Exploring Community from Electricity with Human Community Interaction

Abstract

Communities represent one of the fundamental ways we live in. Exploring a community, named also as social assessment evaluates some social variables and holds a significant role in the field of social science. Traditional social assessment methods require extensive participation from individuals and organizations in society, making it progressively labor-intensive and time-consuming as data scope expands. This, in turn, leads to outdated statistical information in some cases.In order to facilitate faster and broader social assessment, this work locates residential electricity data with strong availability and minimal privacy concerns for inferring social variables at the community level. Nonetheless, conducting this interdisciplinary research presents hurdles stemming from the gap between vast user data and scarce community labels, as well as the interaction between communities and users. To address these challenges, we propose a Time Series Kalman Fusion (TSKF) layer to establish connections between communities and users (human). Furthermore, we introduce a prompt-learner method and a tuning methods with human community interaction to narrow the gap. Our experimental results confirm the effectiveness of this method. By merging time series pre-training techniques with the TSKF layer and prompt-learner, we witness a nearly 10\% loss reduction in downstream social assessment tasks.

Keywords: Social Assessment, Community, Kalman Filtering, Time Series