Knowledge Graphs and Generative AI
0 likes1,872 views
The document discusses the integration of knowledge graphs with generative AI to enhance data retrieval and decision-making processes. It highlights the benefits of using knowledge graphs, including improved search relevance and the ability to automate complex queries, while also addressing challenges like data accuracy and bias. It emphasizes the application of these technologies in various fields, such as finance, life sciences, and customer service.
![© 2023 Neo4j, Inc. All ri hts reserved.
Graph Enrichment
13
Human-cra ted query, human-readable result
MATCH (p1:Person)-[:ENEMY]->(:Person)<-[:ENEMY]-(p2:PERSON)
MERGE (p1)-[:FRIEND]->(p2)
AI-learned ormula, machine-readable result
Predefined ormula, human-readable result
PageRank(Emil) = 13.25
PageRank(Amy) = 4.83
PageRank(Alicia) = 4.75
Node2Vec(Emil) =[5.4 5.1 2.4 4.5 3.1]
Node2Vec(Amy) =[2.8 1.8 7.2 0.9 3.0]
Node2Vec(Katie) =[1.4 5.2 4.4 3.9 3.2]
Queries
Al orithms
Embeddin s
Machine
Learnin
Workflows
Train ML models
based on results](https://image.slidesharecdn.com/katiegovgraphsummitdeck-231030144453-b0199470/85/Knowledge-Graphs-and-Generative-AI-13-320.jpg)
Knowledge Graphs and Generative AI
- 1. Knowledge Graphs and Generative AI Data Science Solutions Architect Neo4j 1 Dr. Katie Roberts
- 2. © 2023 Neo4j, Inc. All ri hts reserved. Agenda Groundin LLMs with Knowled e Graphs ● Knowled e Graphs ● Graph Data Science ● Semantic Search Neo4j Grounded Chatbot Demo 2 Katie Roberts Data Science Solution Architect
- 3. © 2023 Neo4j, Inc. All ri hts reserved. 3 LLMs Give Us an Amazing Opportunity to: Automate data retrieval tasks Improve customer service experiences Expedite readin , understandin , & summarizin 2 Generate content & code 4 1 3
- 4. © 2023 Neo4j, Inc. All ri hts reserved. Example… Q: What was the impact o Hurricane Lee? ● A: Hurricane Lee was a lon -lastin tropical cyclone that ormed in the Atlantic durin the 2017 hurricane season….. 4
- 5. © 2023 Neo4j, Inc. All ri hts reserved. 5 But There Are Challenges… Knowledge cut-off 1 Reasonable answers, not always accurate Can inherit bias throu h trainin data 2 3 Lack of enterprise domain knowledge 4 Inability to veri y or attribute sources 5
- 6. © 2023 Neo4j, Inc. All ri hts reserved. 6 How can you take advantage of this massive opportunity while overcoming these challenges?
- 7. © 2023 Neo4j, Inc. All ri hts reserved. 7 Use a Knowledge Graph to Ground LLMs What was the impact o Hurricane Lee? Hurricane Lee may impact distribution networks in the ollowin areas: MA, NY, NH, ME…
- 8. © 2023 Neo4j, Inc. All ri hts reserved. 8 Grounding with Neo4j Knowledge Graphs Connect Enrich Consume Context rich, connected view o your data that enables easier decision makin Enhance your data with raph data science, text embeddin s, and additional derived context Ground responses with in ormation and context in the raph Improve search relevance combinin vector search and raph traversals
- 9. © 2023 Neo4j, Inc. All ri hts reserved. Connected Data 9 With Knowled e Graphs
- 10. © 2023 Neo4j, Inc. All ri hts reserved. Knowledge Graphs 10 Structured Unstructured Ontolo ies Graph Al orithms and Graph Queries Semantics, Derived Relationships and Additional Context Natural Relationships
- 11. © 2023 Neo4j, Inc. All ri hts reserved. 11 What can you do with a knowledge graph? Finance How many fla ed accounts are in the applicant’s network 4+ hops out? How many lo in / account variables in common? Add these metrics to your approval process What completes the connections rom enes to diseases to tar ets? What enes can be reached 4+ hops out rom a known dru tar et? What mechanisms in common are there between two dru s? Collaborative filterin : users who bou ht X, also bou ht Y What items make you more likely to buy additional items in subsequent transactions? Traverse hierarchies-what items are similar 4+ hops out? Li e Sciences Marketin & Recommendations
- 12. © 2023 Neo4j, Inc. All ri hts reserved. Enrich 12 With Cypher Patterns, Graph Data Science & Text Embeddin s
- 13. © 2023 Neo4j, Inc. All ri hts reserved. Graph Enrichment 13 Human-cra ted query, human-readable result MATCH (p1:Person)-[:ENEMY]->(:Person)<-[:ENEMY]-(p2:PERSON) MERGE (p1)-[:FRIEND]->(p2) AI-learned ormula, machine-readable result Predefined ormula, human-readable result PageRank(Emil) = 13.25 PageRank(Amy) = 4.83 PageRank(Alicia) = 4.75 Node2Vec(Emil) =[5.4 5.1 2.4 4.5 3.1] Node2Vec(Amy) =[2.8 1.8 7.2 0.9 3.0] Node2Vec(Katie) =[1.4 5.2 4.4 3.9 3.2] Queries Al orithms Embeddin s Machine Learnin Workflows Train ML models based on results
- 14. © 2023 Neo4j, Inc. All ri hts reserved. What Are Graph Algorithms? 14
- 15. © 2023 Neo4j, Inc. All ri hts reserved. 15 Insights From Graph Algorithms Outliers, Influencers, Vulnerabilities,.. Recommendations, Homophily, Outliers,.. Recommendations, What-i Analysis, Disambi uation,.. Shortest Path, Optimal path, Route Optimization,... Link prediction, Recommendations, Next-Best Action,.. Centrality Pathfinding Community Detection Similarity Embeddings Link Prediction Dimensionality Reduction, Representation Learnin , ..
- 16. © 2023 Neo4j, Inc. All ri hts reserved. Text Embedding Vectors for Semantic Search 16 Given a question, find the most relevant documents based on a similarity metric (such as Cosine Similarity) between vector o the question and vectors o contents. Movin rom keyword search to similarity (semantic) search. Q: what is a text embeddin ? abstractId similarity 456 0.923445 22 0.892114 … ... Top K by similarity
- 17. © 2023 Neo4j, Inc. All ri hts reserved. 17 Graph Data Science Journey Find the patterns you’re looking for in connected data Identify associations, anomalies, and trends using unsupervised machine learning Learn features in your graph that you don’t even know are important yet Knowled e Graphs Graph Al orithms Graph Native ML → →
- 18. © 2023 Neo4j, Inc. All ri hts reserved. Consume 18 With LLMs and Semantic Search
- 19. © 2023 Neo4j, Inc. All ri hts reserved. 19 Semantic Search Journey Vector Similarity Search Graph Traversals & Pattern Matchin Knowled e Graph In erence & ML Find relevant documents and content or user queries Find people, places, and thin s associated to content. Identi y patterns in connected data. Further improve search relevance and insi hts by enhancin your Knowled e Graph. Use raph al orithms and ML to discover new relationships, entities, and roups. Vector Database Graph Database Neo4j Graph Data Science
- 20. © 2023 Neo4j, Inc. All ri hts reserved. 20 Knowledge Retrieval with Neo4j
- 21. © 2023 Neo4j, Inc. All ri hts reserved. RAG (Retrieval Augmented Generation) Pattern with Neo4j 21 Neo4j LLM API User Cypher Prompt + Relevant In ormation Prompt Response Relevant Results Retrieve relevant results rom Neo4j usin LLM to enerate embeddin s and/or Cypher 2 3 1 1 2 3 Combine relevant results with prompt Instruct LLM to only use the relevant results to enerate response LLM API Embeddin s and/or Cypher Generation Improved ACCURACY and RELEVANCE o responses E. . What is the impact o Hurricane Calvin? Hurricane Calvin caused minor floodin in Hawaii…. vs… 50 policyholders may be at risk o property dama e due to Hurricane Calvin.
- 22. © 2023 Neo4j, Inc. All ri hts reserved. 22 Q: Does claim X have any association with previous fraudulent parties? Q: Provide a list of claims with damage amounts >$100,000 Q: What policyholders may be impacted by Hurricane Calvin? Knowledge Retrieval with Neo4j
- 23. © 2023 Neo4j, Inc. All ri hts reserved. Demo Intro 23 LLM Chatbot
- 24. © 2023 Neo4j, Inc. All ri hts reserved. Typical Business Resilience Data Analyze business impact o ● so tware & OS vulnerabilities, ● hardware & so tware up rades, ● buildin / eo raphic disasters ● chan es to business data ormats …across mission critical applications and business locations 24 hierarchies, flows, relationships…
- 25. © 2023 Neo4j, Inc. All ri hts reserved. Actual Data Model 25 CVE Data Business Data Elements Vendors,So tware Business Tasks Application Instances Data Trans ers People/Roles Locations IT Assets
- 26. © 2023 Neo4j, Inc. All ri hts reserved. LangChain Demo Application ● Translates En lish to Cypher ● Consumption usin LLM model with ew shot promptin ● Data au mentation rom Neo4j response 26
- 27. © 2023 Neo4j, Inc. All ri hts reserved. 27 Demo
- 28. © 2023 Neo4j, Inc. All ri hts reserved. Neo4j Database 30 API Knowled e Graph Graph Data Science Graph DB Knowled e Apps In ormation Extraction & In estion Structured Unstructured Ontolo ies GCP Vertex AI Data Sources API Layer Customer Service Ticket Tria in Recommendations News Content & Discovery Enterprise Knowled e Search Patient Prioritization Clinical Decision Support Systems Pharmacovi ilance Health Assistants FAQ Bots LLMs Amazon Sa eMaker AzureML Bloom Aura Enterprise Edition LLMs API Step 1: Capture Knowled e Step2: Enrich Step 3: Semantic & Contextual Search
- 29. © 2023 Neo4j, Inc. All ri hts reserved. 31 Neo4j Enriches All Phases of an AI Ecosystem Knowled e Graphs Graph Feature En ineerin and Graph ML Graph Analytics, Investi ations and Counter actuals Inte rations and Knowled e Graphs or Heuristic AI Connect Analyze Capitalize Interpret Neo4j Neo4j GDS Neo4j Bloom Neo4j Connectors
- 30. © 2023 Neo4j, Inc. All ri hts reserved. Thank You! katie.roberts@neo4j 32