Daiquiri

Contributor

Tong Jia – cecilio.jia@gmail.com – https://github.com/Cecilio-Jia

Daiquiri is an easy-to-use, end-to-end and large scale scalable toolkit of machine-learning, deep-learning and reinforcement learning based sparse learning algorithm system, which can be used for building your own custom Recommender System and Detection System easily and efficiently.

This package is based on Python3.6, TensorFlow(1.12.0), and using tensorflow high level API Dataset and Estimator for constructing input function and model function, tensorflow-serving for serving the model.

In training phase, the system support three kinds of device topology:

• Single machine CPU version
• Single machine multi GPUs version (e.g. Ring Allreduce)
• Multi machine multi GPUs version (e.g. Parameter Server)

Essential tools:

• Python3.6
• TensorFlow(1.12.0)
• Docker
• TensorFlow-Serving

Architecture

• Retrieval Strategy
  • Collaborative filtering (e.g. SVD)
  • Embedding (e.g. item2vec)
  • Semantic matching (e.g. DSSM)
• Ranking Strategy
  • Click through rate models (e.g. FM)
• Exploration & Exploitation
  • Reinforcement learning models (e.g DQN)

Model list

Collaborative Filtering Based Models

Model	Conference	Paper	Contain
SVD	IEEE Computer Society'09	Matrix Factorization Techniques for Recommender Systems	✔
SVD++	KDD'08	Factorization Meets the Neighborhood: a Multifaceted Collaborative Filtering Model	✖
TrustSVD	AAAI'15	TrustSVD: Collaborative Filtering with Both the Explicit and Implicit Influence of User Trust and of Item Ratings	✖
AutoSVD++	SIGIR'17	AutoSVD++: An Efficient Hybrid Collaborative Filtering Model via Contractive Auto-encoders	✖

Embedding Based Models

Model	Conference	Paper	Contain
Item2vec	RecSys'16	Item2Vec: Neural Item Embedding for Collaborative Filtering	✖
LTR	KDD'18	Learning and Transferring IDs Representation in E-commerce	✖
AirbnbEmbed	KDD'18	Real-time Personalization using Embeddings for Search Ranking at Airbnb	✖

CTR Models

Model	Conference	Paper	Contain
LR (Baseline)		An Introduction to Logistic Regression Analysis and Reporting	✔
FM	ICDM'10	Factorization Machines	✔
FFM	RecSys'16	Field-aware Factorization Machines for CTR Prediction [Criteo]	✔
Wide & Deep	DLRS'16	Wide & Deep Learning for Recommender Systems [Google]	✔
FNN	ECIR'16	Deep Learning over Multi-field Categorical Data: A Case Study on User Response Prediction [RayCloud]	✖
PNN	ICDM'16	Product-based Neural Networks for User Response Prediction	✖
DeepFM	IJCAI'17	DeepFM: A Factorization-Machine based Neural Network for CTR Prediction [Huawei]	✔
AFM	IJCAI'17	Attentional Factorization Machines: Learning the Weight of Feature Interactions via Attention Networks	✖
NFM	SIGIR'17	Neural Factorization Machines for Sparse Predictive Analytics	✖
DCN	KDD'17	Deep & Cross Network for Ad Click Predictions	✔
DIN	KDD'18	Deep Interest Network for Click-Through Rate Prediction [Alibaba]	✔
AutoInt	arxiv'18	AutoInt: Automatic Feature Interaction Learning via Self-Attentive Neural Networks	✖
TEM	WWW'18	TEM: Tree-enhanced Embedding Model for Explainable Recommendation	✖
FNFM	arxiv'19	Field-aware Neural Factorization Machine for Click-Through Rate Prediction	✖