基于LDA主题扩展的个性化电影推荐系统

Personalized Movie Recommendation System Based on LDA Theme Extension

摘要: 依据用户评分数据的传统电影推荐算法存在数据稀疏、评分信息不能够真实有效地表达用户兴趣等问题。而用户评论作为用户兴趣、观点反馈的信息有效载体，通过挖掘分析评论信息可以将产品特征向量化，实现个性化推荐效果。本文利用网站评论文本信息分析电影特征，在给用户推荐时可以根据用户历史评论信息分析用户兴趣，提出一种融合情感分析和LDA主题模型特征拓展的个性化推荐算法，选取主题关键词结合TF-IDF权重进行物品主题关键词特征拓展，情感分析得到的正向评论率结合主题拓展特征向量进行物品相似度计算，用户感兴趣物品相似度较高的产品作为推荐列表进行推荐。通过实验表明本文方法提高了推荐的准确度。

Abstract: Traditional movie recommendation algorithms based on user score data have some problems, such as sparse data and false score information, which cannot really and effectively express user interest. User comments, as an effective carrier of information from users’ interests and opinions, can quantify the product features through mining, analyzing and commenting information, and achieve personalized recommendation effect. Site review analysis of text information based on feature film recommended to the users in time according to the user history information analysis. User interest recommendation algorithm is proposed for expanding fusion sentiment analysis and LDA topic model features personalized selection keywords combined with TF-IDF weight item keywords and the characteristics development; the positive rate of sentiment analysis combined with topic comment expands the feature vector for item similarity calculation; the user is interested in the high similarity of the product as the recommended items list recommended. Experiments show that the proposed method improves the accuracy of recommendation.