3 ms·
'Recall' was at the promising prototype quality level, with results more Cuil-like, centered around linguistic concepts detected during indexing, than Google-li
by gojomo 16y ago
'Recall' was at the promising prototype quality level, with results more Cuil-like, centered around linguistic concepts detected during indexing, than Google-like. So for example the index was compact but there was no phrase-search, and even after receiving a page as a result to one query, a followup query with words on that page might return zero results if that exact set of words hadn't been concept-extracted.
It had nice graphs of concepts over time. But, its cleverness was probably overkill for such visualizations, except to the extent (IIRC) it disambiguated similar concepts from context. As the recent Google Books-based word frequency tool has shown, simple word and n-gram counts provide plemty of similar value.
Ultimately, though, as a bit of advanced, non-open-source Lisp code by a single expert, there was no one appropriate to productionize, maintain, and improve it after she joined Google.
(I work at the Internet Archive.)