Generates quotes. Sometimes nonsenese, sometimes gold
https://presidential-quote-gen.herokuapp.com
Install Python & Flask. Run flask, then browse the local address in your browser
Export flask_env=development; flask runPresidential quotes were pulled from Kaggle, Parade, BrainyQuote, and YourTango. Each of these pages had custom regex applied to remove Presidential indentifying information, webpage garbage, duplicates, and advertisements.
The code uses a 2nd order Markov chain to create a dictionary of words that naturally occur based on the quotes sampled. When a quote is generated, it randomly selects from any of our starter words, then selects randomly from the probable words that follow. While it's possible to select words that occur more frequently, this code selects randomly.
In this example we start with the words "Don't worry". The next step uses our 2nd word 'worry' and looks at all of the pairs that begin with worry. In this case, "worry over" and "worry when". Our random function selected the prior. The following two words didn't have any additional options to select from. This continues to the end, selecting randomly where there were multiple options available until "Congress. STOP". The "STOP" word triggers the end of our quote.
Replacing, adding, editing the quotes used in our Markov Dictionary are specified near the start of our app.py file.
There is the built in option to color code words based on their source text. It is not used in the online demo
Simply specify after the file name, one of the valid color choices as shown above. Our colors are specified (and limited) by the BootStrap theme.
The Markov chain is an introduction into generating text. As seen, often large blocks of text have no other options, and the original quote is reproduced. This is because we do not have any other quotes with similar word patterns.
While we could throw thousands of quotes into our dictionary, it has the drawback of generating text that doesn't make sense, or end up stuck in a circle of repeating text.
Raising the Markov chain order from two to three (or more) will make our text read more naturally but repeats the first problem where we end up creating no significant change to our original text.
- Machine Learning using an LSTM for smarter text generation
- NLTK (Natural Language Tool Kit) which allows for parts of speech tagging
- The ever developing "Natural Language Processing" models that derive from BERT
- The huge pre-trained models Megatron, or GPT-3
Presidential image used under CreativeCommons - NonCommercial https://griffonagedotcom.wordpress.com/2016/07/04/averaged-portraits-of-u-s-presidents-1789-1829/