A few days ago, I watched my favorite movie "Mission Impossible" on Netflix again. This movie is really wonderful. It is called "the best looking spy movie".
Whenever I open this movie on Netflix, Netflix will recommend other movies I may like. These recommended movies really fit my movie taste, such as the recommended movies "", "Mission Impossible 2", "Change Face", "Hurricane Rescue" are also my favorite.
So how does Netflix know what users like, and what is the secret behind these recommendations?
The secret weapon behind recommendation systems like Netflix, Youtube, and Amazon prime is machine learning. Now let's talk in the most understandable words. What exactly is machine learning?
1. How do humans think?
Machine learning is a bit like the human decision-making process. Let's first look at the human thinking process.
Suppose we go to buy oranges, and the lady who sells oranges says that oranges are sweeter than first love. But how do we choose the sweetest orange?
I remember my mother said that the bright yellow oranges are sweeter than the dark yellow ones. So we have a simple criterion: pick only the yellow oranges.
2. Common computer algorithms
How do ordinary computer algorithms implement this process? If you use a computer program to help you pick oranges, you would write this rule:
if (color is tender yellow):Oranges are sweetelse:Oranges are not sweet
We use these rules to pick oranges. But if there are new discoveries in our orange experiment, we have to manually modify this list of rules.
For example, some of the oranges we bought were sour. After tasting various types of oranges, we found that those big oranges are sweet. So we modified the rules:
if (color is yellow and size is large):Oranges are sweetelse:Oranges are not sweet
So how to overcome this shortcoming? Machine learning algorithms can solve this problem.
3. What is a machine learning algorithm?
Machine learning algorithms are evolved from the previous ordinary algorithms. By automatically learning from the data provided, it makes our programs "smarter".
We randomly sample a certain number of oranges from the market (called training data in machine learning) and make a table below, which contains the physical properties of each orange, such as color, size, and origin. (The attributes of these oranges are called characteristics). Also note that this orange is sweet (this is called a label).
We provide this training data to a machine learning algorithm, and it will learn a model about the relationship between the characteristics of the orange and whether it is sweet.
Next time we go to the market to buy oranges, face the new oranges (test data), and then enter the new oranges into this trained model. The model will directly output whether the oranges are sweet or not.
With this model, we can now buy oranges with confidence, without having to consider the details of choosing oranges. You only need to enter the physical properties of oranges into this model to know directly whether oranges are sweet or not.
More importantly, we can make this model better and better over time (reinforcement learning). When this model reads more training data, it will be more accurate and correct itself after making wrong predictions. .
This is not the best part. The best part is that we can use the same machine learning algorithm to train different models. For example, we can use the same machine algorithm to predict apple and watermelon models. This is not possible with conventional computer programs.
This is our exclusive machine learning. Is it cool?
4.Summary
Machine learning is to use machine learning algorithms to build models. When new data comes in, you can use the model to make predictions.
0 Comments