Research On User Preference Analysis Techniques In Personalized Information Retrieval

User preference analysis techniques in personalized information retrieval is animportant issue of IR area. The key task is to obtain the information which meets user’squery tendency and objective law through mining use’s preference by analyzing user’sbehavior on the given queries. For this purpose, we carry out research to analyze user’squery target and explore novel features evaluating the quality of information acquisition.The main contents of this thesis are summarized as follows:Firstly, we carry out research on information retrieval target analysis based on querydividing. User query target identification is a fundament task of the quality of informationacquisition analysis, aiming to mine user’s query intent. According to the internal structureand linguistic phenomenon of query, we propose a novel approach to identify query target.Specifically, firstly, query is divided into query objects and query intents. Secondly, queryintents are clustered to form search intent sets. Then, query object and query intent sets arecombined to represent user’s query target. Finally, we recommend the corresponding querythrough automatically estimate query target. This approach focuses on investigating andpromoting the quality of building query automatically. Thus, by optimizing the descriptionof query, we can improve the quality of query.Secondly, by analyzing user behavior features in the process of search, we propose anovel satisfaction measurement method based on energy consumption of MouseMovement. Firstly, we analyze the relationship between user satisfaction and behaviorfeatures in the process of search. Then, quantitative model is built to simulate user behavior.Using this method, we can compute the energy consumption during mouse movement.Finally, based on energy consumption, user satisfaction in the process of search is inferred,to quantize satisfaction and predict the degree of satisfaction. This method focuses on studying approaches which are used to measure user satisfaction. Using this method, wecan optimize the quality of information retrieval system feedback.Thirdly, we propose a research on discovery and illustration on novel optimal retrievalresult. We develop a labeling platform based on Google search engine. Based-on thelarge-scale data, we attempt to prove that originally correct search results is optimal.Correctness is an inherent objective attribute which meets objective law which neverchanges with subjective cognition of users. On the basis, we propose a metric aboutcorrectness and validate the performance of the reranking algorithm based on user behaviorby using the metric. This work focuses on the research on how the correct informationinfluences the user preference in the process of information retrieval. Experiment showsthat correct retrieval results have bigger positive impact on the user’s cognition. Theresearch can help improve the sorting strategy of retrieval system.In summary, for the user preferences analysis technique in personalized informationretrieval task, we carry out research from three aspects, aims to provide a high quality ofinformation retrieval feedback and return the search results more in line with userpreferences and cognition.