CENTEMENT at SemEval-2018 Task 1: Classification of Tweets using Multiple Thresholds with Self-correction and Weighted Conditional Probabilities

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 200–204 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics CENTEMENT at SemEval-2018 Task 1: Classification of Tweets using Multiple Thresholds with Self-correction and Weighted Conditional Probabilities Tariq Ahmad School of Computer Science University of Manchester Oxford Road, Manchester M13 9PL, U.K. tariq.ahmad@postgrad .manchester.ac.uk Allan Ramsay School of Computer Science University of Manchester Oxford Road, Manchester M13 9PL, U.K. allan.ramsay@ manchester.ac.uk Hanady Ahmed CAS, Arabic Department Qatar University 2713, Al Hala St Doha, Qatar hanadyma@ qu.edu.qa Abstract In this paper we present our contribution to SemEval-2018, a classifier for classifying multi-label emotions of Arabic and English tweets. We attempted “Affect in Tweets”, specifically Task E-c: Detecting Emotions (multi-label classification). Our method is based on preprocessing the tweets and creat- ing word vectors combined with a self cor- rection step to remove noise. We also make use of emotion specific thresholds. The final submission was selected upon the best perfor- mance achieved, selected when using a range of thresholds. Our system was evaluated on the Arabic and English datasets provided for the task by the competition organisers, where it ranked 2nd for the Arabic dataset (out of 14 entries) and 12th for the English dataset (out of 35 entries). 1 Introduction Social network platforms such as Facebook, LinkedIn and Twitter are now at the hub of ev- erything we do. Twitter is one of the most popu- lar social network platforms; as recently as 2013 an incredible 21% of the global internet popu- lation used Twitter actively on a monthly basis (globalwebindex, accessed 05/2016). Twitter is used by celebrities, movie stars, politicians, sports stars and everyday people. Every day, millions of users share their opinions about themselves, news, sports, movies and many many other topics. This makes platforms like Twitter rich sources of data for public opinion mining and sentiment analy- sis (Pak and Paroubek, 2010). However, although these corpora are rich, they are somewhat noisy because tweets can be informal, misspelt and con- tain slang, emoticons (Albogamy and Ramsay, 2015) and made-up words. Furthermore, Arabic tweets have the added complication of dialects in which the same words or expressions can have dif- ferent connotations. Multi-label classification of tweets is a clas- sification problem where tweets are assigned to two or more classes. It is considered more com- plex than traditional classification tasks because the classifier has to predict several classes. There has been much work in the areas of sen- timent detection (Rosenthal et al., 2017), emotion intensity (Mohammad and Bravo-Marquez, 2017) and emotion categorisation (Hasan et al., 2014). Sentiment analysis aims to classify tweets into positive, negative, and neutral categories, emotion intensity is determining the intensity or degree of an emotion felt by the speaker and emotion cate- gorisation is the classification of tweets based on their emotions. The most commonly used clas- sification techniques are Naive Bayes and Sup- port Vector Machines (SVM). Some researchers report that SVMs (Barbosa and Feng, 2010) per- form better …

Content truncated to 500 words