This paper introduces a novel method for time series forecasting using de Bruijn Graphs (dBGs) to represent discretized time series data. Our approach involves (1) encoding time series as a dBG, (2) applying both novel and existing graph encoding algorithms (like struct2vec) to extract features from dBG, and (3) integrating these features into the TimesNet model to enhance short-term univariate forecasting accuracy. Empirical results on the M4 datasets show that our method preserves the dynamics of the time series while improving forecasting performance across various datasets.

Fantasy Sports has a current market size of ${\$}$27B and is expected to grow more than ${\$}$84B in less than a decade. The intent is to create virtual teams that somehow reflect what would happen if the constituent players actually played in a team. Using individual player and team statistics, models can be trained to predict an outcome. But fans are left wanting more. To achieve a more realistic outcome, aspects of what makes live teams win need to be included: (1) transforming player statistics to reflect their relative importance with respect to a player position; (2) team chemistry (TC). In this work, we show a novel characterization of relative position statistics and a new description of TC. Drawn from the NBA’s API, we form a data set to determine whether a fantasy team makes the playoffs using almost two dozen features, including TC. Various Machine Learning models are trained on this data and the best-performing model is offered to the users through a web service. Users can not only inspect fantasy teams and their TC but can also simulate their match-ups with existing 2023 NBA teams and utilize performance visualizations to help improve their team creation process. Our web service can be accessed at https://dalkilic.luddy.indiana.edu/fantasyleague/, and the source code can be found at https://github.com/gany-15/nbafan.

Social programming websites like GitHub and StackOverflow have become an increasingly important aspect of software development and the publicly available datasets provide a rich source of data for exploring challenging NLP problems. One such problem is emotion recognition. This work applies deep NLP methods for detecting emotions in StackOverflow content. Several BERT models were trained and fine-tuned on a small, sparse, hand-labeled and highly-imbalanced dataset of Stack-Overflow comments. Text augmentation techniques were used to balance the data and the model’s vocabulary was enhanced with common domain-specific terms and emoticons. Unsupervised post-training was applied on a large unlabeled StackOverflow dataset to learn representations for added vocabulary before fine-tuning on labeled data. The final model was benchmarked and compared to prior studies on the same dataset.

This study explores the application of deep learning to forecasting the state of Illinois sale tax receipts in ten categories: general merchandise, food, drinking and eating, apparel, furniture, building and hardware, automotive and filling stations, drugs and retail, agriculture and all others, and manufacturers. The state of Illinois has used traditional techniques of economic and tax receipt forecasting in order to project the amount of resources that it will have in order to finance its activities and debts. Such techniques are mostly linear and lack the ability to model more complex non-linear or long term dependencies. Recently, deep learning models have shown promising results in time series forecasting. In this study, we use two types of neural networks (a simple Multi-Layer Perceptron and a Long Short Term Memory network to forecast the state of Illinois sale tax receipts and compare the performance of both models against the more traditional autoregressive integrated moving Average model. Unfortunately, only limited tax receipt data is publicly made available by the state of Illinois which makes it particularly challenging to train a robust neural network model without overfitting. To address this data limitation, we propose to use a global model with an embedding layer for all ten tax categories. The empirical results show that the global Multi-Layer Perceptron model has the best performance in one step forecasting of Illinois sale tax receipts followed by the global Long Short Term memory model. On average, both neural network models outperformed the traditional Integraded Moving Average model.