Can Scholarly Literature and Patents be Represented in a Hierarchy of Topics Structured to Contain 20 Topics per Level? Balancing Technical Feasibility with Human Usability.

Introduction The Intellectual Property & Science division of Thomson Reuters curates millions of records a year covering scholarly literature (Web of Science®), patents and intellectual property (Derwent World Patent Index®) and life sciences discovery (Cortellis®). These millions of records could be connected through billions of potential relationships, such as that represented by a citing relationship between literature and patents, or by different documents that pertain to similar topics. By building these relationships using machine learning techniques we hope to unite information from different data sources to enable extraction of knowledge such that the whole is greater than the sum of the parts, with minimal human effort required. However, connecting these documents in a meaningful way is challenging from both a technological perspective as well as a usability perspective. As shown in Figure 1, studying citation patterns among approximately 250,000 articles from the Web of Science, or 1/200 of the full data set, generates a citation graphthat, while rich with information, is extremely difficult to use to understand knowledge flows. This challenge is the focus of our presentation. For this research project, we have created a graph of the topics represented in a subset of the scholarly literature and granted patents, in order to explore ways to constrain the visualization of this topic graph to emphasize usability. While many additional research areas remain, our initial findings suggest that such constraint enables users to easily explore the knowledge graph in way that maximizes understanding while minimizing user effort. Figure 1. Ball and stick diagram of the citing relationships among a select set of publications from Web of Science®.