News classification
Contact us
- Add: No. 9, North Fourth Ring Road, Haidian District, Beijing. It mainly includes face recognition, living detection, ID card recognition, bank card recognition, business card recognition, license plate recognition, OCR recognition, and intelligent recognition technology.
- Tel: 13146317170 廖经理
- Fax:
- Email: 398017534@qq.com
The value of machine learning AI
The value of machine learning AI
Question 1: what is the value in the end?
The value lies in 1. providing more profit to the company or employer 2., providing users with a better and more convenient experience 3..
Scene 1:
If the merchants who are in the e-commerce platform want to sell more things, they need an e-commerce platform to help with push, SMS and email.
Then how can we find suitable users to recommend suitable businesses? Take mobile phones as an example, the benefits of different ways are 2:
Method conversion rate mono income
Random sampling 1% 2000
Simple logic (reading search Collection) select 2-3% 2000-3000
Harmonious recommendation + cross selling 3%-5% 3000
Monitoring learning 9%-10% 3000-5000
Data have been disposed, but the relative proportions of data are believable.
The above activity, to 10000 people to calculate, business income from 1%*10000*2000 to 10%*10000*4500, business profit of 10% calculation, platform to draw 20% of profit, that is, no material cost (except electricity, except for electricity), business extra profit 56000, platform extra profit 14000, this kind of thing who does not want.
In the platform, the feasible aspects are as many as the bull, the usual advertisement CTR, the merchant recommendation, the user gathering pot, the potential of the community, the user consumption, every time is the money, ask this kind of thing, ask a few machine learning staff, I think no one is not willing, besides, this kind of low end, we are the tens of dollars of the East. West can be done
The value is to provide more profit to the company or the employer, that is, the ROI that is too high to bear.
Scene 2:
Assuming that in X, X, X and X, what is the most direct way to optimize user experience?
Guess what you want, guess what you want to go, guess you want to go
Beijing to Shanghai:
There are a number of tickets here, assuming no logic. The initial sorting is either by time or at the price, in a word, by the programmer's idea.
Here, if there is a window, you will say, no, I think this Lao Wang every time to buy a high price ticket, I want to give him the high price of the ticket each time on it; the product window listen to, where can, this Lao Wang every time at night, I want to put the night ticket on it...
In fact, it makes sense and is feasible. How to make a better decision, or how to present a more suitable way of presentation for different users, requires machine learning assistance.
In short, we can know everyone's real thoughts and intentions.
We can analyze the attribute of the user to judge whether he is a commercial user to buy a high price ticket. In this kind of holiday, we don't have to give priority to high price tickets and should give priority to the high quality travel ticket with 9 points at the same time at the same time. See whether Lao Wang is home on Friday night, so the choice is every time, the next time the next king on Friday in the landing, give priority to push on Friday to go home, but also write a point "home safe" soft text, such a user experience, 99.9% after the human experience is difficult to complete.
Providing users with a better and more convenient experience, making money and letting your customers pay for it is also one of the values.
Scene 3:
14 years ago, the car rental is the entity under the line, the honest open door store, pay the water and electricity, pay the rent, and additional manpower expenses, the users take the car all kinds of inconvenient.
Why do you do something so tired?
Part of the reason is that wind control is partly responsible. Assuming that users can place orders on the Internet and take the car away, who will ensure the safety of vehicles? A car should be about 100 thousand less. Such a risk could not be borne by any enterprise at that time.
To say a data, 2 per thousand, may not be accurate, but also a certain degree of self creation. This is the rate of vehicle loss in traditional open shop car rental industry.
According to 5000 of the daily average orders disclosed by a car rental company, according to the new user +30% of 70% old users, the daily 5000*30% vehicles are the first rental, and the risk vehicles per day are about 2, and the potential risk of 20W is about 2 per thousand.
The daily salary of two hundred thousand, no, 500 yuan per day, you can hire a machine to learn the staff, the usual industry in a car rental, from July 2016, 1 cars have not lost, the use of a large number of travel data information + third party credit data, as to how to do, there are many very mature industry Methods: the score card model, FICO, AHP and so on, any one, can make the original very difficult problem to a certain degree of disposal, and the company is paying for some HC.
These things may not be possible before machine learning, or they may be lacking in talent.
Question two: where is the way out?
In the 11 year when I was in University, I read the mathematics department. Many people asked me what you were going to do when you read the mathematics department. Teach?
To tell you the truth, I didn't know where the way out for my mathematics department was at that time. I didn't want to be a postgraduate student and I didn't want to be a teacher at that time. I was also very contradictory at that time. Two. However, I believe this sentence: it is not impossible, but it has not yet arrived.
Quite a bit of a tongue twister. As usual, more and more people are studying machine learning.
The value lies in 1. providing more profit to the company or employer 2., providing users with a better and more convenient experience 3..
Scene 1:
If the merchants who are in the e-commerce platform want to sell more things, they need an e-commerce platform to help with push, SMS and email.
Then how can we find suitable users to recommend suitable businesses? Take mobile phones as an example, the benefits of different ways are 2:
Method conversion rate mono income
Random sampling 1% 2000
Simple logic (reading search Collection) select 2-3% 2000-3000
Harmonious recommendation + cross selling 3%-5% 3000
Monitoring learning 9%-10% 3000-5000
Data have been disposed, but the relative proportions of data are believable.
The above activity, to 10000 people to calculate, business income from 1%*10000*2000 to 10%*10000*4500, business profit of 10% calculation, platform to draw 20% of profit, that is, no material cost (except electricity, except for electricity), business extra profit 56000, platform extra profit 14000, this kind of thing who does not want.
In the platform, the feasible aspects are as many as the bull, the usual advertisement CTR, the merchant recommendation, the user gathering pot, the potential of the community, the user consumption, every time is the money, ask this kind of thing, ask a few machine learning staff, I think no one is not willing, besides, this kind of low end, we are the tens of dollars of the East. West can be done
The value is to provide more profit to the company or the employer, that is, the ROI that is too high to bear.
Scene 2:
Assuming that in X, X, X and X, what is the most direct way to optimize user experience?
Guess what you want, guess what you want to go, guess you want to go
Beijing to Shanghai:
There are a number of tickets here, assuming no logic. The initial sorting is either by time or at the price, in a word, by the programmer's idea.
Here, if there is a window, you will say, no, I think this Lao Wang every time to buy a high price ticket, I want to give him the high price of the ticket each time on it; the product window listen to, where can, this Lao Wang every time at night, I want to put the night ticket on it...
In fact, it makes sense and is feasible. How to make a better decision, or how to present a more suitable way of presentation for different users, requires machine learning assistance.
In short, we can know everyone's real thoughts and intentions.
We can analyze the attribute of the user to judge whether he is a commercial user to buy a high price ticket. In this kind of holiday, we don't have to give priority to high price tickets and should give priority to the high quality travel ticket with 9 points at the same time at the same time. See whether Lao Wang is home on Friday night, so the choice is every time, the next time the next king on Friday in the landing, give priority to push on Friday to go home, but also write a point "home safe" soft text, such a user experience, 99.9% after the human experience is difficult to complete.
Providing users with a better and more convenient experience, making money and letting your customers pay for it is also one of the values.
Scene 3:
14 years ago, the car rental is the entity under the line, the honest open door store, pay the water and electricity, pay the rent, and additional manpower expenses, the users take the car all kinds of inconvenient.
Why do you do something so tired?
Part of the reason is that wind control is partly responsible. Assuming that users can place orders on the Internet and take the car away, who will ensure the safety of vehicles? A car should be about 100 thousand less. Such a risk could not be borne by any enterprise at that time.
To say a data, 2 per thousand, may not be accurate, but also a certain degree of self creation. This is the rate of vehicle loss in traditional open shop car rental industry.
According to 5000 of the daily average orders disclosed by a car rental company, according to the new user +30% of 70% old users, the daily 5000*30% vehicles are the first rental, and the risk vehicles per day are about 2, and the potential risk of 20W is about 2 per thousand.
The daily salary of two hundred thousand, no, 500 yuan per day, you can hire a machine to learn the staff, the usual industry in a car rental, from July 2016, 1 cars have not lost, the use of a large number of travel data information + third party credit data, as to how to do, there are many very mature industry Methods: the score card model, FICO, AHP and so on, any one, can make the original very difficult problem to a certain degree of disposal, and the company is paying for some HC.
These things may not be possible before machine learning, or they may be lacking in talent.
Question two: where is the way out?
In the 11 year when I was in University, I read the mathematics department. Many people asked me what you were going to do when you read the mathematics department. Teach?
To tell you the truth, I didn't know where the way out for my mathematics department was at that time. I didn't want to be a postgraduate student and I didn't want to be a teacher at that time. I was also very contradictory at that time. Two. However, I believe this sentence: it is not impossible, but it has not yet arrived.
Quite a bit of a tongue twister. As usual, more and more people are studying machine learning.