Study Of Collaborative Web Search Based On CBR

With the rapid development of Internet technology, and Web sites as well as the increasing number of users, WWW (World Wide Web) has become a large-scale complex network of resources and users, and played an important role for us to transfer and share information. However, mass of information has led to "information overload", so users often get lost in the sea of information, increasingly have difficulties to efficiently access specific information they need. For this reason, in recent years, Web search has been paid more and more attention and taken as an importance problem in field of computer science. Among them, personalized Web search, social search, collaborative Web search are the most talked about several major issues.In general, the user generated search history, personal favorites and other information are extremely important. In addition, at the era of Web 2.0, a large number of other forms of user search experience has emerged in the Web, such as various types of tags, voting, comments and so on. By the way, through the reasonalbe percceiving, organizing and using them, it may further improve search quality of users. However, on the one hand, the expert, who has domain knowledge or strong search skills, often can quickly find satisfactory information and produce high quality search experiences which are the most valuable resources; on the other hand, we make use of case-based reasoning (Case-Based Reasoning, CBR) to process the user search experience as an experience information further. To this end, the author of this dissertation explores the acquisition, representation, organization, utilization, and abnormal data processing of users search experience based on CBR in the process of cooperating with experts for search. Specifically, this dissertation has completed the following tasks:(1) In reviewing the current status and development of Web search technology problems, the summary and analysis of the current status of collaborative Web search studies, the study points out the problems and deficiencies, and summarized the focus of future research. On this basis, firstly, through analyzing the development of traditional Web search engines, summarized two methods to achieve collaborative Web search, namely, embedded or plug-in traditional Web search engines for search; Secondly, by analyzing the impact of domain knowledge for Web search, we point out that we can utilize experts and their search experience to further improve search quality of users and it is a good method of collaborative Web search between users. At this basic assumption, we proposed a collaborative Web search model—a model of CBR-based utilizing experts search experience, and introduced its solution and key techniques; Thirdly, we explored a way to implement collaborative Web search utilizing search engine - browser plug-ins - recommendation engine model, and introduced a recommendation system architecture based on CBR(2) The acquisition, representation, organization of users search experience is a core problem to research collaborative Web search. After summarizing the type of user search experience, we firstly analyzed a template-based method to extract return results of search engine; Secondly, after reviewing some problems about case and its represtation, we proposed a query-based and a item-based improved method to represent a user search experience as a case; Finally, we explored the organization of the user search experience, and proposed a community-based method and a method with multi-case bases.(3) We have explored the utilization of the user search experience. Atfer discussing the ways of utilizing the user search experience, we firstly discussed the user modeling using the user search experience and introducd a "keywords weight" based and a "semantic relationship" based model for modeling user; Secondly, we proposed a topic filtering method to identify the experts and their search experience, then introduced a "recommendation frequency" based and a "hierarchical user model" based method to identify the experts search experience; Finally, after summarizing some methods to retrievel and recommend the experts search experience, we intodued a recommendation strategies and an optimized page rank utilizing the experts search experience in ExpertRec, which is a plug-ins collaborative Web search system, and made experiments to prove that our model is effective in order to further improve the search quality of users.(4) Another core problem to research collaborative Web search is anomaly detection problem about the user search experience. By analyzing unsolved problems in processing the user search experience, we points out that real-time anomaly detection approach can be used to process abnormal data in the process of collection and maintenance of the user search experience. Therefore, after the introduction of anomaly detection and peculiarity factor (PF), we proposed sampled peculiarity factor (SPF) and a SPF-based anomaly detection algorithm in order to meet real-time anomaly detection, and then we conducted experiments. Experimental results showed that the SPF could replace other PFs to enhance the performance of the algorithm for real-time anomaly detection. In addition, we proposed a method to recommend community automatically and an online solution to detect outliers of user search experience.(5) Prototype systems are developed. Based on the above-proposed models, algorithms, methods, we have implemented the Whitesun embedded prototype system and ExpertRec plug-in prototype system for collaborative Web search. These systems proved the correctness of the relevant models and methods, to further explore the problems of collaborative Web search as a reference.