CRII: III: Beyond Similarity Learning: Complementarity Learning for Contextual Behavior Modeling

CRII：III：超越相似性学习：情境行为建模的互补学习

• 批准号: 1849816
• 负责人: Meng Jiang
• 金额: $ 17.49万
• 依托单位: University of Notre Dame
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-10-01 至 2021-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1849816&HistoricalAwards=false
• 关键词: CRII; III; Beyond; Similarity; Learning

Given the complexity of human behaviors, it is difficult to develop a successful plan and make right decisions. Behavior data in fields such as social media, education, and academic research have been increasingly available for behavioral pattern discovery, decision making, and planning. Complementarity has been revealed of playing a significant role in many fields: partners need complementary strengths to do successful business; courses need complementary teaching materials to achieve effective student learning. Therefore, the representation of human behaviors should preserve the complementarity information rather than the similarity. The purpose of this project is to develop complementarity learning models to advance our understanding of human behaviors in dynamic, social, and spatiotemporal environments, and practically, to facilitate prediction, recommendation, and decision-making and planning processes towards the effectiveness of behaviors. This project will also support educational and outreach programs that will broaden participation in computer science. Open source software implementations of the new algorithms will be made available to the public, and will also serve as an educational tool for junior researchers. Research supervision and career mentoring will be made available to K-12 students through the development and publication, and a new course in data science and behavior modeling will be offered to undergraduate and graduate students.This project will develop and evaluate novel behavior modeling methods that learn the representation of human behavior by preserving the structure of complementarity among the behavior's components. The idea is that decision makers are looking for not similar but complementary partners, resources, and conditions that provide extra power to make a behavior plan more effective. In this project, principled metrics of complementarity that satisfy intuitive axioms will be proposed; complementarity representation learning methods will be developed, applied, and evaluated on prediction and recommendation tasks. In addition, this project will result in an online recommender system that facilitate young researchers for project teaming and planning. The results will also be disseminated through tutorial and workshop organization at international conferences.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

考虑到人类行为的复杂性，制定一个成功的计划并做出正确的决定是很困难的。社交媒体、教育和学术研究等领域的行为数据越来越多地用于行为模式发现、决策制定和规划。互补性在许多领域发挥着重要作用：合作伙伴需要优势互补才能成功开展业务；课程需要补充教材，以实现学生有效的学习。因此，人类行为的表征应该保留互补性信息而不是相似性。该项目的目的是开发互补性学习模型，以促进我们对动态、社会和时空环境中人类行为的理解，并在实践中促进对行为有效性的预测、推荐、决策和规划过程。该项目还将支持教育和推广计划，以扩大计算机科学的参与。新算法的开源软件实现将向公众开放，也将作为初级研究人员的教育工具。通过开发和出版，将为K-12学生提供研究监督和职业指导，并为本科生和研究生提供数据科学和行为建模的新课程。该项目将开发和评估新的行为建模方法，通过保留行为组件之间的互补性结构来学习人类行为的表示。这个想法是，决策者正在寻找的不是相似的，而是互补的合作伙伴、资源和条件，以提供额外的力量，使行为计划更有效。在这个项目中，将提出满足直观公理的互补性原则度量；互补性表示学习方法将在预测和推荐任务中得到开发、应用和评估。此外，该项目将形成一个在线推荐系统，方便年轻研究人员进行项目团队和规划。研究结果还将通过在国际会议上组织指导和讲习班来传播。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

项目成果

Precise temporal slot filling via truth finding with data-driven commonsense

通过数据驱动的常识发现真相来精确填充时隙

• DOI: 10.1007/s10115-020-01493-w
• 发表时间: 2020
• 期刊: Knowledge and Information Systems
• 影响因子: 2.7
• 作者: Wang, Xueying;Jiang, Meng
• 通讯作者: Jiang, Meng

Modeling Co-Evolution of Attributed and Structural Information in Graph Sequence

• DOI: 10.1109/tkde.2021.3094332
• 发表时间: 2023-02
• 期刊: IEEE Transactions on Knowledge and Data Engineering
• 影响因子: 8.9
• 作者: Daheng Wang;Zhihan Zhang;Yihong Ma;Tong Zhao;Tianwen Jiang;N. Chawla;Meng Jiang

Few-Shot Knowledge Graph Completion

• DOI: 10.1609/aaai.v34i03.5698
• 发表时间: 2019-11
• 作者: Chuxu Zhang;Huaxiu Yao;Chao Huang;Meng Jiang;Z. Li;N. Chawla

Action Sequence Augmentation for Early Graph-based Anomaly Detection

• DOI: 10.1145/3459637.3482313
• 发表时间: 2020-10
• 期刊: Proceedings of the 30th ACM International Conference on Information & Knowledge Management
• 作者: Tong Zhao;Bo Ni;Wenhao Yu;Zhichun Guo;Neil Shah;Meng Jiang

Graph Few-shot Learning via Knowledge Transfer

• DOI: 10.1609/aaai.v34i04.6142
• 发表时间: 2019-10
• 期刊: ArXiv
• 作者: Huaxiu Yao;Chuxu Zhang;Ying Wei;Meng Jiang;Suhang Wang;Junzhou Huang;N. Chawla;Z. Li