Generation of Knowledge Graph

The performance of Kore.ai’s Knowledge Graph is based on the proper organization of the Knowledge Graph based upon key domain terms and establishing a hierarchy.

Building the FAQs might be easy when you start fresh with the Knowledge Graph, but in case you have a list of questions-answer pairs converting the same into a fully functional Knowledge Graph might be a tedious task.

Kore.ai’s Knowledge Graph Generator automatically extracts terms from FAQs, defines the hierarchy between these terms, and also associates the FAQs to the right terms. You can then import the output file from the generator to your Bot’s Knowledge Graph, without having to worry about the hierarchy. You can also edit the hierarchy, after import, to suite your needs. It is highly recommended you review and make changes as the Knowledge Graph generated is a suggestion alone.

NOTE: Knowledge Graph Generator is available from ver7.1 of the platform.

The Kore.ai Knowledge Graph Generator is hosted on the Kore GitHub repository. This document gives the steps in the installation and usage of the generator.

Prerequisites

• Python 3.6: The Knowledge Graph Generator requires python 3.6. This can be downloaded from here.
• Virtual Environment: It is advised to use virtual environment, instead of installing requirements in the system directly. Follow the steps mentioned here to setup virtual environment.
• For Windows developers:
  • Microsoft Visual C++ Build Tools – tested with ver 14.0
  • Windows 10 users should install Windows 10 SDK. You can download it from here.
  • Operating system should be upto date for seamless installation of requirements. Some libraries like SpiCy (internal dependency) need specific DLLs which are available in the latest updates.
• A file containing the FAQs in json or csv format. You can obtain this file in two ways:
  • Export the Knowledge Graph from Kore.ai Bot Builder platform, see here for how.
  • Build the Knowledge Graph in a tabular form, with questions in the first column and answers in the corresponding second column and saving the file in csv format.

Configuration

1. Download the KnowledgeGraphGenerator from Kore.ai GitHub: https://github.com/Koredotcom/KnowledgeGraphGenerator.
2. Extract the zip file into a folder and open the command prompt from that generator folder.
3. Activate virtual environment: Execute the following command replacing the placeholders with actual values to activate the virtual environment:
   • For windows:
     <virtual_environments_folder_location>/<virtualenv_name>/Scripts/activate
   • For Unix/MacOS:
     <virtual_environments_folder_location>/<virtualenv_name>/bin/activate.

   Once the virtual environment is activated, you should see the virtual environment name at the start of every command in the console.

4. Install the requirements: Run the following command from your project root directory (KnowledgeGraphGenerator) in the virtual environment to install requirements
   pip install -r requirements.txt
   You can verify the installation by running the following command and ensuring that the list contains all the components mentioned in requirement.txt file.
   pip list
5. Download spacy english model: Run following command to download spaCy, the NLP model.
   python -m spacy download en

Execution

Now that you have the pre-requisites and have configured the Knowledge Graph Generator, let us see how to generate the Knowledge Graph.

The following command executes the generator:

python KnowledgeGraphGenerator.py --file_path <INPUT_FILE_PATH> --type <INPUT_FILE_TYPE> --language <LANGUAGE_CODE> --v <true/false>

Let us look at each of the options:

Output

Output JSON file is generated and placed under project root directory with name ao_output.json

The output JSON file can directly be imported to Knowledge Graph in Bot. See here for steps in Importing Knowledge Graph.

NOTE: When you try to import the Knowledge Graph it will replace the existing one. We recommend you take a back up before importing.