This repository contains the code and data for the research paper "Improving News Headline Text Generation Quality Through Frequent POS-Tag Patterns Analysis". The paper presents a method for enhancing the quality of generated news headlines by analyzing the frequent POS-tag patterns in the training data and integrating them into the text generation model.
The repository includes the following components:
data: Preprocessed training data used in the papermodels: Source code for the text generation model, implemented in Python with PyTorchHPPE: Headline post-processing script for selecting generated headlines based on POS Tags matching.examples: Generated headlines and their corresponding POS-tag patterns
- Clone this repository using following command:
git clone https://github.com/saifhassan/Headline-Generationg-using-POS.git
The dataset which is used for Urdu News Headlines is taken from here. There are two ways to produce same results as mentioned in paper.
- First, Download original dataset and process through Notebook.
OR
-
Another way is, we have already pre-processed dataset. You can download pre-processed dataset with POS Tags from here
-
Put downloaded dataset in data folder.
- Go to HPPE directory within main directory and open notebook using terminal and execute code.
The repository is open-source and welcomes contributions from the community.
Note: This repository is provided for research purposes only and is not intended for commercial use.