Research On Sentiment Analysis Technology Of Commodity Comments Based On Deep Learning

Text sentiment analysis,also known as text opinion mining,is a process of applying computational linguistics knowledge to analyze and summarize the sentiment information in text.It is one of the research hotspots in the field of natural language processing,with a wide range of application scenarios.The traditional task of sentiment analysis is mainly to study the sentiment polarity of the whole text.But it lacks the deep-seated view of the text corpus.In contrast,fine-grained sentiment analysis can mine tagging the sentiment polarity of some specific objects in the text.However,the refinement of the analysis content makes it more difficult to extract information,especially for comments with multiple attributes.It is difficult to accurately match the expression of related sentiment when there are large numbers of target objects in the evaluation objects.At the same time,there may be overlapping,correlation association among multiple evaluation objects.In addition,it's also difficult to exclude irrelevant information.In the view of the above problems,this thesis proposes to use aspect-category sentiment analysis research method to mine the opinions of specific object in the comment text.Using the deep learning framework,to build a text representation layer that can analyze the dependency context.At the same time,convolution algorithm is used in the upper layer to enhance the expression of category aspects.And non-linear elements are used to match the aspects and sentiment features,so as to achieve the sentiment polarity analysis of the evaluation objects.The main contributions of this thesis are as follows:(1)This thesis proposes an algorithm model of aspect-category sentiment analysis,which integrates dependency parsing.The dependency parsing is constructed into node relation matrix,which is used to indicate the relation of words in context.Through the graph convolution neural network,the text features are extracted under the function of node relation matrix.The features with strong correlation and far distance in the text are fused.On this basis,the category features in the text are classified by pattern recognition.And then,the category features are filtered by the gating unit to match their sentiment expression.In the text representation layer,dynamic word embedding is added as a part of the text representation,which is adjusted according to the algorithm model.After experimental comparison and analysis,the classification effect of the model on the Al challenge 2018 fine-grained sentiment analysis data set is about 1%higher than that of the baseline.(2)Design and implement a commodity review system based on the algorithm above.The system can analyze the sentiment polarity according to the specified category in context,mark the associated position of the opinion expression in the text while displaying the classification results,count the number of different sentiment polarity comments of different categories,and display them directly.This thesis focuses on the research of sentiment analysis method for text category objects,which can effectively improve the accuracy of opinion classification and effectively predict the classification results.In addition,the commodity review system designed based on the model proposed in this paper can effectively process the review text and achieve good display effect.