When I read papers that use LLMs for tasks like IR or RecSys, I always have the data pollution question: we can't guarantee that our evaluation data hasn't been used for training. For example, Lllama 3 confesses that it has been trained on both MSMARCO and TREC DL datasets.

Wrapping up an amazing week with intense work on #recsys at Schloss Dagstuhl but also fun conversations and catching up with great people in the field

Skimming 'MagnusHammer: A Transformer-Based Approach to Premise Selection'

Amazing how similar it looks to a Ads/RecSys/Search/RAG stack. arxiv.org/pdf/2303.04488

Solving math and ads click prediction may not be unrelated problems.

Excellent closing of ECIR2024. Great tutorial on Transformers for RecSys by Craig Macdonald and Aleksandr V. Petrov. Thank you guys!

We finally have an answer to why EC would be investigating consent-or-pay under the DSA! Answer: Article 38, which requires a RecSys (apparently includes ad targeting despite DSA using term '[ads] systems' alongside) alternative w/o profiling. Martin Husovec Fernando Hortal Foronda

🏅Recommender Systems' Primers • recsys.aman.ai

🔹Candidate Generation & Retrieval
🔹Candidate Ranking & Re-ranking
🔹Cold Start Problem
🔹Position Bias
🔹Content Moderation
🔹Evaluation Metrics and Loss
🔹Multi-Armed Bandits
🔹GNNs for RecSys
🔹Transformers for RecSys

📝 Top Papers in Computer Vision, NLP, Speech, Multimodal AI, Core ML, RecSys, & Graph ML
🔗 papers.aman.ai

👉🏼 I’ve put together a summary of key papers in  #AI  and segregated them into (i) need-to-know and (ii) good-to-know.

🔹 Vision
- Image Classification (CNN

Why DOB is useful for music recsys.

Budget efficiency with Causal ML?

Here's how:

1/n

#causaltwitter #causalinference #causalAI #recsys #epitwitter

#RecSys using #MachineLearning Association Rules! #BigData #Analytics #DataScience #AI #MachineLearning #IoT #IIoT #PyTorch #Python #RStats #TensorFlow #Java #GoLang #CloudComputing #Serverless #DataScientist #Linux #Programming #Coding #100DaysofCode
geni.us/M-L-A-Rules

Python Brasil 2014

Estava palestrando de chinelos pq tinha queimado os pés no sol no dia anterior.

Lembro que nessa época o hype era Data Mining ⛏️ e NoSQL.

IA mais avançada na época era recsys e tinha até um framework brasileiro chamado crab.

📢 Excited to release PyG (PyTorch Geometric) v2.5, including distributed training, a dedicated graph tensor representation, RecSys support, PyTorch 2.2 and native compilation support 🎉

Release Notes 👇
github.com/pyg-team/pytor…

🧵 A thread [1/6]

if you're not doing cloud-native development on Modal for a Weaviate • vector database vector DB-backed, Hugging Face embeddings-based recsys app while riding in the back of a Waymo, are you really living in 2024?

RecSys Challenge 2024 started?The task is News Recommendation.Seems interesting but I am not sure it is worth to do as last year was not good,wait for one month to see.

📢We will be hosting the 1st EARL workshop on Evaluating and Applying Recommendation Systems with LLMs ( #earlworkshop ), in Recsys.com 2024! EARL explores LLMs in RSs, focusing on enhancements and challenges.
More info later: earl-workshop.github.io
⭐️ Stay Tuned!
#NLP #LLMs #AI

📈 Graph Neural Networks (GNNs) Primers: Overview, Applications in RecSys and NLP, and Case Studies

➡️ Overview of GNNs in RecSys and NLP: gnn.aman.ai
➡️ GNN-based RecSys with Case-Studies from Snap, Meta, Google, Uber, Pinterest: rec-gnn.aman.ai

#AI

My understanding of why item-knn remains a strong baseline for most recsys tasks.

This was indeed a great and very productive week to move forward #recsys research - thanks to everyone, especially the organizers!

youtube recsys on point