In order to facilitate the processing and understanding of text by computer, this paper uses a triple tuple containing entity and entity relationship to represent the text’s fact in a more formal and concise way. The knowledge base (KBs), which contains a large number of facts, is used in various fields related to natural language processing. KBs usually integrates information from different places, such as manually edited encyclopedia, news articles, and social networks. In this paper, a natural language enhanced association rule mining method (NEARM) combined with KBs is used to deal with network comments. The fragments of fact are found from the pure text, and then the emotion is classified according to the fragments of fact found by the classifiers. Firstly, NEARM clusters the original data containing the pairs of related entities into clusters with different granularity from the data in KBs, and then excavates the rules in each cluster. These rules contain a large number of relational facts, which can reflect the relationship between plain text data, and can be effectively used in emotional classification of text. The experimental results show that the method is feasible. NEARM can deduce the relational triple facts and improve the accuracy of emotional classification.