04.11.2014 Views

elektronická verzia publikácie - FIIT STU - Slovenská technická ...

elektronická verzia publikácie - FIIT STU - Slovenská technická ...

elektronická verzia publikácie - FIIT STU - Slovenská technická ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

User Modeling for Personalized Web-Based Systems 245<br />

The approach uses the co-occurrence of names in close proximity in any documents<br />

publicly available on the Web as an evidence of a direct relationship. Such sources include:<br />

– Links found on home pages<br />

– Lists of co-authors in technical papers and citations of papers<br />

– Exchanges between individuals recorded in netnews archives<br />

– Organization charts (such as for university departments)<br />

The network model is constructed incrementally. When a user first logs into the system, it<br />

uses a general search engine to retrieve Web documents that mention him or her. The names<br />

of other individuals are extracted from the documents. Authors claim that they achieve a high<br />

degree of accuracy (better than 90%), using information extraction techniques, however, they<br />

do not provide additional details. The process is applied recursively for one or two levels,<br />

and the result merged into the global network model.<br />

Jumping Connections. Paper [48] presents an example of Mining and Exploiting Structure<br />

approach. It describes a study algorithms for recommender systems from the perspective<br />

of the combinations of people and artifacts that they bring together. They named the approach<br />

jumping connections.<br />

A recommender dataset R consists of the ratings (e.g., of movies) by a group of people.<br />

It can be represented as a bipartite graph G =(P ⋃ M,E), where P is the set of people, M<br />

is the set of items (movies) and the edges in E represent the ratings. Let N P and N M be<br />

a number of people/items respectively.<br />

A jump is a function J : R ↦→ S; S ⊆ P × P that takes as input a recommender dataset<br />

R and returns a set of (unordered) pairs of elements of P . This means that the two nodes<br />

described in a given pair can be reached from one another by a single jump. Obviously,<br />

jumps are made using the items in the set M. Authors defined the skip jump, which connects<br />

two members in P if they have at least one movie in common.<br />

A jump induces a graph called a social network graph of a recommender dataset R. Itis<br />

a unipartite undirected graph G S =(P, E S ), where the edges are given by E S = J(R). The<br />

graph could be disconnected based on the strictness of the jump function.<br />

Extracting Social Networks from Communication Evidence Paper [20] proposes an<br />

end-to-end system that extracts a user’s social network and its members’ contact information<br />

given the user’s email inbox. Social links are created by extracting mentions of people from<br />

Web pages and creating a link between the owner of the page and the extracted person. The<br />

system is called recursively on each newly extracted people, which result in a large “friends<br />

of friends of friends” network.<br />

The process of social network extraction is depicted in the Figure 8-11. The bootstrapping<br />

set of names is extracted from email headers in user’s inbox. Name coreference resolves<br />

multiple mentions of the same person in different format. Subsequently, the system attempts<br />

to find person’s homepage by submitting queries based on the person’s name and likely<br />

domain to Google search engine. The results are filtered according to URL features and

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!