yunjiaxi/multi-level-interaction-reranking

MIR

The implementation of reranking method proposed in Multi-Level Interaction Reranking with User Behavior History

Requirements

tensorflow-gpu >= 1.9.0,<2
numpy >= 1.16.4
scikit-learn >= 0.21.2
lightgbm >= 2.3.2

Get Started

Download data from Ad and PRM Public, and preprocess

python preprocess_ad.py
python preprocess_prm.py

Run initial ranker

python run_init_ranker.py

Run re-ranker

python run_mir.py

Model parameters can be set by using a config file, and specify its file path at --setting_path, e.g., python run_ranker.py --setting_path config. The config files for the different models can be found in config/ad/ and config/prm/. Moreover, model parameters can also be directly set from the command line. The supported parameters are listed as follows.

Parameters of run_init_ranker.py

argument	usage
--data_dir	The path to the directory where the data is stored
--save_dir	The path to the directory where the models and logs are stored
--model_type	The algorithm of reranker, including DNN, DIN, and LambdaMART PLEASE ATTENTION: Before training lambdaMART, you need to train DNN to get the pre-trained embedding
--setting_path	The path to the json config file, like files in example\config
--max_hist_len	The max length of history
--data_set_name	The name of the dataset, such as ad and prm
--epoch_num	The number of epoch for DNN model
--batch_size	Batch size for DNN model
--lr	Learning rate for DNN and lambdaMART
--l2_reg	The coefficient of l2 regularization for DNN model
--eb_dim	The size of embedding for DNN model
--tree_num	The number of trees for lambdaMART model
--tree_type	The type of tree for lambdaMART model, including lgb and sklearn

Parameters of run_mir.py

argument	usage
--data_dir	The path to the directory where the data is stored
--save_dir	The path to the directory where the models and logs are stored
--setting_path	The path to the json config file, like files in example\config
--data_set_name	The name of the dataset, such as ad and prm
--initial_ranker	The name of initial ranker, including DNN, lambdaMART.
--epoch_num	The number of epoch
--max_hist_len	The max length of history
--batch_size	Batch size
--lr	Learning rate
--l2_reg	The coefficient of l2 regularization
--eb_dim	The size of embedding
--hidden_size	The size of hidden unit, usually the hideen size of LSTM/GRU
--keep_prob	Keep prob in dropout
--metric_scope	The scope of metrics, for example when --metric_scope=[1, 3, 5], MAP@1, MAP@3, and MAP@5 will be computed
--max_norm	The max norm of gradient clip

Structure

Data processing

We process two datasets, Ad and PRM Public, containing user and item features with recommendation lists for the experimentation with personalized re-ranking.

preprocess_ad.py and preprocess_prm.py: process Ad and PRM Public, respectively.

Initial rankers

initial_model.py implements three initial ranking algorithms:

• DNN: a naive algorithm that directly train a multi-layer perceptron network with input labels (e.g., clicks).

• LambdaMART: the implementation of the LambdaMART model in From RankNet to LambdaRank to LambdaMART: An Overview

• DIN: the implementation of Deep Interest Network in Deep Interest Network for Click-Through Rate Prediction

run_init_ranker.py: the main function of initial ranker

Re-ranking algorithm

model.py: the implementation of our proposed reranking method, MIR.

run_mir.py: the main function of reranker.

multi-level-interaction-reranking's Issues

Permutation Invariant Question

Hi,

I am trying to reimplement your code, and I notice that in line 314, 315 in model.py, variable w_v and w_q has a dimension of length v_seq_len.

If a parameter in a model tries to remember some information regarding the position of the candidate list, then how does this model retain permutation invariant?

Thank you !