ashokpant / u.s-presidential-speeches

Textual Analysis of speeches using Google's Word2Vec Model

U.S-Presidential-Speeches

This is an attempt to analyse text of U.S. President's Speeches from 1790 to 2006 using Word2Vec model in gensim.

• Data in enclosed here as speech.txt

• all_speech.txt is the processed version of the speeches such that each line contains one processed speeches

• data_processed.txt is the text file of json.dumps() containing full metadata from speech.txt in the form of list of dictionaries - 'who' : President's name - 'date' : date of speech (example : January 27, 1984) - 'speech' : Full speech - 'what' :'State of the Union Address'

• speech.py contains code to process speech.txt

• w2v_speech.py contains gensim model to learn w2v model of the from speeches. Here I have assumed one speech to constitute one sentence. Speech vector is then calculated by averaging all the word vectors in the speech

• w2v_tsne.py contains the code to plot 2D version of 100 dimensional speech vectors

• speech_vectors.npy is numpy vector of all speeches as processed in w2v_tsne.py

• .png files are respective tsne plots of speech vectors

Here is the t-SNE plot of speech vectors: (labeled version is in the repo. You can download the image to use zoom functionality)

Here is the distance matrix between speech. One might need to zoom in order to see year of speech vectors. Effective visualisation tool for zooming is the next project that I am working on.