| With the rapid popularity of social media and e-commerce,users generate a large amount of social media text and online comment in the process of using social media and Electronic business platform.Data mining of these comment texts can provide comprehensive and scientific decision-making basis for the government,enterprises and consumers.Therefore,sentiment analysis of online reviews has received extensive attention both industry and academia.According to the characteristics of Chinese online reviews,this paper explores a sentiment analysis method that integrates different information such as emotional knowledge graphs,syntactic dependencies,and serialized annotations in deep learning models,using knowledge graphs,dependency parsing,deep learning and so on.The main contents of this paper include:(1)A sentence-level sentiment analysis method integrating knowledge graph and deep learning.This paper firstly analyzes the characteristics of Chinese online reviews by taking car reviews as an example.There are a large number of network terms in the comment text,the default of the evaluation object,the implicit sentiment expression,and the polarity of the sentiment words changes with aspect words.According to the existing problems,this paper proposes a method to construct emotional knowledge graph.Then the paper constructs the sentiment knowledge triple of<opinion-evaluation object-sentiment polarity>,which is different from the previous entity relation triple.Various pretraining models such as BERT and ERNIE have made great achievements in the task of natural language processing,but lack domain-specific knowledge.Knowledge graphs can enhance language representation.Furthermore,knowledge graphs have high entity/concept coverage and strong semantic expression ability.We propose a sentiment analysis knowledge graph(SAKG)-BERT model that combines sentiment analysis knowledge and the language representation model BERT.To improve the interpretability of the deep learning algorithm,we construct an SAKG in which triples are injected into sentences as domain knowledge.The experimental results confirm the effectiveness of the SAKGBERT method on the Chinese sentence-level sentiment analysis task.(2)Aspect-based sentiment analysis method based on AOCP annotation system.This paper proposes an AOCP annotation system for ABSA task,which is different from BIESO annotation system.The AOCP tagging system annotates aspect word(A),opinion word(O),aspect category(C),and sentiment polarity(P).The paper constructs a Chinese aspect-based sentiment analysis corpus,and uses AOCP for serialization annotation.This paper selects a BERT+CRF model to implement an end-to-end aspect-level sentiment analysis method on the corpus.ABSA generally adopts a pipeline method.First,aspect words and opinion words are extracted,and then perform ALSC or triple extraction task by using them as input.The staged ABSA method has different objective functions,and cannot effectively share knowledge and parameters with each other,which is likely to cause false cumulative effects.In the endto-end ABSA training process,the multi-task objective equations are consistent and the parameters are shared,which effectively improves the effect of the fine-grained sentiment analysis model.(3)Aspect-based sentiment analysis(ABSA)method combining syntactic dependencies and graph attention networks(GAT).At present,aspect-level sentiment analysis models based on RNN and attention mechanisms treat sentences as sequences,and do not fully utilize contextual syntactic information such as parts of speech and syntactic dependencies.In order to accurately match aspect words and opinion words,this paper proposes to construct syntactic relation subgraphs centered on opinion words and use GAT to implement ABSA tasks.The experiments were conducted on Chinese reviews for the first time,and the experimental results confirmed the effectiveness of the model.(4)Application of Comment Text Sentiment Analysis in E-commerce Question Answering System.Traditional information retrieval feeds back the results of simply sorted web pages,while the question and answer based on knowledge graph(KGQA)feedback the answer closest to the user’s question needs through semantic analysis,which realizes semantic understanding and knowledge retrieval.The research on KGQA has achieved fruitful results,such as the classification method of question answering,the question answering method of semantic analysis,the answer sorting method,and the knowledge map question answering of multi jump reasoning.This paper mainly studies the application of emotion analysis in e-commerce QA system,trying to solve the contradiction between the untimely answer in the QA area of e-commerce platform,the inability to meet the needs of customers and the commodity evaluation of a large number of users of e-commerce platform.Break the information barrier and gap,use the e-commerce evaluation data sentiment analysis to construct the user evaluation knowledge graph,so as to improve the customer satisfaction and consumer experience of e-commerce QA system. |
