ie-viz 0.3.0

Visualization tool for NLP information extraction: Named entity recognition, Entity attribute extraction, and Relation extration.

Table of Contents

• Overview
• Prerequisite
• Installation
• Quick Start
• Examples
• User Guide

Overview

The ie-viz is a lightweight tool for in-line information extraction (IE) visualization. It supports customizable colors for named entity marking, tooltip for entity attributes display, and relation path that links related named entities. Current version has built-in light and dark themes while customization is available through the CSS. The ie-viz can be deployed as a Flask App that runs on a host:port. It can also be rendered as HTML and display in Browser or Interactive Python environments (e.g., Jupyter Notebook).

The filtering feature supports OR and AND logic for all available entity attributes. The table panel (collapsible) displays the selected entities, attributes, and relations.

Prerequisite

• python ^3.11
• flask >=2.3

Installation

Python package is available on PyPi.

pip install ie-viz

Quick-start

We use a short sample News article synthesized by GPT-4o to demo this quick-start.

text = """
On Monday, the tech giant, Innovia Corp, announced the launch of its new flagship smartphone, the Innovia XPro. 
The device, which comes in both 128GB and 256GB storage variants, is powered by the latest Octa-core Quantum processor 
and features a 6.5-inch AMOLED display. According to the CEO, Lisa Martin, this release marks a significant milestone for the company. 
The phone is expected to compete directly with the recent release from its rival, Nexon Technologies, which unveiled the 
Nexon Ultra series earlier this year.
"""

Entities have attribute type as one of "organization", "person", or "product".

• when type is organization, there is no other attributes
• when type is person, the entity has an optional Role attribute
• when type is product, the entity has an optional Specifications attribute

entities = [
            {'entity_id': '0', 'start': 28, 'end': 40, 'attr': {'Type': 'organization'}}, 
            {'entity_id': '1', 'start': 99, 'end': 111, 'attr': {'Type': 'product', 'Specifications': '128GB, 256GB, Octa-core Quantum processor, 6.5-inch AMOLED display'}}, 
            {'entity_id': '2', 'start': 296, 'end': 307, 'attr': {'Type': 'person', 'Role': 'CEO'}}, 
            {'entity_id': '3', 'start': 452, 'end': 470, 'attr': {'Type': 'organization'}}, 
            {'entity_id': '4', 'start': 492, 'end': 510, 'attr': {'Type': 'product'}}
           ]

Relations (optional) are listed as as entity id pairs.

relations = [
             {'entity_1_id': '0', 'entity_2_id': '1'}, 
             {'entity_1_id': '0', 'entity_2_id': '2'}, 
             {'entity_1_id': '3', 'entity_2_id': '4'}
            ]

The serve() function starts a Flask App that visualizes the entities, attributes, and relations. The theme parameter supports light and dark modes. The color_attr_key parameter specifies that entity color is based on entity "Type" attribute.

from ie_viz import serve

serve(text, entities, relations, theme="light", color_attr_key="Type")

A Flask App starts on the localhost port 5000 (default). The full rendered HTML is available TechNews

* Serving Flask app 'ie_viz.utilities'
 * Debug mode: off
WARNING: This is a development server. Do not use it in a production deployment. Use a production WSGI server instead.
 * Running on all addresses (0.0.0.0)
 * Running on http://127.0.0.1:5000
Press CTRL+C to quit
127.0.0.1 - - [02/Oct/2024 11:44:58] "GET / HTTP/1.1" 200 -

Examples

Below are Jupyter notebooks that demo some use cases:

Tech News: Organization, Product, and Person

Clinical Note: Medication, condition, and Adverse Reaction

User guide

entities = [
            {'entity_id': '<entity id>', 'start': <start char>, 'end': <end char>}, 
            {'entity_id': '<entity id>', 'start': <start char>, 'end': <end char>, 'attr': {'<attribute key>': '<attribute value>'}},
            ...
           ]

relations = [
             {'entity_1_id': '<entity id>', 'entity_2_id': '<entity id>'}, 
             {'entity_1_id': '<entity id>', 'entity_2_id': '<entity id>'}, 
             ...
            ]

from ie_viz import serve

serve(text, entities, color_attr_key="<attribute key to assign color>")

def color_map_func(entity) -> str:
    if entity['attr']['<attribute key>'] == "<a certain value>":
        return "gray"
    else:
        return "#03A9F4"

serve(text, entities, color_map_func=color_map_func)