Abstract: To address diversity and changeability of user interests in current user interest modeling method, a user interest modeling method based on the Dynamic Self-Organizing Map (DSOM) neural network was constructed. This method mapped the process of user interests modeling into a clustering and cluster-maintenance process. Users' interests were aggregated as different information categories, and each cluster represented an interest category of the user. Changes of user interests were depicted by neural cell weight adjustment, new cell insertion and invalid cell deletion. A threshold-controlled trigger rule for insertion and deletion was used to update the user interest model in a real-time and reliable mode. A set of experiments were conducted on a benchmark dataset. Results from experiments showed that the DSOM model could provide reasonable clustering performance and high adaptability for accurately keeping track of users′ multiple interests and changes.

Key words: user interest model, self-organizing map, growing cell structures, neural network, recommended system

摘要： 为解决现有用户兴趣建模方法在处理用户兴趣多样性及动态性方面存在的问题，提出一种利用动态自组织映射神经网络来建立用户兴趣模型的方法。该方法将用户兴趣建模过程映射成一个聚类和类的维护过程。用户感兴趣的信息被聚成不同的信息类，每个信息类对应用户的一种兴趣。用户兴趣的变化通过神经元权重的调整、新神经元的增加和无效神经元的删除来刻画，分别对应用户兴趣基本不变、用户出现全新兴趣和用户原有兴趣消亡的情况。其中，神经元的增删采用阈值控制的触发规则。最后在一个基准数据集上的验证结果表明，该方法能够准确及时地跟踪用户多种兴趣及其变化，保证了用户模型的可靠性。

关键词: 用户兴趣模型, 自组织映射, 成长型单元结构, 神经网络, 推荐系统