BST Tutorial: Behavior Sequence Transformer

BST Tutorial: Behavior Sequence Transformer#

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#

BST applies transformer architecture to user behavior sequences with target item attention for personalized recommendation.

Paper: Chen et al. 2019 - Behavior Sequence Transformer for E-commerce Recommendation

How BST Works#

Architecture#

Transformer for E-commerce:

  1. Behavior Sequence: User’s historical actions

  2. Multi-Head Self-Attention: Capture item dependencies

  3. Target Attention: Attend based on candidate item

  4. Position Encoding: Capture sequence order

  5. Embedding Concat: Combine sequence + features

Key Innovation: Target-aware attention (like DIN but with Transformer)

Architecture:

[Seq] → Transformer → Target Attn → [Other Features] → MLP → Pred
                          ↑
                     Candidate Item

Mathematical Foundation#

Self-Attention:

H = MultiHead(V_seq)

Target Attention:

α_i = softmax(h_i^T W_q e_target / √d)
v_user = Σ α_i · h_i

Final Prediction:

features = [v_user; e_target; context]
ŷ = MLP(features)

Positional Encoding:

PE(pos) = LearnableEmbed(pos)

Learned, not sinusoidal

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.nn_base.BST_base import BST_base as BST
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 = BST(
    name="BST_Model",
    seq_len=20,
    hidden_dim=64,
    num_layers=2,
    num_heads=2,
    use_target_attention=True,
    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/bst_model')

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

Key Takeaways#

When to Use BST#

✅ Perfect For:

  • E-commerce with rich sequences

  • CTR prediction with behavior history

  • Target-aware recommendations

  • Medium-length sequences (10-50 items)

  • When you need attention visualization

❌ Not For:

  • Very short sequences (<5 items)

  • Pure collaborative filtering

  • Real-time constraints (slower than GRU)

  • When SASRec already works

Best Practices#

  1. Sequence Length: 10-50 recent items

  2. Transformer Layers: 1-2 layers (shallow!)

  3. Attention Heads: 1-2 heads

  4. Hidden Dim: 64-128

  5. Target Attention: Essential for performance

  6. Dropout: 0.1-0.3

  7. Position Encoding: Learnable works better

  8. Features: Combine with user/item features

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: Chen et al. 2019 - Behavior Sequence Transformer for E-commerce Recommendation

  • GitHub Repository

Contents