Guest talk / Seminar: John Tait
Guest talk: John Tait
The term Semantic Search is becoming fashionable, but there are a number of problems with the term.
1) There are at least two forms of semantic search. One is based more-or-less hand programmed knowledge sources, like domain ontologies or thesauri. The other is based on emergent properties of the data being searched, using technques like Latent Semantic Analysis or clustering. It is far from clear that the results of applying the two approaches are similar or even compatible.
2) It is often assumed that semantic search is in some sense different from surface text search: which implies that normal old-fashioned Google search (for example) is equivalent to randon string search, when of course the underlying statistics depend critically on the fact that both the queries and copora are natural language (English or German) words with underlying semantics.
3) Semantic Search depends critically on text annotation processes during indexing: but these are potentially corruptable by malefactors. How can this be prevented?
The seminar will explore these three issues, and attempt to find a better definition of the term semantic search and to identify soem ways forward.