Logo image
Back
Discovering innovation synergies across technologies and firms: a data science and machine learning perspective
Dissertation

Discovering innovation synergies across technologies and firms: a data science and machine learning perspective

Junho Yoon
University of Iowa
Doctor of Philosophy (PhD), University of Iowa
Summer 2024
DOI: 10.25820/etd.007715
pdf
JunhoYoon_Thesis1.28 MB
Embargoed Access, Embargo ends: 08/30/2028

Abstract

Technological innovation does not transpire in isolation. Technologies are brought to- gether across boundaries to develop groundbreaking innovations. Firms collaborate through alliances or engage in mergers and acquisitions (M&A) to accelerate their technological development faster than individual R&D. This dissertation examines the synergy of such innovations between technologies or firms. For this purpose, we leverage techniques from machine learning (ML) and data science to address three key questions: 1) how to discover integrations that span technological boundaries, 2) how to locate appropriate alliance partners, and 3) how to predict the market value impact of tech M&As on the merged entities. First, we present a data-driven framework designed to identify boundary-spanning technological innovation systems (TISs) by analyzing textual data from over four million patents. This approach aims to address the vertical structure of the current Cooperative Patent Classification (CPC) system. Our framework’s utility is validated through its ability to predict the quantity and quality of future innovations across different technology classes. Our novel TIS-based innovation metrics, which utilize patent activity in related technology classes, are found to cue future innovation intensity. Experiments with machine learning models are conducted to further tease out the predictive utility of our TIS discovery framework. Second, we propose an automated alliance prediction framework that can facilitate part- ner search as well as provide valuable intelligence to third parties such as analysts and investors. Our graph neural network-based framework utilizes a key alliance theory on relational pluralism, which refers to multitudes of relationships that exist between firms that can provide an understanding of how firms interact and collaborate. To operationalize our prediction framework, we compile a rigorous firm-level network data set derived from 8,739 alliances between 11,499 firms in 11 high-tech industries through the period of 1990 - 2018. Our theory-driven predictive models incorporate multiple innovation-based relations—interfirm collaborations, human capital flow, and knowledge spillovers—into graph neural networks (GNNs). Our prediction results show that the plurality of interfirm relations collectively contributes to superior predictive performance across varying evaluation metrics, highlight- ing the practical utility of our predictive models as a partner search instrument. We discuss the benefits from our prediction framework to different stakeholders in practice, as well as several contributions to the alliance literature. Third, we present a machine learning-based framework designed to forecast M&A synergy in terms of market value. This framework integrates complex innovation profiles using cutting-edge deep learning models, such as graph neural networks (GNNs). Additionally, we incorporate a role-aware mutual attention layer between acquiring and target companies, enabling more expressive representations of their innovation profiles. Our framework has proven highly effective in predicting M&A synergy, as evidenced by our computational experiments.

Details

Metrics

37 Record Views
Logo image