CHORUS - Top-k Composition of Browsing Scripts for efficient social-enabled Usage of Web-based Services

点击翻译按钮获取中文摘要

The Web provides access to a huge amount of information and functionality with billions of websites that are highly distributed and maintained autonomously by their respective providers. For increasingly sophisticated use cases an end user needs to extract, combine, and aggregate information from various web pages belonging to different websites. Often required information lies deep in the web, that is, it is accessible only after performing certain interactions with the web site. Furthermore, it is even more time consuming when such complex tasks are recurring. Current technologies and state of research offer little support for end users in efficiently compiling together the required information from various web sites. Search engines usually focus on recommending websites that a user may or should visit, but not on answering the overall information need of the user. Approaches such as Semantic Web and Linked Data take a static view on the data and rely on cooperation of providers. Web automation scripts, initially developed for test websites, allow end users to capture their browsing activities as executable processes and share them with other end users. Such a technique is very promising for bringing end users into a provider oriented web. However, in order to benefit the full potential of end user scripts, especially the reusability feature, efficient script search and composition techniques are required.In this project we will develop methods for top-k composition of browsing scripts. Our methods will be based on formal semantics of the scripts, their functional and non-functional properties, user preferences and structured queries while still supporting syntactic heterogeneity for the purpose of interoperability. In particular, we will develop methods for ranking scripts and script compositions according to user preferences as well as retrieval and composition of scripts to full the requirement specified in a structured query. In contrast to existing ranking mechanisms that focus on precise computation of a rank and therefore have to compromise on the expressivity to still an acceptable performance, we will allow specification of vague preference by means of fuzzy rules. Fuzzy rules are more expressive than the existing preference languages and there exist efficient inference methods for fuzzy rules. We will investigate whether and how the rank of a script composition can be computed in a compositional fashion from the ranks of the component scripts. In order to retrieve matching scripts for a given query efficiently, we will augment naïve model-checking techniques by appropriate indexes. In order to achieve efficient script composition, we will develop methods based on Plan Space Planning. The composition method will make use of the retrieval method. In both the methods we aim at computing only top-k instead of complete list of answers.