Pattern Mining of Big Data

***本项目的目标有三个：(1)利用高效的聚类技术和概率推理机制分析大规模多层网络，用于社区检测和链路预测任务；(2)为大数据分析设计准确且可扩展的解决方案；(3)为复杂网络分析的多维数据挖掘带来新的贡献。***我们将依靠形式概念分析，这是一个理论上健全的框架知识表示，模式提取和可视化。这意味着我们将继续为这一领域做出新的贡献，同时也将探索数据挖掘、机器学习和概率推理领域的其他突出方法。***为了实现第一个目标，我们将设计新的社区检测和进化算法，要么将基于消息传递的聚类与基于混合密度的聚类相结合，要么将基于消息传递的聚类与基于相似度的聚类相交叉。本文还将设计用于链路预测的新的概率推理算法。对于第二个目标，我们将依靠分布式数据库理论、机器学习和形式化概念分析的理论方法和实践工具来分析大数据问题。这包括数据碎片和分解、并行算法设计和实现以及面向用户的模式挖掘。对于第三个目标，我们将首先丰富我们最近在三元概念分析方面的工作，进一步研究并为多元概念分析在复杂网络分析中的应用带来新的贡献。******短期和长期目标***短期目标包括以下研究活动：(1)继续我们正在进行的工作，定义新的高效算法，用于在三元概念分析、数据和晶格分解中挖掘集群和关联规则；(2)利用大型（社会和生物）网络中经常存在的关键结构属性，研究新的准确和可扩展的概率推理算法，用于推理和挖掘服务；(3)开发新的高效聚类技术，用于社交网络中的社区检测；(4)利用海量数据属性来改进学习目的的推理过程。中期目标旨在(1)为数据聚类和概率推理提供新的可扩展算法，以及(2)探索工业和商业中的高级案例研究。长期目标是建立在短期和中期目标的基础上，进一步解决不那么琐碎的问题，并为发现和管理大数据问题中的模式开发一个强大而灵活的环境。

OBJECTIVES***The objectives of the proposed program are threefold: (1) analyze large-scale multilayer networks for community detection and link prediction tasks using efficient clustering techniques and probabilistic inference mechanisms, (2) design accurate and scalable solutions for big data analytics, and (3) bring new contributions to multidimensional data mining for complex network analysis. ***We will rely on Formal Concept Analysis, which is a theoretically sound framework for knowledge representation, pattern extraction, and visualization. This means that we will continue bringing new contributions to this field, besides also exploring other prominent approaches from data mining, machine learning and probabilistic inference fields.******SCIENTIFIC APPROACH***To achieve the first objective, we will design new community detection and evolution algorithms by either combining message passing-based with mixture density-based clustering or by interleaving message passing-based and similarity-based clustering. New probabilistic inference algorithms for link prediction will be designed as well. For the second objective, we will rely on theoretical methods and practical tools from distributed database theory, machine learning, and formal concept analysis to analyze big data problems. This includes data fragmentation and decomposition, parallel algorithm design and implementation, and user-oriented pattern mining. For the third objective, we will first enrich our recent work on triadic concept analysis to further study and bring new contributions to polyadic concept analysis with applications in complex network analysis.******SHORT AND LONG TERM OBJECTIVES***The short-term objective involves the following research activities: (1) continue our ongoing work on defining new efficient algorithms for mining clusters and association rules in triadic concept analysis, and in data and lattice decomposition, (2) exploit key structural properties that often exist in big (social and biological) networks, to investigate new accurate and scalable probabilistic inference algorithms for reasoning and mining services, (3) develop new efficient clustering techniques for community detection in social networks, and (4) leverage massive data properties to improve inference procedures in learning purposes. The medium-term objective aims at (1) deriving new scalable algorithms for both data clustering and probabilistic inference, and (2) exploring advanced case studies in industry and business. The long-term objective is to build on the results of our short and medium-term objectives to further tackle less trivial issues and develop a robust and flexible environment for the discovery and management of patterns in big data problems.