Dissertation > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Retrieval machine

Establishment and Update of Similar Users’ Cluster in Personalized Information Retrieval

Author ZhengWei
Tutor ZhangYu
School Harbin Institute of Technology
Course Computer Science and Technology
Keywords Personalized Information Retrieval Collaborative Filtering User Clustering Relevance Model
CLC TP391.3
Type Master's thesis
Year 2008
Downloads 75
Quotes 1
Download Dissertation

An important characteristic of next generation search engine is personalization. Personalized Information Retrieval (PIR) focuses on users. It captures users’interest in different kinds (explicit, implicit interest and interest of similar users). These information of users are integrated and used to improve the result of information retrieval system.The establishment and update of similar users’cluster is an important subtask of personalized information retrieval. Its task is to establish clusters of similar users by analyzing users’retrieval and browsing history. The clusters will also be updated with the change of users’information and retrieval areas. The problem in this task is the lack of task’s division and standard evaluation dataset. Therefore, this paper defines four subtasks of PIR, which include the establishment and update of similar users’cluster. The establishment standard evaluation dataset makes it possible to evaluate and compare the systems of user clustering.The data sparseness limits the performance of user clustering because web pages rated by different users are rare. Therefore, the research of this paper focuses on solving the problem of data sparseness. This paper proposes a user clustering method based on relevance model. It uses users’data in similar domains to expand the data of users in current domain by relevance model. The users’clusters will also be updated with the change of retrieval domains. The retrieval information and labeled answers of users are used to establish the experimental dataset. The evaluation matrix includes false alarm rate, miss alarm rate and cost of detection. In the experiment, user clustering based on relevance model improves the result of baseline system by 7.12%. This result proves that the proposed algorithm can alleviate the problem of data sparseness. What’s more, mining users’interest by its cluster can decrease the false information in users’models and improve the result of precision of user clustering.

Related Dissertations
More Dissertations