Abstract: API-based behavior representation is currently an important part of mainstream Android malware detection and classification. However, due to the rough development of Android APIs, this method faces the problem of coarse granularity and is unable to precisely describe application behaviors. In response to this problem, an automatic API parameter semantic analysis method is proposed, which is based on program analysis and natural language processing technology. It refined the representation granularity from API to its parameters and realized a fine-grained representation of application behaviors. Experimental results show that this method can significantly improve the precision of Android application behavior representation, and enhance the effectiveness of tasks such as malware detection and classification.