Computing? Algorithm.–Yanwen

With the rough introduction from previous post, I’d like to elaborate the project I did over the summer and past two weeks.

Started in July, me and other four college students putted our hands on the planning of a Credit Investigation System which is a system that helps banks to determine a specific person’s credibility when banks are asked to lend money to he or she. In the traditional Credit investigation system, the credit degree is mainly depended on the evaluation of the combination of users’ occupation, family situation, social relationship and other elements. Our team, however, decided to cut into users’ daily consume from analyzing their location information and the location’s average cost, and then build the credit system from those analyses. Because, we believed, one’s daily consumption could effectively reflect one’s consuming ability and, correspondingly, one’s loan repayment ability. The estimate target customers of our project are banks and private organizations that need relevant data to manage their potential customers (like Alibaba or Tencent). Basically we are selling data and the way to manipulate data. The system is called TraceCloud.

The basic working model TraceCloud is presented as the following graph:

screen-shot-2016-09-19-at-8-48-27-pm

We grab location information from users’ cell phones to servers, run our algorithms to generate the credit degrees and then send them to our database. The algorithm is the core part of the whole project. Being part of the algorithm team, I started by analyzing weights of different types of consuming. We divided basic consumes into seven parts: car service, diner service, shopping service, sports and relaxation service, medical service, travel, and living service. Since it’s hard for us to grape precise data for medical, and travel service based on users’ location, we mainly depended on the rest five to determine the weight. For example, for diner service, we applied data released by government about the average cost’s division and average salaries’ division to generate the final weight. Data table is presented below:

Cost on meal per person per month (yuan) Average salary (yuan) Weight after normalizing
<100 1266 0
100-200 1975 0.20
200-500 2859 0.46
500-1000 2991 0.49
1000-1500 3165 0.54
>1500 4755 1

With those weights, we were allowed to roughly compute the credit degree of a person based on his or her dinning manner and frequent dinning location.

The function we used to calculate weights was:

screen-shot-2016-09-19-at-11-46-41-pm

W=weights;

Average cost=price;

Scoring probability=rating;

Full mark=full;

Adjusting value=k;

This project opened a relative new way for me to start a project since TraceCloud is a product that need many data to support its operation. Used to be a person who played around database, I discovered during the project that I was actually lack of many knowledge in data managing and researching. Hope during the time when we start running the product and collecting enormous data sets, I can learn more about database. This is what I have for this week.

Citation: No work cited. All images are from the real project.

6 thoughts on “Computing? Algorithm.–Yanwen

    1. yanwenxu Post author

      While updating location information frequently along with newly released information from the government and gaode map which is a widely used mapping system that has a contract with us and constantly provides updated information to us, we will keep running script to test the adjusting value in our credit computing algorithm doing adjustment while needed and make sure it’s appropriate to current finical situation. There is another function for us to increase or decrease a person’s credit.

      Reply
  1. Max

    Completely out of curiosity, what is the process of obtaining location information from target users? Would there be any legal issues?

    Reply
    1. yanwenxu Post author

      Actually most of the apps on the market now are running background code to detect users’ location information and to generate appropriate suggestions for them such as Shanghai is going to rain, bring a jacket–what we called humanization. But there is lack of gathered analysis of those information. Users will be notified anyways while they are being detected–so called they are willing to become long-term credit candidates. =_=

      Reply

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s