中文摘要：

推荐系统是电子商务环境下过滤过载信息的有效工具，但只向用户推荐商品的传统二维推荐方法并不适用于C2C（Customerto Customer）电商情境。在C2C在线情境中，商品的供应者不止一个，消费者不但需要筛选商品，而且需要筛选卖家。针对这一需求，本文提出了一种新的个性化三维推荐方法，同时考虑C2C在线情境下买家、卖家和商品三者之间的相关性，并为买家提供卖家和商品的组合推荐。该方法包括四个阶段：首先基于卖家特征计算卖家相似度；其次依据卖家相似性和销售关系对三维推荐空间中的历史评分数据进行补充，降低其稀疏性；然后依据推导出的商品评分计算买家相似度，寻找具有相似商品偏好的最近邻；最后通过一个三维预测模型，计算买家对“卖家和商品”组合的未知评分，并根据预测评分进行推荐。通过基于淘宝网的真实数据实验，证实在C2C情境中本文提出的三维推荐方法比传统二维推荐方法更加有效。

英文摘要：

Recommender system is an effective instrument to filter overloaded information in e-commerce. However, traditional two-dimensional recommendation methods, featured with product-only recommending to customers, are not applicable in C2C （Customer to Customer） e-commerce context. Due to the diversity of product suppliers, consumers need to filter products as well as their sellers in such online context. In the paper, a new personalized three-dimensional recommendation approach was proposed, in which relevance among buyers, sellers and products are fully considered for providing combined seller-product recommendations to buyers. The approach can be summarized to four continuous stages. Firstly, the seller similarities are calculated based on seller features. Secondly, seller similarities combined with sales relations are used to supplement and reduce the sparsity of the historical rating data in the three-dimensional recommendation space. Thirdly, based on derived product ratings, buyer similarities are calculated so as to find out the buyer neighbors with similar product preferences. Finally, a three-dimensional prediction model is applied to predict theunknown ratings that the buyers may give to candidate seller-product combinations, and recommendation are offered according to these predicted ratings. The results from an experiment of real data in Taobao website prove that the proposed three-dimensional recommendation method is more effective than traditional two-dimensional ones in C2C context.