TL;DR
ANGIE tackles the incompleteness of large knowledge bases by dynamically integrating data from external Web services on-the-fly. This allows users to query and explore a unified knowledge graph, seamlessly blending local data with real-time information from the web without explicit data materialization.
What Problem Does This Paper Address?
Modern knowledge bases, like DBpedia or YAGO, are vast repositories of facts extracted from the web. However, they are inherently incomplete. Users often encounter gaps when querying for specific information, such as the albums of a particular artist or the current ratings of a movie. While crawling the web for missing data is often infeasible due to site restrictions and dynamic content, a growing number of Web services offer access to specialized, structured data. The challenge is to leverage these services to enrich existing knowledge bases in a transparent and efficient manner, making the entire knowledge base appear complete to the user.
Methods
ANGIE introduces an "active knowledge base" concept, where Web services act as dynamic components. The core of the system is a query processing engine that translates user queries into a sequence of function compositions. These compositions involve calls to both the local knowledge base (modeled as an artificial fab function) and external Web services.
Key techniques include:
- RDF Graph Representation: Both the knowledge base and function definitions are represented as RDF graphs.
- Function Modeling: Web services are modeled as functions with pre-conditions (inputs) and post-conditions (outputs), defined as RDF statements with variables.
- Query Translation: A recursive algorithm, similar to Prolog's execution, translates queries into function compositions. This involves matching query edges to function pre-conditions and post-conditions.
- Seamless Integration: The system aims to integrate retrieved data transparently, so users perceive all answers as originating from the local knowledge base.
- Mapping Tool: This component handles the actual invocation of Web services. It translates function pre-conditions into service parameters (using WSDL or REST interfaces) and maps the XML results back into RDF triples for the knowledge base. XSLT is used for this lifting process.
Datasets
Not discussed in the paper. The paper mentions using YAGO as the local knowledge base and integrating with Web services like MusicBrainz, LastFM, LibraryThing, ISBNdb, and IVA.
Key Insights and Ideas
- Web Services as Dynamic Knowledge Sources: The paper frames Web services not just as data providers but as active, dynamic components of a knowledge base, enabling "on-demand" knowledge retrieval.
- Function Composition for Complex Queries: The ability to chain multiple Web service calls, and combine them with local data, is crucial for answering complex queries that span different information sources.
- Transparency is Key: The system's success hinges on making the dynamic retrieval process invisible to the user, maintaining the illusion of a single, comprehensive knowledge base.
- RDF as a Unifying Model: Using RDF for both the knowledge base and function definitions provides a consistent framework for integration and querying.
Key Results
The paper presents the ANGIE system and its architecture, demonstrating its capability to answer queries by combining local knowledge with data from Web services. A demo scenario illustrates how a user can explore Frank Sinatra's information, with albums being retrieved on-demand from MusicBrainz. The system supports interactive exploration and visualization of the active knowledge base. Specific quantitative results are not detailed in this paper, as it focuses on the system's architecture and approach.
Limitations and Future Work
- Function Definition Acquisition: Currently, function definitions for Web services need to be manually identified and cast into RDF graphs. Future work could involve automatically acquiring these definitions from Web service brokers or semi-automatically generating them.
- Cost Estimation: While the system aims to prioritize functions based on cost, a robust model for estimating instantiation costs is still under development.
- Scalability and Performance: The paper mentions setting a limit on the maximum number of function calls to manage performance, implying that handling very large or complex compositions could be a challenge.
- Data Cleaning: Integrating data from diverse Web services poses a data cleaning challenge, which is acknowledged but not deeply detailed in terms of specific solutions.
Summary updated