| With the rapid development of the Internet,people are increasingly expressing their opinions on the online media and expressing their emotions.In this context,textual data on the Internet that contain emotional tendencies and perspectives is exploding.These textual information has extraordinary value.They often include the emotions and opinions of the publisher,which helps people to extract the points of interest and concerns of each individual.In recent years,people often use emotional information from texts to make decisions and begin to use the emotional information in the text to extract the causes of emotions.The main goal of emotional cause extraction is to identify the reason behind a certain emotional expression from the text.This paper mainly studies and implements the method of emotion cause extraction based on deep learning.Through the deep learning method,the relationship between the emotional words and the sentences in the text is established,and the emotional causes in the text are identified.In previous studies,deep learning-based methods usually used to accurately obtain the semantic relevance between sentiment words and sentences in the text,and to identify the emotional causes in the text.This article is mainly divided into four aspects:(1)The corpus for the problem of finding current emotional causes is still small.If we use some complex deep neural networks,it is easy to overfit.So this paper adopts a new semi-automatic annotation method for emotional causes extraction,which doubles the dataset.After that,we verifies the validity of the data.(2)For the current emotion cause extraction method,only the mutual attention mechanism is adopted,the relationship between the emotional words and the text clauses is considered,but the sentence itself is not considered.In addition,the previous methods are not considered to clip the attention weights.It will cause the model still retain unimportant words,which may cause noise.Therefore,this paper proposes a CAES network,combining the mutual attention mechanism and the self-attention mechanism in the model,then using the k-max method to pruning the attention weight.This network obtained an F-score of 0.701 on the new experimental data set.(3)For the modeling methods of the existing research,most of them adopt the direct classification method for modeling.This paper proposes a new modeling method,which considers the emotion cause extraction method as a sorting problem and uses the pairwise rank method.Modeling and using the ideas of the CAES network,the ABSCNN network was constructed,which obtained an F-score of 0.7116 on the experimental dataset.(4)Design and implement the emotion cause extraction prototype system,and integrate the algorithm of this paper into the prototype system. |
