semantic_similarity.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Import Library"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sentence_transformers import SentenceTransformer, util\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Model Selection and Initialization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 1.31G/1.31G [01:36<00:00, 13.6MB/s]  \n"
     ]
    }
   ],
   "source": [
    "# List of models optimized for semantic textual similarity can be found at:\n",
    "# https://docs.google.com/spreadsheets/d/14QplCdTCDwEmTqrn1LH4yrbKvdogK4oQvYO1K1aPR5M/edit#gid=0\n",
    "model = SentenceTransformer('stsb-roberta-large')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Calculate semantic similarity between two sentences"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Sentence 1: I like Python because I can build AI applications\n",
      "Sentence 2: I like Python because I can do data analytics\n",
      "Similarity score: 0.8188022375106812\n"
     ]
    }
   ],
   "source": [
    "sentence1 = \"I like Python because I can build AI applications\"\n",
    "sentence2 = \"I like Python because I can do data analytics\"\n",
    "\n",
    "# encode sentences to get their embeddings\n",
    "embedding1 = model.encode(sentence1, convert_to_tensor=True)\n",
    "embedding2 = model.encode(sentence2, convert_to_tensor=True)\n",
    "\n",
    "# compute similarity scores of two embeddings\n",
    "cosine_scores = util.pytorch_cos_sim(embedding1, embedding2)\n",
    "\n",
    "print(\"Sentence 1:\", sentence1)\n",
    "print(\"Sentence 2:\", sentence2)\n",
    "print(\"Similarity score:\", cosine_scores.item())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Calculate semantic similarity between two lists of sentences"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Sentence 1: I like Python because I can build AI applications\n",
      "Sentence 2: I like Python because I can do data analytics\n",
      "Similarity Score: 0.8188023567199707\n",
      "\n",
      "Sentence 1: I like Python because I can build AI applications\n",
      "Sentence 2: The cat walks on the sidewalk\n",
      "Similarity Score: -0.06005367636680603\n",
      "\n",
      "Sentence 1: The cat sits on the ground\n",
      "Sentence 2: I like Python because I can do data analytics\n",
      "Similarity Score: 0.12721936404705048\n",
      "\n",
      "Sentence 1: The cat sits on the ground\n",
      "Sentence 2: The cat walks on the sidewalk\n",
      "Similarity Score: 0.4131842255592346\n",
      "\n"
     ]
    }
   ],
   "source": [
    "sentences1 = [\"I like Python because I can build AI applications\", \"The cat sits on the ground\"]   \n",
    "sentences2 = [\"I like Python because I can do data analytics\", \"The cat walks on the sidewalk\"]\n",
    "\n",
    "# encode list of sentences to get their embeddings\n",
    "embedding1 = model.encode(sentences1, convert_to_tensor=True)\n",
    "embedding2 = model.encode(sentences2, convert_to_tensor=True)\n",
    "\n",
    "# compute similarity scores of two embeddings\n",
    "cosine_scores = util.pytorch_cos_sim(embedding1, embedding2)\n",
    "\n",
    "for i in range(len(sentences1)):\n",
    "    for j in range(len(sentences2)):\n",
    "        print(\"Sentence 1:\", sentences1[i])\n",
    "        print(\"Sentence 2:\", sentences2[j])\n",
    "        print(\"Similarity Score:\", cosine_scores[i][j].item())\n",
    "        print()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Retrieve Top K most similar sentences from a corpus given a sentence"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "corpus = [\"I like Python because I can build AI applications\",\n",
    "          \"I like Python because I can do data analytics\",\n",
    "          \"The cat sits on the ground\",\n",
    "         \"The cat walks on the sidewalk\"]\n",
    "\n",
    "# encode corpus to get corpus embeddings\n",
    "corpus_embeddings = model.encode(corpus, convert_to_tensor=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "sentence = \"I like Javascript because I can build web applications\"\n",
    "\n",
    "# encode sentence to get sentence embeddings\n",
    "sentence_embedding = model.encode(sentence, convert_to_tensor=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Sentence: I like Javascript because I can build web applications \n",
      "\n",
      "Top 2 most similar sentences in corpus:\n",
      "I like Python because I can build AI applications (Score: 0.6253)\n",
      "I like Python because I can do data analytics (Score: 0.5348)\n"
     ]
    }
   ],
   "source": [
    "# top_k results to return\n",
    "top_k=2\n",
    "\n",
    "# compute similarity scores of the sentence with the corpus\n",
    "cos_scores = util.pytorch_cos_sim(sentence_embedding, corpus_embeddings)[0]\n",
    "\n",
    "# Sort the results in decreasing order and get the first top_k\n",
    "top_results = np.argpartition(-cos_scores, range(top_k))[0:top_k]\n",
    "\n",
    "print(\"Sentence:\", sentence, \"\\n\")\n",
    "print(\"Top\", top_k, \"most similar sentences in corpus:\")\n",
    "for idx in top_results[0:top_k]:\n",
    "    print(corpus[idx], \"(Score: %.4f)\" % (cos_scores[idx]))"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}