Combination of ELMo Representation and CNN Approaches to Enhance Service Discovery

Zhao Huang,Wei Zhao

doi:10.1109/access.2020.3009393

Abstract

With the rapid growth of Web services, the demand for discovering the optimal services to satisfy the users' requirements is no longer an easy task. The critical issue in the process of service discovery is to conduct a similarity calculation. To solve such an issue, this study proposes an effective approach that combines the Embeddings from Language Models (ELMo) representation and Convolutional Neural Network (CNN) to obtain a more accurate similarity score for retrieving target Web services. More specifically, first, the study adopts the ELMo model to generate effective word representations for capturing the sufficient information from services and queries. Then, the word representations are used to compose a similarity matrix, which will be taken as the input for the CNN to learn the matching relationships. Finally, the combination of the ELMo representation and CNN is used to address the representation and interaction processes within the matching task to improve the service discovery performance. The results demonstrate the effectiveness of our proposed approach for retrieving better targeted Web services.

Highlights

Recent years have witnessed the continuous growth of publicly available Web services on the Internet
The following research questions will be investigated in this study: (1) What methods can obtain the word representations that capture sufficiently the syntactic and semantic information and suit the dynamic context? (2) How does the method learn the matching relationships between the descriptions of the service and queries at a lexical level during the service discovery process? (3) What refinements can be made to improve the performance for service discovery according to the specific characteristics of the neural matching models? To answer these questions, this study develops a service discovery method that combines the Embeddings from Language Models (ELMo) representation and Convolutional Neural Network (CNN) to obtain accurate similarity scores for retrieving the optimal target Web services
The results of this study show the great performance of our ELMo-CNN-based discovery method, which suggests that the integration of the ELMo representation and the Convolutional Neural Network can generate great efficiency for service discovery

Summary

Introduction

Recent years have witnessed the continuous growth of publicly available Web services on the Internet. The research that focuses on retrieving the most relevant Web services to meet the specific user’s requirements has received a large amount of attention [2], and it has triggered a considerable amount of effort [3], [4]. Among these studies, service discovery is becoming a critical issue in developing software applications. Service discovery refers to identifying a set of candidate services from a service registry by comparing the matching degree with the given requirements of the users [5]. It can support businesses to better understand users and make quick improvements, to adapt their business to the environment [7]

Objectives

Methods

Findings

Conclusion