At the beginning we were thinking about making something a little different and interesting. Then we came up with the idea of Joke Generator, which we believe is quite unconventional.
What it does
It reads in textual data of jokes (preferably of certain underlying type) and generates new jokes similar to those as user requests via a window GUI.
How we built it
We first parse the input data (tokenization and cleaning). Then we run basic statistical analysis, and use linear interpolation of a trigram and bigram model for the word generation. We used Java Swing framework for the GUI.
Challenges we ran into
Raw input file parsing; coming up with good algorithms for text generation; Storing large data in memory efficiently (we also tried to use SQL database).
Accomplishments that we're proud of
The outcome is quite satisfying. The generated jokes are surprisingly following the English grammar, and are very funny at times.
What we learned
NLP. SQL. JSON. SWING.
What's next for Joke Generator
A more sophisticated model; we are currently thinking about RNN.