Simple Recommendation Engine

• Package for Node.JS
• Implementation of Apriori-like algorithm for mining associations for given antecedent
• Computation in C++

Motivation

• Node.JS is a production ready framework/engine
• Parallel Call Stack
• Ability to parallelize process (fork, spawn etc.)
• C++ integration makes data processing algorithms to perform faster than most of the HLLs
• Following benchmark that shows with what factor C++ performs computation than in native JS

Benchmark

The following benchmark was performed on Ubuntu 16.04 with Node 10.13.0.

Transactions Growth

Itemset Growth

Usage

mine API

Syntax

1Function (Transaction, Options)
2Transaction: Number[][]
3Options: {
4  antecedent: Number,
5  sortingMeasure: String, // default: 'confidence'
6  attachMeasures: Boolean // default false
7}

Sorting measures

Supply one of the following values as String for sortingMeasure parameter in Options. If attachMeasures is set to true, all the following measures are calculated and returned in response, and sorting of associations is done with the sortingMeasure.

• confidence
• lift
• improvedLift
• kulc
• imbalanceRatio
• biConfidence
• biImprove
• jaccard
• conviction
• sebagSchoenauer
• certaintyFactor
• lift
• improve

Note that, currently this package supports only numeric itemset as an array of transactions, where each transaction is an individual array.

Example Usage

1const mine = require('apriorijscpp');
2const assocs = mine([
3  [3, 10],
4  [3, 4, 5],
5  [3, 4, 5],
6  [3, 4],
7  [1, 2],
8  [3],
9  [3, 4, 1],
10  [1, 2],
11  [7, 8, 9]
12], { 
13  antecedent: 3,
14  sortingMeasure: 'lift'
15});
16
17console.log(assocs);

Source

https://github.com/PANDYA21/apriori_nodejs_cpp

ToDo

• Compatibility for newer versions of node.js (newer N-API, breaking changes)
• Parallelization with child process handling
• Add contribution guide
• Add support for string based itemset

Contribution