Recognizing Human Daily Activity Using Social Media Sensors and Deep Learning

Area of Science:

• Computational Social Science
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Human daily activity categories (e.g., sports, shopping) reflect lifestyle and are valuable for health and applications.
• Social media offers low-cost data for activity recognition, surpassing traditional questionnaires.
• Existing methods struggle with contextual information, word order, and capturing activity semantics in posts.

Purpose of the Study:

• To develop a general deep learning model for accurate human activity category recognition from social media posts.
• To address limitations of existing methods by incorporating contextual and temporal information.
• To create a benchmark dataset for training and evaluating human activity recognition models.

Main Methods:

• Proposed a deep learning sequence model to extract higher-level word phrase representations from social media posts.
• Integrated temporal information and external knowledge to enhance the capture of activity semantics.
• Developed and utilized a novel benchmark dataset for model training and evaluation.

Main Results:

• The proposed deep learning model significantly improved the accuracy of human activity category recognition.
• The model effectively captured activity semantics by considering contextual information and word order.
• Experimental results demonstrated superior performance compared to traditional classification methods.

Conclusions:

• Deep learning models offer a powerful approach for recognizing human daily activities from social media data.
• Integrating temporal information and external knowledge is crucial for enhancing semantic understanding.
• The developed dataset and model provide a foundation for future research in social media-based activity recognition.