Implicit Semantics and Business Networks

At the Semantic Web SIG on September 1, 2010 in the Cubberley Community Center, Palo Alto, David Meyer and Michael Bechauf of SAP presented two dramatically different approaches for extracting semantics from business networks.

Meyer, the senior vice president of on-demand, productivity, and sustainability solution management at SAP, began his presentation with the observation that meaning is everywhere, created constantly as we perform our daily work, exchange information, and leave behind context. He proposed enabling the systems with which we work—CRM, HRM, ERP—to feed into and pull from a massive stream that will accumulate this context. SAP has already built this functionality into SAP StreamWork, a collaboration tool that you can try of for free at http://sapstreamwork.com.

These systems, Meyer proposed, will add data to the stream in triplets of actor, verb, and object, forming a network graph. By mining the graph, we will be able to answer questions such as who is working on a particular customer account, even though that data originates from disparate systems.

Meyer explained that the larger the stream, the more the context, the better the system can answer questions. The value of the stream increases further through federation across supply chains, allowing for massive accumulations of context as supply chains grow ever longer and more networked.

Speaking after Meyer, Bechauf, vice president of platform strategy at SAP, also tackled the problem of extracting semantics from business networks but with a different objective in mind: reducing the high cost of integrating business systems. Because of the high cost of integration, Bechauf said, “information in the global supply chain does not flow as smoothly as it could.” And eighty percent of the interoperability problem, according to Bechauf, relates to semantics.

Bechauf explained that XML, once touted as the solution to interoperability, created a plethora of competing schemas. Despite the rise of popular standards such as the Open Applications Group in finance and ISO 20022 in manufacturing, consolidation has been slow and many systems continue to adhere to legacy standards. XML itself defines syntax, the structural relationship among elements within a schema, but not semantics, the meaning of individual elements. XML may define a coupon as part of a purchase order, but it cannot tell us whether a field named coupon in one schema corresponds to a field of the same name in another schema.

Bechauf advocated adding semantic properties to metadata elements based on business context in accordance with the ISO 11179 standard. (ISO 11179 defines metadata registries as well as the classification and semantic description of metadata.) Bechauf and researchers at SAP have created a large database of schemas and have worked with subject matter experts to populate these semantic properties and map fields between the schemas.

This collection of schemas, according to Bechauf, has allowed researchers at SAP to look for patterns and calculate the predictive value of the semantic attributes. The toughest problem, according to Bechauf, is building tools to exploit these patterns and find predicted matches. Bechauf expects that in the near future these tools will remain semi-automatic, finding matches through heuristics but requiring human judgment for validation. Bechauf anticipates the tools will increase in accuracy as researchers collect more schemas for analysis and that this increased accuracy will significantly reduce the cost of integration.

Asked whether their distinct approaches to extracting semantics from business networks had anything in common, Meyer said that the architects for the two programs were in conversation looking for ways to learn from and build upon each other’s work.