中文摘要：

提出了一种自适应于不同题材文本自动确定其包含的潜在主题数K的方法．考虑到大多数文本的潜在主题分布符合段落密度特性，提出以段落为中心的研究策略，通过采用基于K均值的聚类算法联同自定义判别函数的聚类分析方法，实现了段落自适应聚类下的文本潜在主题的自动发现．实验结果表明，该方法在一定程度上能有效处理普遍存在的文风自由且主题表达灵活多样的各式文本．

英文摘要：

A novel approach to discovering topics and determining the number of latent topics K in a text automatically is proposed. By adopting K-means clustering algorithm as well as a clustering analysis algorithm based on self-defined discriminative function, the number of different latent topics in a text is captured and the diverse topics are found accordingly. Experimental results demonstrate that the proposed approach can deal with various texts with free writing style and flexible topic distribution effectively.