aforever / ESRT

A collection of product search embedding models

Overview

This is a collecetion of implements of following models, but it still under development:

• [HEM] Qingyao Ai, Yongfeng Zhang, Keping Bi, Xu Chen, W. Bruce Croft. 2017. Learning a Hierarchical Embedding Model for Personalized ProductSearch. In Proceedings of SIGIR ’17
• [AEM] Qingyao Ai, Daniel Hill, Vishy Vishwanathan and W. Bruce Croft. A Zero Attention Model for Personalized Product Search. Accepted in Proceedings of the 28th ACM International Conference on Information and Knowledge Management (CIKM’19)
• [DREM] Qingyao Ai, Yongfeng Zhang, Keping Bi, W. Bruce Croft. Explainable Product Search with a Dynamic Relation Embedding Model. ACM Transactions on Information Systems (TOIS). 2019

These models are deep neural network models that jointly learn latent representations for queries, products and users and knowledge entites(DREM). They are designed as generative models and the embedding representations for queries, users and items in the models are learned through optimizing the log likelihood of observed user-query-item purchases. The probability (which is also the rank score) of an item being purchased by a user with a query can be computed with their corresponding latent representations. Please refer to the paper for more details.

Requirements:

1. To run the models, python 2.7+ and Tensorflow v1.0+ are needed. (In the paper, we used python 2.7.12 and Tensorflow v2.0.0)
2. To run the jar package in ./seq_utils/AmazonDataset/jar/, JDK 1.7 is needed

Run Models

Data preparation

cd experiment/
bash data_preprocess.sh

Train model

bash hem_run.sh # run hem model
# or
bash aem_run.sh # run aem model
# or
bash drem_run.sh # run drem model

Test model

bash hem_test.sh # run hem model
# or
bash aem_test.sh # run aem model
# or
bash drem_test.sh # run drem model

Evaluate model

bash hem_metric.sh # run hem model
# or
bash aem_metric.sh # run aem model
# or
bash drem_metric.sh # run drem model

Example Parameter Setting

Hyper-parameters	HEM	AEM	DREM
subsampling_rate	0.0001	0.0001	0.0001
max_train_epoch	20	20	20
rank_cutoff	100	100	100
window_size	5	5	5
embed_size	100	100	100
max_gradient_norm	5.0	5.0	5.0
init_learning_rate	0.5	0.5	0.5
L2_lambda	0.005	0.005	0.005
query_weight	0.5	0.5	0.5
negative_sampele	5	5	5
net_struct	"simplified_fs"	"simplified_fs"	"simplified_fs"
similarity_func	"bias_product"	"bias_product"	"bias_product"
batch_size	64	64	64
user_struct		"asin_attention"
num_heads		5
attention_func		'default'
max_history_length		10

Result

Models	MRR	NCDG10	P10
HEM	0.073	0.083	0.016
AEM	0.081	0.091	0.018
DREM	0.101	0.114	0.021