A Chinese Entity Relation Extraction Method Based On Distant Supervision

The expression of entity relationship is a kind of connection between two entities.It plays a very important role in automatic question answering,information retrieval and other fields.The relationship extraction task is to extract this connection accurately and quickly from a large number of unstructured or semi structured data and improve the utilization of information.At the beginning of the Web2.0 era,the network data increased exponentially,including more valuable knowledge and valuable noise,which put forward higher requirements for the relation extraction task,and the research on relation extraction has also been paid more and more attention by experts and scholars.At present,the mature research method of relation extraction needs to determine the field of research,and the work of marking training corpus,specifying the type of relation,selecting features,training classifier and so on need manual processing,and many manual operations are time-consuming and laborious.Accordingly,this paper proposes a relation extraction method based on distant supervision and reinforcement learning to solve the above problems.The research work includes the following three aspects:First of all,we use idea of distant monitoring to extract redundant information from different knowledge bases,corpus annotation,reduce human consumption.At the same time,in view of the noise problem of automatic tagging,a denoising algorithm based on lexical semantic similarity algorithm is proposed to improve the extraction precision.Then,using the features commonly used in the study of relational extraction in recent years as the initial feature,random forest algorithm is used to test feature characterization,and the characteristics of weak characterization are filtered out..Finally,we use Adaboost reinforcement learning method to construct multiple weak classifiers,train the final strong classifier and extract the entity relationship.The method proposed in this paper makes the performance of relation extraction be greatly improved.In the test experiment,the accuracy rate of 71.6% and the recall rate of 76.8% were obtained.