Abstract:In this paper, we propose an approach for humor recognition that is usually formalized as text classification or regression tasks. Algorithms are developed to distinguish whether a text sequence is humorous, or how funny a joke is. Our approach is mainly developed on pre-trained BERT (Devlin et al. 2019), fine-tuned on the task specific dataset. We further propose a heuristic data augmentation method by decomposing a long paragraph into a pair of consecutive paragraphs for three purposes. 1) Increasing the size of the training set. 2) Improving the performance of long distance context modelling of BERT. 3) Automatically find the punchline of a joke if exists. Our approach is evaluated on three datasets including CCL2019 Chinese Humor Computation (XIAONIU Cup) dataset in Chinese, FUN (Blinov, Bolotova-Baranova, and Braslavski 2019) dataset in Russian and HAHA (Chiruzzo et al. 2019) dataset in Spanish, which achieves competitive results on all of them. The experimental results demonstrate that our approach could effectively detect jokes and tag the punchline if exists, in different languages.

ECNU at SemEval-2020 Task 7: Assessing Humor in Edited News Headlines Using BiLSTM with Attention.

Ferryman at SemEval-2020 Task 7: Ensemble Model for Assessing Humor in Edited News Headlines

Duluth at SemEval-2020 Task 7: Using Surprise as a Key to Unlock Humorous Headlines

ECNU at SemEval-2018 Task 1: Emotion Intensity Prediction Using Effective Features and Machine Learning Models

ECNU: Multi-level Sentiment Analysis on Twitter Using Traditional Linguistic Features and Word Embedding Features

ECNUCS: A Surface Information Based System Description of Sentiment Analysis in Twitter in the SemEval-2013 (task 2)

ECNU at SemEval-2018 Task 3: Exploration on Irony Detection from Tweets Via Machine Learning and Deep Learning Methods

ECNU at SemEval-2018 Task 12: an End-to-End Attention-based Neural Network for the Argument Reasoning Comprehension Task

ECNU at SemEval-2016 Task 4: an Empirical Investigation of Traditional NLP Features and Word Embedding Features for Sentence-level and Topic-level Sentiment Analysis in Twitter

ECNU at SemEval-2018 Task 2: Leverage Traditional NLP Features and Neural Networks Methods to Address Twitter Emoji Prediction Task.

ECNU at SemEval-2017 Task 4: Evaluating Effective Features on Machine Learning Methods for Twitter Message Polarity Classification.

ECNU at SemEval 2016 Task 6: Relevant or Not? Supportive or Not? A Two-step Learning System for Automatic Detecting Stance in Tweets.

Classification and Regression Combined Model on Accessing Humor Score with Explanatory Feature

CN-HIT-MI.T at SemEval-2020 Task 8: Memotion Analysis Based on BERT

ECNU at SemEval-2017 Task 7: Using Supervised and Unsupervised Methods to Detect and Locate English Puns

UPB at SemEval-2021 Task 7: Adversarial Multi-Task Learning for Detecting and Rating Humor and Offense

CN-HIT-MI.T at SemEval-2019 Task 6: Offensive Language Identification Based on BiLSTM with Double Attention.

THU_NGN at SemEval-2018 Task 1: Fine-grained Tweet Sentiment Intensity Analysis with Attention CNN-LSTM.

ECNU: Expression- and Message-level Sentiment Orientation Classification in Twitter Using Multiple Effective Features

THU_NGN at SemEval-2018 Task 2: Residual CNN-LSTM Network with Attention for English Emoji Prediction.

Humor Detection based on Paragraph Decomposition and BERT Fine-Tuning