基于内容和用户偏好学习的个性化商品推荐模型

发布时间：2018-05-06 21:41

本文选题：电商场景 + 商品推荐　；参考：《浙江大学》2017年硕士论文

【摘要】：随着信息技术和互联网的发展,信息过载现象使得用户处理海量数据并从中找到有效信息的代价越来越高,个性化推荐应运而生。电子商务是个性化推荐邻域的重要应用之一,如何从海量商品集中高效地为用户推荐个性化商品成为近年来研究热点。由于隐式反馈数据量大且易获取,专家学者逐渐将研究重心从显式评分推荐转移到隐式反馈推荐,其中以BPR(Bayesian Personalized Ranking)为主流模型。电商隐式反馈具有数据量大、存在固有噪声及正负样本极不均衡等特点,且性能要求高,现有模型无法直接应用,因此本文从采样策略和用户偏好定义两方面对BPR进行改进,旨在同时提高商品推荐的精度和效率。本文主要工作包括:1)提出一种基于内容的混合采样策略的BPR改进算法现有采样方式未充分考虑噪声样本对模型准确度和收敛速度的影响,而电商更注重推荐精度和高效性。因此本文研究提出一种基于内容的混合采样策略的BPR改进算法。该算法同时考虑商品对信息值、商品内容及用户潜在偏好三个因素,选择高质量、高可比及高可信的样本进行模型训练,真实电商数据的实验结果表明,其在各项评估指标上都优于BPR且比现有采样改进算法能更快收敛。2)提出一种基于内容和用户偏好学习的个性化商品推荐模型由于电商数据量大、拥有丰富商品内容及用户特殊网购行为特点,现有模型无法准确刻画用户偏好。因此本文研究提出一种基于内容和用户偏好学习的个性化商品推荐模型。模型采用混合采样策略,同时考虑用户潜在偏好、商品内容及用户网购行为特点,重定义用户偏好,并添加相应置信度,真实电商数据的实验结果表明,其相较现有模型能达到更高推荐精度且能更稳定更快的达到收敛状态。
[Abstract]:With the development of information technology and Internet, the phenomenon of information overload makes it more and more expensive for users to process massive data and find effective information. E-commerce is one of the important applications of personalized recommendation neighborhood. How to recommend personalized products efficiently and efficiently from mass commodities has become a hot topic in recent years. Due to the large amount of implicit feedback data and easy to obtain, experts and scholars gradually shift the focus of research from explicit score recommendation to implicit feedback recommendation, in which BPR(Bayesian Personalized ranking is the mainstream model. The implicit feedback of electricity quotient has the characteristics of large amount of data, inherent noise and extreme imbalance of positive and negative samples, and its performance is very high. The existing model can not be directly applied. Therefore, this paper improves BPR from two aspects: sampling strategy and user preference definition. The aim is to improve the accuracy and efficiency of product recommendation at the same time. The main work of this paper includes: (1) A modified BPR algorithm based on mixed sampling strategy based on content is proposed. The existing sampling methods do not fully consider the influence of noise samples on the accuracy and convergence rate of the model, while the ecoquotient pays more attention to the accuracy and efficiency of recommendation. Therefore, an improved BPR algorithm based on content-based mixed sampling strategy is proposed in this paper. The algorithm takes into account the three factors of commodity information value, commodity content and user's potential preference, and selects high-quality, high-comparable and credible samples for model training. The experimental results of real e-commerce data show that, It is superior to BPR in every evaluation index and can converge faster than the existing sampling improved algorithm. 2) A personalized commodity recommendation model based on content and user preference learning is proposed because of the large amount of e-commerce data. The existing models can not accurately depict user preferences because of their rich commodity content and the characteristics of users' special online shopping behavior. Therefore, this paper proposes a personalized commodity recommendation model based on content and user preference learning. The model adopts mixed sampling strategy, taking into account the potential preferences of users, commodity content and the characteristics of users' online shopping behavior, redefining user preferences and adding the corresponding confidence level. The experimental results of real e-commerce data show that: 1. Compared with the existing model, the proposed model can achieve higher recommendation accuracy and more stable and faster convergence.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】