Papers by Wilfried Logerais
A Neural Few-Shot Text Classification Reality Check (2021.eacl-main)
Copied to clipboard
| Challenge: | Modern few-shot text classification models struggle when the amount of annotated data is scarce. |
| Approach: | They compare neural few-shot classification models with NLP and computer vision models with transformers to test their performance. |
| Outcome: | The proposed models perform almost equally on ARSC dataset, but not on the intent detection task. |
PROTAUGMENT: Unsupervised diverse short-texts paraphrasing for intent detection meta-learning (2021.acl-long)
Copied to clipboard
| Challenge: | Recent research considers few-shot intent detection as a meta-learning problem because of labeled data scarcity and the number of classes involved. |
| Approach: | They propose a meta-learning algorithm for short texts classification that limits overfitting on the bias introduced by the few-shots classification objective at each episode. |
| Outcome: | The proposed algorithm limits overfitting on the bias introduced by the few-shots classification objective at each episode. |
Few-shot Pseudo-Labeling for Intent Detection (2020.coling-main)
Copied to clipboard
| Challenge: | Existing methods for labeling intents are expensive and time-consuming. |
| Approach: | They propose a folding/unfolding hierarchical clustering algorithm which assigns weighted pseudo-labels to unlabeled user utterances. |
| Outcome: | The proposed method performs better on multiple intent detection datasets and is stronger than existing methods. |