Project done for CISUC by Hugo Redinho and Carolina Gonçalves.
Article: Malheiro, R., Panda, R., Gomes, P., & Paiva, R. P. (2018). Emotionally-Relevant Features for Classification and Regression of Music Lyrics. IEEE Transactions on Affective Computing – TAFFC, 9(2), 240–254. DOI Link
Consists of a feature extractor for song lyrics consisting of:
- Semantic Features
- Structural Features
- Content-Based Features
- Structural-Based Features
Every one of them has sub-features that can be extracted, either all at once or one at a time.
The project can be run in two ways: By running the main class MainInterface.java and then a interactive menu will pop up:
In this window you can choose the type of feature selection you want, and then a sub-menu will pop up, or you can choose to select a folder to extract the files from.
If none is selected, the default folder is Origem.
One example of the sub-menu (for Semantic Features) is:
Another way the project can be running is by using the MainInterface.jar and running the following command:
java -jar MainInterface.jar typeOfExtraction input outputFile
In terms of input (it can either be a file or a folder containing only the files with the right structure, the program will determine which one it is). Note: Not all feature extraction types will extract to a csv, please look at the list below to see which ones do.
The type of extraction can be one of the following:
- features_gi (for GI features)
- Extracts to csv
- features_synesktech (for Synesktech features)
- Extracts to csv
- features_dal_anew (for DAL & ANEW features)
- Extracts to csv & also outputs a .txt with details
- features_gazeteers (for Gazeteers features)
- Extracts to csv
- features_slang (for Slang features)
- Extracts to csv
- features_capitalltetters (for Capital Letters features)
- Extracts to csv
- features_standardPOS (for standard POS tagger features)
- For each input file outputs a .txt with details
- features_cbf* (for content based features)
- Extracts to csv
- features_titulo* (to extract the number of ocorrences of the title in a song)
- Extracts to csv
- semantic_features (Extracts all semantic features onto a compiled csv file)
- stylistic_features (Similar to above, but for stylistic features)
- all_features (Combines the above extractions, also extracting semantic to "Semantic.csv" and stylistic to "Stylistic.csv")
One example would be: java -jar MainInterface.jar "features_gi" "src/Origem" "outputFile.csv"
Note: The features assigned with * have a different command line structure.
For features_CBF since it has more arguments, it has to be run in a different manner:
java -jar MainInterface.jar features_cbf inputDirectory arg1 arg2 arg3
Note: There's no outputFile path you can specify, the program makes it based on the arguments, check file SaveInstancesInFile.java for details.
In regards to arg1, arg2 and arg3, they are parameters for the program, and they can be:
- arg1: unig, big, trig, 4grams, 5grams
- arg2: nada, st, sw, st+sw
- arg3: freq, bool, tfidf, norm
So one example would be: java -jar MainInterface.jar "features_cbf" "src/Origem/" "unig" "nada" "freq"
The name of the outputFile created is the combination of all three arguments, so for the example above the output file would be "unig_nada_freq.csv".
For features_titulo since you have to input the title of the file, the structure is as such:
*java -jar MainInterface.jar features_titulo inputFile outputFile titleSong"
So an example: java -jar MainInterface.jar "features_titulo" "src/Lirica/tw_bl.txt" "tw_bl.txt" "Blinding Lights"