群体特征提取是发现特定知识群体偏好, 进而提供个性化服务的基础. TFIDF是进行特征提取的常用方法, 然而传统{\rm TFIDF}方法却未考虑 到类间集中度和类内分散度的问题. 针对该情况文章引进了CD因子, 提出了 新的TFIDF算法, 以读者借阅数据为基础开展了图书馆知识群体特征提取研究. 并以某高校建筑与城市规划群体为例, 采用传统和改进两种TFIDF 方法对群体特征进行提取. 实证表明, 改进TFIDF方法效果更佳.

Feature extraction is the basis for discovering the preferences of specific knowledge groups and forms the premise of providing personalized services in libraries. TFIDF is a common method for feature extraction. Aiming at the problem that the traditional TFIDF method fails to take into account the degree of concentration between classes and the degree of decentralization in the same class, the CD factor which means concentration and dispersion is constructed based on inter-class concentration and intra-class dispersion. A new TFIDF algorithm for feature extraction is proposed with CD factor. Based on readers' borrowing data, the new TFIDF algorithm is able to be applied to extract library knowledge group feature. Then an example of group feature extraction is given based on the data of a library in university. The traditional TFIDF and improved TFIDF methods are applied to extract the group features in this example. Empirical evidence shows that the improved TFIDF method is better.