摘要
[目的/意义]基于强化学习的图书内容推荐方法可解决传统推荐系统注重推荐列表的准确率,忽略图书推荐的多样性,且无法解决数据信息缺失的问题。[方法/过程]采用强化学习框架对图书推荐过程进行马尔可夫决策过程建模,实现基于用户不同状态的推荐动作;采用因子分解机和强化Q-Learning作为值函数的近似计算,通过值函数选择最优候选图书;引入随机策略作为最后的推荐结果,提高了图书推荐的多样性。[结果/结论]强化学习算法在图书推荐过程中考虑了用户的动态兴趣,有助于提升推荐结果的多样性。当推荐数量为高于40时,强化学习算法能实现推荐准确率、多样性和覆盖率的平衡,与传统方法在数量为10时最优相比,图书推荐内容的多样性得到大大提高。
[Purpose/significance]The book content recommendation method based on reinforcement learning can solve the existing problems in traditional recommendation system such as only emphasizing the accuracy of recommended list while ignoring the book recommendation diversity and not solving data information missing.[Method/process]The reinforcement learning framework is used to model the Markov decision-making process of the book recommendation process to realize the corresponding recommendation actions according to different user states;The factor decomposition machine and reinforcement Q-Learning are used as an approximate calculation of action value function by which the best candidate books can be selected;A random strategy is also introduced to improve diversity as a final recommendation.[Result/conclusion]The reinforcement learning algorithm takes into account the dynamic interest of users in the book recommendation process and improves the diversity of recommendation results.When the recommendation number is more than 40 the reinforcement learning algorithm can basically achieve the better balance of recommendation accuracy diversity and coverage and it compared with the traditional method optimized when the recommendation number is 10 the diversity of book recommendation content is greatly improved.
作者
宋爱香
马冲
Song Aixiang;Ma Chong(Network&Informatization Management Office Xi’an Polytechnic University,Xi’an Shaanxi 710048;Xi’an Polytechnic University Library,Xi’an Shaanxi 710048)
出处
《情报探索》
2020年第1期9-15,共7页
Information Research
关键词
图书内容推荐
强化学习
多样性
马尔可夫决策
book content recommendation
reinforcement learning
diversity
Markov decision-making
作者简介
宋爱香(1980-),女,硕士,馆员;马冲(1980-),男,本科,馆员。
