The main goal of reading paper is not just understanding it. Try to understand the key concept, but we need to get new ideas and research directions from the paper.
Paper information
title: Attention-based Mixture Density Recurrent Networks for History-based Recommendation
authors: Tian Wang and Kyunghyun Cho
venue: In: Proceedings of the 1st International Workshop on Deep Learning Practice for High-Dimensional Sparse Data. ACM, 2019. p. 5.
abstract: The goal of personalized history-based recommendation is to automatically output a distribution over all the items given a sequence of previous purchases of a user. In this work, we present a novel approach that uses a recurrent network for summarizing the history of purchases, continuous vectors representing items for scalability, and a novel attention-based recurrent mixture density network, which outputs each com- ponent in a mixture sequentially, for modelling a multi-modal conditional distribution. We evaluate the proposed approach on two publicly available datasets, MovieLens-20M and Rec- Sys15. The experiments show that the proposed approach, which explicitly models the multi-modal nature of the pre- dictive distribution, is able to improve the performance over various baselines in terms of precision, recall and nDCG.
Summary: problems to address, key ideas, quick results
They have revealed that the user preference is not static across time.
What you don't like?
How to improve?
They try to treat the user history as a sequence rather than a bag to model the user dynamic preference. However, I think this is not enough. The sequential data combined with time stamps would be more helpful.
Because timestamps can reflect some information. For example, if you buy a guitar at a certain time, you may buy some maintenance supplies for the guitar at intervals, such as strings, cleaning cloth, etc.
Any new ideas?
At present, many of the dynamic networks are done by snapshots. This snapshot is a collection of edges at many times. If we can refine to each moment and do it edge by edge, we should be able to make full use of information.
The main goal of reading paper is not just understanding it. Try to understand the key concept, but we need to get new ideas and research directions from the paper.
Paper information
Summary: problems to address, key ideas, quick results
presentation link
Questions about the paper?
What do you like?
They have revealed that the user preference is not static across time.
What you don't like?
How to improve?
They try to treat the user history as a sequence rather than a bag to model the user dynamic preference. However, I think this is not enough. The sequential data combined with time stamps would be more helpful. Because timestamps can reflect some information. For example, if you buy a guitar at a certain time, you may buy some maintenance supplies for the guitar at intervals, such as strings, cleaning cloth, etc.
Any new ideas?
At present, many of the dynamic networks are done by snapshots. This snapshot is a collection of edges at many times. If we can refine to each moment and do it edge by edge, we should be able to make full use of information.
Reproducing results (if any)