Computational methods frequently used in Analysis of Social Media

These are some computational methods discussed in Social Media Analysis 10-802 in Spring 2010.

k-means clustering

a way to cluster items so that there are items within a cluster are close, and distances between clusters are larger. K-means clustering is sometimes used as a subroutine in spectral clustering. See the Class Meeting for 10-802 02/04/2010

Naive Bayes classifier learning

A fast and simple way to learn a classifier, especially useful for text. See the Class Meeting for 10-802 01/14/2010

pointwise mutual information (PMI)

a particular way of measuring correlation between two items, usually words or phrases in text. Comparing the PMI of a word w to positive terms (e.g., "excellent") to the PMI of word w to negative terms (e.g., "poor") is a good indicator of semantic orientation, if a large corpus is used. See the Class Meeting for 10-802 01/14/2010

spectral clustering

a way to cluster nodes in a graph into "blocks" so that there are many connections within blocks and fewer between blocks, based on clustering in a space defined by the eigenvalues of the graph Laplacian. This is a way of solving a graph cut problem. See the Class Meeting for 10-802 02/04/2010

support vector machine classifier learning (SVMs)

A widely-used way to learn a classifier, especially useful for text. See the Class Meeting for 10-802 01/14/2010

Topic model

A widely-used way to learn the underlying semantic structure of a collection of documents. See the Class Meeting for 10-802 02/16/2010

Relational topic model

A specific topic model that allows links between documents to be modeled. See the Class Meeting for 10-802 02/18/2010

Random walk with restart

A technique to find relavance score between any two nodes in a weighted graph

Submodularity

A property of cost functions that explains the utility of greedy methods and speeds up their performance when used on cost functions that have this trait.

Logistic regression

A technique used for prediction of the probability of occurrence of an event by fitting data to a logistic curve.