A Deep Sentiment Analysis Model Incorporating Multidimensional User Information

Addressing the limitations of single-dimensional classification in traditional sentiment analysis and the lack of inter-task correlation between linear layers in multidimensional classification,this paper proposes a user sentiment analysis model based on multi-dimensional information fusion(MDIF-SAM). First,based on the TopicBERT topic classification model,a multi-dimensional information interaction fusion network is introduced to realize the double classification of user's comment information in category dimension and sentiment dimension, and the data of the same class-sentiment is aggregated to form a specific class-sentiment small sample text set after classification;then, keywords are extracted from the aggregated specific small sample text set, and the relevance of each keyword set is calculated and sorted to generate the main keyword set with no repetition and strong correlation,which achieves a deep analysis of user sentiment and enables merchants to gain insights into users' needs and feelings through the generated keyword set.Verified on two real open source datasets online-shopping and dmsc-v2, the accuracy of MDIF-SAM in category and sentiment classification is 92.23% and 95.23%, the precision is 93.22% and 94.92%, and the F1 value is 91.24%, and a recall of 90% and 94.9%, respectively. Compared with the four models BERT, TopicBERT, ALBERT and RoBERTa, the indicators are significantly improved, and the correlation between keywords extracted by MDIF-SAM and the original text is more than 82.5%. Based on the results of experimental comparisons of real-world cases,the proposed model can provide in-depth analysis and insight into user emotion.