基于“读者—图书”二部图的个性化图书推荐方法

Personalized Recommendation of Books Based on “Reader-Book” Bipartite Graph

个性化图书推荐已成为图书馆领域关注的热点问题,但面临着读者兴趣、图书内容难以获取以及"冷启动"等一系列挑战。文中基于图书借阅行为建立"读者—图书"二部图模型,并基于此提出个性化图书推荐方法。该方法首先根据书名计算图书之间相似度;其次,基于读者兴趣相似度对读者进行聚类,并生成每个读者的获选图书集合;最后计算每个读者与候选图书集合中每本图书的匹配度,并排序后输出推荐图书列表。实验结果表明,该方法能在未知读者兴趣、图书内容的情况下,有效地实现个性化图书推荐,并缓解了"冷启动"问题。

Personalized book recommendations have become a hot area in library science. Current recommending methods,however,are facing the difficulty to automatically acquire reader interests and book topics,and the “cold start” problem. A novel personalized book recommending method based on “Reader-Book” bipartite graph derived from the book lending behavior is proposed. First,the semantic similarities among books are calculated utilizing the book titles. Second,readers are divided into different groups with the use of clustering analysis based on the similarity of reader interests. Every group is assigned a selected book set. Finally,each reader is recommended a preferable book list based on the matching degree between reader and book. Experimental results show that this method can recommend personalized books to a reader without knowing reader interests and book topics,and alleviate the “cold start” problem.