import { pipeline } from "https://cdn.jsdelivr.net/npm/@huggingface/transformers";

​

const extractor = await pipeline(

  "feature-extraction",

  "Xenova/bge-small-en-v1.5"

);

async function getEmbeddings(sentences) {

  let embeddings = [];

  for (let sentence of sentences) {

    let output = await extractor(sentence, {

      pooling: "mean",

      normalize: true,

    });

    embeddings.push(output.data);

  }

  return embeddings;

}

​

const raw = await fetch("thesis-embeddings-193.json");

const { embeddings } = await raw.json();

​

const searchInput = document.getElementById("searchInput");

const results = document.getElementById("searchResults");

​

console.log("ready (ignore errors!)");

searchInput.addEventListener("input", async (event) => {

  const query = event.target.value;

  const queryEmbedding = await getEmbeddings([query]);

​

  const similarities = embeddings.map((embedding) =>

    cosineSimilarity(embedding.embedding, queryEmbedding[0])

  );

  const sortedIndices = similarities

    .map((_, i) => i)

    .sort((a, b) => similarities[b] - similarities[a]);

  const topResults = sortedIndices.slice(0, 5).map((i) => embeddings[i]);

  results.innerHTML = "";

  for (let result of topResults) {

    const li = document.createElement("li");

    li.textContent = result.text;

    results.appendChild(li);

  }

});

​

// Function to calculate dot product of two vectors

function dotProduct(vecA, vecB) {

  return vecA.reduce((sum, val, i) => sum + val * vecB[i], 0);

}

​

// Function to calculate the magnitude of a vector

function magnitude(vec) {

  return Math.sqrt(vec.reduce((sum, val) => sum + val * val, 0));

}

​

// Function to calculate cosine similarity between two vectors

function cosineSimilarity(vecA, vecB) {

  const numerator = dotProduct(vecA, vecB);

  const denominator = magnitude(vecA) * magnitude(vecB);

  // console.log(numerator, denominator);

  return numerator / denominator;

}

​