Stay organized with collections
Save and categorize content based on your preferences.
In the final stage of a recommendation system, the system can re-rank the
candidates to consider additional criteria or constraints. One
re-ranking approach is to use filters that remove some candidates.
Another re-ranking approach is to manually transform the score returned
by the ranker.
This section briefly discusses freshness, diversity, and fairness.
These factors are among many that can help improve your recommendation
system. Some of these factors often require modifying different stages
of the process. Each section offers solutions that you might apply
individually or collectively.
Freshness
Most recommendation systems aim to incorporate the latest usage information,
such as current user history and the newest items. Keeping the model fresh
helps the model make good recommendations.
Solutions
Re-run training as often as possible to learn on the latest training data.
We recommend warm-starting the training so that the model does not have
to re-learn from scratch. Warm-starting can significantly reduce training
time. For example, in matrix factorization, warm-start the embeddings for
items that were present in the previous instance of the model.
Create an "average" user to represent new users in matrix factorization
models. You don't need the same embedding for each user—you
can create clusters of users based on user features.
Use a DNN such as a softmax model or two-tower model. Since the model takes
feature vectors as input, it can be run on a query or item that was not
seen during training.
Add document age as a feature. For example, YouTube can add a video's age
or the time of its last viewing as a feature.
Diversity
If the system always recommend items that are "closest" to the query
embedding, the candidates tend to be very similar to each other. This
lack of diversity can cause a bad or boring user experience. For example,
if YouTube just recommends videos very similar to the video the user is
currently watching, such as nothing but owl videos
(as shown in the illustration), the user will likely lose interest quickly.
Solutions
Train multiple candidate generators using different sources.
Train multiple rankers using different objective functions.
Re-rank items based on genre or other metadata to ensure diversity.
Fairness
Your model should treat all users fairly. Therefore, make sure
your model isn't learning unconscious biases from the training data.
Solutions
Include diverse perspectives in design and development.
Train ML models on comprehensive data sets. Add auxiliary data when
your data is too sparse (for example, when certain categories are
under-represented).
Track metrics (for example, accuracy and absolute error) on each
demographic to watch for biases.
[null,null,["Last updated 2025-08-25 UTC."],[[["\u003cp\u003eRecommendation systems can be improved by re-ranking candidates using filters or score transformations based on criteria like video age or click-bait detection.\u003c/p\u003e\n"],["\u003cp\u003eMaintaining freshness in recommendations involves regularly retraining models, incorporating new user and item data, and potentially using DNNs or age-based features.\u003c/p\u003e\n"],["\u003cp\u003eEnhancing diversity can be achieved by employing multiple candidate generators and rankers, along with re-ranking based on metadata like genre.\u003c/p\u003e\n"],["\u003cp\u003eEnsuring fairness requires diverse development teams, comprehensive training data, and monitoring metrics across demographics to mitigate biases.\u003c/p\u003e\n"]]],[],null,["# Re-ranking\n\n\u003cbr /\u003e\n\nIn the final stage of a recommendation system, the system can re-rank the\ncandidates to consider additional criteria or constraints. One\nre-ranking approach is to use filters that remove some candidates.\n| **Example:** You can implement re-ranking on a video recommender by doing the following:\n|\n| 1. Training a separate model that detects whether a video is click-bait.\n| 2. Running this model on the candidate list.\n| 3. Removing the videos that the model classifies as click-bait.\n\nAnother re-ranking approach is to manually transform the score returned\nby the ranker.\n| **Example:** The system re-ranks videos by modifying the score as a function of:\n|\n| - video age (perhaps to promote fresher content)\n| - video length\n\nThis section briefly discusses freshness, diversity, and fairness.\nThese factors are among many that can help improve your recommendation\nsystem. Some of these factors often require modifying different stages\nof the process. Each section offers solutions that you might apply\nindividually or collectively.\n\nFreshness\n---------\n\nMost recommendation systems aim to incorporate the latest usage information,\nsuch as current user history and the newest items. Keeping the model fresh\nhelps the model make good recommendations.\n\n### Solutions\n\n- Re-run training as often as possible to learn on the latest training data. We recommend warm-starting the training so that the model does not have to re-learn from scratch. Warm-starting can significantly reduce training time. For example, in matrix factorization, warm-start the embeddings for items that were present in the previous instance of the model.\n- Create an \"average\" user to represent new users in matrix factorization models. You don't need the same embedding for each user---you can create clusters of users based on user features.\n- Use a DNN such as a softmax model or two-tower model. Since the model takes feature vectors as input, it can be run on a query or item that was not seen during training.\n- Add document age as a feature. For example, YouTube can add a video's age or the time of its last viewing as a feature.\n\nDiversity\n---------\n\nIf the system always recommend items that are \"closest\" to the query\nembedding, the candidates tend to be very similar to each other. This\nlack of diversity can cause a bad or boring user experience. For example,\nif YouTube just recommends videos very similar to the video the user is\ncurrently watching, such as nothing but owl videos\n(as shown in the illustration), the user will likely lose interest quickly.\n\n### Solutions\n\n- Train multiple candidate generators using different sources.\n- Train multiple rankers using different objective functions.\n- Re-rank items based on genre or other metadata to ensure diversity.\n\nFairness\n--------\n\nYour model should treat all users fairly. Therefore, make sure\nyour model isn't learning unconscious biases from the training data.\n\n### Solutions\n\n- Include diverse perspectives in design and development.\n- Train ML models on comprehensive data sets. Add auxiliary data when your data is too sparse (for example, when certain categories are under-represented).\n- Track metrics (for example, accuracy and absolute error) on each demographic to watch for biases.\n- Make separate models for underserved groups."]]
