Sentiment-based SRT-ELECTRE III algorithm for movie and TV show recommendation optimization

In the context of social networks, the sentiment orientation of online comments become significant factor influencing viewers' decision-makings. This paper proposes a multi-attribute decision-making approach that combines regret theory based on sentiment analysis with the Elimination and Choice Expressing Reality (SRT-ELECTRE III) method. The process begins by collecting a dataset and preprocessing it. Then, the TF-IDF algorithm is employed to extract key terms and determine their weights. Subsequently, the Latent Dirichlet Allocation (LDA) topic model is utilized for clustering user texts based on topics. Next, SnowNLP is employed to calculate the sentiment values of online comments for each category. These sentiment values are then categorized into specific ranges to determine the degree of sentiment. By combining this with a probabilistic linguistic terminology set, a comprehensive evaluation is obtained. Subsequently, a probabilistic linguistic decision matrix is constructed based on the sentiment levels derived from online comments. This matrix is used to establish a regret-joy function for evaluating information. Through a utility function, the regret-joy harmony index and discord index of different alternatives are determined, leading to the establishment of a credibility index for each alternative. Finally, the net credibility of each alternative is calculated, resulting in a ranked list of alternatives. The proposed approach SRT-ELECTRE III is then applied to a case study for movie selection to obtain practical and meaningful decision makings. Furthermore, the proposed approach was compared with the TOPSIS and VIKOR methods to validate its superiority and effectiveness.