song_tagger is a project which uses multiple naive-Bayesian classifiers (via nltk.NaiveBayesClassifier) to tag songs based off of profanity, drug references, sexual references, and violent language.
To run the code, run the main function. Input 1 runs the training function; input 2 runs the console_interactive function.
For training, the output comes in two major parts: the first part, which prints the most important features for each classifier, and the second part, demarcated with the header "Test Results", which tests several songs and prints the results in the format:
# song_name by artist_name tags: {'profanity': w, 'drugs': x, 'violence': y, 'sexual references': z}For console_interactive, the output simply asks for a title and artist, then prints the results in the same format mentioned above.
Container class to generate, handle, and process information relating to a song.
- title
- author
- all_feature_dicts
- final_tags
Constructor for the Song class. Takes in a title, author, and an optional lyrics parameter.
example_song = Song("Fireflies", "Owl City")The constructor uses the title and author to call the Genius API to search for a song, and obtains the first search result's metadata. It then retrieves the html for said first search result and uses bs4 to parse the lyrics found on the page. After retrieving all of the information about the song, it calls self.generate_all_features() to generate all four feature dictionaries for the song.
Takes in a tag and generates a feature dictionary based on that tag. Valid tags include "profanity", "drugs", "violence", and "sexual references". After the feature dictionary is generated, it is assigned to the Song member variable self.all_feature_dicts as a dictionary entry.
example_song.generate_feature_specific("profanity")Defines a list of tags as ['profanity', 'drugs', 'violence', 'sexual references'], then iterates through each tag, calling self.generate_feature_specific(tag) on each tag.
Sets the value of self.final_tags[tag_type] to severity, which should either be "none", "some", or "frequent". This function is needed when pre-assigning values to songs in order to use them as training data for the classifiers.
A container class for a Naive-Bayesian classifier specific towards one tag.
- target_tag: str
- tagged_songs: list
- training_set
- classifier
Assigns tag to self.target_tag and intensity_tags to self.tagged_songs; generates a training set based on self.tagged_songs, and trains an ntlk.NaiveBayesClassifier, assigning the trained model to self.classifier.
intensity_tags is intended to be passed in as a list of (Song, intensity: str) with respect to the current target tag.
tagged_songs = [(profane_song, 'frequent'), (clean_song, 'none')]
profanity_classifier = TagClassifier('profanity', tagged_songs)Uses self.classifier to classify curr_song's intensity with respect to the current tag by getting its feature dictionary for the tag.
Returns the predicted intensity.
r = profanity_classifer.predict(disney_song)
print(r) # 'none'Calls curr_song.assign(self.target_tag, self.predict(curr_song)), thus assigning curr_song a value for the tag in its final_tags.
profanity_classifer.assign(profane_song)
print(profane_song.final_tags['profanity']) # 'frequent'Returns the target tag for the TagClassifier.
print(profanity_classifier.tag_type()) # 'profanity'Uses a list of preset songs as training data and returns a list of four trained TagClassifiers for each corresponding category.
Calls demo_trained_classifiers() to obtain the classifiers, then tests several songs with the trained classifiers to predict their tags.
Calls demo_trained_classifiers() to obtain the classifiers, then allows users to manually request songs with title and artist as inputs. These songs are then tested by the trained classifiers for tag prediction.