Introduction to Recommender Systems

[manisnesan]

RecSys Intro from Google Developers

Why recsys needed? Easier content browsing - helps users find content that they did not thought about asking

Components of recsys Candidate generation - Scoring - Reranking

CG - huge corpus in billions to smaller ones hundreds | 1000s - fast Score - precise model that scores and ranks to get the top 10, can use additional queries Rerank - rerank the top 10 based on filtering and boosting - diversity, freshness, fairness

Refer Deep NN for YouTube ranking

[manisnesan]

Study Group - Recsys Series Date: 2022-06-05

• Building blocks of Recommender Systems - Slides
• ML System Design Jam: Recsys Session 1 (Google’s intro to RecSys ) by Chloe He
• RecSys Series 1 : Intro to RecSys
• Designing Twitter's Trending Hashtags Solution

[manisnesan]

Study Group - Recsys Series Date: 2022-06-12

• ML System Design Jam: Recsys Session 2 - Wide and Deep Learning by Jin Yun Soo and Han-chung Lee
• Paper - Wide & Deep Learning for Recommender Systems

Related

• Wild Wild Tests - walkthrough for recSys use cases, basic metrics with examples and of course a bit of behavioral tests.
• reclist - GitHub - Behavioral "black-box" testing for recommender systems

[manisnesan]

Toy Project to practice

1. RecSys project that covers the bare minimum (retrieval/filtering/scoring/ordering) in some operational toy example. Like the Toy Machine Learning Project by Shreya Shankar (https://github.com/shreyashankar/toy-ml-pipeline)

2. Public dataset for RecSys algorithms (Spotify Million Playlist Dataset https://www.aicrowd.com/challenges/spotify-million-playlist-dataset-challenge)

Competitions

• https://www.kaggle.com/competitions/h-and-m-personalized-fashion-recommendations
• https://www.kaggle.com/competitions/shopee-product-matching

[manisnesan]

real-world (data included) mlops projects with real-world tools , you can check out https://github.com/jacopotagliabue/you-dont-need-a-bigger-boat or the simpler https://github.com/jacopotagliabue/post-modern-stack .

For public datasets, datasets required depending on the use cases: session recs? - https://github.com/coveooss/SIGIR-ecom-data-challenge Item to item? User to item?

https://recsys.trivago.cloud/