Auditing financial documents is a very tedious and time-consuming process. As of today, it can already be simplified by employing AI-based solutions to recommend relevant text passages from a report for each legal requirement of rigorous accounting standards. However, these methods need to be fine-tuned regularly, and they require abundant annotated data, which is often lacking in industrial environments. Hence, we present ZeroShotALI, a novel recommender system that leverages a state-of-the-art large language model (LLM) in conjunction with a domain-specifically optimized transformer-based text-matching solution. We find that a two-step approach of first retrieving a number of best matching document sections per legal requirement with a custom BERT-based model and second filtering these selections using an LLM yields significant performance improvements over existing approaches.
Behavioral profiling in digital games with persistent online worlds are vital for a variety of tasks ranging from understanding the player community to informing design and business decisions. In this paper behavioral profiles are developed for the online multiplayer shooter/role-playing game Destiny, the most expensive game to be launched to date and a unique hybrid incorporating designs from multiple traditional genres. The profiles are based on playstyle features covering a total of 41 features and over 4,800 randomly selected players at the highest level in the game. Four clustering models were applied (k-means, Gaussian mixture models, k-maxoids and Archetype Analysis) across the two primary game modes in Destiny: Player-versus-Player and Player-versus-Environment. The performance of each model is described and cross-model analysis is used to identify four to five distinct playstyles across each method, using a variety of similarity metrics. Discussion on which model to use in different circumstances is provided. The profiles are translated into design language and the insights they provide into the behavior of Destiny's player base described.
Time plays an important role regarding user preferences for products. It introduces asymmetries into the adoption of products which should be considered in the context of recommender systems and business intelligence. We therefore investigate how temporally asymmetric user preferences can be analyzed using a latent factor model called Decomposition Into Directional Components (DEDICOM). We introduce a new scalable hybrid algorithm that combines projected gradient descent and alternating least squares updates to compute DEDICOM and imposes semi-nonnegativity constraints to better interpret the resulting factors. We apply our model to analyze user churn and migration between different computer games in a social gaming environment.