刘海鸥 孙晶晶 苏妍嫄 张亚明

〔摘 要〕大数据环境下，推荐系统项目评分的稀疏性问题愈加突出，单兴趣表示方法也难以对用户多种情境兴趣进行准确描述，导致推荐结果精度大大降低。鉴于此，提出一种多情境兴趣表示方法，在此基础上构建面向图书馆大数据知识服务的多情境兴趣推荐模型，通过对用户多情境兴趣的层次划分，利用蚁群层次挖掘的优势来发现目标用户的若干最近邻类簇，然后根据类簇内相似用户对目标项目的评分对未评分项目进行预测，最后结合MapReduce化的大数据并行处理方法来进行协同过滤推荐。实验结果表明，多情境兴趣的建模方法改善了单兴趣建模存在的歧义推荐问题，有效缓解了大数据环境下项目评分的数据稀疏问题，MapReduce化的蚁群层次聚类方法也大大改善了推荐系统的运行效率。

〔关键词〕大数据知识服务；多情境兴趣；蚁群层次聚类；协同过滤推荐

DOI：10.3969/j.issn.1008-0821.2018.06.009

〔中图分类号〕G203 〔文献标识码〕A 〔文章编号〕1008-0821（2018）06-0062-06

〔Abstract〕Under the big data environment，the sparsity problem of recommendation system project becomes more and more serious.In addition，the traditional single interest representation method is difficult to accurately described，resulting in the reduced accuracy of recommendation result.In view of this，this paper put forward with a kind of multiple interest representation based on recommendation model for library big data knowledge service，by dividing the level of user interest more situations，using ant colony level mining advantage to some target users nearest neighbor cluster.According to the cluster within the same user rating to forecast the goal of the project not scored，this paper finally implemented parallel processing method for collaborative filtering with the MapReduce data.The experimental results showed that the modeling method generates new multiple item clustering interest tree by hierarchical partition mechanism，enhanced the mining depth of situational interest，and the MapReduced ant colony clustering method also greatly reduced the overall computation time，significantly improved the efficiency of the recommendation system.

〔Key words〕library big data knowledge service；multi contextual interest；ACO hierarchical clustering；CF recommendation

随着图书馆海量数据服务资源的不断涌现，项目评分稀疏性[1]、信息语义复杂性与多重性[2]问题大大增加了图书馆大数据知识服务个性化推荐系统实现的难度。图书馆大数据知识服务的个性化推荐是将读者兴趣、知识领域等关联信息加工为能够生动描述读者偏好的知识元，由此来支持数字图书馆各种推荐服务，最终为用户提供满足其个性化需求的知识资源。其中，项目评分的稀疏性问题研究已久，在此不详细赘述；……