Implementation:Recommenders team Recommenders GRU Model

Knowledge Sources	Recommenders
Domains	Recommendation Systems, Deep Learning, Sequential Recommendation
Last Updated	2026-02-10 00:00 GMT

Overview

Implements a GRU (Gated Recurrent Unit) based sequential recommendation model that captures temporal dependencies in user behavior sequences.

Description

The GRUModel class extends SequentialBaseModel and implements _build_seq_graph for sequential recommendation using recurrent neural networks. The model concatenates item_history_embedding and cate_history_embedding into a combined history embedding, then processes this sequence through a GRUCell (from Keras) using TensorFlow's dynamic_rnn with sequence_length masking to handle variable-length sequences. The final_state of the GRU serves as the user representation, which is concatenated with the target_item_embedding to form the model output.

The class also contains an alternative _build_lstm method that follows the same architecture pattern but uses an LSTMCell instead, returning final_state[1] (the cell output) as the user representation. This LSTM variant is available but not used in the default _build_seq_graph flow.

The model is based on the work of Cho et al. (2014) on learning phrase representations using RNN encoder-decoder architectures, adapted for sequential recommendation.

Usage

Use this model as an RNN-based baseline for sequential recommendation when you want to capture temporal order in user interaction sequences. It provides a standard recurrent approach that can be compared against CNN-based (Caser) or attention-based (A2SVD, SLI_REC) alternatives.

Code Reference

Source Location

• Repository: Recommenders
• File: recommenders/models/deeprec/models/sequential/gru.py
• Lines: 1-80

Signature

class GRUModel(SequentialBaseModel):
    def _build_seq_graph(self)

    def _build_lstm(self)

    def _build_gru(self)

Import

from recommenders.models.deeprec.models.sequential.gru import GRUModel

I/O Contract

Inputs

Name	Type	Required	Description
self.item_history_embedding	tf.Tensor	Yes	Embedding tensor of the user's item interaction history
self.cate_history_embedding	tf.Tensor	Yes	Embedding tensor of the corresponding category history
self.target_item_embedding	tf.Tensor	Yes	Embedding tensor of the target item being scored
self.iterator.mask	tf.Tensor	Yes	Binary mask tensor indicating valid sequence positions
self.hidden_size	int	Yes	Dimensionality of the GRU hidden state

Outputs

Name	Type	Description
return (_build_seq_graph)	tf.Tensor	Concatenation of the GRU final hidden state and target item embedding
return (_build_gru)	tf.Tensor	Final hidden state of the GRU representing the user
return (_build_lstm)	tf.Tensor	Final cell output of the LSTM representing the user

Usage Examples

Basic Usage

from recommenders.models.deeprec.models.sequential.gru import GRUModel
from recommenders.models.deeprec.deeprec_utils import prepare_hparams

# Prepare hyperparameters from YAML config
hparams = prepare_hparams("recommenders/models/deeprec/config/gru.yaml")

# Create and train the GRU model
model = GRUModel(hparams, iterator_creator)
model.fit(train_file, valid_file)

# Evaluate the model
eval_results = model.run_eval(test_file)