A Framework For Personalizing Web Search with Multi-Faceted User Profiles

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering


PhD Thesis Defence


Title: "A Framework For Personalizing Web Search with Multi-Faceted User 
Profiles"

By

Mr. Wai-Ting Leung


Abstract

Personalized search is an important means to improve the retrieval 
effectiveness of a search engine, since user queries are normally short 
and ambiguous. Thus, it is hard for a search engine to figure out what the 
users precisely want. Most commercial search engines simply return the 
same set of results to all users who ask the same query. However, 
different users may have different preferences on the result set. Thus, 
personalization is needed in order to rank the results according to a 
user's personal preferences. In this thesis, we develop two methods to 
mine a user's conceptual preferences from search engine clickthrough data, 
and adjust the search result ranking according to the extracted 
preferences to improve the retrieval effectiveness for the user.

We first propose a framework that supports mining a user's conceptual 
preferences from users' clickthrough data resulted from web search. The 
discovered preferences are utilized to adapt a search engine's ranking 
function. In the framework, an extended set of conceptual preferences was 
derived for a user based on the concepts extracted from the search results 
and the clickthrough data. Then, an Ontology-based User Profile (OUP) 
representing the user profile as a concept ontology tree is generated. 
Finally, the OUP is input to a Support Vector Machine (SVM) to learn a 
concept preference vector for adapting a personalized ranking function 
that re-ranks the search results. We confirm that our approach is able to 
improve significantly the retrieval effectiveness for the user.

We then adopt the OUP approach in the area of location-based 
personalization.  We propose a personalized mobile search engine, PMSE, 
that captures the users' preferences in the form of concepts by mining 
their clickthrough data. Due to the importance of location information in 
mobile search, PMSE classifies these concepts into content concepts and 
location concepts. In addition, users' locations (positioned by GPS) are 
used to supplement the location concepts in PMSE. The user preferences are 
organized in an ontology-based, multi-facet user profile, which are used 
to adapt a personalized ranking function for rank adaptation of future 
search results. To characterize the diversity of the concepts associated 
with a query and their relevances to the user's need, four entropies are 
introduced to balance the weights between the content and location facets. 
We also present a detailed client-server architecture of PMSE. In our 
design, the client collects and stores locally the clickthrough data to 
protect privacy, whereas resource consuming tasks such as concept 
extraction, training and reranking are performed at the PMSE server. We 
prototype PMSE on the Google Android platform. Experimental results show 
that PMSE significantly improves retrieval effectiveness comparing to the 
baseline without personalization.


Date:			Wednesday, 25 August 2010

Time:			2:00pm – 4:00pm

Venue:			Room 3501
 			Lifts 25/26

Chairman:		Prof. Chih-Chen Chang (CIVL)

Committee Members:	Prof. Dik-Lun Lee (Supervisor)
 			Prof. Wilfred Ng (Supervisor)
 			Prof. Frederick Lochovsky
 			Prof. Raymond Wong
                         Prof. Bilian Ni Sullivan (MGMT)
                         Prof. Robert Luk (Comp., PolyU)


**** ALL are Welcome ****