Human mobility prediction is a crucial research topic with wide-ranging applications, from urban planning to infectious disease forecasting. While the proliferation of GPS-enabled devices has made large-scale trajectory data available, a lack of standardized open-source datasets has hindered fair comparisons. Recent winning solutions in Human Mobility Prediction Challenges have often leveraged powerful models like BERT and GPT, but these approaches primarily focus on predicting individual movements, often neglecting the broader context of collective human flow.

In the COVID-19 epidemic, balancing a trade-off between preventing the spread of infection and maintaining economic activity is a global challenge. Based on the idea that avoiding crowds leads to the prevention of the spread of infection, we propose to leverage a dynamic pricing method to level out congestion with an aim to balance the trade-off between preventing the spread of infection and economic activity. In our method, reward points are provided according to the degree of congestion in stores to encourage customers to visit stores at less crowded times to avoid crowds. Since store congestion is greatly affected by movement restrictions such as a state of emergency, we propose a demand prediction model that takes into account the biases of the data acquisition circumstances. In an offline evaluation, we validated the effectiveness of the proposed unbiased demand prediction model based on the data from an actual campaign conducted for more than 7 months in Kyushu University. The evaluation results showed that our unbiased model reduced the prediction error by up to relatively 25.0% compared with the model that does not consider biases. Our system has been deployed in our closed service since December, 2021. Online evaluation result showed that our application improved conversion rate by 12.0% and reduced cost per acquisition by up to 11.6%.

Many managers and human resource departments transfer a person in an attempt to increase the performance of an organization or team. When we define performance as team efficiency, the performance is influenced by the density of team communications. However, whether or not the candidate transferee will actually increase the density of team communications is an unknown. In this paper, we propose a new approach that estimates whether or not a person who joins a team will improve the density of team communications based on an instant messaging system (IMS). In the proposed approach, we embed people in feature space using graph neural networks. In this embedding process, we do not use the content of text communications but utilize only communication graph architectural information that expresses who is talking to whom and how often. Additionally, the proposed approach does not require a questionnaire to indicate the density of team communication as in some previous studies. In the proposed approach, we develop a machine learning model classifying whether or not a transferee will improve the density of team communications. The model classifies transferees at an accuracy of 0.57 and precision of 0.58. Since this model does not use text contents from the IMS, it is valuable in actual business situations.

In a disaster situation, local and municipal governments need to distribute relief supplies and provide administrative support to evacuee. Although people are supposed to evacuate to evacuation shelters designated by local governments, some people take refuge at non-designated facilities, called non-designated evacuation shelters, due to unavoidable circumstances such as damages on the access route to designated evacuation shelters. Upon occurrence of a disaster, therefore, it is necessary for the local governments to quickly find the locations of non-designated evacuation shelters. In this paper, we propose a method to detect non-designated evacuation shelters based on autoencoder (AE)-based anomaly detection using real-time population dynamics generated from operation data of cellular phone networks. We assume that reconstruction errors of an AE model include both the errors due to characteristic differences between locations and the errors due to anomalies in population dynamics. Thus, we propose to use the ratio of the reconstruction error before and after the earthquake to determine the threshold of anomaly detection. We evaluate the performance of the proposed method on data from three actual earthquakes in Japan. The evaluation results show that our reconstruction-error-based approach can achieve better accuracy for the actual disaster data compared to a baseline method that exploits statistical anomaly detection.

Storm and flood disasters such as typhoons and torrential rains are becoming more intense and frequent. The national government and municipalities must respond to such natural disasters as soon as possible. When the scale of damage is large; however, it takes much time to investigate the severity of damage, and the initial response can be delayed. If we could precisely and rapidly estimate the severity of damage for each city at an early stage, the national government would be able to better support the municipalities, and consequently respond quickly to help citizens. In this paper, we propose a novel approach to estimate the severity of disaster damage within a short time period after a disaster occurs by exploiting real-time population data generated from cellular networks. First, we investigate the relationship between real-time population data and the severity of damage. Then, we design a Graph Convolutional Networks for Disaster Damage Estimation, called D2E-GCN, which fully exploits the directed and weighted characteristics of human mobility graph. We conduct an offline evaluation on real-world datasets including two typhoons that hit Japan. The evaluation results show that the proposed method outperforms baseline methods which do not consider the graph structure of cities, and the proposed method can estimate the severity of damage approximately 48 hours after typhoons passed. Moreover, we find the experimental insight that the estimation performance can be significantly affected by the graph construction method for GCN models.

We propose to apply graph convolutional networks to a novel domain, mobile app usage. User state estimation from smartphone usage has attracted attention thanks to the prevalence of smartphones. Basic statistics such as the sum of the number of used apps and usage duration have been used for estimating the user state. However, the accuracy of user state estimation can be improved by considering the sequence of apps used and the relationship between each app based on graph convolutional networks. In this paper, we proposed a representation learning method for mobile app usage using graph convolutional networks. We evaluate the proposed method through comparison to another deep learning approach, i. e., long short-term memory, for a classification problem.

Billions of people are using smartphones everyday and they often face problems and troubles with both the hardware as well as the software. Such problems lead to frustrated users and low customer satisfaction. Developing an automatic machine learning-based solution that would detect that the user has a problem and would engage in troubleshooting has the potential to significantly improve customer satisfaction and retention. Here, we design and implement a system that based on the user's smartphone activity detects that the user has a problem and requires help. Our system automatically detects a user has a problem and then helps with the troubleshooting by recommending possible solutions to the identified problem. We train our system based on large-scale customer support center data and show that it can both detect that a user has a problem as well as predict the category of the problem (89.7% accuracy) and quickly provide a solution (in 10.4ms). Our system has been deployed in commercial service since January, 2019. Online evaluation result showed that machine learning based approach outperforms the existing method by approximately 30% regarding the user problem solving rate.