Cold-start

Created: 2022-03-23 11:55
#note

This problem occurs when it is impossible to make valid recommendations due to an initial lack of ratings.

Types:

• new community -> not enough data (for example in a new RS system) -> it occurs in Collaborative filtering systems -> encourage users to make ratings
• new item -> new items don't have ratings, so they are not recommended
• new user -> new user have not rated anything yet, so we can recommend items to them -> can not use Collaborative filtering again

Possible solutions:

• get additional information from new users;
• use cookies;
• geo-ip;
• recommend top-sellers or promotions;
• interview the user;
• use #hybrid_filtering approaches;
• use Content-based filtering in case of new items;
• map attributes to latent features (in case of new items);
• random exploration

References

Tags

#cold_start