Article Preview
TopIntroduction
According to Hammer and Champy (2003), business process re-engineering (BPR) is “fundamental rethinking and radical redesign of business process to achieve dramatic improvements in critical measures of performance such as cost, service and speed”. The aim of BPR is to “better support an organization’s mission and reduce costs of operations”. One major challenge of BPR is to use information technologies to arm business services and coordinate the services of a complex business process to meet the organization’s changing business goals (Oneill & Sohal, 1999). In recent years, business services are usually developed by coordinating existing open APIs or services on the Internet, rather than developed from scratch. There is a huge amount of services and open APIs on the web, and how to discover the most suitable services or APIs from the big data to achieve the constantly changing business requirements is a big problem for business process re-engineers. The open APIs and services are collectively called services in this paper.
There are many established approaches to tackle the problem (Alfazi, et al., 2016; Lo, et al., 2015; Mittal, et al., 2015). These approaches have the advantage of handling service data of a certain scale, but cannot handle the massive service data (Wang & Zhao, 2019). Nowadays, deep learning has made a significant progress in the artificial intelligence task learning, e.g. natural language processing (NLP), image recognition, speech recognition, and so on. Deep learning is widely used in handling big data and achieves significant results due to the fact that: 1) deep learning techniques can extract key features from a variety of complex information that can reflect valuable information about business; 2) deep learning techniques can train a stable neural network model through a large amount of data at the early stage, and thereafter is able to quickly predict and respond to various emerging business processes.
Therefore in this paper, the authors propose a deep learning-based Massive service discovery approach using Recurrent Attention and Feature Fusion (MassRAFF). This approach matches the NL-based business goals to services description text using two novel techniques - Recurrent Attention and Feature fusion. First, MassRAFF uses the similarity of Word2Vec (Mikolov et al., 2013) to calculate the relevant topic of the business goals, then adopts TextRank to extract key sentences from each service description text, and inputs the business goal and corresponding services description text into the semantic sentence matching neural network (SSMNN) to calculate the similarity. Finally, MassRAFF finds the most suitable services through sorting by degree of similarity, which can then be composed and coordinated to implement a new business process. The authors also carried out a set of experiments on the PWeb service dataset to verify the effectiveness of the approach. The rest of the paper is organized as follows: the second section discusses related work on the relationship between BPR, big data, massive service discovery and deep learning techniques. The third section introduces the MassRAFF approach. The fourth section introduces the neural network SSMNN used in MassRAFF. The fifth section shows the experiments with the MassRAFF approach and analyzes the results. The sixth section discusses the threats to validity. The last section concludes this work and introduces the future work.