Caser Tutorial: Convolutional Sequence Embedding Recommendation

Caser Tutorial: Convolutional Sequence Embedding Recommendation#

Sandbox Model — Experimental

This model is part of the CoreRec Sandbox — an experimental collection of algorithms for research and exploration. Sandbox models are not production-tested and may have incomplete implementations. For production-ready models, see the Production Models.

Introduction#

Caser treats user sequence as an “image” applying CNNs with horizontal (recent patterns) and vertical (skip patterns) convolutional filters.

Paper: Tang & Wang 2018 - Personalized Top-N Sequential Recommendation via Convolutional Sequence Embedding

How Caser Works#

Architecture#

CNN for Sequential Patterns:

  1. Sequence Embedding Matrix: L×d (sequence × embedding)

  2. Horizontal Convolution: Capture recent sequential patterns

  3. Vertical Convolution: Capture skip patterns (point-level)

  4. Pooling: Max pooling over all filters

  5. Fully Connected: Combine features for prediction

Key Innovation: CNN treats sequence as “image”

Architecture:

Sequence Matrix → [Horiz Conv] → Pool →                                          Concat → FC → Predict
                  [Vert Conv]  → Pool → /

Mathematical Foundation#

Horizontal Convolution (height h):

c^h = ReLU(W^h * E_{i:i+h} + b^h)

Captures last h items together

Vertical Convolution (width 1):

c^v = ReLU(W^v * E_i + b^v)

Captures single-point patterns

Output:

z = [max(c^h₁); ...; max(c^hₙ); max(c^v₁); ...; max(c^vₘ)]
y = FC(z)

Reference Implementation#

Implementation Note

The code examples below are reference implementations for learning purposes. This is a sandbox model and has not been production-tested. Verify behavior thoroughly before using in production.

Step 1: Import and Load Data#

from corerec.sandbox.collaborative_full.sequential_model_base.caser_base import Caser_base as Caser
from cr_learn import ml_1m
from sklearn.model_selection import train_test_split
import numpy as np

# Load dataset with CORRECT API
data = ml_1m.load()  # Returns dict with 'ratings', 'users', 'movies'
ratings_df = data['ratings']  # DataFrame with user_id, movie_id, rating, timestamp

print(f"Loaded {len(ratings_df)} ratings")

# Split data
train_df, test_df = train_test_split(ratings_df, test_size=0.2, random_state=42)

# Extract arrays for model
train_users = train_df['user_id'].values
train_items = train_df['movie_id'].values
train_ratings = train_df['rating'].values

test_users = test_df['user_id'].values
test_items = test_df['movie_id'].values
test_ratings = test_df['rating'].values

Step 2: Initialize Model#

model = Caser(
    name="Caser_Model",
    seq_len=10,
    horiz_filters=[2, 3, 4],
    num_filters=16,
    dropout=0.5,
    epochs=20,
    batch_size=256,
    learning_rate=0.001,
    verbose=True
)

print(f"Initialized {model.name}")

Step 3: Train#

model.fit(
    user_ids=train_users,
    item_ids=train_items,
    ratings=train_ratings
)

print("Training complete!")

Step 4: Predict#

# Single prediction
score = model.predict(user_id=1, item_id=100)
print(f"Predicted score: {score:.3f}")

# Batch predictions
test_predictions = model.batch_predict(list(zip(test_users[:100], test_items[:100])))

Step 5: Recommend#

# Get top-10 recommendations for user
user_id = 1
recommendations = model.recommend(
    user_id=user_id,
    top_k=10
)

print(f"Top-10 recommendations for User {user_id}:")
for rank, item_id in enumerate(recommendations, 1):
    print(f"  {rank}. Item {item_id}")

Step 6: Evaluate#

from sklearn.metrics import mean_squared_error
import numpy as np

# Predict all test ratings
test_pred = [model.predict(u, i) for u, i in zip(test_users, test_items)]
rmse = np.sqrt(mean_squared_error(test_ratings, test_pred))
print(f"Test RMSE: {rmse:.4f}")

Step 7: Save & Load#

# Save model
model.save('artifacts/caser_model')

# Load model
loaded = Caser.load('artifacts/caser_model')
test_score = loaded.predict(1, 100)
print(f"Loaded model prediction: {test_score:.3f}")

Key Takeaways#

When to Use Caser#

✅ Ideal For:

  • Short-medium sequences (5-50 items)

  • Local sequential patterns

  • Recent behavior matters most

  • E-commerce browsing sessions

  • When RNN/Transformer is overkill

❌ Not For:

  • Very long sequences (>100 items)

  • Global long-range dependencies

  • When position encoding crucial

  • Sparse interaction data

Best Practices#

  1. Sequence Length: L = 5-20 recent items

  2. Horizontal Filters: heights = [2, 3, 4, 5]

  3. Vertical Filters: 4-16 filters

  4. Filter Numbers: 16-64 per size

  5. Dropout: 0.5 on FC layer

  6. Embedding Dim: 50-100

  7. Activation: ReLU standard

  8. Pooling: Max pooling works best

Scaling & Production Considerations#

If you plan to move this sandbox model toward production use, consider:

  • Data Volume: Test with realistic dataset sizes; sandbox implementations may not be optimized for large-scale data.

  • Distributed Training: For datasets exceeding single-machine memory, consider wrapping the model in distributed training frameworks (e.g., PyTorch DDP, Horovod).

  • Hyperparameter Tuning: Use systematic search (Optuna, Ray Tune) rather than manual tuning. The defaults in sandbox models are not tuned for any specific domain.

  • Serving Latency: Profile inference time. Some architectures (e.g., attention-based) may need quantization or ONNX export for low-latency serving.

  • Monitoring: Implement A/B testing and online metrics (CTR, engagement, diversity) before replacing an existing system.

  • Validation: Write comprehensive unit tests and integration tests before deploying. Sandbox models do not have CI coverage.

Further Reading#

  • Paper: Tang & Wang 2018 - Personalized Top-N Sequential Recommendation via Convolutional Sequence Embedding

  • GitHub Repository

Contents