III: Small: Statistical Knowledge Translation and Knowledge Integration Using Markov Logic

随着知识库的快速扩散及其在各种应用程序中的自动推理使用，越来越需要有效的方法来实现(i)知识翻译，即将在一个领域学习或开发的知识应用到另一个语义不同的领域的任务，以及（ii）知识集成，即从不同来源构建统一知识库的任务。虽然数据翻译和集成问题在文献中得到了大量关注，但知识集成问题的探索相对较少。关于这一主题的现有工作主要集中在逻辑框架上，这使得难以考虑知识和用于知识翻译和集成的语义映射中的不确定性。本项目旨在利用马尔可夫逻辑网络同时表达知识和语义映射，获得统一的概率模型，用于共同翻译知识和提炼语义映射。这为解决异构知识与不确定映射的集成挑战、评估翻译知识的正确性、简化结果模型以获得更紧凑的近似翻译以及在实际应用场景中评估方法提供了基础。所建议的工作的结果可能适用于许多需要跨不同知识来源集成或转换知识的应用领域。这项工作加强和促进了跨学科的合作，为研究生和本科生提供了更多的基于研究的培训机会。知识翻译和集成软件以及基准数据将通过项目网站http://aimlab.cs.uoregon.edu/SKTI/向更广泛的社区提供

With the rapid proliferation of knowledge bases and their use in automated inference in a variety of applications, there is a growing need for effective approaches for (i) knowledge translation, i.e., the task of applying knowledge learned or developed in one domain to another semantically different domain, and (ii) knowledge integration, i.e., the task of building a unified knowledge base from disparate sources. While the problem of data translation and integration has received a great deal of attention in the literature, the problem of knowledge integration remains relatively under-explored. Existing work on this topic has focused on primarily logical frameworks which make it difficult to take into account uncertainty in knowledge and in semantic mappings used for knowledge translation and integration. This project aims to use Markov Logic Networks to express both knowledge and semantic mappings to obtain a unified probabilistic model for jointly translating knowledge and refining semantic mappings. This provides a basis for addressing the challenges of integrating heterogeneous knowledge with uncertain mappings, assessing the correctness of translated knowledge, simplifying the resulting models to obtain more compact approximate translations, and evaluating the methods in realistic application scenarios. The results of the proposed work are likely to be applicable in a number of application domains that require integration or translation of knowledge across disparate knowledge sources. The work strengthens and facilitates interdisciplinary collaborations, provides enhanced research-based training opportunities for graduate and undergraduate students. The knowledge translation and integration software, and the benchmark data will be made available to the broader community through the project web site: http://aimlab.cs.uoregon.edu/SKTI/