面向推荐系统的评论分析研究

发布时间：2018-01-31 18:11

本文关键词： 推荐系统用户评论情感分析文本挖掘　出处：《苏州大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着互联网和计算机技术的不断进步,信息的数量不断爆炸式增长,导致了普遍的“信息过载”问题。在信息大爆炸的今天,人们迫切需要解决的问题就是如何高效处理并使用这些信息。除了搜索引擎之外,另外一种可以帮助人们快速发现自己所需信息的强大工具就是推荐系统。推荐系统可以根据用户的历史行为、用户生成内容、商品的描述信息等客观数据来建模,从而推测用户的偏好,进而为用户推荐有价值的信息或者商品。近年来,越来越多的研究者注意到使用用户生成内容(如用户评论,地理位置,好友关系等)来设计推荐算法可以提高推荐性能。本文从不同角度研究分析了用户评论,提高推荐准确度,改善推荐质量。本文的主要工作如下:(1)对推荐系统相关的技术背景做了调研,系统地介绍了推荐系统的原理,并比较了一些常用推荐算法的优势和劣势。(2)相对评分来说,用户对物品的评论从用户和物品的各个角度具体表达了用户的观点。利用这些信息更有助于挖掘用户的喜好,本文提出了一种基于词向量的方法挖掘用户评论信息,并结合协同过滤方法设计新的推荐算法,来改善评分预测的效果。(3)针对现有基于评论的推荐算法没有充分考虑个性化的问题,本文通过对评论进行主题分析,挖掘用户喜好,分别建立基于用户和物品的个性化评分预测模型以提高推荐系统的评分预测性能。(4)主要介绍了基于知识库的评论短语抽取和分层式展示方法。高效、准确、用户友好的评论标签展示不仅可以改善用户查阅用户评论的体验,也可以用于构建更加细致用户画像和物品特征,实现更精确的推荐系统。
[Abstract]:With the continuous progress of the Internet and computer technology, the quantity of information is increasing explosively, which leads to the widespread problem of "information overload". The problem that people urgently need to solve is how to process and use this information efficiently, except for search engines. Another powerful tool that can help people quickly find the information they need is the recommendation system, which can generate content according to the user's historical behavior. Product description information and other objective data to model, so as to speculate the preferences of users, and then recommend valuable information or commodities for users in recent years. More and more researchers have noticed that using user-generated content (such as user comments, geographical location, friend relationships, etc.) to design recommendation algorithms can improve the performance of recommendations. The main work of this paper is as follows: 1) the technical background of recommendation system is investigated and the principle of recommendation system is introduced systematically. The advantages and disadvantages of some commonly used recommendation algorithms are compared. The user's comments on objects express the user's point of view from all aspects of the user and the object. The use of this information can help to mine the user's preferences. In this paper, a word vector based method is proposed to mine user comment information, and a new recommendation algorithm is designed based on collaborative filtering method. To improve the effect of rating prediction. (3) in view of the existing recommendation algorithm based on comments has not fully considered the problem of individuation, this paper analyzes the topic of comments, mining user preferences. In order to improve the performance of rating prediction of recommendation system, this paper mainly introduces the methods of comment phrase extraction and hierarchical presentation based on knowledge base. Accurate, user-friendly comment label display can not only improve the user's experience of checking user comments, but also can be used to build more detailed user portrait and object features, and to achieve more accurate recommendation system.
【学位授予单位】：苏州大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】