基于主题特征的情感分类及推荐算法研究

发布时间：2018-07-31 07:38

【摘要】：随着电子商务的发展,越来越多的人选择使用互联网进行购物、娱乐、交流和工作。用户往往需要花费很多时间查找、浏览和挑选商品或服务。如何针对用户的需求推荐产品与服务成为了产品和服务提供者必须解决的问题。然而,传统的推荐算法只单纯分析产品和用户之间的关系,没有对评论文本进行情感分析和处理,一定程度上影响着推荐效果的好坏。论文从情感分析及产品推荐两个方面,实现向用户更准确地推荐产品。首先从中文文本整体含义的角度,采用文本主题模型LDA,得到中文文本的主题分布。在进行特征选择和抽取时,除了抽取短文本特有的特征外,考虑了短文本的主题特征,在保留半监督学习框架下实现中文文本的情感分类。实验证明通过考虑中文短文本主题特征,文本情感分类准确率比无文本主题特征的情感分类准确率有一定提高。在对产品评论文本进行情感分析和协同过滤基础上,借鉴评分相似度的定义提出情感相似度,采用情感相似度和评分相似度结合的综合相似度进行近邻用户的评判与选择,最终提出了基于情感分析的产品推荐算法。其中,情感相似度计算方法采用论文提出的基于主题特征的半监督学习情感分类算法计算得到。与传统的协同过滤推荐算法进行了对比,实验证明基于情感分析的产品推荐算法较传统的协同过滤推荐算法的推荐效果更好。
[Abstract]:With the development of e-commerce, more and more people choose to use the Internet for shopping, entertainment, communication and work. Users often spend a lot of time searching, browsing, and selecting goods or services. How to recommend products and services to the needs of users has become a problem that must be solved by product and service providers. However, the traditional recommendation algorithm only analyzes the relationship between the product and the user, and does not carry on the emotion analysis and the processing to the comment text, which affects the quality of the recommendation to a certain extent. From the two aspects of emotional analysis and product recommendation, the thesis realizes more accurate product recommendation to users. Firstly, from the perspective of the whole meaning of Chinese text, the topic distribution of Chinese text is obtained by using the text topic model (LDAs). In the process of feature selection and extraction, besides the feature of short text, the subject feature of short text is considered, and the emotion classification of Chinese text is realized under the framework of reserved semi-supervised learning. The experimental results show that the accuracy of text emotion classification is better than that without text theme feature. On the basis of emotion analysis and collaborative filtering of product comment text, using the definition of score similarity, the emotional similarity is put forward, and the comprehensive similarity of emotion similarity and score similarity is used to judge and select the neighboring users. Finally, a product recommendation algorithm based on emotion analysis is proposed. The affective similarity method is based on the semi-supervised learning emotion classification algorithm proposed in this paper. Compared with the traditional collaborative filtering recommendation algorithm, the experiment shows that the product recommendation algorithm based on emotion analysis is better than the traditional collaborative filtering recommendation algorithm.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.1

【相似文献】