4

I'm currently creating a recommender system and there are different types of the systems.

Does anyone know something about the user-to-item and item-to-user recommendations? I have been searching for days, but I couldn't find anything about them. I have only found info about the user-to-user and item-to-item ones.

desertnaut
  • 1,988
  • 2
  • 14
  • 23
ayiram
  • 149
  • 5

2 Answers2

1

You might try several techniques to achieve your goal, I have researched and used these before, so it might help you.

Semi-Supervised Learning

Basically you need some historical data linking the products the users, therefore you can establish a primary knowledge of the relationship between some products and the users after that you cluster the same products/users and you label the rest of unlabeled products based on that.

Association rule learning

is a rule-based machine learning method for discovering interesting relations between variables in large databases. It is intended to identify strong rules discovered in databases using some measures of interestingness.

Collaborative learning/Federated learning

It's better to use it when you have informations about the buyers/sellers, therefore you can study the relationship between the product/(buyers/sellers), then products/shops, then you can supervise the learning between the products/shops.You need a high content quality data.

Cooperative Learning

You run diffrent test on diffrent combination of users/items, and you derive diffrent results from them, then you compare them against each other. For this you need some pre-determined rules to evaluate your own decision making results. You need some historical data of the products/users.

Combine collaborative learning and Cooperative Learning

This is what Google search engine is doing, he know what other people are looking for and build patterns for which is called collaborative learning, but if they give the results of it, it will not be specific for you as a user, so they add Cooperative Learning to emphasize your own search results experience. As interesting as it might sound it not that easy to code.

MXK
  • 184
  • 1
  • 15
0

The mainstream recommender systems use user-to-item scheme. It's natural in the sense that when a user visit, the returned items are ones that the user prefer most. Item-to-user systems are particular inefficient due to the following reasons:

  1. Most items may be directed to a small group of users who may not visit the system or have enough sessions.
  2. If an item find a user, there could be items that the user prefer more.
Jianfei Shen
  • 1
  • 1