Research On Intelligent Question Answering System Based On Plan Recognition

Global data has shown explosive growth in recent years.Although this has provided people with more abundant and comprehensive information resources,it has also caused difficulties in information retrieval.As the main means for people to obtain network information,search engines still have some deficiencies in the new era.For example,there are many noisy data,repeated return results,and not simple enough.In order to solve this problem,question answering system came into being,and have been widely developed and applied by its fast,efficient and accurate features.However,in practical applications,problems such as incomplete input sentences and unclear semantic expressions have a great impact on the performance of the question answering system.Therefore,the idea of plan recognition was used to design a question answering system in this paper.Plan recognition is one of the research hotspots in the field of artificial intelligence.Its characteristic is that it can derive its possible targets based on the observed agent fragments and trivial actions.Based on this characteristic of plan recognition,a question matching method based on weighted target graph was proposed and a question answering system based on plan recognition was designed in this paper.The main work done in this article is as follows:(1)The natural language processing technology used in the question answering system is studied,and the improved S-TF-IDF algorithm is proposed for the problem that the TF-IDF calculation index is single and the synonyms are not considered.S-TF-IDF uses the word frequency,part-of-speech,length,and position information for weight calculation,and uses "Synonym Cilin" as a synonym metric.When calculating the inverse document frequency,all documents containing feature words or their synonyms are counted.Finally,the S-TF-IDF algorithm is applied to Baidu baike word similarity calculation,and verified by the Words-240 data set.The results show that the Baidu baike word similarity calculation based on S-TF-IDF can effectively improve the accuracy of the algorithm and recall rate.(2)In-depth study of some common methods,classifications and applications of plan recognition,through comparative analysis of the advantages and disadvantages of various methods and applicable fields,put forward a problem matching method based on weighted target graph.In this method,the weighted target graph is composed of nodes and weighted edges,where nodes are questions entered by the user,questions in the knowledge base and their constituent words,and the weighted boundary value(degree of support)is word weight and word similarity.Experiments show that this method can effectively realize the question matching in the question answering system.(3)A question answering system based on plan recognition was designed and implemented,and the system was verified to have good performance through experiments.In addition,a scoring function is added to the system to allow users to rate the answers given by the system.When the score is below a certain threshold,the administrator can view the data and modify it.