recommender system tutorials and datasets

They suggest similar items based on a particular item. This system uses item metadata, such as genre, director, description, actors, etc. for movies, to make these recommendations. The general idea behind these recommender systems is that if a person liked a particular item, he or she will also like an item that is similar to it.

Collaborative Filtering

This system matches persons with similar interests and provides recommendations based on this matching. Collaborative filters do not require item metadata like its content-based counterparts.

Same MovieLens dataset, good walkthrough, covers content-based, collaborative, and hybrid of the two systems: https://www.kaggle.com/rounakbanik/movie-recommender-systems *

Another well-written kernel for movies, compares different algorithms and approaches to recommendation systems (more technical than the one above), including deep learning with Keras: https://www.kaggle.com/morrisb/how-to-recommend-anything-deep-recommender/comments *

For Restaurant dataset

Restaurant dataset: simple content-based recommender (writeup is a bit lengthy): https://www.kaggle.com/liyenhsu/simple-content-based-recommenders

For Anime dataset

Anime - collaborative filtering

For collaborative filtering we'll need to create a pivot table of users on one axis and tv show names along the other. The pivot table will help us in defining the similarity between users and shows to better predict who will like what.

https://www.kaggle.com/ajmichelutti/collaborative-filtering-on-anime-data *

Simple content-based anime recommendation system for Anime Recommendations Database - rec based on 'ratings' and 'genre': https://www.kaggle.com/astandrik/simple-anime-recommendation-system-content-based

Same dataset, content-based rec sys using KNN: https://www.kaggle.com/mohammed94/content-based-anime-recommender#KNN-for-finding-similar-animes

For Goodreads-10k books dataset (improved goodreads dataset available on Kaggle, see below):

Content-based: https://www.kaggle.com/bshirude2/goodreads-content-based-book-recommendation

Collaborative: https://www.kaggle.com/sriharshavogeti/collaborative-recommender-system-on-goodreads

Datasets

Kaggle datasets

List of the 7 main recommender datasets on Kaggle (as of late-2018), except for movie datasets which seem to be the most popular https://www.kaggle.com/general/65109#384033 (restaurants, hotels, books, anime, Jester jokes)

Note that goodbooks-10k dataset (Ten thousand books, one million ratings. Also books marked to read, and tags. ) is obsolete. Improved version of the goodbooks dataset is available at https://github.com/zygmuntz/goodbooks-10k. *

It has duplicates removed, and more ratings (six million), sorted by time. Book and user IDs are the same.

9MB download, "a medium sized song list, ideal to practice collaborative filter algorithms" https://www.kaggle.com/rymnikski/dataset-for-collaborative-filters (has one short project using fast.ai for collaborative filtering, but no comments)

dataset for a company to predict ad clicks

(no projects/tutorials available on data.world) https://data.world/hackerearth/predict-the-ad-clicks

Possibly relevant/For Later:

Project with a similar idea (found on July 28):

Produce2Recipe -- article, repo * From image classifier to Flask app; no recommendation system based on user preference like we are planning to do, but the "creating tagging recipes via NLP part could be relevant:

Using Non-negative Matrix Factorization (NMF) to perform topic modeling on scraped recipes (with tags)
focus was on image recognition; no evaluation metric for topic modelling?

Another project: FP3: Recipe (includes ingredients) to cuisine type (one of three sides of our ingredient-cuisine_type-recipe_name triangle)
- hyperparameter optimizing, using bagged classifier, decision tree classifier, and random forest classifier
- its scraped allrecipes dataset has keywords for each recipe (similar to the tags/labels we want to check against)
- Aug 18: installed gensim(documentations) and imbalanced-learn packages in order to run code from forked repo
- Aug 26 discussion, decided code in original repo was problematic: data was not sufficiently cleaned, and no explanation attempted for predicting "Unknown" category the most

2 good papers on rec-sys recommended by a Medium blog post:

Matrix factorization techniques for rec-sys: https://datajobs.com/data-science-repo/Recommender-Systems-[Netflix].pdf

Collaborative Filtering: http://yifanhu.net/PUB/cf.pdf * -- Winner of the IEEE ICDM 10-Year Highest-Impact Paper Award

Alternative apporach

look into using ULMFiT (neural networks and transfer learning), for classifying cuisine before recommendation? https://arxiv.org/abs/1801.06146

Criteo (French company specializing in recsys, and author of 1TB Clicks dataset)

highlighted some recent papers at a recsys conference:

https://labs.criteo.com/2017/10/criteos-take-recsys-2017-10-papers-not-miss/

more about their work:

https://labs.criteo.com/2016/09/product-recommendation-criteo/

Their famous ad dataset here, challenge for predicting ad click-through rate

https://www.kaggle.com/c/criteo-display-ad-challenge/overview

using google cloud ML

https://cloud.google.com/blog/products/gcp/using-google-cloud-machine-learning-to-predict-clicks-at-scale

floraxinru/recsys_resources

recommender system tutorials and datasets

Links of datasets and tutorials for simple recommender system as a precursor to recipe recommendation project

General

Tutorials, ready-to-use:

Short and simple challenge with Youtube video tutorial

https://github.com/llSourcell/recommender_system_challenge

https://www.youtube.com/watch?v=9gBC9R-msAk&list=PL2-dafEMk2A6QKz1mrk1uIGfHkC1zZ6UU&index=3

Walkthrough of song recommender system:

Building a collaborative filtering system with TensorFlow (new)

https://towardsdatascience.com/how-to-build-a-simple-song-recommender-296fcbc8c85

https://github.com/llSourcell/recommender_live

Article of collaborative filtering and SVD -- Theory

https://hackernoon.com/introduction-to-recommender-system-part-1-collaborative-filtering-singular-value-decomposition-44c9659c5e75

Well-documented rec-sys projects on Kaggle *