Application of LSI Based On Relevance Feedback

When we used LSI in EEAS (expert evaluation assistant system), we faced with such an issue: the number of dimensions of the semantic space needed must be determined for each collection. The dimensions of semantic space influence the results Seriously. But the users usually do not have the knowledge to adjust the dimensions. So we must provide a way for the user to adjust the dimensions of semantic space dimension for every topic. We propose such an approach which uses the relevance feedback of the user to get the statistic features of our target documents to modify the semantic space. This paper give a idea to complete of this approach by using expectation of the features values of the document vectors. The experiment proves that the retrieval result is improved.