This part will review the basics of Neo4j and Cypher needed for this book. We will then explore a classic application of graph modeling: recommendation engines.
This section consists of the following chapters:
- Title Page
- Copyright and Credits
- About Packt
- Contributors
- Preface
- Section 1: Graph Modeling with Neo4j
- Graph Databases
- Graph definition and examples
- Graph theory
- A bit of history: the Seven Bridges of Königsberg problem
- Graph definition
- Visualization
- Examples of graphs
- Networks
- Road networks
- Computer networks
- Social networks
- Your data is also a graph
- Moving from SQL to graph databases
- Database models
- SQL and joins
- It's all about relationships
- Neo4j – the nodes, relationships, and properties model
- Building blocks
- Nodes
- Relationships
- Properties
- SQL to Neo4j translator
- Neo4j use cases
- Understanding graph properties
- Directed versus undirected
- Weighted versus unweighted
- Cyclic versus acyclic
- Dense versus sparse
- Graph traversal
- Connected versus disconnected
- Considerations for graph modeling in Neo4j
- Relationship orientation
- Node or property?
- Summary
- Further reading
- The Cypher Query Language
- Technical requirements
- Creating nodes and relationships
- Managing databases with Neo4j Desktop
- Creating a node
- Selecting nodes
- Filtering
- Returning properties
- Creating a relationship
- Selecting relationships
- The MERGE keyword
- Updating and deleting nodes and relationships
- Updating objects
- Updating an existing property or creating a new one
- Updating all properties of the node
- Updating node labels
- Deleting a node property
- Deleting objects
- Pattern matching and data retrieval
- Pattern matching
- Test data
- Graph traversal
- Orientation
- The number of hops
- Variable-length patterns
- Optional matches
- Using aggregation functions
- Count, sum, and average
- Creating a list of objects
- Unnesting objects
- Importing data from CSV or JSON
- Data import from Cypher
- File location
- Local file: the import folder
- Changing the default configuration to import a file from another directory
- CSV files
- CSV files without headers
- CSV files with headers
- Eager operations
- Data import from the command line
- APOC utilities for imports
- CSV files
- JSON files
- Importing data from a web API
- Setting parameters
- Calling the GitHub web API
- Summary of import methods
- Measuring performance and tuning your query for speed
- Cypher query planner
- Neo4j indexing
- Back to LOAD CSV
- The friend-of-friend example
- Summary
- Questions
- Further reading
- Empowering Your Business with Pure Cypher
- Technical requirements
- Knowledge graphs
- Attempting a definition of knowledge graphs
- Building a knowledge graph from structured data
- Building a knowledge graph from unstructured data using NLP
- NLP
- Neo4j tools for NLP
- GraphAware NLP library
- Importing test data from the GitHub API
- Enriching the graph with NLP
- Adding context to a knowledge graph from Wikidata
- Introducing RDF and SPARQL
- Querying Wikidata
- Importing Wikidata into Neo4j
- Enhancing a knowledge graph from semantic graphs
- Graph-based search
- Search methods
- Manually building Cypher queries
- Automating the English to Cypher translation
- Using NLP
- Using translation-like models
- Recommendation engine
- Product similarity recommendations
- Products in the same category
- Products frequently bought together
- Recommendation ordering
- Social recommendations
- Products bought by a friend of mine
- Summary
- Questions
- Further reading
- Section 2: Graph Algorithms
- The Graph Data Science Library and Path Finding
- Technical requirements
- Introducing the Graph Data Science plugin
- Extending Neo4j with custom functions and procedures
- The difference between procedures and functions
- Functions
- Procedures
- Writing a custom function in Neo4j
- GDS library content
- Defining the projected graph
- Native projections
- Cypher projections
- Streaming or writing results back to the graph
- Understanding the importance of shortest path algorithms through their applications
- Routing within a network
- GPS
- The shortest path within a social network
- Other applications
- Video games
- Science
- Dijkstra's shortest paths algorithm
- Understanding the algorithm
- Running Dijkstra's algorithm on a simple graph
- Example implementation
- Graph representation
- Algorithm
- Displaying the full path from A to E
- Using the shortest path algorithm within Neo4j
- Path visualization
- Understanding relationship direction
- Finding the shortest path with the A* algorithm and its heuristics
- Algorithm principles
- Defining the heuristics for A*
- Using A* within the Neo4j GDS plugin
- Discovering the other path-related algorithms in the GDS plugin
- K-shortest path
- Single Source Shortest Path (SSSP)
- All-pairs shortest path
- Optimizing processes using graphs
- The traveling-salesman problem
- Spanning trees
- Prim's algorithm
- Finding the minimum spanning tree in a Neo4j graph
- Summary
- Questions
- Further reading
- Spatial Data
- Technical requirements
- Representing spatial attributes
- Understanding geographic coordinate systems
- Using the Neo4j built-in spatial types
- Creating points
- Querying by distance
- Creating a geometry layer in Neo4j with neo4j-spatial
- Introducing the neo4j-spatial library
- A note on spatial indexes
- Creating a spatial layer of points
- Defining the spatial layer
- Adding points to a spatial layer
- Defining the type of spatial data
- Creating layers with polygon geometries
- Getting the data
- Creating the layer
- Performing spatial queries
- Finding the distance between two spatial objects
- Finding objects contained within other objects
- Finding the shortest path based on distance
- Importing the data
- Preparing the data
- Importing data
- Creating a spatial layer
- Running the shortest path algorithm
- Visualizing spatial data with Neo4j
- neomap – a Neo4j Desktop application for spatial data
- Visualizing nodes with simple layers
- Visualizing paths with advanced layer
- Using the JavaScript Neo4j driver to visualize shortest paths
- Neo4j JS driver
- Leaflet and GeoJSON
- Summary
- Questions
- Further reading
- Node Importance
- Technical requirements
- Defining importance
- Popularity and information spread
- Critical or bridging nodes
- Computing degree centrality
- Formula
- Computing degree centrality in Neo4j
- Computing the outgoing degree using GDS
- Computing the incoming degree using GDS
- Using a named projected graph
- Using an anonymous projected graph
- Understanding the PageRank algorithm
- Building the formula
- The damping factor
- Normalization
- Running the algorithm on an example graph
- Implementing the PageRank algorithm using Python
- Using GDS to assess PageRank centrality in Neo4j
- Comparing degree centrality and the PageRank results
- Variants
- ArticleRank
- Personalized PageRank
- Eigenvector centrality
- The adjacency matrix
- PageRank with matrix notation
- Eigenvector centrality
- Computing eigenvector centrality in GDS
- Path-based centrality metrics
- Closeness centrality
- Normalization
- Computing closeness from the shortest path algorithms
- The closeness centrality algorithm
- Closeness centrality in multiple-component graphs
- Betweenness centrality
- Comparing centrality metrics
- Applying centrality to fraud detection
- Detecting fraud using Neo4j
- Using centrality to assess fraud
- Creating a projected graph with Cypher projection
- Other applications of centrality algorithms
- Summary
- Exercises
- Further reading
- Community Detection and Similarity Measures
- Technical requirements
- Introducing community detection and its applications
- Identifying clusters of nodes
- Applications of the community detection method
- Recommendation engines and targeted marketing
- Clusters of products
- Clusters of users
- Fraud detection
- Predicting properties or links
- A brief overview of community detection techniques
- Detecting graph components and visualizing communities
- Weakly connected components
- Strongly connected components
- Writing the GDS results in the graph
- Visualizing a graph with neovis.js
- Using NEuler, the Graph Data Science Playground
- Usage for community detection visualization
- Running the Label Propagation algorithm
- Defining Label Propagation
- Weighted nodes and relationships
- Semi-supervised learning
- Implementing Label Propagation in Python
- Using the Label Propagation algorithm from the GDS
- Using seeds
- Writing results to the graph
- Understanding the Louvain algorithm
- Defining modularity
- All nodes are in their own community
- All nodes are in the same community
- Optimal partition
- Steps to reproduce the Louvain algorithm
- The Louvain algorithm in the GDS
- Syntax
- The aggregation method in relationship projection
- Intermediate steps
- A comparison between Label Propagation and Louvain on the Zachary's karate club graph
- Going beyond Louvain for overlapping community detection
- A caveat of the Louvain algorithm
- Resolution limit
- Alternatives to Louvain
- Overlapping community detection
- Dynamic networks
- Measuring the similarity between nodes
- Set-based similarities
- Overlapping
- Definition
- Quantifying user similarity in the GitHub graph
- Jaccard similarity
- Vector-based similarities
- Euclidean distance
- Cosine similarity
- Summary
- Questions
- Further reading
- Section 3: Machine Learning on Graphs
- Using Graph-based Features in Machine Learning
- Technical requirements
- Building a data science project
- Problem definition – asking the right question
- Supervised versus unsupervised learning
- Regression versus classification
- Introducing the problem for this chapter
- Getting and cleaning data
- Data characterization
- Quantifying the dataset size
- Labels
- Columns
- Data visualization
- Data cleaning
- Outliers detection
- Missing data
- Correlation between variables
- Data enrichment
- Feature engineering
- Building the model
- Train/test split and cross-validation
- Creating the train and test samples with scikit-learn
- Training a model
- Evaluating model performances
- The steps toward graph machine learning
- Building a (knowledge) graph
- Creating relationships from existing data
- Creating relationships from relational data
- Creating relationships from Neo4j
- Using an external data source
- Importing the data into Neo4j
- Graph characterization
- The number of nodes and edges
- The number of components
- Extracting graph-based features
- Using graph-based features with pandas and scikit-learn
- Extracting graph-based features from Neo4j Browser
- Creating the projected graph
- Running one or several algorithms
- Dropping the projected graph
- Extracting the data
- Automating graph-based feature creation with the Neo4j Python driver
- Discovering the Neo4j Python driver
- Basic usage
- Transactions
- Automating graph-based feature creation with Python
- Creating the projected graph
- Calling the GDS procedures
- Writing results back to the graph
- Dropping the projected graph
- Exporting the data from Neo4j to pandas
- Training a scikit-learn model
- Introducing community features
- Using both community and centrality features
- Summary
- Questions
- Further reading
- Predicting Relationships
- Technical requirements
- Why use link prediction?
- Dynamic graphs
- Applications
- Recovering missing data
- Fighting crime
- Research
- Making recommendations
- Social links (Facebook friends, LinkedIn contacts...)
- Product recommendations
- Making recommendations using a link prediction algorithm
- Creating link prediction metrics with Neo4j
- Community-based metrics
- Path-related metrics
- Distance between nodes
- The Katz index
- Using local neighborhood information
- Common neighbors
- Adamic-Adar
- Total neighbors
- Preferential attachment
- Other metrics
- Building a link prediction model using an ROC curve
- Importing the data into Neo4j
- Splitting the graph and computing the score for each edge
- Measuring binary classification model performance
- Understanding ROC curves
- Extracting features and labels
- Drawing the ROC curve
- Creating the DataFrame
- Plotting the ROC curve
- Determining the optimal cutoff and computing performances
- Building a more complex model using scikit-learn
- Saving link prediction results into Neo4j
- Predicting relationships in bipartite graphs
- Summary
- Questions
- Further reading
- Graph Embedding - from Graphs to Matrices
- Technical requirements
- Why do we need embedding?
- Why is embedding needed?
- One-hot encoding
- Creating features for words – the manual way
- Embedding specifications
- The graph embedding landscape
- Adjacency-based embedding
- The adjacency matrix and graph Laplacian
- Eigenvectors embedding
- Locally linear embedding
- Similarity-based embedding
- High-Order Proximity preserved Embedding (HOPE)
- Computing node embedding with Python
- Creating a networkx graph
- The Neo4j test graph
- Extracting the edge list data from Neo4j
- Creating a networkx graph matrix from pandas
- Fitting a node embedding algorithm
- Extracting embeddings from artificial neural networks
- Artificial neural networks in a nutshell
- A reminder about neural network principles
- Neurons, layers, and forward propagation
- Different types of neural networks
- Skip-graph model
- Fake task
- Input
- Word representation before embedding
- Target
- Hidden layer
- Output layer
- DeepWalk node embedding
- Generating node context through random walks
- Generating random walks from the GDS
- DeepWalk embedding with karateclub
- Node2vec, a DeepWalk alternative
- Node2vec from the GDS (≥ 1.3)
- Getting the embedding results from Python
- Graph neural networks
- Extending the principles of CNNs and RNNs to build GNNs
- Message propagation and aggregation
- Taking into account node properties
- Applications of GNNs
- Image analysis
- Video analysis
- Zero-shot learning
- Text analysis
- And there's more...
- Using GNNs in practice
- GNNs from the GDS – GraphSAGE
- Going further with graph algorithms
- State-of-the-art graph algorithms
- Summary
- Questions
- Further reading
- Section 4: Neo4j for Production
- Using Neo4j in Your Web Application
- Technical requirements
- Creating a full-stack web application using Python and Graph Object Mappers
- Toying with neomodel
- Defining the properties of structured nodes
- StructuredNode versus SemiStructuredNode
- Adding properties
- Creating nodes
- Querying nodes
- Filtering nodes
- Integrating relationship knowledge
- Simple relationship
- Relationship with properties
- Building a web application backed by Neo4j using Flask and neomodel
- Creating toy data
- Login page
- Creating the Flask application
- Adapting the model
- The login form
- The login template
- The login view
- Reading data – listing owned repositories
- Altering the graph – adding a contribution
- Understanding GraphQL APIs by example – GitHub API v4
- Endpoints
- Returned attributes
- Query parameters
- Mutations
- Developing a React application using GRANDstack
- GRANDstack – GraphQL, React, Apollo, and Neo4j Database
- Creating the API
- Writing the GraphQL schema
- Defining types
- Starting the application
- Testing with the GraphQL playground
- Calling the API from Python
- Using variables
- Mutations
- Building the user interface
- Creating a simple component
- Getting data from the GraphQL API
- Writing a simple component
- Adding navigation
- Mutation
- Refreshing data after the mutation
- Summary
- Questions
- Further reading
- Neo4j at Scale
- Technical requirements
- Measuring GDS performance
- Estimating memory usage with the estimate procedures
- Estimating projected graph memory usage
- Fictive graph
- Graph defined by native or Cypher projection
- Estimating algorithm memory usage
- The stats running mode
- Measuring time performances for some of the algorithms
- Configuring Neo4j 4.0 for big data
- The landscape prior to Neo4j 4.0
- Memory settings
- Neo4j in the cloud
- Sharding with Neo4j 4.0
- Defining shards
- Creating the databases
- Querying a sharded graph
- The USE statement
- Querying all databases
- Summary
- Other Books You May Enjoy
-
No Comment
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.