Research On Aspect Sentiment Analysis Based On Deep Learning

Text sentiment analysis is an important research direction in the field of natural language processing,and has extensive application value in many scenarios such as product research,network public opinion discovery,and social hotspot analysis.Aspectbased sentiment analysis is a fine-grained sentiment analysis task,which acquires obtaining deep-level sentiment characteristics by fine-grained analysis and induction of aspects described by text.In recent years,deep learning models have shown unique advantages in the field of natural language processing.Being capable of processing massive text data and automatically capturing deep-semantic features,they are widely used in various downstream tasks.In the field of sentiment analysis,mining deeper information of texts by building appropriate deep learning models has become an important way to improve analysis results.Based on the deep learning model,this paper proposes efficient deep learning models for the subtasks of aspect-based sentiment analysis,which are aspect recognition and aspect sentiment analysis.First of all,through the analysis of annotated corpus,it is found that there exist key words with strong orientation for aspect recognition in a text.However,due to small data set size and the sparsity of data,it is difficult for the current model to fully learn the correspondence between key words and aspects.Based on this,this paper proposes an aspect recognition model based on alignment features and Bi LSTM.The model first uses Bi LSTM with a two-way network structure to capture the two-way semantic information from the text.and rewrites the data according to the aspect recognition-oriented characteristics of the key words in the text at the same time.By forming a bilingual parallel corpus with the original text,it then feeds the alignment algorithm in machine translation with the parallel corpus to extract the bidirectional alignment probability between word and aspect as the alignment feature.Finally,the aspect recognition is performed after joining with the alignment feature at the last hidden state of the Bi LSTM hidden layer.The experimental results show that the performance of the proposed model on different data sets surpasses the baseline model,and this novel model can effectively improve the effectiveness of the recognition task.Secondly,for scenarios where the text contains multiple aspects of information in aspect sentiment analysis tasks,most of the existing studies rely on the temporal features capturing ability of LSTM to model the correlation between aspects.However,the number and positions of the aspects and the sentiment information are accidental in a text,and multiple aspects usually affect each other and appear no regular timing characteristics.Therefore,single LSTM is not suitable for these aspect sentiment analysis tasks.Based on this,this paper proposes an aspect sentiment analysis model that integrates multifaceted attention mechanisms.This model combines a pre-trained language model BERT and Bi LSTM to build a BERT-Bi LSTM text encoding module.Multiple parallel encoding modules are built in the encoding layer to simultaneously encode the text under the target aspect and the adjacent aspect.By taking the encoded text representation under the target aspect as the integration center,and using the attention weight as the aspect weight to fuse the information of all neighboring aspects,the correlation between the aspects is modeled based on attention mechanisms.Then emotional classification feature vectors with multifaceted perception capabilities are generated for prediction.The experimental results show that the performance of the model proposed surpasses all baseline models in different data sets,and the proposed modelling improvements brings a notable effect on sentiment analysis.