An Improved Method of Interrogative Sentence Similarity Compute and Application in Q&A System

Sentence similarity compute is an important part in question answering system based on frequency asking questions. The accuracy of the existing sentence similarity algorithm needs to be improved, so this paper presents a revised question similarity compute method. We combine the word order feature with vector space model algorithm. When we use the VSM to compute the question similarity, we propose a method of extracting topic and focus. The difference between this method and the traditional approach is that this method doesn´t depend on interrogative. Topic and focus can reflect the purport of a question. By identifying it can we better understand the question, and consider the impact of the topic and focus in questions similarity compute. At last, by designing experiment to compare it with other methods, the experiment shows that this method can improve the accuracy.