基于实时反馈的大众点评网团购业务个性化搜索解决方案

Personalized,Search Solution of Group Purchase in Dianping.com Based on Real-Time Feedback

大众点评网是国内领先的餐饮、购物、休闲娱乐及生活服务等领域的商户信息、消费优惠和消费评价的综合互动平台。针对用户个性化搜索的需求,并考虑到用户行为的时效性,大众点评网建立了一套基于Storm框架的个性化实时数据分析平台。该平台目前为团购搜索提供个性化搜索服务。线上大规模实践证明,基于Storm框架的实时数据分析平台对已有系统和HDFS集群具有良好的兼容性,并且具有高效、实时、稳定等优点。该平台完全能够胜任TB级别数据的实时处理和日均上亿次的服务调用。此外,大量实验显示,对用户实时反馈的个性化分析能够显著提升团购附近搜索的性能。预计切换全流量后,月度将新增100K购买用户及10M交易额。

Dianping.com is a domestic leading integrated interactive platform of merchant information,promotion and consumer comments on food,shopping,entertainment and life services.For users' demands on personalized search,regarding the timeliness of user behavior,this paper builds a personalized real-time data analysis platform based on Storm Framework.The platform is used for personalized group purchase search.It is proved by large-scale online test that,the Storm framework-based real-time data analysis platform has good compatibility with both the established system and HDFS,and it operates with high efficiency,good timeliness and stability.The platform is able to handle real-time processing of TB level data and hundreds of millions times service call per day.Besides,experimental results show that the personalized real-time feedback analysis for users has significant incremental impact on group purchase neighbor search.It is estimated that after switching to full flow,a monthly buyer lift of 100 K and sales lift of 10 M can be achieved.