TECHSSN at SemEval-2021 Task 7: Humor and Offense detection and classification using ColBERT embeddings

Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 1185–1189 Bangkok, Thailand (online), August 5–6, 2021. ©2021 Association for Computational Linguistics 1185 TECHSSN at SemEval-2021 Task 7: Humor and Offense detection and classification using ColBERT embeddings Rajalakshmi Sivanaiah, Angel Deborah S, S Milton Rajendram, Mirnalinee T T, Abrit Pal Singh, Aviansh Gupta, Ayush Nanda Department of Computer Science and Engineering Sri Sivasubramaniya Nadar College of Engineering Chennai 603 110, Tamil Nadu, India {rajalakshmis, angeldeborahs, miltonrs, mirnalineett}@ssn.edu.in {abritpal18007, aviansh18028, ayush18031}@cse.ssn.edu.in Abstract This paper describes the system used for de- tecting humor in text. The system developed by the team TECHSSN uses binary classifi- cation techniques to classify the text. The data undergoes preprocessing and is given to ColBERT (Contextualized Late Interaction over BERT), a modification of Bidirectional Encoder Representations from Transformers (BERT). The model is re-trained and the weights are learned for the dataset. This sys- tem was developed for the task 7 of the compe- tition, SemEval 2021. 1 Introduction Natural language processing faces the chal- lenges working with humor as it is a highly subjective phenomena and the age, gender and socio-economic status are known to have an impact on the perception of the joke. It usu- ally involves multiple word senses and cul- tural knowledge to appreciate humor to its best. Now a days chatbots and virtual assistants re- quire automated humor detection systems for a better interaction with the user by understand- ing what a human-like approach to humor is. It is crucial to understand the real motive of the user and provide appropriate answer to have a better experience of the user with the virtual assistants (Chen and Soo, 2018). Based on the general linguistic structure of humor, we pro- pose an approach for detecting humor in short texts using ColBERT in this paper. We have developed a system in the name of TechSSN for previous SemEval tasks (Sivana- iah et al., 2020; Logesh et al., 2019) for offen- sive language detection using various machine learning and deep learning networks. In Se- mEval 2021, we participated in subtask-1a and 1c of humor and offensiveness detection in task 7- HaHackathon (Meaney et al., 2021). 2 Related Work Continuous research is going on in this field of humor detection and the systems are getting better every year. De Oliveira and Rodrigo (2015) developed a model for humor detection in Yelp reviews and used convolutional net- works with a maximum of 81.57% accuracy. The system developed by Ortega-Bueno et al. (2018) used UO UPV, a Attention-based Long Short-Term Memory Network. The model consists of a Bidirectional LSTM neural network with an attention mechanism that al- lows to estimate the importance of each word and then, this context vector is used with an- other LSTM model to estimate whether the tweet is humorous or not. The F1 score for this system is approximately 0.78 with accuracy, 0.84. Mao and Liu (2019) developed a system with BERT, a multi-layer bidirectional trans- former encoder which can help to learn deep …

Content truncated to 500 words