Peter Z. Yeh

Ontology alignment for linked open data

The Web of Data currently coming into existence through the Linked Open Data (LOD) effort is a major milestone in realizing the Semantic Web vision. However, the development of applications based on LOD faces difficulties due to the fact that the different LOD datasets are rather loosely connected pieces of information. In particular, links between LOD datasets are almost exclusively on the level of instances, and schema-level information is being ignored. In this paper, we therefore present a system for finding schema-level links between LOD datasets in the sense of ontology alignment. Our system, called BLOOMS, is based on the idea of bootstrapping information already present on the LOD cloud. We also present a comprehensive evaluation which shows that BLOOMS outperforms state-of-the-art ontology alignment systems on LOD datasets. At the same time, BLOOMS is also competitive compared with these other systems on the Ontology Evaluation Alignment Initiative Benchmark datasets.

Contextual ontology alignment of LOD with an upper ontology: a case study with proton

The Linked Open Data (LOD) is a major milestone towards realizing the Semantic Web vision, and can enable applications such as robust Question Answering (QA) systems that can answer queries requiring multiple, disparate information sources. However, realizing these applications requires relationships at both the schema and instance level, but currently the LOD only provides relationships for the latter. To address this limitation, we present a solution for automatically finding schema-level links between two LOD ontologies - in the sense of ontology alignment. Our solution, called BLOOMS+, extends our previous solution (i.e. BLOOMS) in two significant ways. BLOOMS+ 1) uses a more sophisticated metric to determine which classes between two ontologies to align, and 2) considers contextual information to further support (or reject) an alignment. We present a comprehensive evaluation of our solution using schema-level mappings from LOD ontologies to Proton (an upper level ontology) - created manually by human experts for a real world application called FactForge. We show that our solution performed well on this task. We also show that our solution significantly outperformed existing ontology alignment solutions (including our previously published work on BLOOMS) on this same task.

Alignment-based Querying of Linked Open Data

The Linked Open Data (LOD) cloud is rapidly becoming the largest interconnected source of structured data on diverse domains.The potential of the LOD cloud is enormous, ranging from solving challenging AI issues such as open domain question answering to automated knowledge discovery. However, due to an inherent distributed nature of LOD and a growing number of ontologies and vocabularies used in LOD datasets, querying over multiple datasets and retrieving LOD data remains a challenging task. In this paper, we propose a novel approach to querying linked data by using alignments for processing queries whose constituent data come from heterogeneous sources. We also report on our Alignment based Linked Open Data Querying System (ALOQUS) and present the architecture and associated methods. Using the state of the art alignment system BLOOMS, ALOQUS automatically maps concepts in users’ SPARQL queries, written in terms of a conceptual upper ontology or domain specific ontology, to different LOD concepts and datasets. It then creates a query plan, sends sub-queries to the different endpoints, crawls for co-referent URIs, merges the results and presents them to the user. We also compare the existing querying systems and demonstrate the added capabilities that the alignment based approach can provide for querying the Linked data.

Capturing and answering questions posed to a knowledge-based system

As part of the ongoing project, Project Halo, our goal is to build a system capable of answering questions posed by novice users to a formal knowledge base. In our current context, the knowledge base covers selected topics in physics, chemistry, and biology, and our question set consists of AP (advanced high-school) level examination questions. The task is challenging because the questions are linguistically complex and are often incomplete (assume unstated knowledge), and because the users do not have prior knowledge of the systems contents. Our solution involves two parts: a controlled language interface, in which users reformulate the original natural language questions in a simplified version of English, and a novel problem solver that can elaborate initially inadequate logical interpretations of a question by selecting relevant pieces of knowledge in the knowledge base. An evaluation of the work in 2006 showed that this approach is feasible and that complex, multisentence questions can be posed and answered, thus illustrating novel ways of dealing with the knowledge capture impedance between users and a formal knowledge base, while also revealing challenges that still remain.

Semantic Interpretation of the Web without the Semantic Web: Toward Business-Aware Web Processors

In this paper, we describe a family of systems that we are developing to make systematic use of the Web in support of decision-makers. These systems, which we call business-aware web processors, use semantic representations but do not assume that these representations will come from the Semantic Web. We describe the relationship between our vision and the standard Semantic Web vision. We next present two prototype applications based on the approach that we have been developing to support decision-makers in two different contexts. We then describe the underlying technology platform we have been developing to support this family of applications, and the challenges we are confronting as we advance this research agenda.

Moving beyond SameAs with PLATO: partonomy detection for linked data

The Linked Open Data (LOD) Cloud has gained significant traction over the past few years. With over 275 interlinked datasets across diverse domains such as life science, geography, politics, and more, the LOD Cloud has the potential to support a variety of applications ranging from open domain question answering to drug discovery. Despite its significant size (approx. 30 billion triples), the data is relatively sparely interlinked (approx. 400 million links). A semantically richer LOD Cloud is needed to fully realize its potential. Data in the LOD Cloud are currently interlinked mainly via the owl:sameAs property, which is inadequate for many applications. Additional properties capturing relations based on causality or partonomy are needed to enable the answering of complex questions and to support applications. In this paper, we present a solution to enrich the LOD Cloud by automatically detecting partonomic relationships, which are well-established, fundamental properties grounded in linguistics and philosophy. We empirically evaluate our solution across several domains, and show that our approach performs well on detecting partonomic properties between LOD Cloud data.

An Efficient and Robust Approach for Discovering Data Quality Rules

Poor quality data is a growing problem that affects many enterprises across all aspects of their business ranging from operational efficiency to revenue protection. Moreover, this problem is costly to fix because significant effort and resources are required to identify a comprehensive set of rules that can detect (and correct) data defects along various data quality dimensions such as consistency, conformity, and more. Hence, many organizations employ only basic data quality rules that check for null values, format, etc. in efforts such as data profiling and data cleansing; and ignore rules that are needed to detect deeper problems such as inconsistent values across interdependent attributes. This oversight can lead to numerous problems such as inaccurate reporting of key metrics used to inform critical decisions or derive business insights. In this paper, we present an approach that efficiently and robustly discovers data quality rules -- in particular conditional functional dependencies -- for detecting inconsistencies in data and hence improves data quality along the critical dimension of consistency. We evaluate our approach empirically on several real-world data sets. We show that our approach performs well on these data sets for metrics such as precision and recall. We also compare our approach to an established solution and show that our approach outperforms this solution for the same metrics. Finally, we show that our approach scales efficiently with the number of records, the number of attributes, and the domain size.

SPARQL Query Re-writing Using Partonomy Based Transformation Rules

Often the information present in a spatial knowledge base is represented at a different level of granularity and abstraction than the query constraints. For querying ontologys containing spatial information, the precise relationships between spatial entities has to be specified in the basic graph pattern of SPARQL query which can result in long and complex queries. We present a novel approach to help users intuitively write SPARQL queries to query spatial data, rather than relying on knowledge of the ontology structure. Our framework re-writes queries, using transformation rules to exploit part-whole relations between geographical entities to address the mismatches between query constraints and knowledge base. Our experiments were performed on completely third party datasets and queries. Evaluations were performed on Geonames dataset using questions from National Geographic Bee serialized into SPARQL and British Administrative Geography Ontology using questions from a popular trivia website. These experiments demonstrate high precision in retrieval of results and ease in writing queries.

Exploring Two Enterprise Semantic Integration Systems

An Enterprise Semantic Integration System (ESIS) provides cross-domain and cross-department insights by normalizing and merging structured, semistructured, and unstructured data sources from both internal and external source, into a knowledge model that can then be visualized in an integrated portal. The interface of an ESIS shows data, metadata, and semantic connections. Search and analysis tool and integration points to other enterprise systems can be made available to the knowledge worker in the ESIS. In this paper, we first present an overview of the complex task of data aggregation, cleansing, and import into a knowledge model. We go on to explore two major types of ESIS: a market or competitive intelligence system, with illustrative examples Corporate Radar and Business Event Advisor, and a knowledge management system, the Knowledge Discovery Capability. Each of these case studies provides interesting insights into the challenges involved in the design and deployment of an ESIS.

Capturing the Semantics of Online News Sources for Business Intelligence Applications

In this paper, we present a knowledge based approach to capture rich semantic representations from online news sources for business intelligence (BI) applications that know the representations of interest in advance. Our approach performs this task by generating phrases from these representations and matching these phrases against the news using a set of syntactic and semantic transformations. The representation that best matches a piece of news is selected as its meaning. We present two evaluations showing how our approach performs well on capturing the semantics of online new sources for BI applications.