Machine-Learning based Semantic Similarity Measures to Assist Discovery and Reuse of Data Exchange XML Schema

Interoperability will remain as an unsolvable problem forever at least until when you read this dissertation. Actually, we have spent unbelievable amounts of money due to lack of interoperability among any systems. According to the NIST reports, the estimated interoperability cost nearly reaches to a score billion US dollars just in two industries (i.e., the capital industry and part of the automotive industry) in US only. Can you imagine the total amounts over the world just a single year? To this reason, we have been discussed and agreed upon standard in data-centric integration societies. That was worked for a moment; however, everybody has wanted more than standards. This is an unbreakable chain of interoperability. Even though the standard-based integration approaches have definitely been failed, those experiences also give us much more lessons and opportunities. We recognized that the standard specifications had been evolved over the history. The failure of such approaches is surely due to the failure of consistent tracking of the evolution. In other words, the time is now, not very late to define a formal scheme and assistant tools to manage and support the schema (i.e., standard specifications) evolution. In this sense, the dissertation first aims at defining such a scheme, namely Model Development Life Cycle (MDLC) management, within which we formally identify and clarify the supporting activities and necessary information models. More specifically, the dissertation concentrates on the virtually initial activity of the MDLC, i.e., Schema Discovery activity. In fact, the core functions in schema discovery are reused in the other activities in variant forms. To the end, we designed a schema discovery engine, namely a Semantics Aware Lookup Assistant (SALA). The SALA interprets a set of given data exchange requirements and finds the most appropriate (integration) schema(s) meeting the requirements from a repository. The schema discovery is a new research area; but similar researches have been done under the title of schema matching and integration. We reviewed a bunch of existing works and approaches in the schema matching, and concluded that the main stream is to measure the semantic similarities among schemas, among which the measures indicate the degree of semantic correspondences. The principle in computing the semantic similarities is a convergence of two technology axes separately developed – the schema matching and the machine learning. With the philosophy, the research applies a few favorable machine-learning techniques (e.g., ANN, PCA, PLS) to the schema matching in …