Seller Posting Promotion Platform Architecture and Implementation

Background: Xianyu aims to increase user retention and activity by encouraging sellers to post more items, as higher item counts correlate with higher transaction volume.

Current Situation and Goal: 66% of MAU are non‑sellers; half of active sellers have ≤2 items. Main obstacles are lack of motivation, high publishing cost, and no trigger for resale mindset. Existing temporary promotional activities are inefficient and lack a permanent, targeted intervention mechanism.

Approach: Build a long‑term, operable promotion platform that combines algorithmic supply‑demand analysis with configurable operational strategies to deliver personalized seller prompts.

Main Implementation Methods: The promotion chain is split into two scenarios – configurable operational activities and algorithmic SPU recommendations. Three core modules are designed: operational activity configuration, algorithm recommendation, and conditional search.

Operational Activity Configuration: Developed on the internal Kunpeng platform using two extension points – DataFetcher and MatFilter – to allow parallel development, activity lifecycle management, gray‑scale control, and multi‑condition filtering.

Algorithm Recommendation: Provides supply‑demand analysis and personalized SPU recommendations based on sellers' Taobao order data, optimizing factors such as demand count, price guidance, and estimated sale time. Integration with the corporate TPP service uses a unified template to parse results, reducing coupling and improving extensibility.

Conditional Search: Links search service to retrieve real‑time related items for activity landing pages, supporting keyword, multi‑category, and status‑based queries.

Effect: Early bucket tests show higher click‑through and conversion on landing pages, though overall entry conversion remains modest. Future work will expand to more scenarios to further stimulate seller posting behavior.

Conclusion and Outlook: The platform supports granular audience targeting, multi‑scene deployment (feeds, recommendations, etc.), and continuous iteration of activity and algorithm strategies. Ongoing efforts will explore new entry points, refine seller behavior models, and enhance product supply quality to drive transaction growth.

function <T extends TppResultBaseDO> TppResultDO<T> getResultAndParse(sceneId, userId, params, cls) {
    TppResultDO<T> tppResultDO = new TppResultDO<>();
    // 请求获取tpp结果
    String response = retrieveTppData(sceneId, userId, params);
    try {
        if (condition) {
            // 根据继承模板类解析成基础DO
            tppResultDO = JSONObject.parseObject(response, new TypeReference<TppResultDO<T>>() {});
        } else {
            ...
        }
    } catch (Throwable throwable) {
        ...
    }
    return tppResultDO;
}