Shift Remote: AI: Smarter AI with analytical graph databases - Victor Lee (Tigergraph)
1 like82 views
The document discusses the advantages of analytical graph databases, particularly in the context of AI and machine learning. It highlights the ability of graph databases to connect disparate datasets, enhance insights, and provide real-time analytics across various industries such as healthcare and finance. Key features include superior query performance, schema flexibility, and support for complex algorithms that improve decision-making and operational efficiency.
![© 2020. ALL RIGHTS RESERVED. | TIGERGRAPH.COM | CONFIDENTIAL INFORMATION | 11
Graph-Based Structural Similarity
Use a vertex's neighbors as its feature set
● Cosine: Use edge weights to each neighboring vertex
A
OrderY
B
ZW X
2
3 1
2
14
A's weighted neighbors = {4,1,2,0}
B's weighted neighbors = {0,2,3,1}
Cos(A,B) = 8 / [√21√14] = 0.4666
W,X,Y,Z represent feature vertices, different vertex types than A,B](https://image.slidesharecdn.com/victorlee-smarteraiwithanalyticalgraphdatabasestigergraph-201215141409/85/Shift-Remote-AI-Smarter-AI-with-analytical-graph-databases-Victor-Lee-Tigergraph-11-320.jpg)
![© 2020 TigerGraph. All Rights Reserved
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](https://image.slidesharecdn.com/victorlee-smarteraiwithanalyticalgraphdatabasestigergraph-201215141409/85/Shift-Remote-AI-Smarter-AI-with-analytical-graph-databases-Victor-Lee-Tigergraph-19-320.jpg)
Shift Remote: AI: Smarter AI with analytical graph databases - Victor Lee (Tigergraph)
- 1. Smarter AI with Analytical Graph Databases 17 December 2020 Victor Lee Head of Product Strategy, TigerGraph 1
- 2. © 2020. ALL RIGHTS RESERVED. | TIGERGRAPH.COM | CONFIDENTIAL INFORMATION | 2 “Graph analysis is possibly the single most effective competitive differentiator for organizations pursuing data-driven operations and decisions after the design of data capture.”
- 3. © 2020. ALL RIGHTS RESERVED. | TIGERGRAPH.COM | CONFIDENTIAL INFORMATION | AGENDA ● What is an Analytical Graph Database? ● Integrated Knowledge, More Insight ● Three Basic Approaches for Graph + AI, with Use-Case Examples ○ Unsupervised Learning ○ Feature Enrichment from Graph Features ○ In-Database Learning 3
- 4. © 2020. ALL RIGHTS RESERVED. | TIGERGRAPH.COM | CONFIDENTIAL INFORMATION | Richer, Smarter Data ● Connections-as-data ● Connects different datasets, breaks down silos Deeper, Smarter Questions ● Look for semantic patterns of relationship ● Search far & wide more easily & faster than other DBs More Computational Options ● Graph algorithms ● Graph-enhanced machine learning Explainable Results ● Semantic data model, queries, and answers ● Visual exploration and results 4 Graph+AI Delivers More Value, Better Results Customer Supplier Location 2 Product Payment PURCHASED RESIDESSHIPS TO PURCHASED SHIPS FROM A C C EPTED MAKES Location 1 N O TIFIES Order KNOWS
- 5. © 2020. ALL RIGHTS RESERVED. | TIGERGRAPH.COM | CONFIDENTIAL INFORMATION | Types of Graph Databases 5 Transactional Analytics Property Graphs: • Higher performance for queries, transactions, and advanced analytics • Pattern matching • Schema-free or Schema-based • Schema-based allows application-specific tuning Semantic (RDF) Knowledge Graph: • Collection of facts (RDF triples) • Ontology to model concepts & rules • Pattern matching • Logical inference • Standards-based TigerGraph is a High-Performance and Scalable Property Graph, for both Analytics & Transactions.
- 6. © 2020. ALL RIGHTS RESERVED. | TIGERGRAPH.COM | CONFIDENTIAL INFORMATION | Real-World Better Outcomes from Graph+AI Healthcare: Real-time recommendation ● 1.3TB graph brain ● Real-time care recommendations ● Saving $150M/yr, more 6 Industrial Supply Chain: Analytics for decisions ● Analytics: weeks→ minutes ● Reveal opportunities, optimize tactical & strategic decisions ● Saving $25M+/yr Financial Services: Real-time fraud detection ● Integrates multiple tools ● "Magical" real-time visual results for investigators ● Scalable for growth
- 7. © 2020. ALL RIGHTS RESERVED. | TIGERGRAPH.COM | CONFIDENTIAL INFORMATION | Types of Graph Algorithms ● Path Finding ● Clustering / Community Detection ○ Lenient clustering - connected component: one connection ○ Strict clustering - clique detection: every possible connection ○ Relative density - more connections in-group than between-group ● Ranking and Centrality ○ PageRank, HITS ○ SimRank, RoleSim ● Frequent Pattern Discovery ○ Agglomerative search Case 1: Analytical Queries & Graph Algorithms 7 BOLD indicates more complex tasks, with iterative algorithms, which can be considered unsupervised learning
- 8. © 2020. ALL RIGHTS RESERVED. | TIGERGRAPH.COM | CONFIDENTIAL INFORMATION | 8 How do I find the most influential provider in each region for a particular medical condition? Whole-Graph Compute problem 1. Analyze claims data to identify referral relationships among providers (Time Series Analysis) 2. Create subsets of claims around each condition with a group of healthcare codes (e.g. CPT codes) for each region (e.g. local healthcare market) 3. Utilize PageRank to score hubs within each market Dr. Thomas Condition: Diabetes Healthcare Market: S. San Jose, CA Hub Identified: Dr. Thomas
- 9. © 2020. ALL RIGHTS RESERVED. | TIGERGRAPH.COM | CONFIDENTIAL INFORMATION | 9 Who is influenced by these leaders (e.g. other doctors, chiropractors, physical therapists, facilities)? Utilize Community Detection 1. Identify communities of providers around each hub for each region and for a specific condition 2. Track changes over time to detect significant shifts in communities Dr. Thomas Condition: Diabetes Healthcare Market: S. San Jose, CA Hub Identified: Dr. Thomas Community Detected: Diabetes – S. San Jose – Dr. Thomas
- 10. © 2020. ALL RIGHTS RESERVED. | TIGERGRAPH.COM | CONFIDENTIAL INFORMATION | 10 Find Similar Cases to deliver better healthcare ● Seamlessly integrate multiple sources of data to provide unified and comprehensive view for each journey among 50M Medicare members ● Find similar members with a click of a button in real-time ● Deliver care path recommendations for similar members
- 11. © 2020. ALL RIGHTS RESERVED. | TIGERGRAPH.COM | CONFIDENTIAL INFORMATION | 11 Graph-Based Structural Similarity Use a vertex's neighbors as its feature set ● Cosine: Use edge weights to each neighboring vertex A OrderY B ZW X 2 3 1 2 14 A's weighted neighbors = {4,1,2,0} B's weighted neighbors = {0,2,3,1} Cos(A,B) = 8 / [√21√14] = 0.4666 W,X,Y,Z represent feature vertices, different vertex types than A,B
- 12. © 2020. ALL RIGHTS RESERVED. | TIGERGRAPH.COM | CONFIDENTIAL INFORMATION | Case 2: Graph Feature Extraction Challenge Find and report fraudsters among billions of calls per week. Solution ● Build graph: Real-time operational graph with 600M phone nodes & 15B call detail records. ● Get features and labels: Domain experts write GSQL queries to extract 118 features/phone. Some past calls are labeled for 3 types of unwanted calls. ● Train: Feed machine learning with training data for fraud detection with 118 features/phone for 30M calls. ● Deploy: For each incoming call, extract the current 118 features (subsecond) and apply model for real-time answer. Results ● If unwanted call is predict, display alert on recipient's phone ● Process 2000+ calls/sec ● Improved customer satisfaction 12 Customer: China Mobile
- 13. © 2020. ALL RIGHTS RESERVED. | TIGERGRAPH.COM | CONFIDENTIAL INFORMATION | Case 2: Graph Feature Extraction 13
- 14. © 2020. ALL RIGHTS RESERVED. | TIGERGRAPH.COM | CONFIDENTIAL INFORMATION | Using Graph-Based Features for Machine Learning In Healthcare: Good Doctor - Bad Doctor 14
- 15. © 2020. ALL RIGHTS RESERVED. | TIGERGRAPH.COM | CONFIDENTIAL INFORMATION | Powering Explainable AI with Graph Database Additional details at https://www.tigergraph.com/solutions/ai-and-machine-learning/ 15
- 16. © 2020. ALL RIGHTS RESERVED. | TIGERGRAPH.COM | CONFIDENTIAL INFORMATION | Case 3: In-Database Machine Learning 16 ● Native graph storage and PG model ● Coded once, auto scale-out & scale-up ● Real-time updates ● GSQL Turing-complete language ○ Preprocess data ○ Training: flow-control, accumulator, pattern match ○ Model validation Applications: ● Entity resolution ● Recommendation ● Fraud detection ● ... Training data GSQL ML Algorithm TRAIN DEPLOY Model Prediction Query
- 17. © 2020 TigerGraph. All Rights Reserved In Database ML for Movie Recommendation movie features users ratings Goals: ● Predict users' ratings for movies, based on previous ratings ● Recommend movies to users based on rating prediction
- 18. © 2020 TigerGraph. All Rights Reserved User-Rating-Movie Graph Recommendation Approaches ● Collaborative filtering ● Content based method ● K-nearest neighbors ● Latent factor (model-based) ● Hybrid method ● ... rating: 5 rating: 5 rating:? Toy story ● Disney ● ... Iron man ● Marvel ● Action ● ... Alice ● Disney fan ● Marvel fan ● ... Bob ● Marvel fan ● ... MovieLens dataset https://grouplens.org/datasets/movielens/ ● 100K ratings and 40K tags that 1K users gave to 17K movies ● Ratings are from 0 to 5 stars
- 19. © 2020 TigerGraph. All Rights Reserved 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
- 20. © 2020. ALL RIGHTS RESERVED. | TIGERGRAPH.COM | CONFIDENTIAL INFORMATION | The Future of In-Graph ML Neural Networks for Graphs 20 https://medium.com/@terngoodod/a-comprehensive-survey-on-graph-neural-networks-part-1-types-of-graph-neural-network-1dd93b823c70
- 21. © 2020 TigerGraph. All Rights Reserved Basic Conventional Neural Network Input layer = FeatureMatrix X = FeatureVector * Nodes HiddenLayer H1 = Activation_Function(P * X) HiddenLayer H2 = Activation_Function(P *H1) 21 Propagation P = weighted edges
- 22. © 2020 TigerGraph. All Rights Reserved Graph Convolutional Network Input layer = FeatureMatrix X = FeatureVector * Nodes HiddenLayer H1 = Activation_Function(A*X*P), A = Adjacency Matrix (Edges) HiddenLayer H2 = Activation_Function(A*H1*P) 22 Propagation P = weighted edges
- 23. © 2020. ALL RIGHTS RESERVED. | TIGERGRAPH.COM | CONFIDENTIAL INFORMATION | The TigerGraph Difference Feature Design Difference Benefit Real-Time Deep-Link Querying ● Native Graph design ● C++ engine, for high performance ● Storage Architecture ● Uncovers hard-to-find patterns ● Operational, real-time ● HTAP: Transactions+Analytics Handling Massive Scale ● Distributed DB architecture ● Massively parallel processing ● Compressed storage reduces footprint and messaging ● Integrates all your data ● Automatic partitioning ● Elastic scaling of resource usage In-Database Analytics ● GSQL: High-level yet Turing-complete language ● User-extensible graph algorithm library, runs in-DB ● ACID (OLTP) and Accumulators (OLAP) ● Avoids transferring data ● Richer graph context ● In-DB machine learning 5 to 10+ hops deep 23
- 24. © 2020 TigerGraph. All Rights Reserved Summary for "Why Graph for ML/AI?" 24 ● Natural Data Model - Graph is how we think ● Richer data - connections between entities, graph-based features ● Graphs have always had a natural role in machine learning: ○ Unsupervised learning through graph algorithms, frequent pattern mining ○ Learning through neural networks and deep learning ● Graph data models are uniquely qualified to provide explanatory AI. ● Native Graphs with Massively Parallel Processing like TigerGraph enable large scale feature extraction and in-graph analytics
- 25. © 2020. ALL RIGHTS RESERVED. | TIGERGRAPH.COM | CONFIDENTIAL INFORMATION | 25 Q & A Dr. Victor Lee Head of Product Strategy & Developer Relations victor@tigergraph.com Thank You