SBIR Phase II: Relational Bayesian Modeling for Electronic Commerce

这个小企业创新研究的第二阶段项目将专注于扩大和验证公司的关系模型发现技术，并将具体的应用重点放在网络访问者行为建模上。在第一阶段的研究中，公司开发了一种基于综合变量语言的建模范式，用于关系贝叶斯建模。它的综合变量语言是开发一种原则性方法来表示、发现和执行带有混合表内、跨表和多表关系特征的概率推理的第一次综合努力。这种能力为构建全面的、集成的关系数据模型提供了基础。构建的模型捕获了网站访问者行为的丰富细节，并可用于实时推断网站访问者的意图（例如，是否计划购买）。这些结果是任何其他建模技术都无法获得的。第二阶段项目的技术目标是：(1)开发一种完整的语言，为我们的综合变量能力中突出的问题建立解决方案；(2)设计商业部署所需的基础设施；(3)构建可部署的网络访问者行为模型，以识别干预的机会；(4)进行基于模型的干预的实地试验，以建立我们方法的商业价值。现代社会的一个悖论是，相对于我们收集和存储的数据量，我们拥有的知识太少了。电子商务就是这种悖论的典型例子。电子商务平台收集了前所未有的大量关于客户交互的信息，然而今天的电子商务应用程序并没有提供客户所期望的服务或在线零售商所要求的性能。在线零售商要求越来越复杂的营销和销售技术。拟议中的产品将使在线商家和服务提供商能够有效和综合地理解在线消费者的行为。提议的产品将带来一种新的以客户为中心（而不是以页面为中心）的基于Web的交互，这将有助于万维网作为一种通信媒介的发展。该公司的技术也适用于离线科学分析，作为电子商务领域复杂关系数据的假设生成方法。这项技术使科学家能够更好地利用他们掌握的数据。

This Small Business Innovation Research Phase II project will focus on scale-up and validation of the company's relational model discovery technology, with specific application focus on web-visitor behavior modeling. In Phase I research the company developed a modeling paradigm based a synthetic variable language for relational Bayesian modeling. Its synthetic variable language is the first comprehensive effort to develop a principled way to represent, discover, and perform probabilistic inference with mixed intra-table, cross-table, and multi-table relational features. This capability provides the basis for construction of comprehensive, integrated models of relational data. Models constructed capture the rich detail of web visitor behavior and can be used to make inferences about web visitor intent (e.g., whether or not a purchase in planned) in real- time. These results are not obtainable by any other modeling technology. The technical objectives for the Phase II project are to: (1) develop a complete language to establish solutions to outstanding issues in our synthetic variable capability, (2) engineer the infrastructure needed for commercial deployment, (3) construct deployable models of web visitor behavior to identify opportunities for intervention, and (4) conduct field-trials of model-based interventions to establish the business value of our approach. A paradox of modern society is that we possess so little knowledge relative to the amount of data we collect and store. E-commerce provides a paradigmatic example of this paradox. E-Commerce platforms collect unprecedented amounts of information about customer interactions, yet today's E-commerce applications do not provide the service expected by customers or the performance demanded by online retailers. Online retailers are demanding increasingly sophisticated marketing and merchandising technologies. The proposed product will empower online merchants and service providers by enabling efficient and integrated understanding of online consumer behavior.The proposed product will bring in a new class of customer centric (instead of page-centric) web-based interactions that will contribute to the evolution of the World Wide Web as a communication medium. The company's technology also applies to offline scientific analysis as a method for hypothesis generation in complex relational data as in the E-commerce domain. This technology enables scientists to make better use of the data at their disposal.