A point-of-interest (POI) is a specific location in which someone is interested. In social network services such as Instagram, users share their experiences with text and photos, and link POIs to their posts. POIs can be utilized to understand user preferences and behavior. However, not all posts have POI information. In our study, we found more than half of the posts do not have POI information. The current state-of-the-art POI imputation model adds missing POI information. However, it relies on a conventional machine learning method that requires a substantial amount of laborious feature engineering. To address this problem, we propose DeepPIM, a deep neural POI imputation model that does not require feature engineering. DeepPIM automatically generates textual, visual, user, and temporal features from text, photo, user, and posting time information, respectively. For evaluating DeepPIM, we construct a new large-scale POI dataset. We show that DeepPIM significantly outperforms the current state-of-the-art model on the dataset. Our newly created large-scale POI dataset and the source code of DeepPIM are available at http://github.com/qnfnwkd/DeepPIM.
Bibliographical noteFunding Information:
This research was supported by the National Research Foundation of Korea (No. 2017R1A2A1A17069645 ).
© 2018 Elsevier Inc.
- POI imputation
- Social network
ASJC Scopus subject areas
- Control and Systems Engineering
- Theoretical Computer Science
- Computer Science Applications
- Information Systems and Management
- Artificial Intelligence