Aspect-level User Sentiment Analysis Technology For Online Reviews

With the popularity of online shopping,users are more and more enthusiastic about online shopping and post online reviews of products.These reviews have very important reference value for customers to purchase goods and merchants to improve their products and services.However,an increasing number of online shopping users generate massive user reviews,making it difficult for users and businesses to quickly retrieve desired reviews.Therefore,how to conduct automatic and fine-grained analysis of a large number of online comments is the main problem to be solved.The main work of this article is focused on the analysis of aspect-level opinion analysis oriented to online reviews,including the extraction of aspect-opinion pairs and aspect-based sentiment analysis of users' reviews.The specific work is as follows:(1)Extracting Aspect-Opinion Pairs(EAOP)from online reviews is a fine-grained opinion mining task.Most recent efforts generally extract aspects or opinions while ignoring the relations between them.However,the relations are crucial for downstream tasks,including sentiment classification,commodity recommendation,etc.Recently,the span-based method is used to randomly form candidate pairs from the generated spans for aspect-opinion pairs extraction.The large number of candidate word pairs brings challenges to model training.In this paper,we propose a span-based relationship graph transformer network(RGT)for extracting aspect-opinion pairs.In order to solve the problem of too many candidate word pairs,we first train a discriminator to recognize aspects and opinions,and then combine the recognized aspects and opinions to recognize aspect-opinion pairs.At the same time we use a powerful transformer for encoding,and propose a relational graph convolutional network to capture the dependent relationship between aspects and opinions.Extensive experiments show that our model consistently outperforms state-of-the-art methods.(2)Aspect-based sentiment analysis(ABSA)aims to analyze the sentiment polarity of an input sentence in a certain aspect.Many existing methods of ABSA employ long short-term memory(LSTM)networks and attention mechanism.However,the attention mechanism only models the local certain dependencies of the input information,which fails to capture the global dependence of the inputs.Simply improving the attention mechanism fails to solve the issue of target-sensitive sentiment expression,which has been proven to degrade the prediction effectiveness.In this work,we propose the multi-head self-attention transformation(MSAT)networks for ABSA tasks,which conducts more effective sentiment analysis with target specific self-attention and dynamic target representation.Given a set of review sentences,MSAT applies multi-head target specific self-attention to better capture the global dependence and introduces target-sensitive transformation to effectively tackle the problem of target-sensitive sentiment at first.Second,the part-of-speech(POS)features are integrated into MSAT to capture the grammatical features of sentences.A series of experiments carried on the Sem Eval 2014 and Twitter datasets show that the proposed model achieves better effectiveness compared with several state-of-the-art methods.