#Lyrics Predictor

This is the final project for an Artificial Intelligence class, in which we implemented a word predictor for a corpus of Rock and Pop song lyrics. Our algorithm is a weighted combination of an N-gram model with discounting and back-off, and an N-gram for tags.

Our results and analysis can be found in the project presentation or the report.

To run the project use python 3.4

The following dependencies are needed:

• nltk 3
• pandas
• bokeh

in corpus/raw/ and corpus/lyric_corpus/

1. run cl_client.py using a songlist file to crawl lyrics
2. run corpus_builder.py to clean the raw data
3. manually remove empty files, non english lyrics, etc.
4. run category.py to generate a category file

in analysis/

• run basic_statistics.py for basic meassurements on corpus
• run collocation.py for bigram and trigram collocations in POP and ROCK

in analysis/

• run linearCombination.py
• run perplexity.py
• run predictWord.py
• run testSimpleNgram.py
• run testSmoothing.py
• run tryAlpha.py

in nGram/ the following models and taggers can be found:

• nGramModel.py
• NgramTagModel.py
• trainTagger.py