RecGaze Click Feedback Dataset

This is the RecGaze Click Feedback Dataset from the paper: From Latent to Observable Position-Based Click Models in Carousels

Link to open-acess paper: KDD 2026, Arxiv

This click feedback dataset is made from the RecGaze Dataset, the first eye tracking and user interaction dataset for carousel interfaces. To request access to the original dataset with user and item features along with timestamped feedback (fixations, gaze, clicks, cursor movements) data: RecGaze Dataset Public Version

Please cite the following:

@misc{deleonmartinez2026latentobservablepositionbasedclick,      title={From Latent to Observable Position-Based Click Models in Carousel Interfaces},       author={Santiago de Leon-Martinez and Robert Moro and Maria Bielikova},      year={2026},      eprint={2602.16541},      archivePrefix={arXiv},      primaryClass={cs.IR},      url={https://arxiv.org/abs/2602.16541}, }

@inproceedings{10.1145/3726302.3730301,author = {de Leon-Martinez, Santiago and Kang, Jingwei and Moro, Robert and de Rijke, Maarten and Kveton, Branislav and Oosterhuis, Harrie and Bielikova, Maria},title = {RecGaze: The First Eye Tracking and User Interaction Dataset for Carousel Interfaces},year = {2025},isbn = {9798400715921},publisher = {Association for Computing Machinery},address = {New York, NY, USA},url = {https://doi.org/10.1145/3726302.3730301},doi = {10.1145/3726302.3730301},booktitle = {Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval},pages = {3702–3711},numpages = {10},keywords = {browsing behavior, carousel interfaces, eye tracking},location = {Padua, Italy},series = {SIGIR '25}}

@inproceedings{10.1145/3742413.3789166,author = {de Leon-Martinez, Santiago and Moro, Robert and Kveton, Branislav and Bielikova, Maria},title = {Riding the Carousel: The First Extensive Eye Tracking Analysis of Browsing Behavior in Carousel Recommenders},year = {2026},isbn = {9798400719844},publisher = {Association for Computing Machinery},address = {New York, NY, USA},url = {https://doi.org/10.1145/3742413.3789166},doi = {10.1145/3742413.3789166},booktitle = {Proceedings of the 31st International Conference on Intelligent User Interfaces},pages = {2120–2130},numpages = {11},keywords = {Carousel interfaces, Multi-list recommendations, Browsing behavior, Eye tracking},location = {},series = {IUI '26}}

Conditions/Terms for Download & Use of Dataset

By downloading the dataset, you agree to the following terms: 

You will use the dataset for research (non-commercial) purposes related to: Eye Tracking, Psychology, Human Computer Interaction, Information Retrieval, Recommender Systems.

You will not attempt to identify or deanonymize the users.

You will appropriately cite the papers mentioned in the dataset description in any publication, project, tool using this dataset.

You acknowledge that you are fully responsible for the use of the dataset (data) and for any infringement of rights of third parties (in particular copyright) that may arise from its use beyond the intended purposes. Neither the authors nor Kempelen Institute of Intelligent Technologies (KInIT) are responsible for your actions.

Dataset Description

The RecGaze Click Feedback Dataset is the largest publicly available carousel feedback dataset specifically designed for click modeling. It contains 2,387 sessions and clicks (i.e. users only selected/clicked one movie), 1,307 unique movie items, and 87 unique users. The dataset comprises of interactions from 30 free-browsing task/screens, where participants were instructed to pick one movie they would like to watch on a Netflix-like interface. The 30 task/screens of carousel interfaces were identical accross users (i.e. the items on and ordering found each TaskID is the same for all users). Each screen contains 10 carousel genres (Action, Animation, Comedy, Crime, Drama, Fantasy, Horror, Romance, Sci-Fi, Thriller) randomly ordered accross tasks/screens and 15 items in each carousel for a total of 150 items per task/screen. For a more detailed description of the eye tracked user study, please refer to the the RecGaze paper and the RecGaze GiHub. All features that share a name with the RecGaze dataset are identical. User genre ratings (preferred/non-preferred; top vs rest; 1 to 5 star rating) for each of the 10 genres are also available in the  RecGaze Dataset Public Version. 

The dataset is organized, such that for each session (a unique user and task/screen pairing) there are 150 rows for each of the items present on the carousel interface and indicators if the item was clicked, its position, and if it was impressed or observed by fixation (eye tracker registers that the user fixated/observed the item).  Two filtering steps were used in creating the dataset: (1) sessions that had no registered eye tracking fixations were removed; (2) movies with no clicks accross all sessions were removed. 

The above filtering resulted in the free_browsing_one_click.csv 

In the paper From Latent to Observable Position-Based Click Models in Carousels, the dataset was further filtered to only include impressed items (impressed == 1) resulting in the free_browsing_one_click_impressed.csv

Column Name
Possible Values 
Explanation 

UserID
string (KInIT_1-61 or UvA_1-26)
Institute where the data was gathered, followed by a simple ID for the participant.

TaskID
int (1-40)
Identifier for the screen/task from which data was gathered. 1-30 are the 30 Free-browsing tasks/screens.

MovieID
int
ID for movie in dataset (same as TMDB_id from RecGaze Github).

clicked
int (0, 1) 
Indicator if an item was clicked (1) or not (0)

position
int (1-150)
Position of the item on the whole carousel screen

col_position
int (1-15)
Column position of the item

row_position
int (1-10)
Row position of the item

impressed_by_cascade
int (0, 1) 
Whether the item was impressed (1) or not impressed (0) following a cascade assumption from the click. 

impressed
int (0, 1) 
Whether the item was impressed (1) or not impressed (0). An item is impressed if it was shown on the viewport (visible part of the screen, i.e, it was swiped or scrolled to) during the session 

has_movie_fixations
int (1)
Whether the session has registered fixations on at least one movie. This was used in the pre-processing to create the dataset.

Observed_by_fixation
int (0,1) 
Whether the movie was observed (1) or not (0) given by the eye tracking data. 

session_id
Int (0-2,386)
An identifier given to each unique session.