面向电商精准营销的多维度用户行为画像数据

用于支撑电商平台的个性化推荐、精准广告投放、私域用户运营、内容策略匹配、促销活动分层设计等场景，通过构建动态用户画像，实现“千人千面”的智能营销，有效提升转化率、复购率与客单价。本系统采用“数据融合 → 特征工程 → 混合聚类 → 动态打标 → 实时更新”五阶段算法架构，各阶段与数据字段严格对应：

1.数据输入与预处理：
系统采集用户行为日志，生成以下基础字段：
近7天浏览商品次数（browse_freq_week）：衡量短期活跃度；
距离上次购买天数（last_purchase_days）：识别流失或高活跃用户；
近30天商品分享次数（social_share_cnt）：评估社交影响力；
历史平均客单价（元）（avg_order_value） 与 最常收藏品类（fav_category）：刻画消费能力与兴趣偏好。
2.人群聚类与标签生成：
基于上述字段构建特征向量，采用 K-Means 与 DBSCAN 混合聚类算法，自动发现潜在人群分组；
结合业务规则与机器学习模型，生成 动态人群标签（crowd_label），例如：
若 历史平均客单价（元）（avg_order_value） ≥ 800 且 最常收藏品类（fav_category） 为“数码”或“奢品”，则打标为“高客单价偏好人群”；
若 距离上次购买天数（last_purchase_days） ≤ 30 且 近30天商品分享次数（social_share_cnt） ≥ 3，则打标为“高活跃社交用户”。
3.实时更新机制：
用户每产生一次有效行为（如浏览、下单、分享），系统触发流式计算；
在5分钟内重新计算相关字段，并更新 动态人群标签（crowd_label） 与 画像更新时间（update_timestamp）；
脱敏用户ID（user_id） 作为唯一键，确保画像数据准确关联至对应用户。
通过 “近7天浏览商品次数（browse_freq_week） + 历史平均客单价（元）（avg_order_value） + 距离上次购买天数（last_purchase_days）” 等多维字段联合建模，实现动态、可解释、可运营的用户分群，非静态规则标签

This dataset supports scenarios including personalized recommendation, precise advertising delivery, private domain user operation, content strategy matching, and layered design of promotional activities on e-commerce platforms. By constructing dynamic user profiles, it realizes "one user, one tailored experience" intelligent marketing, effectively improving conversion rate, repurchase rate and average order value. This system adopts a five-stage algorithmic framework of "Data Fusion → Feature Engineering → Hybrid Clustering → Dynamic Tagging → Real-time Update", with each stage strictly corresponding to the data fields:

1. Data Input and Preprocessing:
The system collects user behavior logs and generates the following basic fields:
Number of product views in the past 7 days (browse_freq_week): measures short-term user activity;
Days since last purchase (last_purchase_days): identifies churned or highly active users;
Number of product shares in the past 30 days (social_share_cnt): evaluates social influence;
Historical average order value (CNY) (avg_order_value) and most frequently favorited product category (fav_category): depict consumption capacity and interest preferences.

2. Crowd Clustering and Tag Generation:
Construct feature vectors based on the above fields, adopt a hybrid clustering algorithm combining K-Means and DBSCAN to automatically discover potential crowd groupings; Combine business rules and machine learning models to generate dynamic crowd tags (crowd_label), for example:
If the historical average order value (CNY) (avg_order_value) ≥ 800 and the most frequently favorited product category (fav_category) is "electronics" or "luxury goods", tag the user as "high average order value preference crowd";
If the days since last purchase (last_purchase_days) ≤ 30 and the number of product shares in the past 30 days (social_share_cnt) ≥ 3, tag the user as "highly active social user".

3. Real-time Update Mechanism:
Whenever a user generates a valid behavior (such as browsing, placing an order, sharing), the system triggers stream computing; Recalculate relevant fields and update the dynamic crowd tag (crowd_label) and profile update timestamp (update_timestamp) within 5 minutes; Use anonymized user ID (user_id) as the unique key to ensure that profile data is accurately associated with the corresponding user.

By jointly modeling multi-dimensional fields such as "number of product views in the past 7 days (browse_freq_week) + historical average order value (CNY) (avg_order_value) + days since last purchase (last_purchase_days)", dynamic, interpretable and operable user segmentation is realized, rather than static rule-based tags.

搜集汇总

背景与挑战

背景概述

该数据集包含601条多维度用户行为画像数据，涵盖浏览频率、客单价、收藏品类等8个字段，用于电商精准营销场景。数据采用混合聚类算法动态更新人群标签，支持个性化推荐和广告投放，提升转化率和复购率。