Exploring the Impact of Temporal Bias in Point-of-Interest Recommendation (RecSys 22)
The paper has been submitted to the RecSys conference and upon acceptance, the details of codes, datasets, and results will be made available.
- Folders:
- datasets: including datasets (Gowalla and Yelp)
- In Yelp dataset we do not have the frequecny of ckeck-ins, we have the rating to each location. For example, in Yelp_checkins.txt file we have 39 check-ins for user 0 and the numebr of records in Yelp_train.txt + Yelp_test.txt + Yelp_tune.txt is equal to 39. Should we consdier them as a frequency score?
- item_groups: groups of items (leisure and working items)
- plots: plots of analysis and results
- results: results of models
- user_groups: groups of users
- datasets: including datasets (Gowalla and Yelp)
- Notebooks:
- XXX:
- We use the generated results from the previous experiments (from
ESWA'21
) on Gowalla (i.e., files inresults/Gowalla
)- Each folder indicates a model.
- The main files are
results_top_N.txt
which are the user scores per metric.
- Run
temporalSplit.ipynb
to split users based on the different timestamp and generate some plots to show the correlation between attributes.- Outputs: plots in
\plots
and user and item groups in\groups
- Outputs: plots in
compute_results.py
merge_results.ipynb