**Context** - These datasets were produced as part of a little research project I undertook for a blog post on sentiment analysis, which you can access here: https://bit.ly/32PmWdf - I uploaded a more extensive dataset of presidential speeches on Kaggle here: https://bit.ly/2E7Fmvw **Content** The datasets were created to compare sentiment scores across two text types (tweets versus political speeches) and three sentiment models (Pattern, Vader and the polarity model incorporated in Stanford CoreNLP). ***- sentiment_speeches_Kaggle.csv: *** The dataset contains sentiment codings (Pattern,Vader & Stanford model) for al inauguration and state of the union speeches of US presidents since 1917 on sentence level. **-sentiment_tweets°Kaggle.csv:** The dataset contains sentiment codings (Pattern,Vader & Stanford model) for a sample of around 11500 tweets of US politicians (Donald J. Trump (Rep.), Rand Paul (Rep.),Ted Cruz (Rep.), Alexandria Ocasio-Cortez (Dem.), Nancy Pelosi (Dem.) and Bernie Sanders (Dem.)). The sentiment polarity has been computed on tweet-level, not sentence-level. I simply scraped the last 2000-ish tweets of each timeline using the GetOldTweets module for Python. You can read more about the data collection process in the aforementioned blog post (https://bit.ly/32PmWdf). **Distribution of sentiment scores**  ** Some example analyses ** - The fact that the analyses of presidential speeches has been performed on sentence-level makes an analysis of polarity-development within the same text extremely easy. For example, one can easily plot the polarity scores across a single inaugural speech, like I did here for the inaugural addressees since JFK:  - One could also compare how these models code a particular tweet or speech and index the amount of (dis)agreement betwene the different models:  You can read more about the data collection, wrangling and analysis process in the aforementioned blog post.