III: Medium: Collaborative Research: Mining and Leveraging Knowledge Hypercubes for Complex Applications

III：媒介：协作研究：挖掘和利用知识超立方体进行复杂应用

• 批准号: 2141037
• 负责人: Jing Gao
• 金额: $ 40万
• 依托单位: Purdue University
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-01-01 至 2024-09-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2141037&HistoricalAwards=false
• 关键词: III; Medium; Collaborative; Research; Mining

Knowledge repository refers to a machine-readable structure that stores knowledge about various entities (e.g., organizations, events, genes), which facilitates efficient information seeking. In many domains, knowledge varies with respect to contexts, and a flat structure that is commonly adopted by existing knowledge repositories cannot capture the complicated knowledge associated with different contexts. To make knowledge resources more findable, accessible, interoperable, and reusable (FAIR), this project plans to conceptualize a new structure, Knowledge Hypercube, for organizing and retrieving knowledge that could support complex applications in various domains. A knowledge hybercube organizes knowledge with respect to selected important dimensions (e.g., time, locations, conditions), and thus it allows people to easily access knowledge in any context, encapsulate distinctive entities and facts, and conduct cross-dimensional comparison and inference. This project impacts how people find and use knowledge, advances knowledge-based data analytics approaches, and benefits a wide range of domains which have gigantic literature and unsolved complex tasks by building a bridge between them. Knowledge hypercubes can also support educational innovation and contributes to educational tasks such as knowledge tracing. The major objective of this proposal is to form a paradigm of mining knowledge hybercubes from massive collection of text documents and leveraging such hybercubes for complex exploration and prediction tasks. To meet this goal, this project tackles a series of technical challenges. First, to automatically construct a knowledge hypercube from massive texts, innovative weakly supervised approaches are designed to organize text documents based on the hypercube structure, extract open entity and relationship information and organize cell-specific and cross-cell knowledge in a multi-dimensional manner. Second, novel refinement approaches are developed to automatically verify the information quality within and across cells in knowledge hypercubes by cross-checking within the hypercubes and with external information. Third, knowledge hypercubes motivate the development towards new discovery and learning tasks. In particular, the project introduces an automatic knowledge search pipeline for leveraging knowledge hypercubes for downstream prediction tasks, and a hypothesis generation approach for scoring unknown associations between concepts. The planned paradigm is realized in two specific domains (i.e., biomedical and news events), demonstrating the power of knowledge hypercubes to enable new insights into these domains.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.

知识存储库是指存储有关各种实体的知识的机器可读结构（例如，组织、事件、基因），这有助于有效的信息搜索。 在许多领域中，知识随着上下文而变化，现有知识库通常采用的扁平结构无法捕获与不同上下文相关联的复杂知识。 为了使知识资源更加可查找、可访问、可互操作和可重用（FAIR），该项目计划概念化一种新结构Knowledge Hypercube，用于组织和检索可以支持各个领域复杂应用程序的知识。 知识超立方体相对于所选择的重要维度（例如，时间、地点、条件），因此它允许人们在任何上下文中轻松访问知识，封装独特的实体和事实，并进行跨维度的比较和推理。 该项目影响人们如何查找和使用知识，推进基于知识的数据分析方法，并通过在它们之间建立桥梁，使具有庞大文献和未解决的复杂任务的广泛领域受益。知识超立方体还可以支持教育创新，并有助于知识追踪等教育任务。该建议的主要目标是形成一个从大量文本文档集合中挖掘知识超立方体的范例，并利用这种超立方体进行复杂的探索和预测任务。为了实现这一目标，该项目解决了一系列技术挑战。首先，为了从海量文本中自动构建知识超立方体，提出了基于超立方体结构的弱监督文本组织方法，提取开放的实体和关系信息，并以多维方式组织单元知识和跨单元知识。 第二，新的细化方法，开发自动验证内和跨细胞的知识超立方体内的交叉检查和外部信息的信息质量。 第三，知识超立方体激励开发新的发现和学习任务。特别是，该项目引入了一个自动知识搜索管道，用于利用知识超立方体进行下游预测任务，以及一种假设生成方法，用于对概念之间的未知关联进行评分。 计划的范例在两个特定领域中实现（即，该奖项反映了NSF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

AdaMix: Mixture-of-Adaptations for Parameter-efficient Model Tuning

• DOI: 10.48550/arxiv.2210.17451
• 发表时间: 2022-05
• 作者: Yaqing Wang;Subhabrata Mukherjee;Xiaodong Liu;Jing Gao;Jianfeng Gao

LightToken: a Task and Model-agnostic Lightweight Token Embedding Framework for Pre-trained Language Models

LightToken：用于预训练语言模型的任务和模型无关的轻量级令牌嵌入框架

• 发表时间: 2024
• 期刊: KDD '23: Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining
• 作者: Wang, Haoyu;Li, Ruirui;Jiang, Haoming;Wang, Zhengyang;Tang, Xianfeng;Bi, Bin;Cheng, Monica;Yin, Bing;Wang, Yaqing;Zhao, Tuo
• 通讯作者: Zhao, Tuo

InterHG: an Interpretable and Accurate Model for Hypothesis Generation

InterHG：可解释且准确的假设生成模型

• DOI: 10.1109/bibm52615.2021.9669740
• 发表时间: 2021
• 期刊: IEEE International Conference on Bioinformatics and Biomedicine (BIBM
• 作者: Wang, Haoyu;Wang, Xuan;Wang, Yaqing;Xun, Guangxu;Jha, Kishlay;Gao, Jing
• 通讯作者: Gao, Jing

Multimodal Emergent Fake News Detection via Meta Neural Process Networks

• DOI: 10.1145/3447548.3467153
• 发表时间: 2021-06
• 期刊: Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining
• 作者: Yaqing Wang;Fenglong Ma;Haoyu Wang;Kishlay Jha;Jing Gao

Concept-Level Model Interpretation From the Causal Aspect

• DOI: 10.1109/tkde.2022.3209997
• 发表时间: 2023-09
• 期刊: IEEE Transactions on Knowledge and Data Engineering
• 影响因子: 8.9
• 作者: Liuyi Yao;Yaliang Li;Sheng Li;Jinduo Liu;Mengdi Huai;Aidong Zhang;Jing Gao