Scratch-DKG: A Framework for Constructing Scratch Domain Knowledge Graph

Abstract

With the rapid development of programming platforms, how to utilize the tremendous amount of data produced by the platforms, such as Scatch, has been a big challenge to researchers. The growing data is not only huge, but also heterogeneous and diverse, leading that the existing tools cannot effectively extract valuable information. In this article, considering particular features of Scratch data, we propose an effective framework about constructing a Scratch Domain Knowledge Graph (Scratch-DKG). Our framework includes four modules which are designed to process the semi-structured data, users profile data, projects data and programming knowledge points, respectively. For webpages, we design a template-based wrapper method to extract triples from the semi-structured data. As for users profile data, we improve DeepDive, which is a useful tool to extract information but with the problem of wrong labeling, to extract knowledge triples by the proposed Secondary Labeling Algorithm. For projects data, we propose an advanced keywords extraction method (S-TextRank) to extract keywords triples. For programming knowledge points, we develop a frequently contiguous block combinations mining algorithm to extract the potential domain information of Scratch. Finally, extensive experiments are carried out to evaluate the performance of our proposed methods. The experimental results show that, compared to other competing methods, our proposal can extract more correct and comprehensive Scratch triples.