The TriQL.P Browser: Filtering Information using Context-, Content- and Rating-Based Trust Policies

The TriQL.P browser is a general purpose RDF browser that supports users in exploring RDF datasets containing information from multiple information sources. Information can be filtered using a wide range of user-definable trust policies. Policies can be based on information context, information content, information or information source ratings, and on the presence or absence of digital signatures. In order to help users understand the filtering decisions, the browser can explain why a piece of information fulfils the selected trust policy. 1 Trust Policies for the Semantic Web The Semantic Web is an open, dynamic network of independent information providers all having different views of the world, different levels of knowledge, and different intentions. Thus, information found on the Semantic Web has to be seen as claims rather than as facts. Before using these claims, the information consumer has to evaluate their trustworthiness and determine the subset which he wants to trust for his specific task. In everyday life, we use a wide range of trust assessment policies for evaluating the trustworthiness of information: We might trust Andy on restaurants but not on computers, trust professors on their research field, believe foreign news only when it is reported by several independent sources and buy only from sellers on eBay who have more than 100 positive ratings. Which policy is chosen depends on the specific task, our subjective preferences, our past experiences and the trust relevant information available. For tasks which are economically relevant to the information consumer he might require a very strict trust policy, involving for example recommendations by people he knows. For other tasks, a looser policy like ‘Accept all information that has been asserted by at least two independent information providers, no matter who they are.’ might be acceptable. The future Semantic Web is supposed to be a dense mesh of interrelated information, similar to the information perception situation we face in the offline world. Thus, we argue, a trust policy framework for the Semantic Web can and should support a similarly wide range of trust policies as used offline [4]. Every trust policy employs one or more trust assessment methods. These methods can be classified into three categories: Rating-Based Methods include rating systems like the one used by eBay and Web-Of-Trust mechanisms. Most trust architectures proposed for the Semantic Web so far fall into this category [1][11]. The general problem with these approaches is that they require explicit and topic-specific trust ratings. For many application domains, providing such ratings and keeping them upto-date puts an unrealistically heavy burden on information consumers. Context-Based Methods use meta-data about the circumstances in which information has been claimed, e.g. who said what, when and why. They include role-based trust methods, using the author’s role or his membership in a specific group, for trust decisions. Example policies from this category are: ‘Prefer product descriptions published by the manufacturer over descriptions published by a vendor’ or ‘Distrust everything a vendor says about its competitor.’ Context-based trust mechanisms do not require explicit ratings, but rely on the availability of background information. Within many Semantic Web application areas, such background information might be available. Content-Based Methods do not use meta-data about information, but rules and axioms together with the information content itself and related information about the same topic published by other authors. Example policies following this approach are: ‘Believe information which has been stated by at least 5 independent sources.’ or ‘Distrust product prices that are more than 50% below the average price.’ 2 The TriQL.P Browser The TriQL.P browser is a general purpose RDF browser which shows how Semantic Web content can be filtered using a wide range of trust policies, combining methods from all three categories described above. The TriQL.P browser is based on the Piggy Bank extension for the Firefox browser [10]. Piggy Bank extracts Semantic Web content from Web pages as users browse the Web. On websites where Semantic Web content is not available, Piggy Bank can invoke screen-scrapers to re-structure content into Semantic Web format. The extracted information can be browsed, sorted and searched using a comfortable user-interface, and saved into a local repository for future reference and aggregation. In addition to the functionality provided by Piggy Bank, the TriQL.P browser gives users the ability to: – collect provenance meta-data together with information from the Web; – import information aggregated from multiple sources by a third party into the local repository using the RDF/XML, TriX [7] and TriG [3] syntaxes; – load trust policy suites containing sets of policies; – filter information in the local repository using these policies; – explain on demand why displayed information fulfils a selected policy. Figure 1 shows the user interface of the TriQL.P browser. Information items from the local repository are displayed on the left-hand side. The policy selection box on the right side allows users to select a policy from the policy suite currently loaded. After selecting a policy, the left-hand view updates to show only information matching this policy. There is a ‘Oh, yeah?’-button [2] next to each piece of information. Pressing this buttons opens a new window with an explanation why the piece of information fulfils the selected trust policy. Fig. 1. The TriQL.P user interface. The user selects a trust policy from the right-hand box. The left-hand view updates to show only matching information. The ‘Oh, yeah?’ buttons open new windows with explanations why a piece of information fulfils the