Research And Application Of End-to-end Aspect-based Sentiment Analysis Algorithm Based On Deep Learning

With the rapid development of Internet technology,various social platforms and ecommerce platforms have emerged,which have been greatly expanding people's speech space and generating massive text data with emotional tendencies.The sentiment analysis of these text data is of great significance in the fields of society and business.Aimed at identifying different sentimental tendencies of different target entities in a sentence,aspect-level sentiment analysis has important research value.This paper first uses an integrated labeling strategy,which integrates entity tags and sentiment polarity tags,and defines the aspect-level sentiment analysis task as a sequence labeling task to achieve the end-to-end aspect-level sentiment analysis.In order to solve the problem of lacking of interactive information learning between texts in the end-to-end model,the multi-head attention mechanism is introduced on the basis of the Bi LSTM-CRF framework.Different subspaces are constructed by multi-head attention to enhance the learning of the internal structure of sentences and the ability of the model to focus on different semantic information.Then,in order to utilize the dependency between target entity recognition and its sentiment analysis task,a hierarchical bidirectional long short-term memory network model based on the attention mechanism(Att-HLSTM)is proposed.The proposed Att-HLSTM uses residual connection to establish a hierarchical neural network to make the prediction results of the target entity boundary in the bottom layer act on the upper layer neural network learning,and the gate mechanism is added to optimize the emotional consistency.In addition,there is insufficient research on Chinese aspect-level sentiment analysis tasks.The complex language characteristics of Chinese restrict the development of Chinese sentiment analysis tasks.This paper adds a Chinese data set to the experiment and conducts experiments on English and Chinese respectively.Aiming at the characteristics of Chinese characters,a vocabulary enhancement method based on word boundaries is proposed.The potential word information and word position information are regularly integrated,and the effect of the model is improved by increasing the amount of information represented by the model input.Finally,based on the model proposed in this paper,an aspect-level sentiment analysis system for online e-commerce Chinese reviews was designed and implemented,which completed the target entity recognition and sentiment judgment of the review,and proved the effectiveness of the end-toend model using the integrated annotation strategy.Moreover,such design can further mine the information value of consumer reviews provided help.The experimental results show that the Att-HLSTM model has achieved the best results compared with the baseline model on the Chinese and English data sets.The F1 value of the English data set reached 59.25% and 68.7%,respectively,and the F1 value of the Chinese data set reached 66.16%.Compared with the baseline model F1 value,the vocabulary enhancement representation method proposed in this paper has got 4% improvement.