Validating Syntactic Correctness Using Unsupervised Clustering Algorithms

Abstract

When developing a complex system in an open platform setting, users need to compose and maintain a systematic requirement specification. This paper proposes a solution to guarantee a syntactically accurate requirement specification that minimizes the ambiguity caused by ungrammatical sentences. Our system has a set of standard jargon and templates that are used as a guideline to write grammatically correct sentences. Given a database of standard technical Korean (STK) templates, the system that we have designed and implemented divides a new sentence into a specific cluster. If the system finds an identical template in a cluster, it confirms the new sentence as a sound one. Otherwise, the system uses unsupervised clustering algorithms to return the template that most closely resembles the syntax of the inputted sentence. We tested our proposed system in the field of open platform development for a railway train. In the experiment, our system learned to partition templates into clusters while reducing null attributes of an instance using the autoencoding procedure. Given a set of clusters, the system was able to successfully recommend templates that were syntactically similar to the structure of the inputted sentence. Since the degree of similarity for 500 instances was 97.00% on average, we conclude that our robust system can provide an appropriate template that users can use to modify their syntactically incorrect sentences.