基于标签的强化学习推荐算法研究与应用  

Research and application of tag-based recommendation algorithm based on reinforcement learning

作　　者：李益群[1] 张文生[1] 杨柳[1] 刘琰琼[1] 

机构地区：[1]中国科学院自动化研究所,北京100190

出　　处：《计算机应用研究》2010年第8期2845-2847,2852,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(90924026);国家"863"计划资助项目(2008AA01Z121;2007AA01Z338)

摘　　要：针对协同过滤推荐算法性能稳定性往往受到数据稀疏性影响的问题,在强化学习的框架下提出一种基于标签的协同过滤推荐算法,利用标签模拟用户兴趣来构造非稀疏的个性化数据,并将模拟数据与历史用户访问数据相结合进行协同过滤推荐。实验结果表明,引入基于标签的个性化数据可以有效提升协同过滤算法的性能,且对两种数据的有效结合可以获得最好的效果。In order to solve instability problem in the performance of collaborative filtering recommendation algorithms caused by the data sparseness,this paper proposed an algorithm called tag informed reinforcement learning recommendation model （TIRLR） in framework of reinforcement learning. This paper used the tags to simulate user profiles to construct substantial personalized data,and combined simulated data and historical data to collaborative filtering recommendation. Experimental results show that TIRLR can effectively enhance the performance of collaborative filtering algorithms,and it can get the best result by combining simulated data and historical data.

关 键 词：强化学习 推荐 标签 协同过滤 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]