Adaptive and multi-scale feature fusion for Chinese news headline classification

Yumin Yan · 2026

Read Adaptive and multi-scale feature fusion for Chinese news headline classification by Yumin Yan (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/plos-pone-0345779.txt.

Description

Yumin Yan built two deep learning models for sorting Chinese news headlines into categories, a harder job than it sounds because headlines are short, sparse, and ambiguous. One model targets big datasets and the other small ones. Both start from a Chinese language model, then fuse features from several of its layers so word meaning and local phrasing reinforce each other.

On two standard headline collections, the large-data model edged past a leading lightweight rival by about a point of accuracy. The small-data model, hardened with multi-scale attention and adversarial training, gained roughly three points when trained on only a tenth of the data. The author concludes that matching the architecture to the data volume, especially adaptive feature fusion, pays off in real news applications.

Catalog record

CREATORYumin Yan
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE87914
LICENSECC_BY
RIGHTSCC BY
SOURCEPLOS

Source: https://doi.org/10.1371/journal.pone.0345779

Browse the full library · About the Tower of Babel