Social Analytics Cookbook

Understand your users and drive engagement with graph algorithms.

1. Influence & Connectivity

PageRank for Influencers

gsqlterminal
Copy
// Calculate the relative importance of users based on their follower network
CREATE QUERY user_pagerank(DOUBLE max_change, INT max_iter, DOUBLE damping) FOR GRAPH SocialGraph {
  MaxAccum<DOUBLE> @@max_diff;
  SumAccum<DOUBLE> @received_score;
  SumAccum<DOUBLE> @score = 1.0;

  V = {User.*};
  
  WHILE @@max_diff > max_change LIMIT max_iter DO
    @@max_diff = 0;
    Result = SELECT s FROM V:s -(FOLLOWS:e)-> User:t
             ACCUM t.@received_score += s.@score / s.outdegree("FOLLOWS");
             
    V = SELECT s FROM V:s
        POST_ACCUM @@max_diff += abs(s.@score - (1 - damping) + damping * s.@received_score),
                   s.@score = (1 - damping) + damping * s.@received_score,
                   s.@received_score = 0;
  END;
  
  PRINT V[V.name, V.@score] ORDER BY V.@score DESC LIMIT 100;
}

Community Detection (LPA)

gsqlterminal
Copy
// Group users into communities based on label propagation
CREATE QUERY detect_communities(INT max_iter) FOR GRAPH SocialGraph {
  MapAccum<INT, SumAccum<INT>> @neighbor_labels;
  SumAccum<INT> @label;

  V = {User.*};
  V = SELECT s FROM V:s ACCUM s.@label = getvid(s); // Initialize with unique ID
  
  WHILE i < max_iter DO
    V = SELECT s FROM V:s -(FOLLOWS:e)- User:t
        ACCUM s.@neighbor_labels += (t.@label -> 1)
        POST_ACCUM s.@label = s.@neighbor_labels.argmax().key,
                   s.@neighbor_labels = 0;
    i = i + 1;
  END;
  
  PRINT V.group_by(V.@label, count()) AS community_sizes;
}

2. Recommendation Engines

Collaborative Filtering (Users Like You)

gsqlterminal
Copy
// Recommend products that similar users have purchased
CREATE QUERY recommend_products(VERTEX<User> u, INT k) FOR GRAPH SocialGraph {
  SumAccum<INT> @common_purchases;
  SumAccum<INT> @reco_score;

  Start = {u};
  
  // Find users who bought the same items
  MyItems = SELECT t FROM Start:s -(PURCHASED:e)-> Product:t;
  SimilarUsers = SELECT t FROM MyItems:s -(REVERSE_PURCHASED:e)- User:t
                 WHERE t != u
                 ACCUM t.@common_purchases += 1;
                 
  // Recommend items bought by similar users but not by 'u'
  Recommendations = SELECT t FROM SimilarUsers:s -(PURCHASED:e)-> Product:t
                    WHERE NOT MyItems.contains(t)
                    ACCUM t.@reco_score += s.@common_purchases
                    ORDER BY t.@reco_score DESC
                    LIMIT k;
                    
  PRINT Recommendations;
}

3. Marketing Attribution

Multi-Touch Attribution Path

gsqlterminal
Copy
// Trace the marketing channels a user interacted with before a conversion
CREATE QUERY attribution_path(VERTEX<User> u) FOR GRAPH MarketingGraph {
  ListAccum<STRING> @@channel_path;
  
  Start = {u};
  
  Interactions = SELECT t FROM Start:s -(INTERACTED_WITH:e)- Campaign:t
                 ORDER BY e.timestamp ASC
                 ACCUM @@channel_path += t.channel;
                 
  PRINT @@channel_path;
}

4. Viral Growth

K-Core Decomposition

gsqlterminal
Copy
// Find the "core" group of users who are all connected to at least K others
CREATE QUERY k_core(INT k) FOR GRAPH SocialGraph {
  SumAccum<INT> @degree;
  SetAccum<VERTEX<User>> @@k_core_set;

  V = {User.*};
  V = SELECT s FROM V:s ACCUM s.@degree = s.outdegree();
  
  WHILE V.size() > 0 AND (V.min("@degree") < k) DO
    Removed = SELECT s FROM V:s WHERE s.@degree < k;
    V = V MINUS Removed;
    V = SELECT s FROM V:s -(FOLLOWS:e)- User:t
        WHERE Removed.contains(t)
        ACCUM s.@degree += -1;
  END;
  
  PRINT V;
}

[!TIP] Recommendation Engines in TigerGraph are real-time. Unlike traditional SQL-based recommendations that require pre-computation, TigerGraph can generate "Users Like You" results on-the-fly during a user's session.