2015-12-09

Search is not only for entities

The Knowledge Graph is a great achievement, but its systematic use at the top of search results is sometimes counter-productive. Knowledge Graph nodes are mostly named entities (individuals, particulars) such as people, places, works (movies, books, music tracks), products ... and rarely universals (concepts, topics, common names). And if an ambiguous search sentence can refer to either particular entities or universals, the former seem to always float at the top with their fancy Knowledge Graph display, and relevant results about universals kicked down. The assumption underlying this default behavior is that people search mostly for particular entities (things), not information about some universal (topic). The hijacking of common names as brand names we already pointed here in the past adds to the issue, along with the growing number of work titles using common names. Add to this the magic of the Knowledge Graph knowing entities by various names in different languages, and you end up with examples like the following. 

For a recent post I searched about the Theory of Everything. If instead of going straight to the Wikipedia article I ask Google, here is what I get.


I was searching for information about a theory in physics, and I get all about a movie which happens to have taken as title the name of this theory. And since my browser default language is French, the Knowledge Graph is kind enough to present me the movie under its French adaptation title "Une merveilleuse histoire du temps", which you can imagine even if you don't speak a lot of French, is all but a translation of "Theory of Everything". The silver lining is that if I search for "Théorie du tout" in French, I have not the same problem, since the movie is not known in French under this title which would be the correct translation of the original one. The first result for "Théorie du tout" is the Wikipedia article on this topic, as expected.
You can play the same funny game with "Gravity", "Frenzy" and many more. Given the limited supply of common names, and the exponential growth of named entities in the Knowledge Graph, all tapping into the commons for their names and titles, such ambiguities are likely to end up being rather the rule than exceptions. Search engines should provide a simple way to opt out entities, so that I could ask "Dear Google, give me resources about the topic called gravity, and I don't care about any individual entity with gravity in its name." And yes, Google, you can do it, I'm sure, just take example on BabelNet, where you can sort results by entities, concepts, music, media etc.  A bit of typing goes a long way ...