Supporting Exploratory Search via the Semantic Web

我的工作在很大程度上是关于如何通过有效地展示如何探索信息以及如何表现可以探索的来源的汇流，在正确的时间改善对网络上正确信息的获取。我的目标是找到方法，让任何人都能更容易地从他们所知道的开始，并能够利用这些来帮助他们探索信息，建立他们想要的知识。例如，一个挑战是帮助对古典音乐一无所知的人找到他们可能喜欢的古典音乐。对于这一挑战，我们创建了一种新的界面范例，将音频提示与丰富上下文中的信息联系起来，以便人们可以使用他们所知道的--无论他们是否喜欢他们听到的--以及他们想要的--能够探索领域，找到他们喜欢听的片段。本文档封面显示了该交互的屏幕截图，可在beta.mspace.fm上试用。这项工作既涉及前端需求，也涉及分布式异类信息的设计和信息体系结构。最近，这项工作集中在语义网交付信息的潜力上，通过基于机器的推理，这种信息可以提供比当前Web更丰富的信息点之间的关联。信息不仅可以通过原始搜索结果浏览，还可以通过多个上下文进行浏览，从当前的工作情况到一个人可能给出的返回来源的信任程度。例如，对在语义支持的Web上探索乳腺癌感兴趣的人将能够从多种属性探索这一概念，如病因、治疗方法、历史观点、当前研究、诊断实践等。观察治疗的非专业人员可能会看到谁领导了一种特定的方法，并可能希望了解他们还进行了哪些其他工作，同样地，也可能希望探索人们自己对治疗的体验。他们可能希望主要查看以非专业术语呈现的对这项工作的描述，但不一定将医学期刊排除在他们的视野之外。这里的第一个挑战是如何使这些相关信息网络易于处理，以便能够进行探索，并以最佳方式尊重搜索本身的背景：非专业读者。与信息的表示相关的是以信任的方式表示信息，信任是语义网的关键组件。什么东西什么的。因此，上述方案中的关键挑战是：呈现高度关联的复杂信息网络的一般策略用于跨维度搜索的创新查询。策略/信任规模建议的跨三个站点的协作有三个阶段，这些阶段相互构建。到目前为止，语义网研究中的交互工作已经建立在已知知识库的基础上，其中图形是预先知道的，并且用户界面可以处理这些已知的约束。在麻省理工学院的DIG实验室，我们将研究如何从这些方法中总结出支持更多类似Web的交互。也就是说，如何在没有预先构建的该域部分表示的连通图中进行探索。描绘这个空间的一个挑战是想象一个人在古典音乐的背景下看待贝多芬时，如何能够在探索的同时看到导致贝多芬和拿破仑之间政治联系的途径。麻省理工学院的工作重点将是考虑为这种有意义的探索呈现大规模的Web规模图。在马里兰大学的MindLab和HCIL实验室，我们将研究如何在这样的探索中使政策/信任机制易于处理。为标准的互动实践提出一个模型是不够的，关键是要对其进行评估。Web上的这些新技术提供的新类型的探索需要创新的评估方法：跨(语义)Web等高度关联的空间的任务可以移动

My work has largely been concerned with how to improve access to the right information at the right time on the Web via effective presentation both of how to explore information as well as how to represent the confluence of sources that can be explored. My goal is to find ways to make it ever easier for anyone to start with what they know and be able to use that to help them explore information to build the knowledge they want. For instance, one challenge was to help someone who knew nothing about Classical Music find Classical Music they might like. For this challenge, we created a new interface paradigm that connected audio cues with information in a rich context so that one could use what they did know - whether or not they liked what they heard - with what they wanted - being able to explore the domain to find pieces they would enjoy hearing. The cover of this document shows a screen shot of that interaction, and can be tried at beta.mspace.fm. The work involved both front end requirements and design, information architecture for distributed heterogeneous information Most recently this work has focused on the potential of the Semantic Web to deliver information that can, by machine-based inference, provide richer associations among points of information than the current Web. Information can be explored not only via raw search results but via multiple contexts, from current work situation to the level of trust one may give a returned source. For example, a person interested in exploring breast cancer on a semantically enabled Web will be able to explore this concept from multiple attributes, such as causes, treatments, historical perspectives, current research, diagnostic practices and so on. A lay person looking at treatments may see who has lead a particular approach, and may wish to see what other work they've carried out and, equally, explore what people's own experiences of the treatment have been. They may wish to look primarily at descriptions of this work presented in non-specialist terms, but not have medical journals necessarily excluded from their view. The first challenge here is how to make these networks of associated information tractable in order to be explorable and in a way that best respects the context of the search itself: a lay reader. Related to the representation of the information is presenting this information in terms of Trust, a key component of the Semantic Web. Something something. Key challenges in the above scenario therefore are:General strategies for presenting complex networks of highly associated informationInnovative queries for cross dimensional searches.Policy/Trust ScaleThe proposed collaboration across three sites has three phases which build upon each other. So far, interation efforts in Semantic Web research have been built on top of known knowledge bases, where the graph is known in advance and the UI can work with these known constraints. At MIT's DIG lab, we will look at how to generalize from these approaches to support more Web-like interaction. That is, how to explore across connected graphs where there is no pre-constructed representation of part of that domain. A challenge to illustrate this space is to imagine how a person looking at Beethoven within the context of Classical Music would be able to see paths, while exploring, that would lead to political connections between Beethoven and Napoleon. The focus of work at MIT will be to consider representing massive Web scale graphs for such meaningful exploration. At the University of Maryland's MindLab and HCIL lab, we will be looking at how policy/trust mechanisms can be made tractable within such explorations. It is not sufficient to propose a model for standard interaction practice: it is critical to evaluate it. The new types of exploration afforded by these new technologies on the Web call for innovative methods in evaluation: tasks across a richly associated space like the (Semantic) Web can move