Implicit Sentiment Analysis With Linguistic Features

Sentiment analysis has always been a hot research direction in natural language processing.Implicit sentiment analysis refers to the task of sentiment analysis without explicit sentiment words.Implicit sentiment analysis is faced with problems such as lack of explicit sentiment words,euphemism of expression,and difficulty in understanding semantics.Traditional sentiment analysis methods such as sentiment dictionary and word bag model are difficult to be effective,making task of implicit sentiment classification more difficult.From the perspective of linguistic features,this article delves into the impact of different types of linguistic features on implicit sentiment analysis by combining three different implicit sentiment classification methods,including the following three aspects:Firstly,in response to the difficulties of implicit sentiment analysis,such as the ambiguity and strong subjectivity of sentiment expression,this article obtains sentiment information from different perspectives through different types of text representation models in feature representation.On the one hand,this method can use TextCNN,LSTM,and BERT to capture the text features of different granularity of sentences respectively,which makes up for the problem that the semantics of implicit emotional sentences are difficult to understand.On the other hand,this method can generate part-of-speech embedding and syntactic embedding,and further integrate part-of-speech features and syntactic features in semantic features.The experimental results show that the model proposed by this method has improved accuracy and F1 value by 2.3%and 2.8%respectively compared to baseline.The model that integrates text features of different granularity performs better than a single text feature model,and the fusion of linguistic features can further optimize the model classification effect.Secondly,in response to the lack of explicit sentiment words as sentiment clues in implicit sentiment sentences,this article deeply analyzes the impact of different linguistic features on implicit sentiment analysis based on attention mechanisms,and proposes an implicit sentiment classification method that combines pre trained language models and attention mechanisms.This method generates a mask matrix for the attention mechanism through dependency syntactic relationships,thereby integrating syntactic features into the attention mechanism,making the model more focused on syntactic components closely related to sentiment expression.The experimental results show that the model proposed by this method improves accuracy and F1 value by 1.7%and 1.8%respectively compared to baseline,which can improve the model’s attention to important parts of speech and syntax,thereby improving the performance of implicit sentiment classification.Finally,to address the issue of the impact of syntactic information on implicit sentiment analysis,this paper proposes a graph neural network model that combines text and dependency relationships for implicit sentiment classification.The model can extract the part-of-speech features and dependency features of the text,and construct a graph attention neural network combining linguistic features.Due to the outstanding performance of the pretraining language model in the implicit sentiment classification task,this paper uses the BERT model to extract the text vector features,and further improves it to increase the part of speech features.The expperimental results show that the proposed model improves accuracy and F1 value by 1.5%and 2.3%,respectively,compared to baseline.It can fully utilize part-of-speech and dependency syntactic features,and improve the model’s recognition ability for implicit emotional sentences.