[2C5-OS-7b] OS-7 (2)

[2C5-OS-7b-03] User Clustering based on Distributed Representations for Understanding Tourist Behaviors

Motoi Kubo^1,2, 〇Hiroki Tanaka^1,2, Satoshi Nakamura^1,2 (1. RIKEN Center for Advanced Intelligence Project Tourism Information Analytics Team, 2. Nara Institute of Science and Technology)

Keywords:Tourism behavior analysis, Spatio-temporal data, Distributed representation, Bidirectional LSTM, Hierarchical clustering

In order to promote inbound tourism, we need to analyze behaviors and destinations of tourists, and understand their trends.
In this study, we attempted to cluster tourists' behaviors using a time-series distributed representations. A previous work used the Long-short term memory (LSTM) to predict tourists' next visiting places. In this study, we extended it to the Bi-directional LSTM (Bi-LSTM). To obtain tourist clusters, we calculated the distance of representation vectors derived from the LSTM and the Bi-LSTM. Our results showed that the LSTM grouped tourists who visited similar places, and the Bi-LSTM could also obtain tourist clusters who visited places in reversed order of routes.