Characteristics of a recommender system
Posted: Wed Jan 22, 2025 4:32 am
The key function of a recommendation system is to inform the user about a product that may interest him or her at the moment. The client receives valuable information, and the service earns money by providing quality services. Moreover, the service does not only involve direct sales of the product. Income from services may be a commission fee, as well as an increase in customer loyalty, which will help retain the consumer and encourage him or her to use the resource again.
The business structure also determines the functions assigned to recommendation systems. For example, for TripAdvisor, they form the basis of the business model, and in a typical online store, they help improve user content and simplify the search in the catalog. Individualization of Internet marketing has become a pronounced trend over the past 10 years. McKinsey calculated that Amazon receives up to 35% of its profits from goods and services offered to the user in recommendations. For Netflix, this figure is even more impressive - 85%. The most important task solved with the help of recommendation systems is achieving maximum customer satisfaction.
Characteristics of a recommender system
Let us dwell on the main features of recommender systems that are inherent to each:
Recommendation object. These homeowner database are the goods, services, materials that are offered. In general terms, this is any content that can attract the user's attention. For Wildberries, Yandex.Market, these are goods, for Spotify, audio content, for YouTube, videos selected by an algorithm, for resources like TASS, Interfax, news and articles. The list goes on and on.
The task of the recommendation. What the user is provided with the relevant content for: purchase, information gathering, communication of some information, etc.
Recommendation context: The action that the consumer performs when receiving recommendations, such as watching a movie, reading the news, studying a product or service description, etc.
Recommendation source. The subject from which the data comes. This could be a community, other users (for example, the rating of an establishment on TripAdvisor is formed based on their reviews).
Degree of individualization. An algorithm can generate the same recommendations for all users in a certain location, belonging to a certain age group or gender, but without taking into account customer data. More sophisticated ones already make offers based on information about the consumer: their browsing and purchase history, the content of favorites - any data that directly or indirectly allows us to judge their preferences and needs.
Transparency. Users will trust recommendations more if they understand how they are compiled. This helps avoid systems that show products only because they have a higher (i.e., more profitable for the seller) price.
A well-developed recommendation scheme should detect signs of fraud (artificial rating inflating, custom reviews, etc.). Sometimes manipulations are unintentional. If a new part of a film franchise is released, in the first weeks its viewers are fans whose objectivity is low, and the rating may be much higher than it will be in a couple of months.
Recommendation format. An area of the workspace where suggestions are placed. This could be a product feed, a pop-up window, an automatically compiled playlist recommended for listening, etc.
Read
The business structure also determines the functions assigned to recommendation systems. For example, for TripAdvisor, they form the basis of the business model, and in a typical online store, they help improve user content and simplify the search in the catalog. Individualization of Internet marketing has become a pronounced trend over the past 10 years. McKinsey calculated that Amazon receives up to 35% of its profits from goods and services offered to the user in recommendations. For Netflix, this figure is even more impressive - 85%. The most important task solved with the help of recommendation systems is achieving maximum customer satisfaction.
Characteristics of a recommender system
Let us dwell on the main features of recommender systems that are inherent to each:
Recommendation object. These homeowner database are the goods, services, materials that are offered. In general terms, this is any content that can attract the user's attention. For Wildberries, Yandex.Market, these are goods, for Spotify, audio content, for YouTube, videos selected by an algorithm, for resources like TASS, Interfax, news and articles. The list goes on and on.
The task of the recommendation. What the user is provided with the relevant content for: purchase, information gathering, communication of some information, etc.
Recommendation context: The action that the consumer performs when receiving recommendations, such as watching a movie, reading the news, studying a product or service description, etc.
Recommendation source. The subject from which the data comes. This could be a community, other users (for example, the rating of an establishment on TripAdvisor is formed based on their reviews).
Degree of individualization. An algorithm can generate the same recommendations for all users in a certain location, belonging to a certain age group or gender, but without taking into account customer data. More sophisticated ones already make offers based on information about the consumer: their browsing and purchase history, the content of favorites - any data that directly or indirectly allows us to judge their preferences and needs.
Transparency. Users will trust recommendations more if they understand how they are compiled. This helps avoid systems that show products only because they have a higher (i.e., more profitable for the seller) price.
A well-developed recommendation scheme should detect signs of fraud (artificial rating inflating, custom reviews, etc.). Sometimes manipulations are unintentional. If a new part of a film franchise is released, in the first weeks its viewers are fans whose objectivity is low, and the rating may be much higher than it will be in a couple of months.
Recommendation format. An area of the workspace where suggestions are placed. This could be a product feed, a pop-up window, an automatically compiled playlist recommended for listening, etc.
Read