CataTQA: A Benchmark for Tool-Augmented LLM Question Answering over Heterogeneous Catalysis Tables

Despite their success in general question answering, large language models (LLMs) struggle with hallucinations and inaccurate reasoning in scientific domains. A major challenge stems from experimental data, which are often stored in external sources like supplementary materials and domain-specific databases. These tables are large, heterogeneous, and semantically complex, making them difficult for LLMs to interpret. While external tools show promise, current benchmarks fail to assess LLMs' ability to navigate this data—particularly in locating relevant tables, retrieving key columns, interpreting experimental conditions, and invoking tools. To address this gap, we introduce CataTQA, a new benchmark for catalytic materials. CataTQA features an automated dataset framework and four auxiliary tools. We evaluate tool-enhanced LLMs across five dimensions: table location, column retrieval, condition analysis, tool calling, and question answering, identifying their strengths and weaknesses. Our work sets a new benchmark for evaluating LLMs in scientific fields and paves the way for future advancements. All data and code are publicly available on GitHub.

Dataset Field Description

• question:A table question.
• refer_dataset:Generate a reference dataset of questions and answers.
• column namesThe column name used to generate the problem.
• condition_column:The column names that need to be filled in to generate the problem.
• answer_column:Column name of the answer.
• condition:Conditions contained in the question.
• answer:Answers to questions.
• tool:Tools for answering questions.
• level:The level of the problem.
• question description:Question type description.
• refer_template:Template question.

Example

{
    "question": "Identify the material ID linked to a total energy per atom of -4.093124536666667.",
    "refer_dataset": "table67",
    "column names": ["energy_per_atom", "material_id"],
    "condition_column": ["energy_per_atom"],
    "answer_column": ["material_id"],
    "condition": {"energy_per_atom": "-4.093124536666667"},
    "tool": "search_value",
    "answer": {"material_id": "2dm-6"},
    "level": "simple",
    "question description":"In a tabular data structure, locate the cells that meet the requirements.",
    "refer_template": "Identify the material ID linked to a total energy per atom of {}."
}