work. Beyond two-sided exposure fairness. While we have primarily focused on two-sided exposure fairness so far, we envision that extending that to additional stakeholder may also be important. E.g., in product search exposure fairness may concern with being fair to consumers, manufacturers, and retailers. Incorporating model uncertainty. The stochastic ranking policies we have considered so far involves randomizing a static policy with model-independent sampling of noise. In contrast, the stochasticity could also be informed by the model’s own uncertainty in its prediction. This is an area for potential future work.