Author: Xianyu Technology — Midu

background

With the continuous demand iteration and rapid business development of Xianyu, how to further improve the use, retention and active transformation of users, and then drive the growth of the whole second-hand trading market is the core proposition. According to the internal data analysis of Xianyu, it is found that the number of products released by users is significantly positively correlated with the volume of transactions. Sellers with more products are more active and have higher retention. Therefore, release plays a crucial role in active transaction and user retention. So how do we guide sellers to release more commodities, so as to help the trading innovation business to quickly iterate and promote the supply and trade growth, is one of the key thinking directions of Xianyu at present.

Status and Goals

At present, 66% of The MAU of Xianyu are non-sellers, while nearly half of the users of online sellers only have less than 2 pieces of treasure. The main reasons for sellers not to publish can be divided into the following three categories:

1. Lack of motivation; 2. High cost of publishing; 3

In the past, Xianyu tried to release marketing activities related to idle red envelope, rights and interests in the double Ten promotion activities to improve the motivation of sellers to release; The construction of intelligent publishing (blur, similarity detection, theme recognition) and light publishing (same type prediction, input association, keyboard label) is committed to reducing the user’s publishing cost; However, marketing strategies promoting release on the release link are mainly temporary activities, which require r&d and individual development of each case by case in advance. The scheduling is time-consuming, and the ability of directional intervention in fixed scenarios has been lacking, triggering users’ awareness of resale. Orphan works characteristic of idle second-hand goods market, efficiency and volume efficiency is a far cry from different categories of goods, we hope that the idle market based on the present situation of supply and demand analysis, mining market opportunity, in a fixed lead page set up vendor position, set up a human is recommended for directional intervention + algorithm published guide recommended link, realize one thousand thousand of the sophisticated operations, Guide sellers to release, adjust the category structure of released goods, so as to better provide high-quality supply for idle fish, improve commodity efficiency, promote transaction growth, and realize sellers’ promotion of release closed-loop.

Train of thought

Based on the problems in closed-loop link, how to build a publishing platform with long-term intervention operation? We need to start from the following perspectives:

1. Based on the supply-demand analysis of second-hand idle goods algorithm, different strategies of different sellers can be recommended 2. Based on the multi-algorithm recommendation strategy of the seller’s Taobao order fusion, display the matching purchase and sale details of similar commodities 3. Empower the business side with engineering means, provide real-time intervention ability, accelerate the quick launch of new business and adjust the launch strategy

Based on the above considerations, we and algorithms and search team, through personalized recommendation algorithm based on a key to resell list link, at the same time support operation intervention capabilities, associative search real-time matching associated goods to sell, sell publishes guide for promoting the long-term directional intervention to build a basic link, one thousand thousand of the strategy to achieve guide selling noodles, offers opportunities for the business side quick trial and error.

Main implementation methods

Centering on the core ideas mentioned above, we can split the promotion of release guidance link into two scenarios, one is the activities and strategies of operation configuration, and the other is the personalized SPU recommendation of the algorithm based on the seller’s Taobao order list. Based on UT log collection on mobile terminal, we designed three modules of operation activity configuration, algorithm recommendation and condition search respectively to achieve different functions

1. Operation activity configuration Supports flexible configuration of multiple marketing activity types 2. Algorithm Recommends different policies for obtaining the promotion algorithm 3. Conditional search is used to resolve the data source for the goods associated with the undertaking page

In the implementation process, we followed two principles:

1. Decouple modules from each other, realize multiple external systems that depend on in the process, define the boundary of batches, and achieve minimum coupling 2. The access service within the module is extensible, and the operation strategy and algorithm strategy may be constantly iterated. In terms of code implementation, we need to achieve fast access and easy expansion

The overall hierarchical architecture is as follows:Operation activity configuration construction

In order to provide long-term intervention ability for operations, the realization of thousands of sales strategy. If like the old development mode in the past, need to push through the server to close or even modify the code to achieve, not only high cost and low efficiency. Therefore, the distribution promotion activity was developed based on the Kunpeng system (a platform for business development and operation students) built by The Xianyu technical team, and the special business needs were realized through the development frameworks of DataFetcher and MatFilter provided by Kunpeng’s building blocks. The development mode is upgraded from single-person serial development to multi-person parallel development, isolating the division of labor between development and operation, and realizing functions such as activity cycle management, gray level control, and multi-condition filtering (version, fatigue, platform, etc.).

1.DataFetcher subsystem. By implementing the parent class, the DataFetcher class can be quickly reused when the operation activities are put into the homepage feeds, search results page, guess you like page, etc. It can be registered to the corresponding scene through the console, effectively saving development resources and improving the efficiency of business on-line. 2.MatFilter subsystem, which precipitates a variety of basic components for business use, such as version segmentation, gray flow ratio, search term strict matching, page number filtering, UV fatigue filtering and other pre-filters. Activity configuration can select the corresponding filter on the console according to the relevant operation requirements, and realize the specific business logic by inheriting the MatFilter base class.

Based on the capabilities of the above two subsystems, the promotion of release link expands business requirements on the released extension points, realizes the extensibility of scenes and reuse of material filters, develops the promotion of release unique configuration templates and materials, and provides basic components for operation to achieve the release effects of different configurations.

• The operational configuration design is as follows

Operating in the configuration platform, flexible configuration activities of the life cycle, put in the crowd, fatigue, put on the strategy of priority, and activities of parameters such as specific themes, and then choose a different material according to different activities, the configuration is completed by the style preview, to carry on the examination and approval, the experimental data, link closed-loop implementation activity. After the developer realizes several configuration templates of the design, all subsequent changes of operation classes can be completed by operation and product self-help on the console. If the delivery does not meet expectations, it can also be automatically modified and offline by self-help. The overall process is decouple from the developer.

Algorithm analysis Recommendation algorithm testing mainly provides important basis for operation marketing strategy and personalized SPU product recommendation through supply and demand analysis

1. Analysis of commodity supply and demand. The use of market segmentation and market flow efficiency and commodity efficiency comparison, combined with the dynamic pin situation, according to the length of the release to the transaction, mining the tight supply category, for the operation of the circle to choose the crowd strategy to provide an important basis; 1. Recommend personalized SPU products. Based on users' taobao orders, the optimization is carried out from three aspects: expanding resale orders, optimizing the ordering model, and optimizing the interest points of promoting the release, such as the number of buyers, price guidance, and sales time estimation; Reconstruct the offline data source and online scheme to reveal the real goods purchased and the matching number, and promote the release;Copy the code

Considering the multi-source access of the algorithm and the expansibility of subsequent access, for the unified TPP service of the group (the common JVM code development and hosting platform within the group, which currently carries the main recommendation business), promote the release link to define the result DO according to different algorithm sources (obtaining SPU goods and corresponding matching goods number). Inherits the basic DO (resultBaseDO), uses uniformly encapsulated templates to request and parse TPP service results, greatly reducing access costs and improving code expansibility. The template definition structure class diagram is as follows:Each time, the client only needs to customize the TPP corresponding to the scene ID to return the result structure, and then call the getResultAndParse function in a unified manner. The template implements data structure transformation, log printing, and exception protection in a unified manner, greatly improving access efficiency. A code example is as follows:

function <T extends TppResultBaseDO> TppResultDO<T> getResultAndParse(sceneId, userId, params, cls) { TppResultDO<T> tppResultDO = new TppResultDO<>(); // Request to retrieve TPP result String Response = retrieveTppData(sceneId, userId, params); Try {if (condition) {DO tppResultDO = jsonObject.parseObject (response, new TypeReference<TppResultDO<T>>(cls) { }); } else { ... } } catch (Throwable throwable) { ... } return tppResultDO; }Copy the code

Conditions of the search

To show users with recommended strategy type product details, we search, linkage configuration in operating activities and strategies respectively, algorithm SPU recommend commodities of different activities to undertake page support the associated goods under different conditions of search (keywords, category, status), to solve the problem of the real-time data sources the page to undertake similar goods.

The effect

At present, promote the release guide to sell have online on the barrel experiment, sellers enter page after page hits and undertake conversion efficiency is higher, but the overall entry conversion performance is not obvious, the distribution and trading market promotion co., LTD., will later be released according to the seller behavior characteristics, trying to further explore in more new scenarios for scenario, trigger sellers sell the mind.

Summary and Prospect

This paper mainly introduces the overall architecture of publishing guidance and promoting publishing link and the solutions of several key points, hoping to bring some thinking and inspiration to readers. Promote natural support his people select the release link, kunpeng scene filter, combined operations, strategies and algorithms recommend, not just in the release guide page, home page feeds, guess you like the scene also can rapid access, then we will also try to explore more promote the release of new scene (idle fish market, new village, take pictures and code release, etc.), Continue to mine the behavioral data and characteristics of sellers, improve the accuracy and conversion efficiency of sellers’ user targets, and combine brand power, price power and supply and demand relationship to improve the quality supply of goods and sellers’ activity, so as to achieve transaction growth.