Research And Implementation Of Fine-grained Emotion Classification Method Based On Attention Mechanism

In recent years,as content platforms have grown in vertical websites,a large amount of high-quality user-generated content(UGC)also emerged.User-generated content is characterized by “no threshold," "strong interaction" and "internet traffic advantage," making ways for the continuous growth of content platforms.In order to further understand and meet the actual needs of users,and to bridge the gap between content platforms and users,making good use of user-generated content to discover valuable information has become extremely important.This pressing demand makes the finegrained emotion classification a hotspot and focus in the field of natural language processing.Unlike traditional sentiment classification that focuses on the overall sentiment of a sentence or chapter,the task of fine-grained sentiment classification focuses on a more specific goal.In view of the characteristics of fine-grained sentiment classification,this paper mainly uses the attention mechanism to associate the target entity words with the sentence context,improving the performance and applicability of the model by constructing and using domain-specific sentiment dictionaries.This paper covers the following aspects:1.Domain-specific corpus and domain sentiment dictionary construction: The corpus of finance,mobile phone and hotel industries are constructed by crowdsourcing,which provides a high-quality data source for Chinese fine-grained sentiment classification tasks.We also use the method of fusing simplified and traditional Chinese characters to generate word vectors.After that,we use the three domain-specific word vectors to construct the lexicons respectively.2.Fine-grained sentiment classification model based on attention mechanism: We introduce the basic attention models,and verify the fine-grainedness of attention mechanism by comparing them with models based on gated neural network,confirming the applicability of sentiment classification tasks.In order to further explore the potential of the attention mechanism on the fine-grained emotion classification task,we also use BERT generated features to improve the model performance.In addition,we use the constructed domain sentiment dictionary to guide attention distribution,which improves the performance and interpretability of the model.The results in this paper demonstrate that the model based on attention mechanism has positive effects for fine-grained emotion classification,and the method of guiding attention distribution through the emotion dictionary has improved the performance of emotion classification tasks in three different fields to various extents.3.Design and implementation of the fine-grained emotion classification system based on the attention mechanism: Taking advantage of the strong interpretability and good performance of the attention mechanism,we implement the fine-grained emotion classification system based on the attention mechanism via web services.The system is easy to use and has strong expansibility.To some extent,it meets the production needs of actual NLP tasks.