SSN_NLP at SemEval-2020 Task 7: Detecting Funniness Level Using Traditional Learning with Sentence Embeddings

Proceedings of the 14th International Workshop on Semantic Evaluation, pages 865–870 Barcelona, Spain (Online), December 12, 2020. 865 SSN NLP at SemEval-2020 Task 7: Detecting funniness level using traditional learning with sentence embeddings Kayalvizhi S SSN College of Engineering kayalvizhis@ssn.edu.in Thenmozhi D SSN College of Engineering theni d@ssn.edu.in Aravindan Chandrabose SSN College of Engineering aravindanc@ssn.edu.in Abstract Assessing the funniness of edited news headlines task deals with estimating the humorness in the headlines edited with micro-edits. This task has two sub-tasks in which one has to calculate the mean predicted score of humor level and other deals with predicting the best funnier sentence among given two sentences. We have calculated the humorness level using microtc and predicted the funnier sentence using microtc, universal sentence encoder classifier, many other traditional classifiers that use the vectors formed with universal sentence encoder embeddings, sentence embeddings and majority algorithm within these approaches. Among these approaches, microtc with 6 folds, 24 processes and 3 folds, 36 processes achieve the least Root Mean Square Error for development and test set respectively for subtask 1. For subtask 2, Universal sentence encoder classifier achieves the highest accuracy for development set and Multi-Layer Perceptron applied on vectors vectorized using universal sentence encoder embeddings for the test set. 1 Introduction Assessing the Funniness of Edited News Headlines (Hossain et al., 2020) task of Semeval-2020 deals with estimating the funniness of news headlines that have been modified by humans using a micro-edit to make them funny. The goal of the task is to make the machine predict whether the short edits on the given text makes them funny or not. Thus, the system has to assess the intensity of humorness in the edited sentence. This system has various applications including humor generation where such a system can be used in a generate-and-test scheme to generate many potentially humorous texts and rank them in terms of funniness. This task has two sub tasks that include a regression task, in which the mean funniness of the edited headline has to be predicted and a classification task, in which the funnier of the two edited headlines has to be predicted. 2 Related Work Similar task called Humor Analysis based on Human Annotation (HAHA) has been part of IberLEF 2018 (Castro et al., 2018) and IberLEF 2019 (Chiruzzo et al., 2019) which deals with automatic detection and automatic rating of humor in Spanish tweets. Different methodologies have been reported in these tasks. Convolutional RNN (Giudice, 2019) have been used, in which input tweet is represented as a list with fixed length of sparse vectors and each vector of the list represents an individual character of the tweet and contains flags whose values are either 0 or 1 is fed to the neural network whose output was rounded to 0 or 1 to get a binary value and for the funniness score prediction subtask it was multiplied by five to get a value between 0 and 5. In a BERT based approach (Mao and Liu, 2019; Ismailov, …

Content truncated to 500 words