移动客户端个性化产品资讯推荐算法的研究与实现

发布时间：2018-04-11 17:59

本文选题：协同过滤 + 上下文　；参考：《中国科学院研究生院(沈阳计算技术研究所)》2016年硕士论文

【摘要】：传统电子商务与智能移动设备的有效结合,可显著提高用户购物体验。相较于PC端,智能移动设备具有续航时间短、终端处理能力相对较弱、网络带宽窄等缺陷,这就要求移动电子商务必须重视用户体验。研究表明,现有个性化推荐算法普遍存在数据稀疏性、冷启动等难以克服的问题。本文依托实验室“殨派网电子商务平台”项目,针对上述问题,进行详细讨论并提出相应的解决方案。本文主要研究内容如下:1.提出一种适用于移动端电子商务平台用户冷启动问题的解决方案。用户冷启动解决方案可分为三个步骤:(1)利用已有用户点击浏览记录,提取用户上下文特征;(2)利用提取的用户上下文特征训练用户兴趣分类模型;(3)利用用户兴趣分类模型预测新用户购物兴趣,完成个性化推荐。2.现有协同过滤推荐算法存在推荐准确度和数据稀疏性问题,引入用户信任机制,设计并实现了融合用户信任度模型的协同过滤推荐算法。本文将用户信任的产生方式分为两部分:社会声誉产生的信任和社会相似性产生的信任;提取表征用户信任的上下文特征,利用层次分析法完成模型构建过程。改进的协同过滤推荐算法通过阈值过滤的方式融合用户信任最近邻和用户评分相似最近邻,实现为用户匹配更多邻居的目的;改进现有评分预测公式,使得评分预测过程充分考虑用户信任关系对推荐结果的影响。实验结果表明,本文提出的改进算法有效的解决了用户冷启动问题,显著提高了个性化推荐系统的推荐质量。
[Abstract]:The effective combination of traditional e-commerce and smart mobile devices can significantly improve the user shopping experience.Compared with PC, smart mobile devices have some shortcomings, such as short life, weak terminal processing capacity and narrow network band. Therefore, mobile e-commerce must pay more attention to user experience.The results show that the existing personalized recommendation algorithms are difficult to overcome, such as data sparsity and cold start.Based on the project of Electronic Commerce platform, this paper discusses the above problems in detail and puts forward the corresponding solutions.The main contents of this paper are as follows: 1.This paper presents a solution to the cold start problem of mobile e-commerce platform.The user cold boot solution can be divided into three steps: 1) using existing user click browsing records,Using the extracted user context feature to train the classification model of user interest.) using the model of user interest classification to predict the new user's shopping interest and to complete the personalized recommendation.The problems of recommendation accuracy and data sparsity exist in the existing collaborative filtering recommendation algorithms. The user trust mechanism is introduced to design and implement the collaborative filtering recommendation algorithm which integrates the user trust model.In this paper, the generation of user trust is divided into two parts: trust generated by social reputation and trust generated by social similarity, and the contextual features that represent user trust are extracted, and the process of model construction is completed by using the Analytic hierarchy process (AHP).The improved collaborative filtering recommendation algorithm combines the user trust nearest neighbor and the user rating similar nearest neighbor by threshold filtering to achieve the goal of matching more neighbors for the user, and improves the existing scoring prediction formula.The effect of user trust on recommendation results is fully considered in the scoring prediction process.The experimental results show that the proposed improved algorithm can effectively solve the cold start problem of users and improve the recommendation quality of personalized recommendation system.
【学位授予单位】：中国科学院研究生院(沈阳计算技术研究所)
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】