Research Notes on the Practical Deployment of Semantic Knowledge Bases

Utilising semantic web technologies in knowledge management systems provides an opportunity for news/media providers to enrich their content with information from public datasets such as Linked Data Cloud and develop intelligent retrieval engines to search/browse the content. The semantic web technologies provide applications with machine-understandable metadata representing relevant knowledge domains, which can be reasoned by autonomous software agents to align the discrepancies in knowledge presentation by various contributing information sources and deliver intelligent query methods against the information and the underlying metadata. However, delivering a semantic-based knowledge management system requires the development and integration of processes that utilise a number of constantly evolving technologies ranging from using natural language processing for information extraction to ontology management and intelligent inferencing. The expertise required to develop such complex workflow cannot be provided off-the-shelf and is beyond the reach of most commercial organisations. The proposed paper reflects on the experience of developing an intelligent browsing engine for a commercial media application to propose a methodology for deploying semantic technologies in the construction of knowledge management systems. The developed semantic knowledge management system bootstraps the applications' knowledgebase by leveraging the rich amount of structured knowledge that is publicly available in the Linked Data Cloud using ontology mapping techniques. The knowledge management system also incorporates an information extraction system that aids the labour-intensive semantic tagging process by text-mining the manually annotated free-text image captions. The paper reports on an interesting and novel mutual-benefit workflow between the information extraction system and the knowledgebase. While the knowledgebase plays a crucial role in resolving disambiguation in the extracted information, the information extraction system, in addition to known entity recognition, was developed with the capacity of learning new facts with a confidence rating mechanism that either recommends the direct injection of new knowledge back into the knowledgebase, or its logging for manual verification.