游戏用户特征数据促销分析数据集合

数据清洗规则：采用IQR四分位法剔除参与率、贡献值等指标中的极端异常值，对缺失的促销偏好数据采用“同付费层级+同游戏品类”群体均值填充，保障数据完整性；偏好挖掘规则：基于LDA模型解析用户历史促销参与行为，生成“偏好促销类型-敏感力度-偏好时机”三级标签，标签准确率达92%以上；效果预测算法：构建“用户特征-促销属性-场景适配度”三维输入体系，改良XGBoost模型引入时间衰减因子，解决短期行为干扰问题，预测准确率达90%以上；数据更新规则：用户交互数据实时采集，日度生成促销效果快照，周度更新用户分层及竞品对标数据。Data Cleaning Rules: Adopt the Interquartile Range (IQR) method to eliminate extreme outliers from indicators such as participation rate and contribution value. For missing promotion preference data, impute using the group mean of users with the same payment tier and game category to ensure data integrity. Preference Mining Rules: Analyze users' historical promotion participation behaviors based on the Latent Dirichlet Allocation (LDA) model, and generate three-level labels of "preferred promotion type - sensitivity intensity - preferred timing", with the label accuracy rate exceeding 92%. Effect Prediction Algorithm: Construct a three-dimensional input system of "user characteristics - promotion attributes - scene adaptability". Improve the XGBoost model by introducing a time decay factor to address the interference of short-term behaviors, with the prediction accuracy rate exceeding 90%. Data Update Rules: Collect user interaction data in real time, generate daily promotion effect snapshots, and update user segmentation and competitor benchmarking data weekly.

搜集汇总

背景与挑战

背景概述

该数据集是贵阳一轶科技有限公司自行产生的游戏用户特征数据，规模1G，每日更新，主要用于游戏促销分析，包括用户分层、促销适配和效果优化等场景。它采用LDA和XGBoost等算法进行偏好挖掘和预测，支持精准营销策略迭代。