Manchester Metropolitan at SemEval-2018 Task 2: Random Forest with an Ensemble of Features for Predicting Emoji in Tweets

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 491–496 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics Manchester Metropolitan at SemEval-2018 Task 2: Random Forest with an Ensemble of Features for Predicting Emoji in Tweets Luciano Gerber School of Computing, Mathematics and Digital Technology Manchester Metropolitan University l.gerber@mmu.ac.uk Matthew Shardlow School of Computing, Mathematics and Digital Technology Manchester Metropolitan University m.shardlow@mmu.ac.uk Abstract We present our submission to the Semeval 2018 task on emoji prediction. We used a ran- dom forest, with an ensemble of bag-of-words, sentiment and psycholinguistic features. Al- though we performed well on the trial dataset (attaining a macro f-score of 63.185 for En- glish and 81.381 for Spanish), our approach did not perform as well on the test data. We describe our features and classification proto- col, as well as initial experiments, concluding with a discussion of the discrepancy between our trial and test results. 1 Introduction Written digital communication is increasingly per- vaded by the use of emoji. Classic NLP sys- tems are not well geared to handle them. Lin- guists are still working out how to treat them (Stark and Crawford, 2015; Danesi, 2016). Even their users may disagree on meaning (Tigwell and Flatla, 2016; Miller et al., 2016). A simple ap- proach could be to ignore all emoji and concen- trate on the words of a text, however this approach may miss valuable meaning that can be obtained by treating the emoji as semantic units. The emoji prediction task (Barbieri et al., 2018, 2017), encourages research into the creation of text classification systems which can identify which emoji was present in a tweet. This could lead to automated suggestion systems for emoji, as well as improving the NLP communities under- standing of how to deal with emoji computation- ally. 2 Data Acquisition + Preprocessing The dataset was compiled between October 2015 and May 2016 (Barbieri et al., 2018). Training, trial, and test data emerge from a 80:10:10 split based on chronological order. We followed the Emoji top 5 words love heart my family ve obsessed wcw heaven foodporn view lmao funny lmfao lol hilarious pink breast sanfranciscoengagement loveal- waysyje strides lit fire mixtape heat flames 802-3037 dickensfranklin dickensofachrist- mas bagsbycab 7171 sunglasses shades cool risky coolin sparkle magical pixie magic getonshimmur royals autism bbn autismspeaks forever- royal kisses kiss princessmailyana smooches smooch : :@ bvillain shredforaliving gdlfashion merica usa ivoted imwithher election2016 sunshine sun sunny soakin beachin purple endalz purplerain alzheimer’s relay mividaesunatombola multi-level silver- criketgentlemensclub azek wink facts rns realtalk salute t3t djsty cheesin braces strasberg fcpx christmas merry christmastree tree tis opus : :@ grigsby cred martian neh silly cray jewelrydesigner Table 1: The top 5 words according to our class occur- rence features for each emoji. organisers instructions to obtain the training data, however we were only able to extract 491,486 tweets as some had been removed by their authors. We tokenised the tweets using the NLTK tweet to- keniser (Bird et al., 2009), but …

Content truncated to 500 words