Search Optimization Based On Session Process

With the explosive growth of the Internet information, the search engine plays a crucial role in the network information lookups. Faced with massive amounts of data, traditional search algorithm reveals application limitations. Firstly, keyword-oriented search principle requires users’high ability of formulating queries, which can demonstrate users’information need properly. Secondly, the relevance judgment between short query and massive information results in lower accuracy and recall rate. Finally, the general search algorithm is unable to provide personalized search service. In order to solve problems above, in this paper, we study the search optimization based on session process by treating session information as the main target.Session means a sequence of reformulations along with any user interaction with the retrieved results in service of satisfying an information need, including the click behavior on search results, and the residence time while browsing the webpage. In order to implement search optimization, this paper proposed a session-oriented retrieval model based on Markov Random Field. This retrieval model utilizes session information as the basis. The research of this paper includes following aspects.Firstly, taking Markov Random Field as the theory foundation, we build a session-oriented retrieval model. By the analysis of users’ behavior while searching, we propose a dynamic evolved retrieval model based on timing characteristics of session process.Secondly, we study the term dependences in session retrieval model based on the analysis of linguistic characteristics. In this paper, we propose two session retrieval models based on different term dependence assumptions, one is Fully Independence Session Model (FISM), the other is Sequential Dependence Session Model (SDSM).The impact on session retrieval models’accuracy which brought by term correlation assumptions is studied also.Thirdly, different session information’s influence on retrieval process is studied based on session information classification. In this paper, we divide session information into two categories:historical queriesHQand historical clicked web pagesHc. Based on the session retrieval model theory, we can construct the query elements byE(Qi),E(Ci),E(Qi+Ci) and E(WAFi) methods, which finally achieve the effective integration of different historical information with session retrieval process.Finally, taking Word Activation Force as the theory foundation, we make the query expansion based on session information to study the effectiveness of the session retrieval model, which is integrated with Word Activation Force.In response to the research points above, this paper designs and implements a series of experiments. The experiments’results show that the search optimization target can be achieved by the session-oriented retrieval model we proposed.