SlideShare a Scribd company logo

Recommendation Engine with In-Database Machine Learning

TigerGraph, profile picture
TigerGraph

The document discusses a recommendation engine utilizing in-database machine learning for efficient model training and prediction in various applications like movie recommendations and fraud detection. It highlights the advantages of in-situ machine learning, including continuous model training, scalability, and the implementation of a latent factor recommendation model. A detailed approach to movie rating prediction and the training methodology using a graph database and GSQL implementation is also presented.

Data & Analytics
Related topics:
Graph Database Insights
1 of 29
Download to read offline
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
Recommendation Engine with In-Database Machine Learning Changran Liu, Mingxi Wu
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | Outline ● In-database model training ● Latent factor recommendation model ● Distributed model training in Graph ● Demo ● GSQL implementation
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | Traditional Model Training Pipeline training data model request results Database: ● data storage ● data update ● preprocess data Machine learning platform ● model training ● model validation Applications: ● place order ● recommendation ● ...
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | In-situ machine learning in database: ● No need for exporting data ● Better support continuous model training over evolving data ● Less limitation on model size ● Support distributed model training In-Database Model Training request results Applications: ● recommendation ● fraud detection ● ...
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | Movie Recommendation movie features users ratings Goals: ● Predict users' ratings for movies they haven't seen, based on previous ratings ● Recommend movies to users based on rating prediction
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | Movie Rating Prediction (Latent factors model) Movie Alice Bob Carol Dave Love at last 5 5 0 0 Romance forever 5 ? ? 0 Cute puppies of love ? 4 0 ? Toy story ? ? ? 5 Sword vs. karate 0 0 5 ? Nonstop car chases 0 0 5 4 ● Each movie has a latent factor vector: θ(j) ● Each user has a latent factor vector: x(i) ● Predict the user j’s rating to movie i by: (θ(j) )T x(i) θ(1) = [5, 0] θ(2) = [5, 0] θ(3) = [0, 5] θ(4) = [0, 5] x(1) = [0.9, 0] x(2) = [1, 0.1] x(3) = [0.9, 0] x(4) = [0.1, 1] x(5) = [0.1, 1] x(6) = [0, 0.9] 4.5 5 4.5 0.5 0.5 0
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | Movie Rating Prediction (Latent factors model) Movie Alice Bob Carol Dave Love at last 5 5 0 0 Romance forever 5 ? ? 0 Cute puppies of love ? 4 0 ? Toy story ? ? ? 5 Sword vs. karate 0 0 5 ? Nonstop car chases 0 0 5 4 θ(1) = [5, 0] ● Each movie has a latent factor vector: θ(j) ● Each user has a latent factor vector: x(i) ● Predict the user j’s rating to movie i by: (θ(j) )T x(i) θ(2) = [5, 0] θ(3) = [0, 5] θ(4) = [0, 5] x(1) = [0.9, 0] x(2) = [1, 0.1] x(3) = [0.9, 0] x(4) = [0.1, 1] x(5) = [0.1, 1] x(6) = [0, 0.9] action romance 4.5 5 4.5 0.5 0.5 0
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | Cost Function RMSE regularization User 2 Movie 1 Movie 2 Movie 3 User 1 rating: y (1,1) rating: y (1,2) rating: y (2,2) rating: y (2,3) θ(1) θ(2) x(1) x(2) x(3)
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | Model Training in Graph User 2 Movie 1 Movie 2 Movie 3 User 1 rating: y(1,1) rating: y (1,2) rating: y (2,2) rating: y(2,3) θ(1) θ(2) x(1) x(2) x(3) Phase 1: ● Collect x(i) , y(i,j) from the movies that each user rated
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | Model Training in Graph User 2 Movie 1 Movie 2 Movie 3 User 1 rating: y(1,1) rating: y (1,2) rating: y (2,2) rating: y(2,3) [θ(1) , x(1) , y(1,1) ] [θ(1) , x(2) , y(1,2) ] x(1) x(2) x(3) [θ(2) , x(2) , y(2,2) ] [θ(2) , x(3) , y(2,3) ] Phase 1: ● Collect x(i) , y(i,j) from the movies that each user rated
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | Model Training in Graph User 2 Movie 1 Movie 2 Movie 3 User 1 rating: y(1,1) rating: y (1,2) rating: y (2,2) rating: y(2,3) [θ(1) , x(1) , y(1,1) ] →( (θ(1) )T x(1) -y(1,1) ) x(1) [θ(1) , x(2) , y(1,2) ] →( (θ(1) )T x(2) -y(1,1) ) x(2) x(1) x(2) x(3) [θ(2) , x(2) , y(2,2) ] → ((θ(2) )T x(2) -y(2,2) ) x(2) [θ(2) , x(3) , y(2,3) ] → ((θ(2) )T x(3) -y(2,3) ) x(3) Phase 1: ● Collect x(i) , y(i,j) from the movies that each user rated ● Compute the gradient contributed by each movies
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | Model Training in Graph User 2 Movie 1 Movie 2 Movie 3 User 1 rating: y(1,1) rating: y (1,2) rating: y (2,2) rating: y(2,3) ( (θ(1) )T x(1) -y(1,1) ) x(1) + ( (θ(1) )T x(2) -y(1,1) ) x(2) x(1) x(2) x(3) ((θ(2) )T x(2) -y(2,2) ) x(2) + ((θ(2) )T x(3) -y(2,3) ) x(3) Phase 1: ● Collect x(i) , y(i,j) from the movies that each user rated ● Compute the gradient contributed by each movies Phase 2: ● Aggregate the gradient
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | Model Training in Graph User 2 Movie 1 Movie 2 Movie 3 User 1 rating: y(1,1) rating: y (1,2) rating: y (2,2) rating: y(2,3) x(1) x(2) x(3) Phase 1: ● Collect x(i) , y(i,j) from the movies that each user rated ● Compute the gradient contributed by each movies Phase 2: ● Aggregate the gradient ● Update the feature vector using gradient descent
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | Training Split data Initialize latent factor vectors diff. between prediction and label converged? no finish yes update latent vectors using gradient descent (splitData.gsql) (initialization.gsql) (training.gsql)
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | Demo
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | GSQL Training Block USERs = SELECT s FROM USERs:s -(rate:e)-> MOVIE:t ACCUM DOUBLE prediction = dotProduct(s.@theta,t.@x), DOUBLE delta = prediction-e.rating, s.@Gradient += product(t.@x,delta), t.@Gradient += product(s.@theta,delta) POST-ACCUM s.@theta += product(s.@Gradient,-alpha), t.@x += product(t.@Gradient,-alpha); Dave Romance forever Love at last Nonstop car chases Alice rating: 5 rating: 5 rating: 0 rating: 4 θ = [1.5, 1.7] θ = [1.0, 1.5] x = [2.0, 2.3] x = [2.0, 1.3] x = [1.0, 1.3]
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | GSQL Training Block USERs = SELECT s FROM USERs:s -(rate:e)-> MOVIE:t ACCUM DOUBLE prediction = dotProduct(s.@theta,t.@x), DOUBLE delta = prediction-e.rating, s.@Gradient += product(t.@x,delta), t.@Gradient += product(s.@theta,delta) POST-ACCUM s.@theta += product(s.@Gradient,-alpha), t.@x += product(t.@Gradient,-alpha); Dave Romance forever Love at last Nonstop car chases Alice rating: 5 rating: 5 rating: 0 rating: 4 θ = [1.5, 1.7] θ = [1.0, 1.5] x = [2.0, 2.3] x = [2.0, 1.3] x = [1.0, 1.3]
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | GSQL Training Block USERs = SELECT s FROM USERs:s -(rate:e)-> MOVIE:t ACCUM DOUBLE prediction = dotProduct(s.@theta,t.@x), DOUBLE delta = prediction-e.rating, s.@Gradient += product(t.@x,delta), t.@Gradient += product(s.@theta,delta) POST-ACCUM s.@theta += product(s.@Gradient,-alpha), t.@x += product(t.@Gradient,-alpha); Dave Romance forever Love at last Nonstop car chases Alice rating: 5 rating: 5 rating: 0 rating: 4 θ = [1.5, 1.7] θ = [1.0, 1.5] x = [2.0, 2.3] x = [2.0, 1.3] x = [1.0, 1.3] prediction: 6.9 prediction: 5.2 prediction: 4.0 prediction: 3.0
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | GSQL Training Block USERs = SELECT s FROM USERs:s -(rate:e)-> MOVIE:t ACCUM DOUBLE prediction = dotProduct(s.@theta,t.@x), DOUBLE delta = prediction-e.rating, s.@Gradient += product(t.@x,delta), t.@Gradient += product(s.@theta,delta) POST-ACCUM s.@theta += product(s.@Gradient,-alpha), t.@x += product(t.@Gradient,-alpha); Dave Romance forever Love at last Nonstop car chases Alice ẟ: 1.9 ẟ: 0.2 ẟ: 4.0 ẟ: -1.1 θ = [1.5, 1.7] θ = [1.0, 1.5] x = [2.0, 2.3] x = [2.0, 1.3] x = [1.0, 1.3]
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | GSQL Training Block USERs = SELECT s FROM USERs:s -(rate:e)-> MOVIE:t ACCUM DOUBLE prediction = dotProduct(s.@theta,t.@x), DOUBLE delta = prediction-e.rating, s.@Gradient += product(t.@x,delta), t.@Gradient += product(s.@theta,delta) POST-ACCUM s.@theta += product(s.@Gradient,-alpha), t.@x += product(t.@Gradient,-alpha); Dave Romance forever Love at last Nonstop car chases Alice ẟ: 1.9 ẟ: 0.2 ẟ: 4.0 ẟ: -1.1 θ = [1.5, 1.7] grad(θ) = [4.2, 4.7] θ = [1.0, 1.5] grad(θ) = [6.9, 3.8] x = [2.0, 2.3] x = [2.0, 1.3] x = [1.0, 1.3]
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | GSQL Training Block USERs = SELECT s FROM USERs:s -(rate:e)-> MOVIE:t ACCUM DOUBLE prediction = dotProduct(s.@theta,t.@x), DOUBLE delta = prediction-e.rating, s.@Gradient += product(t.@x,delta), t.@Gradient += product(s.@theta,delta) POST-ACCUM s.@theta += product(s.@Gradient,-alpha), t.@x += product(t.@Gradient,-alpha); Dave Romance forever Love at last Nonstop car chases Alice ẟ: 1.9 ẟ: 0.2 ẟ: 4.0 ẟ: -1.1 θ = [1.5, 1.7] grad(θ) = [4.2, 4.7] θ = [1.0, 1.5] grad(θ) = [6.9, 3.8] x = [2.0, 2.3] grad(x) = [2.9, 3.2] x = [2.0, 1.3] grad(x) = [4.3, 6.3] x = [1.0, 1.3] grad(x) = [-1.1, -1.6]
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | GSQL Training Block USERs = SELECT s FROM USERs:s -(rate:e)-> MOVIE:t ACCUM DOUBLE prediction = dotProduct(s.@theta,t.@x), DOUBLE delta = prediction-e.rating, s.@Gradient += product(t.@x,delta), t.@Gradient += product(s.@theta,delta) POST-ACCUM s.@theta += product(s.@Gradient,-alpha), t.@x += product(t.@Gradient,-alpha); Dave Romance forever Love at last Nonstop car chases Alice ẟ: 1.9 ẟ: 0.2 ẟ: 4.0 ẟ: -1.1 θ = [1.5, 1.7] θ’ = [1.46, 1.65] θ = [1.0, 1.5] θ’ = [0.93, 1.46] x = [2.0, 2.3] x’ = [1.97, 2.27] x = [2.0, 1.3] x’ = [1.96, 1.24] x = [1.0, 1.3] x’ = [1.01, 1.32] * alpha = 0.01
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | GSQL Training Block USERs = SELECT s FROM USERs:s -(rate:e)-> MOVIE:t ACCUM DOUBLE prediction = dotProduct(s.@theta,t.@x), DOUBLE delta = prediction-e.rating, s.@Gradient += product(t.@x,delta), t.@Gradient += product(s.@theta,delta) POST-ACCUM s.@theta += product(s.@Gradient,-alpha), t.@x += product(t.@Gradient,-alpha); Dave Romance forever Love at last Nonstop car chases Alice ẟ: 1.9 ẟ’: 1.6 ẟ: 0.2 ẟ’: -0.1 ẟ: 4.0 ẟ’: 3.6 ẟ: -1.1 ẟ’: -1.1 θ = [1.5, 1.7] θ’ = [1.46, 1.65] θ = [1.0, 1.5] θ’ = [0.93, 1.46] x = [2.0, 2.3] x’ = [1.97, 2.27] x = [2.0, 1.3] x’ = [1.96, 1.24] x = [1.0, 1.3] x’ = [1.01, 1.32]
Q&A
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | User-Rate-Movie Graph ● Content based method Toy story ● Disney ● ... Iron man ● Marvel ● Action ● ... Alice ● Disney fan ● Marvel fan ● ... Bob ● Marvel fan ● ... rating: 5 rating: 5 rating:4.5 rating:?
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | User-Rate-Movie Graph ● Content based method Toy story ● Disney ● ... Iron man ● Marvel ● Action ● ... Alice ● Disney fan ● Marvel fan ● ... Bob ● Marvel fan ● ... rating: 5 rating: 5 rating:4.5 rating:? ● K-nearest neighbors
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | User-Rate-Movie Graph ● Content based method ● K-nearest neighbors Toy story ● Disney ● ... Iron man ● Marvel ● Action ● ... Alice ● Disney fan ● Marvel fan ● ... Bob ● Marvel fan ● ... rating: 5 rating: 5 rating:? ● Latent factor (model-based)
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | User-Rate-Movie Graph ● Content based method ● K-nearest neighbors ● Latent factor (model-based) ● Hybrid method ● ... Toy story ● Disney ● ... Iron man ● Marvel ● Action ● ... Alice ● Disney fan ● Marvel fan ● ... Bob ● Marvel fan ● ... rating: 5 rating: 5 rating:?
| GRAPHAIWORLD.COM | #GRAPHAIWORLD | User-Rate-Movie Graph ● Content based method ● K-nearest neighbors ● Latent factor (model-based) ● Hybrid method ● ... Toy story ● Disney ● ... Iron man ● Marvel ● Action ● ... Alice ● Disney fan ● Marvel fan ● ... Bob ● Marvel fan ● ... rating: 5 rating: 5 rating:?

More Related Content

PDF
Image-to-Image Translation with Conditional Adversarial Nets (UPC Reading Group)
Universitat Politècnica de Catalunya
 
PDF
Image-to-Image Translation
Junho Kim
 
PDF
Finding connections among images using CycleGAN
NAVER Engineering
 
PDF
1시간만에 GAN(Generative Adversarial Network) 완전 정복하기
NAVER Engineering
 
PDF
그림 그리는 AI
NAVER Engineering
 
PDF
Unsupervised learning represenation with DCGAN
Shyam Krishna Khadka
 
PDF
Generative adversarial networks
남주 김
 
PPTX
Generative Adversarial Networks (GAN)
Manohar Mukku
 
Image-to-Image Translation with Conditional Adversarial Nets (UPC Reading Group)
Universitat Politècnica de Catalunya
 
Image-to-Image Translation
Junho Kim
 
Finding connections among images using CycleGAN
NAVER Engineering
 
1시간만에 GAN(Generative Adversarial Network) 완전 정복하기
NAVER Engineering
 
그림 그리는 AI
NAVER Engineering
 
Unsupervised learning represenation with DCGAN
Shyam Krishna Khadka
 
Generative adversarial networks
남주 김
 
Generative Adversarial Networks (GAN)
Manohar Mukku
 

What's hot (20)

PPTX
Machine Learning with R
Barbara Fusinska
 
PPTX
Adversarial learning for neural dialogue generation
Keon Kim
 
PDF
DiscoGAN - Learning to Discover Cross-Domain Relations with Generative Advers...
Taeksoo Kim
 
PDF
Generative Adversarial Network (+Laplacian Pyramid GAN)
NamHyuk Ahn
 
PPTX
Clean, Learn and Visualise data with R
Barbara Fusinska
 
PDF
[GAN by Hung-yi Lee]Part 1: General introduction of GAN
NAVER Engineering
 
PDF
Generative Adversarial Networks
Mustafa Yagmur
 
PDF
Introduction to Generative Adversarial Networks
BennoG1
 
PDF
Basic Generative Adversarial Networks
Dong Heon Cho
 
PDF
Generative adversarial networks
Yunjey Choi
 
PDF
Generative adversarial text to image synthesis
Universitat Politècnica de Catalunya
 
PPTX
Reading group gan - 20170417
Shuai Zhang
 
PPTX
Gan seminar
San Kim
 
PDF
GAN - Theory and Applications
Emanuele Ghelfi
 
PDF
ddpg seminar
민재 정
 
PPTX
ICASSP 2018 Tutorial: Generative Adversarial Network and its Applications to ...
宏毅 李
 
PDF
Deep Generative Models
Mijung Kim
 
PPTX
3D Multi Object GAN
Yu Nishimura
 
PDF
Generative Adversarial Networks 2
Alireza Shafaei
 
PDF
Variational Autoencoded Regression of Visual Data with Generative Adversarial...
NAVER Engineering
 
Machine Learning with R
Barbara Fusinska
 
Adversarial learning for neural dialogue generation
Keon Kim
 
DiscoGAN - Learning to Discover Cross-Domain Relations with Generative Advers...
Taeksoo Kim
 
Generative Adversarial Network (+Laplacian Pyramid GAN)
NamHyuk Ahn
 
Clean, Learn and Visualise data with R
Barbara Fusinska
 
[GAN by Hung-yi Lee]Part 1: General introduction of GAN
NAVER Engineering
 
Generative Adversarial Networks
Mustafa Yagmur
 
Introduction to Generative Adversarial Networks
BennoG1
 
Basic Generative Adversarial Networks
Dong Heon Cho
 
Generative adversarial networks
Yunjey Choi
 
Generative adversarial text to image synthesis
Universitat Politècnica de Catalunya
 
Reading group gan - 20170417
Shuai Zhang
 
Gan seminar
San Kim
 
GAN - Theory and Applications
Emanuele Ghelfi
 
ddpg seminar
민재 정
 
ICASSP 2018 Tutorial: Generative Adversarial Network and its Applications to ...
宏毅 李
 
Deep Generative Models
Mijung Kim
 
3D Multi Object GAN
Yu Nishimura
 
Generative Adversarial Networks 2
Alireza Shafaei
 
Variational Autoencoded Regression of Visual Data with Generative Adversarial...
NAVER Engineering
 
Ad

Similar to Recommendation Engine with In-Database Machine Learning (20)

PDF
C3_W2.pdf
ShaheenKolimi
 
PDF
Graph Gurus Episode 28: In-Database Machine Learning Solution for Real-Time R...
TigerGraph
 
PDF
[한국어] Safe Multi-Agent Reinforcement Learning for Autonomous Driving
Kiho Suh
 
PDF
[系列活動] Data exploration with modern R
台灣資料科學年會
 
PPTX
Gradient descent optimizer
Hojin Yang
 
PDF
Deep Convolutional GANs - meaning of latent space
Hansol Kang
 
PDF
Gradient Boosted Regression Trees in scikit-learn
DataRobot
 
PDF
The Gremlin Graph Traversal Language
Marko Rodriguez
 
PDF
Recent Progress on Utilizing Tag Information with GANs - StarGAN & TD-GAN
Hao-Wen (Herman) Dong
 
PPTX
Feature Extraction
Revathi Krishnan
 
PPT
Generation of Deepfake images using GAN and Least squares GAN.ppt
DivyaGugulothu
 
PPTX
Leveraging R in Big Data of Mobile Ads (R在行動廣告大數據的應用)
Craig Chao
 
PDF
Building Streaming Recommendation Engines on Apache Spark with Rui Vieira
Databricks
 
PPTX
India software developers conference 2013 Bangalore
Satnam Singh
 
PDF
Matrix Factorization
Yusuke Yamamoto
 
PDF
Learning from Computer Simulation to Tackle Real-World Problems
NAVER Engineering
 
PPTX
Raccomender engines
Alessio Palma
 
PDF
A Walk in the GAN Zoo
Larry Guo
 
PDF
Hadoop France meetup Feb2016 : recommendations with spark
Modern Data Stack France
 
PDF
机器学习Adaboost
Shocky1
 
C3_W2.pdf
ShaheenKolimi
 
Graph Gurus Episode 28: In-Database Machine Learning Solution for Real-Time R...
TigerGraph
 
[한국어] Safe Multi-Agent Reinforcement Learning for Autonomous Driving
Kiho Suh
 
[系列活動] Data exploration with modern R
台灣資料科學年會
 
Gradient descent optimizer
Hojin Yang
 
Deep Convolutional GANs - meaning of latent space
Hansol Kang
 
Gradient Boosted Regression Trees in scikit-learn
DataRobot
 
The Gremlin Graph Traversal Language
Marko Rodriguez
 
Recent Progress on Utilizing Tag Information with GANs - StarGAN & TD-GAN
Hao-Wen (Herman) Dong
 
Feature Extraction
Revathi Krishnan
 
Generation of Deepfake images using GAN and Least squares GAN.ppt
DivyaGugulothu
 
Leveraging R in Big Data of Mobile Ads (R在行動廣告大數據的應用)
Craig Chao
 
Building Streaming Recommendation Engines on Apache Spark with Rui Vieira
Databricks
 
India software developers conference 2013 Bangalore
Satnam Singh
 
Matrix Factorization
Yusuke Yamamoto
 
Learning from Computer Simulation to Tackle Real-World Problems
NAVER Engineering
 
Raccomender engines
Alessio Palma
 
A Walk in the GAN Zoo
Larry Guo
 
Hadoop France meetup Feb2016 : recommendations with spark
Modern Data Stack France
 
机器学习Adaboost
Shocky1
 
Ad

More from TigerGraph (20)

PDF
MAXIMIZING THE VALUE OF SCIENTIFIC INFORMATION TO ACCELERATE INNOVATION
TigerGraph
 
PDF
Better Together: How Graph database enables easy data integration with Spark ...
TigerGraph
 
PDF
Building an accurate understanding of consumers based on real-world signals
TigerGraph
 
PDF
Care Intervention Assistant - Omaha Clinical Data Information System
TigerGraph
 
PDF
Correspondent Banking Networks
TigerGraph
 
PDF
Delivering Large Scale Real-time Graph Analytics with Dell Infrastructure and...
TigerGraph
 
PDF
Deploying an End-to-End TigerGraph Enterprise Architecture using Kafka, Maria...
TigerGraph
 
PDF
Fraud Detection and Compliance with Graph Learning
TigerGraph
 
PDF
Fraudulent credit card cash-out detection On Graphs
TigerGraph
 
PDF
FROM DATAFRAMES TO GRAPH Data Science with pyTigerGraph
TigerGraph
 
PDF
Customer Experience Management
TigerGraph
 
PDF
Graph+AI for Fin. Services
TigerGraph
 
PDF
Davraz - A graph visualization and exploration software.
TigerGraph
 
PDF
Plume - A Code Property Graph Extraction and Analysis Library
TigerGraph
 
PDF
TigerGraph.js
TigerGraph
 
PDF
GRAPHS FOR THE FUTURE ENERGY SYSTEMS
TigerGraph
 
PDF
Hardware Accelerated Machine Learning Solution for Detecting Fraud and Money ...
TigerGraph
 
PDF
How to Build An AI Based Customer Data Platform: Learn the design patterns fo...
TigerGraph
 
PDF
Machine Learning Feature Design with TigerGraph 3.0 No-Code GUI
TigerGraph
 
PDF
Supply Chain and Logistics Management with Graph & AI
TigerGraph
 
MAXIMIZING THE VALUE OF SCIENTIFIC INFORMATION TO ACCELERATE INNOVATION
TigerGraph
 
Better Together: How Graph database enables easy data integration with Spark ...
TigerGraph
 
Building an accurate understanding of consumers based on real-world signals
TigerGraph
 
Care Intervention Assistant - Omaha Clinical Data Information System
TigerGraph
 
Correspondent Banking Networks
TigerGraph
 
Delivering Large Scale Real-time Graph Analytics with Dell Infrastructure and...
TigerGraph
 
Deploying an End-to-End TigerGraph Enterprise Architecture using Kafka, Maria...
TigerGraph
 
Fraud Detection and Compliance with Graph Learning
TigerGraph
 
Fraudulent credit card cash-out detection On Graphs
TigerGraph
 
FROM DATAFRAMES TO GRAPH Data Science with pyTigerGraph
TigerGraph
 
Customer Experience Management
TigerGraph
 
Graph+AI for Fin. Services
TigerGraph
 
Davraz - A graph visualization and exploration software.
TigerGraph
 
Plume - A Code Property Graph Extraction and Analysis Library
TigerGraph
 
TigerGraph.js
TigerGraph
 
GRAPHS FOR THE FUTURE ENERGY SYSTEMS
TigerGraph
 
Hardware Accelerated Machine Learning Solution for Detecting Fraud and Money ...
TigerGraph
 
How to Build An AI Based Customer Data Platform: Learn the design patterns fo...
TigerGraph
 
Machine Learning Feature Design with TigerGraph 3.0 No-Code GUI
TigerGraph
 
Supply Chain and Logistics Management with Graph & AI
TigerGraph
 

Recently uploaded (20)

PPTX
Business Acumen Training GuidePresentation.pptx
stephvshelton19
 
PPTX
1_Introduction to advance data techniques.pptx
AshutoshDeshmukh33
 
PPTX
AI Strategy room jwfjksfksfjsjsjsjsjfsjfsj
ankitpratapsingh28
 
PPTX
Computer network topology notes for revision
BatoolRawat
 
PPT
Reliability_Chapter_ presentation 1221.5784
abobaker13
 
PPT
ISS -ESG Data flows What is ESG and HowHow
lucandthemachine
 
PDF
Foundation of Data Science unit number two notes
sanguleumeshit
 
PPTX
Database Infoormation System (DBIS).pptx
TefferiMekonnen2
 
PDF
"Python Programming for Geospatial Data Science." ...
institute of Geoinformatics and Earth Observation at PMAS ARID Agriculture University of Rawalpindi
 
PDF
168300704-gasification-ppt.pdfhghhhsjsjhsuxush
sriram270905
 
PPTX
01_intro xxxxxxxxxxfffffffffffaaaaaaaaaaafg
Vittaya Boon Suwansiri
 
PDF
.pdf is not working space design for the following data for the following dat...
jerinjoy242
 
PPTX
Introduction to Basics of Ethical Hacking and Penetration Testing -Unit No. 1...
AshutoshDeshmukh33
 
PPTX
Introduction to machine learning and Linear Models
ShivanshMehta17
 
PPTX
The THESIS FINAL-DEFENSE-PRESENTATION.pptx
hshakobe3
 
PPTX
Supervised vs unsupervised machine learning algorithms
agarwal18harsh08
 
PPTX
IBA_Chapter_11_Slides_Final_Accessible.pptx
SrikantKapoor1
 
PDF
annual-report-2024-2025 original latest.
nityanema19
 
PDF
Recruitment and Placement PPT.pdfbjfibjdfbjfobj
Alka392097
 
PDF
Mega Projects Data Mega Projects Data
saidsalah8181
 
Business Acumen Training GuidePresentation.pptx
stephvshelton19
 
1_Introduction to advance data techniques.pptx
AshutoshDeshmukh33
 
AI Strategy room jwfjksfksfjsjsjsjsjfsjfsj
ankitpratapsingh28
 
Computer network topology notes for revision
BatoolRawat
 
Reliability_Chapter_ presentation 1221.5784
abobaker13
 
ISS -ESG Data flows What is ESG and HowHow
lucandthemachine
 
Foundation of Data Science unit number two notes
sanguleumeshit
 
Database Infoormation System (DBIS).pptx
TefferiMekonnen2
 
"Python Programming for Geospatial Data Science." ...
institute of Geoinformatics and Earth Observation at PMAS ARID Agriculture University of Rawalpindi
 
168300704-gasification-ppt.pdfhghhhsjsjhsuxush
sriram270905
 
01_intro xxxxxxxxxxfffffffffffaaaaaaaaaaafg
Vittaya Boon Suwansiri
 
.pdf is not working space design for the following data for the following dat...
jerinjoy242
 
Introduction to Basics of Ethical Hacking and Penetration Testing -Unit No. 1...
AshutoshDeshmukh33
 
Introduction to machine learning and Linear Models
ShivanshMehta17
 
The THESIS FINAL-DEFENSE-PRESENTATION.pptx
hshakobe3
 
Supervised vs unsupervised machine learning algorithms
agarwal18harsh08
 
IBA_Chapter_11_Slides_Final_Accessible.pptx
SrikantKapoor1
 
annual-report-2024-2025 original latest.
nityanema19
 
Recruitment and Placement PPT.pdfbjfibjdfbjfobj
Alka392097
 
Mega Projects Data Mega Projects Data
saidsalah8181
 

Recommendation Engine with In-Database Machine Learning

  • 1. Recommendation Engine with In-Database Machine Learning Changran Liu, Mingxi Wu
  • 2. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | Outline ● In-database model training ● Latent factor recommendation model ● Distributed model training in Graph ● Demo ● GSQL implementation
  • 3. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | Traditional Model Training Pipeline training data model request results Database: ● data storage ● data update ● preprocess data Machine learning platform ● model training ● model validation Applications: ● place order ● recommendation ● ...
  • 4. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | In-situ machine learning in database: ● No need for exporting data ● Better support continuous model training over evolving data ● Less limitation on model size ● Support distributed model training In-Database Model Training request results Applications: ● recommendation ● fraud detection ● ...
  • 5. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | Movie Recommendation movie features users ratings Goals: ● Predict users' ratings for movies they haven't seen, based on previous ratings ● Recommend movies to users based on rating prediction
  • 6. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | Movie Rating Prediction (Latent factors model) Movie Alice Bob Carol Dave Love at last 5 5 0 0 Romance forever 5 ? ? 0 Cute puppies of love ? 4 0 ? Toy story ? ? ? 5 Sword vs. karate 0 0 5 ? Nonstop car chases 0 0 5 4 ● Each movie has a latent factor vector: θ(j) ● Each user has a latent factor vector: x(i) ● Predict the user j’s rating to movie i by: (θ(j) )T x(i) θ(1) = [5, 0] θ(2) = [5, 0] θ(3) = [0, 5] θ(4) = [0, 5] x(1) = [0.9, 0] x(2) = [1, 0.1] x(3) = [0.9, 0] x(4) = [0.1, 1] x(5) = [0.1, 1] x(6) = [0, 0.9] 4.5 5 4.5 0.5 0.5 0
  • 7. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | Movie Rating Prediction (Latent factors model) Movie Alice Bob Carol Dave Love at last 5 5 0 0 Romance forever 5 ? ? 0 Cute puppies of love ? 4 0 ? Toy story ? ? ? 5 Sword vs. karate 0 0 5 ? Nonstop car chases 0 0 5 4 θ(1) = [5, 0] ● Each movie has a latent factor vector: θ(j) ● Each user has a latent factor vector: x(i) ● Predict the user j’s rating to movie i by: (θ(j) )T x(i) θ(2) = [5, 0] θ(3) = [0, 5] θ(4) = [0, 5] x(1) = [0.9, 0] x(2) = [1, 0.1] x(3) = [0.9, 0] x(4) = [0.1, 1] x(5) = [0.1, 1] x(6) = [0, 0.9] action romance 4.5 5 4.5 0.5 0.5 0
  • 8. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | Cost Function RMSE regularization User 2 Movie 1 Movie 2 Movie 3 User 1 rating: y (1,1) rating: y (1,2) rating: y (2,2) rating: y (2,3) θ(1) θ(2) x(1) x(2) x(3)
  • 9. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | Model Training in Graph User 2 Movie 1 Movie 2 Movie 3 User 1 rating: y(1,1) rating: y (1,2) rating: y (2,2) rating: y(2,3) θ(1) θ(2) x(1) x(2) x(3) Phase 1: ● Collect x(i) , y(i,j) from the movies that each user rated
  • 10. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | Model Training in Graph User 2 Movie 1 Movie 2 Movie 3 User 1 rating: y(1,1) rating: y (1,2) rating: y (2,2) rating: y(2,3) [θ(1) , x(1) , y(1,1) ] [θ(1) , x(2) , y(1,2) ] x(1) x(2) x(3) [θ(2) , x(2) , y(2,2) ] [θ(2) , x(3) , y(2,3) ] Phase 1: ● Collect x(i) , y(i,j) from the movies that each user rated
  • 11. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | Model Training in Graph User 2 Movie 1 Movie 2 Movie 3 User 1 rating: y(1,1) rating: y (1,2) rating: y (2,2) rating: y(2,3) [θ(1) , x(1) , y(1,1) ] →( (θ(1) )T x(1) -y(1,1) ) x(1) [θ(1) , x(2) , y(1,2) ] →( (θ(1) )T x(2) -y(1,1) ) x(2) x(1) x(2) x(3) [θ(2) , x(2) , y(2,2) ] → ((θ(2) )T x(2) -y(2,2) ) x(2) [θ(2) , x(3) , y(2,3) ] → ((θ(2) )T x(3) -y(2,3) ) x(3) Phase 1: ● Collect x(i) , y(i,j) from the movies that each user rated ● Compute the gradient contributed by each movies
  • 12. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | Model Training in Graph User 2 Movie 1 Movie 2 Movie 3 User 1 rating: y(1,1) rating: y (1,2) rating: y (2,2) rating: y(2,3) ( (θ(1) )T x(1) -y(1,1) ) x(1) + ( (θ(1) )T x(2) -y(1,1) ) x(2) x(1) x(2) x(3) ((θ(2) )T x(2) -y(2,2) ) x(2) + ((θ(2) )T x(3) -y(2,3) ) x(3) Phase 1: ● Collect x(i) , y(i,j) from the movies that each user rated ● Compute the gradient contributed by each movies Phase 2: ● Aggregate the gradient
  • 13. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | Model Training in Graph User 2 Movie 1 Movie 2 Movie 3 User 1 rating: y(1,1) rating: y (1,2) rating: y (2,2) rating: y(2,3) x(1) x(2) x(3) Phase 1: ● Collect x(i) , y(i,j) from the movies that each user rated ● Compute the gradient contributed by each movies Phase 2: ● Aggregate the gradient ● Update the feature vector using gradient descent
  • 14. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | Training Split data Initialize latent factor vectors diff. between prediction and label converged? no finish yes update latent vectors using gradient descent (splitData.gsql) (initialization.gsql) (training.gsql)
  • 15. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | Demo
  • 16. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | GSQL Training Block USERs = SELECT s FROM USERs:s -(rate:e)-> MOVIE:t ACCUM DOUBLE prediction = dotProduct(s.@theta,t.@x), DOUBLE delta = prediction-e.rating, s.@Gradient += product(t.@x,delta), t.@Gradient += product(s.@theta,delta) POST-ACCUM s.@theta += product(s.@Gradient,-alpha), t.@x += product(t.@Gradient,-alpha); Dave Romance forever Love at last Nonstop car chases Alice rating: 5 rating: 5 rating: 0 rating: 4 θ = [1.5, 1.7] θ = [1.0, 1.5] x = [2.0, 2.3] x = [2.0, 1.3] x = [1.0, 1.3]
  • 17. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | GSQL Training Block USERs = SELECT s FROM USERs:s -(rate:e)-> MOVIE:t ACCUM DOUBLE prediction = dotProduct(s.@theta,t.@x), DOUBLE delta = prediction-e.rating, s.@Gradient += product(t.@x,delta), t.@Gradient += product(s.@theta,delta) POST-ACCUM s.@theta += product(s.@Gradient,-alpha), t.@x += product(t.@Gradient,-alpha); Dave Romance forever Love at last Nonstop car chases Alice rating: 5 rating: 5 rating: 0 rating: 4 θ = [1.5, 1.7] θ = [1.0, 1.5] x = [2.0, 2.3] x = [2.0, 1.3] x = [1.0, 1.3]
  • 18. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | GSQL Training Block USERs = SELECT s FROM USERs:s -(rate:e)-> MOVIE:t ACCUM DOUBLE prediction = dotProduct(s.@theta,t.@x), DOUBLE delta = prediction-e.rating, s.@Gradient += product(t.@x,delta), t.@Gradient += product(s.@theta,delta) POST-ACCUM s.@theta += product(s.@Gradient,-alpha), t.@x += product(t.@Gradient,-alpha); Dave Romance forever Love at last Nonstop car chases Alice rating: 5 rating: 5 rating: 0 rating: 4 θ = [1.5, 1.7] θ = [1.0, 1.5] x = [2.0, 2.3] x = [2.0, 1.3] x = [1.0, 1.3] prediction: 6.9 prediction: 5.2 prediction: 4.0 prediction: 3.0
  • 19. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | GSQL Training Block USERs = SELECT s FROM USERs:s -(rate:e)-> MOVIE:t ACCUM DOUBLE prediction = dotProduct(s.@theta,t.@x), DOUBLE delta = prediction-e.rating, s.@Gradient += product(t.@x,delta), t.@Gradient += product(s.@theta,delta) POST-ACCUM s.@theta += product(s.@Gradient,-alpha), t.@x += product(t.@Gradient,-alpha); Dave Romance forever Love at last Nonstop car chases Alice ẟ: 1.9 ẟ: 0.2 ẟ: 4.0 ẟ: -1.1 θ = [1.5, 1.7] θ = [1.0, 1.5] x = [2.0, 2.3] x = [2.0, 1.3] x = [1.0, 1.3]
  • 20. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | GSQL Training Block USERs = SELECT s FROM USERs:s -(rate:e)-> MOVIE:t ACCUM DOUBLE prediction = dotProduct(s.@theta,t.@x), DOUBLE delta = prediction-e.rating, s.@Gradient += product(t.@x,delta), t.@Gradient += product(s.@theta,delta) POST-ACCUM s.@theta += product(s.@Gradient,-alpha), t.@x += product(t.@Gradient,-alpha); Dave Romance forever Love at last Nonstop car chases Alice ẟ: 1.9 ẟ: 0.2 ẟ: 4.0 ẟ: -1.1 θ = [1.5, 1.7] grad(θ) = [4.2, 4.7] θ = [1.0, 1.5] grad(θ) = [6.9, 3.8] x = [2.0, 2.3] x = [2.0, 1.3] x = [1.0, 1.3]
  • 21. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | GSQL Training Block USERs = SELECT s FROM USERs:s -(rate:e)-> MOVIE:t ACCUM DOUBLE prediction = dotProduct(s.@theta,t.@x), DOUBLE delta = prediction-e.rating, s.@Gradient += product(t.@x,delta), t.@Gradient += product(s.@theta,delta) POST-ACCUM s.@theta += product(s.@Gradient,-alpha), t.@x += product(t.@Gradient,-alpha); Dave Romance forever Love at last Nonstop car chases Alice ẟ: 1.9 ẟ: 0.2 ẟ: 4.0 ẟ: -1.1 θ = [1.5, 1.7] grad(θ) = [4.2, 4.7] θ = [1.0, 1.5] grad(θ) = [6.9, 3.8] x = [2.0, 2.3] grad(x) = [2.9, 3.2] x = [2.0, 1.3] grad(x) = [4.3, 6.3] x = [1.0, 1.3] grad(x) = [-1.1, -1.6]
  • 22. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | GSQL Training Block USERs = SELECT s FROM USERs:s -(rate:e)-> MOVIE:t ACCUM DOUBLE prediction = dotProduct(s.@theta,t.@x), DOUBLE delta = prediction-e.rating, s.@Gradient += product(t.@x,delta), t.@Gradient += product(s.@theta,delta) POST-ACCUM s.@theta += product(s.@Gradient,-alpha), t.@x += product(t.@Gradient,-alpha); Dave Romance forever Love at last Nonstop car chases Alice ẟ: 1.9 ẟ: 0.2 ẟ: 4.0 ẟ: -1.1 θ = [1.5, 1.7] θ’ = [1.46, 1.65] θ = [1.0, 1.5] θ’ = [0.93, 1.46] x = [2.0, 2.3] x’ = [1.97, 2.27] x = [2.0, 1.3] x’ = [1.96, 1.24] x = [1.0, 1.3] x’ = [1.01, 1.32] * alpha = 0.01
  • 23. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | GSQL Training Block USERs = SELECT s FROM USERs:s -(rate:e)-> MOVIE:t ACCUM DOUBLE prediction = dotProduct(s.@theta,t.@x), DOUBLE delta = prediction-e.rating, s.@Gradient += product(t.@x,delta), t.@Gradient += product(s.@theta,delta) POST-ACCUM s.@theta += product(s.@Gradient,-alpha), t.@x += product(t.@Gradient,-alpha); Dave Romance forever Love at last Nonstop car chases Alice ẟ: 1.9 ẟ’: 1.6 ẟ: 0.2 ẟ’: -0.1 ẟ: 4.0 ẟ’: 3.6 ẟ: -1.1 ẟ’: -1.1 θ = [1.5, 1.7] θ’ = [1.46, 1.65] θ = [1.0, 1.5] θ’ = [0.93, 1.46] x = [2.0, 2.3] x’ = [1.97, 2.27] x = [2.0, 1.3] x’ = [1.96, 1.24] x = [1.0, 1.3] x’ = [1.01, 1.32]
  • 24. Q&A
  • 25. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | User-Rate-Movie Graph ● Content based method Toy story ● Disney ● ... Iron man ● Marvel ● Action ● ... Alice ● Disney fan ● Marvel fan ● ... Bob ● Marvel fan ● ... rating: 5 rating: 5 rating:4.5 rating:?
  • 26. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | User-Rate-Movie Graph ● Content based method Toy story ● Disney ● ... Iron man ● Marvel ● Action ● ... Alice ● Disney fan ● Marvel fan ● ... Bob ● Marvel fan ● ... rating: 5 rating: 5 rating:4.5 rating:? ● K-nearest neighbors
  • 27. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | User-Rate-Movie Graph ● Content based method ● K-nearest neighbors Toy story ● Disney ● ... Iron man ● Marvel ● Action ● ... Alice ● Disney fan ● Marvel fan ● ... Bob ● Marvel fan ● ... rating: 5 rating: 5 rating:? ● Latent factor (model-based)
  • 28. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | User-Rate-Movie Graph ● Content based method ● K-nearest neighbors ● Latent factor (model-based) ● Hybrid method ● ... Toy story ● Disney ● ... Iron man ● Marvel ● Action ● ... Alice ● Disney fan ● Marvel fan ● ... Bob ● Marvel fan ● ... rating: 5 rating: 5 rating:?
  • 29. | GRAPHAIWORLD.COM | #GRAPHAIWORLD | User-Rate-Movie Graph ● Content based method ● K-nearest neighbors ● Latent factor (model-based) ● Hybrid method ● ... Toy story ● Disney ● ... Iron man ● Marvel ● Action ● ... Alice ● Disney fan ● Marvel fan ● ... Bob ● Marvel fan ● ... rating: 5 rating: 5 rating:?