feat: add cosine similarity algorithm

__tests__/unit/bfs.spec.ts

@@ -1,4 +1,4 @@
-import breadthFirstSearch from "../../packages/graph/src/bfs";
+import { breadthFirstSearch } from "../../packages/graph/src";
 import { Graph } from "@antv/graphlib";
 
 const data = {

__tests__/unit/cosine-similarity.spec.ts

@@ -0,0 +1,41 @@
+import { cosineSimilarity } from "../../packages/graph/src";
+
+describe('cosineSimilarity abnormal demo: ', () => {
+    it('item contains only zeros: ', () => {
+        const item = [0, 0, 0];
+        const targetTtem = [3, 1, 1];
+        const cosineSimilarityValue = cosineSimilarity(item, targetTtem);
+        expect(cosineSimilarityValue).toBe(0);
+    });
+    it('targetTtem contains only zeros: ', () => {
+        const item = [3, 5, 2];
+        const targetTtem = [0, 0, 0];
+        const cosineSimilarityValue = cosineSimilarity(item, targetTtem);
+        expect(cosineSimilarityValue).toBe(0);
+    });
+    it('item and targetTtem both contains only zeros: ', () => {
+        const item = [0, 0, 0];
+        const targetTtem = [0, 0, 0];
+        const cosineSimilarityValue = cosineSimilarity(item, targetTtem);
+        expect(cosineSimilarityValue).toBe(0);
+    });
+});
+
+describe('cosineSimilarity normal demo: ', () => {
+    it('demo similar: ', () => {
+        const item = [30, 0, 100];
+        const targetTtem = [32, 1, 120];
+        const cosineSimilarityValue = cosineSimilarity(item, targetTtem);
+        expect(cosineSimilarityValue).toBeGreaterThanOrEqual(0);
+        expect(cosineSimilarityValue).toBeLessThan(1);
+        expect(Number(cosineSimilarityValue.toFixed(3))).toBe(0.999);
+    });
+    it('demo dissimilar: ', () => {
+        const item = [10, 300, 2];
+        const targetTtem = [1, 2, 30];
+        const cosineSimilarityValue = cosineSimilarity(item, targetTtem);
+        expect(cosineSimilarityValue).toBeGreaterThanOrEqual(0);
+        expect(cosineSimilarityValue).toBeLessThan(1);
+        expect(Number(cosineSimilarityValue.toFixed(3))).toBe(0.074);
+    });
+});

__tests__/unit/dfs.spec.ts

@@ -1,4 +1,4 @@
-import depthFirstSearch from "../../packages/graph/src/dfs";
+import { depthFirstSearch } from "../../packages/graph/src";
 import { Graph } from "@antv/graphlib";
 
 

__tests__/unit/nodes-cosine-similarity.spec.ts

@@ -0,0 +1,109 @@
+import { nodesCosineSimilarity } from "../../packages/graph/src";
+import propertiesGraphData from '../data/cluster-origin-properties-data.json';
+import { NodeSimilarity } from "../../packages/graph/src/types";
+
+describe('nodesCosineSimilarity abnormal demo', () => {
+    it('no properties demo: ', () => {
+        const nodes = [
+            {
+                id: 'node-0',
+                data: {},
+            },
+            {
+                id: 'node-1',
+                data: {},
+            },
+            {
+                id: 'node-2',
+                data: {},
+            },
+            {
+                id: 'node-3',
+                data: {},
+            }
+        ];
+        const { allCosineSimilarity, similarNodes } = nodesCosineSimilarity(nodes as NodeSimilarity[], nodes[0]);
+        expect(allCosineSimilarity.length).toBe(3);
+        expect(similarNodes.length).toBe(3);
+        expect(allCosineSimilarity[0]).toBe(0);
+        expect(allCosineSimilarity[1]).toBe(0);
+        expect(allCosineSimilarity[2]).toBe(0);
+    });
+});
+
+
+describe('nodesCosineSimilarity normal demo', () => {
+    it('simple demo: ', () => {
+        const nodes = [
+            {
+                id: 'node-0',
+                data: {
+                    amount: 10,
+                }
+            },
+            {
+                id: 'node-2',
+                data: {
+                    amount: 100,
+                }
+            },
+            {
+                id: 'node-3',
+                data: {
+                    amount: 1000,
+                }
+            },
+            {
+                id: 'node-4',
+                data: {
+                    amount: 50,
+                }
+            }
+        ];
+        const { allCosineSimilarity, similarNodes } = nodesCosineSimilarity(nodes as NodeSimilarity[], nodes[0], ['amount']);
+        expect(allCosineSimilarity.length).toBe(3);
+        expect(similarNodes.length).toBe(3);
+        allCosineSimilarity.forEach(data => {
+            expect(data).toBeGreaterThanOrEqual(0);
+            expect(data).toBeLessThanOrEqual(1);
+        })
+    });
+
+    it('complex demo: ', () => {
+        const { nodes } = propertiesGraphData;
+        const { allCosineSimilarity, similarNodes } = nodesCosineSimilarity(nodes as NodeSimilarity[], nodes[16]);
+        expect(allCosineSimilarity.length).toBe(16);
+        expect(similarNodes.length).toBe(16);
+        allCosineSimilarity.forEach(data => {
+            expect(data).toBeGreaterThanOrEqual(0);
+            expect(data).toBeLessThanOrEqual(1);
+        })
+    });
+
+
+    it('demo use involvedKeys: ', () => {
+        const involvedKeys = ['amount', 'wifi'];
+        const { nodes } = propertiesGraphData;
+        const { allCosineSimilarity, similarNodes } = nodesCosineSimilarity(nodes as NodeSimilarity[], nodes[16], involvedKeys);
+        expect(allCosineSimilarity.length).toBe(16);
+        expect(similarNodes.length).toBe(16);
+        allCosineSimilarity.forEach(data => {
+            expect(data).toBeGreaterThanOrEqual(0);
+            expect(data).toBeLessThanOrEqual(1);
+        })
+        expect(similarNodes[0].id).toBe('node-11');
+    });
+
+    it('demo use uninvolvedKeys: ', () => {
+        const uninvolvedKeys = ['amount'];
+        const { nodes } = propertiesGraphData;
+        const { allCosineSimilarity, similarNodes } = nodesCosineSimilarity(nodes as NodeSimilarity[], nodes[16], [], uninvolvedKeys);
+        expect(allCosineSimilarity.length).toBe(16);
+        expect(similarNodes.length).toBe(16);
+        allCosineSimilarity.forEach(data => {
+            expect(data).toBeGreaterThanOrEqual(0);
+            expect(data).toBeLessThanOrEqual(1);
+        })
+        expect(similarNodes[0].id).toBe('node-11');
+    });
+});

package.json

@@ -21,7 +21,7 @@
     "build:ci": "pnpm -r run build:ci",
     "prepare": "husky install",
     "test": "jest",
-    "test_one": "jest ./__tests__/unit/dfs.spec.ts",
+    "test_one": "jest ./__tests__/unit/nodes-cosine-similarity.spec.ts",
     "coverage": "jest --coverage",
     "build:site": "vite build",
     "deploy": "gh-pages -d site/dist",

packages/graph/src/bfs.ts

@@ -24,7 +24,7 @@ Performs breadth-first search (BFS) traversal on a graph.
 @param startNodeId - The ID of the starting node for BFS.
 @param originalCallbacks - Optional object containing callback functions for BFS.
 */
-const breadthFirstSearch = (
+export const breadthFirstSearch = (
   graph: Graph,
   startNodeId: NodeID,
   originalCallbacks?: IAlgorithmCallbacks,
@@ -65,5 +65,3 @@ const breadthFirstSearch = (
     previousNodeId = currentNodeId;
   }
 };
-
-export default breadthFirstSearch;

packages/graph/src/cosine-similarity.ts

@@ -0,0 +1,27 @@
+import { Vector } from "./vector";
+
+/**
+Calculates the cosine similarity
+@param item - The element.
+@param targetItem - The target element.
+@returns The cosine similarity between the item and the targetItem.
+*/
+export const cosineSimilarity = (
+    item: number[],
+    targetItem: number[],
+): number => {
+    // Vector of the target element
+    const targetItemVector = new Vector(targetItem);
+    // Norm of the target element vector
+    const targetNodeNorm2 = targetItemVector.norm2();
+    // Vector of the item
+    const itemVector = new Vector(item);
+    // Norm of the item vector
+    const itemNorm2 = itemVector.norm2();
+    // Calculate the dot product of the item vector and the target element vector
+    const dot = targetItemVector.dot(itemVector);
+    const norm2Product = targetNodeNorm2 * itemNorm2;
+    // Calculate the cosine similarity between the item vector and the target element vector
+    const cosineSimilarity = norm2Product ? dot / norm2Product : 0;
+    return cosineSimilarity;
+}

packages/graph/src/dfs.ts

@@ -41,7 +41,7 @@ function depthFirstSearchRecursive(
   });
 }
 
-export default function depthFirstSearch(
+export function depthFirstSearch(
   graph: Graph,
   startNodeId: NodeID,
   originalCallbacks?: IAlgorithmCallbacks,

packages/graph/src/index.ts

@@ -5,4 +5,6 @@ export * from "./iLouvain";
 export * from "./k-core";
 export * from "./floydWarshall";
 export * from "./bfs";
-export * from "./dfs";
+export * from "./dfs";
+export * from "./cosine-similarity"
+export * from "./nodes-cosine-similarity";

packages/graph/src/nodes-cosine-similarity.ts

@@ -0,0 +1,43 @@
+import { clone } from '@antv/util';
+import { getAllProperties, oneHot } from './utils';
+import { NodeSimilarity } from './types';
+import { cosineSimilarity } from '.';
+
+/**
+Calculates the cosine similarity based on node attributes using the nodes-cosine-similarity algorithm.
+This algorithm is used to find similar nodes based on a seed node in a graph.
+@param nodes - The data of graph nodes.
+@param seedNode - The seed node for similarity calculation.
+@param involvedKeys - The collection of keys that are involved in the calculation.
+@param uninvolvedKeys - The collection of keys that are not involved in the calculation.
+@returns An array of nodes that are similar to the seed node based on cosine similarity.
+*/
+export const nodesCosineSimilarity = (
+  nodes: NodeSimilarity[] = [],
+  seedNode: NodeSimilarity,
+  involvedKeys: string[] = [],
+  uninvolvedKeys: string[] = [],
+): {
+  allCosineSimilarity: number[],
+  similarNodes: NodeSimilarity[],
+} => {
+  const similarNodes = clone(nodes.filter(node => node.id !== seedNode.id));
+  const seedNodeIndex = nodes.findIndex(node => node.id === seedNode.id);
+  // Collection of all node properties
+  const properties = getAllProperties(nodes);
+  // One-hot feature vectors for all node properties
+  const allPropertiesWeight = oneHot(properties, involvedKeys, uninvolvedKeys) as number[][];
+  // Seed node properties
+  const seedNodeProperties = allPropertiesWeight[seedNodeIndex];
+  const allCosineSimilarity: number[] = [];
+  similarNodes.forEach((node: NodeSimilarity, index: number) => {
+    const nodeProperties = allPropertiesWeight[index];
+    // Calculate the cosine similarity between node vector and seed node vector
+    const cosineSimilarityValue = cosineSimilarity(nodeProperties, seedNodeProperties);
+    allCosineSimilarity.push(cosineSimilarityValue);
+    node.cosineSimilarity = cosineSimilarityValue;
+  });
+  // Sort the returned nodes according to cosine similarity
+  similarNodes.sort((a: NodeSimilarity, b: NodeSimilarity) => b.cosineSimilarity - a.cosineSimilarity);
+  return { allCosineSimilarity, similarNodes };
+}

packages/graph/src/types.ts

@@ -37,4 +37,7 @@ export interface IAlgorithmCallbacks {
   allowTraversal?: (param: { previous?: NodeID; current?: NodeID; next: NodeID }) => boolean;
 }
 
-export type NodeID = string | number;
+export type NodeID = string | number;
+export interface NodeSimilarity extends Node<{ [key: string]: any }> {
+  cosineSimilarity?: number;
+}

packages/graph/src/utils.ts

@@ -50,7 +50,7 @@ export const oneHot = (dataList: PlainObject[], involvedKeys?: string[], uninvol
   // 获取所有的属性/特征值
   const allValue = Object.values(allKeyValueMap);
   // 是否所有属性/特征的值都是数值型
-  const isAllNumber = allValue.every((value) => value.every((item) => (typeof(item) === 'number')));
+  const isAllNumber = allValue.every((value) => value.every((item) => (typeof (item) === 'number')));
 
   // 对数据进行one-hot编码
   dataList.forEach((data, index) => {
@@ -65,7 +65,7 @@ export const oneHot = (dataList: PlainObject[], involvedKeys?: string[], uninvol
         subCode.push(keyValue);
       } else {
         // 进行one-hot编码
-        for(let i = 0; i < allKeyValue.length; i++) {
+        for (let i = 0; i < allKeyValue.length; i++) {
           if (i === valueIndex) {
             subCode.push(1);
           } else {

tsconfig.json

@@ -7,6 +7,7 @@
     "allowJs": true,
     "moduleResolution": "node",
     "resolveJsonModule": true,
-    "allowSyntheticDefaultImports": true
+    "allowSyntheticDefaultImports": true,
+    "esModuleInterop": true,
   }
 }