This repo contains:
- The official dataset NewsInt for news intent recognition
- The official method DMint for news intent application
of the research paper "Exploring news intent and its application: A theory-driven approach"
๐๐๐ Accepted to Information Processing & Management [J]
Zhengjia Wang, Danding Wang, Qiang Sheng, Juan Cao, Siyuan Ma, Haonan Cheng (2025). Exploring news intent and its application: A theory-driven approach. Information Processing & Management, 62(6), 104229.
- ๐ ๏ธ Project: https://github.com/ICTMCG/NewsInt
- ๐ Paper: https://doi.org/10.1016/j.ipm.2025.104229
- ๐ PDF: https://arxiv.org/pdf/2312.16490
๐ฆ NewsInt Dataset: a fine-grained labeled dataset for news intent recognition.
โ๏ธ DMint Method: a plug-in method for news intent application.
Intent refers to a cognitive state that emerges from rational planning, grounded in the agent's desires and beliefs. News creation intent refers to the purpose or intention behind the creation of a news article.
We present the first Conceptual Deconstruction-based News INTent Understanding framework (NINT), deconstructing news intent from perspectives of rational action and outcomes.
NINT deconstructs the concept of news intent into beliefs, desires, and plans using interdisciplinary theories. It further situates these elements within the specific context of news to investigate the concrete manifestations of news intent.
To the best of our knowledge, this is the first news creation intent dataset through a deconstruction approach.
The overall process of building the NewsInt dataset is shown below:
We collect raw data from 511 news domains, resulting in 12,959 news articles with an average of 5.45 discussion posts from Reddit for each news article. The obtained news articles show diverse distribution on contemporary topics (such as general politics or the US presidential race, the COVID-19 pandemic, womenโs and menโs rights, climate change, vaccines, abortion, gun control, 5G, etc.).
Data distribution in factuality (left) and political bias level (right):
An Instance from the NewsInt dataset:
For more details on data collection, annotation, and dataset analysis, please refer to our original paper "Exploring news intent and its application: A theory-driven approach".
Please fill out this form: Application to Use the Dataset NewsInt to request access. [dataset_readme]
DMint introduces a cognitive-inspired architecture that decomposes news intent through multi-view extractors and dynamic view-gated aggregation, enabling joint modeling of compositional semantics and contextual writing patterns via localโglobal interactions.
Text encoder: Given the news topic and news content, we concatenate them using the [๐ ๐๐] token as input. Specifically, RoBERTa (Liu et al., 2019) is employed as the text encoder.Multi-View Extractor: Three multi-view extractors are developed to explicitly capture different dimensions of news intent.Intent Aggregator: builds an adaptive approach, allowing the proposed DMint to adjust and combine these representations (of elements of news intent) dynamically.
Commands for training and inference have been written in the following bash file. Run by:
bash run.sh
bash run_infer.sh
Results will be automatically saved in ./param_model/
Suggested requirements:
numpy==2.3.5
scikit_learn==1.7.2
torch==2.5.1+cu121
torchvision==0.20.1+cu121
transformers==4.45.2
If you find our paper useful, please cite:
@article{wang2025exploring,
title = {Exploring news intent and its application: A theory-driven approach},
author = {Wang, Zhengjia and Wang, Danding and Sheng, Qiang and Cao, Juan and Ma, Siyuan and Cheng, Haonan},
journal = {Information Processing \& Management},
volume = {62},
number = {6},
pages = {104229},
year = {2025},
publisher = {Elsevier},
issn = {0306-4573},
doi = {https://doi.org/10.1016/j.ipm.2025.104229},
url = {https://www.sciencedirect.com/science/article/pii/S0306457325001700}
}