Kreiswolke/JokesNet

JokesNet

1) First task: Building a database of oneliner jokes.

1.1) onelinefun.com -> Completed {~2700 items}

1.2) funn2.com -> Completed {~300 items}

1.3) jokesoftheday.net -> Completed, but site's html code is very inconsistent and further exceptions might have to be added {~7000 items}

2) Train a network to produce funny jokes

2.1) Train an LSTM model on the oneliner jokes corpus: -> Done using https://github.com/sherjilozair/char-rnn-tensorflow implementation, Results are rather mixed so far:

2.1.1) Problems: - Generated words are readable but misspellings still likely

- The content is often i) confusing ii) not funny iii) lacking consistent oneliner properties(e.g. 'Q: What is a ...?' But 'A: ...' is missing)

-> Corpus too small and generally basic word-relations are not covered (which results in a seemingly random concatenation of words)

Idea: Use a progressive network approach: First train network on basic text data (e.g. wikipedia) and then add a network extension trained on the oneliner jokes task