Mining Context-Specific Web Knowledge : an Experimental Dictionary-based Approach

This work presents an experimental semantic approach for mining knowledge from the World Wide Web (WWW). The main goal is to build a context-specific knowledge base from web documents. The basic idea is to use a reference knowledge provided by a dictionary as the indexing structure of domain-specific computed knowledge instances organised in the form of interlinked text words. The WordNet lexical database has been used as reference knowledge for the English web documents. Both the reference and the computed knowledge are actually conceived as word graphs. Graph is considered here as a powerful way to represent structured knowledge. This assumption has many consequences on the way knowledge can be explored and similar knowledge patterns can be identified. In order to identify context-specific elements in knowledge graphs, the novel semantic concept of “minutia” has been introduced. A preliminary evaluation of the efficacy of the proposed approach has been carried out. A fair comparison strategy with other non-semantic competing approaches is currently under investigation.