A leaked memo from Satya Nadella, Microsoft’s SVP of R&D, Online Services re Project Kumo (c.f. my comments on technewsworld) yields some very interesting ‘facts’: “40 percent of queries go unanswered; half of queries are about searchers returning to previous tasks; and 46 percent of search sessions are longer than 20 minutes. These and many other learnings suggest that customers often don’t find what they need from search today”. Setting aside the question of how an unanswered search is defined / quantified and if we accept the validity of the stats, it says that too many searches are abandoned, too many re-started and too much time wasted.
How much more problematic then the search for images which relies on text tagging to describe content that given it’s semantic richness is fundamentally unsuited to ‘description’. The hollow boast in search results that the engine has retrieved x million image search results in y milliseconds is to me more depressive than impressive. Surely it’s time for a different approach.
The more I and the Imprezzeo team ‘freestyle search’ image collections by clicking on examples of what we want, refining and traversing content to find what we want; the more we build features that allow you for example to highlight part of an image as the example upon which to base the search (only possible with an underlying visual search platform); the more we hone our engine to provide great results across a diverse range of content - the more we are convinced that the days of text-based-only image search are numbered. Shameless plug perhaps but reality, definitely.