Computers are starting to understand our world more and more, as dozens of artificial intelligence projects have already proven. But they are still incapable of connecting the dots. The reason is machines’ lack of background knowledge we have had years and millions of experiences to accumulate. Their interpretation is wrong in most cases, because they lack the ability to grasp multiple meanings. Microsoft is trying to fix that with Concept Graph #softwaremagic
Concept Graph is a public tool based on Probase, Microsoft’s knowledge database. This database has 5.4 million concepts registered, harnessed from millions of web pages and search logs. The idea is to mix and match these concepts so they result in different interpretations, each of those with a certain degree of probability.
This analysis is the key to the Concept Tagging Model, a way of mapping text into semantic categories. The model takes into account both extremes – the very probable relationships and the highly unlikely ones.
For now, you can depend on the Graph to rank the relevance for every text entry. “This can benefit various text processing applications including search engines, automatic question-answering, online advertising, recommendation systems and artificial intelligence system”, wrote Microsoft. In the future, the company hopes to release “single instant conceptualization with context”.