协同对称蕴涵熵推理及其在情感计算中的应用研究

In the field of fuzzy reasoning, the triple I method has the unique advantage from the viewpoint of logic, which has laid its international status. However, the triple I method is not perfect at the meaning of fuzzy system, resulting in that the triple I method is faced with its development bottleneck. Focusing on this problem, the symmetric implicational method (previously proposed by us as a generalization of the triple I method) is regarded as the starting point. Aiming at the intuitionistic fuzzy environment, the principle of maximum information entropy is introduced, and then the symmetric implicational entropy inference idea is put forward. Moreover, the multiple and multi-dimensional fuzzy reasoning has the problems of heavy calculation burden and curse of dimensionality. For these problems, the ideas of cooperative computing and multi-level mode are successively introduced, and then the cooperative symmetric implicational entropy inference method is proposed. This proposal will investigate the new fuzzy reasoning method, which includes four parts. First, the properties and relationship of various operators are analyzed, and the solving mechanism and rationality of the symmetric implicational entropy inference method are explored. Second, collaborative task partitioning is carried through via fuzzy clustering. The multi-level model and similarity activation strategy are researched, and based on them the solving of the collaborative symmetric implication reasoning entropy method is realized. Third, the fuzzy system via the proposed method is established, and more ideal system are obtained by virtue of performance analyses. Lastly, on the basis of our large-scale Chinese emotional corpus, the emotion deduction system, which the new emotion is recognized by the basic emotions, is researched with utilizing the mentioned-above achievement. Then the emotional corpus is expanded, and all of these are integrated into an emotional service robot. This purpose of this proposal is to try to bring a breakthrough for the framework of triple I method, and promote the theory establishment and application generalization of fuzzy reasoning and affective computing.

在模糊推理领域中，逻辑层面的独特优势奠定了三I方法的国际地位，但在模糊系统层面的不理想导致其遭遇瓶颈期。对此，以我们先前提出的对称蕴涵方法（三I方法的推广）为起点，本申请面向直觉模糊环境，融入最大信息熵原则，建立对称蕴涵熵推理的新构想；并针对多重多维推理的计算量大和维数灾难问题，引入协同计算、多级模式的理念，提出协同对称蕴涵熵推理方法。研究内容包括：①剖析各类算子的性质关联，探究对称蕴涵熵推理方法的求解机制及合理性；②通过模糊聚类进行协同任务划分，探寻多级模式、相似度激活策略，实现协同对称蕴涵熵推理方法的求解；③建立以该方法为内核的模糊系统，凭借性能分析，总结出更多的理想系统。④以我们的大规模中文情感语料库为基础，基于上述成果探究从基本情感识别新情感的情感推断系统，扩充情感语料库，集成到情感服务机器人中。试图籍此给三I方法的体系带来突破，带动模糊推理、情感计算的理论建设与应用辐射。

逻辑层面的独特优势奠定了三I方法的国际地位，但在模糊系统层面的不理想导致其遭遇瓶颈期。对此，以我们先前提出的对称蕴涵方法（三I方法的推广）为起点，本申请面向直觉模糊环境，建立对称蕴涵熵推理的新构想；并针对计算量大和维数灾难问题，引入协同计算、多级模式的理念，提出协同对称蕴涵熵推理方法。研究成果包括：①剖析了各类算子的性质关联，面向单重单维的情形，获得对称蕴涵熵推理方法的求解策略，从还原性、连续性、鲁棒性等角度验证了对称蕴涵熵推理方法的合理性。②面向多重多维的情形，探讨协同对称蕴涵熵推理方法的解的属性，通过模糊C均值算法的直觉模糊集版本进行协同任务划分，形成多级模式、相似度激活策略，对协同对称蕴涵熵推理方法形成了统一形式的求解模式，推动了模糊推理理论的进步。③通过叠加模糊器、解模糊器，建立了以协同对称蕴涵熵推理方法为内核的模糊系统，发现其总体响应性能较优，并总结了较为优异的模糊系统。④将协同对称蕴涵熵推理方法应用于情感计算中，扩充获得更全面的语料库，实现从已有基本情感推出新情感的情感推断系统，促进了基于该语料库的情感计算理论与应用体系的进步，最后集成到情感服务机器人中。通过本项目的实施，在IEEE Transactions on Cybernetics、Applied Soft Computing等期刊和会议上发表（或录用）论文23篇（均标注基金号）。完成先前语料库的扩充，其中情感类别达到20种，词规模达到4500万。申请国家发明专利7项，其中5项获得授权。获得国家软件著作权登记5项。团队举办了2次会议。唐益明新增为SCI期刊Journal of Intelligent & Fuzzy Systems和IEEE Access的Associate Editor、SCI期刊PLOS ONE编委、中国人工智能学会机器学习专业委员会委员、ISKE 2019国际会议程序委员会委员等。

内容获取失败，请点击重试