# Incremental Few-shot Text Classification with Multi-round New Classes: Formulation, Dataset and System

Source: [Incremental Few-shot Text Classification with Multi-round New Classes: Formulation, Dataset and System](<https://www.sota2.com/research/paper/68fa69d81e209131cea24910>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-25T23:10:38.141Z

Authors: Congying Xia, Wenpeng Yin, Yihao Feng, Philip Yu

Published: 2021-04-24T04:41:15.000Z

## Abstract

Text classification is usually studied by labeling natural language texts with relevant categories from a predefined set. In the real world, new classes might keep challenging the existing system with limited labeled data. The system should be intelligent enough to recognize upcoming new classes with a few examples. In this work, we define a new task in the NLP domain, incremental few-shot text classification, where the system incrementally handles multiple rounds of new classes. For each round, there is a batch of new classes with a few labeled examples per class. Two major challenges exist in this new task: (i) For the learning process, the system should incrementally learn new classes round by round without re-training on the examples of preceding classes; (ii) For the performance, the system should perform well on new classes without much loss on preceding classes. In addition to formulating the new task, we also release two benchmark datasets in the incremental few-shot setting: intent classification and relation classification. Moreover, we propose two entailment approaches, ENTAILMENT and HYBRID, which show promise for solving this novel problem.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Intent Detection on SNIPS Type 2 (test)](<https://www.sota2.com/research/sota/intent-detection-on-snips-type-2-test.md>): Intent Detection; SNIPS Type 2 (test); Accuracy: 73.6.

- [Intent Detection on SNIPS Type 3 (test)](<https://www.sota2.com/research/sota/intent-detection-on-snips-type-3-test.md>): Intent Detection; SNIPS Type 3 (test); Accuracy: 68.5.

- [Intent Detection on SNIPS Type 1 (test)](<https://www.sota2.com/research/sota/intent-detection-on-snips-type-1-test.md>): Intent Detection; SNIPS Type 1 (test); Accuracy: 69.1.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2104.11882>)

[PDF](<https://arxiv.org/pdf/2104.11882.pdf>)
