Research And Prototype Implementation Of Aspect-Opinion Pair Extraction

The popularity and rapid development of the internet have had a subtle impact on people’s consumption habits and lifestyles.People are increasingly inclined to complete shopping,recharging and other consumption on the e-commerce platform.In this scenario,user comments in the e-commerce platform play an important reference role in user consumption and also have key guidance for merchants to improve their products.However,the massive and increasing comment information makes it difficult for people to extract comprehensive summary information of goods from comments.Based on this background,a finegrained sentiment analysis technology,which can analyze the views and emotions contained in the comment text is proposed.This paper mainly makes an in-depth study on the task of aspectopinion pair extraction,and mines the fine-grained key subjective emotion expression contained in user comments.Firstly,this paper summarizes the critical challenges not considered in the existing work in this task,and puts forward a new set of problem modeling ideas according to the analysis results.Specifically,this paper proposes to decompose the aspect-opinion pair extraction task into two subtasks:aspect term extraction and aspectspecified opinion extraction,and complete the modeling of the overall task by combining these two subtasks.In addition,a new sequence labeling scheme and extraction scheme are designed for each subtask,and the subtasks promote each other through a multi-task learning mechanism to improve the overall extraction performance.To prove the effectiveness and interpretability of the method proposed in this paper,a variety of quantitative comparison experiments and qualitative analysis experiments are further designed.In this paper,a large number of experiments are carried out on four data sets for the three tasks involved in this paper(one main task and two sub-tasks).Quantitative experimental results show that the proposed method is significantly foremost to the most advanced research on these tasks.The qualitative experimental results also prove the scientificity and effectiveness of the components contained in the proposed method.Based on the good performance of technical research,this paper further designs and develops a system for aspect-opinion pairs extraction.By analyzing the needs of target users and combining design constraints,the system realizes the extraction of user comments and the visual display of results through bottom-up hierarchical design.It is worth mentioning that the model reasoning service of this system supports the three models involved in this paper.Finally,through a series of system tests and examples,this paper proves that the system fully meets the needs of users and design requirements.