Research On Application Of Dependency Tree In Chinese Question Classification

Question classification is a key technology in the automatic question answering system.It can be employed to effectively reduce the search space of the answer and improve the accuracy and efficiency of the system.For the Chinese question,it is more difficult to extract classification features due to the Chinese language parataxis,non-tenses,word order flexibility.This dissertation is mainly discussed the Chinese question classification feature extraction from the dependency tree structure of questions.The main works of this dissertation are as follows:(1)The headword is an important feature in question classification.Aiming at the problem that the recognition precision of headword is not high,this dissertation is proposed a method of mining frequent subtree pattern from the dependency trees of question set and reducing the pattern to obtain relationship between the headword and the local structural features of dependency trees.Based on this,the method of combining the reduced frequent subtree pattern with the bidirectional gated recurrent unit(BiGRU)to recognite the headword is designed,that is,BiGRU is used to recognite the headword of question set first;then the high confidence frequent subtree rules are selected to correct the initial recognition results.The experimental results show that this method can effectively improve the recognition precision of headword.(2)By analyzing the dependency tree of Chinese question,this paper is found that there are differences in the importance of different words in the judgment of answer categories.There is a relationship between the word dependency distance and its importance.The existing question classification models based on deep learning does not introduce this information into the classification process.Therefore,this dissertation is designed a attention mechanism that combines the dependency distance feature and semantic feature,which can be used to calculate the role of each word in the question classification.Based on this attention mechanism and bidirectional GRU network,a Chinese question classification model is proposed which can highlight the role of key words in question classification.The experimental results show that the model effectively improves the accuracy of question classification.