Multi-Agent-Based Information Retrieval System Using Information Scent in Query Log Mining for Effective Web Search

Introduction

Information on the web is huge and the retrieval of web documents relevant to a user information need is a big challenge. Search engines are used for web information retrieval and retrieves large collection of documents for a given query. The user query issued for web search contains few keywords therefore difficult to infer information need of the user. This is because of small size user query, the search engine retrieves large collection of documents out of which few are relevant. There is the need for personalizing the web search to the information need of the user by retrieving more and more relevant documents in search results. The data mining techniques have been applied to web usage data to get knowledge of web user’s browsing pattern for effective information retrieval. The search engines store the search history of web users in query log which contains the search query and its associated clicked URLs (Broder, 2002; Jansen & Spink, 2006). There are four broad categorization of personalization techniques based on various factors such as context, behavior, location and history. In Saravanakumar and Deepa (2011), WordNet 3.1 is used to find different synonyms of query keywords to form alternate queries. In Gulati and Sharma (2010), the context of query is applied to retrieve the document containing the synonyms of user query which is otherwise missed in traditional search system. Personalization system based on capturing the behavior of user such as short term and long term interest has been proposed in Gao, Xi, and Im (2013). History based personalization has been proposed in Rastafari and Shamsuddin (2010), which uses the factors such as user query logs history, pages visited by user, action performed on that page, time spent on that page etc. Google generates the customized search result based on user search activity linked to cookie saved in browser. In Sethi and Dixit (2015), the personalized search system is proposed based on user previous history. The user behavior on the web like surfing pattern, search queries submitted and some explicitly collected information is used to re–rank the search results. In Li, Yang and Kitsuregawa (2009) and Morris, Teevan and Bush (2008), search engine search results are re-ranked based on user preferences. In Carman et al. (2010) and Lv et al. (2006), user implicit feedback is used to fine-tune the search results. In Stermsek, Strembeck, and Neumann (2007) and Ghosh and Dekhil (2009), web search personalization is based on user profile built from different resources, but it ignores the vocabulary problem and involves users in maintaining the profile. In location based personalization, a model for web personalization is proposed based on current location of users in (Mokbel et al., 2011). In Leung, Lee and Lee (2010); Bouidghaghen, Tamine, and Boughanem (2011) and Weber and Castillo (2010), personalized search techniques are proposed based on user’s geographical location. In Moawad et al. (2012), multiagent system is proposed that uses profile and WordNet ontology for web search personalization. In Lieberman (1995) and Turner et al. (2001), software agent track the user browsing and generates the user profile in order to assist the user web search. In Menczer (2003) and Wei, Moreau and Jennings (2003), auction protocol and reward techniques are used for collaboration among agents in order to answer single and multiple queries. In Blanzieri et al. (2001); Chau et al. (2003) and Yu and Singh (2002), personal agents collaborate with another in order to improve the user browsing where users were asked to specify the areas of interest or analyze the similar search results. In Birukov, Blanzieri, and Giorgini (2005), agent system is used to generate the personalized search results using collaborative approach and generates the suggestions from the members of the community in addition to search engine results. In comparative analysis of personalized search techniques such as context, behaviour, location and history oriented. In context oriented technique synonyms and polysemy of keywords are used for context disambiguation with no user involvement. In behavior oriented, the user interest area is tracked using explicit or implicit involvement of user. In location oriented, geographic location is inferred automatically by the system but sometimes generates the misleading results. In history oriented, the user's clicks, action and time spent during web surfing is tracked implicitly and provides the relevant results. Google generates the personalized search results based on context, behavior, location and history. Yahoo uses the behavior approach and Bing generates the results based on user behavior and history (Mittal & Sethi, 2015). There are issues related to personalized search techniques such as there are many techniques but all are not applicable to all users at the same time. It is realized that hybrid of personalized search techniques can generate better results than any individual techniques. Thus the system proposed in this chapter provides the hybrid solution to various issues associated with effective information retrieval in one system. It uses query expansion and related queries recommendation for context of user input query and builds the user profile based on search behavior. At the same time web structure mining is used to identify the high quality content(Authorities) and resource web pages(Hubs) for web page recommendations and personalize the web search effectively. This chapter describes multi agent based information retrieval system in Bedi and Chawla (2010), which personalizes the web search based on Hubs & Authorities recommendations using information scent and HITS (Hyper linked Induced Topic search) in query sessions mining. The information scent is used in query log mining to generate the clusters of similar information need web user search sessions. The clusters are processed using HITS to generate hubs and authorities. Hubs are high quality resource web pages which point to many good authorities and good authorities are high quality content web pages pointed by many good hub web pages. During web search, the cluster is selected to generate the hubs and authorities for recommendation to the users who search with similar intent as that of user sessions in clusters. The recommendation of hubs and authorities according to user search pattern on the web continues till search is personalized to the information need of the user. This multi agent Information retrieval system contains user agent, interface agent, related queries generator agent, query sessions miner agent using information scent and high scent hubs and authorities recommender agent. The entire processing is divided into two phases: