Graphics Reference
In-Depth Information
There are different ways of computing product recommendations, such as
machine learning, collaborative filtering, and market basket analysis. A
market basket analysis
is a graph created by connecting all the items
purchased in a single transaction (that is, all the items in that market basket
are linked together).
In the case of an e-mail, a market basket is all the people involved in each
e-mail. An example of e-mail data may look like this:
To, From, CC, Date, Size
"Joe", "Zoe", "Tim", 12/09/2014, 156kb
"Joe", "Ben", "Ann, Tim, Zoe", 11/09/2014, 2048kb
"Joe", "Tim", "Ben, Zoe", 11/09/2014, 805kb
...
In this example, each row is a single e-mail, and all the people in the From,
To, Cc, and Bcc fields form a set of links. Chapter 8 provides an example
programming script to turn email data into a graph data set. See the Python
EmaildataintheSupplementaryMaterialonthisbook'scompanionwebsite
for an example of raw, anonymous e-mail data similar to what was
previously transformed into node and link graph data and then visualized.
Sequence Data (for Example, Customer Paths, Patent Citations)
Sequence data is very similar to transaction data.
In a web server log, each row indicates a particular web page served to a
particular user. Within each row, a client IP address and possibly a session
identifier is used to indicate a particular session for a user accessing the
website. By collecting all the records corresponding to a particular session,
all the web pages for that session are identified, in the sequence that the
user went through the website. That is, this is a set of nodes (web pages)
and the associated links (sequence) that the user traveled through the site.
By combining multiple paths, you might see if people take common routes
through a website:
Time Client_Address Requested_File Status
09:55:15 12.34.56.78 GET /index.html 200
09:57:35 12.34.56.78 GET /images/logo.gif 200
09:58:22 12.34.56.78 GET /flash/splash.swf 200
