Acquisition of Typical Occupancy Schedules for Commercial Buildings from Social Networks

Occupancy behavior in buildings has been a hotspot of research as building systems become more sophisticated. Traditional sensing technologies suffer from high costs, inevitable sensor errors, and scalability issues and thus are not widely implemented in buildings. In this notes paper, two different approaches for extracting the typical occupancy schedules for the input to the building energy simulation are explored based on the data from social networks. The first approach is to use text classification algorithms to identify whether people are present in space where they are making posts on social networks. To achieve this, word embedding and machine learning algorithms for the classification are used. On top of that, we could extract the typical occupancy schedules by assuming certain people counting rules. The second approach is to utilize the processed GPS location tracking data provided by social network giants such as Facebook and Google Map. Web scraping techniques are used to obtain the data and extract the building typical occupancy schedules. Two preliminary case studies demonstrate these two approaches as a proof of concept using a museum building, the Art Institute of Chicago. The results show that the extracted building occupancy schedules from different social network data sources (Twitter, Facebook, and Google Map) share a similar trend but slightly distinct with each other, which requires more explorations and further validations and corrections. However, the methodology and promising results from this preliminary study will lay the foundation of the occupancy sensing through the social network data mining, which aims to provide another data source for occupancy sensing in buildings at the building level. This will provide another alternative to estimate the occupancy at the community- and urban-scale.