This paper presents a novel framework for dynamic circle recommendation for a query user at a given time point from historical communication logs. We identify the fundamental factors that govern interactions and aim to automatically form dynamic circle for scenarios, such as, who should I dial to in the early morning? whose mail would I reply first at midnight? We develop a time-sensitive probabilistic model (TCircleRank) that not only captures temporal tendencies between the query user and candidate friends but also blends frequency and recency into group formation. We also utilize the model to support two types of dynamic circle recommendation: Seedset Generation: single-interaction suggestion and Circle Suggestion: multiple interactions suggestion. We further present approaches to infer relevant time interval in determining circles for a query user at a given time. Experimental results on Enron dataset, Call Detail Records and Reality Mining Data prove the effectiveness of dynamic circle
recommendation using TCircleRank.
The 18th Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD 2014), pp.25-37