UTILIZING NATURAL LANGUAGE PROCESSING, OBSERVATIONAL METRICS FOR PREDICTIVE ANALYSIS OF GENERATIVE ARTIFICIAL INTELLIGENCE MODELS

Abstract

In the era of massive language models, there is a growing need to determine the future of the enduring desire to use ChatGPT, an example of a generative artificial intelligence (GenAI) model. It's doubtful that consumers' present feelings and degree of interest will last over time. This work investigated the predictive analytics of observational metrics on GenAI models using online data and natural language processing techniques to forecast future sentiments and search interest. Time-bound web analytics data and Twitter metrics related to GenAI were collected using Google Trend and the Twitter API on Orange Data Mining Toolkit. Google trend data was forecasted using Autoregressive Integrated Moving Average (ARIMA), whereas sentiment polarities and search interest time series were predicted using Naive Bayes. The experiment's results indicated a limited correlation between tweet sentiment polarity scores and engagement metrics. Five subjects in all were returned by the topic modeling: doubts or skepticism about OpenAI and Microsoft, Microsoft and AI Use, French discussions on ChatGPT, ChatGPT arguments and usage, and making something funny in relation to intelligence and analysis. The findings revealed a predominantly positive sentiment tendency among the 50 anticipated sentiment instances, with 41 (82%), 4 (8%) and 5 (10%) denoting good, neutral, and negative sentiments, respectively. This implies a generally positive outlook. These findings showed how important it is to look at sentiment and interest trends to fully understand the evolution of GenAI models.