Author Name: Grace Celestina K Date: 27-03-2026
Sentiment analysis has emerged as one of the most widely deployed computational tools for inferring psychological states from digital media behavior, enabling real-time measurement of audience emotional responses at scales impossible through traditional self-report methods. This paper provides a comprehensive evaluation of sentiment analysis as a psychological assessment tool in digital news and media research, examining its theoretical foundations in affective psychology, technical implementation across lexicon-based and deep learning approaches, validation evidence against ground-truth psychological measures, and methodological limitations that constrain its use in scientific research. The paper reviews the psychometric properties of major sentiment analysis tools including VADER, SentiWordNet, BERT-based fine-tuned models, and large language model zero-shot classification, comparing their performance against expert human raters and validated affective self-report measures across diverse media content types. The distinction between sentiment as a linguistic property of text and affect as a psychological state of the reader is theorized, with implications for interpreting sentiment analysis findings in audience research. The paper advances a multilevel model of media sentiment assessment that integrates text-level sentiment, platform-level behavioral aggregation, and individual-level psychological validation. Applications in real-time audience sentiment monitoring for news organizations, public health communication surveillance, and political communication research are evaluated. Ethical considerations including surveillance risks, manipulation potential, and representativeness biases in sentiment data are systematically addressed. The paper concludes with methodological recommendations and a research agenda for improving the psychometric validity of computational sentiment assessment in media contexts.
Keywords: sentiment analysis; affective computing; audience psychology; BERT; NLP; real-time monitoring; emotional states; media audience measurement.
