Forecasting future bigrams and promising patents: introducing text-based link prediction

Abstract

Purpose In recent years patents have become a very popular data source for forecasting technological changes. However, since a vast amount of patents are “worthless” (Moore, 2005), there is a need to identify the promising ones. For this purpose, previous approaches have mainly used bibliographic data, thus neglecting the benefits of textual data, such as instant accessibility at patent disclosure. To leverage these benefits, this study aims to develop an approach that uses textual patent data for predicting promising patents. Design/methodology/approach For the identification of promising patents, the authors propose a novel approach which combines link prediction with textual patent data. Thereby the authors are able to predict the emergence of hitherto unmentioned bigrams. By mapping these future bigrams to recent patents, the authors are able to distinguish between promising and nonpromising patents. To validate this approach, the authors apply the methodology to the case example of camera technology. Findings The authors identify stochastic gradient descent as a suitable algorithm with both a receiver operating characteristic area under curve score and a positive predictive value of 78%, which outperforms chance by a factor of two. In addition, the authors present promising camera patents for diverse application fields, such as cameras for surgical systems, cameras for rearview vision systems in vehicles or light amplification by stimulated emission of radiation detection and ranging cameras for three-dimensional imaging. Research limitations/implications This study contributes in at least three directions to scholarship. First, the authors introduce a novel approach by combining link prediction with textual patent analysis and, in this way, leverage the benefits of both worlds. Second, the authors add to all theories that regard novel technologies as a recombination of existing technologies in presenting word combinations from textual data as a suitable instrument for revealing recombination in patents. And third, the approach can be used by scholars as a complementary or even integrative tool with conventional forecasting methods like the Delphi technique or Scenario planning. Practical implications At least three practical implications arise from the study. First, incumbent firms of a technology branch can use this approach as an early-warning system to identify technological change and to identify opportunities related to their company’s technological competence and provide inspiration for new ideas. Second, companies seeking to tap into new markets may also be interested in the approach as managers could anticipate whether their company’s technological competences are in line with upcoming trends. Third, the approach may be used as a supportive tool for various purposes, such as investment decisions or technology life cycle analysis. Originality/value The approach introduces textual patent data as suitable means for forecasting activities. As the statistical validation reveals, the promising patents identified by the approach are cited significantly more often than patents with less promising prospects.

Big data, machine learning and uncertainty in foresight studies

“…The mix of methods used in foresight studies, as illustrated in Figure 5, clearly demonstrates how BDML techniques are integrated with both quantitative and qualitative methodologies. Many authors (Denter et al. , 2022; Kayser et al.…”

“…The mix of methods used in foresight studies, as illustrated in Figure 5, clearly demonstrates how BDML techniques are integrated with both quantitative and qualitative methodologies. Many authors (Denter et al, 2022;Kayser et al, 2014;Kayser and Shala, 2020;Nazarenko et al, 2021;Santo et al, 2006;Yufei et al, 2016) describe examples of how BDML could support foresight practice and decision-making, by providing up-to-date information, guiding creative processes and potentially avoiding expert biases.…”

The Watchful Waiting Strategy in Standard-Essential Patents: The Case of 5G Technology

“…13 We include two control variables which allow us to adjust for other factors that affect the regression results. Annual particularities are controlled by computing dummy variables for the filing year [62]. Additionally, we control for technology subclasses by computing dummy variables for each of the WIPO-35 technology classes [63].…”

“…Stratified random sampling ensures a similar distribution of SEPs and non-SEPs among both datasets [60]. Next, z-score standardization is applied to the training dataset and adapted to the test dataset in order to reduce computational resources and enable the deep neural network models to converge more quickly [62,85]. In the learning phase, deep neural networks are trained on the basis of the 80% training dataset.…”

Identifying the Technology Opportunities and the Technology Taxonomy for Railway Static Inverters With Patent Data Analytics

“…The building of a Document-Term Matrix (DTM) is important because it is used in LDA topical modeling, technology mapping analysis, and social network analysis, which are analysis methodologies used in this study [35][36][37]. For building the Document-Term Matrix (DTM), we have identified technical keywords through TF-IDF analysis, targeting text data from documents [38], [39].…”