融合辅助文本信息的项目冷启动推荐研究

Research on Auxiliary Text Information based Recommendation System for Cold Start Items

传统的推荐系统主要通过用户、项目的历史交互信息来学习用户、项目的特征信息,从而实现推荐。但是对于刚进入市场的新项目,因为没有或缺乏足够的历史记录,使得传统推荐系统的推荐性能受到局限,这便是经典的项目冷启动问题。在经典的矩阵分解算法基础上,进一步融合使用卷积神经网络提取的文本信息,提出两个针对不同项目冷启动情况的推荐系统,有效缓解项目冷启动问题。在一个真实的电影推荐数据集上的实验结果证明,所提出的推荐算法具有更优异的推荐性能。

The traditional recommender systems learn the user＇s interests and item＇s properties through their historical interactive data. But for the newcome items, the recommendation performances of these models are limited by the lack of historical records. It is also known as item cold start. To tackle this problem, adds the text information, which is extracted by Convolutional Neural Network, to the classical matrix factorization model and proposes two recommendation models to deal with two different cold start problems. Experimental results in realworld dataset show that the proposed models lead to significant improvement.