The document discusses an app called StayConnected that measures friend "connectedness" using interactions on social media. It analyzes likes, comments, tags and messages to determine meaningful interactions between a user and their friends. Using principal component analysis, it identifies closest friends based on common interests and comments. A graph is shown plotting meaningful interactions against total interactions to visualize connectedness - a positive slope means friends communicate more with the user, while a negative slope means the user communicates more with friends. The end provides brief biographical information about Megan Cartwright, the creator of StayConnected.
5. How do you re-connect?
Go to: See the friends
Login with
stayconnected- you can
Facebook
app.com reconnect with!
6. How do you measure connectedness?
Likes on Number
Number of
your of
Comments on tags in Comments
photos messages
your photos photos and on friends’
statuses status
Likes on your Likes on friend’s
status status
Comments on Comments on
your status friend’s photos
PCA Likes on
Common likes
friend's
photos
7. PCA determines the most meaningful interactions!
Likes on Number
Number of
your of
Comments on tags in Comments
photos messages
your photos photos and on friends’
statuses status
Likes on your Likes on friend’s
status status
Comments on Comments on
your status friend’s photos
PCA Likes on
Common likes
friend's
photos
How do you re-connect? Which friends are the friends you haven’t talked to lately, that are still important to you?
My friends mean a lot to me as I’m sure your friends mean a lot to you.We enjoy spending time with our friends, participating in each others lives, But life often gets in the way and despite best intentions we lose track of our friends.Social networks do a great job of making us feel connected, we can see what's going on in people’s livesBut how connected are we really?
My friends mean a lot to me as I’m sure your friends mean a lot to you.We enjoy spending time with our friends, participating in each others lives, But life often gets in the way and despite best intentions we lose track of our friends.Social networks do a great job of making us feel connected, we can see what's going on in people’s livesBut how connected are we really?
My friends mean a lot to me as I’m sure your friends mean a lot to you.We enjoy spending time with our friends, participating in each others lives, But life often gets in the way and despite best intentions we lose track of our friends.Social networks do a great job of making us feel connected, we can see what's going on in people’s livesBut how connected are we really?
How do you re-connect? Which friends are the friends you haven’t talked to lately, that are still important to you?
The app will measure how connected you are to your friends – but how does it do that?There are so many ways we interact with our friends on Facebook…what are the important features?Using a procedure called principal component analysis we can determine which features are most important.This PCA analysis uses an orthogonal transformation that converts a set of potentially correlated variables into a set of linearly uncorrelated variables called principal components, where the first principal component is the largest variance and so far down to the smallest variance component. This is based on a normal distribution, which is not necessarily the case here – but it is a convenient way to reduce the dimensionality of our features space (where we are finding a direction onto which to project the data which minimizes the error).
This application found that the most important features (by an order of magnitude) are commenting and liking our statuses. The interesting thing about this is we can use that information to find how much we communicated with our friends and vice versa. (next plot)Describe how these red boxes are the meaningful interactions
Using the most important features:I normalized them by Subtracting the number of your likes from the number of your friend likes for each friendNormalized this by using a sigmoid function (1 / 1 +exp(-constant*subtraction))Summed the two normalized features (# of likes and # of comments) = Ratio of CommunicationFound the k-means clusteringCluster at the top of my closest friends – zoom in that Cluster at the bottom of my least close friends – zoom in on thatBut these are the friends I most care about the ones in the middle
Using the most important features:I normalized them by Subtracting the number of your likes from the number of your friend likes for each friendNormalized this by using a sigmoid function (1 / 1 +exp(-constant*subtraction))Summed the two normalized features (# of likes and # of comments) = Ratio of CommunicationFound the k-means clusteringCluster at the top of my closest friends – zoom in that Cluster at the bottom of my least close friends – zoom in on thatBut these are the friends I most care about the ones in the middle
Using the most important features:I normalized them by Subtracting the number of your likes from the number of your friend likes for each friendNormalized this by using a sigmoid function (1 / 1 +exp(-constant*subtraction))Summed the two normalized features (# of likes and # of comments) = Ratio of CommunicationFound the k-means clusteringCluster at the top of my closest friends – zoom in that Cluster at the bottom of my least close friends – zoom in on thatBut these are the friends I most care about the ones in the middle
Using the most important features:I normalized them by Subtracting the number of your likes from the number of your friend likes for each friendNormalized this by using a sigmoid function (1 / 1 +exp(-constant*subtraction))Summed the two normalized features (# of likes and # of comments) = Ratio of Communication
Using the most important features:I normalized them by Subtracting the number of your likes from the number of your friend likes for each friendNormalized this by using a sigmoid function (1 / 1 +exp(-constant*subtraction))Summed the two normalized features (# of likes and # of comments) = Ratio of Communication
Using the most important features:I normalized them by Subtracting the number of your likes from the number of your friend likes for each friendNormalized this by using a sigmoid function (1 / 1 +exp(-constant*subtraction))Summed the two normalized features (# of likes and # of comments) = Ratio of CommunicationFound the k-means clusteringCluster at the top of my closest friends – zoom in that Cluster at the bottom of my least close friends – zoom in on thatBut these are the friends I most care about the ones in the middle