Hirohiko Suwa

Students in an online Introduction to Programming (CS 1) class suffer a severe disadvantage. In a face-to-face learning environment, teachers can see if students struggle with the assigned programming task through their body language or overall demeanor. In online learning, all that the teachers see is a small image of their student, if at all, and are thus unable to provide the needed intervention to help the student overcome their stumbling block. Our research investigates extrapolating student emotion through a web camera and code logs gathered while students work on programming exercises, with a hardware-based gaze tracker to serve as ground truth for software-based gaze tracking. Code, face, and gaze data, together with annotated emotion data gathered from our experiments, were used to train learning models via XGBoost. Our best model can predict a student’s emotional state at a precision, recall, and F1-score of 72.7%, 77.3%, and 74.9%, respectively. We achieved these results from analyzing 8 hours and 20 minutes of experimental data from 26 participants using only software-based gaze tracking, with statistically similar results to hardware-based gaze tracking to within a ±8% equivalence margin.

In cycle tourism, where cyclists can freely experience regional attractions, there exist Cyclist Points of Interest (CPoI) that cyclists prefer. However, these CPoIs are known only among cyclists and are not widely shared as general tourist information. This study proposes a semi-automated method for identifying and evaluating CPoIs using cyclist ride log data. The proposed method consists of three stages: automatic extraction using stop position estimation from GPS ride logs and Density-Based Spatial Clustering of Applications with Noise, facility identification using crowdsourcing, and evaluation using crowdsourcing. To evaluate the effectiveness of the proposed method, we collected ride data from 286 cyclists in Maebashi City, Gunma Prefecture, from April to December 2023, and successfully identified 48 CPoIs and assigned evaluation values to each CPoI. Expert validation of the identified CPoIs revealed high agreement rates: 89.6% for selection validity and 95.8% for the credibility of evaluation results. The detected CPoIs included many facilities often overlooked in conventional tourist destination selection, such as convenience stores, roadside stations, and cyclist-specialized cafes, demonstrating the practical utility of the proposed method.

Options trading is a type of financial trading that can offer significant profit opportunities depending on market conditions, but its profit and loss structure is complex. Traditional trading decisions are often made based on predictions of market indicators such as price fluctuations and volatility, and improving the predictive accuracy of these indicators may not necessarily lead to increased trading profits. In recent years, there has been growing interest in the use of supplementary information that can reflect the attention, expectations, and psychology of market participants, and it has been reported that text information such as newspaper articles and message board posts is related to short-term price formation and returns. Newspaper articles is useful for understanding events and social conditions that affect the economy and finance, and message board posts are useful for capturing investor interests, opinions, and emotions that cannot be captured by newspaper articles alone. Based on the above, this study proposes a framework for directly predicting profit opportunities in Nikkei 225 options trading using newspaper articles, message board posts, and financial time series, using supervised learning. Furthermore, under the same evaluation period, trading rules, and verification procedures, we systematically compare the forecasting performance and profit characteristics based on trading simulations for each combination of information sources, with and without them. In addition, we implemented a rule-based baseline based on predicting increases in the Nikkei 225 Volatility Index (VI) and a reinforcement learning baseline, and compared their profitability with the proposed model. Experimental results showed that in the options market, classification performance and profit characteristics based on trading simulations change depending on the presence or absence and combination of text information. In particular, when market data was combined with newspaper articles and message board posts, cumulative profits and losses in trading simulations tended to improve and maximum drawdowns tended to be smaller than when market data was used alone. The contribution of the information source also varied depending on the threshold, and it became clear that when market data and message board posts were combined, cumulative profits and losses improved in settings that aimed for high profit opportunities.