SeanGolliher.com

Machine Learning & Support Vector Machines (SVM) Lecture 9 CSCSI 494

admin — Thu, 08 Mar 2012 01:50:24 +0000

Last lecture we discussed the major topics in machine learning and an important classification algorithm entitled Support Vector Machines (SVM). Modern search engines use a combination of these fundamental techniques to find the relevance of documents w.r.t a given search query.
Papers to read: Standard SVM [Cortes and Vapnik, 1995]

We also discussed a two dimensional problem and showed how to expand a machine learning problem to infinite dimensions.

Lecture Notes

Probabilistic Retrieval Models - Lect. 8 CSCS 494

admin — Thu, 01 Mar 2012 17:00:57 +0000

Lecture 8 covered probabilistic retrieval models and some review of basic probability theory.
Also covered were some of the early retrieval models. Vector space models and cosine similarity.

Slides for Lect 8.

Lecture 7 - Text Statistics & Document Parsing

admin — Thu, 23 Feb 2012 02:36:02 +0000

Today we discussed the fundamentals behind text statistics and how to calculate probabilities of n-grams appearing in a document. Here are the slides for Lect. 7.

Some useful links.
Google’s n-grams data
Text REtreival conference
A corpus of twitter data:Tweets2011

Information Retrieval, Indexing, BigTable Lect 6 CSCI 494

admin — Thu, 16 Feb 2012 02:55:23 +0000

Lecture 6 covered crawling issues, indexing, Google’s BigTable, detecting near duplicate and duplicate content.
Also, the next paper was handed out.
R. Song et al,. “Learning Block Importance Models for Web Pages”

Here are the slides from lecture 6

You Are Worth ~1.18/year Profit to Facebook - Facebook Files an S-1

admin — Thu, 02 Feb 2012 04:32:43 +0000

Facebook filed their S-1 today and there are plenty of interesting numbers in the report. I quickly looked at the numbers to see if there was anything interesting to graph. While looking at these numbers I defined two new metrics. Since Facebook is going public we have, for the first time, data on yearly revenue for a large social network with a display ad model. Until now we haven’t had an accurate value for earnings per user on a free social network. I calculate that Facebook made a profit of approximately $1.18/MAU in 2011. MAU is defined as “Monthly Active Users”. We can define this as PMAU (profit per monthly active user). We can also define RMAU ( revenue per monthly active user). These numbers give us our first estimates on the real value of social data.

This may establish a reasonable “upper bound” on what you are able to make per user using the same monetization model. Unless, of course, you have a better ad network or other applications generating revenue within your network. What advertisers are willing to spend on and ad network has more variables to it than just MAUs.

From an investment standpoint it is interesting to look at these two metrics. I have graphed both RMAU and PMAU below. You can see that RMAU has nearly doubled to $4.50 per MAU (monthly active user) while Profit per MAU (monthly active user) is flat. Profits, as we know, are skewed because of hiring and other expenses. These can be controlled by the company. RMAU is a better indication of how Facebook’s earnings are scaling as it’s user base grows. Judging by the RMAU graph, below, Facebook is scaling up fast in terms of revenue per active user. RMAU has doubled since 2010.