InteliWISE AVATAR

Efficiency issues forced us to change the so called „extensive“ paradigm, characteristic to AIML approach, because of large rule sets and a large token tree to be built and to be scanned. That‘s also why we didn‘t choose to use neural networks: they are not so efficient for large data sets and their topology management and maintenance is not a polynomial problem. For now, 98% of responses are produced in less than 0.5 s, with none above 4 s (single Intel Xeon 3.0 GHz). We must consider InteliWISE AVATAR is not a simple web search engine. It maintains not only semantic and lexical knowledge, but also clients‘ knowledge bases, dictionary definitions, flexing dictionaries and at most it provides interlocutions services, so we measure efficiency in the terms of the whole system but by single factor: time of responses to requests of the user.