Research On Aspect-level Sentiment Analysis Algorithm Based On Deep Learning

With the rapid development of the internet and the large-scale growth of users,how to extract people's opinions and emotional tendencies in big data has become an urgent problem to be studied and solved.Since the introduction of sentiment analysis in this century,it has gradually become a research hotspot in many fields,especially in the field of natural language processing.There are many practical applications for sentiment analysis,such as product pricing,international relationship analysis,and banking system risk monitoring.In recent years,with the rise of e-commerce platforms,people are increasingly used to leaving their own comments after purchasing goods.These evaluations of users will help others understand products and make purchasing decisions.And these evaluations also provide market decision-making for merchants.Thus how to automatically extract and analyze users',emotion from their review texts becomes an important research topic.Sentiment analysis tasks have different levels,including document level,sentence level and aspect level.Aspect level sentiment analysis can provide more accurate and detailed information than document level and sentence level sentiment analysis,but it is also more difficult.This paper studied the aspect level sentiment analysis algorithm based on deep learning.The main research contents of this paper are as follows:1.We propose an automatic extraction algorithm for aspect words and opinion words combining automatic labeling algorithm and deep learning.In order to solve the problem of lack of large-scale annotation corpus,this paper proposes an aspect level sentiment analysis algorithm combining automatic annotation algorithm and deep learning.The unlabeled train data is first labeled by an automatic labeling algorithm to solve the problem of lack of labeling of train data.Through the joint deep learning model,the relationship between aspect words and opinion words is fully explored,and the abstract semantic information of the text is fully captured.Finally,our algorithm can implement the automatic extraction of aspect word and opinion words by combining rule-based automatic labeling method and deep learning,and achieve good results.2.We propose two joint extraction models for aspect words,opinion words and their align relations based on CMLA and Attention.A variety of neural network-based joint extraction models,such as CMLA(Coulpled Multi-Layer Attentions)can implement joint extraction of aspect words and opinion words,but can't extract their align relations.It is impossible to learn which opinion words modify which aspect words.And this brings difficulties for subsequent sentiment analysis,In order to solve this problem,this paper propose two deep learning models combing CMLA and Attention,namely CMLA +Attention and CMLA +Attention + Label Embedding models.