An investigation on different behaviours during entire song listening sessions with regards to the users' session-based skipping activity. The analysis is performed on the Spotify's Music Streaming Sessions (MSSD) Dataset.
This repository contains the source code for the approach outlined in the Short Paper On Skipping Behaviour Types in Music Streaming Sessions, accepted at the 30th ACM International Conference on Information and Knowledge Management (CIKM2021).
For the YouTube presentation on the submitted version of the paper, please click here.
To know more about our research activities at NeuraSearch Laboratory, please follow us on Twitter (@NeuraSearch) and to get notified of future uploads please subscribe to our YouTube channel!
The required Python packages can be found in requirements.txt. Using a package manager such as pip, they can be easily installed as follows:
pip3 install -r requirements.txt
The main scripts (experiment.py and analysis.py) require the entire MSSD training dataset to be in a specific format. Specifically, they require every day as a SQLite database. To do so, the following instructions have to be followed (only once):
Part 1. In data/training_set/, sub-folders for every day have to be created. The following structure is expected:
data/training_set/20180715/
data/training_set/20180716/
data/training_set/20180717/
...
data/training_set/20180918/
Part 2. In each of these newly created subfolders, copy all original csv files for that day from the original dataset. This means that, for example, in folder data/training_set/20180715/, it is expected to have log_0_20180715_000000000000.csv, log_1_20180715_000000000000.csv, log_2_20180715_000000000000.csv, ..., log_9_20180715_000000000000.csv.
Part 3. Run python data_preparation.py, and all individual databases should be automatically created. Depending on the amount of selected data, this process may make some time.
Having now completed the prior step (Data Preparation), it is now possible to run experiments. This can be done via the following command:
python experiment.py --name MyAllExperiment --type all -l 20 --pca 7
This will create an experiment in results, named MyAllExperiment, with an all experimental condition (meaning all sessions and on all days), for listening sessions of length 20, and with 7 PCA components. Further, individual boxplots for each skipping types are generated and available in the figures sub-folder.
The available experimental conditions flags are: all, weekday, weekend, morning, afternoon, evening, and night. Additionally, to perform an experiment on playlist types (e.g. editorial playlist), the array attribute context_types in experiment.py has to be modified accordingly. If empty (default), no playlist types filtering is applied when collecting listening sessions.
Finally, to modify the number of clusters, the N_CLUSTERS attribute in constants.py can be changed accordingly.
This last script allows for comparison, via clusters matching, on the identified types for experiments of a same session length. The metric used for matching clusters is the Euclidean distance. The analysis can be performed via the following command:
python analysis.py
Important to note is the fact that, when comparing distributions for different session lengths (via stacked histogram), it is required to manually rearrange the distr_dict rows to the desired sequence of skipping types. This is a necessary step if uou want to correctly report distributions on different lengths and for the same sequence of types, such as "listener, listen-then-skip, skip-then-listen, skipper".
Please, cite this work as follows:
@inproceedings{10.1145/3459637.3482123,
author = {Meggetto, Francesco and Revie, Crawford and Levine, John and Moshfeghi, Yashar},
title = {On Skipping Behaviour Types in Music Streaming Sessions},
year = {2021},
isbn = {9781450384469},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3459637.3482123},
doi = {10.1145/3459637.3482123},
booktitle = {Proceedings of the 30th ACM International Conference on Information & Knowledge Management},
pages = {3333–3337},
numpages = {5},
keywords = {spotify, skipping, session, music, listening, user behaviour},
location = {Virtual Event, Queensland, Australia},
series = {CIKM ’21}
}
Francesco Meggetto, Crawford Revie, John Levine, and Yashar Moshfeghi. 2021. On Skipping Behaviour Types in Music Streaming Sessions. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management(CIKM ’21). Association for Computing Machinery, New York, NY, USA, 3333–3337. DOI:https://doi.org/10.1145/3459637.3482123