Cold start in collaborative filtering

Collaborative filtering (CF)

Collaborative filtering recommends items to a user based on the ratings and behaviour of other similar users (or items). It needs a user-item rating matrix.

Cold start problem

The cold start problem is the difficulty of giving good recommendations when the system has little or no data about a new user, a new item or a new system. Because CF depends on past ratings, it cannot find similar users or items without them.

Types

1. New user problem: a user with no ratings has no neighbours, so no recommendation can be computed.
2. New item problem: a new item has no ratings, so it is never recommended; the more popular items get all the attention.
3. New system (new community) problem: at the start, there are no ratings at all.

Example

A movie website has users Ram, Sita and Hari who rated many films.

CF can compare Ram and Sita because they have rated common films. It cannot compare Gita (no ratings) with anyone. Film D will not be recommended to anyone because nobody has rated it.

Solutions

• Ask new users to rate a few items or choose genres on sign-up.
• Use demographic data (age, location) or the user's social-media information.
• Use content-based filtering for new items (genre, actor, description).
• Hybrid methods that combine content-based and CF.
• Recommend popular or trending items at first.
• Let experts rate new items; use active learning.