La strada verso il successo con i database a grafo, la Graph Data Science e l’AI generativa

La strada verso il successo
con i database a grafo, la
Graph Data Science e l’AI
generativa
Marco Bessi, PreSale Engineer
Neo4j Inc. All rights reserved 2023

Your friendly instructor
Neo4j
Italy
PreSale
Engineer
PhD
@PoliMI

Neo4j

Pioneered the
graph database
in 2002
4 Neo4j Inc. All rights reserved 2023

Neo4j is a DB
Causal
clustering
ACID
transactions
High
availability
Binary
& HTTP
protocol
Ofﬁcial
drivers

Neo4j is a GRAPH DB
Causal
clustering
ACID
transactions
High
availability
Binary
& HTTP
protocol
Ofﬁcial
drivers
Native
graph DB
Property
graph
model
Schema
free
Index-free
adjacency
Cypher

Labeled property graph
model components
• Nodes
- Represent objects in the graph
• Relationships
- Relate nodes by type and direction
• Properties
- Name-value pairs that can go
on nodes and relationships
- Can have indexes and composite
indexes
• Labels
- Group nodes & shape the domain

Hybrid workload duality
Intelligent
Operational
Systems
Graph Transactions,
Storage & Querying
Built for operational and analytical workloads
Better
Predictions
for Analytics
Graph Analytics, ML,
& Data Science

Graph Data Science

GDS evolution
Local
Matching
Use embeddings to learn
the features in your
graph that you don’t even
know are important yet.
Train in-graph supervise
ML models to predict
links, labels and missing
data.
Global
Patterns
Graph
Representations
Use unsupervised
machine learning
techniques to identify
associations, anomalies,
and trends.
Graph analytics
Graph feature
engineering
Find the patterns
you’re looking for in
connected data.
Knowledge graphs

65+ Graph Data Science Techniques in Neo4j
Pathfinding &
Search
• Shortest Path
• Single-Source Shortest Path
• All Pairs Shortest Path
• A* Shortest Path
• Yen’s K Shortest Path
• Minimum Weight Spanning Tree
• K-Spanning Tree (MST)
• Random Walk
• Breadth & Depth First Search
Centrality &
Importance
• Degree Centrality
• Closeness Centrality
• Harmonic Centrality
• Betweenness Centrality & Approx.
• PageRank
• Personalized PageRank
• ArticleRank
• Eigenvector Centrality
• Hyperlink Induced Topic Search (HITS)
• Influence Maximization (Greedy, CELF)
Community
Detection
• Triangle Count
• Local Clustering Coefficient
• Connected Components (Union Find)
• Strongly Connected Components
• Label Propagation
• Louvain Modularity
• K-1 Coloring
• Modularity Optimization
• Speaker Listener Label Propagation
Supervised
Machine Learning
• Node Classification
• Link Prediction
… and more!
Heuristic Link
Prediction
• Adamic Adar
• Common Neighbors
• Preferential Attachment
• Resource Allocations
• Same Community
• Total Neighbors
Similarity
• Node Similarity
• K-Nearest Neighbors (KNN)
• Jaccard Similarity
• Cosine Similarity
• Pearson Similarity
• Euclidean Distance
• Approximate Nearest Neighbors (ANN)
Graph
Embeddings
• Node2Vec
• FastRP
• FastRPExtended
• GraphSAGE
• Synthetic Graph Generation
• Scale Properties
• Collapse Paths
• One Hot Encoding
• Split Relationships
• Graph Export
• Pregel API (write your own algos)

Where is graph data science applied
Predictive
Maintenance
Churn
Prediction
Fraud
Detection
Life Sciences
Personalized
Recommendations
Cybersecurity
Disambiguation &
Segmentation
GenAI and KG

Generative AI &
Knowledge Graphs

In a world of inﬁnite
content, knowledge
becomes valuable
Denny
Vrandečić
[WikiData]
[Semantic MediaWiki]
[Wikifunctions]

$6.6 Trillion
In Economic Value
Generative AI is Predicted to Unlock
Up to 3.3% productivity improvement annually

Generative AI
A branch of artiﬁcial intelligence that
focuses on creating models and
algorithms capable of generating new
and original content.
ChatGPT is a well known example of a
generative AI.
A type of Generative AI that is trained on
vast amounts of content.
Currently seen as the “GenAI for
language/text”.
ChatGPT is a Large Language Model.
Large Language Model

LLMs give us an amazing opportunity to:
Automate data
retrieval tasks 1
3
2
4
Expedite reading,
understanding, &
summarizing
Improve customer
service experiences
Generate content
& code

Save time and money
Improve growth and retention:
Customer
Operations 1
3
2
4
Software Engineering
Marketing & Sales
R&D
75% of GenAI value will come from four areas

But there are challenges:
1
3
2
4
5
Knowledge
cut-off
Can inherit bias
through training data
Reasonable answers,
not always accurate
Lack of enterprise
domain knowledge
Inability to verify or
attribute sources

In summary
Learns random sentences from
random people
Talks like a person but doesn’t
really understand what it’s saying
Occasionally speaks absolute
non sense
Is a cute little bird

How can you take advantage of
this massive opportunity while
overcoming these challenges?

LLM + Graphs
I want to use
LLM’s in my
Graph for…

LLM + Graphs
Front-end
Natural language
to cypher
I want to use
LLM’s in my
Graph for…

LLM + Graphs
Back-end
KG-enhanced
LLMs
Front-end
Natural language
to cypher
I want to use
LLM’s in my
Graph for…

How to help LLM do better?
Fine tuning
Few-shot
learning
Grounding
Provide additional
training data to better
tune GenAI to your
use case
Provide completed
examples “shots” to
the AI as context in
prompts. a.k.a
In-Context Learning
Provide AI with the
information to use for
generating responses
Neo4j as the data
source for Grounding

Ground LLMs in Neo4j’s Knowledge Graph
Retrieval Augmented
Generation (RAG):
retrieve data from outside a
foundation model and augment
your prompts by adding the
relevant retrieved data in
context

Neo4j’s Knowledge Graph
LLM + Neo4j’s
Knowledge
Graph
Improve
accuracy
Deploy with
conﬁdence
Unlock
innovation

Combine the power of LLM with the stored data of
your knowledge graph for more accurate responses
and to reduce hallucinations.
Improve Accuracy

LLMs can help generate more accurate responses by considering the connections
and dependencies within the graph and mapping new links as new data is identiﬁed.
Flexible schema means it is
easy to grow your knowledge
base whenever new
information is available
Relationships are data that is
used to return explicit results
Accuracy
Vector search adds implicit
results showcasing similar
responses

Knowledge retrieval with Neo4j
CALL db.index.vector.queryNodes('products', 5, $embedding)
CALL apoc.ml.openai.embedding([$question], $apiKey)
CALL apoc.ml.openai.chat([{role:'system', content: $system},
{role:'user', content: $user}], $apiKey)
1
2
3

Add a layer of context
over your LLM for
accuracy and
specificity
Neo4j knowledge graphs supply the LLM with information about your company so answers
are specific to your business, giving more context for more accurate responses.
Specificity

Ensure your grounding-partner meets security,
scalability, and regulatory requirements so you
can deploy enterprise-wide with conﬁdence.
Deploy with Conﬁdence

Deﬁne policies by role or identity
Integrates with identity and access
management provider with SSO
LLM retrieves and returns information governed by your enterprise security and
access control policies-down to the node level.
Security & Privacy
Build constraints on nodes, labels,
relationships, properties, speciﬁc
parts of the graph, and even
traversal depth

Easily scale with autonomous
clustering
Reliability
Neo4j knowledge graphs scale and are battle-tested to thousands of concurrent users,
get answers quickly with incredibly fast query speeds.
Incredibly fast traversals with index
free adjacency

Add metadata or annotations
Map relationships between
search results and data source
nodes
Represent data sources as nodes
Verify the enriched responses from your LLM because each piece of information is
linked to its sources and origins.
Explainability

Adopt innovation and iterate as your LLM
implementations evolve with full interoperability
across all cloud and data providers.
Unlock Innovation

Any cloud, any data provider
Graph Data Science
BI & VISUALIZATIONS
INGEST
STORE
PROCESS
Apache
Kafka
MACHINE LEARNING
Cloud
Functions
Neo4j Bloom
PubSub
DataProc
Analytics
Feature
Engineering
Data
Exploration
Graph
Data
Science
Business
Applications &
Existing Systems
Files (unstructured,
structured)
TensorFlow
KNIME Python
Cloud Storage
AWS
Lambda
Graph Database

Ground LLMs to improve accuracy and explainability with Neo4j’s
enterprise-ready knowledge graph built on a ﬂexible schema that
integrates seamlessly with your GenAI tech platform.
Stay Grounded with Neo4j
Improve Accuracy Deploy with Conﬁdence Unlock Innovation

Nhow Milano
Via Tortona 35, Milano
Graph Talk Milano
Giovedi, 5 Ottobre 2023
9.00-13.30

Thank you
Marco Bessi, PreSale Engineer, marco.bessi@neo4j.com

Support

Better predictions with data you already have
● Traditional ML ignore network structure because it’s difﬁcult to extract
● Add graphy data to existing ML pipelines to increase accuracy, or
● Graphs use relationships to unlock otherwise unattainable predictions
Machine Learning Pipeline

Why do data science in Neo4j (VS Python/R )?
1
2
3
Performance: Specialized data structure; Good concurrency;
In-memory optimization & compression productions-ready.
Easy of use: Consistent syntax, including Cypher;
Comprehensive library (65+ vetted algorithms);
Persistence, every stage of the data science process
can be stored/not stored.
Product support: GDS / AuraDS is supported
by Neo4j, well documented, with a large
team of support engineers, and a sizeable
user community.

Neo4j should be the database for Grounding
Vector DB Limitations
Knowledge Graph
Strengths
Neo4j Differentiators
Similarity ≠ Relevance or
Accuracy
Black-Box (Sub-Symbolic)
Duplicate & incomplete
results
Missing reference information
Challenging to answer
multi-hop questions
Difﬁcult for SME to correct
Relevancy beyond just
similarity
Transparent symbolic
representation
Condensed information
storage
References between
documents calculated before
query time
Enables human correction
LLMs understand Cypher
Vectors + Cypher
Index for many data types
(numeric, geopoints, dates)
Fine-grained security and
access control
ACID transactions,
high-availability, and scale
Ecosystem integration
Available on all clouds
Graph Data Science for
enhanced ML

New Neo4j features
From Neo4j v5.11
1
2
APOC OpenAI API:
CALL apoc.ml.openai.embedding(['Some Text'], $apiKey, {})
CALL apoc.ml.openai.completion('Question', $apiKey, {config}) yield
value;
CALL apoc.ml.query("Question") yield value, query
...
Vector index & search:
Index nodes on ﬂoat or double array properties.
Based on Lucene which uses Hierarchical Navigable Small-World graphs (HNSW).
Created and queried using procedures.

Creation
CALL db.index.vector.createNodeIndex(
)
indexName, // Name of the index to create (STRING)
label, // Label of nodes to index (STRING)
propertyKey, // Property key to index (STRING)
dimensions, // Dimensionality of vectors to index (INTEGER)
similarityFunction // “EUCLIDEAN” or “COSINE”
(case-insensitive STRING)

Querying
CALL db.index.vector.queryNodes(
)
indexName, // Name of the index to query (STRING)
k, // Number of neighbors to query for (INTEGER)
vector // Query vector whose neighbors we are searching for
(LIST<FLOAT>)

Writing vector efﬁciently
CALL db.create.setVectorProperty(
)
node, // Nodes to add the property to (NODE)
propertyKey, // Property key to write to (STRING)
vector, // Property value to write (LIST<FLOAT>)

La strada verso il successo con i database a grafo, la Graph Data Science e l’AI generativa

Recommended

Recommended

More Related Content

Similar to La strada verso il successo con i database a grafo, la Graph Data Science e l’AI generativa

Similar to La strada verso il successo con i database a grafo, la Graph Data Science e l’AI generativa (20)

More from Neo4j

More from Neo4j (20)

Recently uploaded

Recently uploaded (20)

La strada verso il successo con i database a grafo, la Graph Data Science e l’AI generativa