Research Of Inquires Model Based On Weighted Grammar Dependence

The technology about processing Chinese information is important in daily life. But the progressing of Chinese semantic similarity algorithms research was slow, which is very important status in Chinese information processing technology. And at present the methods about the Chinese semantic similarity algorithms has some disadvantages more or less. So the Chinese sentence similarity algorithms was explored and researched in this thesis. According to analyzing the characteristics of Chinese sentence similarity algorithms and information inquiry, improving the semantic dependence algorithm, we propose the weighted semantic similarity algorithm in this thesis. It is clear that the algorithm achieved good results in the information query model. The main research contents were as follows:Improving sentence similarity algorithm: Through analyzing the semantic similarity algorithm, the weighted semantic similarity algorithm was proposed. According to incomprehensive problem in the definition of the weight of matching the sentence effectively, we use the results of analyses and add the information of semantic role tagging when judged the weight of matching the sentence. According to the problem of discarding the effective keywords, semantic dependency was weighted, when combining the characteristics of word form and word meaning unifies similarity algorithm. The results showed that the compute accuracy of sentence similarity had been raised using the improved algorithm.Application of the improved algorithm (start sorting): It is a difficulty problem that how to improve the recall and the accuracy ratio simultaneously on information searching. In view of this question, we first proposed one kind of improved sentence similarity algorithm which based on the semantic interdependence degree, and has proven this method feasibility with the example, and then made re-sorting with the improved method for inquiry results, which founded on enhancing the inquiry system recall. Through the similarity computing which are about the original query sentence and top-K document title, the similarity value is the measure of re-sorting, thus enhance the accuracy ratio of the first K returns results. The experiment proved that this method improved system's accuracy ratio.