In FinArg-1, the tasks centered on argument-based sentiment analysis and argument relation identification. FinArg-2 shifted the focus to a more fine-grained examination of arguments, specifically targeting argument-based temporal inference. As the final shared task in the FinArg series, we now invite participants to leverage insights from both FinNum and FinArg to evaluate the quality of arguments, with particular attention to financial forward-looking statements.
In FinArg-3, we aim to investigate the quality of financial arguments, continuing to work with data from earnings calls, analyst reports, and social media. This task is divided into three subtasks. The first subtask assesses the quality of arguments in earnings calls from a linguistic perspective, focusing on four dimensions. The second subtask evaluates causal reasoning from the perspective of forecasting skill—specifically, which inferences are more likely to come true in the future. For this, we conduct pairwise comparisons of social media posts, aiming for models to recommend the higher-quality post to users. The third subtask involves evaluating analyst reports under a single-scenario setting. Here, the goal is to assess whether the forward-looking statements made by analysts are likely to materialize based on their causal reasoning.
Overall, we aim to discuss and evaluate the financial argument, building upon the findings from the previous five tasks during the evaluation process. Given the current strong performance of LLMs and their ability to generate diverse forms of analysis, we also intend to leverage the findings from FinArg-3 to support future research in evaluating LLM-generated text.
FinArg-1 in NTCIR-17 (2023): Fine-grained Argument Understanding in Financial Analysis
FinArg-2 in NTCIR-18 (2025): Temporal Inference of Financial Arguments