摘要： 通过提取玩家在网络游戏数据库中的数据特征，用基于BP神经网络的数据挖掘方法挖掘出玩家对虚拟物品的各种属性的偏好，为设计开发个性化的虚拟物品提供决策支持。针对传统神经网络中很难获取有广泛代表性的训练样本、常常导致普通神经网络对陌生样本推荐时精度不高的问题，提出改进的BP神经网络，依据专家知识对神经网络的权重进行初始化，并根据训练样本对权重加以微调。仿真案例验证了该方法的有效性。

关键词: 数据挖掘, 网络游戏, 虚拟物品, BP神经网络

Abstract: This paper extracts gamers’ personal information from the database of online game and mines to find their potential demands of virtual items by BP Neural Network (NN). To solve the problem of inaccuracy caused by lacking of supervising sample using the traditional NN, an advanced BP model with initial weights in BP NN set by expertise and adjusted by the trained samples is put forward, whose result is more accurate with fewer supervising sample. Simulation results show that the model is efficient and useful.

Key words: data mining, online game, virtual item, BP Neural Network(NN)

中图分类号: 

• TP18