GeminiDB Redis API for Product Correlation Analysis

Application Scenarios

The correlation can be analyzed based on products added to the shopping cart of a user. The results are crucial for the e-commerce industry and can be used to analyze user's shopping habits. For example, merchants can:

• Recommend related items to the user who is browsing on the details page of a specific item.
• Recommend related items to a user who just added an item to the shopping cart.
• Display highly correlated items together.

Data Structure Design

• Product browse records
  Store IDs of products that users have browsed in a set of a Redis instance.

  • Key: user:<user_id>:viewed
  • Value: set of product IDs

  Example:

  User 123 browsed products 1001, 1002, and 1003.

  user:123:viewed -> {1001, 1002, 1003}

• Purchase records
  Store IDs of products that users have purchased in a set.

  • Key: user:<user_id>:purchased
  • Value: set of product IDs

  Example:

  User 123 purchased products 1001 and 1005.

  user:123:purchased -> {1001, 1005}

• Product correlation
  Store co-occurrence times of products in a zset (sorted set) of a Redis instance.

  • Key: product:<product_id>:related
  • Value: IDs of other related products and co-occurrence times

  Example:

  Other product ID related to product 1001 and their co-occurrence times

  product:1001:related -> {1002:5, 1003:3, 1005:2}

Data Collection

• User browse behavior

Each time a user browses a product, add its ID to the user's browse records. The following is an example of C++ code:

void recordView(int user_id, int product_id) {
    std::cout << "User ID " << user_id << " has viewed Product ID " << product_id << std::endl;
    redis.sadd("user:" + std::to_string(user_id) + ":viewed", std::to_string(product_id));
} 

• User purchase behavior

Each time a user purchases a product, add its ID to the user's purchase records. The following is an example of C++ code:

void recordPurchase(int user_id, int product_id) {
    std::cout << "User ID " << user_id << " has purchased Product ID " << product_id << std::endl;
    redis.sadd("user:" + std::to_string(user_id) + ":purchased", std::to_string(product_id));
} 

Query API

The following is an example of C++ code for querying other products most related to a specified product based on its ID:

std::vector<std::pair<std::string, double>> getRelatedProducts(int product_id) {
    std::vector<std::pair<std::string, double>> result;
    redis.zrevrange("product:" + std::to_string(product_id) + ":related", 0, -1, std::back_inserter(result));
    return result;
} 

Complete Example of C++ Code

The following is a complete example of C++ code, which is implemented using C++ Redis client redis++.

#include <iostream>
#include <iterator>
#include <set>
#include <vector>
#include <string>
#include <utility>
 
#include "sw/redis++/redis++.h"
 
using namespace sw::redis;
 
auto redis = Redis("tcp://127.0.0.1:6379");
 
void recordView(int user_id, int product_id) {
    std::cout << "User ID " << user_id << " has viewed Product ID " << product_id << std::endl;
    redis.sadd("user:" + std::to_string(user_id) + ":viewed", std::to_string(product_id));
}
 
void recordPurchase(int user_id, int product_id) {
    std::cout << "User ID " << user_id << " has purchased Product ID " << product_id << std::endl;
    redis.sadd("user:" + std::to_string(user_id) + ":purchased", std::to_string(product_id));
}
 
void updateRelatedProducts(int user_id) {
    std::set<std::string> viewed_products;
    redis.smembers("user:" + std::to_string(user_id) + ":viewed", std::inserter(viewed_products, viewed_products.end()));
 
    std::set<std::string> purchased_products;
    redis.smembers("user:" + std::to_string(user_id) + ":purchased", std::inserter(purchased_products, purchased_products.end()));
 
    for (const auto& product_id : viewed_products) {
        for (const auto& related_product_id : viewed_products) {
            if (product_id != related_product_id) {
              //Sets the browse weight to 1.
                redis.zincrby("product:" + product_id + ":related", 1, related_product_id);
            }
        }
    }
 
    for (const auto& product_id : purchased_products) {
        for (const auto& related_product_id : purchased_products) {
            if (product_id != related_product_id) {
             //Higher weight of purchase behavior
                redis.zincrby("product:" + product_id + ":related", 2, related_product_id);
            }
        }
    }
}
 
std::vector<std::pair<std::string, double>> getRelatedProducts(int product_id) {
    std::vector<std::pair<std::string, double>> result;
    redis.zrevrange("product:" + std::to_string(product_id) + ":related", 0, -1, std::back_inserter(result));
    return result;
}
 
int main() {
    int user_id = 123;
 
    //Simulates user behavior.
    recordView(user_id, 1001);
    recordView(user_id, 1002);
    recordView(user_id, 1003);
    recordView(user_id, 1004);
    recordView(user_id, 1005);
    recordPurchase(user_id, 1001);
    recordPurchase(user_id, 1005);
 
     //Updates product correlation information based on user behavior.
    updateRelatedProducts(user_id);
 
    //Queries products related to product 1001.
    int product_id = 1001;
    auto related_products = getRelatedProducts(product_id);
    std::cout << "Product ID " << product_id << "'s related product is: " << std::endl;
    for (const auto& [product_id, score] : related_products) {
        std::cout << "Product ID: " << product_id << ", Score: " << score << std::endl;
    }
 
    return 0;
}

Returned results: