NextRec

中文版

A Unified, Efficient, and Scalable Recommendation System Framework

Introduction

NextRec is a modern recommendation system framework built on PyTorch, providing a unified modeling, training, and evaluation experience for researchers and engineering teams. The framework adopts a modular design with rich built-in model implementations, data-processing tools, and production-ready training components, enabling quick coverage of multiple recommendation scenarios.

This project draws on several open-source recommendation libraries, with the general layers referencing the mature implementations in torch-rechub. These part of codes is still in its early stage and is being gradually replaced with our own implementations. If you find any bugs, please submit them in the issue section. Contributions are welcome.

Key Features

• Multi-scenario Recommendation: Supports ranking (CTR/CVR), retrieval, multi-task learning, and generative recommendation models such as TIGER and HSTU — with more models continuously added.
• Unified Feature Engineering & Data Pipeline: Provides Dense/Sparse/Sequence feature definitions, persistent DataProcessor, and optimized RecDataLoader, forming a complete “Define → Process → Load” workflow.
• Efficient Training & Evaluation: A standardized training engine with optimizers, LR schedulers, early stopping, checkpoints, and logging — ready out-of-the-box.
• Developer-friendly Engineering Experience: Modular and extensible design, full tutorial support, GPU/MPS acceleration, and visualization tools.

---

Installation

NextRec supports installation via UV or traditional pip/source installation.

Option 1: Using UV (Recommended)

UV is a modern, high-performance Python package manager offering fast dependency resolution and installation.

git clone https://github.com/zerolovesea/NextRec.git
cd NextRec

# Install UV if not already installed
pip install uv

# Create virtual environment and install dependencies
uv sync

# Activate the virtual environment
source .venv/bin/activate  # macOS/Linux
# or
.venv\Scripts\activate     # Windows

# Install the package in editable mode
uv pip install -e .

Note: Make sure to deactivate any other conda/virtual environments before running uv sync to avoid environment conflicts.

Option 2: Using pip/source installation

git clone https://github.com/zerolovesea/NextRec.git
cd NextRec

# Install dependencies
pip install -r requirements.txt
pip install -r test_requirements.txt

# Install the package in editable mode
pip install -e .

---

5-Minute Quick Start

The following example demonstrates a full DeepFM training & inference pipeline using the MovieLens dataset:

import pandas as pd

from nextrec.models.ranking.deepfm import DeepFM
from nextrec.basic.features import DenseFeature, SparseFeature, SequenceFeature

df = pd.read_csv("dataset/movielens_100k.csv")

target = 'label'
dense_features = [DenseFeature('age')]
sparse_features = [
    SparseFeature('user_id', vocab_size=df['user_id'].max()+1, embedding_dim=4),
    SparseFeature('item_id', vocab_size=df['item_id'].max()+1, embedding_dim=4),
]

sparse_features.append(SparseFeature('gender', vocab_size=df['gender'].max()+1, embedding_dim=4))
sparse_features.append(SparseFeature('occupation', vocab_size=df['occupation'].max()+1, embedding_dim=4))

model = DeepFM(
    dense_features=dense_features,
    sparse_features=sparse_features,
    mlp_params={"dims": [256, 128], "activation": "relu", "dropout": 0.5},
    target=target,
    device='cpu',
    model_id="deepfm_with_processor",
    embedding_l1_reg=1e-6,
    dense_l1_reg=1e-5,
    embedding_l2_reg=1e-5,
    dense_l2_reg=1e-4,
)

model.compile(optimizer="adam", optimizer_params={"lr": 1e-3, "weight_decay": 1e-5}, loss="bce")
model.fit(train_data=df, metrics=['auc', 'recall', 'precision'], epochs=10, batch_size=512, shuffle=True, verbose=1)
preds = model.predict(df)
print(f'preds: {preds}')

More Tutorials

The tutorials/ directory provides examples for ranking, retrieval, multi-task learning, and data processing:

• movielen_match_dssm.py — DSSM retrieval on MovieLens 100k
• movielen_ranking_deepfm.py — DeepFM ranking on MovieLens 100k
• example_ranking_din.py — DIN (Deep Interest Network) example
• example_match_dssm.py — DSSM retrieval example
• example_multitask.py — Multi-task learning example

---

Data Processing Example

NextRec offers a unified interface for preprocessing sparse and sequence features:

import pandas as pd
from nextrec.data.preprocessor import DataProcessor

df = pd.read_csv("dataset/movielens_100k.csv")

processor = DataProcessor()
processor.add_sparse_feature('movie_title', encode_method='hash', hash_size=1000)
processor.fit(df)

df = processor.transform(df, return_dict=False)

print("\nSample training data:")
print(df.head())

---

Supported Models

Ranking Models

Model	Paper	Year	Status
FM	Factorization Machines	ICDM 2010	Supported
AFM	Attentional Factorization Machines: Learning the Weight of Feature Interactions via Attention Networks	IJCAI 2017	Supported
DeepFM	DeepFM: A Factorization-Machine based Neural Network for CTR Prediction	IJCAI 2017	Supported
Wide&Deep	Wide & Deep Learning for Recommender Systems	DLRS 2016	Supported
xDeepFM	xDeepFM: Combining Explicit and Implicit Feature Interactions	KDD 2018	Supported
FiBiNET	FiBiNET: Combining Feature Importance and Bilinear Feature Interaction for CTR Prediction	RecSys 2019	Supported
PNN	Product-based Neural Networks for User Response Prediction	ICDM 2016	Supported
AutoInt	AutoInt: Automatic Feature Interaction Learning	CIKM 2019	Supported
DCN	Deep & Cross Network for Ad Click Predictions	ADKDD 2017	Supported
DIN	Deep Interest Network for CTR Prediction	KDD 2018	Supported
DIEN	Deep Interest Evolution Network	AAAI 2019	Supported
MaskNet	MaskNet: Feature-wise Gating Blocks for High-dimensional Sparse Recommendation Data	2020	Supported

Retrieval Models

Model	Paper	Year	Status
DSSM	Learning Deep Structured Semantic Models	CIKM 2013	Supported
DSSM v2	DSSM with pairwise BPR-style optimization	-	Supported
YouTube DNN	Deep Neural Networks for YouTube Recommendations	RecSys 2016	Supported
MIND	Multi-Interest Network with Dynamic Routing	CIKM 2019	Supported
SDM	Sequential Deep Matching Model	-	Supported

Multi-task Models

Model	Paper	Year	Status
MMOE	Modeling Task Relationships in Multi-task Learning	KDD 2018	Supported
PLE	Progressive Layered Extraction	RecSys 2020	Supported
ESMM	Entire Space Multi-task Model	SIGIR 2018	Supported
ShareBottom	Multitask Learning	-	Supported

Generative Models

Model	Paper	Year	Status
TIGER	Recommender Systems with Generative Retrieval	NeurIPS 2023	In Progress
HSTU	Hierarchical Sequential Transduction Units	-	In Progress

---

Contributing

We welcome contributions of any form!

How to Contribute

1. Fork the repository
2. Create your feature branch (git checkout -b feature/AmazingFeature)
3. Commit your changes (git commit -m 'Add AmazingFeature')
4. Push your branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

Before submitting a PR, please run tests using pytest test/ -v or python -m pytest to ensure everything passes.

Code Style

• Follow PEP8
• Provide unit tests for new functionality
• Update documentation accordingly

Reporting Issues

When submitting issues on GitHub, please include:

• Description of the problem
• Reproduction steps
• Expected behavior
• Actual behavior
• Environment info (Python version, PyTorch version, etc.)

---

License

This project is licensed under the Apache 2.0 License.

---

Contact

• GitHub Issues: Submit issues on GitHub
• Email: zyaztec@gmail.com

---

Acknowledgements

NextRec is inspired by the following great open-source projects:

• torch-rechub - A Lighting Pytorch Framework for Recommendation Models, Easy-to-use and Easy-to-extend.
• FuxiCTR — Configurable and reproducible CTR prediction library
• RecBole — Unified and efficient recommendation library
• PaddleRec — Large-scale recommendation algorithm library

Special thanks to all open-source contributors!

---

Back to Top