nlp_similarity_crowdflower.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# NLP Similarity Approaches\n",
    "\n",
    "In this notebook i'll show alternative approaches to determine textual similarity. Basic approach from IR is keyword matching (tf/idf), but there is more options that can work as well or better. Combining approaches can give best resuts/\n",
    "\n",
    "Dataset and competition for search relevance is here: https://www.kaggle.com/c/crowdflower-search-relevance"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "\n",
    "df = pd.read_csv('data/crowdflower-search-relevance.csv', index_col=0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Dataset\n",
    "\n",
    "Let's load the data and have a quick look at the available variables."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>query</th>\n",
       "      <th>product_title</th>\n",
       "      <th>product_description</th>\n",
       "      <th>median_relevance</th>\n",
       "      <th>relevance_variance</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>id</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>bridal shower decorations</td>\n",
       "      <td>Accent Pillow with Heart Design - Red/Black</td>\n",
       "      <td>Red satin accent pillow embroidered with a hea...</td>\n",
       "      <td>1</td>\n",
       "      <td>0.000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>led christmas lights</td>\n",
       "      <td>Set of 10 Battery Operated Multi LED Train Chr...</td>\n",
       "      <td>Set of 10 Battery Operated Train Christmas Lig...</td>\n",
       "      <td>4</td>\n",
       "      <td>0.000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>projector</td>\n",
       "      <td>ViewSonic Pro8200 DLP Multimedia Projector</td>\n",
       "      <td>NaN</td>\n",
       "      <td>4</td>\n",
       "      <td>0.471</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>wine rack</td>\n",
       "      <td>Concept Housewares WR-44526 Solid-Wood Ceiling...</td>\n",
       "      <td>Like a silent and sturdy tree, the Southern En...</td>\n",
       "      <td>4</td>\n",
       "      <td>0.000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>light bulb</td>\n",
       "      <td>Wintergreen Lighting Christmas LED Light Bulb ...</td>\n",
       "      <td>WTGR1011\\nFeatures\\nNickel base, 60,000 averag...</td>\n",
       "      <td>2</td>\n",
       "      <td>0.471</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                        query  \\\n",
       "id                              \n",
       "1   bridal shower decorations   \n",
       "2        led christmas lights   \n",
       "4                   projector   \n",
       "5                   wine rack   \n",
       "7                  light bulb   \n",
       "\n",
       "                                        product_title  \\\n",
       "id                                                      \n",
       "1         Accent Pillow with Heart Design - Red/Black   \n",
       "2   Set of 10 Battery Operated Multi LED Train Chr...   \n",
       "4          ViewSonic Pro8200 DLP Multimedia Projector   \n",
       "5   Concept Housewares WR-44526 Solid-Wood Ceiling...   \n",
       "7   Wintergreen Lighting Christmas LED Light Bulb ...   \n",
       "\n",
       "                                  product_description  median_relevance  \\\n",
       "id                                                                        \n",
       "1   Red satin accent pillow embroidered with a hea...                 1   \n",
       "2   Set of 10 Battery Operated Train Christmas Lig...                 4   \n",
       "4                                                 NaN                 4   \n",
       "5   Like a silent and sturdy tree, the Southern En...                 4   \n",
       "7   WTGR1011\\nFeatures\\nNickel base, 60,000 averag...                 2   \n",
       "\n",
       "    relevance_variance  \n",
       "id                      \n",
       "1                0.000  \n",
       "2                0.000  \n",
       "4                0.471  \n",
       "5                0.000  \n",
       "7                0.471  "
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(10158, 5)"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Preprocessing \n",
    "\n",
    "So from the dataset we saw two columns describing the product (title, description) and one column with a query. We'll combine the two product columns. Also there is some empty values, so we'll fill those up."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "df['query'].fillna('', inplace=True)\n",
    "df['product_title'].fillna('', inplace=True)\n",
    "df['product_description'].fillna('', inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "def concatenate_title_and_description(row):\n",
    "    return row['product_title'] + ' ' + row['product_description']\n",
    "\n",
    "df['product_title_and_description'] = df.apply(concatenate_title_and_description, axis=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Text cleaning\n",
    "\n",
    "Next step is cleaning of the texts. The gensim library has nice functions for this, so we'll use those. Basically it will remove punctuation, symbols, stopwords and numbers. This should help determining textual similarity."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "from gensim.parsing.preprocessing import preprocess_documents\n",
    "\n",
    "query_texts = preprocess_documents(df['query'])\n",
    "product_title_and_description_texts = preprocess_documents(df['product_title_and_description'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's check out a few examples:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['bridal', 'shower', 'decor']"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "query_texts[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['girl', 'halloween', 'costum']"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "query_texts[23]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Dictionary & transforming to numbers\n",
    "\n",
    "Now we have the cleaned texts, so we'll transform them to numbers as gensim library requires that for further processing."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "from gensim.corpora import Dictionary\n",
    "\n",
    "dictionary = Dictionary(query_texts + product_title_and_description_texts)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "18303"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(dictionary)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "dictionary.filter_extremes(no_below=5, no_above=0.2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "5784"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(dictionary)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "query_corpus = [dictionary.doc2bow(x) for x in query_texts]\n",
    "product_title_and_description_corpus = [dictionary.doc2bow(x) for x in product_title_and_description_texts]\n",
    "\n",
    "all_corpus = query_corpus + product_title_and_description_corpus"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Process with Tf/Idf and LSI\n",
    "\n",
    "So now we'll apply two possible similarity models to our data:\n",
    "\n",
    "- Tf/Idf transformation: This counts word frequencies in documents (query/product) and overall. The logic is that uncommon words that appear multiple times in a product carry a lot of meaning. (This principle is common in all search engines)\n",
    "- LSI transformation: This is an SVD decomposition on top of tf/idf. It finds co-occurring words and decomposes the texts into co-occurring topics/dimensions. The idea is that each topic/diminsion has a specific semantic meaning.\n",
    "- Doc2Vec embedding: An embedding technique for words/docs. It analyzes which words occur in the same context and maps them close in a high dimensional space. The idea is that similar context means the words share meaning.\n",
    "\n",
    "So let's go transform out data!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "tfidf_dimensions = len(dictionary)\n",
    "lsi_dimensions = 150\n",
    "doc2vec_dimensions = 75"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "from gensim.models import TfidfModel\n",
    "\n",
    "tfidf = TfidfModel(all_corpus, id2word=dictionary)\n",
    "\n",
    "query_tfidf = tfidf[query_corpus]\n",
    "product_title_and_description_tfidf = tfidf[product_title_and_description_corpus]\n",
    "all_tfidf = tfidf[all_corpus]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "from gensim.models import LsiModel\n",
    "\n",
    "lsi = LsiModel(all_tfidf, id2word=dictionary)\n",
    "\n",
    "query_lsi = lsi[query_tfidf]\n",
    "product_title_and_description_lsi = lsi[product_title_and_description_tfidf]\n",
    "all_lsi = lsi[all_tfidf]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "from gensim.models.doc2vec import Doc2Vec, TaggedDocument\n",
    "\n",
    "query_tagged_docs = [TaggedDocument(x, [i]) for i, x in enumerate(query_texts)]\n",
    "title_and_description_tagged_docs = [TaggedDocument(x, [i]) for i, x in enumerate(product_title_and_description_texts, len(query_texts))]\n",
    "\n",
    "all_tagged_docs = query_tagged_docs + title_and_description_tagged_docs\n",
    "\n",
    "d2v_model = Doc2Vec(all_tagged_docs, vector_size=doc2vec_dimensions, epochs=40, workers=4)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Quick intermezzo, what did doc2vec learn? What words are similar to \"pillow\"? Does it make sense?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[('bed', 0.57822585105896),\n",
       " ('sleep', 0.538665235042572),\n",
       " ('sleeper', 0.5267432928085327),\n",
       " ('sham', 0.5035320520401001),\n",
       " ('surya', 0.49588605761528015),\n",
       " ('stomach', 0.48900169134140015),\n",
       " ('topper', 0.4878620505332947),\n",
       " ('blanket', 0.45482903718948364),\n",
       " ('luxuri', 0.44911447167396545),\n",
       " ('snore', 0.4265357255935669)]"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "d2v_model.wv.most_similar('pillow')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "So now we've trained the models, so let's transform our data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "from gensim.matutils import sparse2full\n",
    "\n",
    "df['query_tfidf'] = [sparse2full(x, tfidf_dimensions) for x in query_tfidf]\n",
    "df['product_title_and_description_tfidf'] = [sparse2full(x, tfidf_dimensions) for x in product_title_and_description_tfidf]\n",
    "\n",
    "df['query_lsi'] = [sparse2full(x, tfidf_dimensions) for x in query_lsi]\n",
    "df['product_title_and_description_lsi'] = [sparse2full(x, tfidf_dimensions) for x in product_title_and_description_lsi]\n",
    "\n",
    "df['query_doc2vec'] = [d2v_model.infer_vector(x) for x in query_texts]\n",
    "df['product_title_and_description_doc2vec'] = [d2v_model.infer_vector(x) for x in product_title_and_description_texts]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Similarity\n",
    "\n",
    "We've learned/trained our models and transformed our data. Next step is to determine similarity between product and query, this then could be used in a search engine to sort results!\n",
    "\n",
    "We'll determine similarity by cosine distance:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/dehling/Documents/machine-learnings/lib/python3.5/site-packages/scipy/spatial/distance.py:644: RuntimeWarning: invalid value encountered in float_scalars\n",
      "  dist = 1.0 - uv / np.sqrt(uu * vv)\n"
     ]
    }
   ],
   "source": [
    "from scipy.spatial.distance import cosine\n",
    "\n",
    "def cosine_distance_tfidf(row):\n",
    "    return cosine(row['query_tfidf'], row['product_title_and_description_tfidf'])\n",
    "\n",
    "def cosine_distance_lsi(row):\n",
    "    return cosine(row['query_lsi'], row['product_title_and_description_lsi'])\n",
    "\n",
    "def cosine_distance_doc2vec(row):\n",
    "    return cosine(row['query_doc2vec'], row['product_title_and_description_doc2vec'])\n",
    "\n",
    "df['distance_tfidf'] = df.apply(cosine_distance_tfidf, axis=1)\n",
    "df['distance_lsi'] = df.apply(cosine_distance_lsi, axis=1)\n",
    "df['distance_doc2vec'] = df.apply(cosine_distance_doc2vec, axis=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Lets look at some examples to see which products would match a query with our approach:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>query</th>\n",
       "      <th>product_title</th>\n",
       "      <th>median_relevance</th>\n",
       "      <th>distance_tfidf</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>id</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>14692</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>Cisco - flash memory card - 16 GB - SD</td>\n",
       "      <td>4</td>\n",
       "      <td>0.377292</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>23279</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>EP Memory - flash memory card - 16 GB - Compac...</td>\n",
       "      <td>4</td>\n",
       "      <td>0.392852</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4058</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>Samsung EVO MB-SP16D/AM 16GB Class 10 SD Memor...</td>\n",
       "      <td>4</td>\n",
       "      <td>0.399038</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17873</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>Samsung 16gb Micro SDHC Class 2 Memory Card wi...</td>\n",
       "      <td>4</td>\n",
       "      <td>0.411266</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9497</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>Lexar 600x 16GB MicroSDHC Memory Card with USB...</td>\n",
       "      <td>4</td>\n",
       "      <td>0.417127</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20526</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>SanDisk 16GB microSDHC Memory Card - Black (SD...</td>\n",
       "      <td>4</td>\n",
       "      <td>0.442482</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19846</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>Dane 16GB microSD Memory Card with Adapter - B...</td>\n",
       "      <td>4</td>\n",
       "      <td>0.445012</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18757</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>16GB MicroSD Memory Card with SD Adapter (Pack...</td>\n",
       "      <td>4</td>\n",
       "      <td>0.449338</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>24971</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>SanDisk Ultra 16GB microSDHC Memory Card</td>\n",
       "      <td>4</td>\n",
       "      <td>0.463154</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>23804</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>SanDisk Ultra 16GB UHI-I Class 10 Micro SDHC M...</td>\n",
       "      <td>4</td>\n",
       "      <td>0.464085</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                   query                                      product_title  \\\n",
       "id                                                                            \n",
       "14692  16 gb memory card             Cisco - flash memory card - 16 GB - SD   \n",
       "23279  16 gb memory card  EP Memory - flash memory card - 16 GB - Compac...   \n",
       "4058   16 gb memory card  Samsung EVO MB-SP16D/AM 16GB Class 10 SD Memor...   \n",
       "17873  16 gb memory card  Samsung 16gb Micro SDHC Class 2 Memory Card wi...   \n",
       "9497   16 gb memory card  Lexar 600x 16GB MicroSDHC Memory Card with USB...   \n",
       "20526  16 gb memory card  SanDisk 16GB microSDHC Memory Card - Black (SD...   \n",
       "19846  16 gb memory card  Dane 16GB microSD Memory Card with Adapter - B...   \n",
       "18757  16 gb memory card  16GB MicroSD Memory Card with SD Adapter (Pack...   \n",
       "24971  16 gb memory card           SanDisk Ultra 16GB microSDHC Memory Card   \n",
       "23804  16 gb memory card  SanDisk Ultra 16GB UHI-I Class 10 Micro SDHC M...   \n",
       "\n",
       "       median_relevance  distance_tfidf  \n",
       "id                                       \n",
       "14692                 4        0.377292  \n",
       "23279                 4        0.392852  \n",
       "4058                  4        0.399038  \n",
       "17873                 4        0.411266  \n",
       "9497                  4        0.417127  \n",
       "20526                 4        0.442482  \n",
       "19846                 4        0.445012  \n",
       "18757                 4        0.449338  \n",
       "24971                 4        0.463154  \n",
       "23804                 4        0.464085  "
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df[['query', 'product_title', 'median_relevance', 'distance_tfidf']].sort_values(by=['query', 'distance_tfidf']).head(n=10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Looks pretty sensible, right? For a query about a memory card various memory cards are suggested. I'm happy!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>query</th>\n",
       "      <th>product_title</th>\n",
       "      <th>median_relevance</th>\n",
       "      <th>distance_lsi</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>id</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>19182</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>Transcend 16GB MicroSD Class 4 Flash Memory Card</td>\n",
       "      <td>4</td>\n",
       "      <td>0.007564</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>32384</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>Transcend 16gb Class 10 SDHC Flash Memory Card</td>\n",
       "      <td>4</td>\n",
       "      <td>0.010354</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>23279</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>EP Memory - flash memory card - 16 GB - Compac...</td>\n",
       "      <td>4</td>\n",
       "      <td>0.012274</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>29717</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>Kingston 4GB 8GB 16GB 32GB Class 4 microSD SDH...</td>\n",
       "      <td>4</td>\n",
       "      <td>0.016376</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20967</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>SanDisk Ultra 16G 32GB 64GB 128GB 40MB/S SD SD...</td>\n",
       "      <td>4</td>\n",
       "      <td>0.024250</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14692</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>Cisco - flash memory card - 16 GB - SD</td>\n",
       "      <td>4</td>\n",
       "      <td>0.028702</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13919</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>SanDisk - flash memory card - 16 GB - microSDHC</td>\n",
       "      <td>4</td>\n",
       "      <td>0.041562</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19317</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>SanDisk 16GB Class 4 microSDHC Memory Card</td>\n",
       "      <td>4</td>\n",
       "      <td>0.042414</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>23804</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>SanDisk Ultra 16GB UHI-I Class 10 Micro SDHC M...</td>\n",
       "      <td>4</td>\n",
       "      <td>0.054703</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4058</th>\n",
       "      <td>16 gb memory card</td>\n",
       "      <td>Samsung EVO MB-SP16D/AM 16GB Class 10 SD Memor...</td>\n",
       "      <td>4</td>\n",
       "      <td>0.059089</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                   query                                      product_title  \\\n",
       "id                                                                            \n",
       "19182  16 gb memory card   Transcend 16GB MicroSD Class 4 Flash Memory Card   \n",
       "32384  16 gb memory card     Transcend 16gb Class 10 SDHC Flash Memory Card   \n",
       "23279  16 gb memory card  EP Memory - flash memory card - 16 GB - Compac...   \n",
       "29717  16 gb memory card  Kingston 4GB 8GB 16GB 32GB Class 4 microSD SDH...   \n",
       "20967  16 gb memory card  SanDisk Ultra 16G 32GB 64GB 128GB 40MB/S SD SD...   \n",
       "14692  16 gb memory card             Cisco - flash memory card - 16 GB - SD   \n",
       "13919  16 gb memory card    SanDisk - flash memory card - 16 GB - microSDHC   \n",
       "19317  16 gb memory card         SanDisk 16GB Class 4 microSDHC Memory Card   \n",
       "23804  16 gb memory card  SanDisk Ultra 16GB UHI-I Class 10 Micro SDHC M...   \n",
       "4058   16 gb memory card  Samsung EVO MB-SP16D/AM 16GB Class 10 SD Memor...   \n",
       "\n",
       "       median_relevance  distance_lsi  \n",
       "id                                     \n",
       "19182                 4      0.007564  \n",
       "32384                 4      0.010354  \n",
       "23279                 4      0.012274  \n",
       "29717                 4      0.016376  \n",
       "20967                 4      0.024250  \n",
       "14692                 4      0.028702  \n",
       "13919                 4      0.041562  \n",
       "19317                 4      0.042414  \n",
       "23804                 4      0.054703  \n",
       "4058                  4      0.059089  "
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df[['query', 'product_title', 'median_relevance', 'distance_lsi']].sort_values(by=['query', 'distance_lsi']).head(n=10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "LSI also looks sensible, similar products are suggested."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Evaluation\n",
    "\n",
    "Let's compare the similarity we calculated against the human assessed relevance. For this we'll use correlation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "-0.364037545605384"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df['median_relevance'].corr(df['distance_tfidf'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "-0.3669726247025937"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df['median_relevance'].corr(df['distance_lsi'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "-0.08686668817840917"
      ]
     },
     "execution_count": 72,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df['median_relevance'].corr(df['distance_doc2vec'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's also make some plots to visualize the relation between calculated similarity and assessed relevance."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f30389ab940>"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import seaborn as sns\n",
    "%matplotlib inline\n",
    "\n",
    "sns.violinplot(x='median_relevance', y='distance_tfidf', data=df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f2fe3aea940>"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.violinplot(x='median_relevance', y='distance_lsi', data=df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f2fe08bbe10>"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.violinplot(x='median_relevance', y='distance_doc2vec', data=df)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Each of the similarity approaches shows different results, and perform differently good/bad.\n",
    "\n",
    "* TF/IDF and LSI do quite well on correlation and find relevant product for a query\n",
    "* Doc2Vec doesn't do well, probably it needs better training data or more finetuning. With more energy spent it might work, but not out-of-the-box at least.\n",
    "\n",
    "I hope you enjoyed!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.5.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}