Inspiration
As both crossword lovers and computer nerds, we jumped at the opportunity to do something that involved our interests as well as providing an educational tool to students and teachers.
What it does
Given a user submitted topic, Cyber-Crossology generates a crossword puzzle with answers related to that topic. This crossword comes complete with hints and a layout.
How we built it
Cyber-Crossology was made using Python. To determine answer-topic similarity, Universal Sentence Encoder was used. Clues and answers are from a dataset containing 50+ years worth of crossword answers and hints from major US newspapers.
Challenges we ran into
Evaluating crossword semantic similarity was difficult, as crosswords are often designed to be a little indirect in the relation between hint and clue. Cleaning the large datasets for crossword hint and clues also took quite a while
Accomplishments that we're proud of
We have a working project! First time building a web app <3 First time working with ML and made a working product <3
What we learned
Several methods to obtain and clean large datasets, how to use natural language processing packages, how to work with Tensorflow models
What's next for Cyber-Crossology
Add support for phrase/sentence length crossword answers, improve the front end to support users editing the ML generated crossword
Built With
- justpy
- machine-learning
- natural-language-processing
- pandas
- python
- tensorflow
Log in or sign up for Devpost to join the conversation.