基于机器学习的互联网用户满意度个性化评估

发布时间：2018-07-06 08:28

本文选题：支持向量机 + 贝叶斯模型　；参考：《中国地质大学(北京)》2017年硕士论文

【摘要】：互联网飞速发展至今,经历了四个具有清晰界限的、并且拥有历史意义的阶段,而目前发展到第四阶段则是由于互联网的深刻发展,用户满意度将会是互联网巨头们在未来发展中稳中求胜的首要问题,用户满意度(User Satisfaction)将会是在竞争中决定胜负的关键因素。提升用户满意度是提升产品在互联网市场上的竞争力的重要手段,关于用户满意度评估的研究已经有了很多优秀的研究成果。但大多数研究成果都局限于影响产品的客观因素,而忽略了用户满意度与用户主观因素的关系。针对这一研究现状,本文主要做了以下研究:首先,对用户满意度评估中主观影响最大的个性化推荐系统进行了调查和研究。充分分析了个性化推荐系统的架构、功能、产品设计等相关问题,同时有效解决了信息过载给现代互联网发展带来的庞大压力。本课题将深入互联网内容类产品—图书这一具体场景,对此类问题进行了相关的实验和分析,通过对用户期望的量化分析,合理的提出了用户期望对用户满意度的影响,最终构建了以用户期望预测为基础的用户满意度个性化评估模型。其次,本文表达了内容类互联网产品的用户个性化满意度,讨论每个用户的主观期望的个性化。本课题基于统计概率分析的角度,综合可观测特征和不可观测特征建立了后验概率模型,使用图书类互联网产品的开源数据集,不停调整训练参数,对该模型进行多次迭代和训练,用户期望预测的目标得以实现。最后,本文通过构建并训练RVM模型,对用户期望进行预测来优化图书个性化推荐系统。相关向量机算法采取全概率框架,先验概率引入模型权重,超参数对权重进行一对一及一对多的分类支配,而超参数的值可由稀疏训练向量模型训练的处。本文的研究,为用户接受个性化服务提供了理论支撑,通过建立包含用户期望的用户满意度评估模型,根据已有的研究理论总结出用户期望的性质及用户期望与用户满意度之间的关系,更加精准地对用户的期望进行预测。
[Abstract]:So far, the rapid development of the Internet has gone through four distinct and historic stages, and the fourth stage is due to the profound development of the Internet. User satisfaction will be the most important issue for Internet giants to win in the future, and user satisfaction will be the key factor to decide the success or failure of the competition. Improving customer satisfaction is an important means to enhance the competitiveness of products in the Internet market. There have been many excellent research results on the evaluation of user satisfaction. However, most of the research results are limited to the objective factors that affect the product, while ignoring the relationship between user satisfaction and user subjective factors. In view of the present situation of this research, this paper mainly does the following research: first, the personal recommendation system which has the most subjective influence in the evaluation of user satisfaction is investigated and studied. The structure, function and product design of personalized recommendation system are fully analyzed. At the same time, the enormous pressure of information overload on the development of modern Internet is effectively solved. This topic will go deep into the concrete scene of Internet content product-book, and carry on the related experiment and analysis to this kind of problem, through the quantitative analysis of the user's expectation, put forward reasonably the influence of the user's expectation on the user's satisfaction. Finally, a personalized evaluation model of user satisfaction based on user expectation prediction is constructed. Secondly, this paper expresses the user satisfaction of content Internet products, and discusses the individuation of each user's subjective expectation. Based on the statistical probability analysis, a posteriori probabilistic model is established based on the observable and non-observable features, and the training parameters are adjusted continuously by using the open source data set of the books internet products. The model is iterated and trained for many times, and the target of user expectation prediction can be achieved. Finally, this paper optimizes the book personalization recommendation system by constructing and training RVM model to predict user expectation. The algorithm of correlation vector machine adopts the full probability frame, the priori probability introduces the weight of the model, and the super parameter classifies the weight by one to one and one to many, and the value of the superparameter can be trained by the sparse training vector model. The research in this paper provides a theoretical support for users to accept personalized services, through the establishment of user satisfaction evaluation model including user expectations, According to the existing research theory, we summarize the nature of user expectation and the relationship between user expectation and user satisfaction, and predict user expectation more accurately.
【学位授予单位】：中国地质大学(北京)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP181;TP391.3

【相似文献】