AI Extractive Text Summarizer Project

Artificial Intelligence project 03

Last updated: 4 September 2026

Build a complete extractive text summarizer workflow with reproducible Python code, transparent inputs, reviewable output, and responsible-use limits.

Explore AI training in Vizag View all project ideas

AI objective

Select the most informative source sentences without generating facts that are absent from the input.

Data or knowledge source

User-provided plain text split into sentences; the demo article discusses AI data, evaluation, monitoring, and review.

Requirements

Python 3.10 or later
A terminal or command prompt
python -m pip install numpy scikit-learn
About 45-60 minutes to build and review

How the system works

Split sentences, calculate TF-IDF features, normalise sentence scores for length, select the highest scores, and restore original order.

Validation checklist

The requested sentence count is respected
Every summary sentence is copied from the source
Short documents are returned without failure
No external model or network request is required

Complete Python code

Save the program as ai_extractive_text_summarizer.py. The code runs locally and requires no paid API key or model download.

"""Create an extractive summary by ranking source sentences with TF-IDF."""

from __future__ import annotations

import re
import numpy as np
from sklearn.feature_extraction.text import TfidfVectorizer


def split_sentences(text: str) -> list[str]:
    return [part.strip() for part in re.split(r"(?<=[.!?])\s+", text.strip()) if part.strip()]


def summarize(text: str, sentences: int = 3) -> str:
    source = split_sentences(text)
    if not source or sentences < 1:
        return ""
    if len(source) <= sentences:
        return " ".join(source)
    matrix = TfidfVectorizer(stop_words="english").fit_transform(source)
    scores = np.asarray(matrix.sum(axis=1)).ravel() / np.sqrt(np.maximum(1, (matrix != 0).sum(axis=1).A1))
    selected = sorted(scores.argsort()[-sentences:].tolist())
    return " ".join(source[index] for index in selected)


def main() -> None:
    article = (
        "Artificial intelligence systems learn patterns from data. Good datasets need clear ownership and documentation. "
        "Evaluation should reflect the real task and user population. A high average score can hide weak performance for important groups. "
        "Teams should monitor deployed systems because data and behaviour change. Human review is essential for high-impact decisions."
    )
    print(summarize(article, sentences=3))
    print("This method selects source sentences; it does not generate new facts.")


if __name__ == "__main__":
    main()

Run the project

Create and activate a virtual environment.
Install dependencies with python -m pip install numpy scikit-learn when packages are required.
Run python ai_extractive_text_summarizer.py.
Review confidence, fallbacks, sources, or error metrics rather than accepting output automatically.
Test additional normal, edge, unsupported, and adversarial inputs.

Expected output

A concise extractive summary containing only original sentences.

Accuracy, privacy, and responsible-use limits

TF-IDF importance does not guarantee complete or balanced coverage. Extractive summaries can omit qualifications and should not replace review of legal, medical, financial, or safety-critical text.

Ways to extend the project

Add section-aware scoring, redundancy removal, query-focused summaries, evaluation against human references, and citation offsets.

Continue learning Artificial Intelligence

Try the next project, return to the Softenant project library, or explore the AI training in Vizag for guided NLP, retrieval, evaluation, automation, and responsible AI practice.