Recommender systems that can learn from cross-session data to dynamically
predict the next item a user will choose are crucial for online platforms.
However, existing approaches often use out-of-the-box sequence models which are
limited by speed and memory consumption, are often infeasible for production
environments, and usually do not incorporate cross-session information, which
is crucial for effective recommendations. Here we propose Hierarchical Temporal
Convolutional Networks (HierTCN), a hierarchical deep learning architecture
that makes dynamic recommendations based on users' sequential multi-session
interactions with items. HierTCN is designed for web-scale systems with
billions of items and hundreds of millions of users. It consists of two levels
of models: The high-level model uses Recurrent Neural Networks (RNN) to
aggregate users' evolving long-term interests across different sessions, while
the low-level model is implemented with Temporal Convolutional Networks (TCN),
utilizing both the long-term interests and the short-term interactions within
sessions to predict the next interaction. We conduct extensive experiments on a
public XING dataset and a large-scale Pinterest dataset that contains 6 million
users with 1.6 billion interactions. We show that HierTCN is 2.5x faster than
RNN-based models and uses 90% less data memory compared to TCN-based models. We
further develop an effective data caching scheme and a queue-based mini-batch
generator, enabling our model to be trained within 24 hours on a single GPU.
Our model consistently outperforms state-of-the-art dynamic recommendation
methods, with up to 18% improvement in recall and 10% in mean reciprocal rank.
%0 Conference Paper
%1 you2019hierarchical
%A You, Jiaxuan
%A Wang, Yichen
%A Pal, Aditya
%A Eksombatchai, Pong
%A Rosenberg, Chuck
%A Leskovec, Jure
%B Proceedings of the international conference on the World Wide Web.
%D 2019
%K cnn hierachical item recommendation rnn system temporal
%T Hierarchical Temporal Convolutional Networks for Dynamic Recommender Systems
%U http://arxiv.org/abs/1904.04381
%X Recommender systems that can learn from cross-session data to dynamically
predict the next item a user will choose are crucial for online platforms.
However, existing approaches often use out-of-the-box sequence models which are
limited by speed and memory consumption, are often infeasible for production
environments, and usually do not incorporate cross-session information, which
is crucial for effective recommendations. Here we propose Hierarchical Temporal
Convolutional Networks (HierTCN), a hierarchical deep learning architecture
that makes dynamic recommendations based on users' sequential multi-session
interactions with items. HierTCN is designed for web-scale systems with
billions of items and hundreds of millions of users. It consists of two levels
of models: The high-level model uses Recurrent Neural Networks (RNN) to
aggregate users' evolving long-term interests across different sessions, while
the low-level model is implemented with Temporal Convolutional Networks (TCN),
utilizing both the long-term interests and the short-term interactions within
sessions to predict the next interaction. We conduct extensive experiments on a
public XING dataset and a large-scale Pinterest dataset that contains 6 million
users with 1.6 billion interactions. We show that HierTCN is 2.5x faster than
RNN-based models and uses 90% less data memory compared to TCN-based models. We
further develop an effective data caching scheme and a queue-based mini-batch
generator, enabling our model to be trained within 24 hours on a single GPU.
Our model consistently outperforms state-of-the-art dynamic recommendation
methods, with up to 18% improvement in recall and 10% in mean reciprocal rank.
@inproceedings{you2019hierarchical,
abstract = {Recommender systems that can learn from cross-session data to dynamically
predict the next item a user will choose are crucial for online platforms.
However, existing approaches often use out-of-the-box sequence models which are
limited by speed and memory consumption, are often infeasible for production
environments, and usually do not incorporate cross-session information, which
is crucial for effective recommendations. Here we propose Hierarchical Temporal
Convolutional Networks (HierTCN), a hierarchical deep learning architecture
that makes dynamic recommendations based on users' sequential multi-session
interactions with items. HierTCN is designed for web-scale systems with
billions of items and hundreds of millions of users. It consists of two levels
of models: The high-level model uses Recurrent Neural Networks (RNN) to
aggregate users' evolving long-term interests across different sessions, while
the low-level model is implemented with Temporal Convolutional Networks (TCN),
utilizing both the long-term interests and the short-term interactions within
sessions to predict the next interaction. We conduct extensive experiments on a
public XING dataset and a large-scale Pinterest dataset that contains 6 million
users with 1.6 billion interactions. We show that HierTCN is 2.5x faster than
RNN-based models and uses 90% less data memory compared to TCN-based models. We
further develop an effective data caching scheme and a queue-based mini-batch
generator, enabling our model to be trained within 24 hours on a single GPU.
Our model consistently outperforms state-of-the-art dynamic recommendation
methods, with up to 18% improvement in recall and 10% in mean reciprocal rank.},
added-at = {2019-04-14T21:54:23.000+0200},
author = {You, Jiaxuan and Wang, Yichen and Pal, Aditya and Eksombatchai, Pong and Rosenberg, Chuck and Leskovec, Jure},
biburl = {https://www.bibsonomy.org/bibtex/28cb4a302e87a734bb524816a9b19ec9d/nosebrain},
booktitle = {Proceedings of the international conference on the World Wide Web.},
interhash = {82aa2f448c0a3adbeb55a7fb1b0ec1cd},
intrahash = {8cb4a302e87a734bb524816a9b19ec9d},
keywords = {cnn hierachical item recommendation rnn system temporal},
timestamp = {2019-04-14T21:56:46.000+0200},
title = {Hierarchical Temporal Convolutional Networks for Dynamic Recommender Systems},
url = {http://arxiv.org/abs/1904.04381},
year = 2019
}