Design And Implementation Of A Fine-grained Sentiment Analysis System For Review Texts In E-commerce Domain

The aspect based sentiment analysis in e-commerce domain mainly focuses on the analysis of the sentiment tendency of each attribute,i.e.,aspect,in each e-commerce platform review text,and no longer analyzes the sentiment tendency of the whole review text,which is closer to the actual needs of users.In this thesis,we study the finegrained sentiment analysis of Chinese e-commerce review texts,and design and implement a fine-grained sentiment analysis system for e-commerce reviews on this basis.The specific work done in this thesis is as follows.(1)To address the basic problem of the lack of Chinese e-commerce review dataset in the aspect sentiment triplet extraction task research,this thesis crawls 5602 reviews from an e-commerce platform in the field of laptops,and performs pre-processing operations,such as filtering,cleaning,Chinese word segmentation and annotated by words according to the annotation strategy proposed in this thesis,and 824 review texts were collated and build a Chinese e-commerce review dataset for the aspect sentiment triplet extraction task.(2)In the realm of fine-grained sentiment analysis research,the aspect sentiment triplet extraction task holds great significance.This thesis tackles this task by introducing a novel fine-grained sentiment analysis model.The proposed model is designed to improve the accuracy and granularity of aspect sentiment triplet extraction,resulting in a more comprehensive and detailed understanding of the sentiment expressed in the text.The model mainly contains three parts: sentence coding module,attribute sentiment word classification and pruning module,and attribute sentiment word pairing and sentiment prediction module,and solves the sub-task of fine-grained sentiment analysis through an end-to-end approach.Unlike previous models that only consider the relationships of individual words,this model not only considers the relationships of individual words,but also captures the interrelationships between word spans,improving the performance of the model for multi-word triplet extraction.The findings from the experiments provide compelling evidence for the efficacy of the proposed fine-grained sentiment analysis model based on span interaction.The model’s performance surpasses that of existing models,with an F1 value increase of 23.55%,14.49%,and 6.35% in comparison with the Li-unified-R model,JET model,and BMRC model,respectively.These results underscore the feasibility and superiority of the proposed model,thereby highlighting its potential impact in the field of sentiment analysis.(3)In this thesis,we utilize the Python programming language and Flask development framework to construct a fine-grained sentiment analysis system tailored to the e-commerce domain.This thesis approach is informed by real user requirements.The system is capable of accurately extracting and analyzing sentiment information from e-commerce reviews,providing valuable insights into customer opinions and preferences.The system is built based on Python language and Flask development framework,and can perform fine-grained sentiment analysis based on user’s input text.The system is tested to prove that the system can run stably and reliably and meet the actual needs of fine-grained sentiment analysis of users.