Read Schöpper and Kersten 2021

[divilian]

(See Zotero, "General background" folder.)

[vgcagle]

Companies consider their supply chain data as proprietary and are cautious about sharing it with others. Vertical mapping dimension: contains supplier-customer relations Horizontal mapping dimension: competitors and complementors. Lack of transparency in supply chains is the main barrier to risk management in supply chains. Horizontal mining approaches had two categories: content-based and link-based procedures. Link-based used supervised learning techniques. Content-based procedures primarily used rule-based approaches. Vertical mining approaches are newer. All approaches are content-based, and the majority use supervised learning techniques. They created a taxonomy for supply chain mapping approaches.

[harmonyrose]

Schopper and Kersten discuss the challenges that come with the current lack of supply chain visibility and review the existing literature that seeks to solve this problem. Some of the main challenges of supply chain mapping include sources in different languages, ambiguous information, limited data, and low recall. The authors define some basic requirements of supply chain mapping, including directed relations, relation dynamics, location, multiple company roles, and more. They discover that the current literature essentially only addresses directed relations. They also found that research in this field is at an early stage, and it is fragmented in terms of its time scale and conceptual knowledge basis. This has limited recent approaches’ ability to learn from prior research’s strengths and weaknesses.

The authors define the two dimensions of a supply network: horizontal and vertical. The horizontal dimension refers to the number of tiers across the whole supply chain. Content-based horizontal approaches assume business entities and their relations are present in news articles’ content and link-based approaches assume that companies and their competitors are more likely to be linked in their web sources. Both strategies present challenges, and more recent literature has focused more on vertical approaches.

The vertical dimension refers to the number of suppliers and customers at each level. Vertical mining approaches are all content-based, supervised, and are akin to our research. The methodology typically involves labeling sentences that contain certain types of business relations and then using classification techniques to label new sentences. It seemed like the Wichmen et. al. paper is the most recent, state-of-the-art approach, although it has limitations like high potential for overfitting, noisy data, and having too many classes with not enough data.