@invinc4u@sscdotopen Yes, the GitHub link is in the paper. We also evaluated on one Instacart dataset you guys open sourced a few years back, looking forward to the real test! Let's keep in touch if you have further questions:)
@invinc4u@jnvinagre@olivierjeunen By within basket, I assume you mean next item reco is a single session. Whether machine unlearning can be applied efficiently depends the modelling approaches. As in our paper if one formulates within basket reco as NBR w/ a single-item basket as target, then answer is yes.
@sscdotopen I did some digging, and found one interesting section (15) in the guideline stating "the algorithms recommendation service providers should provide users the capabilities to select, modify or delete user tags/labels"
Source (in Chinese): https://t.co/2db0DiLfYr
There are many models in recommender systems. It is not our interest to propose yet another new model, but to look at how the models are defined/trained/evaluated. We report our understandings in these two "revisit" papers. **Warning**: no formulas, no new models.
Nice to see industry folks, often from @GoogleAI, dropping me notes of thanks saying my @ucsd_cse grad course on Data Systems for ML was helpful for them. Thank you for the kind act folks! ๐
All lectures/videos/materials are public: https://t.co/SyCjpupPrN I welcome feedback.
"PhD Position on Causal Inference & Machine Learning"
I was struck by this AD from TU Delft, The Netherlands, for putting CI first and ML second :
https://t.co/kPrtfGoXfW
Evidently, the faculty at TU Delft understand where the future of ML lies. If qualified, I would apply.
Nice writeup by @Analyticsindiam on why I think we should move from model-centric AI development (where the emphasis is on improving models) toward #DataCentricAI (where we systematically improve the data, using MLOps tools). https://t.co/7ClnrLKVGC