TRADEOFFS IN GOOGLE DISTANCE ONLY VS A WORDNET HYBRID FOR QOS-ENABLED WEB SERVICE COMPOSITION

Abstract

This thesis proposes a hybrid approach in using Google Distance and WordNet together in a new method, called the Domain Independent Quality of Service (DIQOS) method for QoS-enabled web services discovery. Comparisons, using delay, recall, and precision metrics, between this hybrid approach and an earlier lightweight Google Distance-only based approach for web services discovery are provided. Further, our performance evaluation demonstrates as of yet undocumented trade-offs between Google distance, Google-WordNet distance, and WordNet distance approaches for similarity matching in the web services discovery phase. The impact of all approaches on QoS-enabled web service composition is described for representative web transactions in the travel domain. Findings include that the recall of signature matching increases by 5 to 15% for WordNet-assisted Google Distance DIQOS approach over the pure Google-distance DIQOS variant. WordNet-assisted Google Distance also shows 20-40 % increases in recall for signature matching compared to the WordNet only approach. Also, bigram-based, short sentence, and WordNet-based vector optimizations in specification similarity matching show an average of 25% increase in recall over a previous competitive method called the Flexible Ontology-Independent QOS-enabled method (FOIQOS). Our WordNet-assisted Google Distance method shows 34% increase in recall compared to FOIQOS. Our approaches produce 11% lower precision than FOIQOS, but FOIQOS is speedier as delays are 5 to 15 % lower.