{ "nbformat": 4, "nbformat_minor": 5, "metadata": { "kernelspec": { "display_name": "Python 3", "language": "python", "name": "python3" }, "language_info": { "name": "python", "version": "3.11" }, "colab": { "provenance": [] } }, "cells": [ { "cell_type": "markdown", "id": "cell-0000", "metadata": {}, "source": [ "# Content-Based Filtering\n", "\n", "Recommend \"more like this\" by turning movie descriptions into TF-IDF vectors and ranking them with cosine similarity — a full search-style recommender in the browser.\n", "\n", "*Part of the free [Machine Learning](https://ramadnsyh.dev/courses/machine-learning) course by [Muhammad Ramadiansyah](https://ramadnsyh.dev). This notebook is generated from the interactive lesson — [read it online](https://ramadnsyh.dev/courses/machine-learning/content-based-filtering).*" ] }, { "cell_type": "markdown", "id": "cell-0001", "metadata": {}, "source": [ "Popularity charts treat everyone the same. Content-based filtering takes the\n", "first step toward personalization: if you just watched *Toy Story*, recommend\n", "movies that *are like* Toy Story. To do that we need two ingredients — a way\n", "to represent each item as numbers, and a way to measure how close two items\n", "are. This lesson builds both, then wires them into a working\n", "`get_recommendations()` function." ] }, { "cell_type": "markdown", "id": "cell-0002", "metadata": {}, "source": [ "## From \"more like this\" to vectors\n", "\n", "Content-based filtering compares **item attributes**: genres, keywords, cast,\n", "director, plot synopsis. For movies, the most flexible trick is to mash all of\n", "that text into a single string per movie — often called a **metadata soup**:" ] }, { "cell_type": "markdown", "id": "cell-0003", "metadata": {}, "source": [ "```text\n", "\"animation family comedy toys jealousy a cowboy doll is threatened\n", " when a new spaceman figure becomes the favorite toy\"\n", "```" ] }, { "cell_type": "markdown", "id": "cell-0004", "metadata": {}, "source": [ "Once every movie is a bag of words, the problem becomes a *document search*\n", "problem: encode each soup as a vector, then find the vectors nearest to the\n", "one the user just watched. It's the same machinery behind a search engine —\n", "except the \"query\" is a movie instead of typed keywords." ] }, { "cell_type": "markdown", "id": "cell-0005", "metadata": {}, "source": [ "## Counting words — and why raw counts mislead\n", "\n", "The simplest encoding is a word count: one dimension per vocabulary word, and\n", "each movie's vector holds how often each word appears (scikit-learn's\n", "`CountVectorizer`). The flaw: common words dominate. If half your catalog's\n", "synopses contain \"world\" or \"life\", those words contribute big counts to many\n", "pairs of movies — inflating similarity without carrying any real signal.\n", "\n", "**TF-IDF** (term frequency × inverse document frequency) fixes this by\n", "down-weighting words that appear in many documents and up-weighting rare,\n", "distinctive ones. A word like \"dinosaur\" that appears in only two synopses\n", "becomes a strong link between exactly those two movies; a word like \"story\"\n", "that appears everywhere gets weight near zero. That's `TfidfVectorizer`." ] }, { "cell_type": "markdown", "id": "cell-0006", "metadata": {}, "source": [ "## Cosine similarity: direction, not length\n", "\n", "How do we compare two vectors? Euclidean distance is tempting, but it punishes\n", "*length* — a long synopsis would look far from a short one even if they use\n", "identical vocabulary. **Cosine similarity** measures only the *angle* between\n", "vectors: 1 means \"pointing the same way\", 0 means \"nothing in common\".\n", "See it in two dimensions first:" ] }, { "cell_type": "code", "id": "cell-0007", "metadata": {}, "execution_count": null, "outputs": [], "source": [ "import numpy as np\n", "\n", "# Two dimensions: counts of \"robot\" and \"love\" in a synopsis\n", "short_scifi = np.array([3.0, 1.0]) # short synopsis, mostly robots\n", "long_scifi = np.array([9.0, 3.0]) # 3x longer, same vocabulary mix\n", "romance = np.array([1.0, 4.0]) # mostly love\n", "\n", "def cosine(a, b):\n", " return a @ b / (np.linalg.norm(a) * np.linalg.norm(b))\n", "\n", "def euclid(a, b):\n", " return np.linalg.norm(a - b)\n", "\n", "print(\"cosine(short sci-fi, long sci-fi) =\", round(cosine(short_scifi, long_scifi), 3))\n", "print(\"cosine(short sci-fi, romance) =\", round(cosine(short_scifi, romance), 3))\n", "print()\n", "print(\"euclidean(short sci-fi, long sci-fi) =\", round(euclid(short_scifi, long_scifi), 2))\n", "print(\"euclidean(short sci-fi, romance) =\", round(euclid(short_scifi, romance), 2))" ] }, { "cell_type": "markdown", "id": "cell-0008", "metadata": {}, "source": [ "Cosine gets it right: the two sci-fi synopses are a perfect 1.0 match because\n", "they point in the same direction, while Euclidean distance claims the romance\n", "synopsis is *closer* to the short sci-fi one — purely because the long synopsis\n", "has bigger numbers. For text, always think angles." ] }, { "cell_type": "markdown", "id": "cell-0009", "metadata": {}, "source": [ "## Building the recommender\n", "\n", "Now the real thing: 15 movies, each with a genre + keyword + overview soup.\n", "Pipeline: `TfidfVectorizer` → similarity matrix → look up a title → return the\n", "five nearest neighbors (skipping the movie itself, which is always its own\n", "best match)." ] }, { "cell_type": "code", "id": "cell-0010", "metadata": {}, "execution_count": null, "outputs": [], "source": [ "import pandas as pd\n", "from sklearn.feature_extraction.text import TfidfVectorizer\n", "from sklearn.metrics.pairwise import cosine_similarity\n", "\n", "data = [\n", " (\"Toy Story\", \"animation family comedy toys friendship jealousy a cowboy doll feels threatened when a new spaceman action figure becomes the favorite toy\"),\n", " (\"Finding Nemo\", \"animation family adventure ocean fish father son a timid clownfish crosses the ocean to rescue his son captured by a diver\"),\n", " (\"The Incredibles\", \"animation family action superhero secret identity a family of undercover superheroes is forced back into action to save the world\"),\n", " (\"Shrek\", \"animation comedy fantasy ogre fairy tale swamp an ogre rescues a princess to reclaim his swamp from banished fairy tale creatures\"),\n", " (\"The Dark Knight\", \"action crime thriller superhero vigilante joker batman faces a criminal mastermind who plunges gotham city into anarchy and chaos\"),\n", " (\"Batman Begins\", \"action crime superhero origin fear training bruce wayne travels the world and returns to gotham to fight crime as batman\"),\n", " (\"Inception\", \"action science fiction heist dreams subconscious a thief who steals secrets from dreams is given a final job planting an idea\"),\n", " (\"Interstellar\", \"science fiction space wormhole time dilation family explorers travel through a wormhole to find humanity a new home among the stars\"),\n", " (\"The Martian\", \"science fiction space survival mars botany an astronaut stranded on mars must grow food and signal earth to survive\"),\n", " (\"Gravity\", \"science fiction space survival debris astronaut two astronauts drift in orbit after debris destroys their shuttle\"),\n", " (\"Titanic\", \"romance drama disaster ship iceberg class a poor artist and a rich young woman fall in love aboard a doomed ocean liner\"),\n", " (\"The Notebook\", \"romance drama love letters memory an elderly man reads the story of a lifelong summer love to a woman with dementia\"),\n", " (\"La La Land\", \"romance drama music jazz dreams ambition an aspiring actress and a jazz pianist chase their dreams and fall in love in los angeles\"),\n", " (\"Get Out\", \"horror thriller mystery hypnosis suspicion a young black man uncovers a sinister secret while visiting his girlfriend family estate\"),\n", " (\"A Quiet Place\", \"horror thriller monsters silence family a family must live in total silence to hide from creatures that hunt by sound\"),\n", "]\n", "df = pd.DataFrame(data, columns=[\"title\", \"soup\"])\n", "\n", "tfidf = TfidfVectorizer(stop_words=\"english\")\n", "matrix = tfidf.fit_transform(df[\"soup\"]) # shape: (15 movies, vocab)\n", "sim = cosine_similarity(matrix) # 15 x 15 similarity matrix\n", "indices = pd.Series(df.index, index=df[\"title\"]) # title -> row lookup\n", "\n", "def get_recommendations(title, k=5):\n", " idx = indices[title]\n", " scores = pd.Series(sim[idx], index=df[\"title\"])\n", " return scores.drop(title).sort_values(ascending=False).head(k)\n", "\n", "print(\"Vocabulary size:\", len(tfidf.get_feature_names_out()), \"\\\\n\")\n", "for title in [\"Toy Story\", \"The Dark Knight\", \"Titanic\"]:\n", " print(f\"Because you watched {title}:\")\n", " print(get_recommendations(title).round(3).to_string(), \"\\\\n\")" ] }, { "cell_type": "markdown", "id": "cell-0011", "metadata": {}, "source": [ "The clusters fall out beautifully: *Toy Story* pulls the other animated\n", "family films, *The Dark Knight* finds *Batman Begins* through shared words\n", "like \"gotham\", \"batman\", and \"superhero\", and *Titanic* lands in the romance\n", "corner. Nobody told the model about genres as a concept — the vocabulary\n", "overlap alone encodes it." ] }, { "cell_type": "markdown", "id": "cell-0012", "metadata": {}, "source": [ "> **CountVectorizer or TfidfVectorizer?**\n", "> \n", "> For free-text like synopses, TF-IDF is almost always better — it silences\n", "> filler words automatically. But for curated metadata (genre tags, cast names,\n", "> director), plain `CountVectorizer` is often the right call: every token was\n", "> chosen deliberately, and you may *not* want \"Steven Spielberg\" down-weighted\n", "> just because he directed many films. A common design: count-vectorize the\n", "> structured tags, TF-IDF the synopsis, and combine. You can also repeat\n", "> important tokens in the soup to weight them manually." ] }, { "cell_type": "markdown", "id": "cell-0013", "metadata": {}, "source": [ "## Strengths and the filter bubble\n", "\n", "What content-based filtering gets right:\n", "\n", "- **No other users needed.** It works from day one with a catalog of one user\n", " — similarity comes from item attributes, not crowd behavior. New *items*\n", " are no problem either: as soon as a movie has a synopsis, it can be\n", " recommended.\n", "- **It explains itself.** \"Recommended because you watched Toy Story — both\n", " are animated family films\" is a legible, trust-building reason.\n", "\n", "Its central weakness is **over-specialization**. The system can only\n", "recommend things similar to what you've already consumed. Watch three space\n", "movies and your homepage becomes an airlock — it will never discover that\n", "you'd also love jazz documentaries, because no vocabulary connects them. Users\n", "get trapped in a *filter bubble* of their own history. Breaking out requires\n", "information the item descriptions don't contain: what *other people* with\n", "tastes like yours enjoyed. That's collaborative filtering, next lesson." ] }, { "cell_type": "markdown", "id": "cell-0014", "metadata": {}, "source": [ "### 🏋️ Exercise — More like A Quiet Place\n", "\n", "Extend the 15-movie corpus with **two movies of your own** — one horror film\n", "described with words overlapping *A Quiet Place* (monsters, silence,\n", "survival…) and one family film. Rebuild the TF-IDF matrix and check\n", "`get_recommendations(\"A Quiet Place\")`: does your horror film appear in the\n", "top 5? Which shared words do you think made the match?" ] }, { "cell_type": "code", "id": "cell-0015", "metadata": {}, "execution_count": null, "outputs": [], "source": [ "# Your solution here" ] }, { "cell_type": "markdown", "id": "cell-0016", "metadata": {}, "source": [ "
✅ Show solution\n", "\n", "```python\n", "# Append these two movies to the data list from the lesson, then rerun:\n", "extra = [\n", " (\"Bird Box\", \"horror thriller monsters blindfold survival a mother and two children travel a river blindfolded to escape creatures that make you look\"),\n", " (\"Paddington\", \"family comedy adventure bear london marmalade a polite young bear from peru searches for a home with a kind london family\"),\n", "]\n", "data = data + extra\n", "df = pd.DataFrame(data, columns=[\"title\", \"soup\"])\n", "\n", "tfidf = TfidfVectorizer(stop_words=\"english\")\n", "matrix = tfidf.fit_transform(df[\"soup\"])\n", "sim = cosine_similarity(matrix)\n", "indices = pd.Series(df.index, index=df[\"title\"])\n", "\n", "print(get_recommendations(\"A Quiet Place\").round(3).to_string())\n", "# Bird Box should now rank near the top via shared words like\n", "# monsters / creatures / survival / family, while Paddington\n", "# joins the animated-family cluster instead.\n", "```\n", "\n", "
" ] }, { "cell_type": "markdown", "id": "cell-0017", "metadata": {}, "source": [ "Next up: collaborative filtering — dropping item descriptions entirely and\n", "learning taste from the ratings matrix itself." ] } ] }