statistics

SPARQL dataset statistics

This subproject (part of QADO question answering dataset RDFizer) address the need of a good understanding of the characteristics of Knowledge Graph Question Answering (KGQA) benchmark datasets. For this purpose, a tool is provided here to create CSV files containing statistics about the RDFized data in a triplestore.

Table of Contents

• SPARQL dataset statistics
  • Requirements
  • Usage
  • Configuration
    • Script configuration
    • Statistics
      • Boxplot data
  • Default statistics

Requirements

• Docker needs to be installed on your system

• your triplestore needs to be accessible via SPARQL endpoint (cf. Section Configuration)

Usage

./createSpreadsheet.sh

This script creates new CSV files with all statistics and stores them following the pattern QADO-statistics-${statistic_name}-${current_date}.csv in the folder /tmp/QADO-statistics/ (by default). To change the output directory, you can provide a parameter to the script, e.g.:

./createSpreadsheet.sh myOutputDirectory

Configuration

Script configuration

You might edit the config.py to select the triplestore where your data is stored.

Statistics

A number of existing statistics are already implemented. To create your own statistics, just store your SPARQL queries into the queries directory. This tool creates a new CSV file for each SPARQL query file.

Boxplot data

If you want to generate a boxplot chart, you need to store your query with the file ending _boxplot.sparql. Additionally, the data for the boxplot chart needs to be selected via (GROUP_CONCAT(?YOUR_VARIABLE ; SEPARATOR = ",") AS ?concat).

Default statistics

The tool generates per default the following statistics for RDFized Question Answering benchmarks:

• Query length per benchmark

• Query modifiers per benchmark

• Question length per

  • answer type and benchmark (boxplot)

  • benchmark (boxplot)

  • language and benchmark (boxplot)

• Questions per

  • answer type

  • benchmark

  • language

  • language and benchmark

• Question type per benchmark

• Statistics of used resources inside the SPARQL queries per benchmark