Detecting mental illness from short social media posts is challenging because these texts are often brief, fragmented, and lack explicit descriptions of the user’s mental state. Prior studies using encoder-based models such as BERT show promise but struggle when key contextual information is missing. To address this, we propose a method that augments posts with interpretive sentences generated by MentaLLaMA-chat, a generative model specialized in mental health, and fine-tunes BERT on the augmented dataset. We curated 1,525 Japanese posts containing the word “mental” (in katakana) from X (formerly Twitter) and manually annotated them according to Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition criteria, labeling 557 posts as positive and 968 as negative. Our method improved recall by 2.4 percentage points compared to models trained on the original posts alone, while maintaining comparable accuracy and precision. Shapley Additive Explanations analysis revealed that tokens introduced by the interpretive sentences—including both negative and positive expressions—enhanced the model’s ability to identify mental-distress posts. These results demonstrate that generative-model-based text augmentation effectively provides additional context, enabling more accurate detection of mental illness indicators in short, ambiguous social media posts.

This study addresses the optimal inventory management problem for new smartphone products as an effective example of a supply chain with a short product life cycle. The determination of the optimal inventory level leads to a reduction of lost opportunities and defective inventory, which is an important issue from a profit improvement perspective. Mathematical optimization and reinforcement learning approaches have been proposed for inventory management; however, most of these approaches focus on products that are regularly sold over a long period. Thus, when the target is a new product, it is difficult to optimize inventory control from its day of release due to a lack of sufficient data for learning. To solve this problem, we focus on model-based deep reinforcement learning with high sample efficiency and propose an inventory management method for new products that combines model learning in an offline environment and planning in an online environment. Simulations using real-world historical sales datasets demonstrate that the proposed method outperforms existing methods in terms of profitability, efficiency, and customer satisfaction. In particular, the proposed method improves total rewards and inventory turnover by ¿5% each than the heuristic method while maintaining the same stock-out rate. In addition, the results demonstrate that the proposed method can maintain stable inventory control for multiproduct and multistore supply chains.

There are more than 2,000 Karaoke stores in Japan and approximately 50 million users visit every year for a singing experience. For Karaoke, recommendations of the next songs play a significant role in leading customers to explore more songs and make the most of time in Karaoke. However, few recommendation functions are provided due to the difficulty and uniqueness of the Karaoke environment, in which unknown groups choose songs based on with whom they are sharing the Karaoke experience and the group mood; solid information of these factors is not available. In this paper, we propose a session-based multivariate recurrent neural network (RNN) to consider an occasion for a Karaoke store visit and any possible mood changes; these factors are assumed to be embedded in the sequential data. The model uses metadata from previously played songs, the conditions present, and performs several tasks that predict the songs and artists to be played at time t to t+n. It outperformed the item-kNN algorithm for prediction tasks (offline) of the next song by 127.94% regarding MAP@20 and the next artist by 142.8% regarding MAP@10. In addition, we analyzed the performance of the model for three types of groups, namely individuals, older groups, and younger groups that were manually categorized from played songs, and it outperformed the item-kNN algorithm by 89.59%, 483.11%, and 50.88% regarding MAP@20 respectively. The model also outperformed the baselines for all positions in a session, which demonstrates that it has the potential to quickly capture the preferences of a target group and to recognize mood changes from the sequence. We have launched a recommender system that is designed for Japanese Karaoke stores. The system has demonstrated that the model used in this system is also appliable to business applications due to its ability to handle vast numbers of requests in a reasonable time and enhance user experience by helping to choose more diverse songs faster.

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.

Due to the spread of smartphones, mobile applications (app) have been widely used in daily life. Several apps which provide real-world related information (called "local apps") are useful for not only tourist but also residents. There is a category that seems to contain local apps in app market such as "Travel & Local" in Google Play, but many local apps are categorized into other categories. Thus, we present a method to classify local apps based on app market data using deep neural network (DNN). We leverage the fact that each app is manually labeled by developer to pre-train the DNN. In addition, we create features from an app market data because app markets involve multi-modal data such as app name, category and number of installs. We conducted an experiment on a real-world dataset crawled from Google Play to validate the effectiveness of the proposed method. Our evaluation shows that the proposed method outperforms the baseline method by 5.5% regarding F1 score.

Task-models concretize general requests to support users in real-world scenarios. In this paper, we present an IR based algorithm (IRTML) to automate the construction of hierarchically structured task-models. In contrast to other approaches, our algorithm is capable of assigning general tasks closer to the top and specific tasks closer to the bottom. Connections between tasks are established by extending Turney’s PMI-IR measure. To evaluate our algorithm, we manually created a ground truth in the health-care domain consisting of 14 domains. We compared the IRTML algorithm to three state-of-the-art algorithms to generate hierarchical structures, i.e. BiSection K-means, Formal Concept Analysis and Bottom-Up Clustering. Our results show that IRTML achieves a 25.9% taxonomic overlap with the ground truth, a 32.0% improvement over the compared algorithms.

A topic model capable of assigning word pairs to associated topics is developed to explore people's activities. Considering that the form of word pairs led by verbs is a more effective way to express people's activities than separate words, we incorp

We propose a topic model capable of generating tri-layer clusters, each of which is composed of a topic layer, an activity layer and a word layer. The objective is to better predict activities involved in documents by considering general topics of the activities for clustering. The proposed model is a supervised topic model based on the Latent Dirichlet Allocation (LDA). As a follow-up study of word-pair generation LDA (wpLDA) model, the model introduces the topic-specific activity distribution as an external input, with an activity node inserted into the main generation thread. In addition, we refer to D. Ramage et al.'s one-to-one correspondence to directly learn word-activity tags. An experiment was conducted to prove the feasibility of this model. We chose ten top-listed activities from the wish clusters obtained by the previous wpLDA research, and used each as the key words to extract thirty tweets for training and five for testing, respectively, tagging the tweets with the corresponding activities. By applying the proposed model, we obtained the expected tri-layer clusters in the training phase. Then, in the testing phase, we utilized the activity-specific word distribution derived from the training results to learn the activities of the testing documents. The Stanford Classifier was put forward as the control group, and the activity prediction accuracy demonstrates that the proposed model exhibits the superiority in multi-activity prediction.

We demonstrate a generative model that incorporates word-pair connection into the smoothed LDA model to intuitively discover people's wish related activities. The widely used model, LDA topic model, generally generates clusters in the form of separate words. However, this form is not intuitive enough to express people's activities. Therefore, we consider the word-pairs led by verbs can better describe users' intentions and activities, and we prefer to present this collocation under topics as the clustering results. We mathematically present the relatedness between verbs and non-verb words through association rule, and build the physical connection of word-pairs and possible topics. By incorporating the connection lattice into the smoothed LDA, the word-pair LDA model is created. In the experiments, Twitter posts about “new year's resolutions” were chosen as the data source. The results show that the proposed model performs well on perplexity, and presents excellent intuitive character.

Mobile devices are becoming more and more difficult to use due to the sheer number of functions now supported. In this paper, we propose a menu customization system that ranks functions so as to make interesting functions, both frequently used functions and rarely used functions, easy to access. Concretely, we define the features of phone functions by extracting keywords from the manufacturer's manual, and propose the method that ranks the functions based on user operation history by using Ranking SVM (Support Vector Machine). We conduct a home-use test for one week to evaluate the efficiency of customization and the usability of menu customization. The results show that the average rank of used functions on the last day of the test is half of that of first day and almost 70 % of the users are satisfied with the ranking provided by menu customization and the usability of menus. In addition, interviews show that automatic mobile menu customization is more appropriate for mobile phone beginner rather than the master users.

An algorithm that acquires the intermediate goals between the initial and goal states is proposed for an agent executing multiple tasks. We demonstrate the algorithm in the problem of rearranging multiple objects. The result shows that the moving distance to transfer the entire objects to their goal configuration is 1/15 of that without using intermediate goals. We experiment using a real robot to confirm that the intermediate goal can be adapted to a real environment. Our experimental results showed that an agent could adapt the intermediate goals, which were acquired in the simulation, to the experimental environment.

In this paper, region exploration path planning algorithm is proposed. In order for a mobile robot to perform this task, appropriate measures with the shape of the working environment, which may be intricate or curved, is necessary. In addition, a robot must be able to react and be flexible when confronted with obstacles. With this algorithm, these challenges can be met by approximately expressing the working environment in grid points and regenerating the path using one that was planned beforehand. Simulations are used to demonstrate proposed exploration path planning and re-planning algorithm.

In this paper, we propose cooperative exploration method for mobile robots in a working area. The method will generate and share the minimal-cost path for each mobile robot. Each mobile robot goes through several observation points to accomplish the exploration. The observation points should be arranged at a fixed distance from at least on corresponding observation point. In addition, the number of observation points and the lengths of paths for each mobile robot are to be minimized. The proposed method should have the efficiency in computational cost concurrently with the adaptability to dynamic environmental changes.

This paper offers a proposal for an algorithm of controlling a mobile robot that searches for and rearranges objects with unknown locations and shape. In this paper, we divide the task into two parts: exploration task and rearrangement task. The algorithms for each part of the task are presented with respect to the effectiveness of the path length and computational cost. Additionally integration algorithm that effectively combines exploration and rearrangement is presented. Experiments with a real robot are conducted to demonstrate the effectiveness of the proposed algorithm.