基于深度学习加强的混合推荐方法.PDFVIP

基于深度学习加强的混合推荐方法* 丁弼原，张敏，谭云志，刘奕群，马少平 (清华信息科学与技术国家实验室（筹），清华大学计算机系, 智能技术与系统国家重点实验室，北京，100084) z-m@ 摘要：近年来基于矩阵分解的协同过滤算法在评分预测上取得的显著成果，但冷启动、数据稀疏等问题仍 然未能得到很好的解决，因此如何将评论信息引入推荐系统以缓解上述问题，开始成为研究的热点之一。 另一方面深度学习在图像处理、自然语言处理等领域已经发挥重要作用，其对文本的编码与分析能力恰好 适应推荐领域当前正在研究的问题。本文尝试基于深度学习来加强个性化推荐，提出将层叠降噪自动编码 器（SDAE）与隐含因子模型（LFM）相结合的混合推荐方法，综合考虑评论文本与评分，以提升推荐模型对 潜在评分预测的准确性。在常用大规模公开数据集 Amazon 上进行的测试结果表明，与传统推荐模型相比， 所提出方法可有效提高评分预测的准确性，最高可达到 64.43%的性能提升。 关键词：深度学习；协同过滤；混合推荐 Hybrid Recommendation Approach Enhanced by Deep Learning DING Biyuan, ZHANG Min, TAN Yunzhi, LIU Yiqun, MA Shaoping (State Key Laboratory of Intelligent Technology and Systems Tsinghua National Laboratory for Information Science and Technology Department of Computer Science and Technology, Tsinghua University, Beijing 100084, China) Abstract: Collaborative filtering based on matrix factorization has achieved great success, while cold-start and data sparseness problems have not been well-solved. Hence it has been a significant trend to involve review information into rating prediction as one of the dominant solutions. One the other hand, deep learning has played a vital part in image and natural language processing recently, which meets the requirement of current recommendation research. This paper proposes a hybrid model that introduces deep learning in recommendation system with collaborative filtering. It combines Stacked Denoising Auto Encoder (SDAE) with Latent Factor Model (LFM) to make use of both review and rating information so as to improve the performance on rating prediction. Comparative evaluation has been made on large-scale commonly-used Amazon dataset. Results show that the proposed approach improves the rating prediction precision greatly in comparison with traditional models, revealing a up to 64.43% perf