Automated Service Selection Using Natural Language Processing

Abstract

With the huge number of services that are available online, requirements analysts face an overload of choice when they have to select the most suitable service that satisfies a set of customer requirements. Both service descriptions and requirements are often expressed in natural language (NL), and natural language processing (NLP) tools that can match requirements and service descriptions, while filtering out irrelevant options, might alleviate the problem of choice overload faced by analysts. In this paper, we propose a NLP approach based on Knowledge Graphs that automates the process of service selection by ranking the service descriptions depending on their NL similarity with the requirements. To evaluate the approach, we have performed an experiment with 28 customer requirements and 91 service descriptions, previously ranked by a human assessor. We selected the top-15 services, which were ranked with the proposed approach, and found 53% similar results with respect to top-15 services of the manual ranking. The same task, performed with the traditional cosine similarity ranking, produces only 13% similar results. The outcomes of our experiment are promising, and new insights have also emerged for further improvement of the proposed technique.