Matt Bai at SemEval-2023 Task 5: Clickbait spoiler classification via BERT

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1067–1068 July 13-14, 2023 ©2023 Association for Computational Linguistics Matt Bai at SemEval-2023 Task 5: Clickbait spoiler classification via BERT Nukit Tailor International Institute of Information Technology, Hyderabad nukit.t@research.iiit.ac.in Radhika Mamidi International Institute of Information Technology, Hyderabad radhika.mamidi@iiit.ac.in Abstract The Clickbait Spoiling shared task aims at tack- ling two aspects of spoiling: classifying the type of spoiler based on its length and generat- ing the spoiler. This paper focuses on the task of classifying the spoiler type. A better classifi- cation of the spoiler type would eventually help to generate a better spoiler for the post. We use BERT-base (cased) to classify the clickbait posts. The model achieves a balanced accuracy of 0.63 as we give only the post content as the input to our model instead of the concatenation of the post title and post content to find out the differences that the post title might be bringing in. 1 Introduction Clickbaits are becoming extremely popular nowa- days, and people are falling for them way too easily. Since it plays on people’s innate curiosity and need for rapid satisfaction, clickbait may be an effective tactic for driving traffic to a website or increasing interaction on social media, hence resulting in rev- enue. The general conception about clickbaits is that it is often seen as a bad strategy, since it fre- quently misleads users and gives little value after they click through to the content. This can lead to dissatisfaction, disillusionment, and a lack of faith in the clickbait’s source. However, not all click- bait posts are bad and misleading, some people genuinely use clickbaits to gain more engagement, they also have good content. The Clickbait spoiling shared task (Fröbe et al., 2023a) aims at generating short texts that satisfy the curiosity induced by a clickbait post. Spoilers are needed to briefly let the reader know what to expect when the article is clicked on or whether to click on it or not. The task has two subtasks, one in which we have to classify the clickbait spoiler in 3 classes namely "phrase", "passage", and "multi" which are short phrase spoilers, longer passage spoilers, and multiple non-consecutive pieces of text, respectively. The second task is to generate the spoiler itself, which is not discussed here. The main reference paper (Hagen et al., 2022) concluded that improving the spoiler type classifi- cation would also improve the generation of spoil- ers. The dataset contains 5000 English posts which have a manually extracted spoiler for each. In our approach, we have taken the BERT-base model and given it the content of the clickbait posts as the input as opposed to the the concatenation of the title and content of the post as done in the original approach. We find that removing the post title has a negative overall impact on the scores. 2 Background Even though clickbaits have been around here for a long time now and topics like clickbait identi- …

Content truncated to 500 words