Insights into knowledge evolution based on semantic representation and dynamic visual analytics

Public summary

* Maps 40k books in semantic space, revealing hotspot migration.

* Interactive KnowFlowViz displays ideas traveling across disciplines.

* Semantic embeddings expose hidden links beyond citation networks.

* Knowledge transfer flow charts rising and fading research themes.

Abstract

In the field of knowledge science, understanding the structure and dynamic evolution of knowledge is essential for advancing disciplinary development and anticipating research trends. However, current methodologies lack a unified semantic framework for the structured representation of knowledge, which impedes the quantitative analysis of its evolution and limits the ability to uncover complex relationships among knowledge entities. To bridge these gaps, we propose a structured knowledge representation method based on semantic embedding, enabling a deeper and more consistent understanding of semantic relationships within knowledge units. Building on this foundation, we introduce the concept of knowledge transfer flow to quantitatively analyze and visualize the dynamic evolution of knowledge hotspots over time, revealing the underlying mechanisms that drive knowledge transformation. Furthermore, we develop the KnowFlowViz system, which leverages interactive visual analytics to uncover intricate structural patterns and evolutionary dynamics within knowledge systems, thereby supporting decision-making and guiding future research directions. Our study reveals that established knowledge domains (such as long-standing disciplines) tend to maintain their dominant positions, while newly emerging knowledge entities often preferentially connect with these domains to form interdisciplinary linkages. This phenomenon of advantage accumulation and preferential attachment accelerates the growth and recognition of newcomers. The findings underscore the importance of fostering a more equitable and inclusive knowledge network, and they support the development of policies that nurture emerging disciplines and sustain a diverse, vibrant knowledge ecosystem.

Introduction

In the era of information explosion, the volume of knowledge generated and accumulated within academic disciplines has reached unprecedented levels. As the boundaries between fields become increasingly blurred and interdisciplinary research flourishes, the challenge of organizing, navigating, and comprehending this vast knowledge landscape poses a significant obstacle for scholars and practitioners. Visualization of structured knowledge, positioned at the intersection of information science, data mining, and human-computer interaction, has emerged as an effective means to represent and interpret complex information in an intuitive manner. By translating intricate knowledge systems into accessible visual forms, researchers can identify patterns, trace formation processes, and uncover migration trajectories of ideas.

Traditional approaches to knowledge representation, typically based on citation networks and bibliometric methods, offer valuable snapshots of knowledge structures but are limited in capturing the dynamic processes of knowledge creation and transformation. Static representations illustrate relationships and hierarchies at a single point in time, yet they fail to convey the temporal dimension of intellectual progress. Understanding the dynamic evolution of knowledge is essential, as it reveals how concepts emerge, diffuse, and interact with existing paradigms. However, citations themselves are not always reliable indicators of intellectual influence, since they may be shaped by author preferences, journal policies, or disciplinary norms. Furthermore, due to the overwhelming volume of publications, no author can cite every relevant work, which often leaves gaps in citation networks and obscures significant intellectual connections, as illustrated in Figure 1A.