NHRec:一种基于长短期兴趣的神经混合推荐模型

Neural Hybrid Recommendation Model Based on Long-term and Short-term Interests

用户建模是推荐系统中的一项基本任务,传统的方法使用协同过滤(CF)建模用户的潜在兴趣,但用户的兴趣往往是复杂多样且会随时间而变化,单一的模型无法准确建模用户的兴趣特征,针对此问题,本文提出一种新的自适应融合用户长短期兴趣的混合推荐模型(NHRec).该模型根据用户的历史信息,利用融合注意力机制的门控循环单元(GRU)建模用户的短期兴趣,兼顾时序信息和内容上的相关性,同时采用卷积神经网络(CNN)对用户的全局信息进行提取得到用户长期兴趣,并使用基于时间间隔信息的自适应方式融合两类兴趣进行推荐计算.实验结果表明,提出的推荐算法NHRec相较于目前比较流行的推荐算法表现出更为优越的推荐性能.

User modeling is a basic task in recommendation systems.Traditional methods use collaborative filitering(CF)to model the potential interests of users,but user interests are always complex and change over the time that a single model can hardly be accurately constructed.In order to solve this problem,we proposes a new hybrid recommendation model(NHRec)that adaptively fuses the short and long term interests of users.According to the user′s historical information,the user′s short-term interests are modeled by a gated loop unit(GRU)that integrates the attention mechanism,taking into account the temporal information and content correlation,while using the convolutional neural network(CNN)to extract the user′s global information as the user′s long-term interests.Finally we use the adaptive method based on the time interval information to fuse two types of interests for recommendation calculation.The experimental results show that the our recommendation algorithm has more superior recommendation performance than the current popular recommendation algorithms.