This paper addresses the challenge of micro-action recognition, which involves identifying subtle and rapid movements that convey emotional states and intentions. The authors introduce a new dataset, Micro-action-52 (MA-52), which includes 52 micro-action categories and 22,422 video instances from 205 participants. MA-52 captures a wide range of whole-body movements, including gestures, upper and lower limb movements, and complex bodily interactions. To enhance micro-action recognition, the authors propose the Micro-Action Network (MANet), which integrates squeeze-and-excitation (SE) and temporal shift module (TSM) into the ResNet architecture. MANet is evaluated on nine existing action recognition methods and demonstrates superior performance in micro-action recognition. The study also explores the application of micro-action recognition in emotion analysis, showing that capturing micro-actions significantly improves emotion recognition accuracy. The dataset and source code are publicly available, providing a valuable resource for future research in human behavior analysis.This paper addresses the challenge of micro-action recognition, which involves identifying subtle and rapid movements that convey emotional states and intentions. The authors introduce a new dataset, Micro-action-52 (MA-52), which includes 52 micro-action categories and 22,422 video instances from 205 participants. MA-52 captures a wide range of whole-body movements, including gestures, upper and lower limb movements, and complex bodily interactions. To enhance micro-action recognition, the authors propose the Micro-Action Network (MANet), which integrates squeeze-and-excitation (SE) and temporal shift module (TSM) into the ResNet architecture. MANet is evaluated on nine existing action recognition methods and demonstrates superior performance in micro-action recognition. The study also explores the application of micro-action recognition in emotion analysis, showing that capturing micro-actions significantly improves emotion recognition accuracy. The dataset and source code are publicly available, providing a valuable resource for future research in human behavior analysis.