基于梯度提升回归算法的O2O推荐模型研究

发布时间：2018-01-26 22:23

本文关键词： 梯度提升回归树位置服务个性化推荐行为日志分析　出处：《安徽理工大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着移动智能手机的普及,移动020电子商务迎来了广阔的市场空间,移动端的购物消费也更加普遍,基于此020电子商务的迅猛发展对推荐系统提出了更高的需求。针对当前020场景中的推荐模型,要求更准确更及时的个性化推荐系统,以及基于移动位置迅速变化产生的推荐内容更新。本文在研究020移动电子商务模型和阿里巴巴生活类服务商品一个月的真实交易数据的基础上,分析出用户基于自身的生活习惯会对某一类服务商品产生周期性消费；其次在热门消费场景区域会形成马太效应,移动用户的平均消费倾向会大大提高；最后每个用户在当前位置场景的消费倾向的不同,影响了用户在当前场景的商品转化率。根据上述三个假设,建立用户基于位置信息特征的偏好模型,提出了一种改进的基于梯度提升回归算法的020电子商务推荐模型,在推荐算法中引入当前时间参数和位置参数。同时深入挖掘移动020电子商务中的用户行为日志,抽取能辨别用户对商品服务购买行为的多个特征,然后将这些特征融入到梯度提升回归算法中,预测用户的购物行为。实验结果表明,改进的基于梯度提升回归算法的020电子商务推荐模型在实时性和准确性方面明显优于传统的推荐算法。
[Abstract]:With the popularity of mobile smartphones, mobile 020 e-commerce ushered in a broad market space, mobile shopping consumption is more common. Based on the rapid development of E-commerce of 020, the recommendation system is required to be more accurate and timely for the recommendation model in the current 020 scenario. This paper studies the 020 mobile electronic commerce model and the real transaction data of Alibaba's life service goods for one month. It is analyzed that the consumer will consume a certain kind of service goods periodically on the basis of their own living habits. Secondly, the Matthew effect will be formed in the popular consumption scenario area, and the average consumption tendency of mobile users will be greatly improved. Finally, each user's consumption tendency in the current location scene is different, which affects the product conversion rate of the user in the current scene. According to the above three assumptions, the user preference model based on location information features is established. An improved recommendation model for E-commerce 020 based on gradient lifting regression algorithm is proposed. The current time and location parameters are introduced into the recommendation algorithm, and the user behavior logs in mobile 020 e-commerce are deeply mined to extract multiple features that can distinguish the user's purchase behavior of goods and services. Then these features are integrated into the gradient lifting regression algorithm to predict the user's shopping behavior. The experimental results show that. The improved 020 E-commerce recommendation model based on gradient lifting regression algorithm is superior to the traditional recommendation algorithm in real-time and accuracy.
【学位授予单位】：安徽理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】