shenweichen/DeepCTR
8,039 estrellas·2,223 forks·Python·Apache-2.0·6 vistasdeepctr-doc.readthedocs.io/en/latest/index.html↗
DeepCTR
DeepCTR is a specialized software framework and deep learning model library designed for predicting click-through rates and implementing recommendation systems. It provides a suite of tabular data models and architectures tailored for binary classification and sparse feature processing.
The framework includes dedicated toolkits for multi-task learning and sequential interest modeling. It allows for the simultaneous estimation of multiple related targets through shared-bottom and gated expert neural networks, while capturing evolving user behavior using attention mechanisms and transformers.
The library covers a broad range of capabilities, including sparse feature engineering, user behavior modeling, and the implementation of various neural network architectures for tabular data. These are supported by modular components for feature interaction, embedding-based representations, and sequence-pooling aggregation.
Features
- Click-Through Rate Predictors - Provides deep learning architectures specifically for estimating click probability using feature interactions and behavior sequences.
- Deep Learning Libraries - Offers a modular collection of deep learning implementations, including DeepFM and xDeepFM, tailored for tabular data.
- Sparse and Dense Feature Declarations - Handles high-cardinality categorical data through sparse and dense feature declarations and embedding layers.
- Feature Interaction Models - Implements neural network architectures specifically designed to learn complex high-order relationships between input features.
- Multi-Task Learning Models - Implements model architectures that predict multiple related targets, such as click and conversion rates, simultaneously.
- Recommender Systems - Builds deep learning models designed to predict user preferences and rank items for personalized recommendations.
- User Behavior - Models evolving user interests by processing historical interaction sequences with attention and recurrent structures.
- Tabular Feature Embeddings - Maps high-cardinality categorical and numerical tabular features into a shared high-dimensional latent space via embeddings.
- Tabular Predictive Models - Provides a suite of predictive models for structured tabular data using sparse and dense feature processing.
- Gated Expert Routing Layers - Uses gated expert routing layers to combine shared expert modules for multiple target tasks.
- Shared-Bottom Architectures - Implements shared-bottom architectures to estimate multiple related targets using shared neural representations.
- Modular Layer Assembly - Allows assembling complex models using reusable, pre-optimized blocks like cross-networks and multi-head attention.
- Linear-Deep Hybrids - Combines linear components for feature memorization with deep neural networks for high-order generalization.
- Pooling Mechanisms - Implements pooling mechanisms to reduce variable-length user behavior sequences into fixed-length vectors.
- Sequential - Processes user behavior sequences using attention and transformers to capture evolving interests for personalized predictions.
