Joint Modeling Based On Multi-Task Learning For Aspect Term Extraction And Sentiment Classification

Aspect-level sentiment analysis is an important task in the field of natural language processing,which is used to identify sentiment tendencies in different aspects of evaluation in online reviews.When a comment statement contains multiple opinion subjects,how to extract each opinion subject,match the opinion term correctly and sentiment polarity is the key to this task.Most methods require multi-step operation in a pipelined manner to complete this task,which is easy to cause error propagation and redundant training.Therefore,how to filter unrelated words,solve the high-order representation problem of long-distance dependencies in comment sentences,overcome the phenomenon of "overfitting",and correctly match aspect words and their sentiment trigger words are the key research content of this project.This paper conducts localization experiments on the main methods of aspect-level sentiment analysis,and builds a joint model under the framework of multi-task learning to solve the problems of aspect term extraction,opinion extraction and aspect sentiment classification.⑴ The fine-grained sentiment analysis of user comments processes the two tasks of extraction and sentiment classification separately.To solve the above problems,the model proposes a joint model based on position embedding and graph convolutional network under the framework of multi-task learning,which is an end-to-end approach to the overall solution of aspect-level sentiment analysis.Firstly,the method learns the semantic feature representation of sentence through a bidirectional gated recur-rent unit network.Then exploits positional embedding to enhance the recognition of aspect terms in sentence,and uses the graph convolutional network to generate a contextual representation containing syntactic in-formation.Finally,interactive attention network is used to model the semantic relationship between context and aspect terms,and the sentiment polarity of aspect terms is output through softmax.Experimental results on the Sem Eval-2014 public datasets show that the performance of the proposed model has a significant improvement compared with other existing models.⑵ The Triplet elements neglect the semantic similarity and syntactic.To address the above problems,the model proposes a method based on gated attention and distance-weighted graph text for aspect sentiment triplet extraction.Firstly,the method uses a bidirectional long-short-term memory network to learn the sequence feature representation of sentences.Secondly,uses a gated attention unit to learn linear connections between words.Thirdly,uses a grammatical distance-weighted graph convolutional network to enhance the interactions between triplet elements.Finally,uses a grid tagging inference strategy to predict triples.Experimental results on four public datasets show that the proposed method can effectively enhance the interaction between triple elements and improve the accuracy of triple extraction.Moreover,the F1 values are all improved compared to the baseline model.