基于迭代回归树模型的跨平台长尾商品购买行为预测

发布时间：2018-10-24 08:01

【摘要】：长尾商品是指单种商品销量较低,但是由于种类繁多,形成的累计销售总量较大,能够增加企业盈利空间的商品。在电子商务网站中,用户信息量较少且购买长尾商品数量较少、数据稀疏,因此对用户购买长尾商品的行为预测具有一定的挑战性。该文提出预测用户购买长尾商品的比例,研究单一用户购买长尾商品的整体偏好程度。利用社交媒体网站上海量的文本信息和丰富的用户个人信息,提取用户的个人属性、文本语义、关注关系、活跃时间等多个种类的特征;采用改进的迭代回归树模型MART(Multiple Additive Regression Tree),对用户购买长尾商品的行为进行预测分析;分别选取京东商城和新浪微博作为电子商务网站和社交媒体网站,使用真实数据构建回归预测实验,得到了一些有意义的发现。该文从社交媒体网站抽取用户特征,对于预测用户购买长尾商品的行为给出一个新颖的思路,可以更好地理解用户个性化需求,挖掘长尾市场潜在的经济价值,改进电子商务网站的服务。
[Abstract]:Long tail commodity is a commodity with a low sales volume, but it can increase the profit space of an enterprise because of the variety of goods, the accumulated total sales amount is larger and it can increase the profit space of the enterprise. In e-commerce websites, the amount of user information is less, the quantity of long-tailed goods is less, and the data is sparse, so it is challenging to predict the behavior of users buying long-tailed goods. This paper proposes to predict the proportion of long-tailed goods purchased by users and to study the overall preference of a single user to buy long-tailed goods. By using the text information of Shanghai social media website and the rich personal information of users, we can extract the user's personal attributes, text semantics, attention relationship, active time and other kinds of features. The improved iterative regression tree model (MART (Multiple Additive Regression Tree),) is used to predict and analyze the behavior of users buying long-tailed items. JingDong Mall and Sina Weibo are selected as e-commerce sites and social media sites respectively. Using real data to construct regression prediction experiment, some meaningful findings are obtained. This paper extracts user features from social media sites and provides a novel way to predict the behavior of users buying long-tailed products, which can better understand the individual needs of users and tap the potential economic value of long-tailed markets. Improve the service of e-commerce website.
【作者单位】：中国人民大学信息学院;大数据管理与分析方法研究北京市重点实验室;
【基金】：国家自然科学基金青年科学基金(61502502) 国家重点基础研究发展计划(2014CB340403) 北京市自然科学基金(4162032) 中国人民大学2016年度拔尖创新人才培育资助计划
【分类号】：TP391.1;TP393.092

【相似文献】