Abstract: It has been the focus of academia and application field since knowledge discovery in data is proposed, but facing the increasingly complex application environment, most of the existing research methods are based on statistical learning, which often neglect the relationship between knowledge and the knowledge change over time, leading to that the effect of application is often just passable. How to grasp the statistical relationship between knowledge accurately to make the correct knowledge learning, have become a problem in knowledge research. Recently, with the development of research on statistical relational learning, Markov logic networks, which combines the probability graph and first-order logic theory, is applied successfully to natural language processing, machine learning, social network analysis and so on. In this paper, a Markov Logic Networks is presented for knowledge learning, which brings in first-order logic to learn the relationship between knowledge, and finish logical reasoning. Compared with SVM, this proposal method gets about 10 percent higher effect by classification knowledge learning in the text categorization fields.

Key words: knowledge learning, Markov logic networks, classification, Support Vector Machine（SVM）

摘要： 现有的知识学习多基于统计方法，常常忽略了知识间的关系以及随时间的变化情况，在应用效果方面往往差强人意。如何准确把握知识间的统计关系，进行正确的知识学习，成为知识研究的一个重点和难点。近几年，随着统计关系学习研究的兴起，结合概率图模型和一阶逻辑理论的马尔可夫逻辑网被成功应用于自然语言处理、机器学习、社会关系分析等领域中。基于马尔可夫逻辑网技术，提出一种知识学习方法，在传统知识获取方法的基础上，引入一阶逻辑来学习知识间的关系，进行逻辑推理。在文本分类的应用实验中，通过对分类知识学习，与传统的SVM相比，所提出方法的准确率提高10%左右。

关键词: 知识学习, 马尔可夫逻辑网, 分类, 支持向量机