Computable Social Patterns from Sparse Sensor Data

Dinh Phung, Brett Adams, Svetha Venkatesh


We present a computational framework to automatically discover high-order social patterns from very noisy and sparse location sensing information. The key idea in this work is to construct a social codebook, transform raw data into a 'book of life' — a collection of social pages, each in turn is a sequence of encoded footprints. Computable patterns are then defined as repeated structures found in these sequences. Tackling this problem poses several challenges such as how to construct the codebook, how to define patterns and extract them. In particular, extracting high-order patterns are known to be a very hard problem. We address these questions and propose a Latent Social theme Dirichlet Allocation (LSDA) model, which is a personalized version of the Ngram topic models proposed recently. This model can be viewed as a Baysian clustering method, jointly discovering temporal collocation of footprints and exploiting statistical strength across social pages. Alternatively it can be viewed as dimensionality reduction method where the reduced latent space corresponds to hidden social themes or plans of user's daily activities. Applying to a real-world noisy dataset collected over 1.5 years, we show that many useful and interesting patterns can be computed automatically.


Bibliography Navigation: Reference List; Author Index; Title Index; Keyword Index

Generated by sharef2html on 2011-04-15, 02:00:41.