关键词: 文本评论, 在线评分, 情感分析, 三角模糊数, 随机多属性可接受度分析

Abstract: As the representative of human wisdom and creativity, film is the product of modern technology and cultural development, which enriches and improves the quality of people’s leisure life. As a rising star in the arts, film has become an indispensable spiritual nourishment in world culture. With the deep integration of the internet and the film industry, a vast amount of diverse online review data has emerged. How to scientifically and accurately mine viewer preferences and market demand information has become a current hot topic of research. This paper proposes a movie ranking method based on big data environment. The evaluation results can help both viewers choose quality movies and producers understand viewer preferences and market trends so that they can make more targeted marketing decisions. In addition, this paper also make contribution to the evaluation of high-quality films, which helps to enhance the influence and competitiveness of the Chinese film internationally and strengthen the national cultural soft power.
   Firstly, this paper establishes a film comprehensive evaluation index system based on text review data. Web crawling techniques are utilized to obtain online review data of films from media websites and preprocess them. We use the Latent Dirichlet Allocation (LDA) theme model to identify the underlying choice preference of the viewer, and the decision-maker can determine the evaluation index system according to the underlying theme and high frequency topic words generated by LDA topic model, and the experience of the decision-maker. Secondly, the paper combines sentiment analysis with triangular fuzzy number to accurately reflect the fuzziness and subjectivity of the text reviews, avoiding the loss of evaluation information. We construct a sentiment analysis model based on sentiment dictionary, calculate the percentages of negative, neutral and positive sentiment tendencies in text reviews and construct triangular fuzzy number to represent the sentiment analysis results, which make the conversion of text data closer to actual values. Finally, we combine text reviews with online rating information to make up for the lack of considering only one of them and construct a film decision model based on obtaining enough information. The comprehensive ranking function is given by integrating the ranking function based on text information and the average online rating results. A new film evaluation framework is constructed by using the Stochastic Multicriteria Acceptability Analysis - Triangular Fuzzy Number (SMAA-TFN) method to calculate the overall acceptability index of each film and achieve the ranking of different films. In addition, the proposed method can provide the decision-maker with the assistant decision information by using center weight vectors to distinguish movie advantages and disadvantages, identifying competitors through analyzing the dominant relationship between the films. The comprehensive method extends the evaluation dimensions and improves the quality of decision-making.
   In this paper, the real online rating and text reviews of five different types films from the Douban website are selected as the experimental data to illustrate the implementation process of the method. Through a comparative analysis, it is found that the results of the proposed method are more accurate than other models, verifying the effectiveness and superiority of the proposed method.
   In theory, constructing a film ranking model that fully combines online rating with text reviews reduces information bias. The use of triangular fuzzy number to represent sentiment analysis results overcomes the ambiguity and uncertainty characteristics of the text. In addition, the SMAA-TFN evaluation framework proposed in this paper does not require the pre-assignment of subjective weight information or the calculation of objective weight information, overcoming the limitations of existing weight determination methods. In practice, this study leverages a new data source (big data) and integrates it into traditional multi-attribute decision-making methods to evaluate different types of films, providing effective management insights for audiences, producers, etc.

Key words: text reviews, online rating, sentiment analysis, triangular fuzzy number, stochastic multi-criteria acceptability analysis