# Case Study: Content-Based Recommendations
Build a recommendation engine using text similarity and HNSW indexing.
## Use Case
Find similar movies based on plot descriptions.
## Implementation
```rust,ignore
use aprender::recommend::ContentRecommender;
fn main() {
// Create recommender with HNSW parameters:
// - M=16: connections per node
// - ef_construction=200: build quality
// - decay_factor=0.95: IDF decay
let mut recommender = ContentRecommender::new(16, 200, 0.95);
// Add movie descriptions
let movies = vec![
("inception", "A thief steals secrets through dream-sharing technology"),
("matrix", "A hacker discovers reality is a simulation"),
("interstellar", "Astronauts travel through a wormhole to save humanity"),
("avatar", "A marine explores an alien world called Pandora"),
("terminator", "A cyborg assassin is sent back in time"),
("blade_runner", "A detective hunts rogue replicants in dystopian future"),
];
for (id, description) in &movies {
recommender.add_item(id, description);
}
// Build the index
recommender.build_index();
// Find similar movies
let query = "science fiction about artificial intelligence and reality";
let recommendations = recommender.recommend(query, 3);
println!("Query: {}\n", query);
println!("Recommendations:");
for (id, score) in recommendations {
println!(" {} (score: {:.3})", id, score);
}
}
```
Output:
```text
Query: science fiction about artificial intelligence and reality
Recommendations:
matrix (score: 0.847)
blade_runner (score: 0.723)
terminator (score: 0.691)
```
## How It Works
1. **TF-IDF Vectorization**: Convert descriptions to sparse vectors
2. **Incremental IDF**: Update vocabulary as items are added
3. **HNSW Index**: Fast approximate nearest neighbor search
4. **Cosine Similarity**: Rank by vector similarity
## Key Features
- **Incremental updates**: Add items without rebuilding
- **Scalable**: HNSW provides O(log n) search
- **No training required**: Pure content-based filtering
## Run
```bash
cargo run --example recommend_content
```