Research On Person-event Relation Extraction Oriented Interactive Question Answering

In recent years, the data produced by the Internet have expended very fast. Under the situation of such a huge data scale, how to get the information people need accurately has become more and more urgent. How to return the information to the user accurately and smartly becomes a hot research point in recent years. The question answering system can satisfy this need to some degree. The traditional question answering system`s knowledge base is organized by the pair of the question and answer. It searches the answer of the question by the inverted index retrieval technique. The pair of question and answer source becomes the bottleneck of this kind of system. So the question answering system based on the entity and attribute become a popular structure. Especially, it is suitable for the field of person-relation extraction. Most of the research does not specify what the person-relation is and what the events happened between them. According to this phenomenon, this paper puts forward the extraction method for the person-relation, extracting the person-relation by the event dimension.This research has four main contents,(1)Words extraction of the person-relation,(2)Person-relation extraction based on weakly supervised learning,(3)Person-relation extraction based on semi supervised learning,(4)Construction of question answering system for the person-relation. First of all, the subject did some analysis on the small scale of data to construct the initial heuristic rules, and built a classification system. Then it used the classification system to label the train set, which was used to train a model to extract person-relation. But weakly supervised method had the problem of inaccuracy. The semi supervised method was put forward for this phenomenon. The semi supervised method firstly needed a small labeling set, and then used the co-training algorithm to extend training set. The extended training set could be used to train a new model to extract the person-relation. By this way, it could solve the lack of training set to some degree. Finally, it constructed the knowledge base according to the extracted content, and combined it with the question answering system to build a question answering system.The subject achieved a person-relation extracting method which oriented to huge scale of data, and a scheme that combined the person-relation and question answering system. The experimental results showed that the method could extract person-relation from the web page text efficiently, and the question answering system that combined with the person-relation knowledge could work well.