Sentiment Analysis in Code-Mixed Telugu-English Text with Unsupervised Data Normalization

Proceedings of Recent Advances in Natural Language Processing, pages 753–760 Sep 1–3, 2021. https://doi.org/10.26615/978-954-452-072-4_086 753 Sentiment Analysis in Code-Mixed Telugu-English Text with Unsupervised Data Normalization Kusampudi Siva Subrahamanyam Varma, Preetham Sathineni, Radhika Mamidi Language Technologies Research Centre, IIIT Hyderabad, Telangana, India {siva.subrahamanyam, preethamreddy.s}@research.iiit.ac.in radhika.mamidi@iiit.ac.in Abstract In a multilingual society, people communi- cate in more than one language, leading to Code-Mixed data. Sentimental anal- ysis on Code-Mixed Telugu-English Text (CMTET) poses unique challenges. The unstructured nature of the Code-Mixed Data is due to the informal language, in- formal transliterations, and spelling errors. In this paper, we introduce an annotated dataset for Sentiment Analysis in CMTET. Also, we report an accuracy of 80.22% on this dataset using novel unsupervised data normalization with a Multilayer Percep- tron (MLP) model. This proposed data normalization technique can be extended to any NLP task involving CMTET. Fur- ther, we report an increase of 2.53% ac- curacy due to this data normalization ap- proach in our best model. 1 Introduction In recent times, huge volumes of data are being generated on social media and online blogging. The usage of multiple languages in day-to-day conversations and the minimal linguistic re- strictions on the online content lead to the gen- eration of Code-Mixed data. Code-Mixing or Code-Switching is the usage of two or more languages in a single sentence or a conversa- tion. The generated code-mixed data can be used to extract potential knowledge like emo- tion, news or information. Sentimental analysis on Code-Mix social me- dia data helps us to understand the underlying sentiment of the phrase or sentence, which can have many practical use cases in the real world. For example, it can be used to understand the sentiment of restaurant or movie reviews. Code-Mixed Data can be easily extracted from various online platforms like social media and blogs using APIs and various web-scraping tools. In spite of having huge amounts of data, performing sentiment analysis on it is still chal- lenging because of its unstructured and noisy nature (Arora and Kansal (2019), Gautam and Yadav (2014), Barik et al. (2019)). Hence, this requires robust preprocessing techniques to normalize this unstructured data. In this paper, we have performed sentimen- tal analysis on Code-Mixed Telugu-English Text (CMTET) with an unsupervised data normalization. We analyse and classify each sentence into three sentiments, positive, nega- tive and neutral. To the best of our knowledge this is the first work to propose a framework to perform sentiment analysis on CMTET. The reason for choosing English as our secondary language for the Code-Mixed data is because we observed that most of the Telugu data avail- able in social media is very often Code-Mixed with English. The rest of the paper is organized as follows. In section 2, we discuss the existing Sentiment Analysis approaches in other Code-Mixed lan- guages. Section 3 introduces the dataset and describes the methodology involved in prepro- cessing and annotating this dataset. In section 4, we discuss in detail various challenges faced in performing sentiment analysis in CMTET. …

Content truncated to 500 words