© 2023 Neo4j, Inc. All rights reserved.
Nouveautés
produits
Nicolas Rouyer, Field Engineer, Neo4j
© 2023 Neo4j, Inc. All rights reserved.
Agenda
Nouvelles fonctionnalités Neo4j 5
2
© 2023 Neo4j, Inc. All rights reserved.
Nouveautés Neo4j 5
Synthèse
3
© 2023 Neo4j, Inc. All rights reserved.
Neo4j 5 -Téléchargement
4
https://neo4j.com/download-center/
© 2023 Neo4j, Inc. All rights reserved.
NEO4J 5
Performances, Passage à l’échelle,
agilité
5
Performances
Élargies
Scalabilité
illimitée
sans effort
Flexibilité
Totale
des
Opérations
Répondre + vite à
des requêtes
+ complexes
Scalabilité H/V
+ automatique
-de coûts
d’infrastructure
Déployer,
exécuter et
exploiter Neo4j
simplement
et partout
© 2023 Neo4j, Inc. All rights reserved.
Schéma du graphe
Nouvelles contraintes sur les nœuds,
les relations et les propriétés :
● Unique relationship property,
● Relationship key,
● Property data types
NOUVELLES FONCTIONNALITÉS NEO4J 5
Améliorations Base de
Données
6
© 2023 Neo4j, Inc. All rights reserved.
Graph Pattern Matching
“La REGEX de la traversée”
Amélioration de l'expressivité de la navigation
dans les graphes avec les
Quantified Path Patterns (QPP).
Syntaxe + puissante et + performante pour
naviguer et parcourir votre graphe..
NOUVELLES FONCTIONNALITÉS NEO4J 5
Améliorations Base de
Données
7
© 2023 Neo4j, Inc. All rights reserved.
NOUVELLES FONCTIONNALITÉS NEO4J 5
Améliorations Base de
Données
Exemple Graph Pattern Matching
8
MATCH path=(a:Account)-[rel:TRANSACTION*3..6]->(a)
WHERE size(apoc.coll.toSet(nodes(path))) = size(nodes(path)) - 1
AND ALL(idx in range(0, size(rel)-2)
WHERE (rel[idx]).date < (rel[idx+1]).date
AND (rel[idx+1].amount / rel[idx].amount) * 100 <= 20
)
RETURN path
Avant-QPP
© 2023 Neo4j, Inc. All rights reserved.
NOUVELLES FONCTIONNALITÉS NEO4J 5
Améliorations Base de
Données
Exemple Graph Pattern Matching
9
QPP
MATCH path=(a:Account)-[:PERFORMS]->(first_tx)
((tx_i)-[:BENEFITS_TO]->(a_i)-[:PERFORMS]->(tx_j)
WHERE tx_i.date < tx_j.date
AND 0.80 <= tx_i.amount / tx_j.amount <= 1.00
){3,6}
(last_tx)-[:BENEFITS_TO]->(a)
WHERE size(apoc.coll.toSet([a]+a_i)) = size([a]+a_i)
RETURN path
© 2023 Neo4j, Inc. All rights reserved.
Parallel
Runtime
10
© 2023 Neo4j, Inc. All rights reserved.
NOUVELLES FONCTIONNALITÉS NEO4J 5
Parallel Runtime : requêtes analytiques
accélérées x100
11
© 2023 Neo4j, Inc. All rights reserved.
NOUVELLES FONCTIONNALITÉS NEO4J 5
Parallel Runtime : requêtes analytiques
accélérées x100
12
+ de cores
Requêtes
+ rapides
© 2023 Neo4j, Inc. All rights reserved.
Change Data
Capture (CDC)
13
© 2023 Neo4j, Inc. All rights reserved.
NOUVELLES FONCTIONNALITÉS NEO4J 5
CDC : Suivi automatisé en temps réel des
modifications
14
© 2023 Neo4j, Inc. All rights reserved.
Perspectives :
Auto-Sharding
15
© 2023 Neo4j, Inc. All rights reserved.
NOUVELLES FONCTIONNALITÉS NEO4J 5
Auto-Sharding Fabric
16
A venir 2024+
Mapping
Règles de
Sharding
● Exploiter les indications du client pour créer une
stratégie de répartition
● Permettre la création de schéma/métadonnées pour le
sharding
● Augmenter automatiquement le nombre de shards en
fonction de la configuration du client
● Créer un planner Fabric pour transférer les données
et les requêtes vers les bons shards
● Prendre en charge les transactions distribuées et les
proxy nodes
● Rendre disponible l’auto-sharding dans Aura
© 2023 Neo4j, Inc. All rights reserved.
Neo4j 5
Détails
17
© 2023 Neo4j, Inc. All rights reserved.
18
XP
Développeur
Data Scientist
Langage
et
Modèle
Performances
en
lecture/expor
t
Neo4j 5
Performances
en
écriture/import
Passage à l’
échelle et
Haute Dispo
Exploitation
Cloud Native
Gestion de
versions et
Support
Administratio
n
Exploitation
et
Observabilité
Sécurité
© 2023 Neo4j, Inc. All rights reserved.
XP Développeur/Data Scientist
19
• Python driver export API (Aura)
• Export Neo4j result sets to Pandas dataframes
• Python driver support for asyncio
• ‘Retryable’ flag on Error objects (Aura)
• Query String and Parameters in Go driver results objects renamed
• Procedures can specify analyzer in Full-text queries (Aura)
• Notifications (Aura)
• Updates severity on some of the existing notifications
© 2023 Neo4j, Inc. All rights reserved.
Langage et Modèle
20
• New Element ID for graph objects (Aura)
ID to uniquely identify the object and the database it is from.
• Graph Pattern Matching-relationship and label patterns (Aura)
(part of the Graph Pattern Matching long-term effort in Cypher)
• Additions to Cypher:
Underscores in numeric literals, count, exists(), isNaN
• APOC 5 Compatible Plugin (Aura)
Extended has been moved out to a community supported library
• Traversal API
Improved documentation and examples
© 2023 Neo4j, Inc. All rights reserved.
Graph Pattern Matching
21
Neo4j 5
Simpler alternative syntax to navigate and traverse your
graph
- Disjunction (v:L1|L2)
- Negation (v:!L1)
- Grouping (v: (L1&L2)|(L3&!L4))
- Repetition ()* // 0..n ()+ //1..n
- Conjunction (v:L1&L2)
© 2023 Neo4j, Inc. All rights reserved.
Performances en lecture/export
22
• Faster TEXT indexes with Trigram analyzer (Aura)
• Include Lists in FULLTEXT indexes (Aura)
• New RANGE and POINT indexes (Aura)
• Auto-parameterize partial parameterized queries (Aura)
• Faster K-Hop Queries in Cypher (Aura)
• Slotted Runtime for Community Edition
© 2023 Neo4j, Inc. All rights reserved.
K-Hop Query optimization
23
In Neo4j 5: breadth first
Before: depth first
Benefit: orders of magnitude faster!
K-Hop query: variable length patterns.
Ex: MATCH (:LabelA)-[:RELTYPE*1..5]->(:LabelB)
© 2023 Neo4j, Inc. All rights reserved.
Performances en écriture/import
24
• neo4j-admin import can now perform incremental
load
into an existing database
• CALL … {} … IN TRANSACTION subquery support
writes
in PIPELINED Runtime (Aura)
• No Eager for DETACH DELETE (n) in CALL IN
TRANSACTIONS (Aura)
© 2023 Neo4j, Inc. All rights reserved.
Passage à l’échelle et Haute Disponibilité
25
• Autonomous Clustering (Enterprise)
The new clustering infrastructure decouples
servers from databases, improving scalability and
cloud-readiness.
• Create Database from URI
• COMPOSITE databases (Enterprise)
Fabric technology is used to improve the setting
and management of sharded and federated
databases with dynamic compositions of databases
© 2023 Neo4j, Inc. All rights reserved.
Exploitation Cloud Native
26
• Immutable privileges (Aura)
Cloud operators can use sidecar design pattern to control
RBAC-based permissions.
• Server-side routing enabled by default (Aura)
Simplify operations in VPC, behind firewalls & network appliances
• Docker & Cloud Marketplace Publishing
Neo4j 4.4 and 5 published as a single image in AWS, GCP and Azure.
• ARM and Corretto JVM on AWS
Recommend AWS Corretto JVM for consistent write throughput
© 2023 Neo4j, Inc. All rights reserved.
Gestion de versions et Support
27
• Neo4j 5 lifecycle and release cadence
• New supported platforms, require JDK 17 and Scala upgrade
• Any-to-any Neo4j 5.X version and patch release
• Neo4j 4.4 to Neo4j 5.X migration
• Improved eval/ commercial messaging
• Db.tx_state.memory_allocation
• Reorder transaction log file list in debug.log
• Cypher-shell Java Driver Logging
• Transaction log file rotation and retention policy change
© 2023 Neo4j, Inc. All rights reserved.
Neo4j 5 Continuous Release
Support Model
28
5.LTS.0
Fixes ONLY
Fixes
Frequent
5.0.0
5.LTS+n.0
Features
LTS in June 2024
Final Features Released
End of Support Nov
2027
Upgrade to the latest version to receive latest fixes from Support
Frequent (~monthly) releases for features and fixes
© 2023 Neo4j, Inc. All rights reserved.
Administration, Exploitation et Observabilité
29
• neo4j-admin refresh
Improved features, more control and consistent set of arguments for
the admin
• Configuration settings refresh
Settings have been reviewed, old and unnecessary settings have been
removed. A more consistent namespace structure has been introduce.
Settings in neo4j.conf have now strict validation.
• Improved procedures to check and set config values
• Fully integrated Log4j
The logging framework is fully integrated with Log4j, provides better
control of all the database logs.
© 2023 Neo4j, Inc. All rights reserved.
Administration, Exploitation et
Observabilité
30
• Cypher-shell improvements
Access available when there is no leader, plus auto completion for
commands.
• SHOW/TERMINATE TRANSACTION improvements (Aura)
More information and control about running and terminating
transaction with the CALL … IN TRANSACTIONS
• Neo4j Operations Manager (Enterprise)
The new UI based tool to monitor, administer and operate Neo4j 4.4
and 5.x.
• Enhanced Backup (Enterprise)
Backup provides new compressed file format, an API, differential
backup, ‘restore until’ capabilities
© 2023 Neo4j, Inc. All rights reserved.
Sécurité
31
• Execute and ExecuteBoosted privilege review (Enterprise)
Refresh of permissions for the execution of admin procedures
• Cypher-Shell supports impersonation (Enterprise)
© 2023 Neo4j, Inc. All rights reserved.
Futur moteur de stockage au format block
32
MATCH (:Person {name:
'Bob'})-[r:LIKES]->(:Person {name:
'Alice'})
© 2023 Neo4j, Inc. All rights reserved.
Neo4j Vector Index
● Les Neo4j vector indexes utilisent Apache Lucene (search, indexing).
● Lucene implémente l’algorithme Hierarchical Navigable Small World (HNSW) pour effectuer une
requête k~approximate nearest neighbors (k-ANN) sur les propriétés vectorielles.
CALL db.index.vector.createNodeIndex('abstract-embeddings', 'Abstract',
'embedding', 1536, 'cosine');
MATCH (title:Title)<--(:Paper)-->(abstract:Abstract)
WHERE toLower(title.text) = 'efficient and robust approximate nearest
neighbor search using hierarchical navigable small world graphs'
CALL db.index.vector.queryNodes('abstract-embeddings', 10, abstract.embedding)
YIELD node AS similarAbstract, score
MATCH (similarAbstract)<--(:Paper)-->(similarTitle:Title)
RETURN similarTitle.text AS title, score;
33
© 2023 Neo4j, Inc. All rights reserved.
Intégration
Google Dataflow
34
© 2023 Neo4j, Inc. All rights reserved.
Intégration GCP DataFlow
35
https://neo4j.com/blog/loading-data-neo4j-google-cloud-dataflow/
https://cloud.google.com/dataflow/docs/guides/templates/provided/google-cloud-to-neo4j?hl=fr
© 2023 Neo4j, Inc. All rights reserved.
Nouveautés
Graph Data Science
36
Edge embeddings
Découvrir les relations
manquantes
● Users generate edge embeddings
using DGL, PyKeen, or PyTorch
Geometric
● Bring them into Neo4j Graph Data
Science or AuraDS stored as a
property.
● Use similarity functions to identify
possible relationship candidates
and check node pairs.
● Create relationships
Help predict missing
relationships between entities
Infer new relationships
Improve semantic
understanding and reasoning
Neo4j Inc. All rights reserved 2023
37
How it Helps You
Longest Path Algo
Find the path of maximum
length between two nodes
in a directed acyclic graph
Neo4j Inc. All rights reserved 2023
Longest Path
Once a directed acyclic graph
is built, use either Graph Data
Science or the Python client to
call the Longest Path
Algorithm.
Find the path of maximum
length between two nodes in a
graph.
Help solve critical scheduling and
optimization problems by
understanding longest paths
How it Helps You
Topological Sort Algo
Sort nodes in a directed
acyclic graph by
traversing them in
order of dependency
Identify dependencies
Neo4j Inc. All rights reserved 2023
Identify
Dependencies
Once a directed acyclic graph
is built, use either Graph Data
Science or the Python client to
call the topological sort
algorithm.
Find dependencies in the
graph.
Ensure that dependent tasks are
performed in the correct order
How it Helps You
© 2023 Neo4j, Inc. All rights reserved.
Rappels LLM+Neo4j
42
Neo4j et l’IA Générative
Neo4j Inc. All rights reserved 2023
1
2
3
L'IA générative adaptée à votre organisation
Utilisez Neo4j pour adapter l’IAGen à votre entreprise,
améliorer la précision et réduire les erreurs.
Démarrer un Graphe de Connaissances avec les LLMs
Utiliser LLMs + Neo4j pour l'extraction d'entités/relations
et la création de graphes de connaissances
Partenariats poussés
avec les fournisseurs cloud
pour l'IA générative
Neo4j Fiabilise les LLMs
44
© 2023 Neo4j, Inc. All rights reserved.
https://data.world/blog/generative-
ai-benchmark-increasing-the-accur
acy-of-llms-in-the-enterprise-with-
a-knowledge-graph/
Neo4j
Fiabilise
les LLMs
45
© 2023 Neo4j, Inc. All rights reserved.
Merci!
nicolas.rouyer@neo4j.com
46

Autodesk Netfabb Ultimate 2025 free crack

  • 1.
    © 2023 Neo4j,Inc. All rights reserved. Nouveautés produits Nicolas Rouyer, Field Engineer, Neo4j
  • 2.
    © 2023 Neo4j,Inc. All rights reserved. Agenda Nouvelles fonctionnalités Neo4j 5 2
  • 3.
    © 2023 Neo4j,Inc. All rights reserved. Nouveautés Neo4j 5 Synthèse 3
  • 4.
    © 2023 Neo4j,Inc. All rights reserved. Neo4j 5 -Téléchargement 4 https://neo4j.com/download-center/
  • 5.
    © 2023 Neo4j,Inc. All rights reserved. NEO4J 5 Performances, Passage à l’échelle, agilité 5 Performances Élargies Scalabilité illimitée sans effort Flexibilité Totale des Opérations Répondre + vite à des requêtes + complexes Scalabilité H/V + automatique -de coûts d’infrastructure Déployer, exécuter et exploiter Neo4j simplement et partout
  • 6.
    © 2023 Neo4j,Inc. All rights reserved. Schéma du graphe Nouvelles contraintes sur les nœuds, les relations et les propriétés : ● Unique relationship property, ● Relationship key, ● Property data types NOUVELLES FONCTIONNALITÉS NEO4J 5 Améliorations Base de Données 6
  • 7.
    © 2023 Neo4j,Inc. All rights reserved. Graph Pattern Matching “La REGEX de la traversée” Amélioration de l'expressivité de la navigation dans les graphes avec les Quantified Path Patterns (QPP). Syntaxe + puissante et + performante pour naviguer et parcourir votre graphe.. NOUVELLES FONCTIONNALITÉS NEO4J 5 Améliorations Base de Données 7
  • 8.
    © 2023 Neo4j,Inc. All rights reserved. NOUVELLES FONCTIONNALITÉS NEO4J 5 Améliorations Base de Données Exemple Graph Pattern Matching 8 MATCH path=(a:Account)-[rel:TRANSACTION*3..6]->(a) WHERE size(apoc.coll.toSet(nodes(path))) = size(nodes(path)) - 1 AND ALL(idx in range(0, size(rel)-2) WHERE (rel[idx]).date < (rel[idx+1]).date AND (rel[idx+1].amount / rel[idx].amount) * 100 <= 20 ) RETURN path Avant-QPP
  • 9.
    © 2023 Neo4j,Inc. All rights reserved. NOUVELLES FONCTIONNALITÉS NEO4J 5 Améliorations Base de Données Exemple Graph Pattern Matching 9 QPP MATCH path=(a:Account)-[:PERFORMS]->(first_tx) ((tx_i)-[:BENEFITS_TO]->(a_i)-[:PERFORMS]->(tx_j) WHERE tx_i.date < tx_j.date AND 0.80 <= tx_i.amount / tx_j.amount <= 1.00 ){3,6} (last_tx)-[:BENEFITS_TO]->(a) WHERE size(apoc.coll.toSet([a]+a_i)) = size([a]+a_i) RETURN path
  • 10.
    © 2023 Neo4j,Inc. All rights reserved. Parallel Runtime 10
  • 11.
    © 2023 Neo4j,Inc. All rights reserved. NOUVELLES FONCTIONNALITÉS NEO4J 5 Parallel Runtime : requêtes analytiques accélérées x100 11
  • 12.
    © 2023 Neo4j,Inc. All rights reserved. NOUVELLES FONCTIONNALITÉS NEO4J 5 Parallel Runtime : requêtes analytiques accélérées x100 12 + de cores Requêtes + rapides
  • 13.
    © 2023 Neo4j,Inc. All rights reserved. Change Data Capture (CDC) 13
  • 14.
    © 2023 Neo4j,Inc. All rights reserved. NOUVELLES FONCTIONNALITÉS NEO4J 5 CDC : Suivi automatisé en temps réel des modifications 14
  • 15.
    © 2023 Neo4j,Inc. All rights reserved. Perspectives : Auto-Sharding 15
  • 16.
    © 2023 Neo4j,Inc. All rights reserved. NOUVELLES FONCTIONNALITÉS NEO4J 5 Auto-Sharding Fabric 16 A venir 2024+ Mapping Règles de Sharding ● Exploiter les indications du client pour créer une stratégie de répartition ● Permettre la création de schéma/métadonnées pour le sharding ● Augmenter automatiquement le nombre de shards en fonction de la configuration du client ● Créer un planner Fabric pour transférer les données et les requêtes vers les bons shards ● Prendre en charge les transactions distribuées et les proxy nodes ● Rendre disponible l’auto-sharding dans Aura
  • 17.
    © 2023 Neo4j,Inc. All rights reserved. Neo4j 5 Détails 17
  • 18.
    © 2023 Neo4j,Inc. All rights reserved. 18 XP Développeur Data Scientist Langage et Modèle Performances en lecture/expor t Neo4j 5 Performances en écriture/import Passage à l’ échelle et Haute Dispo Exploitation Cloud Native Gestion de versions et Support Administratio n Exploitation et Observabilité Sécurité
  • 19.
    © 2023 Neo4j,Inc. All rights reserved. XP Développeur/Data Scientist 19 • Python driver export API (Aura) • Export Neo4j result sets to Pandas dataframes • Python driver support for asyncio • ‘Retryable’ flag on Error objects (Aura) • Query String and Parameters in Go driver results objects renamed • Procedures can specify analyzer in Full-text queries (Aura) • Notifications (Aura) • Updates severity on some of the existing notifications
  • 20.
    © 2023 Neo4j,Inc. All rights reserved. Langage et Modèle 20 • New Element ID for graph objects (Aura) ID to uniquely identify the object and the database it is from. • Graph Pattern Matching-relationship and label patterns (Aura) (part of the Graph Pattern Matching long-term effort in Cypher) • Additions to Cypher: Underscores in numeric literals, count, exists(), isNaN • APOC 5 Compatible Plugin (Aura) Extended has been moved out to a community supported library • Traversal API Improved documentation and examples
  • 21.
    © 2023 Neo4j,Inc. All rights reserved. Graph Pattern Matching 21 Neo4j 5 Simpler alternative syntax to navigate and traverse your graph - Disjunction (v:L1|L2) - Negation (v:!L1) - Grouping (v: (L1&L2)|(L3&!L4)) - Repetition ()* // 0..n ()+ //1..n - Conjunction (v:L1&L2)
  • 22.
    © 2023 Neo4j,Inc. All rights reserved. Performances en lecture/export 22 • Faster TEXT indexes with Trigram analyzer (Aura) • Include Lists in FULLTEXT indexes (Aura) • New RANGE and POINT indexes (Aura) • Auto-parameterize partial parameterized queries (Aura) • Faster K-Hop Queries in Cypher (Aura) • Slotted Runtime for Community Edition
  • 23.
    © 2023 Neo4j,Inc. All rights reserved. K-Hop Query optimization 23 In Neo4j 5: breadth first Before: depth first Benefit: orders of magnitude faster! K-Hop query: variable length patterns. Ex: MATCH (:LabelA)-[:RELTYPE*1..5]->(:LabelB)
  • 24.
    © 2023 Neo4j,Inc. All rights reserved. Performances en écriture/import 24 • neo4j-admin import can now perform incremental load into an existing database • CALL … {} … IN TRANSACTION subquery support writes in PIPELINED Runtime (Aura) • No Eager for DETACH DELETE (n) in CALL IN TRANSACTIONS (Aura)
  • 25.
    © 2023 Neo4j,Inc. All rights reserved. Passage à l’échelle et Haute Disponibilité 25 • Autonomous Clustering (Enterprise) The new clustering infrastructure decouples servers from databases, improving scalability and cloud-readiness. • Create Database from URI • COMPOSITE databases (Enterprise) Fabric technology is used to improve the setting and management of sharded and federated databases with dynamic compositions of databases
  • 26.
    © 2023 Neo4j,Inc. All rights reserved. Exploitation Cloud Native 26 • Immutable privileges (Aura) Cloud operators can use sidecar design pattern to control RBAC-based permissions. • Server-side routing enabled by default (Aura) Simplify operations in VPC, behind firewalls & network appliances • Docker & Cloud Marketplace Publishing Neo4j 4.4 and 5 published as a single image in AWS, GCP and Azure. • ARM and Corretto JVM on AWS Recommend AWS Corretto JVM for consistent write throughput
  • 27.
    © 2023 Neo4j,Inc. All rights reserved. Gestion de versions et Support 27 • Neo4j 5 lifecycle and release cadence • New supported platforms, require JDK 17 and Scala upgrade • Any-to-any Neo4j 5.X version and patch release • Neo4j 4.4 to Neo4j 5.X migration • Improved eval/ commercial messaging • Db.tx_state.memory_allocation • Reorder transaction log file list in debug.log • Cypher-shell Java Driver Logging • Transaction log file rotation and retention policy change
  • 28.
    © 2023 Neo4j,Inc. All rights reserved. Neo4j 5 Continuous Release Support Model 28 5.LTS.0 Fixes ONLY Fixes Frequent 5.0.0 5.LTS+n.0 Features LTS in June 2024 Final Features Released End of Support Nov 2027 Upgrade to the latest version to receive latest fixes from Support Frequent (~monthly) releases for features and fixes
  • 29.
    © 2023 Neo4j,Inc. All rights reserved. Administration, Exploitation et Observabilité 29 • neo4j-admin refresh Improved features, more control and consistent set of arguments for the admin • Configuration settings refresh Settings have been reviewed, old and unnecessary settings have been removed. A more consistent namespace structure has been introduce. Settings in neo4j.conf have now strict validation. • Improved procedures to check and set config values • Fully integrated Log4j The logging framework is fully integrated with Log4j, provides better control of all the database logs.
  • 30.
    © 2023 Neo4j,Inc. All rights reserved. Administration, Exploitation et Observabilité 30 • Cypher-shell improvements Access available when there is no leader, plus auto completion for commands. • SHOW/TERMINATE TRANSACTION improvements (Aura) More information and control about running and terminating transaction with the CALL … IN TRANSACTIONS • Neo4j Operations Manager (Enterprise) The new UI based tool to monitor, administer and operate Neo4j 4.4 and 5.x. • Enhanced Backup (Enterprise) Backup provides new compressed file format, an API, differential backup, ‘restore until’ capabilities
  • 31.
    © 2023 Neo4j,Inc. All rights reserved. Sécurité 31 • Execute and ExecuteBoosted privilege review (Enterprise) Refresh of permissions for the execution of admin procedures • Cypher-Shell supports impersonation (Enterprise)
  • 32.
    © 2023 Neo4j,Inc. All rights reserved. Futur moteur de stockage au format block 32 MATCH (:Person {name: 'Bob'})-[r:LIKES]->(:Person {name: 'Alice'})
  • 33.
    © 2023 Neo4j,Inc. All rights reserved. Neo4j Vector Index ● Les Neo4j vector indexes utilisent Apache Lucene (search, indexing). ● Lucene implémente l’algorithme Hierarchical Navigable Small World (HNSW) pour effectuer une requête k~approximate nearest neighbors (k-ANN) sur les propriétés vectorielles. CALL db.index.vector.createNodeIndex('abstract-embeddings', 'Abstract', 'embedding', 1536, 'cosine'); MATCH (title:Title)<--(:Paper)-->(abstract:Abstract) WHERE toLower(title.text) = 'efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs' CALL db.index.vector.queryNodes('abstract-embeddings', 10, abstract.embedding) YIELD node AS similarAbstract, score MATCH (similarAbstract)<--(:Paper)-->(similarTitle:Title) RETURN similarTitle.text AS title, score; 33
  • 34.
    © 2023 Neo4j,Inc. All rights reserved. Intégration Google Dataflow 34
  • 35.
    © 2023 Neo4j,Inc. All rights reserved. Intégration GCP DataFlow 35 https://neo4j.com/blog/loading-data-neo4j-google-cloud-dataflow/ https://cloud.google.com/dataflow/docs/guides/templates/provided/google-cloud-to-neo4j?hl=fr
  • 36.
    © 2023 Neo4j,Inc. All rights reserved. Nouveautés Graph Data Science 36
  • 37.
    Edge embeddings Découvrir lesrelations manquantes ● Users generate edge embeddings using DGL, PyKeen, or PyTorch Geometric ● Bring them into Neo4j Graph Data Science or AuraDS stored as a property. ● Use similarity functions to identify possible relationship candidates and check node pairs. ● Create relationships Help predict missing relationships between entities Infer new relationships Improve semantic understanding and reasoning Neo4j Inc. All rights reserved 2023 37 How it Helps You
  • 38.
    Longest Path Algo Findthe path of maximum length between two nodes in a directed acyclic graph Neo4j Inc. All rights reserved 2023
  • 39.
    Longest Path Once adirected acyclic graph is built, use either Graph Data Science or the Python client to call the Longest Path Algorithm. Find the path of maximum length between two nodes in a graph. Help solve critical scheduling and optimization problems by understanding longest paths How it Helps You
  • 40.
    Topological Sort Algo Sortnodes in a directed acyclic graph by traversing them in order of dependency Identify dependencies Neo4j Inc. All rights reserved 2023
  • 41.
    Identify Dependencies Once a directedacyclic graph is built, use either Graph Data Science or the Python client to call the topological sort algorithm. Find dependencies in the graph. Ensure that dependent tasks are performed in the correct order How it Helps You
  • 42.
    © 2023 Neo4j,Inc. All rights reserved. Rappels LLM+Neo4j 42
  • 43.
    Neo4j et l’IAGénérative Neo4j Inc. All rights reserved 2023 1 2 3 L'IA générative adaptée à votre organisation Utilisez Neo4j pour adapter l’IAGen à votre entreprise, améliorer la précision et réduire les erreurs. Démarrer un Graphe de Connaissances avec les LLMs Utiliser LLMs + Neo4j pour l'extraction d'entités/relations et la création de graphes de connaissances Partenariats poussés avec les fournisseurs cloud pour l'IA générative
  • 44.
  • 45.
    © 2023 Neo4j,Inc. All rights reserved. https://data.world/blog/generative- ai-benchmark-increasing-the-accur acy-of-llms-in-the-enterprise-with- a-knowledge-graph/ Neo4j Fiabilise les LLMs 45
  • 46.
    © 2023 Neo4j,Inc. All rights reserved. Merci! nicolas.rouyer@neo4j.com 46