Sequential Recommender Systems

Sequential Recommender Systems
By Aryan December 10, 2024 min read

Sequential Recommender Systems tries to recommend items based on changing / evolving behaviour of the user and the consumption trends. One simple example for this like a person buying printers will have a need of printer inks after next few months.

Conventional content-based and collaborative filtering recommendation system wont be effective in sequential recommendation problems because they wont be considering the current state for recommendation. Now the current state can be dependent on some of the following contexts :

  • Choice or situation of the user

  • Market popularity of the Item at that point

  • Price of the Item at that point

Algorithms for Solving Sequential Recommendation Problems (SRPs)

Learning long time sequential dependency and high order sequential dependency are tougher problems in SRPs.

Below is a table wise summary of various popular sequential recommender systems.


Characteristics

Example

Solution

Long User-Item Interaction Sequences

User Going for Vacation and order for multiple things

Higher Order Markov Chains

User-Item Interaction with flexible Order Sequences

Purchase of Butter Milk and Flour

CNN

User-Item Interaction with Noise

Purchasing of bacon, a rose, eggs, bread. Here is rose like noise item.

Attention Models and Memory Networks

User-Item Interaction with heterogeneous relations

Purchasing heterogeneous items

Mixture Models

User-Item Interaction with hierarchical Structures

Learning hierarchical Items

Hierarchical RNN, Hierarchical Attention

Reference : https://arxiv.org/pdf/2001.04830.pdf


That’s all for this blog, next we will see implementation of some of the above algorithms and will deep dive into more details. Until then stay tuned!


About the Author

Aryan

Machine Learning Expert

Aryan is the Machine Learning Expert at FutureWebAI, specializing in developing and implementing advanced machine learning models and AI-driven solutions. With deep expertise in data science, algorithm optimization, and neural networks, Aryan is dedicated to pushing the boundaries of what AI can achieve.