Dataset for Application of Recommender Systems on Police Photo Lineup Assembling task

For more information about the Recommender Systems for Police Photo Lineup project, please visit: http://www.ksi.mff.cuni.cz/~peska/lineup/paper.pdf

The dataset consists of two parts:

The raw dataset contains visual and attribute-based desriptors of the list of candidate persons:

- personsCB_IDs.csv: ordered list of persons IDs
- personsData.csv: raw attribute features of the persons
- personsCBVectors.csv: derived binary attribute-features with TF-IDF applied. The ordering of records is the same as in personsCB_IDs.csv
- personsVectors.csv: derived visual descriptors of persons' images. Probability layer of VGG-Face CNN was applied for this task. The ordering of records is the same as in personsCB_IDs.csv.

The implicit feedback dataset (feedbackDatasetCB-RSFeatures.csv) contains information received from the user-study on lineup assembling task performed by seven domain experts. This table has following structure:

- evaluatorID;
- lineupID (id of the suspect);
- candidateID (person recommended to this particular suspect);
- calculated content-based similarity
- selection (1 if evaluator selected this candidate, 0 otherwise);
- substraction of CB features of the suspect and the candidate (this can be simply modified by accessing raw dataset to contain visual descriptors or any other combination of features)