Rise 2014 st requier

Post on 18-Dec-2014

43 views 0 download

description

 

Transcript of Rise 2014 st requier

Contextual Web Service

Suggest a query and a search engine

Aurélien Saint Requier, LITIS Rouen

Table of contents

� What is search on the Internet?What is search on the Internet?What is search on the Internet?What is search on the Internet?� Our proposal

● Modelize user interests● Suggest pairs of conceptual query and search engine

� Evaluation� Conclusion

Search the web

1.Select a search engine

2.Formulate your need3.Hope to find a relevant result in the result list

Select a search engine

Formulate a need

➔Users express query in few words (2-3)➔Between 20% and 30% of queries contain a

single word➔Users often reformulate their queries➔For novice users, the formulation of queries

is a difficult task➔For a complex information task, users

formulate more and longer queries in a same

session

Problems

Analyze results

➔Users show interest on the first and second

results➔Users do not go beyond the first result page➔For a complex information task, users spend

more time on the result page

Proposal

Goal: ➔Help the user to formulate his need and

suggest a search engine according to his

need

How:➔Get interests of users➔Suggest a pair composed of a conceptual

query and a search engine

Get interests of a user

➔Use a weighted conceptual user profile: ● a long term profile = knowledge of the user● a short term profile = context of the search

➔Corpus:● LP : web pages mark as favorite, saved web pages and

documents provide by users to avoid cold start problem.● SP : all visited web pages

Get interests of a user

➔Represent an interest by a DBPedia category➔Weight is equal to the probability of

occurrence of the concept in the corpus

Technical issues to profile construction

●Use Zemanta to extract DBPedia concepts

from text●Encode profile in Attention Profiling Markup

Language (APML)●Develop a Firefox extension to track user web

activities

From concepts to thematic profile

Profile fusion

●Function

Profile fusion

●Result

Suggest pairs of conceptual query and search engine

Process :

1.Get keyword user query2.Translate keyword query in conceptual

queries

3.Match conceptual queries with search

engines

4.Suggest pairs of conceptual query and search engines

Translate keyword query to conceptual query

Determine relevant search engine to the conceptual query (1)

●Define a semantic description of a Search

Engine : <SearchEngine> <Id>e018</Id> <Name>LastFM</Name> <Url>http://www.lastfm.fr/music/</Url> <Description>Last.fm is a music recommendation service. </Description> <Specialized>true</Specialized> <Thematic> <Subject> … </Subject> </Thematic> <ContentType> <Type>http://dbpedia.org/ontology/Band</Type> <Type>http://dbpedia.org/ontology/Single</Type> <Type>http://dbpedia.org/ontology/MusicalWork</Type> <Type>http://dbpedia.org/ontology/Album</Type> </ContentType> <Popularity>5</Popularity> <Searchable>true</Searchable></SearchEngine>

Determine relevant search engine to the conceptual query (2)

●Use a similarity measure based on the types

and the categories of conceptual queries

Finaly

Finaly

Experimental system

●Based on WebLab and Liferay● Use Web services and portlet

Experimental system

●Based on WebLab and Liferay● Use Web services and portlet

Conclusion

� Modelize user interests by a thematic profile� Use this thematic profile to translate keyword queries into conceptual queries

� Suggest pairs of conceptual query and search engine

� Upcoming evaluation● Compare our approach of (conceptual query / search

engine) pair suggestion to (google suggestion / google

search engine) pair suggestion