metadata

language:
  - en
license: mit
tags:
  - tables
  - benchmark
  - qa
  - llms
  - document-understanding
  - multimodal
pretty_name: Human Centric Tables Question Answering (HCTQA)
size_categories:
  - 10K<n<100K
task_categories:
  - question-answering
task_ids:
  - document-question-answering
  - visual-question-answering
annotations_creators:
  - expert-generated
configs:
  - config_name: default
    data_files:
      - split: train
        path: train.parquet
      - split: validation
        path: val.parquet
      - split: test
        path: test.parquet
dataset_info:
  - config_name: default
    features:
      - name: table_id
        dtype: string
      - name: table_csv_path
        dtype: string
      - name: table_image_url
        dtype: string
      - name: table_image_local_path
        dtype: string
      - name: table_csv_format
        dtype: string
      - name: table_properties
        dtype: string
      - name: question_id
        dtype: string
      - name: question
        dtype: string
      - name: question_template
        dtype: string
      - name: question_properties
        dtype: string
      - name: answer
        dtype: string
      - name: prompt
        dtype: string
      - name: prompt_without_system
        dtype: string
      - name: dataset_type
        dtype: string
    description: >
      Human Centric Tables Question Answering (HCTQA) is a benchmark designed
      for evaluating the performance of LLMs on question answering over complex,
      real-world and synthetic tables. This dataset contains both real-world and
      synthetic tables with associated images, CSVs,  and structured metadata.
      The dataset includes questions with varying levels of complexity, 
      requiring models to handle reasoning across complex structures, numeric
      aggregation, and context-dependent  understanding. The `dataset_type`
      field indicates whether a sample is from the real world data sources
      (`realWorldHCTs`) or synthetically created (`syntheticHCTs`).

HCT-QA: Human-Centric Tables Question Answering

HCT-QA is a benchmark dataset designed to evaluate large language models (LLMs) on question answering over complex, human-centric tables (HCTs). These tables often appear in documents such as research papers, reports, and webpages and present significant challenges for traditional table QA due to their non-standard layouts and compositional structure.

The dataset includes:

2,188 real-world tables with 9,835 human-annotated QA pairs
4,679 synthetic tables with 67,500 programmatically generated QA pairs
Logical and structural metadata for each table and question

📄 Paper: [Title TBD]
The associated paper is currently under review and will be linked here once published.

📊 Dataset Splits

Config	Split	# Examples (Placeholder)
RealWorld	Train	7,500
RealWorld	Test	2,335
Synthetic	Train	55,000
Synthetic	Test	12,500

🏆 Leaderboard

Model Name	FT (Finetuned)	Recall	Precision
Model-A	True	0.81	0.78
Model-B	False	0.64	0.61
Model-C	True	0.72	0.69

📌 If you're evaluating on this dataset, open a pull request to update the leaderboard.

Dataset Structure

Each entry in the dataset is a dictionary with the following structure:

Sample Entry

{
  "table_id": "arxiv--1--1118",
  "table_info": {
    "table_csv_path": "../tables/csvs/arxiv--1--1118.csv",
    "table_image_url": "https://hcsdtables.qcri.org/datasets/all_images/arxiv_1_1118.jpg",
    "table_image_local_path": "../tables/images/arxiv--1--1118.jpg",
    "table_properties": {
      "Standard Relational Table": true,
      "Row Nesting": false,
      "Column Aggregation": false,
      ...
    },
    "table_formats": {
      "csv": ",0,1,2\n0,Domain,Average Text Length,Aspects Identified\n1,Journalism,50,44\n..."
    }
  },
  "questions": [
    {
      "question_id": "arxiv--1--1118--M0",
      "question": "Report the Domain and the Average Text Length where the Aspects Identified equals 72",
      "gt": "{Psychology | 86} || {Linguistics | 90}",
      "question_properties": {
        "Row Filter": true,
        "Aggregation": false,
        "Returned Columns": true,
        ...
      }
    }
    ...
  ]
}

Ground Truth Format

Explain the GT format here
Example: {value1 | value2} || {value3 | value4}

Table Properties

Property Name	Definition
Standard Relational Table	TBD
Multi Level Column	TBD
Balanced Multi Level Column	TBD
Symmetric Multi Level Column	TBD
Unbalanced Multi Level Column	TBD
Asymmetric Multi Level Column	TBD
Column Aggregation	TBD
Global Column Aggregation	TBD
Local Column-Group Aggregation	TBD
Explicit Column Aggregation Terms	TBD
Implicit Column Aggregation Terms	TBD
Row Nesting	TBD
Balanced Row Nesting	TBD
Symmetric Row Nesting	TBD
Unbalanced Row Nesting	TBD
Asymmetric Row Nesting	TBD
Row Aggregation	TBD
Global Row Aggregation	TBD
Local Row-Group Aggregation	TBD
Explicit Row Aggregation Terms	TBD
Implicit Row Aggregation Terms	TBD
Split Header Cell	TBD
Row Group Label	TBD

Question Properties

Property Name	Definition
Row Filter	TBD
Row Filter Condition Type Lookup	TBD
Row Filter Condition Type Expression	TBD
Row Filter Involved Columns Single	TBD
Row Filter Involved Columns Multiple	TBD
Row Filter Max Depth Of Involved Columns	TBD
Row Filter Retained Rows Single	TBD
Row Filter Retained Rows Multiple	TBD
Row Filter Num Of Conditions	TBD
Returned Columns	TBD
Returned Columns Project On Plain	TBD
Returned Columns Project On Expression	TBD
Returned Columns Max Depth	TBD
Returned Columns Expression In Table Present	TBD
Returned Columns Expression In Table Not Present	TBD
Returned Columns Num Of Output Columns	TBD
Yes/No	TBD
Aggregation	TBD
Aggregation Type Sum	TBD
Aggregation Type Avg	TBD
Aggregation Grouping Global	TBD
Aggregation Grouping Local	TBD
Rank	TBD
Rank Type	TBD