Research On Aspect-based Sentiment Analysis Method Based On Senticnet

In the era of big data,all kinds of information are filled in all aspects of human production and life.Compared with written records,people seem to be more inclined to express their true opinions on objective things,express their true attitude towards each event,or record the state of mind at that time.Many of these contents exist in the form of text,which hides great commercial and social value.This requires the elimination of a large amount of information,the screening of useful information,and analysis or utilization of useful information.As a sub field of natural language processing,sentiment analysis is an important way to transform information into value,which has attracted extensive attention in recent years.As the subtask with the most detailed analysis granularity in sentiment analysis,aspect level sentiment analysis has a deeper mining level and a more specific and deeper description,so it has a certain research value.With its inherent feature extraction ability and powerful feature representation ability,deep learning has made amazing achievements in the field of aspect-based sentiment analysis.In addition,the attention mechanism is becoming more and more mature,and its strong focusing ability makes the model more competitive in sentiment classification tasks.Aspect-based sentiment analysis has broad prospects,but there are still some technical problems:when the target text is too long and the syntactic relationship is relatively complex,it is difficult to accurately capture the dependency between the target and other words;Ignoring the role of common sense knowledge in goal relationship modeling;There are obscure polysemy in the comment text,but the model is prone to misjudgment.To solve the above problems,the main research contents of this thesis are as follows:When the existing view quintuple describes the commentator’s emotional color on a certain side of things,there may be some problems such as incomplete elements and unclear expression.Therefore,taking the category of aspect words and the place mentioned in the text as the entry point,a new emotion representation model is constructed,that is,aspect category items and place items are added on the basis of view quintuple.When the existing view quintuple describes the commentator’s emotional color on a certain side of things,there may be some problems such as incomplete elements and unclear expression.Therefore,taking the category of aspect words and the place mentioned in the text as the entry point,a new emotion representation model is constructed,which adds aspect category items and location items on the basis of view quintuple.In order to make full use of the contextual semantic information,accurately capture the syntactic information and long-distance constraints between words in sentences,and consider the possible polysemy of one word in the comment text,the hybrid embedding method based on character embedding,dependency embedding and context embedding is used in this thesis to extract aspects so as to further improve the accuracy of aspect extraction.Based on graph convolution neural network,a method based on Syn-in-GCN and SenGCN is proposed to solve the problem of aspect-based sentiment analysis.On the one hand,the graph convolution operation on the dependency parsing tree can capture the syntactic dependency between the target word and the context with high quality.By considering the relative syntactic distance,the model can focus on important words.Secondly,Senticnet,the sentiment knowledge base,is embedded into the graph convolution neural network as an external knowledge source to expand the emotional semantics.To sum up,the sentiment representation model proposed in this thesis expands the existing viewpoint representation model from different aspects,and further explores the extraction methods of various elements.For sentiment classification,it reflects the importance of syntactic relative distance on the basis of syntactic dependence,and integrates external knowledge to perceive emotion more finely.These improvements make the model get higher accuracy in the final extraction task and classification task,which is better than the existing models.Hope to bring some inspiration to other researchers.