Recommendation systems: car tips
Recommendation Systems:
- Car Tips
- Cold start
- Introduction to hybrid systems
- artificial immune systems and the effect of idiotypes
Many current trends in web design are aimed at helping the user to better navigate and, if possible, facilitate selection and decision making. There are many explicit and implicit ways to direct and manipulate the user's behavior, to focus his attention on the desired aspects of the system and influence his decisions. But is it possible, and most importantly, is it necessary to save the user from having to choose? Is it possible to entrust part of the solution to automated systems of recommendations?
A good and concise definition of a recommendation system is provided by Wikipedia :
Recommendation systems are a special kind of information filtering technique that tends to present information (movies, music, books, images, web sites, and so on) that most likely will interest the user. Usually the recommendation system compares the user profile with some kind of reference information and tries to predict the “rating” that the user will give to the object he has not even thought about.
Systems of recommendations (hereinafter - CP), can be classified in different ways:
According to the method of collecting information, they can be divided into implicit data ( Implicit data collection ). Explicit ways include, for example, voting, sorting objects according to how much they like or dislike them, and directly compiling a list of what the user likes. Implicit methods are to analyze user activity and try to identify any dependencies.
')
According to the CP information processing method, one of the most common approaches can be distinguished: collaborative filtering . These systems analyze user behavior and build recommendations based on the opinions of people like him: after all, those who agreed in the past are likely to agree in the future. This method requires a huge amount of input data, but there are alternative ways to predict the user's opinion, regardless of the collective mind, using various mathematical algorithms.
I'm not a big fan of classifications, so I will refrain from further deepening this topic. Classification can always be replenished as needed.
The scope of CP is very wide, including on the Internet. Here are some famous examples of using such systems:
Even the well-known tag cloud, in fact, is also a recommendation. The size of the tag shows what others are reading and what it is worth reading to the user.
If this topic is interesting to readers, I will try to continue writing about it, because the use and implementation of such systems are really interesting, I speak from my own experience. In the meantime, I advise you to think about the following questions:
- Can you rely on a similar system?
- Is it important for you to know about the principles of operation and the structure of the system, or is it enough to see only the result?
- How much can you trust such systems? How much are you willing to let them into your personal life? How do you choose the line between functionality and ethical standards?
You can answer these questions out loud in the comments.
Original on my blog
- Car Tips
- Cold start
- Introduction to hybrid systems
- artificial immune systems and the effect of idiotypes
Many current trends in web design are aimed at helping the user to better navigate and, if possible, facilitate selection and decision making. There are many explicit and implicit ways to direct and manipulate the user's behavior, to focus his attention on the desired aspects of the system and influence his decisions. But is it possible, and most importantly, is it necessary to save the user from having to choose? Is it possible to entrust part of the solution to automated systems of recommendations?
Brief introduction
A good and concise definition of a recommendation system is provided by Wikipedia :
Recommendation systems are a special kind of information filtering technique that tends to present information (movies, music, books, images, web sites, and so on) that most likely will interest the user. Usually the recommendation system compares the user profile with some kind of reference information and tries to predict the “rating” that the user will give to the object he has not even thought about.
Systems of recommendations (hereinafter - CP), can be classified in different ways:
According to the method of collecting information, they can be divided into implicit data ( Implicit data collection ). Explicit ways include, for example, voting, sorting objects according to how much they like or dislike them, and directly compiling a list of what the user likes. Implicit methods are to analyze user activity and try to identify any dependencies.
')
According to the CP information processing method, one of the most common approaches can be distinguished: collaborative filtering . These systems analyze user behavior and build recommendations based on the opinions of people like him: after all, those who agreed in the past are likely to agree in the future. This method requires a huge amount of input data, but there are alternative ways to predict the user's opinion, regardless of the collective mind, using various mathematical algorithms.
I'm not a big fan of classifications, so I will refrain from further deepening this topic. Classification can always be replenished as needed.
Application
The scope of CP is very wide, including on the Internet. Here are some famous examples of using such systems:
- On amazon.com and many other online stores when choosing a product, you can see a list of products that were bought by users who previously bought your selected product. With its help, the buyer can quickly find a product similar to the one that he liked. Such a system works well with complementary goods, but it is unlikely to offer the user an alternative choice.
- Last.fm is a music site that uses the Audioscrobbler program. One of my favorite examples. All that is required of the user is to install the program on his computer, and it will automatically analyze the music that users listen to and give him advice. It is because of the word “automatically” that I love it: it can work in the background and the user does not need to do any unnecessary actions.
- Some information resources, such as digg.com ( Digg Recommendation Engine ) and Habrahabr itself ( Habralenta 3.0 ), use CP to determine which articles and news may be of interest to a specific user. This area to me, personally, is especially interesting because I am currently working on such a project . Such systems analyze a variety of indirect and explicit manifestations of user interest, such as viewing news, voting, and entering into the "favorites." It is important that such systems use not only explicit expressions of interest, especially if they are limited, like voting at Habrahabr, but also all possible implicit factors.
Even the well-known tag cloud, in fact, is also a recommendation. The size of the tag shows what others are reading and what it is worth reading to the user.
Why all this?
If this topic is interesting to readers, I will try to continue writing about it, because the use and implementation of such systems are really interesting, I speak from my own experience. In the meantime, I advise you to think about the following questions:
- Can you rely on a similar system?
- Is it important for you to know about the principles of operation and the structure of the system, or is it enough to see only the result?
- How much can you trust such systems? How much are you willing to let them into your personal life? How do you choose the line between functionality and ethical standards?
You can answer these questions out loud in the comments.
Original on my blog
Source: https://habr.com/ru/post/57759/
All Articles