The collaborative filtering is a technique used by recommender systems to solve the problems of overinformation. This trend is growing every day, due to its enormous functionality, more users use this tool in their searches.
Before the birth of the Internet, the consumer had no source of information except the product’s own advertising. The market has gone from this scarcity of information to its saturation.
Companies incorporate these tools on their website and the users themselves build a collective intelligence through a system of recommendations that are then studied and translated using statistical algorithms.
Types of collaborative Filters:
There are different types of filtering when establishing recommendations, they are classified into four:
- Filtering based on content: recommendations are made based on content that may be liked or interesting.
- Demographic filters: they are carried out by the characteristics of the users (age, sex, studies …).
- Collaborative Filtering – Recommendations are based on upvoted searches from similar users.
- Hybrid Filters : Mix the two or three of the previous filters for a better experience.
Difficulties in collaborative filtering:
Data scarcity: Collaborative filtering systems are based on data sets. If this data sample is small, it can be very expensive and ineffective. Sometimes a common problem is starting from scratch, as preferences cannot be collected accurately and reliably.
Synonyms: The diversity of tags with similar names are sometimes not recognized by the filtering system when in reality the user is looking for the same element and information.
Shilling attacks: In the recommendation systems anyone can make evaluations, a user being able to vote positively only on their products and services and give negative to their competitors, falsifying the effectiveness of this tool.
Diversity: The filters try to find a diversity when recommending multiple options. Sometimes this filter only gives visibility to the most popular products.