Team Name: Data Don                                                                                                                     

Team: V. Dinesh Kumar

Institution: SSN School of Advanced Career Education

Phone: 9962255456

Topic: Next USA President

DPL - Methodology Document

Step 1: Exploring the data sets

In our Scenario the significant data is PRE DEBATE and POST DEBATE of USA election.

·         Pre Debate and Post Debate data are filtered according the sources, for instance Twitter, FB, Forum, Blogs spot, Mainstream and other sources respectively.

·          Each sources in the data set handled independently to filter using “Text Filter” in “excel” tool.

·         I used keywords like “Donald Trump, trump, donald” and “Hillary Clinton, hillary, clinton”.

·         One of the option I used for text filter contain “Donald Trump” does not contain “Hillary Clinton” and other combination were carried out to filter the sentiment of Trump and Hillary separately.

·         Consolidated all the sources for Trump and Hillary respectively.

·         Previous step resulted in giving Trump and Hillary’s pre and post-debate sentiments in separate files.

Step 2: Loading the data sets in R Studio

·         Data pulled into R script.

·         Identified how many languages are in the data set by using “Textcat” package.

In total 30 languages are present in the given data set.

·         Introduced “Natural Language Processing” algorithm” to mark score each sentences.

·         Scoring is applied to all the sentences and explored the distribution of the sentiments.

·         Summary of the distribution is checked for both Trump and Hillary in the both events of pre and post-debate.

·         In both event before debate and post-debate Trump has sound sentiments among the data given comparatively than Hillary.

Step 3: Summary of data for Hillary and Trump

·      Hillary pre debate sentiment score range lies between -8 to 6 and median score is 0 which indicates the comments for her is “Neutral”.

·      Trump pre debate sentiment score range lies between -7 to 11 whereas the Trumps score have more spread than Hillary both in negative and positive as well, and his median score is 1 which is most of his sentiments from the people are “Positive”.

·      Hillary’s score after the debate has actually reduced in positive term -8 to 5 and her median score still continues to be 0.

·      Trump’s score has increased in negative score -10 but the positive score has boosted up to 21 and median value continues to 1, where these sign supports the Trump strongly in USA election.

Step 4: Code / Output and Model Validation

file:///C:/Users/ssn/AppData/Local/Temp/Codes_and_Outputs%20(2).html

·         The Model accuracy is 54% (500 sentiment samples were considered based on Stratified Random Sampling and scored manually and crossed checked against the predicted score).

Step 5: Tools

·         Ms Excel - Data Classification.

·         R Studio - Data Cleaning, Sentiments and words for word cloud picked for Trump.

·         Adobe Photoshop - Word cloud generated and merged on Trump image.

·         Tableau- Created charts for report.(https://public.tableau.com/profile/dinesh.kumar7575#!/)

·         Piktochart – Info graphic Visualization report.    

PS: The results are double checked using the artificial intelligence tool provided by IBM (Watson Analytics). The results are coincidental with my analysis.