2. Approaches?
Trust Management
2
Trust models
■ serve as a decision criterion for an agent to engage in activities
One of the approaches
■ Reputation-based approach
□ use reputation as a base for trust
□ closed domains: each has its own method to query, store, aggregate,
infer, interpret and represent reputation
□ Used in:
– Web communities (e-Markets, blogs, social networks)
– Services
– Software agents
Reputation.Interoperability (Semantic Technologies);
3. Concepts?
Reputation Approach
3
■ Reputation Target
users, movies, products, blog posts, tags, companies, services, software
agents, and IP addresses
■ Reputation Model
all of the reputation statements, events, and processes for a particular
context
■ Reputation Context
the relevant category for a specific reputation
Reputation.Interoperability (Semantic Technologies);
4. Concepts?
Reputation Approach
4
■ Reputation Target
users, movies, products, blog posts, tags, companies, services, software
agents, and IPComputation Function
addresses
Communication Model
■ Reputation Model
all of the reputation statements, events, and processes for a particular
Participants
context
Resources
■ Reputation Context
Representation Model
the relevant category for a specific reputation
Storage
Functionalities and Applications
Reputation.Interoperability (Semantic Technologies);
5. Where…
Rep. Approach
5 Reputation of
Reputation of
users service providers
In Service-oriented Arch.
Reputation of
Reputation of business domains
services
Social/entertain
Slashdot E-Markets ment
News Online Reputation
Systems
Opinion & Business/Jobs
Activities network
10. Ratings and Reviews
Online Markets
10
■ After buying, the consumer is asked to give his feedback in two ways:
a) stars ratings
b) by answering a seller-feedback questions with an option of leaving a
comment.
■ No obvious distinction -at the rating page- of what exactly being rated or
reviewed.
■ In one of our user studies:
Differentiate between 5 Stars and reviews
35% Did not know, thought
65% maybe product quality
Customer Service
Reputation.Interoperability (Semantic Technologies);
11. Ratings in Online Markets
Online Markets
11
■ Only at the description page (policies)
□ stars rating is an overall rating of the product
□ detailed review page is for the buying experience = reviewing the
seller (order fulfillment, customer service, correct item description)
”If your comments include any of the following, your feedback is subject to
removal:
Product reviews: It is more appropriate to review product on the product detail
page....Customers reviews are for products”.
■ Three reputation attributes:
□ product quality
□ seller reputation
□ customer service
Reputation.Interoperability (Semantic Technologies);
12. Our User Studies
12
Online Survey + interviews: Online Chocolate Store
200 users, different + detailed ratings
countries
■ test
□ how users perceive reputation
□ how many of reputation attributes the users consider
□ which of them the users focus on
□ used in the decision process with each other or separately?
□ their relation to the overall rating
Reputation.Interoperability (Semantic Technologies);
13. User Survey:
http://www.kwiksurveys.com?s=ILNING_865460b6
Study 1
13
14. User Survey:
http://www.kwiksurveys.com?s=ILNING_865460b6
Study 1
14
15. Results
15 eBay Rating Frequency
(number,
40 Gold 14
Sometimes
60 %
stars)
21%
Stars and
% 65 Never
% detailed
Reviews %
other
No. of
20 Reviews
No. of
Reviews
% 40
80 High-Value 60 % Detailed
ratings
%
Stars %
Reputation.Interoperability (Semantic Technologies);
16. Results
16 eBay Rating Frequency
(number,
40 Gold 14
Sometimes
60 %
stars)
21%
Stars and
% 65 Never
% detailed
Reviews %
other high seller reputation
conjoint
measure of
11 price & quality
% 50 Good Customer
39 Service
%
% other
No. of
20 Reviews
No. of
Reviews
% 40
80 High-Value 60 % Detailed
ratings
%
Stars %
Reputation.Interoperability (Semantic Technologies);
18. Results: comparison bet. Rating styles
18
no ”reviews” option available
users explicitly asked for it to be added stating that this is the only
way to gather more information on a provider before selecting him
Reputation.Interoperability (Semantic Technologies);
19. What does this mean?
Analysis
19
■ Confusion in interpreting the meaning of
□ rating styles
□ reputation values
■ Reputation of a seller or a product means more than one attribute
□ combination of attributes: neither represented nor clear from current
rating methods
■ Detailed ratings were preferred over stars ratings and high number-of-
reviews
■ Users tend to read reviews (comments and feedback) in order to decide on
a product or a seller
Reputation.Interoperability (Semantic Technologies);
20. ChocStore
Study 2
20 ■ Online Chocolate Store
■ For our institution personal
■ Normal Online store functionalities
■ 2 choices of payments
■ 3 choices of delivery
■ prices were changed
significantly compared to
the procurement cost
(underpriced, overpriced)
Reputation.Interoperability (Semantic Technologies);
22. Online Chocolate Store
Objectives
22
■ How many attributes are suitable for a review
■ What is the important aspect of each user’s rating
■ Based on the previous study, we show that several attributes - delivery
time for instance- affects rating
■ To examine categorized ratings with multiple attributes vs. overall ratings
Reputation.Interoperability (Semantic Technologies);
23. Online Chocolate Store
Results
23
■ The overall rating does not always relate to the same attribute (i.e.
delivery, quality, price) -> overall rating does not convey or show the
meaning behind it
■ Delivery time affects delivery rating and sometimes overall rating
■ Prices always affect overall rating
■ Pick one attribute that is most important to your overall rating:
Rating Attributes
Delivery
23%
38.50% Price
38.50%
Quality
Reputation.Interoperability (Semantic Technologies);
24. What does this mean?
Analysis
24
■ Users gave the same overall rating for different reasons
■ Some cared more about product quality, others cared for how fast the
delivery is
■ Average 4 attributes in the form was acceptable by all users
■ Singular formats of reputation is not enough
■ ignore the reasons and information behind the ratings
■ Users use several pieces of information to decide on a service provider
Reputation.Interoperability (Semantic Technologies);
25. So?
Discussion
25
■ A more aware user
■ read textual reviews to find what he/she is looking for
■ Possible -> for human users though time consuming
Not Possible -> in other domains e.g. software agents or web services,
■ text analysis: a highly expensive task that can not be performed for every
transaction
■ A user seeking a provider checks for high reputed ones
■ assuming that the high reputation interprets into his own attribute of
selection
Reputation.Interoperability (Semantic Technologies);
26. Why Rating is not enough?
No context
26 Bad E-Shop
Review
Business
Relying Party Owner/Seller/Facto
ry
Delivery
User Service
Delayed
Package
Context excluded from the reputation value
□ reputation query is too general
□ 3 different contexts
□ delivery, quality, price
Reputation.Interoperability (Semantic Technologies);
27. Why Rating is not enough?
Single Rating
27
■ Rating “used books”
□ is the rating for the book itself -> the user liked what he read
□ or the quality of the book -> was new and good printing
□ or the service provided by Amazon for example -> offering the
book, price, delivery, payment method, etc.
Reputation.Interoperability (Semantic Technologies);
28. Why Rating is not enough?
Different perceptions
28
Different representations, interaction styles and
trust rating scales
Reputation.Interoperability (Semantic Technologies);
29. Why Rating is not enough?
Different perceptions
29
Isolated reputation communities that have different:
□ perception of reputation
□ calculation of reputation
□ interpretation of reputation
□ overall reputation – not context related
Different representations, interaction styles and
trust rating scales
Reputation.Interoperability (Semantic Technologies);
30. Why Rating is not enough?
No portability
30
Reputation.Interoperability (Semantic Technologies);
31. Why Rating is not enough?
No portability
31
□ Starting from scratch for each domain
□ Cold start problem
□ No reputation information exchange
Reputation.Interoperability (Semantic Technologies);
32. Why Rating is not enough?
No portability
32
□ Starting from scratch for each domain
□ Cold start problem
□ No reputation information exchange
Solution
Unify the representation not the calculation
Define a generic reputation Ontology
Embed more information- relating semantics
Facilitate knowledge exchange
Reputation.Interoperability (Semantic Technologies);
33. Ontologies
Why Ontologies?
33
Ontologies
Concepts & relationships used to describe & represent an area of
knowledge
■ creates a common understanding
■ specifies the factors -their explicit semantics - involved in computing
reputation
■ separates the definition of reputation from how it is calculated
■ enables the mapping between reputation concepts in different models
■ facilitates the use of existing mapping &integration techniques in IS for
reusing reputation info
■ reputation interoperability & cross community sharing of reputation
information
Reputation.Interoperability (Semantic Technologies);
34. Competency Questions
Reputation Requirements
34
Q1 Reputation definition
define the notion of reputation within the domain?
Q2 Reputation Identity
entities? reputation roles such as source, target, evaluator, etc.?
Q3 Reputation representation
in a single format? is it enough to express its meaning? how reputation will be
represented, communicated?
Q4 Reputation statement
a reputation statement? what information does a reputation transaction hold?
Q5 Reputation computation mechanism
is there a property that defines and describes the computation mechanism?
Reputation.Interoperability (Semantic Technologies);
35. Cont. Competency Questions
Reputation Requirements
35
Q6 Reputation context
a property that expresses the relation between a reputation value and the context of its
creation? combine its reputation in different contexts?
Q7 Reputation factors
factors affecting reputation? does the source’s reputation affect reputation calculation?
Q8 Reputation dynamics and temporal effect
change through time? properties that reflect the change in reputation values? time
validity? is the new value time-stamped?
Q9 Reputation history
can we maintain the history of reputation values that an entity owned?
Q10 Reputation expressiveness
can we define and describe the semantics of the involved factors, contexts, relations,
and concepts? is there a way to communicate the semantics of a reputation context?
Reputation.Interoperability (Semantic Technologies);
36. Reputation Object Model
Representation
36
The RO model
■ Uses more information about the domain
□ the contexts and/or relevant quality criteria
■ Using this information, reputation is represented differently
□ as a developed object
■ The Reputation Object profiles an entity’s performance and has knowledge about
□ contexts
□ ratings values/reviews/feedback
□ computation functions
□ collecting method
Reputation Object
a profile of an entity’s performance
Reputation.Interoperability (Semantic Technologies);
41. Which Technology?
Used Technology
41
■ Developing interoperable reputation objects requires
■ structure and standardize reputation info and its relevant data
■ enable data integration
■ provide ways to relate the data to its explicit semantics
■ provide common data representation framework in order
to facilitate the integration of multiple sources to draw new
conclusions
Semantic
Technologies
Reputation.Interoperability (Semantic Technologies);
42. What is Semantic Web?
Semantic Technologies
42
■ extension of the current web in which information is given well-defined
meaning, better enabling computers and people to work in cooperation
■ collection of standard technologies to realize a Web of Data where they are
linked & are understandable by machines
■ provide common data representation framework in order to facilitate the
integration of multiple sources to draw new conclusions
■ Goals
■ Standard Representation
■ Linkability and Integration
■ Automation
■ Reuse across applications
Reputation.Interoperability (Semantic Technologies);
43. …define and structure
Semantic Technologies
43
Define Structure
■ Ontologies ■ RDF
Concepts &relationships used to ■ RDFa, microformats
describe & represent an area of ■ OWL
knowledge ■ …
Data
Integration
Phases
Reputation.Interoperability (Semantic Technologies);
44. RO Ontology: OWL
RO Ontology
44
■ Developed using Protégé 3.4.4 OWL-DL
■ Vocabulary of RO Ontology:
■ to represent an entity's (foaf:Agent) reputation
■ an object (ReputationObject) has one or multiple
instances of class Criterion or QualityAttribute
■ each criterion instance has a ReputationValue
(currentValueand historyList) that has a set of
PossibleValues (as literals or resources URI)
■ a criterion is collected by a CollectingAlgorithm&
computed using a ComputationAlgorithm
■ Employing also known vocabulary
OWL, RDFS, FOAF, XSD, RDF Review, ..
Reputation.Interoperability (Semantic Technologies);
46. Using Semantic Technologies
Goals
46
enabling reputation information exchange
facilitate the integration of multiple sources to draw new conclusions,
connecting data to its definitions and to its context
achieving reputation interoperability
Context-aware reputation
ensuring understandability and reusability of the embedded information
Semantic
Technologies
Reputation.Interoperability (Semantic Technologies);
47. A seller RO in e-Markets
Applications
47
■ Using GoodRelations ontologies to describe a seller and RO ontology to
describe its reputation
Criterion 1
Criterion 2
Reputation.Interoperability (Semantic Technologies);
48. Usage Control in E-Markets
Applications
48
■ Using ROs for decisions during runtime allows revoking participants due to
their former behavior
■ Security settings is one of the RO criteria
Reputation.Interoperability (Semantic Technologies);
49. Rule-based Reputation Systems
Applications
49
■ Using Rule Responder (Multi Agent Reasoning system) to deploy distributed
rule inference services
■ Agents/services communicate reputation objects or specific measures in
them
■ Reputation values used in the agent’s rule logic,
o e.g. deciding on a seller based on delivery method and review
Reputation.Interoperability (Semantic Technologies);
50. Cloud Provider Selection
Applications
50
■ Selecting cloud providers based on their reputation & a consumer preference
list
■ Reducing the risks by selecting reputable SPs
From the detailed reputation
profile,
□ cross reference the quality
parameters requested by the
consumer and the
performance parameters
extracted from the providers’
reputation objects.
Reputation.Interoperability (Semantic Technologies);
51. Personalized News Network
38
■ Real Experience
We want to stay informed with real trusted news
Online Social Network was the answer
■ 2 motivating facts:
Delayed news in the mainstream media
Fabricated news
Reputation.Interoperability (Semantic Technologies);
52. Personalized News Network
38
■ Real Experience
We want to stay informed with real trusted news
Online Social Network was the answer
■ 2 motivating facts:
Delayed news in the mainstream media
Fabricated news
Reputation.Interoperability (Semantic Technologies);
53. Personalized News Network
38
■ Real Experience
We want to stay informed with real trusted news
Online Social Network was the answer
■ 2 motivating facts:
Delayed news in the mainstream media
Fabricated news
Reputation.Interoperability (Semantic Technologies);
54. Personalized News Network
38
■ Real Experience
We want to stay informed with real trusted news
Online Social Network was the answer
■ 2 motivating facts:
Delayed news in the mainstream media
Fabricated news
■ Filtered
□ Trends
□ Trusted networks
Reputation.Interoperability (Semantic Technologies);
-So the targets of reputation statements can be…. As you can see it can be “I like” button, or “Stars” in another community-- Context: is a category in which this reputation is earned.A high ranking for a user of Yahoo! Chess doesn't really tell you whether you should buy something from that user on eBay, but it might tell you something about how committed the user is to board gaming tournaments.
-So the targets of reputation statements can be…. As you can see it can be “I like” button, or “Stars” in another community-- Context: is a category in which this reputation is earned.A high ranking for a user of Yahoo! Chess doesn't really tell you whether you should buy something from that user on eBay, but it might tell you something about how committed the user is to board gaming tournaments.
no obvious distinction -at the rating page- of what exactly being rated or reviewed.
no obvious distinction -at the rating page- of what exactly being rated or reviewed.
no obvious distinction -at the rating page- of what exactly being rated or reviewed.
no obvious distinction -at the rating page- of what exactly being rated or reviewed.
no obvious distinction -at the rating page- of what exactly being rated or reviewed.
Objectiveshow users perceive reputationDoes users use more information than the usual stars and reviews? Which attributes do they actually use to decide on a seller?compare between ratings representations does detailed-ratings style confirm with the social notion more than single-ratings style? frequency and size of cooperation between users
Objectiveshow users perceive reputationDoes users use more information than the usual stars and reviews? Which attributes do they actually use to decide on a seller?compare between ratings representations does detailed-ratings style confirm with the social notion more than single-ratings style? frequency and size of cooperation between users
The form gave them the opportunity to express their perception of rating on different levels.
Excluding context from reputation value, this can be illustrated with an e-market example, where a buyer wants to buy a TV set frpm an e-shop, the seller hands the product to a delivery service, which delays the package.As a result the customer is not satisfied and is giving the seller a bad reviewRating used books is the rating for the book itself -> the user liked what he reador the quality of the book ->was new and good printing) or the service provided by Amazon for example -> offering the book, price, delivery, payment method, etc.Such problem raised legal hassles: eBay (California, Grace vs. eBay) Amazon (cases in UK and USA)reason: rating ambiguity
We have done an analysis on some of communities that use Reputation and we found out that each one has:…slide Slide - In that sense, reputation was modeled in a simple way. Although some of these models are based on complicated mathematical calculations, they still do not reflect the real cognitive nature of reputation because they do not represent all the parameters that affect it.
We have done an analysis on some of communities that use Reputation and we found out that each one has:…slide Slide - In that sense, reputation was modeled in a simple way. Although some of these models are based on complicated mathematical calculations, they still do not reflect the real cognitive nature of reputation because they do not represent all the parameters that affect it.
We have done an analysis on some of communities that use Reputation and we found out that each one has:…slide Slide - In that sense, reputation was modeled in a simple way. Although some of these models are based on complicated mathematical calculations, they still do not reflect the real cognitive nature of reputation because they do not represent all the parameters that affect it.
- Following the methodology of Gruninger for ontology development we construct a set of comp. questions- Competency Questions: These questions act as requirements in the form of queries that an ontology should be able to answer.
In our previous work and in this one, we show our data model that focus on facilitating the standardization of reputation information: Reputation Object model The model is based on the idea of:Reputation is the notion of profiling an entity’s performance In this model we: Uses more information about the domain and the contextsand relevant quality criteria in which a reputation can be earnedUsing this information, reputation is represented differentlyas a developed reputationobjectThe Reputation Object profiles an entity’s performance and has information aboutContexts in the domain in which a reputation can be earnedRatings values given (or reviews or opinions) Computation reputation functions that is used to aggregate the ratingsHow this rating is collected
- The model structure (in the Ontology Figure) contains a description of how this value is collected (e.g. by community ratings or moni- toring service), the computation function (for this criterion) used to aggregate the values each time a new one is entered, and a history list (previous values dated back to a certain time slot). Our model describes a more complex, yet easy to comprehend, reputation representation.
- The model structure (in the Ontology Figure) contains a description of how this value is collected (e.g. by community ratings or moni- toring service), the computation function (for this criterion) used to aggregate the values each time a new one is entered, and a history list (previous values dated back to a certain time slot). Our model describes a more complex, yet easy to comprehend, reputation representation.
- The model structure (in the Ontology Figure) contains a description of how this value is collected (e.g. by community ratings or moni- toring service), the computation function (for this criterion) used to aggregate the values each time a new one is entered, and a history list (previous values dated back to a certain time slot). Our model describes a more complex, yet easy to comprehend, reputation representation.
- The model structure (in the Ontology Figure) contains a description of how this value is collected (e.g. by community ratings or moni- toring service), the computation function (for this criterion) used to aggregate the values each time a new one is entered, and a history list (previous values dated back to a certain time slot). Our model describes a more complex, yet easy to comprehend, reputation representation.
In RDF everything is a resource “classes” are also resources, but……they are also a collection of possible resources (i.e., “individuals”)“fiction”, “novel”, …Relationships are defined among classes and resources:“typing”: an individual belongs to a specific class “«The Glass Palace» is a novel”to be more precise: “«http://.../000651409X» is a novel”“subclassing”: all instances of one are also the instances of the other (“every novel is a fiction”)Linked open data: data are linked from one source to the other by defining relations between them, are self-describing and open -> enables the discovery of new data sources
How to get RDF data: GRDL, RDFa, microformatsBy adding some “meta” information, the same source can be reused for, eg, data integration, better mashups, etctypical example: your personal information, like address, should be readable for humans andprocessable by machinesTwo solutions have emerged:extract the structure from the page and convert the content into RDFadd RDF statements directly into XHTML via RDFaRDFa extends (X)HTML a bit by:defining general attributes to add metadata to any elements provides an almost complete “serialization” of RDF in XHTML
- For declarative processing of the semantic reputation objects we make use of rules