Summarizing Learning Materials Using Graph Based Multi-Document Summarization

Introduction

To access relevant information quickly in today’s vast amount of online information, the automatic text summarization (ATS) is an important and timely too1. The ATS produces a summary text from the given input text whose size is less than half of the original text and contains important information (Radev et al., 2002). The ATS process can be either single document summarization (SDS) or multi-document summarization (MDS). The single document summarization generates the summary of a single document, whereas the multi-document summarization generates the summary of a group of related or unrelated documents. The multi-document summary should contain the relevant information shared among all the documents, plus the unique information about some of the documents which are essential (Goldstein et al., 2000).

This article generates a summary of learning materials using the MDS with sentence similarity graphs. It uses the graph structure, maximal clique to provide a concept oriented summary (Tomita et al., 2011). A clique is a complete sub graph of a graph. Since all nodes (sentences) in a clique are related to each other, each clique represents one concept or main idea of the given text. A maximal clique is a clique which is not a proper subset of any other clique. This article covers all important concepts of the given text based learning materials by selecting summary sentences from a diverse set of maximal cliques of the sentence graphs of the given input text. Throughout this article clique means maximal clique.

The same teacher teaches for all the students in the class at the same time in traditional classroom teaching. But, different students have different learning capacity due to the knowledge difference among the students (Wang & Cai, 2009). Some students may not have an interest in getting more details of a topic. Some may not understand if the content is too long. Therefore, the E-learning is used nowadays for improving the learning interests and efficiency of the learners.

In E-learning, the learning material is delivered to the remote learners through a computer network. The learning object of an E-learning environment is a chunk of electronic content that can be accessed individually. Since each learning resource contains a vast amount of information to be read, the learners feel difficult when they are reading at their earlier stage. About 60% of students wanted the summarized material rather than the entire content (Shimada et al., 2015). Hence, providing a summary for either a single learning material or a group of learning materials on a specific topic can help the learners understand the resource better. This is the reason why does this article aim to summarize the learning materials.

This article aims to provide a dynamic multi-document summary of academic learning materials especially computer science theory subjects in the form of textual documents from various sources such as an E-learning resource, lecture notes, a learner’s class notes, previous year notes, book chapters, and some other online documents based on a learner’s requests. Why does this article aim to provide dynamic summaries? To satisfy the varying summary requirements of different kinds of learners, dynamic summaries are provided. A learner can decide the summary size up to fifty percent of the length of the document-set (set of input documents). Once get a summary, if the learner is not satisfied, he may go for further levels of summaries. This summary can be used as a preview before reading the text first time and can be useful during revision (Baralis & Cagliero, 2016). Also, it can be used as a summarized learning material to learners with fewer skills, students studying in part-time mode, and students studying in distance education.

The structure of this work is framed as follows. The Background section discusses the related works on MDS, graph based MDS, and the E-learning context and compares them with the proposed work. The Proposed work section describes the problem statement, overview, feature extraction, sentence score calculation, summary generation, and summary evaluation of the proposed work. The Evaluation results and discussion section shows the results of the intrinsic content based evaluation and the extrinsic task based evaluation and discusses these results. The Conclusion section concludes the current work and suggests the future work. The reference section lists the references cited in this work.