Research And Application Of Aspect Based Sentiment Analysis Based On Deep Learning

With the advent of 5G era,a large amount of unstructured text data has emerged explosively.In order to quickly grasp the views on political events,public topics,commercial products,social mechanisms and so on.Aspect based sentiment analysis has become a hot scientific research in the field of natural language processing.There are two main tasks in aspect based sentiment analysis,one is classification task,the other is extraction task.At present,the classification task is mostly implemented by graph convolution neural network model,which can ensure that the structure of word relationship in sentences does not change with training.However,the complexity of the model is high,and the relationship between aspect words and context is ignored.The extraction task mainly extracts aspect words,opinion words and sentiment through multi task learning method.It can effectively distinguish the feature representation of the three,but it also leads to poor coupling between aspect words and opinion words,which affects the subsequent sentiment classification.In view of the above problems,this thesis has proposed a simplified graph convolution method based on specific masking mechanism and a multi task learning method with location information,and make a sentiment analysis software based on the two methods.The main innovations and research contents of this thesis are as follows:(1)Aiming at the problems of high complexity of graph convolution neural network and poor relevance between aspect words and context in classification task,this thesis proposed a simple graph convolution sentiment analysis model integrating specific masking mechanism.The model used bidirectional gated round unit to obtain timing information,and then used position-aware transformation to strengthen the learning of words around aspect words.Then,it reduced the complexity of the network by removing nonlinearity and folding the weight matrix between successive layers.Then,it was fused with the specific aspect masking layer to realize the simple graph convolution masking structure,and then combined with the feature representation of the hidden layer in the gated round unit to generate an attention mechanism based on the retrieval context.Finally,the classification vector was obtained through the full connection layer to predict the sentiment polarity.A large number of experiments have been carried out on five datasets,and the experimental results show that the performance of the model is better.(2)Aiming at the lack of coupling between aspect words and opinion words in the extraction task,this thesis proposed an opinion triplet sentiment analysis model integrating position information.The model used the bi-directional long short-term memory network to obtain the text representation,then used the self-attention mechanism to enhance the correlation between aspect words and opinion words,and then extracted the opinion triples in the multi task learning method.Then the extracted representation and position information were weighted and fused.Finally,the biaffine score was used to analyze the sentiment dependence between the weighted aspect words and opinion words,and then used stop-on-non-I algorithm to decode triples and output triples.A large number of experiments on four datasets show that the method is significantly better than a series of baseline models.(3)In order to verify the usability of the sentiment analysis model proposed in this thesis in real life,this thesis constructed a sentiment analysis model applied to Chinese and English texts based on the proposed method.The simple graph convolution neural network was applied to the opinion triple structure fused with position information to extract tags,and combined with the attention mechanism to realize the sentiment classification of aspect words.Finally,the model was deployed to the local server and combined with Client to realize the functions of testing,training and updating data.