在头部互联网电商大厂当数据分析师，是种什么体验？

Hello, everyone. I am currently working in the product operation department of a large Internet e-commerce company.

Today, I would like to share with you the interview process of the big Internet e-commerce companies, some of my daily work, and introduce some related work cases to you.

But for most people, the resume screening stage can be a very big problem when sending resumes to big companies.

With regard to the content of daily work, first of all, it is some procedural. For example, there are morning meetings, followed by weekly meetings, daily newspapers and weekly newspapers, which are also submitted every day.

Both internally and externally, you should follow the process step by step according to the process you are in. Basically work according to this flow program.

First of all, the basic attributes of the portrait, such as user login name, user level, gender, age, marital status, education, occupation and other indicators are easily available.

In addition, there are some indicators that need to be excavated deeply. For example, whether there is a child, how old the child is, the gender of the child, and whether the family has a car, what is the purchasing power.

Its main use is to help sellers better understand their customers, and then facilitate accurate designated marketing activities to increase the conversion rate, as well as to help sellers analyze the sales situation.

Here is part of our company's portrait system, its tag name, some algorithms and application scene model interpretation, and so on.

Users' age, marital status, education, occupation and so on are all through some algorithms, which are divided by algorithms, and we will not elaborate on some specific application scenarios.

For e-commerce companies, a very big pain point it has to solve is to buy, sell and stock, that is, I want to make sure that the goods in the warehouse are enough for me to sell, and at the same time, I can't buy too much, because the warehousing capacity and funds are limited.

There are basically three algorithms for sales forecast: statistical algorithm, calculation algorithm and deep learning algorithm.

So we can establish such a weight, divided by the sum of all the weights, then we can get a more similar sales forecast.

There are some data that may fluctuate seasonally, and then we can also use, for example, the year-on-year moving average in the same month of the year. It is the same month in different years, its sales can be used to eliminate seasonal errors, or according to the principle of proximity, it may have a higher weight on these dates closer to this year.

The specific modeling process is to collect data first, and then clean data preprocessing and feature engineering. The main goal of these data cleaning, data preprocessing and feature engineering is to clean the data set, extract the model index after cleaning, and then start model development after listing it. After the model is built, the model is tuned, and then our data are verified and iterated.