基于电商评论的个性化产品推荐系统研究

发布时间：2018-04-17 16:17

本文选题：电商评论 + 相似度　；参考：《哈尔滨工业大学》2017年硕士论文

【摘要】：随着电子商务的迅猛发展,商城的多样性,商品种类不断细化,物流的便捷性,消费者的网络购物也越来越频繁,人们慢慢从传统的实体店消费模式转移到网络购物。消费者在购买和使用产品之后,对商品发表评分及评论,在网络上分享自己的使用体验。这些评论日益增多,呈现信息过载的趋势,同时评论中包含了大量信息,也体现了消费者对产品的评价及喜好程度。越来越多的人在做出购买决策前喜欢先到网上参考产品的评论信息。用户的评价通常包含一个数值评分和一段文本形式的评论,这些评论体现了用户对产品不同属性特征的偏好和用户的情感倾向。随着购物数量的增大,评论信息的数量也会不断增加,如果全部阅读来帮助决策是十分困难的。电商评论近几年也成为研究的热点。评论数量的巨增,导致了信息过载。推荐系统已经被电子商务行业得到广泛的使用。在各大电商平台的界面都可见到“猜你喜欢”。推荐系统通过提供个性化的推荐帮助人们克服信息过载的问题,其核心是通过个性化的算法,利用不同用户对于不同商品的评价信息,发现他们的喜好。协同过滤推荐是目前应用广泛的推荐算法,由于它跟具体的领域相关性比较弱,所以在电商,新闻,评论,音乐等行业取得了比较成功的应用。简单讲就是把和目标用户具有相似喜好的用户喜欢的商品进行推荐。本文主要研究内容如下:对用户的评论文本进行结构化处理,通过中文分词进行提取特征值,通过对词频进行统计,提取倾向词袋,对潜在用户进行个性化推荐。通过对比对有共同爱好和基于用户的协同过滤推荐算法,使用对有共同爱好个性化推荐算法,并对相似性的计算进行了优化,提出了基于综合相似度的计算方法,分别从用户对商品评分的相似度,基于情感倾向评分的相似度,基于标签的相似度三个方面的综合相似度进行计算,提出相应的推荐公式。实验根据综合相似度与传统相似度的对比结果表明,在量化指标的衡量下本文中的算法优于传统相似度计算方法,改善了推荐的质量。
[Abstract]:With the rapid development of electronic commerce, the diversity of shopping malls, the variety of goods, the convenience of logistics, the consumers' online shopping is becoming more and more frequent, and people are slowly moving from the traditional consumption mode of physical stores to online shopping.After purchasing and using the product, consumers rate and comment on the product and share their experience online.These comments are increasing, showing the trend of information overload, while the comments contain a lot of information, and also reflect the degree of consumers' evaluation and preference for products.More and more people prefer to go online to refer to product reviews before making purchase decisions.The user's evaluation usually includes a numerical rating and a textual comment, which reflects the user's preference for the different attributes of the product and the user's emotional tendency.As the amount of shopping increases, so will the number of comments, which can be difficult to read to help make decisions.In recent years, ecommerce reviews have also become the focus of research.The huge increase in the number of comments has led to information overload.Recommendation system has been widely used in e-commerce industry.You can see "guess you like" on the interface of all major ecommerce platforms.Recommendation system provides personalized recommendation to help people overcome the problem of information overload. The core of recommendation system is to find out the preferences of different users through personalized algorithm and using the evaluation information of different users for different products.Collaborative filtering recommendation is a widely used recommendation algorithm. Because of its weak correlation with specific fields, collaborative filtering recommendation has been successfully applied in e-commerce, news, commentary, music and other industries.Simply put, it is to recommend the products that the users like to have similar preferences with the target users.The main contents of this paper are as follows: structured processing of user's comment text, extraction of feature value by Chinese word segmentation, extraction of propensity word bag by statistics of word frequency, individualized recommendation to potential users.By comparing the collaborative filtering recommendation algorithms with common interests and users, using personalized recommendation algorithms with common interests, and optimizing the calculation of similarity, a method based on comprehensive similarity is proposed.This paper calculates the similarity of the user's product score, the similarity based on the emotion tendency score and the similarity based on the label, and puts forward the corresponding recommendation formula.The experimental results show that the proposed algorithm is superior to the traditional similarity calculation method and improves the quality of recommendation.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】