Collaborative filtering Collaborative filtering

Every time I want to see a movie, I ask my friend, Xiaojian. Generally he recommended the film, I prefer. Obviously not everyone has the ability to Xiaojian. Because I happen to have similar tastes with Xiaojian.

The experience of this life, in fact, has a wide range of uses.

When a system needs to make recommendations for someone, a machine learning algorithm works like this: To find a large group of people who are similar to his tastes, sort the things they like, and then recommend them to him.

Two questions naturally arise:

1. Who is similar to his taste of the people;
2. How to sort the things that these people like;

There are many different answers to these two questions, and different answers mean different algorithms.

---

There are two answers to question 1:

1. Euler distance; Given n objects, the two are scored {S1}{S2} respectively, then the Euler distance is the sum of the squares of each element of S1 and S2;
2. Pearson correlation coefficient. Given n physics, the degree to which two people have a consistent rating can be measured by this factor. Each noun represents a point that represents two people with a given fraction (x, y), and you can see that the midpoint of the right is closer to the line, which indicates that the two people in the right are getting closer to their scoring, although the absolute value of the score is different. ：

It can be imagined that if two people, a more mean, a more generous, using the # # # is easier to find the characteristics of the evaluated object.

-------

What is the answer to question 2?

A common approach is to add a weighted sum to the recommended value for each recommended object: for example, a personal similarity of 90%, the person who recommended the value of this object is 80 points, then he contributed to the 80*0.9=72.

For example, the most recommended object can be obtained by sequencing.

Obviously the logic above is that what is recommended for someone like you is more in line with your mind.

Technorati Tags: Collaborative filtering, collaborative filtering, machine learning, life experience

Collaborative filtering Collaborative filtering