KISDI Publishes Report on “Analyzing Changes in Generative AI Text Trends by Period Using BERTopic” (detail)

※ URL(Korean): https://www.kisdi.re.kr/bbs/view.do?bbsSn=114776&key=m2101113055776&pageIndex=2&sc=&sw=

KISDI Publishes Report on “Analyzing Changes in Generative AI Text Trends by Period Using BERTopic”

Examines the structure of generative AI discourse across key periods using BERTopic -

The Korea Information Society Development Institute (KISDI, President Sangkyu Rhee) recently published “Analyzing Changes in Generative AI Text Trends by Period Using BERTopic” (KISDI Perspectives 25-10-03).

The report analyzes changes in generative AI text trends across key periods using BERTopic, a topic modeling method based on BERT, a context-aware natural language model developed by Google in 2018. BERTopic analyzes semantic similarity across documents and automatically derives interpretable topics from large-scale text data. Using this method, the report examines changes in generative AI discourse reflected in news data across different periods defined by major technological turning points.

The analysis divides the timeline according to key developments in generative AI, including the releases of GPT-3.5, GPT-4, and Gemini, and compares structural changes in media discourse across periods.

The findings indicate that generative AI discourse has shifted from a focus on “technological shock and new service experiences” (Phase 1: November 2022–February 2023) to “industrialization and institutionalization” (Phase 2: March 2023–November 2023), and then to “efficiency and governance” (Phase 3: December 2023–July 2025).

In Phase 1, triggered by the release of GPT-3.5, public attention increased sharply around technological innovation and service experiences, and discourse exhibited a single-centered structure integrating performance, user experience, and public response. In Phase 2, following the release of GPT-4, adoption expanded across industries, and discourse developed in parallel across technological, industrial, and institutional domains alongside intensified corporate competition in commercialization and infrastructure. In Phase 3, marked by the emergence of innovative models such as DeepSeek, the competitive paradigm shifted from performance toward efficiency and governance. The topic structure became more differentiated, reflecting the structural embedding of generative AI across society and the economy.

Jeong Min JIN (Research Fellow) stated that generative AI discourse has undergone a structural evolution from early technology-centered interest to industrial application and institutional integration, and then to efficient operation and ecosystem coordination. He added that generative AI is increasingly recognized as a structural infrastructure and operating system reshaping society and the economy, with implications extending beyond technological development to broader institutional and governance frameworks.

The report is available for download on the KISDI website (www.kisdi.re.kr)

KISDI News

KISDI Publishes Report on “Analyzing Changes in Generative AI Text Trends by Period Using BERTopic”