ULD@NUIG at SemEval-2020 Task 9: Generative Morphemes with an Attention Model for Sentiment Analysis in Code-Mixed Text

Proceedings of the 14th International Workshop on Semantic Evaluation, pages 968–974 Barcelona, Spain (Online), December 12, 2020. 968 ULD@NUIG at SemEval-2020 Task 9: Generative Morphemes with an Attention Model for Sentiment Analysis in Code-Mixed Text Koustava Goswami, Priya Rani, Bharathi Raja Chakravarthi, Theodorus Fransen, and John P. McCrae Insight SFI Research Centre for Data Analytics, Data Science Institute, National University of Ireland Galway {koustava.goswami, priya.rani, bharathi.raja, theodorus.fransen, john.mccrae}@insight-centre.org Abstract Code mixing is a common phenomena in multilingual societies where people switch from one language to another for various reasons. Recent advances in public communication over different social media sites have led to an increase in the frequency of code-mixed usage in written language. In this paper, we present the Generative Morphemes with Attention (GenMA) Model sentiment analysis system contributed to SemEval 2020 Task 9 SentiMix. The system aims to predict the sentiments of the given English-Hindi code-mixed tweets without using word-level language tags instead inferring this automatically using a morphological model. The system is based on a novel deep neural network (DNN) architecture, which has outperformed the baseline F1-score on the test data-set as well as the validation data-set. Our results can be found under the user name “koustava” on the “Sentimix Hindi English”1 page. 1 Introduction Sentiment analysis refers to a process of predicting the emotion content from a given text. Sentiment analysis is usually seen as a categorization problem over a variable with three values: positive, negative, neutral (Phani et al., 2016). With the increase in the popularity of social media such as Twitter, a new area of study to the field of natural language processing and thus, sentiment analysis has been explored. Most of the data extracted from social media are code-mixed (Ranjan et al., 2016; Priyadharshini et al., 2020), which have become a common approach in most cases but also pose unique challenges. Analysis of short texts from micro-blogging platforms such as Twitter is in high demand as the analysis of these text distill and evaluate the moods and the sentiment of the users and are very useful for different organisations, be it government or business or NGO. Sentiment analysis for Indian code-mixed languages is relatively new (Jose et al., 2020; Chakravarthi et al., 2020a; Chakravarthi et al., 2020b; Rani et al., 2020). The significant difference in style of language, orthography (Chakravarthi et al., 2019) and grammar used in tweets presents specific challenges for English-Hindi code-mixed data. In this paper we aim to introduce a novel deep neural network system which was submitted for SemEval 2020 Task 9, Sub Task A for English-Hindi data (Patwa et al., 2020). We will also compare the system with other state-of-the-art systems and describe how the system has outperformed others. The systems were trained using only the Twitter data provided by the organisers excluding the word-level language tags provided in the data. 2 Related Work Although the field of sentiment analysis is growing and several systems have advanced the state-of-the-art, the overall performance of systems to predict sentiment in code-mixed …

Content truncated to 500 words