Extraction Details

To capture relevant information from studies, we expanded BiBTeX fields for each study with additional fields. For reproducibility, these instructions provide information on the process followed for each field.

Fields

IDENTIFIER

Unique identifier of article. Contains last name of lead author, year of publication and first two letters of article title. Hyphenated last names collapsed.

AUTHOR

Names of all authors. Last name precedes first name and separated by comma. For multiple authors “and” precedes each listed subsequent author. E.g., Sorussa, Kanawat and Choksuriwong, Anant and Karnjanadecha, Montri

JOURNAL

Title of journal containing article.

NOTE

Includes number of citing articles and open access details. E.g., Cited by: 4; All Open Access, Gold Open Access, Green Open Access

TITLE

Title of article.

VOLUME

Volume number of publication.

YEAR

Publication year.

DOI

Digital object identifier of article.

ABSTRACT

Complete text of article abstract.

SOURCE

Database article was sourced from. Scopus, Web of Science (WoS) or OpenAlex.

AUTHOR_KEYWORDS

Corresponding keywords for article indicated by author.

NOTES_AUTHORINITIALS

Decision and comments by respective author

STIMULUS_TYPE

Metadata pertaining to stimuli employed in paradigm. Can be listed as genres of music stimuli employed, or if stimuli come from a standard database, name of standard.

STIMULUS_DURATION

Duration of stimuli, if applicable. Unit of measurement (seconds, measures) specified in STIMULUS_DURATION_UNIT

STIMULUS_DURATION_UNIT

Unit of measurement pertaining to STIMULUS_DURATION. E.g., seconds, measures, etc.

STIMULUS_N

Number of stimuli employed in experiment. If multiple experimental conditions reported, separate \(n\) by conditions where possible.

FEATURE_N

Number of features included in data modeling (if available).

PARTICIPANT_N

Total number of participants in experiment.

PARTICIPANT_EXPERTISE

Expertise of annotators. E.g., experts, non-experts, not specified.

PARTICIPANT_ORIGIN

Origin country of participants, or online platform participants were recruited from (e.g., MTurk)

PARTICIPANT_SAMPLING

How participants were recruited (e.g., convenience, random sampling, crowdsourcing)

PARTICIPANT_TASK

Nature of rating/classification task undertaken by participants. E.g., rate, annotate.

FEATURE_CATEGORIES

Names of categories analyzed features pertain to, based on names in Panda (2021). Includes names of all pertinent categories: Melody, Rhythm, Timbre, Pitch, Tonality, Expressivity, Texture, Form, Vocal, High-Level

FEATURE_SOURCE

Name(s) of feature analysis toolbox(es).

FEATURE_REDUCTION_METHOD

Name(s) of feature reduction or feature selection methods employed.

MODEL_CATEGORY

Name of model type (regression, classification, or both).

MODEL_DETAIL

Additional information pertaining to predictive model, such as the name of algorithm used and other pertinent parameters. E.g., Random Forest, Commonality Analysis, Multiple Regression, Neural Networks, LDSM.

MODEL_MEASURE

Metric used in model evaluation. E.g., \(R^2\), \(MSE\), \(CCC\), Classification accuracy, etc.

MODEL_COMPLEXITY_PARAMETERS

Additional information pertaining to predictive model. E.g., training epochs: 100; n layers: 1, 2; LSTM units: 124,248.

MODEL_RATE_EMOTION_NAMES

Names of predicted emotions. E.g., valence, arousal, happy, sad, angry, fearful, etc.

MODEL_RATE_EMOTION_VALUES

Pertinent prediction of model summaries. Report as R named arrays, including summary statistics in variables. When reporting results of multiple models, concatenate multiple entries with bind_field. When reporting results for different toolboxes or feature subsets, assign each to a new BiBTeX field with relevant identifier following final underscore. See additional details below.

MODEL_VALIDATION

Validation method used (if applicable). E.g., 10-fold cross validation, leave one out cross validation.

Encoding study results:

Regression

Report relevant summary statistics in the MODEL_RATE_EMOTION_VALUES field in R code using the following format. Parameters delimited by spaces are later assigned to separate columns in the resulting dataframe.

bind_field(
library.model.features.data.experiment = c(dimension_measure.summaryStat = 0, ...),
...
)

Classification

Confusion matrices can be encoded using the unflatten function, which assigns relevant meta-parameters to the model_parameters attribute of the output matrix. The function automatically builds a \(n\) by \(n\) matrix and takes individual values as arguments. Values can be specified for the first row, and are recycled in subsequent rows.

unflatten(A = .5, B = .2, C = .3,
          .3, .5, .2,
          .1, .1, .8)

Summary parameters can be extracted from the resulting matrix by calling the summarize_matrix function. This function calls on caret to extract several summary statistics at once. Including bind_field in the call ensures consistent nomenclature with other study encodings.

bind_field(
  'library.model.features.data.exp' = summarize_matrix(
    unflatten(Q1=185.85,Q2=14.4,Q3=8.6,Q4=18.15,
              23.95,190.55,7,3.5,
              14.2,8.4,157.25,45.15,
              24.35,1.65,45.85,153.15)
  )
)

Multiple matrices in a study can be summarized by putting them in a list and using the lapply function to extract relevant statistics:

bind_field(
  lapply(
    list(
      'library.model.features.data.exp1' = unflatten(
        Angry=.984,Happy=.007,Relax=.007,Sad=0,
        0,.987,0,.007,
        .008,.007,.987,0,
        .008,0,.007,.993
        ),
      'library.model.features.data.exp2' = unflatten(
        Angry=.976,Happy=.013,Relax=.014,Sad=0,
        .008,.967,0,.02,
        .008,.013,.98,.007,
        .008,.007,.007,.974
       )
      ),
    summarize_matrix
  )
)

When confusion matrices are not available, encode available parameters (accuracy, precision, recall, \(F\) scores, etc.) using the standard nomenclature to distinguish relevant outcomes for each class:

bind_field(
library.model.features.data.experiment = c(class_measure.summaryStat = 0, ...),
...
)
Back to top