Impact of the spatial context on human communication activity

Abstract

Technology development produces terabytes of data generated by human activity in space and time. This enormous amount of data often called big data becomes crucial for delivering new insights to decision makers. It contains behavioral information on different types of human activity influenced by many external factors such as geographic information and weather forecast. Early recognition and prediction of those human behaviors are of great importance in many societal applications like health-care, risk management and urban planning, etc. In this paper, we investigate relevant geographical areas based on their categories of human activities (i.e., working and shopping) in order to understand human-environmental relationships. We use spectral clustering approach followed by k-means algorithm based on TF/IDF cosine similarity metric. We evaluate the quality of the relevant area clusters with a use of silhouette coefficients which are estimated based on the similarities of the mobile communication activity temporal patterns. The area clusters are further used to explain typical or exceptional communication activities. We demonstrate our study using a real dataset containing 1 million Call Detailed Records. This type of analysis and application is important for discovering hidden relationships and unknown correlations.