Automated-Headline-and-Sentiment-Generator

Overview

Digital content is expanding at a very rapid pace. Many activities that experts undertake today involve the ability to process digital content and synthesize them to make decisions.

We were provided with a dataset consisting of news articles and tweets which belong to either mobile technology. The problem statement consisted of 3 parts :

1. Develop an intelligent system that could first identify the theme of tweets and articles.
2. If the theme is mobile technology then it should identify the sentiments against a brand (at a tweet/paragraph level).
3. Finally a one-sentence headline of max of 20 words for articles that follow the mobile technology theme was needed to be generated. A headline for tweets is not required.

We approached each of these sub problems separately, using three different models for each task. We used state of the art Transformers for the purpose of classification of tweets/articles and generation of headlines. We used VADER for Sentiment Extraction which is a sentiment analysis model. The specific approaches are discussed in more detail in later sections.

Models Used

• [:heavy_check_mark:] Text Classification : XLM-RoBERTa
• [:heavy_check_mark:] Sentiment Extraction : VADER
• [:heavy_check_mark:] Headline Generation : MT5

Features provided

• Implementation of script and complete procedure for Text Classification with instructions.
• Implementation of scripts for headline generation and sentiment extraction.
• Implementation of complete single pipeline performing all the three steps in order.

Setup

Run the following commands to set-up environment:

git clone https://github.com/AniketRajpoot/Automated-Headline-and-Sentiment-Generator.git  
cd Automated-Headline-and-Sentiment-Generator  
pip install -r requirements.txt  

Pre-trained checkpoints:

XLM-RoBERTa

Run the following command:

gdown --id 1mBhGHYOTnikOJD3KOBK1s_FuCjaiUR1a

Alternatively, the link to the same is given below:

https://drive.google.com/file/d/1mBhGHYOTnikOJD3KOBK1s_FuCjaiUR1a/view?usp=sharing

MT5

Run the following command:

!gdown --id 1ncA3AMBPEFvfv8xMdLrl6vCSzNUu3sw9

Alternatively, the link to the same is given below:

https://drive.google.com/file/d/1ncA3AMBPEFvfv8xMdLrl6vCSzNUu3sw9/view?usp=sharing

Scripts

Text Classification

python -u scripts/predict_class.py --file <FILEPATH> <OR>  
python -u scripts/predict_class.py --sen <SENTENCE>  

Sample run:

python -u scripts/predict_class.py --file 'sample_article.txt' 

Headline Generation

python -u scripts/predict_headline.py --file <FILEPATH> <OR>  
python -u scripts/predict_headline.py --sen <SENTENCE>  --num_sentences <NO OF HEADLINES>  

Sample run:

python -u scripts/predict_headline.py --file 'sample_article_2.txt' --num_sentences 5 

Support

There are many ways to support a project - starring⭐️ the GitHub repo is just one.