RETUYT-InCo at EmoEvalEs 2021: Multiclass Emotion Classification in Spanish Luis Chiruzzo and Aiala Ros´a Universidad de la Rep´ ublica Montevideo, Uruguay {luischir,aialar}@fing.edu.uy Abstract. This paper presents the results for the team RETUYT-InCo of the participation in the EmoEvalEs 2021 challenge. We trained several systems using classical ML techniques and neural networks, and using a diverse set of features including word embeddings and features from Spanish BERT. Our best system achieved 0.6573 macro weighted average F1 score (position 10 in the ranking) and 0.6781 accuracy (position 9) over the test set. The most difficult classes to classify were surprise, disgust and fear, which are also the classes with fewer examples in the corpus. Keywords: Emotion classification · Spanish · LSTM · BERT · word embeddings. 1 Introduction Within the area of subjectivity analysis in texts, emotion analysis presents greater challenges and has been less studied than the more traditional task of classifying texts according to their polarity. It is necessary to define the set of categories and to have larger datasets than for polarity classification, where the different categories are sufficiently represented. This implies a more complex annotation process due to greater subtlety in choosing the category for each example, making it more difficult to assess inter-annotator agreement. An important antecedent on emotion annotation is the corpus created by [11], used at SemEval-2018 Task 1: Affect in Tweets [10]. In this task, a subtask on emotion classification was proposed for three languages: English, Arabic and Spanish. The corpus was annotated according to a set of eleven cate- gories: anger, anticipation, disgust, fear, joy, love, optimism, pessimism, sad- ness, suprise, trust, and a neutral or no emotion extra class. This year, for the second time in a row, the IberLEF workshop includes a task addressing this problem for Spanish texts. In IberLEF 2020, an emotion IberLEF 2021, September 2021, M´alaga, Spain. Copyright © 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).