基于案例、社会化媒体评论与知识的突发事件应对流程的智能生成方法研究

Intelligent generation of the emergency response process has been becoming the urgent problem to be solved. First, for the problem of incompleteness of response knowledge and lack of fine-grained response knowledge, the methods of data mining and machine learning are adopted to the study on fine-grained response rule mining and transfer on emergency cases from the perspective of "Micro-scenario—Response". It is useful to improve the completeness and generalization ability of response knowledge. Second, for the problem of incomplete response evaluation for existing emergency cases, especially lack of public satisfaction evaluation on the emergency response, a new idea for supplement the response evaluation using emotional analysis results of online reviews in social media is proposed. Based on the idea mentioned above, the response evaluation method based on the D-S evidence theory is studied by considering the blog (or post) and corresponding online reviews in the social media as the evidence. Third, the intelligent generation method of alternative response processes based on knowledge-based and case-based reasoning is studied. Furthermore, multi-source evaluation and optimal selection methods of alternative response processes are studied, considering expert evaluation, response evaluation in similar cases, and satisfaction evaluation of public on similar cases. Finally, the prototype of experimental platform is developed, and the application of related research results is conducted. Related research results can not only rich the theoretical method system of emergency management and business process management, but also help to improve the rationality of response process and timeliness of its generation.

突发事件应对流程的智能生成已成为急需解决的问题。首先，针对应对知识不完备、缺乏细粒度应对知识的问题，运用数据挖掘及机器学习的方法，研究基于案例的“微情景—应对活动”的细粒度应对规则的挖掘与迁移方法，以提高应对知识的完备性与泛化能力。其次，针对现有应急案例中应对效果评价不全面，尤其缺乏公众对突发事件应对效果满意度评价等问题，提出利用社会化媒体评论的情感分析结果补充案例应对效果评价的新思路，将社会化媒体中的博文(或帖子)及其评论作为证据，研究基于D-S证据理论的案例应对效果评价方法。进一步的，研究基于知识和案例推理的备选应对流程智能生成方法，进而在考虑专家评价与相似案例中应对效果的基础上，结合公众对相似案例应对效果的满意程度，研究备选应对流程的多源评价与优选方法。最后，进行实验平台原型研发和应用验证。研究成果既可丰富应急管理和业务流程管理的理论方法体系，亦可提升应对流程的合理性与生成及时性。

突发事件具有破坏性、复杂性和不确定性等，频繁威胁着人们的生命和财产安全。针对突发事件应对知识不完备、应急专家的主观应对经验有限，现有应急案例中应对效果评价不全面，尤其缺乏公众对突发事件应对效果满意度测度等问题，项目组围绕基于案例、社会化媒体评论与知识的突发事件应对方案的智能生成方法开展了研究，取得了如下具有理论价值和应用价值的研究成果。其一，从数据与知识驱动的视角，分别提出了基于改进Bond指标的突发事件应对规则挖掘方法和基于改进遗传算法的突发事件应对规则挖掘方法，并借鉴迁移学习的思想，提出了数据与知识驱动的基于领域相似度的突发事件应对规则迁移方法。其二，从公众视角出发，提出了基于多视角集成学习的微博情感分类方法、基于三支决策和深度相互学习的微博情感分类方法，进一步地，提出了基于微博舆情分析的突发事件应对效果测度方法。其三，结合专家们进行不确定决策时的实际情形，提出了知识引导下考虑风险态度和自信行为的应急响应等级决策方法，以及双信度下考虑专家风险态度和研判调整行为的群体性突发事件等级确定方法；此外，综合考虑专家群体和公众的意见，提出公众关注视角下考虑专家影响力和评价值一致性的应急决策方法，并针对专家应对经验较为匮乏的情形，提出了案例驱动的基于改进RIMER的备选应对方案评价与优选方法。其四，基于上述研究，研发构建了实验平台原型系统。本项目研究成果不仅可以丰富突发事件应急管理的理论与方法，也能够为实际应急决策提供参考与支持，有助于提高应急决策的科学性，进而有效降低突发事件所造成的危害，因此具有较强的理论意义和较为广阔的应用前景。