Hi Everyone… I’ve written a quick and short article on AI Entity Searches following a couple of recent discussions with the aim of helping you understand how it works and ultimately appreciate the incredible amounts of time it can save and value it can deliver because it goes further than we can ourselves.
What is Entity Search?
AI entity search takes into consideration the definition of the terms that you are searching for.
Say for example that you wanted to search for the brand Apple. A manual keyword search only containing the word "apple" would also return mentions of fruits and any other apple based product names or brand names, so you would need to add additional context to your search in order to get to the relevant content (e.g. Apple AND (iphone OR ipad… etc)). However, using AI-powered entity search, you pick the concept that you are searching for — in this case the brand Apple, not the fruit — and the system will return mentions of the entity that you have selected.
- Noun extraction. First, we run all mentions in our data library through noun-phrase extraction, which pulls out all of the important noun phrases from a mention. In the example below, we find iphone, apple, logo, and a few others.
- Entity candidate generation. Secondly, we generate a list of possible candidate definitions for each identified entity. For example, Apple the company and apple the fruit are both entity candidates for the word "apple."
- Entity disambiguation. In the final and most comprehensive step, we work out which entity candidate is the one being used in the given mention.
For more information and a detailed breakdown of how it works, read our Help Center article.
How does it work?
Brandwatch Consumer Research pulls in Entities from https://www.wikidata.org/ Specifically, the “Q” Number (aka Entity ID) from the URL / Title of the page
Some associated examples:
iPad: 2796
iPhone: 2766
iMac: 14091
Your query might look like this, note Apple is an Entity ID found through Brandwatch.
Apple AND (entityId:2796 OR entityId:2766 OR entityId:14091)
Brandwatch should pull in all entities, should it not - now you know how you can do this yourself.
Flip it!
Now, consider how you can take what yo’ve learned to improve your Exclusions as well as improving the accuracy of the data you do want to bring in.
Eg, NOT: Apple Records
Let me know how you get on and if you have any tips to share (just leave them in the comments below and I’ll throw some points or a badge your way!)
