Taiwanese LOM (TW LOM) annotation: Automatic description generation for learning objects

Owning to the great growth of e-learning objects, authorities (e.g. ADL and IEEE) have developed some metadata standards to facilitate the keyword search for various e-learning applications. However, too much fields, such as 58 blank fields in IEEE LOM, waiting for authors or annotators to fill up become an endless nightmare. In order to reach our vision of sharing and reusing valuable assets, the needs for an intelligent and automatic annotation system become more and more urgent. Among these 58 elements, it is the most difficult to extract the fittest solutions for Description, which calls for the advanced Chinese language processing technologies to generate the suitable value. We also adopted the Self-Organizing Map clustering method from Neural Networks, feature selection from Information Retrieval, and Latent Semantic Analysis from Linguistics to cope with the automatic annotation problem. In this paper, we proposed a novel approach called Clustering Descriptor, CD, to automatically generate the description metadata in TW LOM - a Learning Object Metadata application profile in Taiwan. Then, we conducted two experiments to evaluate the annotation quality for Description data element in terms of three parameters: clustering, feature weight, and semantics. Because of the benefits from clustering and feature weight, Clustering Descriptor achieved improvement in precision rate: 6.30% (clustering) and 8.60% (clustering plus feature weight) compared with the baseline.