Multi-Lingual Semantic-Based Sentiment Analysis Using Generative AI
Praloy Biswas, Daniel Arockiam, Subhrendu Guha Neogi, Somenath Sengupta, Sanjeev Saraswat, Rishit Chanda · 2024
Generative AI models have extensive applications in Natural Language Processing (NLP) and large language models (LLMs) across various domains. They demonstrate high performance in numerous NLP tasks such as summarization, language understanding, reasoning, language generation, question-answering, sentiment categorization, and translation. Newer LLMs are projected to analyse and generate text in multiple languages because to their training on multilingual datasets, like ChatGPT, BLOOMZ, and other similar versions. Considering how frequently LLMs are used, it is critical to assess their effectiveness in multilingual environments. A major issue to be noted is that in a zero-shot context, the present generative models are not very good at producing text in Indian languages. Several scholarly articles studied generative LLMs in the English language. In contrast, LLM for Indic languages is not meant to be used in a zero-shot manner in downstream applications due to poor generating performance. Generative LLMs evaluate models on standard NLP benchmarks covering NLP datasets in diverse languages. We compared the performance of generative LLMs with non-autoregressive models on these tasks to evaluate the generative AI model's performance compared to the previous generation of LLMs. A framework for evaluating generative LLMs in the multilingual setting has been evaluated to provide future directions.