PewePro 2

Domain Model Enrichment Based on External Sources

A method for enriching domain model with relations acquired from annotations inserted by users into learning content.

Overview
Inputs: Annotations inserted into learning texts
Outputs: Relations between portions of educational content and concepts from domain model.

Addressed Problems

Creation and enrichment of a domain model is a complex and time consuming task, which requires an expert knowledge of a domain and significant effort. We use annotations inserted into educational content as source of metadata which we use to enrich content and domain model of educational course.

Description

Annotations inserted by students and teachers into the educational content represent metadata attached to certain part of educational text. While annotations represent metadata, they also can be used to discover relations between various parts of the educational text. By inserting the similar annotations into several parts of the document user expresses that these parts describe similar topics.

We believe that it is necessary to provide specific types of annotations to users in order to successfully extract metadata. Simple text comments used to insert any type of information lead to inconsistent and ambiguous metadata, while specialized types of annotations result in metadata which are easier to interpret correctly. We currently use following types of annotations to extract metadata:

Students use highlight annotations to select important parts of educational texts. Highlights are private; students cannot see highlights of other students. Therefore students are not influenced by highlights of their peers when selecting important parts of the text. We currently collect highlights from students for further processing and identification of important fragments of the texts.

Our effort is currently focused on links to external sources. External sources inserted into related educational text describe similar topics as the text does. External sources also typically explain more concepts than learning objects do, hence one external source can be inserted into various learning objects.

We designed a method for discovering relations from external sources, which consist from following steps:

  1. Analysis of external sources
  2. Metadata (relations) creation

Since external source can be in any format or language, we extract readable text from external source and translate it to English using Google Translate web service. In further processing if the source we use the translation as content. In analysis step we detect known concepts from domain model within the content of external source. Goal of this step is to create relations between external sources and concepts, therefore integrating them into educational course.

In metadata creation step we build a graph using known relations between learning objects, concepts and external sources (e.g. external source is inserted into learning object) resulting in tripartite graph. We apply spreading activation algorithm to the graph, spreading activation from the objects to count similarity between objects resulting in relations between most similar objects.

Publications

  1. Mihál, V., Bieliková, M.: Domain Model Relations Discovering in Educational Texts based on User Created Annotations. In: ICL 2011: 14th International Conference on Interactive Collaborative Learning and 11th International Conference Virtual University. Piešťany, Slovakia, September 21-23,2011. Piscataway : IEEE, 2011, pp. 542-547.