Daily Routines Inference Based on Location History

Discovering User’s Trends and Routines from Location Based Social Networks

“…For the temporal context, we discretize continuous values of time to obtain different time intervals inside a day. We introduced in [17] different strategies to create time intervals: by fixed size values (e.g., each interval of one-hour size), by means of data distribution (e.g., computing percentiles to obtain intervals with an equally distributed timestamps) or by means of the data density (e.g., applying a clustering algorithm). In this case, we choose a fixed size strategy with 6 intervals of 4 h. We named those intervals, from 00:00 to 23:59, as follows: late night, early morning, morning, afternoon, evening and night.…”

“…We also defined two more simple Bayesian networks along with the one with all features (BN-A): a Bayesian network with only the temporal context (BN-T) and another one with the spatial and temporal context (BN-ST). In addition of the Bayesian ones, we use other models for benchmarking: random choice of the user's location types (referred to as RND), the most frequent location type choice (referred to as MODE), a first-order Markov chain model (referred to as MC) and a Probabilistic Finite Automaton (equivalent to a Weighted Finite Automaton) for the transitions between location types depending on the time interval (see [17] for more details about this model). Semantic (i) The models' evaluation is based on the assessment of the efficiency for predicting the type of visited site, using information of the current context and previous states.…”

User identification from mobility traces

“…Thus, we group initial time stamps with small differences and facilitate the generation of a mobility model according to daily routines. We introduced in (Salomón et al 2017) three different strategies to address segmentation based on the time interval of the day: by fixed size values (e.g., each interval of one-hour size), by means of data distribution (e.g., computing percentiles to obtain intervals with an equally distributed number of records) or by means of data density (e.g., using a clustering algorithm). In Example 1 we present the three strategies applied to the same time stamps set.…”