Research On Key Techniques Based User Preference Analysis In Personal Information Retrieval

Personal Information Retrieval aims to obtain the information which meets user 's query intention through user preference analysis and detection. The challenge is to recognize individual behavior characters during the process of information retrieval. Especially, it is very difficult to be aware of the user preference and the personal query modeling. The main contents of this thesis include: Satisfaction Prediction Oriented Quantitative Mouse Movement AnalysisIn current study on Information Retrieval, the determination of quality of knowledge learning and precision of information acquisition heavily depend on the relevance between user 's need and information. However, the quantity normally is incapable of reflecting the satisfactoriness degree of users to pseudo information feedback. Compared to the scalable relevance degree among multi- medias, such as linguistic texts, images, audios and videos, the satisfactoriness, which is triggered and driven by human's subjective recognition is not easily reachable. In particular, it is difficult to directly measure the satisfactoriness degree. To solve the problem, the paper proposes a Mouse-Movement- Law based satisfactoriness analysis and measurement method. The method concentrates on trajectory analysis of mouse movement that is driven by human's physical activity. More importantly, the method detects the impetus of mouse movement(i.e., momentum) during the course of sliding, by which it indirectly reflects the activity degree of the mind. The quantity of momentum, accordingly, is favorable for measurement of satisfactoriness toward pseudo information feedback. Experiments show that the method is effective in supporting analysis of user experience in the process of Information Retrieval.Research on Query Intention Boundary DetectionIn general, several queries will be submitted by a user to capture specific query intention. Boundary detection is extremely helpful in decoding query intention. Moreover, it is conductive to query suggestion by identifying the full query intention; it can be also used in the query expansion and user profile construction. Benefited from the advantages mentioned above, we introduce the boundary detection into user query stream analyzing, especially, we propose topic distribution based similarity method, as well as the solution for the stream of query intention including SVM and CRF models. The SVM model usually be used in the task of text categorization, while the C RF is a sequence labeling model which can label the continuous query intent ion. The exprimetal results show that our F1-measure is improved by 2% in comparison on the baseline system.Correlation analysis between Social Network and Query IntentionSocial network contains large amounts of user preference. We annotate large-scale data of user interest, and we emprically found that some query intentions may occur in the process of information retreival in both search engine and social media.This thesis argues that a user 's query intention has a strong correlation with the corresponding social network contents, which focuses on two aspects. First, we analyze the proportion of social network websites in user 's retrieval records, and evaluate the degree of relationship between information retrieval behavior and social network behavior from a statistic perspective. Second, a similarity between the two components of query intent ion, retrieval content and social network content, is calculated employing a topic model to measure the correlation on a ground-truth data of retrieval records indicates that a long-term retrieval procedure is usually accompanied with accessing social network websites. More importantly, the experimental results show that more than 57% of the content from social network has a strong correlation with users' query intention.In summary, for the user preference analysis technique in personal information retrieval task, we carray out research from three aspects, aims to provide a high quality of information retrieval feedback and return the search results more in line with user preferences and cognition.