Problems frequently addressed in Analysis of Social Media

These are some technical problems discussed in Social Media Analysis 10-802 in Spring 2010.

fitting parameters of a random graph model

Since random graph models have parameters (e.g., the mean degree parameter for an Erdos-Renyi graph), this leads to the problem of finding the best values of these parameters, given a particular graph. Depending on the complexity of the graph model this can be simple or complex. See the Class Meeting for 10-802 01/28/2010

identifying the semantic orientation of words

In sentiment analysis, the problem of automatically classifying words or phrases (often, just adjectives) as typically "thumbs up" or "thumbs down" (of course sometimes the actual orientation is context-dependent). See the Class Meeting for 10-802 01/14/2010

mincut problem for a graph

The problem of finding a small set of edges to delete from a graph in order to "best" break it into 2 (or sometimes more) components. There are a number of variants of this problem, which make slightly different assumptions about the kind of links in the graph and/or different choices about how to measure the "best" cut. See the Class Meeting for 10-802 02/04/2010

normalized cut or NCut problem

A variant of the graph cut problem. See the Class Meeting for 10-802 02/04/2010

opinion question-answer

In sentiment analysis, the problem of automatically answering questions about opinions (private states) based on analysis of text. See the Class Meeting for 10-802 01/14/2010

product review mining

In sentiment analysis, the problem of automatically preprocessing product reviews to facilitate shopping. See the Class Meeting for 10-802 01/14/2010

ratio cut problem

A variant of the graph cut problem. See the Class Meeting for 10-802 02/04/2010

review classification

In sentiment analysis, the problem of automatically classifying product reviews, usually as "thumbs up" or "thumbs down". See theClass Meeting for 10-802 01/14/2010

tag recommendation

Given a resource (photo, link, etc.) and tags assigned to this resource, recommend tags to enrich the annotation.

word sense disambiguation

Given a word's appearance in context, the problem of automatically determining which sense of a word is intended. See theClass Meeting for 10-802 01/14/2010 for a discussion of this problem as it relates to finding the semantic orientation of words.

neighbourhood formation(NF) and anomaly detection(AD)

NF refers to finding out close neighbours of a node in large social graph. AD refers to finding the nodes which connects nodes in different neighborhoods. See the Neighborhood formation and Anomaly detection for discussion of the methods to solve these problems on bipartite graphs.

Correlation between various network attributes

This refers to finding out which attributes are related to which other attributes and how strong the dependence is. A widely used method for this task is finding out Pearson correlation coefficient.

Community Detection

In network we often find nodes whose mutual interaction is stronger with some nodes than the rest of the network. To find groups of tightly interactive nodes is commonly known as community detection.