Filippo Betello

PhD student in Data Science at Sapienza

💛❤🥋

[Sumit @_reachsumit]

Iterative Reranking as a Compute-Scaling Method for LLM-based Rankers

Proposes iteratively applying listwise LLM rankers to refine search results, trading compute for quality gains on difficult queries like comparative and multi-attribute searches.

📝https://t.co/Bij1DkL3ef

[RSTLess group @RSTLessGroup]

Today @giulio_deras presented two works in the poster session of #ICTIR25
Eclipse: Contrastive Dimension Importance Estimation with Pseudo-Irrelevance Feedback for Dense Retrieval
https://t.co/PMHvo5HbXp
and as a proxy presenter:
QPP-RA: Aggregating Large Language Model Rankings

[RSTLess group @RSTLessGroup]

@andreag_difra00, @giuditeodoro and @AntonioPuri00 presenting their work in the special session for Neural IR and RecSys at #IJCNN25, where @giuditeodoro also served as session chair! Lots of insights on Recommender systems. Link to the papers in the comments! #IR #Research

[RSTLess group @RSTLessGroup]

With just 3 lines of code, our lightweight saver lets you update EcoTaskSet with new training dynamics. Simple, fast, and ready to use:

https://t.co/Sq2D8BWbqf

[RSTLess group @RSTLessGroup]

GREEN: an inference-time method using our new dataset EcoTaskSet to recommend Pareto-optimal models balancing accuracy and emissions. Our UI lets users choose dataset & accuracy-emissions trade-off and recommends the best model
📚https://t.co/PJ4pnQJ9JF
💻 https://t.co/Fw5GDh3AF8

[RSTLess group @RSTLessGroup]

Our paper "Are Convolutional Sequential Recommender Systems Still Competitive? Introducing New Models and Insights" was accepted at #IJCNN! 🎉
With the right modifications, CNN-based SRSs can in some cases outperform attention-based models like SASrec by up to +53% in NDCG@10! 📈

[Filippo Betello @filippobetello]

Come to join and listen us !! #RecSys2024 #Bari #ACM

[RSTLess group @RSTLessGroup]

Tomorrow 4 presentations for our group.
@filippobetello & @_FedeSiciliano_ will present their works at the @RobustRecSys workshop (Room I, 14.15 CET).
@AntonioPuri00 will present at the #RecSoGood workshop (Room C, 12.15 CET).
See you at the @ACMRecSys! #Papers in the comments.

[RSTLess group @RSTLessGroup]

“The Role of Fake Users in Sequential Recommender Systems”, authored by @filippobetello has been accepted as an #oral presentation at the @RobustRecSys workshop, co-located with @ACMRecSys 24 in Bari. Filippo’s second work on #recsys #robustness after a paper accepted at #ECIR.

[RSTLess group @RSTLessGroup]

A workshop organized by members of our group, co-located with @ACMRecSys 24. Don’t miss the opportunity to submit your paper! They are waiting for LBR& DEMO notifications of acceptance!

[Filippo Betello @filippobetello]

Thanks @_reachsumit for sharing our work! kudos to all the authors @AntonioPuri00 @_FedeSiciliano_ @GioTrappolini @Andrea_Bacciu @ntonellotto and @fabreetseo
#RecSys #IR #Reproducibility

[Sumit @_reachsumit]

A Reproducible Analysis of Sequential Recommender Systems

Addresses reproducibility issues for Sequential Recommender Systems and challenges existing benchmarks, revealing several surprising insights about model performance.

📝https://t.co/5cZYQfpcTX
👨🏽‍💻https://t.co/Q0yvIKZN8Z

[RSTLess group @RSTLessGroup]

Say hello to @filippobetello , a 1st year #PhD student in #DataScience at Sapienza. He works with #RecommenderSystems and their #Robustness, #LLMs and #AI application to #medicine. He has a https://t.co/DLk7RxvXT0. in #AI & #Robotics
#PhD #RecSys

[Filippo Betello @filippobetello]

Special thanks to my colleague and co-author Salvatore Falciglia, and to Professor Christian Napoli for his many suggestions along the journey.

[Filippo Betello @filippobetello]

Unraveling the mysteries of the mind just got a little easier! Have a look at our latest paper "Learning Visual Stimulus-Evoked EEG Manifold for Neural Image Classification" published in Neurocomputing, Elsevier.

Link: https://t.co/Ns8E0HYclK

#EEG #DeepLearning #Elsevier

[Filippo Betello @filippobetello]

Using a novel architecture, that employs riemann geometry and deep learning techniques, we were able to classify images seen by a subject with up to 86% accuracy using only the EEG signal.

[Filippo Betello @filippobetello]

@ImPushMish @fabreetseo and the resulting ranked lists of suggestions share merely 10% of the ground truth items. These insights underscore significant implications for real-world recommender system applications. Access the preprint version here https://t.co/B0zZdx2KHI to read the full details.

[Filippo Betello @filippobetello]

Thrilled to announce that our full research paper, "Investigating the Robustness of Sequential Recommender Systems Under Training Data Perturbation" (co-authored with Federico Siciliano, @ImPushMish, and @fabreetseo ), has been accepted at the #ECIR24 conference.
🧵

[Filippo Betello @filippobetello]

@ImPushMish @fabreetseo In our study, we delve into the effects of item positioning within chronologically ordered training sequences. Our findings reveal a substantial impact: removing the most recent items can degrade NDCG@20 by as much as 60%..