SlideShare a Scribd company logo
1 of 7
Download to read offline
2010 CRC PhD Student Conference




 Supporting multimodal media recommendation
  and annotation using social network analysis
                                    Adam Rae


                                  a.rae@open.ac.uk


     Supervisors            Stefan RĆ¼ger, Suzanne Little, Roelof van Zwol
     Department             The Knowledge Media Institute
     Status                 Full Time
     Probation Viva         After
     Starting Date          October 2007


Research Hypothesis
      By analysing and extracting information from the social graphs de-
      scribed by both explicit and implicit user interactions, like those
      found in online media sharing systems like Flickr1 , it is possible
      to augment existing non-social aware recommender systems and
      thereby signiļ¬cantly improve their performance.

Large scale web based systems for sharing media continue to tackle the problem
of helping their users ļ¬nd what they are looking for in a timely manner. To do
this, lots of good quality metadata is required to sift through the data collection
to pick out exactly those documents that match the information need of the
user. In the case of ļ¬nding images from the online photo sharing website Flickr,
this could be from over 4 billion examples. How can we help both the system
and the user in enriching the metadata of the media within the collection in
order to improve the experience for the user and to reduce the burden on the
underlying data handling system? Can modelling users, by themselves and
within the context of the wider online community help? Can this modeling be
used to improve recommender systems that improve the experience and reduce
cognitive burden on users?
    Existing approaches tend to treat multimedia in the same way they have
dealt with text documents in the past, speciļ¬cally by treating the textual meta-
data associated with an image as a text document, but this ignores the inherently
diļ¬€erent nature of the data the system is handling. Images are visual data, and
while they can be described well by textual metadata, they cannot be described
completely by it. Also, the user cannot be ignored in the retrieval process, and
learning more about a user provides information to the system to tailor results to
their speciļ¬c requirements. Users interact online, and these interactions form a
  1 http://www.ļ¬‚ickr.com/




                                     Page 91 of 125
2010 CRC PhD Student Conference




new type of data that has yet to be fully explored nor exploited when modelling
users.
    The work presented here combines the mining of social graphs that occur in
Flickr with visual content and metadata analysis to provide better personalised
photo recommender mechanisms and the following experiment and its analysis
are a major component in my overall thesis.

Interaction Scenario
In order to address this research question, multiple experiments have been car-
ried out, one of which I present here:
       Envisage an incoming stream of photos made available to a user. In
       systems of a scale similar to Flickr, this could be thousands of im-
       ages per second. Can a system that uses cues from the social, visual
       and semantic aspects of these images perform better than one that
       uses the more traditional approach of using only semantic informa-
       tion, according to speciļ¬cally deļ¬ned objective metrics? How does
       performance vary between users?
An experiment was carried out that mines data from the social communities in
Flickr, from the visual content of images and from the text based metadata and
uses a machines learning mechanism to merge these signals together to form a
classiļ¬er that, given a candidate image and prospective viewing user, decides
whether the user would label that image as a ā€˜Favouriteā€™2 - see Figure 1.


Related Work
The signiļ¬cant inļ¬‚uence that our peers can have on our behaviour online has
been studied by researchers such as Lerman and Jones[3], and the particular
interaction that occurs between users and visual media in particular in the
work of Nov et al.[4]and Kern et al[2]. Their insights into the importance of
understanding more about a user in order to best fulļ¬l their information need
supports the hypothesis that this kind of information can be usefully exploited
to improve systems that try to match that need to a data set supported by social
interaction. Here I extend their ideas by incorporating this valuable social data
into a complementary multimodal framework that takes advantage of multiple
types of data.
    The use of social interaction features in the work of Sigurbjƶrnsson and van
Zwol[7] and Garg and Weber[1] inspired my more comprehensive feature set
and its analysis. Their ļ¬ndings that aggregating data generated from online
communities is valuable when suggesting tags is important and I believe also
transfers to recommendation in general as well as to the speciļ¬c task of recom-
mending images. In fact, I demonstrated this in previous work on social media
tag suggestion[6].
    I use some of the human perception based visual features outlined in the
work of San Pedro and Siersdorfer[5], as these have been shown to work well
in similar experimental scenarios and cover a range of visual classes. I extend
them further with a selection of other high performing visual features.
  2A   binary label Flickr users can use to annotate an image they like.




                                        Page 92 of 125
2010 CRC PhD Student Conference




                                         Incoming stream of previously unseen candidate images




                                                     Textual         Social        Visual
                               User information                                                    User information


                                                           Feature Extraction



                   User A                                                                                             User B
           Has tagged beaches before                                                                        Member of urban animals group
                                                               Trained Classiļ¬er




                                                                                            Potential Favourite Images
                        Potential Favourite Images                                                  for User B
                                for User A




 Figure 1: Diagram of the image classiļ¬cation system used with Flickr data.


Experimental Work
400 users of varying levels of social activity were selected from Flickr and their
ā€˜Favouriteā€™ labelled images collected. This resulted in a collection of hundreds
of thousands of images. To train my classiļ¬er, these images were treated as
positive examples of relevant images. I generated a variety of negative example
sets to reļ¬‚ect realistic system scenarios. For all photo examples we extracted
visual and semantic features, and social features that described the user, the
owner of the photo, any connection between them as well as other behaviour
metrics. We then tested our classiļ¬er using previously unseen examples and
measured the performance of the system with a particular emphasis on the
information retrieval metric of precision at 5 and 10 to reļ¬‚ect our envisaged use
case scenario.


Results
An extract of the results from the experiment are shown in Table 1. They can
be summarised thus:
   ā€¢ It is possible to achieve high levels of precision in selecting our positive
     examples, especially by using social features. This performance increase
     is statistically signiļ¬cantly higher than the baseline Textual run. These
     social signals evidently play a signiļ¬cant rĆ“le when a user labels an image
     a ā€˜Favouriteā€™ and can be usefully exploited to help them.
   ā€¢ The value of individual types of features is complex, but complementary.
     The combined systems tend to perform better than the individual ones.
   ā€¢ It is far easier to classiļ¬er photos that are not ā€˜Favouritesā€™ than those that
     are, as shown by the high negative values. This can be used to narrow
     down the search space for relevant images by removing those that are
     obviously not going to interest the user, thus reduing load on both the
     user and the system.




                                                          Page 93 of 125
2010 CRC PhD Student Conference




      System            Accuracy      + Prec.      + Rec.     - Prec.    - Rec.
     Textual              0.87         0.48         0.18        0.88       0.97
      Visual              0.88         1.00         0.09        0.88       1.00
      Social              0.92         0.80         0.56        0.94       0.98
 Textual+Visual           0.88         0.62         0.27        0.90       0.97
 Textual+Social           0.92         0.77         0.62        0.94       0.97
  Visual+Social           0.93         0.89         0.56        0.94       0.99
 Text+Vis.+Soc.           0.93         0.84         0.62        0.94       0.98

Table 1: Accuracy, precison and recall for various combinations of features using
the experiments most realistic scenario data set. Photos labelled as ā€˜Favouritesā€™
are positive examples, and those that are not are negative examples. Higher
numbers are better.


   ā€¢ As is typical in this style of information retrieval experiment, we can trade-
     oļ¬€ between precision and recall depending on our requirements. As we are
     interested in high precision in this particular experiment, we see that the
     combination of the Visual+Social and Text+Visual+Social runs
     give good precision without sacriļ¬cing too much recall.



References
[1] Nikhil Garg and Ingmar Weber. Personalized, interactive tag recommenda-
    tion for ļ¬‚ickr. In Proceedings of the 2008 ACM Conference on Recommender
    Systems, pages 67ā€“74, Lausanne, Switzerland, October 2008. ACM.
[2] R. Kern, M. Granitzer, and V. Pammer. Extending folksonomies for image
    tagging. In Workshop on Image Analysis for Multimedia Interactive Services,
    2008, pages 126ā€“129, May 2008.
[3] Kristina Lerman and Laurie Jones. Social browsing on ļ¬‚ickr. In Proceedings
    of ICWSM, December 2007.
[4] Oded Nov, Mor Naaman, and Chen Ye. What drives content tagging: the
    case of photos on ļ¬‚ickr. In Proceeding of the twenty-sixth annual SIGCHI
    conference on Human factors in computing systems, pages 1097ā€“1100, Flo-
    rence, Italy, 2008. ACM.
[5] Jose San Pedro and Stefan Siersdorfer. Ranking and classifying attractive-
    ness of photos in folksonomies. In WWW, Madrid, Spain, April 2009.

[6] Adam Rae, Roelof van Zwol, and Bƶrkur Sigurbjƶrnsson. Improving tag
    recommendation using social networks. In 9th International conference on
    Adaptivity, Personalization and Fusion of Heterogeneous Information, April
    2010.
[7] Roelof van Zwol. Flickr: Who is looking? In IEEE/WIC/ACM Inter-
    national Conference on Web Intelligence, pages 184ā€“190, Washington, DC,
    USA, 2007. IEEE Computer Society.




                                  Page 94 of 125
2010 CRC PhD Student Conference



     The effect of Feedback on the Motivation of Software
                          Engineers

                                      Rien Sach
                                r.j.sach@open.ac.uk

Supervisors          Helen Sharp
                     Marian Petre
Department/Institute Computing
Status               Fulltime
Probation viva       After
Starting date        October 2009

Motivation is reported as having an effect on crucial aspects of software engineering
such as productivity (Procaccino and Verner 2005), software quality (Boehm 1981),
and a projectā€™s overall success (Frangos 1997). Feedback is a key factor in the most
commonly used theory in reports published on the motivation of software engineers
(Hall et al. 2009), and it is important that we gain a greater understanding of the effect
it has on the motivation of software engineers.

My research is grounded in the question ā€œWhat are the effects of feedback on the
motivation of software engineers?ā€, and focuses on feedback conveyed in human
interactions. I believe that before I can focus my question further I will need to begin
some preliminary work to identify how feedback occurs, what types of feedback
occur, and the possible impact of this feedback.

Motivation can be understood in different ways. For example, as a manager you might
consider motivation as something you must maintain in your employees to ensure
they complete work for you as quickly as possible. As an employee you might
consider motivation as the drive that keeps you focused on a task, or it might simply
be what pushes you to get up in the morning and go to work.

Herzberg (1987) describes motivation as ā€œa function of growth from getting intrinsic
rewards out of interesting and challenging workā€. Thatā€™s quite a nice definition; and
according to Herzberg motivation is intrinsic to oneā€™s self. Ryan and Deci (2000)
describe intrinsic motivation as ā€œthe doing of activity for its inherent satisfaction
rather than for some separable consequenceā€ (Page 60).

Herzberg (1987) defines extrinsic factors as movement and distinguishes it from
motivation, stating that ā€œMovement is a function of fear of punishment or failure to
get extrinsic rewardsā€. Ryan and Deci (2000) state that ā€œExtrinsic motivation is a
construct that pertains whenever an activity is done in order to attain some separable
outcomeā€.

There are 8 core motivational theories (Hall et al. 2009) and some of the theories
focus on motivation as a ā€œa sequence or process of related activitiesā€ (Hall et al. 2009)
called process theories, while others focus on motivation ā€œat a single point in timeā€
(Couger and Zawacki 1980) called content theories.


                                        Page 95 of 125
2010 CRC PhD Student Conference




As reported in a systematic literature review conducted by Beecham et al (2007), and
their published review of the use of theory inside this review in 2009 (Hall et al 2009),
the three most popular theories used in studies of motivation in Software Engineering
were Hackman and Oldmanā€™s Job Characteristics Theory (68%), Herzbergā€™s
Motivational Hygiene Theory (41%), and Maslowā€™s Theory of Needs (21%)1.

Hackman and Oldmanā€™s Job Characteristics Theory focuses on the physical job, and
suggests five characteristics (skill variety, task identity, task significance, autonomy,
and feedback) that lead to three psychological states which in turn lead to higher
internal motivation and higher quality work. Herzbergā€™s Hygiene Theory suggests that
the only true motivation is intrinsic motivation, and this leads to job satisfaction,
where extrinsic factors are only useful in avoiding job dissatisfaction.

One of the five key job characteristics in Hackman and Oldmanā€™s theory is feedback.
Feedback is not explicitly mentioned in Herzbergā€™s Motivational Hygiene Theory, but
he notes that it is a part of job enrichment, which he states is ā€œkey to designing work
that motivates employeesā€ (Herzberg 1987). However this is a managerial view point.

Software Engineers are considered to be current practitioners working on active
software projects within the industry. This includes programmers, analysts, testers,
and designers who actively work and produce software for real projects in the real
world.

From a management perspective, gaining a greater understanding of what motives
employees could prove invaluable in increasing productivity and software quality, and
from an individual perspective the prospect of being given feedback that motivates
you and makes your job more enjoyable and improves the quality of your work
experience could lead to a more successful and enjoyable work life.

My proposed research is divided into stages. In the first stage I plan to conduct
interviews and diary studies to identify the types of feedback in software engineering
and how feedback is experienced by software engineers. I then plan to conduct
additional studies to identify what impact this feedback has on software engineers and
how that impact is evident. Finally, I plan to observe software engineers at work to
see feedback in context, and to compare those observations to the information
gathered during the first two stages.

At the end of my PhD I hope to accomplish research that leads to a greater
understanding of what feedback is inside software engineering and how it is given or
received. Subsequently I wish to gain an understanding of how this feedback alters the
motivation of software engineers and how this manifests as something such as
behaviour, productivity or attitude.




1
  The percentages are a representative of how many of 92 papers the theories were found to be
explicitly used in. There can be multiple theories used in any one paper, and the 92 papers were part of
a systematic literature review conducted by Hall et al (2007) sampling over 500 players.


                                             Page 96 of 125
2010 CRC PhD Student Conference



References
   B.W. Boehm, Software Engineering Economics, Prentice-Hall, 1981.
   COUGER, J. D. AND ZAWACKI, R. A. 1980. Motivating and Managing Computer Personnel.
   John Wiley & Sons.
   S.A. Frangos, ā€œMotivated Humans for Reliable Software Products,ā€ Microprocessors and
   Microsystems, vol. 21, no. 10, 1997, pp. 605ā€“610.
   Frederick Herzberg, One More Time: How Do You Motivate Employees? (Harvard Business
   School Press, 1987).
   J. Procaccino and J.M. Verner, ā€œWhat Do Software Practitioners Really Think about Project
   Success: An Exploratory Study,ā€ J. Systems and Software, vol. 78, no. 2, 2005, pp. 194ā€“203.
   Richard M. Ryan and Edward L. Deci, ā€œIntrinsic and Extrinsic Motivations: Classic Definitions
   and New Directions,ā€ Contemporary Educational Psychology 25, no. 1 (January 2000): 54-67.
   Tracy Hall et al., ā€œA systematic review of theory use in studies investigating the motivations of
   software engineers,ā€ ACM Trans. Softw. Eng. Methodol. 18, no. 3 (2009): 1-29.
   Sarah Beecham et al., ā€œMotivation in Software Engineering: A systematic literature review,ā€
   Information and Software Technology 50, no. 9-10 (August 2008): 860-878.




                                            Page 97 of 125

More Related Content

Similar to Rae

Harvesting Intelligence from User Interactions
Harvesting Intelligence from User Interactions Harvesting Intelligence from User Interactions
Harvesting Intelligence from User Interactions
R A Akerkar
Ā 
A NOVEL WEB IMAGE RE-RANKING APPROACH BASED ON QUERY SPECIFIC SEMANTIC SIGNAT...
A NOVEL WEB IMAGE RE-RANKING APPROACH BASED ON QUERY SPECIFIC SEMANTIC SIGNAT...A NOVEL WEB IMAGE RE-RANKING APPROACH BASED ON QUERY SPECIFIC SEMANTIC SIGNAT...
A NOVEL WEB IMAGE RE-RANKING APPROACH BASED ON QUERY SPECIFIC SEMANTIC SIGNAT...
Journal For Research
Ā 
Privacy Policy Inference of User-Uploaded Images on Content Sharing Sites
Privacy Policy Inference of User-Uploaded Images on Content Sharing SitesPrivacy Policy Inference of User-Uploaded Images on Content Sharing Sites
Privacy Policy Inference of User-Uploaded Images on Content Sharing Sites
1crore projects
Ā 
Privacy Policy Inference of User-Uploaded Images on Content Sharing Sites
Privacy Policy Inference of User-Uploaded Images on Content Sharing SitesPrivacy Policy Inference of User-Uploaded Images on Content Sharing Sites
Privacy Policy Inference of User-Uploaded Images on Content Sharing Sites
1crore projects
Ā 
Techniques Used For Extracting Useful Information From Images
Techniques Used For Extracting Useful Information From ImagesTechniques Used For Extracting Useful Information From Images
Techniques Used For Extracting Useful Information From Images
Jill Crawford
Ā 
iaetsd Adaptive privacy policy prediction for user uploaded images on
iaetsd Adaptive privacy policy prediction for user uploaded images oniaetsd Adaptive privacy policy prediction for user uploaded images on
iaetsd Adaptive privacy policy prediction for user uploaded images on
Iaetsd Iaetsd
Ā 

Similar to Rae (20)

Adaptive Search Based On User Tags in Social Networking
Adaptive Search Based On User Tags in Social NetworkingAdaptive Search Based On User Tags in Social Networking
Adaptive Search Based On User Tags in Social Networking
Ā 
TAG BASED IMAGE SEARCH BY SOCIAL RE-RANKING
TAG BASED IMAGE SEARCH BY SOCIAL RE-RANKINGTAG BASED IMAGE SEARCH BY SOCIAL RE-RANKING
TAG BASED IMAGE SEARCH BY SOCIAL RE-RANKING
Ā 
Harvesting Intelligence from User Interactions
Harvesting Intelligence from User Interactions Harvesting Intelligence from User Interactions
Harvesting Intelligence from User Interactions
Ā 
Ko3419161921
Ko3419161921Ko3419161921
Ko3419161921
Ā 
Social Re-Ranking using Tag Based Image Search
Social Re-Ranking using Tag Based Image SearchSocial Re-Ranking using Tag Based Image Search
Social Re-Ranking using Tag Based Image Search
Ā 
A NOVEL WEB IMAGE RE-RANKING APPROACH BASED ON QUERY SPECIFIC SEMANTIC SIGNAT...
A NOVEL WEB IMAGE RE-RANKING APPROACH BASED ON QUERY SPECIFIC SEMANTIC SIGNAT...A NOVEL WEB IMAGE RE-RANKING APPROACH BASED ON QUERY SPECIFIC SEMANTIC SIGNAT...
A NOVEL WEB IMAGE RE-RANKING APPROACH BASED ON QUERY SPECIFIC SEMANTIC SIGNAT...
Ā 
Socially Shared Images with Automated Annotation Process by Using Improved Us...
Socially Shared Images with Automated Annotation Process by Using Improved Us...Socially Shared Images with Automated Annotation Process by Using Improved Us...
Socially Shared Images with Automated Annotation Process by Using Improved Us...
Ā 
Privacy Policy Inference of User-Uploaded Images on Content Sharing Sites
Privacy Policy Inference of User-Uploaded Images on Content Sharing SitesPrivacy Policy Inference of User-Uploaded Images on Content Sharing Sites
Privacy Policy Inference of User-Uploaded Images on Content Sharing Sites
Ā 
Privacy Policy Inference of User-Uploaded Images on Content Sharing Sites
Privacy Policy Inference of User-Uploaded Images on Content Sharing SitesPrivacy Policy Inference of User-Uploaded Images on Content Sharing Sites
Privacy Policy Inference of User-Uploaded Images on Content Sharing Sites
Ā 
Bn35364376
Bn35364376Bn35364376
Bn35364376
Ā 
benchmarking image retrieval diversification techniques for social media
benchmarking image retrieval diversification techniques for social mediabenchmarking image retrieval diversification techniques for social media
benchmarking image retrieval diversification techniques for social media
Ā 
HABIB FIGA GUYE {BULE HORA UNIVERSITY}(habibifiga@gmail.com
HABIB FIGA GUYE {BULE HORA UNIVERSITY}(habibifiga@gmail.comHABIB FIGA GUYE {BULE HORA UNIVERSITY}(habibifiga@gmail.com
HABIB FIGA GUYE {BULE HORA UNIVERSITY}(habibifiga@gmail.com
Ā 
VU University Amsterdam - The Social Web 2016 - Lecture 5
VU University Amsterdam - The Social Web 2016 - Lecture 5VU University Amsterdam - The Social Web 2016 - Lecture 5
VU University Amsterdam - The Social Web 2016 - Lecture 5
Ā 
Lecture 5: Personalization on the Social Web (2014)
Lecture 5: Personalization on the Social Web (2014)Lecture 5: Personalization on the Social Web (2014)
Lecture 5: Personalization on the Social Web (2014)
Ā 
Techniques Used For Extracting Useful Information From Images
Techniques Used For Extracting Useful Information From ImagesTechniques Used For Extracting Useful Information From Images
Techniques Used For Extracting Useful Information From Images
Ā 
A Meteoroid on Steroids: Ranking Media Items Stemming from Multiple Social Ne...
A Meteoroid on Steroids: Ranking Media Items Stemming from Multiple Social Ne...A Meteoroid on Steroids: Ranking Media Items Stemming from Multiple Social Ne...
A Meteoroid on Steroids: Ranking Media Items Stemming from Multiple Social Ne...
Ā 
iaetsd Adaptive privacy policy prediction for user uploaded images on
iaetsd Adaptive privacy policy prediction for user uploaded images oniaetsd Adaptive privacy policy prediction for user uploaded images on
iaetsd Adaptive privacy policy prediction for user uploaded images on
Ā 
Service rating prediction by exploring social mobile usersā€™ geographical loca...
Service rating prediction by exploring social mobile usersā€™ geographical loca...Service rating prediction by exploring social mobile usersā€™ geographical loca...
Service rating prediction by exploring social mobile usersā€™ geographical loca...
Ā 
Personalized geo tag recommendation for community contributed images
Personalized geo tag recommendation for community contributed imagesPersonalized geo tag recommendation for community contributed images
Personalized geo tag recommendation for community contributed images
Ā 
benchmarking image retrieval diversification techniques for social media
benchmarking image retrieval diversification techniques for social mediabenchmarking image retrieval diversification techniques for social media
benchmarking image retrieval diversification techniques for social media
Ā 

More from anesah (20)

Aizatulin slides-4-3
Aizatulin slides-4-3Aizatulin slides-4-3
Aizatulin slides-4-3
Ā 
Aizatulin poster
Aizatulin posterAizatulin poster
Aizatulin poster
Ā 
Abraham
AbrahamAbraham
Abraham
Ā 
Mouawad
MouawadMouawad
Mouawad
Ā 
Pantidi
PantidiPantidi
Pantidi
Ā 
Wilkie
WilkieWilkie
Wilkie
Ā 
Van der merwe
Van der merweVan der merwe
Van der merwe
Ā 
Thomas
ThomasThomas
Thomas
Ā 
Taubenberger
TaubenbergerTaubenberger
Taubenberger
Ā 
Sach
SachSach
Sach
Ā 
Pantidi
PantidiPantidi
Pantidi
Ā 
Corneli
CorneliCorneli
Corneli
Ā 
Collins
CollinsCollins
Collins
Ā 
Xambo
XamboXambo
Xambo
Ā 
Ullmann
UllmannUllmann
Ullmann
Ā 
Tran
TranTran
Tran
Ā 
Quinto
QuintoQuinto
Quinto
Ā 
Pluss
PlussPluss
Pluss
Ā 
Pawlik
PawlikPawlik
Pawlik
Ā 
Overbeeke
OverbeekeOverbeeke
Overbeeke
Ā 

Recently uploaded

Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers:  A Deep Dive into Serverless Spatial Data and FMECloud Frontiers:  A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
Safe Software
Ā 
Understanding the FAA Part 107 License ..
Understanding the FAA Part 107 License ..Understanding the FAA Part 107 License ..
Understanding the FAA Part 107 License ..
Christopher Logan Kennedy
Ā 

Recently uploaded (20)

Repurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost Saving
Repurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost SavingRepurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost Saving
Repurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost Saving
Ā 
Rising Above_ Dubai Floods and the Fortitude of Dubai International Airport.pdf
Rising Above_ Dubai Floods and the Fortitude of Dubai International Airport.pdfRising Above_ Dubai Floods and the Fortitude of Dubai International Airport.pdf
Rising Above_ Dubai Floods and the Fortitude of Dubai International Airport.pdf
Ā 
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers:  A Deep Dive into Serverless Spatial Data and FMECloud Frontiers:  A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
Ā 
DEV meet-up UiPath Document Understanding May 7 2024 Amsterdam
DEV meet-up UiPath Document Understanding May 7 2024 AmsterdamDEV meet-up UiPath Document Understanding May 7 2024 Amsterdam
DEV meet-up UiPath Document Understanding May 7 2024 Amsterdam
Ā 
Strategies for Landing an Oracle DBA Job as a Fresher
Strategies for Landing an Oracle DBA Job as a FresherStrategies for Landing an Oracle DBA Job as a Fresher
Strategies for Landing an Oracle DBA Job as a Fresher
Ā 
Corporate and higher education May webinar.pptx
Corporate and higher education May webinar.pptxCorporate and higher education May webinar.pptx
Corporate and higher education May webinar.pptx
Ā 
Understanding the FAA Part 107 License ..
Understanding the FAA Part 107 License ..Understanding the FAA Part 107 License ..
Understanding the FAA Part 107 License ..
Ā 
FWD Group - Insurer Innovation Award 2024
FWD Group - Insurer Innovation Award 2024FWD Group - Insurer Innovation Award 2024
FWD Group - Insurer Innovation Award 2024
Ā 
MINDCTI Revenue Release Quarter One 2024
MINDCTI Revenue Release Quarter One 2024MINDCTI Revenue Release Quarter One 2024
MINDCTI Revenue Release Quarter One 2024
Ā 
AI in Action: Real World Use Cases by Anitaraj
AI in Action: Real World Use Cases by AnitarajAI in Action: Real World Use Cases by Anitaraj
AI in Action: Real World Use Cases by Anitaraj
Ā 
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...
Apidays New York 2024 - Accelerating FinTech Innovation by Vasa Krishnan, Fin...
Ā 
Apidays New York 2024 - The value of a flexible API Management solution for O...
Apidays New York 2024 - The value of a flexible API Management solution for O...Apidays New York 2024 - The value of a flexible API Management solution for O...
Apidays New York 2024 - The value of a flexible API Management solution for O...
Ā 
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
Ā 
AWS Community Day CPH - Three problems of Terraform
AWS Community Day CPH - Three problems of TerraformAWS Community Day CPH - Three problems of Terraform
AWS Community Day CPH - Three problems of Terraform
Ā 
DBX First Quarter 2024 Investor Presentation
DBX First Quarter 2024 Investor PresentationDBX First Quarter 2024 Investor Presentation
DBX First Quarter 2024 Investor Presentation
Ā 
CNIC Information System with Pakdata Cf In Pakistan
CNIC Information System with Pakdata Cf In PakistanCNIC Information System with Pakdata Cf In Pakistan
CNIC Information System with Pakdata Cf In Pakistan
Ā 
Apidays New York 2024 - APIs in 2030: The Risk of Technological Sleepwalk by ...
Apidays New York 2024 - APIs in 2030: The Risk of Technological Sleepwalk by ...Apidays New York 2024 - APIs in 2030: The Risk of Technological Sleepwalk by ...
Apidays New York 2024 - APIs in 2030: The Risk of Technological Sleepwalk by ...
Ā 
ICT role in 21st century education and its challenges
ICT role in 21st century education and its challengesICT role in 21st century education and its challenges
ICT role in 21st century education and its challenges
Ā 
Six Myths about Ontologies: The Basics of Formal Ontology
Six Myths about Ontologies: The Basics of Formal OntologySix Myths about Ontologies: The Basics of Formal Ontology
Six Myths about Ontologies: The Basics of Formal Ontology
Ā 
Spring Boot vs Quarkus the ultimate battle - DevoxxUK
Spring Boot vs Quarkus the ultimate battle - DevoxxUKSpring Boot vs Quarkus the ultimate battle - DevoxxUK
Spring Boot vs Quarkus the ultimate battle - DevoxxUK
Ā 

Rae

  • 1. 2010 CRC PhD Student Conference Supporting multimodal media recommendation and annotation using social network analysis Adam Rae a.rae@open.ac.uk Supervisors Stefan RĆ¼ger, Suzanne Little, Roelof van Zwol Department The Knowledge Media Institute Status Full Time Probation Viva After Starting Date October 2007 Research Hypothesis By analysing and extracting information from the social graphs de- scribed by both explicit and implicit user interactions, like those found in online media sharing systems like Flickr1 , it is possible to augment existing non-social aware recommender systems and thereby signiļ¬cantly improve their performance. Large scale web based systems for sharing media continue to tackle the problem of helping their users ļ¬nd what they are looking for in a timely manner. To do this, lots of good quality metadata is required to sift through the data collection to pick out exactly those documents that match the information need of the user. In the case of ļ¬nding images from the online photo sharing website Flickr, this could be from over 4 billion examples. How can we help both the system and the user in enriching the metadata of the media within the collection in order to improve the experience for the user and to reduce the burden on the underlying data handling system? Can modelling users, by themselves and within the context of the wider online community help? Can this modeling be used to improve recommender systems that improve the experience and reduce cognitive burden on users? Existing approaches tend to treat multimedia in the same way they have dealt with text documents in the past, speciļ¬cally by treating the textual meta- data associated with an image as a text document, but this ignores the inherently diļ¬€erent nature of the data the system is handling. Images are visual data, and while they can be described well by textual metadata, they cannot be described completely by it. Also, the user cannot be ignored in the retrieval process, and learning more about a user provides information to the system to tailor results to their speciļ¬c requirements. Users interact online, and these interactions form a 1 http://www.ļ¬‚ickr.com/ Page 91 of 125
  • 2. 2010 CRC PhD Student Conference new type of data that has yet to be fully explored nor exploited when modelling users. The work presented here combines the mining of social graphs that occur in Flickr with visual content and metadata analysis to provide better personalised photo recommender mechanisms and the following experiment and its analysis are a major component in my overall thesis. Interaction Scenario In order to address this research question, multiple experiments have been car- ried out, one of which I present here: Envisage an incoming stream of photos made available to a user. In systems of a scale similar to Flickr, this could be thousands of im- ages per second. Can a system that uses cues from the social, visual and semantic aspects of these images perform better than one that uses the more traditional approach of using only semantic informa- tion, according to speciļ¬cally deļ¬ned objective metrics? How does performance vary between users? An experiment was carried out that mines data from the social communities in Flickr, from the visual content of images and from the text based metadata and uses a machines learning mechanism to merge these signals together to form a classiļ¬er that, given a candidate image and prospective viewing user, decides whether the user would label that image as a ā€˜Favouriteā€™2 - see Figure 1. Related Work The signiļ¬cant inļ¬‚uence that our peers can have on our behaviour online has been studied by researchers such as Lerman and Jones[3], and the particular interaction that occurs between users and visual media in particular in the work of Nov et al.[4]and Kern et al[2]. Their insights into the importance of understanding more about a user in order to best fulļ¬l their information need supports the hypothesis that this kind of information can be usefully exploited to improve systems that try to match that need to a data set supported by social interaction. Here I extend their ideas by incorporating this valuable social data into a complementary multimodal framework that takes advantage of multiple types of data. The use of social interaction features in the work of Sigurbjƶrnsson and van Zwol[7] and Garg and Weber[1] inspired my more comprehensive feature set and its analysis. Their ļ¬ndings that aggregating data generated from online communities is valuable when suggesting tags is important and I believe also transfers to recommendation in general as well as to the speciļ¬c task of recom- mending images. In fact, I demonstrated this in previous work on social media tag suggestion[6]. I use some of the human perception based visual features outlined in the work of San Pedro and Siersdorfer[5], as these have been shown to work well in similar experimental scenarios and cover a range of visual classes. I extend them further with a selection of other high performing visual features. 2A binary label Flickr users can use to annotate an image they like. Page 92 of 125
  • 3. 2010 CRC PhD Student Conference Incoming stream of previously unseen candidate images Textual Social Visual User information User information Feature Extraction User A User B Has tagged beaches before Member of urban animals group Trained Classiļ¬er Potential Favourite Images Potential Favourite Images for User B for User A Figure 1: Diagram of the image classiļ¬cation system used with Flickr data. Experimental Work 400 users of varying levels of social activity were selected from Flickr and their ā€˜Favouriteā€™ labelled images collected. This resulted in a collection of hundreds of thousands of images. To train my classiļ¬er, these images were treated as positive examples of relevant images. I generated a variety of negative example sets to reļ¬‚ect realistic system scenarios. For all photo examples we extracted visual and semantic features, and social features that described the user, the owner of the photo, any connection between them as well as other behaviour metrics. We then tested our classiļ¬er using previously unseen examples and measured the performance of the system with a particular emphasis on the information retrieval metric of precision at 5 and 10 to reļ¬‚ect our envisaged use case scenario. Results An extract of the results from the experiment are shown in Table 1. They can be summarised thus: ā€¢ It is possible to achieve high levels of precision in selecting our positive examples, especially by using social features. This performance increase is statistically signiļ¬cantly higher than the baseline Textual run. These social signals evidently play a signiļ¬cant rĆ“le when a user labels an image a ā€˜Favouriteā€™ and can be usefully exploited to help them. ā€¢ The value of individual types of features is complex, but complementary. The combined systems tend to perform better than the individual ones. ā€¢ It is far easier to classiļ¬er photos that are not ā€˜Favouritesā€™ than those that are, as shown by the high negative values. This can be used to narrow down the search space for relevant images by removing those that are obviously not going to interest the user, thus reduing load on both the user and the system. Page 93 of 125
  • 4. 2010 CRC PhD Student Conference System Accuracy + Prec. + Rec. - Prec. - Rec. Textual 0.87 0.48 0.18 0.88 0.97 Visual 0.88 1.00 0.09 0.88 1.00 Social 0.92 0.80 0.56 0.94 0.98 Textual+Visual 0.88 0.62 0.27 0.90 0.97 Textual+Social 0.92 0.77 0.62 0.94 0.97 Visual+Social 0.93 0.89 0.56 0.94 0.99 Text+Vis.+Soc. 0.93 0.84 0.62 0.94 0.98 Table 1: Accuracy, precison and recall for various combinations of features using the experiments most realistic scenario data set. Photos labelled as ā€˜Favouritesā€™ are positive examples, and those that are not are negative examples. Higher numbers are better. ā€¢ As is typical in this style of information retrieval experiment, we can trade- oļ¬€ between precision and recall depending on our requirements. As we are interested in high precision in this particular experiment, we see that the combination of the Visual+Social and Text+Visual+Social runs give good precision without sacriļ¬cing too much recall. References [1] Nikhil Garg and Ingmar Weber. Personalized, interactive tag recommenda- tion for ļ¬‚ickr. In Proceedings of the 2008 ACM Conference on Recommender Systems, pages 67ā€“74, Lausanne, Switzerland, October 2008. ACM. [2] R. Kern, M. Granitzer, and V. Pammer. Extending folksonomies for image tagging. In Workshop on Image Analysis for Multimedia Interactive Services, 2008, pages 126ā€“129, May 2008. [3] Kristina Lerman and Laurie Jones. Social browsing on ļ¬‚ickr. In Proceedings of ICWSM, December 2007. [4] Oded Nov, Mor Naaman, and Chen Ye. What drives content tagging: the case of photos on ļ¬‚ickr. In Proceeding of the twenty-sixth annual SIGCHI conference on Human factors in computing systems, pages 1097ā€“1100, Flo- rence, Italy, 2008. ACM. [5] Jose San Pedro and Stefan Siersdorfer. Ranking and classifying attractive- ness of photos in folksonomies. In WWW, Madrid, Spain, April 2009. [6] Adam Rae, Roelof van Zwol, and Bƶrkur Sigurbjƶrnsson. Improving tag recommendation using social networks. In 9th International conference on Adaptivity, Personalization and Fusion of Heterogeneous Information, April 2010. [7] Roelof van Zwol. Flickr: Who is looking? In IEEE/WIC/ACM Inter- national Conference on Web Intelligence, pages 184ā€“190, Washington, DC, USA, 2007. IEEE Computer Society. Page 94 of 125
  • 5. 2010 CRC PhD Student Conference The effect of Feedback on the Motivation of Software Engineers Rien Sach r.j.sach@open.ac.uk Supervisors Helen Sharp Marian Petre Department/Institute Computing Status Fulltime Probation viva After Starting date October 2009 Motivation is reported as having an effect on crucial aspects of software engineering such as productivity (Procaccino and Verner 2005), software quality (Boehm 1981), and a projectā€™s overall success (Frangos 1997). Feedback is a key factor in the most commonly used theory in reports published on the motivation of software engineers (Hall et al. 2009), and it is important that we gain a greater understanding of the effect it has on the motivation of software engineers. My research is grounded in the question ā€œWhat are the effects of feedback on the motivation of software engineers?ā€, and focuses on feedback conveyed in human interactions. I believe that before I can focus my question further I will need to begin some preliminary work to identify how feedback occurs, what types of feedback occur, and the possible impact of this feedback. Motivation can be understood in different ways. For example, as a manager you might consider motivation as something you must maintain in your employees to ensure they complete work for you as quickly as possible. As an employee you might consider motivation as the drive that keeps you focused on a task, or it might simply be what pushes you to get up in the morning and go to work. Herzberg (1987) describes motivation as ā€œa function of growth from getting intrinsic rewards out of interesting and challenging workā€. Thatā€™s quite a nice definition; and according to Herzberg motivation is intrinsic to oneā€™s self. Ryan and Deci (2000) describe intrinsic motivation as ā€œthe doing of activity for its inherent satisfaction rather than for some separable consequenceā€ (Page 60). Herzberg (1987) defines extrinsic factors as movement and distinguishes it from motivation, stating that ā€œMovement is a function of fear of punishment or failure to get extrinsic rewardsā€. Ryan and Deci (2000) state that ā€œExtrinsic motivation is a construct that pertains whenever an activity is done in order to attain some separable outcomeā€. There are 8 core motivational theories (Hall et al. 2009) and some of the theories focus on motivation as a ā€œa sequence or process of related activitiesā€ (Hall et al. 2009) called process theories, while others focus on motivation ā€œat a single point in timeā€ (Couger and Zawacki 1980) called content theories. Page 95 of 125
  • 6. 2010 CRC PhD Student Conference As reported in a systematic literature review conducted by Beecham et al (2007), and their published review of the use of theory inside this review in 2009 (Hall et al 2009), the three most popular theories used in studies of motivation in Software Engineering were Hackman and Oldmanā€™s Job Characteristics Theory (68%), Herzbergā€™s Motivational Hygiene Theory (41%), and Maslowā€™s Theory of Needs (21%)1. Hackman and Oldmanā€™s Job Characteristics Theory focuses on the physical job, and suggests five characteristics (skill variety, task identity, task significance, autonomy, and feedback) that lead to three psychological states which in turn lead to higher internal motivation and higher quality work. Herzbergā€™s Hygiene Theory suggests that the only true motivation is intrinsic motivation, and this leads to job satisfaction, where extrinsic factors are only useful in avoiding job dissatisfaction. One of the five key job characteristics in Hackman and Oldmanā€™s theory is feedback. Feedback is not explicitly mentioned in Herzbergā€™s Motivational Hygiene Theory, but he notes that it is a part of job enrichment, which he states is ā€œkey to designing work that motivates employeesā€ (Herzberg 1987). However this is a managerial view point. Software Engineers are considered to be current practitioners working on active software projects within the industry. This includes programmers, analysts, testers, and designers who actively work and produce software for real projects in the real world. From a management perspective, gaining a greater understanding of what motives employees could prove invaluable in increasing productivity and software quality, and from an individual perspective the prospect of being given feedback that motivates you and makes your job more enjoyable and improves the quality of your work experience could lead to a more successful and enjoyable work life. My proposed research is divided into stages. In the first stage I plan to conduct interviews and diary studies to identify the types of feedback in software engineering and how feedback is experienced by software engineers. I then plan to conduct additional studies to identify what impact this feedback has on software engineers and how that impact is evident. Finally, I plan to observe software engineers at work to see feedback in context, and to compare those observations to the information gathered during the first two stages. At the end of my PhD I hope to accomplish research that leads to a greater understanding of what feedback is inside software engineering and how it is given or received. Subsequently I wish to gain an understanding of how this feedback alters the motivation of software engineers and how this manifests as something such as behaviour, productivity or attitude. 1 The percentages are a representative of how many of 92 papers the theories were found to be explicitly used in. There can be multiple theories used in any one paper, and the 92 papers were part of a systematic literature review conducted by Hall et al (2007) sampling over 500 players. Page 96 of 125
  • 7. 2010 CRC PhD Student Conference References B.W. Boehm, Software Engineering Economics, Prentice-Hall, 1981. COUGER, J. D. AND ZAWACKI, R. A. 1980. Motivating and Managing Computer Personnel. John Wiley & Sons. S.A. Frangos, ā€œMotivated Humans for Reliable Software Products,ā€ Microprocessors and Microsystems, vol. 21, no. 10, 1997, pp. 605ā€“610. Frederick Herzberg, One More Time: How Do You Motivate Employees? (Harvard Business School Press, 1987). J. Procaccino and J.M. Verner, ā€œWhat Do Software Practitioners Really Think about Project Success: An Exploratory Study,ā€ J. Systems and Software, vol. 78, no. 2, 2005, pp. 194ā€“203. Richard M. Ryan and Edward L. Deci, ā€œIntrinsic and Extrinsic Motivations: Classic Definitions and New Directions,ā€ Contemporary Educational Psychology 25, no. 1 (January 2000): 54-67. Tracy Hall et al., ā€œA systematic review of theory use in studies investigating the motivations of software engineers,ā€ ACM Trans. Softw. Eng. Methodol. 18, no. 3 (2009): 1-29. Sarah Beecham et al., ā€œMotivation in Software Engineering: A systematic literature review,ā€ Information and Software Technology 50, no. 9-10 (August 2008): 860-878. Page 97 of 125