Analysing the content of Web 2.0 documents by using a hybrid approach

Abstract

User involvement in Web 2.0 has made a significant contribution to the increase in the amount of multimedia content on the Web. Images are one of the most used media, shared across the network to mark user experience in daily life. Interactive applications have allowed users to participate in describing these images, usually in the form of free text, thus gradually enriching the images' descriptions. Nevertheless, often these images are left with crude or no description. Web search engines such as Google and Yahoo provide text based searching to find images by mapping query concepts with the text description of the image, thus limiting the information discovery to material with good text descriptions. A similar issue is faced by text based search provided by Web 2.0 applications. Images with less description might not contain adequate information while images with no description will be useless as they will become unsearchable by a text based search. Therefore, there is an urgent need to investigate ways to produce high quality information to provide insight into the document content. The aim of this research is to investigate a means to improve the capability of information retrieval by utilizing Web 2.0 content, the Semantic Web and other emerging technologies. A hybrid approach is proposed which analyses two main aspects of Web 2.0 content, namely text and images. The text analysis consists of using Natural Language Processing and ontologies. The aim of the text analysis is to translate free text descriptions into a semantic information model tailored to Semantic Web standards. Image analysis is developed using machine learning tools and is assessed using ROC analysis. The aim of the image analysis is to develop an image classifier exemplar to identify information in images based on their visual features. The hybrid approach is evaluated based on standard information retrieval performance metrics, precision and recall. The example semantic information model has structured and enriched the textual content thus providing better retrieval results compared to conventional tag based search. The image classifier is shown to be useful for providing additional information about image content. Each of the approaches has its own strengths and they complement each other in different scenarios. The thesis demonstrates that the hybrid approach has improved information retrieval performance compared to either of the contributing techniques used separately.

More information

Identifiers

ORCID for Wendy Hall: orcid.org/0000-0003-4327-7811

Catalogue record

Export record