Associations
- Josh Liss (Unlicensed)
- andrew johnston (Unlicensed)
Overview
The association object is intended to provide a minimal but generic description of various activities and relationships. It can be thought of as "subject verb object at location over time", where the subjects and objects can be free text and/or point to entities within the document. In this representation:
The subject is "entity1" (free form text), or "entity1_index" (the "index" of an entity in the "entities" array of the document or the query)
The object is "entity2"/"entity2_index"
The verb is the "verb" field, with the option of provider a higher level "verb_category" field, which allows grouping of related fields (i.e. "walk", "drive" would both have category "travel")
Times and locations are represented by "time_start", "time_end", "geo_index", and "geotag".
Note: events may have neither a "time_start" nor a "time_end" - in general the document "publishedDate" field can be used (this is what happens automatically in the event timeline aggregation)
Conversely, if an event does not have any geo information, it does not follow that fields such as the document geotag can be used.
Associations are sub-categorized into one of three types: Event, Fact, or Summary.
Event: link multiple entities (via "entity1_index", "entity2_index", "geo_index") and represent a transient activity (i.e. travel)
Fact: link multiple entities like "Events" but represent (transient or permanent) relationships (i.e. being president)
Summary: generally link one entity to a free text (i.e. a quotation: "Obama says...").
Associations As Aggregations
There are a few other minor differences between the association object in as an aggregation vs document child:
"entity1", "entity2", and "verb" fields are not present for aggregations
Only "Event" and "Fact" association types appear in aggregations (no Summaries)
"geotag" fields are not present for aggregations (implementation limitation)
In aggregations, the entity/geo significances may be set to 0 for some entities (implementation limitation)
Event Timeline
There is a third representation of associations in the Event Timeline widget. In this case, associations are the same as those within the documents, except that scoring is different:
The entity significances are not included
The association significance ("assoc_sig") is simply the Pythagorean significance of all documents in which the association is included
Where times are not included intrinsically in the association, they are filled in to span the time range of the documents including them
Association Types
The "Associations" view shows the different "verb categories", ie the top level type of the association between entities.
The following association types are possible on the IKANOW platform.
Association Type | Example | Description | Source |
---|---|---|---|
theme | "President Obama/Person is thematically linked to (theme) Mortgage Advice/Keyword" | A link in the content between keywords and entities such as people/places/companies (eg an important sentence containing both) | Salience |
topic | "Alex Piggott/Person writes about (topic) Software/Topic" | An association from the author of a post to its topics | Salience |
hashtag | "gamerd00d/TwitterUser tweets about (hashtag) guildwars2/Hashtag" | An association from the author a tweet to the hashtags contained therein, eg @gamerdood: "wow #guildwars2 is the best game ever" | Datasift |
mentions (twitter) | "upAllNight/TwitterUser mentions (mentions) gamerd00d/TwitterUser" | An association from one twitter user to another based on a mention, eg @upAllNight: "@gamerdood that was crazy!!!" | Datasift |
mentions (facebook) | "Alex Piggott/FacebookUser mentions (mentions) IKANOW/FacebookUser" | When one Facebook user mentions another in a post. | |
retweets | "gameReviews/TwitterUser retweets gamerd00d/TwitterUser" | An associated from one twitter user to another based on a retweet, eg @gameReviews: "Seems like the punters are enjoing it @gamerdood RT wow #guildwars2 is the best game ever" | Datasift |
likes | "IKANOW/FacebookUser likes (likes) Open Analytics Group/FacebookUser" | An association from a Facebook user who likes a post by another Facebook user. | Datasift |
In this section: