Analysis And Research Of Emotional Tendency Based On Entity-Relation Joint Extraction Method

As the rapid development of e-commerce,fine-grained sentiment analysis of online comment data has become a hot topic.At present,fine-grained emotional tendency analysis tasks usually use multi-label classification technology as a solution,but this solution can't obtain the emotional entity and emotional attributes in the instance,thus affecting the performance of emotional tendency analysis task.Therefore,inspired by the method of building the emotional knowledge map,this paper tries to apply the task of entity relationship extraction to the scene of the emotional tendency analysis to improve the recognition effect of the emotional tendency analysis task.However,in the entity relationship extraction task,the series entity relationship extraction method has the error propagation,and neglects the relationship between entity recognition and relationship extraction task,in addition,there is no effective solution to the problem of relationship overlap which exists objectively in the real data.In order to solve the problems of the above-mentioned entity relationship extraction,this paper proposes a method of joint extraction of entity relationships,which combines Conditional Random Fields with Convolution Neural Networks,and introduces Self-Attention mechanism and word location information to achieve accurate extraction of entity relationships.The main research contents and research results are as follows:(1)In view of the problem of error propagation,the lack of connection between multi-tasking and the overlap of relationships in the multi-tasking entity relationship extraction on the Chinese entity relationship extraction,this paper presents the joint extraction algorithm of the entity relationship between the fusion Condition Random Field and the Convolution Neural Network.The algorithm uses the Conditional Random Field model to identify the entities in the text,extract the subject position information,combine the Bi-directional Long Short-Term Memory Network,Self-Attention mechanism and word position information to obtain the text feature,extract the object and relationship with the subject by Convolutional Neural Network.The algorithm can achieve 79.79% on the F1 score of entity relationship extraction tasks,and perform better than the classic Multi-head Selection and DGCNN models in the same Chinese data set.(2)At present,in the task of fine-grained emotion analysis of text,there is a lack of method of emotional analysis of the extraction of emotional entities and attributes and the classification of emotional polarity as joint tasks.At present,in the task of fine-grained emotion analysis of text,there is a lack of method of emotional analysis of the extraction of emotional entities and attributes and the classification of emotional polarity as a joint task.In response to this problem,a multi-label classification algorithm based on entity-relationship joint extraction is presented.The algorithm uses the joint extraction algorithm of the complex Condition Random Field and the Convolution Neural Network to extract the emotional entity and emotional attribute of the commodity,introduces the position characteristics of the emotional entity and the emotional attribute,and carries on the fine-grained emotional multi-label classification of the text through the Bi-directional Gated Recurrent Unit.The algorithm can achieve 72.83% of the accuracy,which is better than the performance of the classical Bi-directional Gated Recurrent Unit,Bi-directional Long Short-Term Memory Network and other multi-label classification methods in fine-grained emotional tendency analysis tasks.Therefore,the effectiveness of the combined extraction algorithm of the entity relationship between the fusion condition random field and the convolutional neural network is verified in this paper.