ELMo-NB at SemEval-2020 Task 7: Assessing Sense of Humor in EditedNews Headlines Using ELMo and NB

Proceedings of the 14th International Workshop on Semantic Evaluation, pages 1001–1007 Barcelona, Spain (Online), December 12, 2020. 1001 ELMo-NB at SemEval-2020 Task 7: Assessing Sense of Humor in Edited News Headlines Using ELMo and NB Enas Khwaileh Dept. of Computer Science Jordan University of Science and Technology-Irbid, Jordan ekhwaileh18@cit.just.edu.jo Muntaha Al-as’ad Dept. of Computer Science Jordan University of Science and Technology-Irbid, Jordan maalasaad17@cit.just.edu.jo Abstract In this paper, we present our submission for SemEval-2020 competition subtask 1 in Task 7 (Hossain et al., 2020a): Assessing Humor in Edited News Headlines. The task consists of estimating the hilariousness of news headlines that have been modified manually by humans using micro-edit changes to make them funny. Our approach is constructed to improve on a couple of aspects; preprocessing with an emphasis on humor sense detection, using embeddings from state-of-the-art language model (ELMo), and ensembling the results came up with using machine learning model Na¨ıve Bayes (NB) with a deep learning pretrained models. ELMo-NB participation has scored (0.5642) on the competition leader board, where results were measured by Root Mean Squared Error (RMSE). 1 Introduction Checking the degree of sentence sense of Humor through understanding and analyzing humans natural language and by connecting the text to an intelligent system is considered a critical task (Rastogi et al., 2020). Hence Expressing the readers and writers opinions can increase several emotions, we still need to expand the positive texts and establish a way for analyzing it (Salminen et al., 2020). Humor is considered a great way to the reader, it is resembling therapy (Ziabari and Treur, 2020). Producing machines that can determine whether the sentence contains some degree of sense of humor or not is gaining a great attention recently (Abdullah and Shaikh, 2018). Since social media is taking over most of people’s daily life routines , the culture and environment affect the content greatly (Downey et al., 2006)(Zhao et al., 2020). Numerous factors have brought increasing attention to real-life tasks such as text classification (Howard and Ruder, 2018) (Conneau et al., 2016) (Al-Omari et al., 2020) and other text analysis like pun classification (Diao et al., 2020). Moreover, (Miller et al., 2020) came up with an idea to detect the tweets humorousness using Gaussian Process. Several Natural Language Processing (NLP) applications and tools are proliferating recently (Kumar and Garg, 2020), (Hirschberg and Manning, 2015), especially with the rise of Deep Learning (DL) and Machine Learning (ML) enhancements (Duerr and Ramdeen, 2017) (Young et al., 2018) (Gardner et al., 2018). One of NLP state of the art approaches is ELMo language preprocessing model as a pretrained model on general NLP tasks of language modeling (Reimers and Gurevych, 2019). ELMo can be fine- tuned on specific tasks like next word prediction (Siddiqui and Hassan, 2019), translation (Li and Chen, 2019) or question answering (McCann et al., 2018) and semantic text (Al-Asa’d et al., 2019). Another NLP state of the art is BERT (Peters et al., 2018). SemEval-2020 competition in Task 7 1 has 313 as a total …

Content truncated to 500 words