互动大数据驱动的购买意愿预测与个性化营销方法研究

Interaction through online communities has been an important way for customers to collect product information and form their purchase decisions. Predicting customers’ purchase intentions based on the characteristics of the interaction behaviors and attracting them to purchase are very important issues for the big data-driven marketing management. Based on the interaction big data, this proposal employs the theories such as text mining methods, social network analysis methods, Bayesian data analysis methods and intelligent optimization methods to mine the characteristics and the evolution mechanism of customers’ interaction behaviors, predict their purchase intentions and conduct personalized marketing strategies. This proposal conducts researches from the following three aspects. First, we mine the characteristics and the evolution mechanisms of customers’ interaction behaviors, and investigate their relations with the purchase decisions. The behavior characteristics, evolution mechanisms and their relations with the purchase decisions provide prior knowledge for the purchase prediction. Second, to be consistent with the background that the products discussed in online communities are substitutes and customers’ purchase decisions are dependent on each other, we study the modeling methods of the competitive network for substitute products and investigate the purchase prediction method by considering products’ competitive network and customers’ purchase dependence. The proposed models can obtain customers’ purchase probabilities for different products. Third, to attract customers’ purchase behaviors, we study the analysis methods of the negative emotions and reveal the emotional stimuli. With each customer’s negative emotions and the corresponding stimuli, we study the personalized marketing strategy and the multi-objective optimization of the personalized marketing strategy. This project is significant for researchers and practitioners because it enriches the theoretical framework of the big data-driven marketing management and decision-making and provides guidelines for companies to innovate marketing strategies.

通过在线社区与他人互动已经成为消费者收集产品信息、形成购买决策的重要途径，利用消费者互动行为预测其购买意愿，并通过营销策略将其转换为实际客户是大数据环境下营销管理的重要研究问题。本项目以互动大数据为基础，以“反映购买意愿的行为特征及演化分析、消费者购买意愿预测、个性化营销激励”为主线，应用文本分析、社交网络分析、贝叶斯数据分析和智能优化等理论，挖掘反映消费者购买意愿的行为特征及其演化规律，构建行为特征及演化与购买决策的关系模型，为购买意愿预测提供先验知识；针对产品竞争和购买依赖等问题背景，研究产品竞争网络的构建方法和消费者购买意愿的预测方法，获得消费者购买不同产品的概率；基于个性化营销引导消费者的购买行为，研究消费者负面情绪的挖掘方法，揭示负面情绪产生的原因，在此基础上，研究个性化营销方法和多目标营销优化方法。本项目对丰富大数据驱动的营销管理与决策理论，创新企业营销实践具有理论和实践意义。

通过在线社区与他人互动已经成为消费者收集产品信息、形成购买决策的重要途径，利用消费者互动行为预测其购买意愿，并通过营销策略将其转换为实际客户是大数据环境下营销管理的重要研究问题。本项目以互动大数据为基础，以“反映购买意愿的行为特征分析、消费者购买意愿预测、个性化营销激励”为主线，研究了互动文本的挖掘方法，提出了融合专家知识的主题模型、基于众包策略的主题模型和基于数据增强的非参数贝叶斯模型，为基于互动文本分析消费者偏好提供了方法基础；提出了用户角色的划分方法和基于不同角色用户的社会营销策略，构建了人格特征到行为偏好的生成模型，研究了考虑选择与影响的社交网络的生成与演化机制，项目成果提高了互动情境下消费者行为的可解释性；提出了面向个体消费者和消费者群体的个性化需求预测模型，设计了多源数据融合的个性化推荐策略，为准确预测消费个性化需求提供了方法基础；研究了社会化营销中的影响力最大化问题，提出了基于电导中心性的影响力度量方法，构建了考虑种子节点成本和非种子节点收益的影响力最大化模型，将消费者的线上线下转换引入双渠道促销优化，构建了多阶段多市场的协同价格优化模型，综合销量最大、利润最高、不满意率最低等目标的影响，构建了双渠道促销的多目标优化方法。项目组与美国匹兹堡大学Katz商学院、宾夕法尼亚大学Wharton商学院、澳大利亚迪肯大学计算机系的多位专家合作，在EJOR、Information Sciences等国内外学术期刊发表和录用论文21篇，其中，SCI/SSCI检索论文16篇，获得国家发明专利授权4项，实审发明专利8项。项目组开发汽车品牌管理原型系统一套，为安徽省人民政府提供《安徽省大数据产业发展路径》智库报告1份，培养硕士研究生3名，博士研究生4名。