Soft Contextualized Encoder For User Defined Text Classification

Charu Maheshwari, Vyas Raina · 2026

Read Soft Contextualized Encoder For User Defined Text Classification by Charu Maheshwari, Vyas Raina (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/arxiv-2601-03450.txt.

Description

Charu Maheshwari and Vyas Raina tackle classification tasks where the user defines the labels on the spot, including labels the system has never seen. Such tasks appear in business analytics, content moderation, and search. Their model, a soft contextualized encoder, reads each candidate label in light of the full label set and a soft prompt built from the input text.

Training used varied data from many sources, which helped the model handle new topic sets from any domain. On held-out tests and on unseen benchmarks, it matched or beat prior systems and set new best results. The work matters because real label sets change often. A classifier that needs no retraining for new labels saves time and labeling cost.

Catalog record

CREATORCharu Maheshwari, Vyas Raina
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE48 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-03450

Browse the full library · About the Tower of Babel