Leveraging Taxonomy and LLMs for Improved Multimodal Hierarchical Classification

Shijing Chen 1 , Mohamed Reda Bouadjenek 2 , Shoaib Jameel 3 , Usman Naseem , Basem Suleiman 1 , Flora D. Salim 1 , Hakim Hacid 5 , Imran Razzak 6,1 1 University of New South Wales, Australia 1 Deakin University, Australia 3 University of Southampton, United Kingdom 4 Macquarie University, Sydney, A · 2025

Read Leveraging Taxonomy and LLMs for Improved Multimodal Hierarchical Classification by Shijing Chen 1 , Mohamed Reda Bouadjenek 2 , Shoaib Jameel 3 , Usman Naseem , Basem Suleiman 1 , Flora D. Salim 1 , Hakim Hacid 5 , Imran Razzak 6,1 1 University of New South Wales, Australia 1 Deakin University, Australia 3 University of Southampton, United Kingdom 4 Macquarie University, Sydney, A (2025) 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/arxiv-2501-06827.txt.

Description

Shijing Chen and co-authors teach classifiers to respect the family tree of their classes, in a 2025 paper. Sorting goods into layered classes, like department, then aisle, then shelf, breaks down when each layer votes alone. A model may pick a shelf that does not sit under its picked aisle. The team weaves the class tree into a large language model so each call must fit the levels above it.

Their frame works with goods described by both text and images, and it does not depend on one brand of language model. Tests used a large online-shop dataset with several layers of classes. Scores rose sharply over both classic layered sorters and plain language-model setups. The authors say baking the tree into the model is the key step, and that the trick should carry to other layered sorting tasks, from medical codes to law.

Catalog record

CREATORShijing Chen 1 , Mohamed Reda Bouadjenek 2 , Shoaib Jameel 3 , Usman Naseem , Basem Suleiman 1 , Flora D. Salim 1 , Hakim Hacid 5 , Imran Razzak 6,1 1 University of New South Wales, Australia 1 Deakin University, Australia 3 University of Southampton, United Kingdom 4 Macquarie University, Sydney, A
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE36 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-06827

Browse the full library · About the Tower of Babel