Cartic Ramakrishnan

Semantics for the Semantic Web: The Implicit, the Formal and the Powerful

Enabling applications that exploit heterogeneous data in the Semantic Web will require us to harness a broad variety of semantics. Considering the role of semantics in a number of research areas in computer science, we organize semantics in three forms — implicit, formal, and powerful — and explore their roles in enabling some of the key capabilities related to the Semantic Web. The central message of this article is that building the Semantic Web purely on description logics will artificially limit its potential, and that we will need to both exploit well-known techniques that support implicit semantics, and develop more powerful semantic techniques.

Ranking complex relationships on the semantic Web

Industry and academia are both focusing their attention on information retrieval over semantic metadata extracted from the Web, and it is increasingly possible to analyze such metadata to discover interesting relationships. However, just as document ranking is a critical component in todays search engines, the ranking of complex relationships would be an important component in tomorrows semantic Web engines. This article presents a flexible ranking approach to identify interesting and relevant relationships in the semantic Web. The authors demonstrate the schemes effectiveness through an empirical evaluation over a real-world data set.

Geospatial Ontology Development and Semantic Analytics

Geospatial ontology development and semantic knowledge discovery addresses the need for modeling, analyzing and visualizing multimodal information, and is unique in offering integrated analytics that encompasses spatial, temporal and thematic dimensions of information and knowledge. The comprehensive ability to provide integrated analysis from multiple forms of information and use of explicit knowledge make this approach unique. This also involves specification of spatiotemporal thematic ontologies and populating such ontologies with high quality knowledge. Such ontologies form the basis for defining the meaning of important relations terms, such as near or surrounded by, and enable computation of spatiotemporal thematic proximity measures we define. SWETO (Semantic Web Technology Evaluation Ontology) and geospatial extension SWETO-GS are examples of these ontologies. The Geospatial Semantics Analytics (GSA) framework incorporates: (1) the ability to automatically and semi-automatically tract metadata from syntactically (including unstructured, semi-structured and structured data) and semantically heterogeneous and multimodal data from diverse sources; and (2) analytical processing that exploits these ontologies and associated knowledge bases, with integral support for what we term spatiotemporal thematic proximity (STTP) reasoning and interactive visualization capabilities. This paper discusses the results of our geospatial ontology development efforts as well as some new semantic analytics methods on this ontology such as STTP.

TaxaMiner: an experimentation framework for automated taxonomy bootstrapping

Construction of domain ontologies on the semantic web is a human and resource intensive process, efforts to reduce which are crucial for the Semantic Web to scale. We present a framework for automated taxonomy construction, that involves: (a) generation of a cluster hierarchy from a document corpus using statistical clustering and NLP techniques; (b) extraction of a topic hierarchy from this cluster hierarchy; and (c) assignment of labels to nodes in the topic hierarchy. Metrics for estimating topic hierarchy quality and parameters of an experimentation framework are identified. MEDLINE was the document corpus and MeSH thesaurus was the gold standard.

Unsupervised Discovery of Compound Entities for Relationship Extraction

In this paper we investigate unsupervised population of a biomedical ontology via information extraction from biomedical literature. Relationships in text seldom connect simple entities. We therefore focus on identifying compound entities rather than mentions of simple entities. We present a method based on rules over grammatical dependency structures for unsupervised segmentation of sentences into compound entities and relationships. We complement the rule-based approach with a statistical component that prunes structures with low information content, thereby reducing false positives in the prediction of compound entities, their constituents and relationships. The extraction is manually evaluated with respect to the UMLS Semantic Network by analyzing the conformance of the extracted triples with the corresponding UMLS relationship type definitions.

Relationship Web: Blazing Semantic Trails between Web Resources

Using keywords as inputs to search engines and receiving documents as responses remains the prevalent way to access information on the Web. Although a shift toward entity awareness is a fairly recent trend in information access, such methods remain devoid of semantics, which are increasingly recognized as the lynchpin of search, integration, and analysis. We argue that relationships are at the heart of semantics, and, as such, we envision a Web of relationships to relate content across Web resources. Under this powerful new paradigm, information access over the Web would switch from a mere document-retrieval operation to an information framework that supports insight elicitation and semantic analytics over Web resources. In this column, we outline our vision and discuss how recent Improvements in content extraction and semantic annotation will ultimately help us realize this relationship Web.

Joint Extraction of Compound Entities and Relationships from Biomedical Literature

In this paper we identify some limitations of contemporary information extraction mechanisms in the context of biomedical literature. We present an extraction mechanism that generates structured representations of textual content. Our extraction mechanism achieves this by extracting compound entities, and relationships between them, occuring in text. A detailed evaluation of the relationship and compound entities extracted is presented. Our results show over 62% average precision across 8 relationship types tested with over 82% average precision for compound entity identification.