TopBench: A Benchmark for Implicit Prediction and Reasoning over Tabular Question Answering
For researchers in table-based reasoning, this benchmark highlights a critical gap in LLMs' ability to handle real-world predictive queries, showing that intent disambiguation is a prerequisite for accurate prediction.
The paper introduces TopBench, a benchmark with 779 samples for evaluating LLMs on implicit predictive reasoning over tables, finding that current models struggle with intent recognition and often default to simple lookups instead of predictive inference.
Large Language Models (LLMs) have advanced Table Question Answering, where most queries can be answered by extracting information or simple aggregation. However, a common class of real-world queries is implicitly predictive, requiring the inference of unobserved answers from historical patterns rather than mere retrieval. These queries introduce two challenges: recognizing latent intent and reliable predictive reasoning over massive tables. To assess LLMs in such Tabular questiOn answering with implicit Prediction tasks, we introduce TopBench, a benchmark consisting of 779 samples across four sub-tasks, ranging from single-point prediction to decision making, treatment effect analysis, and complex filtering, requiring models to generate outputs spanning reasoning text and structured tables. We evaluate diverse models under both text-based and agentic workflows. Experiments reveal that current models often struggle with intent recognition, defaulting to just lookups. Deeper analysis identifies that accurate intent disambiguation serves as the prerequisite for leading these predictive behaviors. Furthermore, elevating the upper bound of prediction precision requires the integration of more sophisticated modeling or reasoning capabilities.