Abstract: To address the issues of redundant recommendations and high time complexity in community-based recommendation methods under network formal contexts, Expert Community Construction Algorithm (ECCA) and Expert Community Based Recommendation Algorithm (ECBRA) were proposed. Firstly, the screening method for expert nodes was determined, and the comprehensive influence of nodes, which is composed of network structure and positive ratings, was designed. Nodes with higher comprehensive influence were selected as expert nodes. Secondly, the expert community for each user was constructed. A user node and all its adjacent expert nodes jointly formed an expert community, ensuring that all users have their own independent expert communities. Finally, recommendations were made based on expert communities. Combined with user preferences, dynamic and static recommendation confidence thresholds were designed, and the sum of the two thresholds served as the recommendation confidence threshold. Personalized recommendations for each user were realized through the calculation of recommendation confidence. ECBRA was compared with the Group Recommendation Algorithm based on Weaken-concept Similarity (GRAWS). Experimental results show that ECCA has fewer recommendation redundancies and lower time complexity. The total time consumption of ECBRA is only 0.1% of that of the GRAWS algorithm, with no recommendation redundancy. Comparisons were also conducted between ECBRA and classic collaborative filtering algorithms, including K-Nearest Neighbor (kNN), Item-Based Collaborative Filtering (IBCF), Concept Set Based Recommendation (CSBR), Concept Set Based-Personalized Recommendation Algorithm (CSPR), and the combination of GreConD and kNN (GreConD-kNN) algorithms on real datasets. The results also demonstrate that ECBRA achieves better performance. Specifically, on one of the datasets, compared with the CSPR algorithm, the recall rate is increased by 63.3%, and the F1-score is increased by 9.3%; compared with the kNN algorithm, the F1-score is increased by 62.7%.

Key words: Keywords: Network Formal Context, Expert community, Static confidence threshold, Dynamic confidence threshold, Recommendation system

摘要： 针对网络形式背景下基于社区推荐的方法会造成冗余推荐、时间复杂度过高的问题，提出一种以用户为中心的专家社区构建算法 (ECCA)和基于专家社区的推荐算法(ECBRA)。首先，确定专家节点的筛选办法，设计由网络结构和正向评分共同构成的节点综合影响力，综合影响力较大的节点选为专家节点；其次，构建每个用户的专家社区，用户节点与它的邻接的所有专家节点共同组建专家社区，保障所有用户均有自己独立的专家社区；最后，基于专家社区进行推荐，结合用户偏好设计了动态和静态推荐置信度阈值，两阈值相加得到推荐置信度阈值，通过推荐置信度计算实现针对每个用户的个性化推荐。ECBRA与基于传统专家社区的推荐算法GRAWS 进行了对比，实验结果表明ECBRA的推荐冗余少、时间复杂度低，ECBRA综合用时仅为GRAWS算法的0.1%，无推荐冗余。在9个真实数据集与经典的协同过滤算法k最近邻(kNN)、基于项目的协同过滤(IBCF)以及基于概念集合的推荐(CSBR)算法、基于概念集的个性化推荐算法(CSPR)、结合形式概念分析与kNN的推荐(GreConD-kNN)算法进行了对比，结果也展示了ECBRA效果较好，特别是其中一个数据集上，相较于CSPR算法的召回率提高了63.3%，F1值提高了近9.3%，相较于kNN算法F1值提高了近62.7%。

关键词: 关键词: 网络形式背景, 专家社区, 静态置信度阈值, 动态置信度阈值, 推荐系统