After reading an interesting tutorial on Kaggle about Sentiments Analysis applied to American Political Speeches, I decided to create my own version.
Instead of using R, I used Python and I applied my model to French Political Speeches.
I used the lexicons available here, which are from the following paper : Chen, Y., & Skiena, S. (2014). Building Sentiment Lexicons for All Major Languages. In ACL (2) (pp. 383-389).
I used the speeches available on this page.
I used a Class Speech which contains the following:
- name of the speaker (Here, the name of the President)
- year of the speech
- number of positive words in the speech
- number of negative words in the speech
- number of words in the speech
- a ratio, (number of positive words - number of negative words)/(number of words).
I created a function getLexicons(positiveList, negativeList) which put all the positive words from the positive lexicon into a list, and put all the negative words from the negative lexicon into another list.
I then created another function getSpeechAnalysis(speechName, name, year, positiveList, negativeList) which return a speech class with all the information we need (number of words, number of positive/negative words...).
Then, after initializing the environment with the first function, I applied this last function to a list of speeches (the end-of-the-year speech given by the Frencg President).
I obtained the following results:
-
On this graphe, we can see the positivness ratio evolving among time. We can see where the different Presidents position themselves compared to the moving average.
-
On this graphe, I draw a boxplot for every President. Thus, we can see the median, the first and third quartile, the maximum and the minimum for every President.
