This document discusses a framework for acquiring vague knowledge from socially generated content in an enterprise setting. It involves setting up a microblogging platform for employees to discuss topics related to the enterprise. Vague knowledge assertions are extracted from posts and used to determine fuzzy degrees and membership functions for concepts, relations, and datatypes in a fuzzy ontology representing the enterprise's knowledge. The strength of each assertion is calculated based on social characteristics of the discussions. Future work involves applying the framework in a real enterprise to evaluate its ability to acquire vague knowledge and accuracy of the learned fuzzy ontology.

1. Learning Vague Knowledge From Socially
Generated Content in an Enterprise Framework
Panos Alexopoulos, John Pavlopoulos, Phivos Mylonas
1st Mining Humanistic Data Workshop,
Halkidiki, Greece, September 27th, 2012

2. 2
 Introduction
 Background and Problem Definition
 Approach Overview and Rationale
 Vague Knowledge Acquisition
 Conceptualization and Initialization
 Microblogging Framework
 Extraction of Vague Knowledge
Assertions
 Assertion Strength Assessment
 Generation of Membership
Functions and Fuzzy Degrees
 Conclusions and Future Work
Agenda

3. 3
Background
Introduction
●Knowledge Management is a discipline that aims to enable enterprises and
organizations to fully leverage their knowledge in their effort to grow more
efficient and competitive.
●This leverage involves several key objectives such as:
● Identification, gathering and organization of existing knowledge
● Sharing and reusing of this knowledge for different applications and
users and facilitation of new knowledge creation
●However, a dimension of this knowledge that has so far been inadequately
considered is vagueness.

4. 4
Vagueness
Introduction
● Vagueness is manifested through
predicates that admit borderline cases,
i.e. cases where it is unclear whether or
not the predicate applies
● E.g. Tall, High, Experienced etc.
Definition
● Degree Vagueness: Lack of crisp
boundaries between application and non
application in some dimension.
● E.g. Tall, Rich, Recent
● Combinatory Vagueness: Inability to
clearly define adequate applicability
criteria.
● Π.χ. Modern, Expert, Religion
Types of Vagueness
● Uncertainty: E.g. Today it might rain
● Inexactness: E.g. Someone has height
between 170 and 180 cm.
Frequently Confused Concepts
● A person can be tall with respect to the
average population height and not tall
with respect to professional basketball
players
● This doesn’t mean that a vague predicate
can stop being vague in a different
context but merely that the interpretation
of its vagueness can change.
Context Dependence

5. 5
Fuzzy Ontologies
Introduction
●Fuzzy Ontologies are extensions of classical ontologies that allow the
assignment of truth degrees to vague ontological elements.
●For example:
● “The project's budget is satisfactory to a degree of 0.7"
● “Jane is an expert at Artificial Intelligence to a degree of 0.5".

6. 6
Fuzzy Ontological Elements
Introduction
● A fuzzy ontology concept may have
instances that belong to it at certain degrees.
● E.g. John is a TallPerson to a degree of 0.5.
Fuzzy Concepts
● A fuzzy ontology relation links concept
instances at certain degrees.
● Π.χ. John is expert at Machine Learning to a
degree of 0.9.
● Similarly, a fuzzy attribute assigns literal
values to concept instances at certain
degrees.
Fuzzy Relations and Attributes
● A fuzzy datatype consists of a set of vague
terms which may be used within the
ontology as attribute values.
● Π.χ. Low, Average, High for the
attribute Project Budget.
● Each term is mapped to a fuzzy set that
defines the term’s meaning.
Fuzzy Datatypes

7. 7
Problem Definition
Introduction
● An important bottleneck in developing fuzzy ontologies is the definition of the degrees and
membership functions of the fuzzy elements.
● High level of subjectivity.
● Context dependence.
● Thus the problem is defined as follows: Given a fuzzy enterprise ontology, what are the
optimal fuzzy degrees and membership functions that should be assigned to its
elements (concepts, relations and datatypes) in order to represent the domain’s vagueness
as accurately as possible?
● E.g. given the fuzzy concept CompanyCompetitor and a set of individual components,
what is the degree to which each of these companies is considered a competitor?
● E.g. given the fuzzy relation isExpertAt and a set of related through it pairs of
instances (e.g. persons related to business areas), what is the degree to which the
relation between these pairs actually stands?
● E.g. given the fuzzy datatype ProjectBudget and the terms it consists of (e.g. low,
average, high), what are the membership functions of the fuzzy sets that best reflect
the meaning of each of these terms?

8. 8
Process Overview
Vague Knowledge Acquisition
1. Identification within the enterprise of vague knowledge and conceptual modeling of it
in the form of a fuzzy enterprise ontology.
2. Setting up of a microblogging platform in which the members of the enterprise are
expected to participate and perform discussions and information exchange on all
aspects regarding the enterprise and its environment.
3. Detection and extraction from the user generated platform’s content of vague
knowledge assertions, namely statements related to the elements already defined in
the fuzzy enterprise ontology.
4. Calculation for each vague assertion of a strength value based on the utilization of
various characteristics of the discussions they are involved in.
5. Aggregation of these assertions and automated generation of fuzzy degrees and
membership functions.

9. 9
Conceptualization and Initialization with IKARUS-Onto
Vague Knowledge Acquisition
Acquire
Crisp Ontology
Define Fuzzy
Ontology Elements
Formalize Fuzzy
Elements
Validate Fuzzy
Ontology
● Establish a basis for the development
of the fuzzy ontology.
● Develop or acquire the crisp ontology.
● Justify and estimate the necessary
work for the fuzzy ontology
development.
● Ensure existence of vagueness in the
domain.
● Ensure vagueness is a requirement.
● Conceptualization of vagueness in an
explicit way and shareable way.
● Definition of fuzzy ontology elements
● Specification of fuzzy degrees and
membership functions by experts.
● Make fuzzy ontology machine-
processable.
● Select fuzzy ontology language and
use it to represent the defined
elements.
● Ensure adequate and correct
capturing of the domain’s vagueness
● Check correctness, accuracy,
completeness and consistency.
Step Goals Actions
Establish Need
for Fuzziness

10. 10
Microblogging Platform
Vague Knowledge Acquisition
● The microblogging platform we adopt for the purposes of this work is miKrow, an
intra-enterprise semantic microblogging tool that allows its end-users to share short
messages expressing what are they working at.
● The platform works mostly like Twitter, with two important enhancements:
● When users reply to a message they are able to denote the nature of their reply
by using the predefined hashtags #support and #attack.
● Users are also able to denote their agreement or disagreement to a message
through a rating functionality
● These two features allows us to use the platform as an argumentation tool and
capture the disagreements and debates over vague knowledge statements that may
occur.

11. 11
Microblogging Platform
Vague Knowledge Acquisition

12. 12
Extraction of Vague Knowledge Assertions
Vague Knowledge Acquisition
● Vague knowledge assertions are practically statements related to the elements of
fuzzy ontology.
● E.g. The assertion “The budget for the project X is low" is related to the fuzzy
datatype “ProjectBudget”
● E.g. The assertion “John is expert at ontologies" is related to the fuzzy relation
“isExpertAt”.
● Our goal is to detect and extract such assertions from the user messages so that we
can use them for determining the fuzzy degrees of their respective elements.
● To achieve this, we use an in-house developed semantic annotation tool that, given a
fuzzy ontology, is able to recognize such assertions within a piece of text.
● An important factor that contributes to higher levels of precision for this detection is
the fact that microblogging messages are short.
● In any case, the detection process may be performed in a semi-automatic fashion
where the correctness of the extracted assertions could be checked by the system’s
administrator.

13. 13
Knowledge Assertion Strength Calculation
Vague Knowledge Acquisition
● To calculate the strength of the extracted vague assertions we aggregate the strength
of messages that are directly or indirectly related to these assertions.
● Strength of a message depends on:
● On the number of agreements and disagreements it has received by the users.
● On the number and strength of attacking and supporting messages.
● On the overall influence of the user who has published the message. This is
generally relevant to the number of users that follow the message publisher but
also to the person’s expertise on the messages topic.

14. 14
Generation of Membership Functions and Fuzzy Degrees
Vague Knowledge Acquisition
● Given an instance and a fuzzy concept we consider all the relevant assertions along
with their strengths:
● A1: “Accenture is a Competitor to a strength of 0.5.”
● A2: “Accenture is a Competitor to a strength of 0.7.”
● A3: “Accenture is a Competitor to a strength of 0.4.”
● To aggregate these strengths into a single degree we:
● Compute the mean value of all the strength values
● We estimate confidence intervals and we only allow those mean values with
significance level of no less than 0.05.
● In most cases we expect to have most assertions gathered very close to a single
mean value which can then be considered as the degree of the relevant ontological
statement.
● In case many assertions seem to be out of the confidence intervals, then that’s an
indication that the statement’s interpretation might be context-dependent.
Fuzzy Concept and Relation Assertions

15. 15
Generation of Membership Functions and Fuzzy Degrees
Vague Knowledge Acquisition
● Given a term and a fuzzy datatype we consider all the relevant assertions along with
their strengths:
● A1: “A budget of 20,000 is low to a strength of 0.5.”
● A2: “A budget of 24,000 is low to a strength of 0.3.”
● A3: “A budget of 26,000 is average to a strength of 0.6.”
● A4: “A budget of 30,000 is high to a strength of 0.4.”
● Based on the pairs of these values and strengths, we determine the optimal function
fuzzy membership function that links them.
● This is a well-studied problem in the area of fuzzy expert systems and several related
methods that construct such functions from training data are available.
Fuzzy Datatypes

16. 16
Key Points
Conclusions and Future Work
● We proposed a framework for automatic vague knowledge acquisition in enterprise
settings based on:
● A semantically enhanced microblogging system.
● A fuzzy ontology learning process that acts upon the social content produced by
the enterprise’s people.
● The key characteristic of our approach is the utilization of the content’s social
features in order to assign strengths to vague assertions.
● The relative agreement and support that microposts enjoy.
● The status and influence of the users.

17. 17
Future Work
Conclusions and Future Work
● In the future we intend to apply our framework in an actual enterprise setting and
evaluate its effectiveness in acquiring vague knowledge.
● This evaluation will focus on two dimensions:
● The ability of the microblogging approach in producing rich social context over
the vague knowledge.
● The accuracy of the fuzzy ontology degrees and membership functions learned
using this context.

18. 18
Contact iSOCO
Where we are
Questions?
Barcelona
Tel +34 935 677 200
Edificio Testa A
C/ Alcalde Barnils, 64-68
St. Cugat del Vallès
08174 Barcelona
Valencia
Tel +34 963 467 143
Oficina 107
C/ Prof. Beltrán Báguena, 4
46009 Valencia
Pamplona
Tel +34 948 102 408
Parque Tomás
Caballero, 2, 6º-4ª
31006 Pamplona
Dr. Panos Alexopoulos
Senior Researcher
palexopoulos@isoco.com
Madrid
Tel +34 913 349 797
Av. del Partenón, 16-18, 1º7ª
Campo de las Naciones
28042 Madrid