基于商品特征挖掘的在线评论有用性分类研究

发布时间：2019-06-20 07:57

【摘要】：随着电子商务的快速发展,越来越多的消费者习惯于网上购物。消费者在发生购买行为后,可以对己购买的商品进行评论,这些评论不仅是消费者对商品卖家的反馈,同时也能对其他消费者提供建议和指导。商品的热销意味着商品评论的大量增加,某些火爆的商品动辄数万条的评论让卖家和买家都难以处理,这就需要双方从海量的商品评论中快速地筛选出有用的评论,从大量冗余的信息中提取出真正可以指导销售和购买的有用信息。对海量在线评论中有用信息的迫切需求使得国内外研究者都不约而同地关注起了评论挖掘的一个具体的应用领域——评论有用性分类。本研究考虑到各大电商网站普遍无法提供全面的评论信息这一现实情况,从评论内容本身及商品特征信息入手,通过商品特征挖掘为评论有用性分类特征的选取提供参考；为了充分利用海量的评论,本研究采用半监督学习的方法对分类模型进行训练,最终得到有优异性能的评论有用性分类模型。论文首先研究已有商品特征挖掘方法的不足,从分词、剪枝和特征选取等方面进行有效改进,最后得到优化的商品特征挖掘结果；在此基础上,深入研究评论有用性的影响因素,将商品特征信息作为一个重要参考因素加入到有用性分类特征集合中；最后利用支持向量机的重要扩展——直推式支持向量机进行半监督学习,综合利用有标签评论和无标签评论,训练出在线评论有用性的半监督分类模型。结果显示该分类模型表现优于传统的监督学习模型,在只考虑评论内容信息条件下有较好的表现,进而说明商品特征信息是影响评论有用性的重要因素,而半监督学习可以有效地提升分类结果。
[Abstract]:With the rapid development of e-commerce, more and more consumers are used to online shopping. After buying, consumers can comment on the goods they buy. These comments are not only feedback from consumers to sellers, but also provide advice and guidance to other consumers. The hot sale of goods means a large increase in commodity reviews. Tens of thousands of comments on some popular goods make it difficult for sellers and buyers to deal with them, which requires both parties to quickly screen useful comments from a large number of commodity reviews and extract useful information from a large number of redundant information that can really guide sales and purchase. The urgent need for useful information in massive online reviews has led researchers at home and abroad to pay attention to a specific application field of comment mining-comment usefulness classification. Considering the fact that major e-commerce websites are generally unable to provide comprehensive comment information, this study provides a reference for the selection of useful classification features through commodity feature mining from the review content itself and commodity feature information. In order to make full use of massive reviews, this study uses semi-supervised learning to train the classification model, and finally obtains a useful review classification model with excellent performance. Firstly, this paper studies the shortcomings of the existing commodity feature mining methods, improves effectively from the aspects of word segmentation, pruning and feature selection, and finally obtains the optimized results of commodity feature mining. On this basis, it deeply studies the influencing factors of the usefulness of the review, and adds the commodity feature information as an important reference factor to the useful classification feature set. Finally, the semi-supervised learning is carried out by using the direct support vector machine, which is an important extension of support vector machine, and the semi-supervised classification model of the usefulness of online comments is trained by using tagged comments and untagged comments. The results show that the classification model is superior to the traditional supervised learning model, and has a better performance under the condition of only considering the content information of the review, which shows that the commodity characteristic information is an important factor affecting the usefulness of the review, and semi-supervised learning can effectively improve the classification results.
【学位授予单位】：大连理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：F724.6

【参考文献】