Project 2nd draft Derry Reyyan

Social Media Analysis Project Ideas

Team Members

Derry Wijaya [dwijaya@cs.cmu.edu]

Reyyan Yeniterzi [reyyan@cs.cmu.edu]

Project Idea

Understanding change

Given an entity of interest, we would like to model and analyze its change in terms of words and phrases that co-occur with it over time. In other words, we would like to understand the change in Social Network Attribute over time, where the social network is defined over words.

We propose to construct a social network, but instead of people, we put words as nodes and edges are weighted based on the number of co-occurrences between the words. Using this social network of words, we propose to analyze:

(1) how co-occurrence with other words change over time

(2) how the change influences the state (semantic or sentiment) associated with the entity

(3) how the change may correspond to events that occur during the same period of time

For example, the entity 'BP' frequently co-occurred with negatively associated words during and after the Gulf-spill event.

Dataset

Google Books Ngram Data.

Motivation

The co-occurrence of words changes over time

It will be interesting to model this change and analyze:

(1) How the state (semantic or sentiment) of a given entity changes over time depending on its neighbors (i.e. co-occurring words/phrases)

(2) How such changes relate to events that occur in the same period of time

(3) Whether we can find a natural sequence of events that define a change of state (semantic or sentiment) of a given entity

(4) Whether we can use (3) to predict the change of state of a given entity

Techniques and Related Works

(1) Linear regression analysis to measure tendency of a word to become negative/positive in meaning over time, when co-occurred with negative/positive words

Related paper: Nicholas A. Christakis, M.D., Ph.D., M.P.H., and James H. Fowler, Ph.D. (2007) The Spread of Obesity in a Large Social Network over 32 Years: External Link - techniques to analyze the spread of obesity in a network of people

(2) Identify break points in the states of an entity (based on its co-occurrence changes) and find events that correspond to the break points

Related paper: Akcora et.al. (2010) Identifying Breakpoints in Public Opinion: External Link - technique to identify break points in sentiments found in tweets (Twitter), using a set of manually constructed emotion words (Vector space models)

Related paper: Michel et.al. (2010) Quantitative Analysis of Culture Using Millions of Digitized Books: External Link - measure usage frequency over time of a given n-gram (such as "slavery", "great war", etc) that represents an entity of interest

(3) Graph modeling of co-occurrences (where nodes are words and edges are weighted by the number of co-occurrences between words). Use of techniques from dynamic network evolution, link analysis or clustering to model and analyze changes in this graph over time

Related paper: Cemal Cagatay Bilgin and Bülent Yener (2010) Dynamic Network Evolution: Models, Clustering, Anomaly Detection: External Link

(4) For baseline, we plan to use Bayes' Law to measure probability that a word will co-occur with an entity, given the entity and other words that have co-occurred with the entity:

$p(word_{i}\vert entity,word_{1},\dots ,word_{i-1})={\frac {p(word_{i})\ p(entity,word_{1},\dots ,word_{i-1}\vert word_{i})}{p(entity,word_{1},\dots ,word_{i-1})}}.\,$

Using Naive Bayes conditional independence assumption,

$p(word_{i}\vert entity,word_{1},\dots ,word_{i-1})={\frac {1}{Z}}p(word_{i})p(entity|word_{i})\prod _{j=1}^{i-1}p(word_{j}\vert word_{i})$

where $Z$ (the evidence) is the normalization factor. We then need to model how this probability changes over time.

Evaluation

Our project will be mainly quantitative analysis in nature.