Theme

• Sensing and analyzing real world activity and event from web data such as social media and sensors/logs of smartphone using machine learning.
• Its application for activity support such as intelligent personal assistant.

Location-related Social Media Analysis

Toponym (Location Name) Disambiguation Method for Microblogs using Time-Varying Location-related Words [PDF]

We propose a disambiguation method for toponyms using words related to the location. Conventionally, toponym ambiguation has been resolved by using nearby toponyms based on the hypothesis that geographically-closed toponyms are appeared frequently in the same context. In the case of microblogs, however, words other than toponyms are preferable to be used because short texts of microblogs have less information. To this end, we consider that microblogs have a topic related to the location and propose a method which uses words related to the location (”location-related words”) as disambiguators for each toponym. The location-related words are categorized into two groups. One is static words independent of seasonal variations and so on. The other is dynamic one which depends on seasonal variations etc. The dynamic location-related words reflect immediacy of microblog(i.e., the dynamic location-related words vary with time). We evaluated our proposed method by recall and precision using manually labeled data. The result showed that the recall of our proposed method is higher than that of the conventional method.

Point-of-Interest Official Twitter Account Extraction

Twitter event detection

User Behavior Analysis

Analyzing the Effects of Location-based Services for Location Prediction

Predicting user location is one of the most important topics in the field of location data analysis. While it is reasonable that human mobility is predictable for frequently visited places such as home and the workplace, location prediction for novel places is much more difficult. However, location-based services (LBSs) such as Pokémon Go can influence user destination and we can exploit this to achieve more accurate location prediction even for new locations. In this paper, we conduct an experiment that assesses the behavior difference of Pokémon Go users and non-users.Then we perform a simple machine learning experiment to analyze how Pokémon Go usage impacts location predictability. We assume that users who use the same LBS tend to visit similar locations. We find that the novel location predictability of Pokémon Go users is 53.8% higher than that of non-user.

Keiichi Ochiai, Yusuke Fukazawa, Wataru Yamada, Hiroyuki Manabe, Yutaka Matsuo: “Pokémon Go Influences Where You Go: Analyzing the Effects of Location-based Services for Location Prediction.” Asian CHI Symposium: Emerging HCI Research Collection in ACM Conference on Human Factors in Computing Systems (CHI) 2018. Montréal, Canada.