export const EMBEDDING_DIM = 64;
export type EmbeddingVector = number[];
function stableHash(value: string): number {
let hash = 2166136261;
for (let index = 0; index < value.length; index += 1) {
hash ^= value.charCodeAt(index);
hash = Math.imul(hash, 16777619);
}
return hash >>> 0;
}
export function tokenize(text: string): string[] {
const normalized = text.toLowerCase();
const tokens = normalized.match(/[a-z0-9_./:-]{2,}/g);
return tokens ?? [];
}
export function normalizeVector(vector: EmbeddingVector): EmbeddingVector {
const magnitude = Math.sqrt(vector.reduce((acc, value) => acc + value * value, 0));
if (magnitude === 0) {
return vector;
}
return vector.map((value) => value / magnitude);
}
export function textToEmbedding(text: string, dimension = EMBEDDING_DIM): EmbeddingVector {
const vector = new Array<number>(dimension).fill(0);
const tokens = tokenize(text);
for (const token of tokens) {
const hash = stableHash(token);
const index = hash % dimension;
const sign = (hash & 1) === 0 ? 1 : -1;
const weight = Math.min(3, Math.max(1, token.length / 5));
vector[index] += sign * weight;
}
return normalizeVector(vector);
}
export function averageEmbeddings(vectors: EmbeddingVector[]): EmbeddingVector {
if (vectors.length === 0) {
return new Array<number>(EMBEDDING_DIM).fill(0);
}
const dimension = vectors[0].length;
const centroid = new Array<number>(dimension).fill(0);
for (const vector of vectors) {
for (let index = 0; index < dimension; index += 1) {
centroid[index] += vector[index] ?? 0;
}
}
const averaged = centroid.map((value) => value / vectors.length);
return normalizeVector(averaged);
}
export function cosineSimilarity(a: EmbeddingVector, b: EmbeddingVector): number {
const dimension = Math.min(a.length, b.length);
let dot = 0;
let normA = 0;
let normB = 0;
for (let index = 0; index < dimension; index += 1) {
dot += (a[index] ?? 0) * (b[index] ?? 0);
normA += (a[index] ?? 0) * (a[index] ?? 0);
normB += (b[index] ?? 0) * (b[index] ?? 0);
}
if (normA === 0 || normB === 0) {
return 0;
}
return dot / Math.sqrt(normA * normB);
}
