When you’re the host of a convention, who do you introduce to a person you’ve just met? Are the person’s knowledge and experience the main predictors of a valuable connection? How do you leverage technology to facilitate serendipitous encounters?
These are the questions we are facing. In a small trilogy of blogs I’ll shed light on the process of translating a grand vision into actual physical encounters using artificial intelligence. In the second blog I want to talk about how we created a knowledge ontology to find out which topics are related and how we leverage these insights to make better connections between people.
Hey, what do you know?
When you check in at a Seats2meet location you are asked to indicate which topics you are knowledgeable about and what you are going to work on that day. This is the data we use to connect you to like-minded people. The challenge is there are as many topics one can be knowledgeable about as there are stars in the sky. Chances are, none of your topics occurs in the knowledge clouds of the other people present at the location. But that’s not the end of the story. What if we can automatically learn which themes are related, so we can still match you meaningfully?
From a network of knowledge…
The beautiful blob above this post is a collection of knowledge tags and their related tags. We taught the machine that when two tags are used together, they are probably related. If you check out the lower right corner of the image, you will find a cluster of education related tags. When someone checks in with “moocs” in her scope, we can safely match her to someone interested in e-learning.
Here’s what’s really cool: the ontology is highly dynamic; every connection has a continuously changing strength. When two concepts are mentioned together, the strength is increased. At the end of each day the strength of every connection is decreased. In this manner the system can forget about relations that are no longer relevant. Maybe ‘big data’ was related to ‘artificial intelligence’ last week, but I think the connection between ‘data science’ and ‘artificial intelligence’ is much more relevant today.
…towards a network of people…
Which supermarket is the closest to your home? Finding people close to you in the sense of common knowledge can be regarded as a similar problem. All we do is find the shortest path through the knowledge cloud. Not only does the system discover the strength of the relationship between concepts, it also discovers the strength of the relationship between a person and a concept. Maybe Steve (his number is 159565, can you find him?) says he knows about programming, but in practice all he really wants to talk about is painting. Based on Steve’s behaviour in terms of answering questions, connecting to other people and so on, the system tries to find out exactly how much Steve cares about the topics he says he knows about.
…in a network of physical spaces.
Although the matchmaking is designed to transcend physical space, we do record which knowledge is frequently present at every location. This helps us recommend locations to people who are maximizing the chance of encountering people with a certain skillset and it might help our operators understand the type of environment they’ve created, knowledge-wise.