A Multi-Label Recommendation Algorithm Based on Graph Attention and Sentiment Correction

Abstract

Traditional recommendation algorithms overlook the abundant information present in user feedback and are limited to the binary relationship between items and users. In contrast, this paper proposes an algorithm that takes advantage of item labels and user feedback to generate faster and more precise recommendation lists. The algorithm employs graph attention link prediction and sentiment analysis to calibrate parameters and enhance traditional recommendation algorithms by utilizing user information comprehensively. Empirical results on publicly available datasets indicate that the proposed algorithm outperforms both traditional and sentiment-uncorrected recommendation algorithms in terms of accuracy.