基于主题的聚类检索技术研究计算机科学与技术专业论文.docxVIP

哈尔滨工业大学工学硕士学位论文 哈尔滨工业大学工学硕士学位论文 - - II - Abstract With the progress of computer and network technology, information distribution and sharing are no longer restricted by time and space. Meanwhile, it brings the problem of information overload, namely, the valuable information and knowledge is relatively deficient while the information resource is in great abundance. As a global information space, not only the Web provides an unprecedented experimental environment and application possibility, but also brings new challenges to information retrieval. Web information retrieval systems such as search engines attempt to solve this problem. However, their effects are far from satisfying for the low precision, hard to use and so on. Therefore, this paper explores the topic on clustered Web information, especially the founded of index, the extracted of salient phrases, the classified and ranked of themes, the displayed of themes, which is a research issue with great significance in theory and wide perspective in application. Speed is the key factor considered in clustering retrieval system, so, the traditional cluster method is not suitable in this condition. A cluster method is introduced based on themes in this paper. This method first extracts salient phrases as candidate cluster theme, then documents are assigned to relevant salient phrases to form candidate clusters, and the final clusters are generated by merging these candidate cluster. This paper focuses on the topic of how to extract the salient phrases of documents, combines the statistical information and semantic information of phrases, considers the properties of POS, phrase’s length, phrase’s frequency, phrase’s position, lexical chain, and uses these properties to train a regression model of SVR (Support Vector Regression) learned from human labeled training data. This paper uses this model to rank and extract the salient phrases of documents. Also this paper uses several