Author: Tong Hui

background

During its rapid development in recent years, Xianyu has accumulated hundreds of millions of online products, which may lead to a bad shopping experience for buyers. Therefore, we hope to introduce the ability to understand leisure fish products, select high-quality products, and improve the shopping experience of leisure fish.

Pain points

Existing free fish commodities have some pain points:

  • Seller attributes: There are mainly individual sellers on Xianyu, and the sellers are not online for a long time, which leads to normal transactions.

  • Commodity attributes: there are inferior commodities, commodity drainage and repeated shop and other problems.

  • Price attribute: there are false price tags, low cost performance problems.

  • Supply and demand: there are a lot of unbalance goods in idle fish.

In view of these pain points, from the above four dimensions to understand the free fish commodities, and according to the understanding of the content of the selection.

Commodity hierarchical dimension

Commodity attribute

  • New product attributes: describe brand information, SKU/SPU/model, etc.

  • Second-hand attribute: used to estimate the damage of goods. This different category is different, such as the purchase date of a beauty care category, surplus and other attributes. The date of purchase of the phone, whether the screen is broken, whether there are scratches, etc.

The coverage of existing new and second-hand attributes is not high. We will continue to improve the accuracy and coverage of product attributes through user guidance and algorithm association in the future. This is also the basis for subsequent analysis of supply and demand and price attributes. In addition, we also can judge in commodity attribute side, whether contain pornographic, inferior, drainage and repeat the situation of shop goods, and do relevant management to these goods.

Supply and demand

Below is a classic supply and demand curve, where line D represents the demand curve and line S represents the supply curve. The point E where supply and demand are equal is defined as the equilibrium point, and the price level corresponding to point E is defined as the equilibrium price (P), that is, the price when supply and demand balance; The yield corresponding to point E is defined as the equilibrium yield Q. That’s the quantity that’s balanced between supply and demand.

Similarly, there is a large imbalance between supply and demand for idle fish commodities. For example, a large number of women’s clothing products accumulated, supply exceeds demand; Mobile phone category of goods consumption is relatively fast, supply is less than demand. By analyzing the supply and demand relationship and commodity attributes of idle fish products, we can know the corresponding supply and demand situation of products in specific categories, specific brands and specific discount situations. In view of these situations, we can relieve the situation of overstocking of idle fish products, and also replenish some goods whose supply is less than demand.

Property prices

The main mentality of second-hand platform buyers is to buy low-cost, cost-effective goods. According to the new and second-hand attributes obtained from the analysis of product attributes, we can know the value of idle fish and its cost performance ratio among similar commodities. In this way, we can know the “value” of the product, so as to better guide the “price”, and analyze whether the product is competitive.

The seller properties

Seller attributes include active rate, response rate, selling rate, dispute rate, praise rate and so on. The particularity of Xianyu itself — Xianyu is dominated by individual sellers. Just imagine that if the sellers are not online, do not reply or unwilling to deliver goods, it is impossible to complete the normal transaction process. Therefore, the seller attribute is the necessary condition for whether goods can be sold normally, and the attribute is also an important link in the commodity attribute.

Hierarchical implementation of goods

Understand the goods

We analyze commodities from the above four commodity dimensions, find the factors related to these commodity dimensions and design the commodity information index, supply and demand index, price index and seller index, so that these four indexes are positively correlated with the main indicators (such as exposure rate of payment, turnover rate, etc.).

Finally, regression analysis is made on these indexes, and the final weighted score of high quality commodities is obtained. The pie chart below is the score of quality products obtained (the ratio is for demonstration only).

Selection of quality goods

So with the quality commodity index, how to select quality commodities?

We will define objectives and restrictions. For example, we will select the highest quality goods and 20 million goods. We also hope that the categories of goods will be balanced and the dispute rate will not be too high. Then it becomes a combinatorial optimization problem:

  • Define the Target Target Max(Quality product points)

  • Define the restriction s.T. Quantity of commodity <= 20 million S.T. leaf category commodity equilibrium s.T. Dispute rate <= 0.3%

By solving the combinatorial optimization problem, the high-quality commodity pool can be obtained.

Data back to the test

The purpose of data backtesting is to hope that our results have been validated over time. If one month is selected as the backtest cycle, we will backtest the difference between the main index of high-quality goods and the average value of the market. If the main index of high-quality goods is better than the average value of the market for a continuous period of time, it can be proved that the selected high-quality goods are better than ordinary goods.

The simulation system

If the data backtest is to verify the effect of “past” selection, then the simulation system is to predict the impact of high-quality product strategy on “future” after it goes online.

The idea of the simulation system is to predict the outcome of the strategy after it goes live by combining the primary, intermediate, and advanced knowledge gained from the previous analysis. Among them

  • Primary knowledge refers to knowledge gained from a particular activity, for example, men and people in their 20s and 30s are more interested in that activity.

  • Intermediate knowledge refers to the knowledge gained from certain activities, such as red envelope activities, which work better for certain groups of people.

  • Advanced knowledge refers to more general knowledge, such as the effect of buying on activity.

The effect

Quality products selected by layers of Xianyu products have been applied in the scenes of first purchase of Xianyu by new users and promotion of Xianyu, and indicators such as click rate and purchase rate have been significantly improved. The highest click-through rate increased by 20%, and the highest interactive payment rate increased by 100%. (For the sake of data security, specific effect indicators will not be published)

Looking forward to

This paper introduces the idle fish knowledge understand part of the goods, we later will continue to improve the idle fish commodity hierarchical system, at the same time hope to continue and run classmates together, mining more knowledge of goods but and entry to idle fish commodity knowledge base, and through the user guide, enlarge the number of idle fish high quality goods, so as to further improve the user experience.