钱炘祺 xinqiqian@gmail.com 一种面向实体浏览中属性融合的人机交互的设计与实现 Designing Human-Computer Interaction of Property Consolidation for Entity Browsing 钱炘祺 xinqiqian@gmail.com.

Presentation on theme: "钱炘祺 xinqiqian@gmail.com 一种面向实体浏览中属性融合的人机交互的设计与实现 Designing Human-Computer Interaction of Property Consolidation for Entity Browsing 钱炘祺 xinqiqian@gmail.com."— Presentation transcript:

1 钱炘祺 xinqiqian@gmail.com
一种面向实体浏览中属性融合的人机交互的设计与实现 Designing Human-Computer Interaction of Property Consolidation for Entity Browsing 钱炘祺

2 Outline Scenario Task & Approach Technical Problems entity browsing
property consolidation crowdsourcing HCI related work Task & Approach baseline graph display property pair & questioning Technical Problems

3 Entity browsing semantic browsing organizing and presenting properties
retrieving, presenting and traversing information and data on semantic web organizing and presenting properties order of properties facet: property partition simple: property consolidation 语义网浏览简单的说是检索、展示和遍历信息和语义web数据 落实到实体上，实体浏览则主要在呈现上。 除了解决实体共指等挑战，对某一确定实体进行浏览时，我们更关注如何对该实体的信息进行展示。 换句话说就是，如何组织并呈现property 这里为了达到一个有效且友好的浏览体验，至少有这几方面： 属性呈现的顺序/属性的划分以将更相关的属性聚在一起/属性融合已消减冗余属性保持简洁

4 Property consolidation
equivalent property different property   same meaning average f-measure alignment for properties is quite low (0.21) versus classes (0.68) (OAEI) False Positive In some cases the domain or range of the matched properties indicate that they are not being used in the same way. False Negative lower string similarity domain and range have strong syntactic similarity //我希望研究的目标对象就是这里的等价属性，我们希望能够对其进行融合以达到 对属性简洁且有组织地呈现 的目的。 property consolidation/alignment 说到这里的属性融合 其所谓要融合的等价属性 一般即指语义上等价 简单来说 不同的属性实际上表达的是实体同一个方面的特征。 Workshop on Ontology Matching.机器自动融合的效果目前不佳。f-measure只有0.21。 主要是误报和漏报都比较高。根结还是目前大多使用字符串匹配为基础，而有时字符相似的属性不一定等价，而有些等价的属性在字面上也不一定相似。

5 Crowdsourcing balance of complexity & effect
simple participating considerable improvement show user current consolidation situation (group/graph/tree) and let them do further consolidation manually list property pairs for user to confirm (questioning model) 复杂性与效率的平衡 轻度的参与/可观的效果 呈现+用户手动参与 提问方式

6 HCI related work entity coreference resolution HCI

7 HCI related work entity coreference resolution HCI

8 HCI related work entity coreference resolution HCI

9 Task Design a human-computer interaction for crowdsourcing of property consolidation baseline graph visualization & manually consolidation list property pairs & questioning

10 baseline

11 graph & manually

12 property pair & questioning

13 Technical Problems what kind of data to send back to the system for improving by machine learning information for learning params of a model? information for active learning? the order of questions in the property pairs questioning approach answer type of questions in the property pairs questioning approach programming method for graph display on web page 目前想到的技术难点有以下几个： 前台界面获取用户的什么信息反馈给后台系统，简洁的说就是采取什么大众参与方案 例如若后台选择机器学习学一个将机器算的和大众参与的平均考虑的模型 那如何提供数据来学习参数 若后台选择是主动学习 那前台应该返回什么结果 如果采取property对提问的方案，那么如何选择提问的顺序来取得尽量少的提问次数 回答的种类 是/否/不确定 表示相同的事情/一个更宏观或者更细致/在某种其他方面相关/无关 they mean the same thing, one is a more general or more specic term than the other, they are related in some other way, or there is no relation. 在可视化方面如何通过网页编程展现图以及交互效果等

14 Reference [1] Cheatham M, Hitzler P. The Properties of Property Alignment[C]//Proceedings OM-2014, The Ninth International Workshop on Ontology Matching, at the 13th International Semantic Web Conference, ISWC [2] 语义网实体链接的人机交互界面设计与实现 仇宏磊 2013 [3] Vesdapunt N, Bellare K, Dalvi N. Crowdsourcing algorithms for entity resolution[J]. Proceedings of the VLDB Endowment, 2014, 7(12).

15 Thanks Q & A ?