Transcript Trust Web Service

Jennifer Golbeck, Bijan Parsia, James Hendler
Trust Networks on the Semantic Web
Trust Network
Ratings and Ontology
Using our Trust Ontology (an extension of FOAF), users
create trust ratings for their acquaintances
•Individuals are identified by email address
•Ratings are made on a scale of 1-9
•Ratings can be made in general, or with respect to a
specific concept
Calculating Trust
Trust from the source to sink is calculated using a BFS
based algorithm.
<Person rdf:ID="Bob">
<mbox rdf:resource="mailto:[email protected]"/>
where tij is the trust between node i and node j
n is the number of nodes in the graph
<trustsSomewhat rdf:resource=“#Joe”/>
<trustsHighlyRe>
<TrustsRegarding>
<trustsPerson rdf:resource="#Dan"/>
<trustsOnSubject
rdf:resource="http://example.com/ont#AutoRepair"/>
</TrustsRegarding>
</trustsHighlyRe>
</Person>
Trust Web Service
Via a web service that hooks in to a spidered graph of the
network, users can
1. Write their own applications that use the trust network
2. Create their own trust metrics based on the properties
of the graph
3. Submit class files with their own code to be used in
existing applications that use the service
TrustBot
Using the web service as a back-end, we have created
TrustBot. This is an IRC bot running on irc.freenode.net what
allows users to access and control the in-memory graph. They
can query, get help and explanations, and reload the graph
after making changes to their local files
1. If there is a direct edge connecting source to sink
that value is used as the trust value
2. If there is no direct edge
1. Find all paths from source to sink (
2. Find the max flow on each of those paths
3. Use a weighted average of the maximum trust
capacity of each path as given in the formula
above
TrustMail
Built on Mozilla Messenger, TrustMail displays the sender’s
trust rating next to each message.
This provides something akin to the opposite of a spam filter: a
relevance rating that allows users to sort messages to see
those from the most trusted individuals first.
Consider two research groups collaborating on a project. The
professors in charge of each group know one another, and
each knows her own graduate students, but neither knows the
students from the other group. With TrustMail, messages from
students
in one group to members of the other group will be
recognizable and highly rated because of the intermediate link
between professors.
Calculated trust values can be overridden by simply creating a
personal rating for a given individual. This is always used in
place of the network rating.