时间上下文优化的协同过滤图书推荐

Research on a Collaborative Filtering Algorithm of Time Context Optimization for Book Recommendation

针对高校图书馆场景存在的无显式反馈、借阅数据稀疏和传统推荐算法效果不好问题,提出基于时间上下文优化协同过滤的推荐算法,包含读者阅读行为评分、时间上下文和内容兴趣变迁3个要素。在数据准备阶段,通过制定评分转化规则、设计标准化函数来构建一种基于用户行为操作的兴趣评分模型,以解决用户评分缺失问题;在推荐召回阶段,提出一种非线性的时间衰减模型来对评价矩阵进行优化,以提高推荐效果;在推荐排序阶段,提出一种兴趣捕捉模型对召回结果按照图书类别进行精排序,以缓解数据稀疏问题并进一步提高推荐效果。实验结果表明,文章提出的优化算法在Top5的F值较未经优化的协同过滤提升增幅达141%。

University libraries have been facing such problems as little explicit feedback,decreasing borrowing data and poor performance of traditional recommendation algorithms.Thus,a recommendation algorithm for collaborative filtering optimization based on time context is proposed here.It consists of three elements,i.e.,reading behavioral score,time context,and content interest.In the stage of data preparation,in view of the lack of user ratings,an interest rating model(IRM)based on user behavior operations is constructed,so that scoring conversion rules are formulated,and the normalization function could be realized.In the stage of recommendation recall,in order to improve the performance of collaborative filtering recall results,a nonlinear time decay model(TDM)is used to optimize the evaluation matrix.In the stage of recommendation ranking,in order to alleviate the data sparse problem and further improve the recommendation performance,an interest capture model(ICM)is proposed to rank the recall results according to book categories.The final result shows that compared to traditional collaborative filtering,the proposed optimization algorithm could lead to a 141%increase in the F-Measure of Top5 recommendation.