Notable types

From Freebase

Jump to: navigation, search

Contents

Overview

Topics in Freebase can have many different types applied to them. For example, Barack Obama is a Person, Polician, Author, Award Winner, etc. Often times, in its useful to be able to know which one of these types is the the best disambiguator or the most notable type for that topic.

Freebase computes notable types for every topic and uses them in things like Freebase Suggest to help disambiguate topics with the same name. In this case, Barack Obama is most notable for being a US President.

Notable types are associated with topic in Freebase via the /common/topic/notable_types property. The notable types are not stored explicitly in the graph as individual links but they are available in the new Topic API and data dumps as /common/topic/notable_types.

Algorithm

The way a topic's notable types is calculated is rather complex.

First, it prefers Commons types and only uses user types if no Commons types were available. Then it does graph analysis at both the type and the topic level -- including factors such as: what properties are linked to the topic, how many instances does the type have, what is the group membership we have defined for the type, etc. -- to arrive at a score for each type that the topic has.

A few words on type groups mentioned above. All Freebase Commons types have been manually curated to fall into groups depending on their annotative role and how they're used. For example, we say that "City/Town/Village" is a definition type (What is Boston? It's a city). We say that "Award Winner" is a description type (winning an award does not fundamentally change a topic -- if it's a city before the award, it does not become a country after the award). This information is used to help rank and score a topic's notable types.

For topics that are people, we also scan the Wikipedia article for a list of known professions, to avoid obvious mistakes caused by holes in our data, like saying someone is most notable as a "Deceased Person" or "Building Occupant". Depending on what data we have, we may defer to WP to determine what a person's most notable for.

As a last resort, the notable types algorithm also supports manual override to correct blatant and persistent mis-categorization on a topic. If you notice an inappropriate notable_for value, feel free to file a bug.

Search filters

You can filter the Search API results to only include topics with a given notable type like this:

 filter=(all notable:Politician)

Try it out

Data dumps

If you'd like to download a list of every topic in Freebase and its notable types. You can get that data from our TSV topic dumps for the /common/topic type.

Download here: http://download.freebase.com/datadumps/latest/browse/common/topic.tsv

The notable types are given as a comma separated list of types in the 11th column of the spreadsheet. The types are ordered from least notable to most notable.

Personal tools