Loc2Vec: Learning location embeddings with triplet-loss networks Sentiance

At Sentiance, we developed a platform that takes in smartphone sensor data such as accelerometer, gyroscope and location information, and extracts behavioral insights. Our AI platform learns about the user's patterns and is able to predict and explain why and when things happen, allowing our customers to coach their users and engage with them in the right way, at the right time. An important component of our platform is the venue mapping algorithm. The goal of the venue mapper is to figure out what venue you are visiting, given an often inaccurate location measurement coming from the smartphone's location subsystem. Figure 1: Left: Venue mapping means estimating which of the neighboring venues a user was actually visiting. Right: Human intuition helps us to quickly discard unlikely venues, such as the lifeguard station when a user is visiting the beach. Although venue mapping is a difficult problem altogether and will be material for a future blog post, a simple sense of human intuition based on the surrounding geography of the area often goes a long way.