Loading...

Colbert at haha 2021: parallel neural networks for rating humor in spanish tweets

Annamoradnejad, I ; Sharif University of Technology | 2021

187 Viewed
  1. Type of Document: Article
  2. Publisher: CEUR-WS , 2021
  3. Abstract:
  4. Previously, we proposed ColBERT, a humor detection model based on the general linguistic structure of humor for formal English texts. ColBERT uses BERT model to produce embeddings for the text sentences, which will be put as inputs into a parallel neural network. In this paper, we utilized the proposed model on informal Spanish texts to detect humor and rate its level. The current task has three differences compared to the original humor detection task on the ColBERT dataset: (1) rating humor is a regression task rather than binary classification, (2) texts are informal, and (3) texts are in a different language. Using our general model and without any knowledge of the Spanish language, we participated in HAHA shared task at IberLEF 2021 Forum and achieved 2nd place for humor rating and 3rd place for binary humor detection. The results confirm robustness of our proposed model. © 2021 CEUR-WS. All rights reserved
  5. Keywords:
  6. Linguistics ; Neural networks ; Current ; Computational humor ; Detection models ; Embeddings ; Humor rating ; Informal text ; Linguistic structure ; Model-based OPC ; Parallel neural networks ; Spanish tweet ; Classification (of information)
  7. Source: 2021 Iberian Languages Evaluation Forum, IberLEF 2021, 21 September 2021 ; Volume 2943 , 2021 , Pages 860-866 ; 16130073 (ISSN)
  8. URL: http://ceur-ws.org/Vol-2943/haha_paper7.pdf