差异化客户需求的提取及比较研究：基于产品在线评论的挖掘分析

Fast development of the e-commerce offers consumers more options on similar products. To gain more customers, companies are advised to comprehend advantages about similar products, pinpoint comparative customer requirements and launch target-directed competitive products. However, conventional studies in the comparison of customer requirements focus only on a small number of structural questionnaire data and fail to handle a big volume of unstructured data. Notice that, millions of online reviews which contain valuable customer concerns are widely available in e-commerce and social network websites. It motivates researchers in different areas to explore the value of product reviews on the acquisition and analysis of customer requirements. Nonetheless, coarse static customer requirements are explored in these studies, which potentially induces that fine-grained ever-evolving customer concerns are neglected. Accordingly, in this research, product online reviews are analyzed from the perspective of designers, in which classical theories on customer requirements acquisition and analysis are referred. First, algorithms for the extraction of multi-level customer concerns are developed based on models for sequential data labeling and co-clustering algorithms. Next, techniques for new word identification and multitask learning are applied to detect fresh customer requirements and compare regional customer concerns. Finally, approaches for customer concern extraction and comparison over similar products are proposed by using models for time series analysis and integer programming. All these endeavors contribute to providing an integrated intelligent solution for comparisons of customer requirements by mining product online reviews. The ultimate goal of this research is to ease the recognition and analysis of comparative customer concerns in market-driven new product design, promote the theoretical development in requirement acquisition and highlight new pathways for research investigations in requirement management as well. Also, it helps to support decision-makers with critical data for the comparison of customer requirements, which will smooth the well-focused product development with intelligent decision support.

电商的发展为消费者提供许多相似产品。为赢得客户，企业应理解同类产品的优势，识别需求差异，推出有针对性的产品。目前，传统客户需求差异化服务的研究大都基于少量格式化的调查问卷，很难处理海量非结构化数据。鉴于电商网站及社交网络等有大量可获取的富含客户反馈的产品评论，有学者开始研究这些海量非结构化数据对需求获取的价值。但不少研究仅停留于分析粗粒度的、静态的需求，遗漏了许多较细节的、变化的需求。本课题以产品评论为研究对象，借鉴需求获取等理论，开发基于序列标注及联合聚类分析的多层次需求偏好识别算法，构建基于新词发现及多任务学习的需求时空差异比较模型，提出基于时间序列分析及整数规划模型的同类产品对比方法，形成面向需求差异比较的方法体系。研究成果将推动市场驱动环境中需求差异化数据的发现，完善需求获取及分析的相关理论，为需求管理等研究提供新思路，为企业在新产品开发中准确识别差异化需求及精准开发提供决策支持。

电子商务的高速发展为消费者提供许多相似产品。为赢得客户，企业应当充分理解消费者需求，清楚消费者对产品在不同时间不同地区的需求差异，分析同类产品的竞争优势，并根据此推出有针对性的符合用户需求的产品。当前，传统的客户需求差异化服务的研究大都是基于少量的格式化的调查问卷而展开的，没有考虑海量的具有丰富消费者观点态度的非结构化评论文本数据。鉴于电子商务网站及社交网络等有大量可获取的富含客户反馈的产品评论，有学者开始研究这些海量非结构化数据对需求获取的价值。但不少研究仅停留于分析粗粒度的、静态的需求，遗漏了许多较细节的、变化的需求。本课题以产品评论为研究对象，借鉴需求获取等理论，设计了基于概率图模型的客户需求提取的方法、基于词向量相似度和卡诺模型结合的感性工学视角下的用户需求挖掘研究方法、基于评论需求分析的领域文本数据驱动下本体构建和可视化展示的方法、基于语义相似和结构相似的跨领域文本数据驱动下本体构建和本体对齐的方法、基于词嵌入和主题模型相结合的短文本聚类的方法、基于序列标注的面向用户需求分析的产品评论用例提取的方法、基于新词发现的不同时间消费者客户需求差异分析的方法、基于层次贝叶斯模型的不同地区消费者客户需求差异分析的方法、基于整数规划的系列产品客户需求分析比较的方法等，形成面向需求差异分析比较的消费者观点大数据挖掘的方法体系。研究成果将从数据来源方面丰富市场驱动环境中需求差异化的发现路径，完善需求获取及分析中有关海量非结构化数据处理的相关理论，为需求管理等研究提供新思路，为企业在新产品开发中准确识别差异化的客户需求及面向消费者需求驱动的产品精准开发提供决策支持。