Social Activity Recognition and Recommendation Based on Mobile Sound Sensing

Nowadays, as social activities in the real world are getting more and more popular, it is important to understand the social interaction features and the social contexts of users (meetings, social gatherings, etc.) to support social activity organization. By leveraging the social context and activities of users, various services can be provided (e.g., recommendation of friends, activities, etc.). In this paper, we present a mobile phone based social activity recommendation system based on background sound recognition. We first propose a method to analyze background sound signals of different activities, where a combination of the Mel frequency cepstral coefficients (MFCCs) and several other sound features are extracted. Based on these features, the Dynamic Time Warping (DTW) with limited-path searching algorithm is used to recognize different human activities. The information online (friend relationship and online interaction history) is adopted to measure the intimacy among users. Finally, we present an algorithm that combines the distance, user preference and intimacy features for activity ranking and recommendation. Experimental results show that the proposed approach is effective and useful.