wkstools

wkstools is a small convenience library that provides utilities to efficiently work with entities and relations provided by IBM Natural Language Understanding.

Intuition

When using IBM Watson Knowledge Studio (WKS) models in IBM Natural Language Understanding (NLU), in most cases the resulting JSON requires post-processing to work with the extracted entities and relations. This is where wkstools comes to help:

Let's assume you have trained a machine learning model in IBM Watson Knowledge Studio that recognizes relations between quantifiers, numbers and units. "I want a smartphone that weighs no more than 160 g."

If we want to build a tool that allows natural language searches for smartphone against a product database, you need to understand the relevant concepts in a structured manner. For the above example it will be helpful to extract the relations that point from the operator (no more than) to the (value) and the unit (g).

See the intuition.py and nlu_response.json in the exmaples folder to have a running example.

Requirements

• Python 3.6+
• Pydantic

Installation

$ pip install wkstools

Usage

Parse entities and relations

To parse the NLU JSON response retrieved from IBM Natural Language Understanding, use:

import wkstools

# Your NLU JSON response to process
nlu_response = '{..., "relations": [{"type": "specifiesValue", ...], "entities": [...]}' 

entities = wkstools.parse_entities(nlu_response)
relations = wkstools.parse_relations(nlu_response)

See the entity and relation models for available fields.

Access specific relations

import wkstools

# Your NLU JSON response to process
nlu_response = '{..., "relations": [{"type": "specifiesValue", ...], "entities": [...]}' 

relations = wkstools.parse_relations(nlu_response)

value_relations = wkstools.get_relations_by_type(relations, "specifiesValue")

Testing

To lint and test, run:

$ ./scripts/test.sh

To get the html coverage report run:

$ ./scripts/test-cov-report.sh

License

This project is licensed under the terms of the MIT license.