Twitter Classification

Background

}Goal: Mesh the viewing of RSS Articles with related Twitter Message

◦Use Machine Learning to determine the most significant Tweets based on User’s preferences.

◦Determine the best ML model to use that can handle dynamic content

Process

}Article Processing

◦Clean Article Text

◦Keyword Extraction (Regular Expressions)

–Proper Nouns

–Twitter Hash Tags / User Names

}Keyword Analysis

}Twitter Querying

Multinomial Naive Bayes

}Training / Learning

◦Consider T to be the Set of Tweets used in Training

1.) Extract Keywords from Tweets (K)

2.) For Each T

2a.) Calculate the Prior Based on previous User Feedback

2b.) For Each Extracted Keyword In T

3.) Aggregate Keyword Weights and compute   the Conditional Probability

}Classification

1.Extract the Keywords from the Article

2.For Each Classifying Tweet

2a. Go Through Each Keyword in the Article

3a. Calculate the Conditional Probability

4a. Sum this value for all Keywords in a Tweet

3. Sort Tweets by this Value to get Ranking

Implementation

}MALLET

 Java-based package for statistical natural language processing, document classification, clustering, topic modeling, information extraction, and other machine learning applications to text.

http://mallet.cs.umass.edu/

Machine Learning Final Report.pdf