Table QA

Task Overview

The goal of the Table QA (TQA) task is to extract the answer value to a question, given a single table from the securities reports and the question.

Task Definition

• Input: Question (question), a table from annual securities report HTML (table_id)

• Output: Value (value) and cell (cell_id)

• Evaluation: Accuracy

Dataset

An example from questions_tqa.json:

{

"question_train2": {

  "question": "大和ハウス工業の2019年の個別のShareholdersEquityにおける「自己株式の処分」を示すセルは？",

  "table_id": "S100ITAZ-0000000-tab1",

  "cell_id": "S100ITAZ-0000000-tab1-r1c1",

    "value": "1033000000000"
  },

}

The template is structured as follows.

• question_id (object)

  • question_{data type}{sequence number}

  • The data type is either train, valid or test.

  • The serial numbers are not consecutive for each task to prevent duplication of question texts between tasks.

• table_id (str)

  • {doc_id}-{HTML file number (7 digits)}-tab{serial number}

  • This is an ID assigned to each table in HTML files by the Task Organizer.

  • It is provided as the "table-id" attribute of the table tag in the HTML.

• cell_id (str)

  • {TableID}-r{row number}c{column number}

  • This is an ID assigned to each cell in HTML files by the Task Organizer.

  • It is provided as the "cell-id" attribute to the th and td tags in the HTML.

• value (str)

  • This is the raw value obtained from the XBRL report.

  • The unit is considered if the value is a number, so please note that it will not exactly match the string contained in the HTML cell.