A hybrid approach to detecting technological recombination based on text mining and patent network analysis

Abstract

Detecting promising technology groups for recombination holds the promise of great value for R&D managers and technology policymakers, especially if the technologies in question can be detected before they have been combined. However, predicting the future is always easier said than done. In this regard, Arthur's theory (The nature of technology: what it is and how it evolves, Free Press, New York, 2009) on the nature of technologies and how science evolves, coupled with Kuhn's theory of scientific revolutions (Kuhn in The structure of scientific revolutions, 1st edn, University of Chicago Press, Chicago, p 3, 1962), may serve as the basis of a shrewd methodological framework for forecasting recombinative innovation. These theories help us to set out quantifiable criteria and decomposable steps to identify research patterns at each stage of a scientific revolution. The first step in the framework is to construct a conceptual model of the target technology domain, which helps to refine a reasonable search strategy. With the model built, the landscape of a fieldits communities, its technologies, and their interactions-is fleshed out through community detection and network analysis based on a set of quantifiable criteria. The aim is to map normal patterns of research in the domain under study so as to highlight which technologies might contribute to a structural deepening of technological recombinations. Probability analysis helps to detect and group candidate technologies for possible recombination and further manual analysis by experts. To demonstrate how the framework works in practice, we conducted an empirical study on AI research in China. We explored the development potential of recombinative technologies by zooming in on the top patent assignees in the field and their innovations. In conjunction with expert analysis, the results reveal the cooperative and competitive relationships among these technology holders and opportunities for future innovation through technological recombinations.

Tracking developments in artificial intelligence research: constructing and applying a new search strategy

“…Such papers were not relevant to our purpose of developing a conceptual search strategy. A similar observation is found in Zhou et al (2019) in their search of "artificial intelligence" in the WoS "Title" field. To anchor our search for additional keywords germane to the core of research on artificial intelligence, we focused on the WoS subject categories of "Computer Science, Artificial Intelligence", "Computer Science, Information Systems", "Computer Science, Interdisciplinary Applications", "Computer Science, Theory & Methods", "Computer Science, Software Engineering", "Computer Science, Hardware & Architecture", "Computer Science, Cybernetics", and "Robotics".…”

Discovering Technology Opportunity by Keyword-Based Patent Analysis: A Hybrid Approach of Morphology Analysis and USIT

“…Since patent documents consist of two parts: structured part (e.g., patent citation, patent classification) and non-structured part (e.g., abstracts, claims, and description) [45], the techniques of patent analysis for TOD could largely be classified into two categories: the target of analysis aiming to structured part (e.g., patent co-classification analysis [46,47], patent citation analysis [48], etc.) and non-structured part (e.g., keyword analysis [10,49], patent network analysis [50], etc.). Among these techniques, keyword analysis combined with text mining techniques is particularly useful for scrutinizing the technical content of patents with the aim of decomposing technology to capture the overall technology features [30].…”

Bayesian Structure Learning and Visualization for Technology Analysis

“…To analyze XR patent documents, we have to preprocess the retrieved patent documents, that is, transform the patent documents into structured data such as table of database management system, because statistics and machine learning algorithms require structured data type for patent data analysis [8,10,13]. There are various approaches to patent data preprocessing, such as text mining and natural language processing [14][15][16]. The structured data that has been preprocessed is a matrix that has patent document and keyword for its row and column, respectively.…”