Knowledge-Enhanced Sentiment Analysis Of Review Texts

In the era of digital economy,the construction of various infrastructures in China has been perfected,and people can surf the Internet anytime and anywhere using smart devices.The flourishing development of various social media platforms and electronic commerce has provided a wide range of channels for Internet users to express their reviews and opinions,and has also brought about a huge amount of text data containing rich emotion in reviews.Efficient analysis of these opinion information has positive effects on improving the service quality of online platforms,enhancing social opinion monitoring,and improving business marketing strategies.In this paper,we take social media review texts as the research object and conduct research on aspect-based sentiment classification tasks.Due to the various forms of expressions of online review sentence and the problems of short length,insufficient background information and irregular wording,it is difficult to correctly model the semantic relationships of sentences from the perspective of review texts alone,especially in complex contexts.Therefore,we consider using linguistic knowledge such as vocabulary and syntax,and external structured knowledge together as auxiliary information to participate in model training to enhance its semantic parsing ability of review texts.In order to give full play to the role of different types of feature information,two different knowledge fusion models are designed in this paper to analyze the influence of prior knowledge on sentiment classification models from different perspectives.The specific work is as follows:(1)In this paper,an interactive attention neural network model incorporating conceptual knowledge is proposed.The model firstly uses the information provided by the knowledge graph to enhance the semantic recognition of aspect words by the model to avoid their ambiguity in different contexts;secondly,the parser is used to obtain the display syntactic features of the text and combine with the graph convolutional network to generate a contextual representation with knowledge of the syntactic structure;finally,the interaction attention mechanism is used to learn the similarity and relative importance between the context of the review text and the evaluation aspect Finally,the similarity and relative importance between the context of the review text and the evaluation aspect are learned through the interactive attention mechanism,and the coordination between the context of the review text and the evaluation aspect is optimized.The experimental results show that the incorporation of external knowledge can effectively improve the classification performance of the model.(2)For the aspect-based sentiment analysis task,sentiment knowledge is a very important factor,so more attention should be paid to sentiment knowledge in the process of introducing knowledge.To address the above issues,this paper coordinates and optimizes this paper’s model by incorporating multi-source knowledge information,including syntactic dependency information,word co-occurrence information,sentiment knowledge and concept mapping embedding,to enhance the representation of the contextual and evaluative aspects of review sentence.And the connection between the context and evaluative aspects of review sentence is strengthened by a dual-interaction attention model.The experimental results show that this knowledge can help the model better understand the meaning and sentiment tendency of review sentence,and the accuracy of sentiment analysis is improved on multiple datasets.Therefore,the incorporation of multi-source knowledge can enhance the model’s expressive power and generalization ability to a certain extent,and improve the recognition of the sentiment information of review sentence.