An Intelligent Web Service Discovery Framework Based on Improved Biterm Topic Model

Abstract

Given the proliferation of Web services, effectively identifying the most suitable ones based on user queries poses a formidable challenge. In response to the challenges posed by the reduced efficiency of service discovery methods as the number of services continues to grow and the limited co-occurrence of word frequencies in service description documents, this study proposes an intelligent service discovery framework based on probabilistic topic distribution. The framework utilizes the Biterm Topic Model to extract the probabilistic topic distribution from both service description documents and user requirements. It then performs functional clustering and service matching based on this probabilistic topic distribution, resulting in a set of candidate services. To expedite the training process of the topic model, a topic model training algorithm employing sampling recombination is introduced, which reorganizes the topic sampling process and reduces training time. Additionally, a functional clustering algorithm based on weighted connected graphs is presented to enhance the quality of clustering. Experimental results validate the effectiveness of the proposed framework, which significantly reduces the training time required for the topic model and service discovery while improving the accuracy of service discovery and the normalized discounted cumulative gain.