Neo4j Social Graph

In a relational database, questions like 'friends of friends within 3 hops' or 'the shortest path between two people' need repeated self-JOINs that get slower as the network grows. Graph databases store each relationship directly on its nodes, so traversals stay fast no matter how connected the data is.

Screenshots from the Neo4j Browser after seeding and running the queries. Tap any figure to zoom.

1. Tap to zoom

   Fig. 01 · Graph view

   The social network as a graph

   Each circle is a User node and each arrow a FOLLOWS relationship from follower to followed. The network is accurately modelled as a directed graph.

   • A dense community sits in the middle (Alice, Bob, Carol, Eve, Frank, Dave), where many users follow each other.
   • Peripheral users such as Trent, Peggy, Mallory, and Niaj connect only through longer chains.
2. Tap to zoom

   Fig. 02 · Cypher · UNION ALL

   Validate the model: all users

   Returns every distinct name property on nodes (and, via UNION ALL, on relationships if any exist). All 15 expected users are present.

3. Tap to zoom

   Fig. 03 · Node properties

   Where users live

   Seven distinct cities. Properties stored on nodes support location-based queries such as 'find all users from Colombo' or regional recommendations.

4. Tap to zoom

   Fig. 04 · Temporal properties

   When users joined

   Join years from 2020 to 2024 show how the network grew over time and allow comparing early adopters with newer users.

• Q1 · connected users

  variable-length path

  MATCH p = (target)-[*..3]-(other) and return min(length(p)) per user: everyone reachable within 3 hops and how far away they are.

• Q2 · most connected

  degree

  Two OPTIONAL MATCHes count outgoing and incoming FOLLOWS per user, then rank by total degree (in + out).

• Q3 · shortest path

  shortestPath()

  shortestPath((a)-[:FOLLOWS*..15]-(b)) between two named users, bounded to avoid expansion blow-up.

• Recommendation query

  collaborative filtering

  Find items liked by users who share likes with User 501, exclude items they already like, and rank by how many similar users liked them.

• Q1 (Alice, 3 hops): Bob, Carol, Dave, Eve, Frank and Judy at 1 hop; Grace, Heidi, Ivan and Olivia at 2; Mallory and Niaj at 3. Only Peggy and Trent sit further out.
• Q2: Alice and Bob tie as the most connected (degree 6, each followed by 4 users), followed by Carol (5), Eve (4) and Frank (4).
• Q3 (Alice → Mallory): Alice → Eve ← Heidi ← Mallory, a 3-hop path found by treating FOLLOWS as undirected.

• Social networks: 'friends of friends within 3 hops', 'people you may know', and finding influencers or communities are direct traversals instead of repeated JOINs.
• Fraud detection: accounts, devices, IPs, cards, and transactions become nodes, so rings of accounts sharing one device or IP show up as connected clusters within a few hops.
• Recommendations: users, items, and genres linked by WATCHED, LIKED, PURCHASED, and SIMILAR_TO make 'users who liked this also liked that' a short walk through the graph.

• Nodes: users, items (movies, products, songs), and categories or genres.
• Relationships: [:VIEWED], [:PURCHASED], [:RATED] and [:LIKED] for interactions; [:BELONGS_TO_GENRE] and [:IN_CATEGORY] for content; [:SIMILAR_TO] from shared features or co-purchases.
• Collaborative filtering becomes one query: start at the user, walk to items they liked, out to other users who liked those items, then to new items, and rank by how many similar users liked each one.