Graph Gurus 23: Best Practices To Model Your Data Using A Graph Database
2 likes447 views
The document discusses various approaches to database design, particularly focusing on entity-relationship modeling and edge types in graph databases. It highlights differences between undirected and directed edges, exploring their advantages and disadvantages in representing relationships and querying data. Additionally, it provides use cases for structuring and accessing data efficiently, including hierarchical datetime structures and transaction analytics.
![User 1 Tran 1 User 2
Tran 2Trans
Attr1
Trans
Attr2
OR User 1 User 2
Edge Attribute:
LastTransactionDate:2018-01-01
FirstTransactionDate:2015-08-21
TotalTransactionAmount:12233.23
event_dates:[2018-01-01, 2018-02-01,
2015-08-21]
● Create a vertex for each transaction event.
● Connect transactions with the same
attributes via attribute vertices.
● Connect users who have transactions with
a single edge
● Aggregate historical info or use a container
to hold a set of values
Method 1: Each Event as a Vertex Method 2: Events aggregated into one
Edge](https://image.slidesharecdn.com/graphgurus23bestpractices-graphtomodelyourdata-191113225732/85/Graph-Gurus-23-Best-Practices-To-Model-Your-Data-Using-A-Graph-Database-27-320.jpg)
Graph Gurus 23: Best Practices To Model Your Data Using A Graph Database
- 2. ● ● ● ● ● ●
- 3. ● ● ●
- 4. ● ● ●
- 6. ● is a graphical approach to database design." - guru99.com ● ● ● ●
- 10. ● ● ● ● ●
- 11. ● ● ●
- 12. Graph Schema Input Data Format Query Use Case
- 15. Peter Paula is_coworker Steve iPhone X purchased_product apple fruit is_a Peter A93450 has_account
- 16. A B A can traverse to B, and B can traverse to A A B A can traverse to B, but B cannot traverse to A A B A can traverse to B via e1, B can traverse to A via e2 (e2 is automatically created upon creation of e1. e2's attributes have the same values as e1's) e2 e1 ● What is the difference between undirected edge and directed edge + reverse edge? ● What to know when making edge type choices？
- 17. A B A can traverse to B, and B can traverse to A Pros: Simple when working with undirected (symmetric) or bidirectional relationships. Example: "A friend_of B" ⇔ "B friend_of A" Cons: Does not carry directional info. A B A can traverse to B, but B cannot traverse to A Pros: Saves memory and correctly describes a direction-restricted relationship Example: "A parent_of B" ⇏ "B parent_of A" Cons: Can not traverse back from target to source. A B A can traverse to B via e1, B can traverse to A via e2 Pros: Flexibility to traverse in either designated direction. Example: e1 type is "parent_of" and e2 type is "child_of" Cons: Need to remember two edge types. e2 e1
- 18. User User_Has_Email with undirected edge: user_share_email = SELECT t FROM start-(User_Has_Email*2)-:t; with directed edge + reverse edge user_share_email = SELECT t FROM start-(User_Has_Email>)-email-(reverse_User_Has_Email>)-:t; Find users share the same email In this case, it is more concise to use undirected edge EmailOr
- 19. In this use case, the question is hard to answer by using undirected edges, since they do not provide any directional info (parent vs. child) However it can be easily solved by using directed edge + reversed edge. When querying for parent companies it can use the red edge, and when querying for child companies it can use the reverse edge (blue). c2 c1 c3 c4 c5 c6 c7 c8 c2 c1 c3 c4 c5 c6 c7 c8 Use Case: Given an enterprise graph and an input company, find its ultimate parent company and the ultimate child branches
- 20. 1. 2. 3. 4.
- 21. Person Is_Coworker Is_Relative Is_Friend Person Related_To Option 1: Person vertexes are interconnected via different edge types Pros: Efficient when traversing specific edge types, uses less memory Cons: Less concise when traversing all types, makes schema very large when having many relationship types. Person Person Person Option 2: Person vertexes are interconnected via one edge type Pros: Easy to traverse all relationships Cons: Attribute for relationship type consumes extra memory, relationship type checking is slower Person Relationship_Type (string)
- 22. Product Name Color Brand Type Price product 1 blue Apple phone 1000 product 2 red LG laptop 2000 ... ... ... ... ... Product Table Product 1 Product 2 Color: blue Brand: Apple Type: phone Price: 1000 Color: red Brand: LG Type: laptop Price: 2000 … ... … ... … ... Searching for a red product scans over ALL products
- 23. Apple LG Blue Red Phone Lapto p Product 1 Product 2 Product 3 Product 4 Product 5 Brand Color Type
- 24. Year 2018 Month Jan Month Feb Days 1 to 15 Days 1 to 15 Days 16 to 31 Days 16 to 31 Product 1 Product 2 Product 3 Product 4 Product 5 Similar to creating vertices for attributes. Hierarchical datetime structures can be created for faster time series querying speed. The levels and partitioning of each level can be customized to best suit your use case.
- 25. Year 2018 Month Jan Month Feb Days 1 to 15 Days 1 to 15 Days 16 to 31 Days 16 to 31 Product 1 Product 2 Product 3 Product 4 Product 5 Similar to creating vertices for attributes. Hierarchical datetime structures can be created for faster time series querying speed. The levels and partitioning of each level can be customized to best suit your use case. Example: Search for products that are produced in the first 15 days of Feb 2018 Traversed vertexes
- 26. Inventory 1 Car Wheel Set Tires Car Rims Rubber Belts Steel Rods ABS Module Alumi num Rods Contro ler Color White Grey Floral White Ghost White Light Grey Dim Grey NA USA CA OH SF Red wood City CANA DA BOM Graph Product Attributes Location QC
- 27. User 1 Tran 1 User 2 Tran 2Trans Attr1 Trans Attr2 OR User 1 User 2 Edge Attribute: LastTransactionDate:2018-01-01 FirstTransactionDate:2015-08-21 TotalTransactionAmount:12233.23 event_dates:[2018-01-01, 2018-02-01, 2015-08-21] ● Create a vertex for each transaction event. ● Connect transactions with the same attributes via attribute vertices. ● Connect users who have transactions with a single edge ● Aggregate historical info or use a container to hold a set of values Method 1: Each Event as a Vertex Method 2: Events aggregated into one Edge
- 28. User 1 Tran 1 User 2 Tran 2Trans Attr1 Trans Attr2 User 1 User 2 ● Create vertex for each transaction event. ● Connect transactions with same attribute via attribute vertices. Pros: Easy to do transaction analytics, such as finding transaction community and similar transactions. Able to do filtering on the transaction vertex attributes. Cons: Uses more memory, takes more steps to traverse between users. ● Connect users who had transactions with a single edge ● Aggregate historical info to edge attributes Pros: Significantly less memory usage (if without container). Takes fewer steps to traverse between users. Cons: Searching on transactions is less efficient. Slower update/insert when using a container.
- 29. Doctor Claim ClaimPatient Doctor Edge for discovered "referral" relationship Doctor Doctor Doctor Doctor Doctor Referral Network Edges from input data
- 30. Event ID IP Server Device UserId EventType Message 001 50.124.11.1 s001 dev001 u001 et1 mmmmmmm 002 50.124.11.2 s002 dev002 u002 et2 mmmmmmm ... ... ... ... ... ... ... IP user server device Event Type Event IP user server device Event Type Event IP user server device Event Type Event
- 31. IP user server device Event Type Event Event-centered schema Pros: All info of an event is in its 1-hop neighborhood. Cons: Users are 2 hops away from the device or IP she used IP user server device Event Type Event User-centered schema Pros: Easy to analyze the connectivities between the users. Cons: Events are 2 hops away from their related server and IP. It is hard to tell which IP is used for which event. Suitable use cases: 1. Starting from an input user, detect blacklisted users in k hops. 2. Given a set of blacklisted users, identify the whitelisted users similar to them. 3. Given two input users, are they connected with paths? Suitable use cases: 1. Finding communities of events 2. Finding the servers that processed the most events of a given event type 3. Finding the servers visited by a given IP
- 35. Start Free at TigerGraph Cloud Today! https://www.tigergraph.com/cloud/ Test Drive Online Demo https://www.tigergraph.com/demo Download the Developer Edition https://www.tigergraph.com/download/ Guru Scripts https://github.com/tigergraph/ecosys/tree/master/guru_scripts Join our Developer Forum https://groups.google.com/a/opengsql.org/forum/#!forum/gsql-users