Development of the lyrics-based deep learning algorithm for identifying alcohol-related words (LYDIA)

Citation: Abraham Albert Bonela, Zhen He, Dan-Anderson Luxford, Benjamin Riordan, Emmanuel Kuntsche, Development of the lyrics-based deep learning algorithm for identifying alcohol-related words (LYDIA), Alcohol and Alcoholism, Volume 59, Issue 2, March 2024, agad088, https://doi.org/10.1093/alcalc/agad088

Lyrics Dataset

This dataset has the following structure that includes essential files and folders required to train a classification model.

Structure:

1. Lyrics Data
2. MergedPreprocessedAnnotationData.csv
3. FilenamesAndIndices.csv
4. KeywordsFinal3.1.xlsx

Lyrics Data: This folder contains all the song lyrics of top-100 songs across the years 1959 to 2020. The song lyrics are grouped by year. Each year may have at least 100 songs (in some years there are fewer number of song lyrics) and at most slightly greater 100 (indicating ties about the ranks of songs). Lyrics for each song is stored in two file formats: json and txt. You may use the .json format files as they are updated with correct lyrics.

MergedPreprocessedAnnotationData.csv: This file contains all the annotations for each occurence of alcohol keywords in this above lyrics corpus.

This file contains 5 columns:

1. fileIndex
2. lineIndex
3. word
4. wordLabel
5. wordContext

fileIndex: Represents the index of the lyrics file corresponding to the corpus.

lineIndex: Represents the line number of the lyrics file in which there is an occurrence of alcohol keyword

word: Represents the alcohol keyword

wordLabel: Represents the annotated label (alcohol-relation) of the alcohol keyword for that occurrence. Each keyword corresponds to one of the three categories: alcohol-related, non-alcohol related, or unsure.

Label "0" represents --> non-alcohol-related
Label "1" represents --> alcohol-related
Label "0" represents --> unsure

wordContext: Represents the annotated label (alcohol-related-context) of the alcohol keyword for that occurrence. Each keyword corresponds to one of the three categories: positive, negative, or neutral.

Label "0" represents --> neutral
Label "1" represents --> positive
Label "-1" represents --> negative

FilenamesAndIndices.csv: This file contains the fileIndex which corresponds to the song lyrics file in the corpus.

This file contains 3 columns:

1. fileIndex
2. filenames
3. year

fileIndex: Represents the index of the lyrics file in the corpus.

fileName: Represents the name of the lyrics file in the corpus.

year: Represents the year in which the song was released.

KeywordsFinal3.1.xlsx: Represents the alcohol keywords that were manually scraped by the annotators from various sources including academic research articles, and Wikipedia.

For more information please refer to the above citation or the below link to the published paper.

It would be highly appreciated if you can cite this article when you use this dataset for your research or commerial purposes.