Research On Negative Emotion Opinion Target Extraction Algorithm For User Comments

People leave many comments containing emotional information on the Internet when the Internet,especially the mobile Internet,develops rapidly.Through text sentiment analysis on massive user comments,people’s real emotions,attitudes,and opinions on products and services can be mined.Such information is helpful for consumers to make choices and businesses and service platforms to optimize products and services,especially the negative comments targets in user comments known as negative-emotion opinion targets.Merchants can improve their products or services more specifically when consumers can make better choices with information of the defects in products or services,and service platforms can supervise merchants effectively by extracting negative-emotion opinion targets from user comments.This paper proposes two models for the new finegrained sentiment analysis task of target extraction of negative emotional comments for user reviews.We first propose a negative-emotion opinion targets extraction network based on word mixed embedding and attention mechanism.The network uses a word representation method based on word mixed embedding to enhance the generalization ability of user comment data with poor text quality.The affective feature capture method based on the attention mechanism was added to assist the judgment of the affective polarity of the opinion targets.In ablation experiments on three datasets and comparison with 11 baseline models,the proposed model achieved the best results,verifying the effectiveness of the word-mixed embedding and attentional mechanisms.The accuracy of the network to judge the polarity of the target emotion is reduced due to the mutual interference between the target emotional features.This paper further proposes a new method based on fusion dependency syntactic features and multigranularity local features.The new multi-feature fusion-based negative-emotion opinion targets extraction network can alleviate the mutual interference between the emotional features of the opinions and enhance the ability of the model to capture the emotional features of the opinions correctly.Comparison experiments and ablation experiments on three datasets verify the validity of the proposed dependency syntactic relationship features and multi-granularity local features.