From Opinion Mining to Deep Learning: Mapping the Knowledge Landscape of Sentiment Analysis

Abstract

The field of sentiment analysis has evolved dramatically over the past two decades, progressing from lexicon- and rule-based opinion mining toward advanced deep learning and transformer-driven approaches. This study maps the evolution of sentiment analysis research from 2004 to 2024 using a bibliometric analysis of 2,394 publications indexed in Scopus. Using growth indicators, citation metrics, and visualisation tools such as VOSviewer, the research maps publication trends, influential authors, leading countries, and thematic developments. The results reveal steady growth in research output, with significant surges after 2014 coinciding with adopting deep learning and transformer-based models. China (703 papers) and India (660) emerged as the most productive countries, while the United States led in citation impact with an average of 61 citations per paper. Conference proceedings (48.9%) and journal articles (44.6%) dominated as primary publication formats. Influential authors such as Zhang Y. and Bhattacharyya P. shaped the knowledge base, while thematic analysis highlighted “social media,” “deep learning,” and “computational linguistics” as core research clusters. Overall, sentiment analysis has matured into a multidisciplinary field with expanding applications in politics, healthcare, business, and education, though challenges remain in multilingual analysis, bias, and explainability.