What is the best way to manage repetitveness of recommendations in a tensorflow recommender system?

Research & Models
#1

In a recommender system, what is the best way to manage repetiveness of items being recommended?

This may come about, for example, if the user provides no additional input and so the same items are recommended. In some cases, re-recommended items are desirable to give the user more time to consider the item.

What is the best way to manage the trade off between items being under and over recommended in a tensorflow recommender system? Do you manage this in tensorflow itself or do you need a custom filter afterwards?

#3

Hi @James_Hope ,

Here is my thought about under- and over-recommending in a tensorflow recommender system.

Regarding the tradeoff between under and over-recommending, there is no one-size-fits-all approach as it depends on the specific use case and user preferences. In some cases, re-recommending items may be desirable to give the user more time to consider the item, as you mentioned. In other cases, it may be more important to avoid over-recommending items to prevent user fatigue or annoyance. It is important to monitor user feedback and adjust the recommender system accordingly.

These techniques can be implemented in TensorFlow using custom loss functions, sampling strategies, or using libraries such as TensorFlow Recommenders. Whether a custom filter is needed afterwards depends on the specific use case and how well the techniques are able to manage repetitiveness.

Please let me know your opinion on the above thought.

Thanks.

#4
  1. Use post-processing techniques to filter out items that are too similar to each other. This can be done by clustering similar items and selecting representative items from each cluster, or by applying a diversity score to the recommended items and removing those that fall below a certain threshold.
#6

First of all, collect the explicit or implicit feedback from users regarding the recommended they receive. This feedback can be used to improve the recommender system and reduce repetitiveness. such as you can incorporate feedback signals like ratings, clicks, or dwell time to adapt the recommendation to individual prefrences.