Essays in Innovation Management
- Gao, Xiaochen Kev
- Advisor(s): Krishnan, Vish
Abstract
Innovation management is increasingly recognized as critical to the growth and sustenance of firms, and researchers have also begun focusing on innovation as a sub-discipline of business administration. In recent years, advancements in machine learning (ML) and deep learning (DL) have produced remarkable results in various domains of applications. The dissertation comprises three chapters, two of which apply the latest methodologies of ML and DL to innovation management research.
The first chapter studies the examination process of patent applications, by which innovators seek to protect their inventions and intellectual properties (IP) from infringements, and how novelty can be assessed. We collect data published by the U.S. Patent and Trademark Office (USPTO) and apply a DL model for text classification. We find that the texts alone cannot yield good predictive performance under even an advanced text classification model. To tackle this problem, we take two major steps. First, we constructed handcrafted features to aid the model. Second, we design a graph/tree representation for patent claims, both taking account of the internal and external structures within and among claims. We then apply a customized graphical neural network (GNN) model. Experimental tests show that the graph approach dramatically increases the prediction performance.
The second chapter considers the implementation of innovation projects. Based on comprehensive data from USASpending.gov about federal government contracts, we aim to perform predictive and causal analysis of the contract performances in the form of cost overrun and schedule overrun. We find that a gradient-boosting decision tree (GBDT) model specially designed to tackle categorical variables suits the dataset well and yields very good predictive performance. We then use the statistically robust metric SHAP (SHapley Additive exPlanations) to interpret the model. Furthermore, we explore causal machine learning. We affirm that several contract characteristics, including the type of contract pricing terms have a causal impact on the contract performance, and the level of bidding competition does not causally influence it.
The third chapter theoretically investigates disruptive innovation with a stylized economic model and focuses on the incumbent's strategies in managing incoming disruptive threats. We formalize and operationalize disruptive innovation theory with a novel modeling approach of disruptive accessibility. We further study how the incumbent can leverage sustaining innovation and product strategies to foil disruptive threats. Contrary to conventional beliefs that offering a product portfolio cannot be optimal for a fixed-cost firm, we find that it can be optimal when the firm is not a monopoly due to competition mitigation and innovation incentivization. In addition, the advantage of offering a product portfolio can be more prominent when disruption intensifies, and the product strategy can be an effective measure to combat a disruptive competitor.