Automatic Bangla Headline Generation Using an Attention-Based Sequence-to-Sequence Model

Abstract

Online Bangla content is growing quickly, yet Bangla remains a low-resource language for natural language processing (NLP). This study builds an automatic Bangla headline generation system that produces a short headline from a longer Bangla text. We compiled a dataset of 1,027 Bangla texts from news articles, blogs and social media, each paired with a human-written headline, and applied cleaning, contraction expansion, tokenization and pre-trained Bengali Word2Vec embeddings. An LSTM/GRU-based encoder-decoder with attention was implemented in TensorFlow and trained for 150 epochs on Google Colab, reaching a training loss of 0.0006. Qualitative examples show headlines that keep the core meaning of the source text. No ROUGE or BLEU evaluation was run, so the results should be read as a feasibility study rather than a benchmark. We discuss the limits of the current system and outline a path toward Transformer based models such as BanglaT5.

BanglaDiaHallu: Hallucinations of large language models and the effect of guideline-grounded retrieval-augmented generation in Bangla diabetes question answering
K. Nahid, M. Anam, M. T. H. Saon +3
medRxiv14 Oct 2026PreprintCC BYFull textEpidemiology & public healthmedRxiv
CSPT-SleepNet: An Automatic Sleep Staging Framework Integrating Center-Driven Sequence Supervision and Multimodal Pyramid Token Encoding
Xiaoqin Lian, Peiyi Wang, Chao Gao +1
Brain sciences1 Oct 2026Peer-reviewedCC BYFull textEurope PMC
An example of an automatic sequence with non-regular abelian complexity
Jean-Paul Allouche, Narad Rampersad, Michel Rigo +2
arXiv2 Oct 2026PreprintCC0Full textCombinatorics & discrete mathematicsarXiv
Automatic music emotion tag generation for short-video platforms: a self-supervised audio-lyric fusion network with valence-arousal regularization
Zhiqi Liu, Huangfeng Hong
Scientific Reports21 Oct 2026Peer-reviewedCC BY-NC-ND 4.0Full textCrossref
Deep co-supervision and attention fusion strategy for automatic COVID-19 lung infection segmentation on CT images
Haigen Hu, Leizhao Shen, Qiu Guan +3
Pattern Recognition5 Oct 2026Peer-reviewedCC BY 4.0Full text32 citedMedical imaging & radiologyHAL

Cite

BibTeX RIS

BibTeX

@misc{mim2026automatic,
  title = {{Automatic Bangla Headline Generation Using an Attention-Based Sequence-to-Sequence Model}},
  author = {Raisa Ahmed Mim},
  howpublished = {Zenodo},
  year = {2026},
  month = oct,
  doi = {10.5281/zenodo.23633117},
  url = {https://zenodo.org/records/23633117},
  note = {CC BY 4.0},
}

RIS

TY  - UNPB
TI  - Automatic Bangla Headline Generation Using an Attention-Based Sequence-to-Sequence Model
AU  - Raisa Ahmed Mim
PY  - 2026
DA  - 2026/10/25
PB  - Zenodo
DO  - 10.5281/zenodo.23633117
UR  - https://zenodo.org/records/23633117
AB  - Online Bangla content is growing quickly, yet Bangla remains a low-resource language for natural language processing (NLP). This study builds an automatic Bangla headline generation system that produces a short headline from a longer Bangla text. We compiled a dataset of 1,027 Bangla texts from news articles, blogs and social media, each paired with a human-written headline, and applied cleaning, contraction expansion, tokenization and pre-trained Bengali Word2Vec embeddings. An LSTM/GRU-based encoder-decoder with attention was implemented in TensorFlow and trained for 150 epochs on Google Colab, reaching a training loss of 0.0006. Qualitative examples show headlines that keep the core meaning of the source text. No ROUGE or BLEU evaluation was run, so the results should be read as a feasibility study rather than a benchmark. We discuss the limits of the current system and outline a path toward Transformer based models such as BanglaT5.
ER  -

Abstract

Related

Cite