Help will be provided for this task: Ontology-Based Annotator Web Service

Clément Jonquet, Mark A. Musen, Nigam H. Shah2012

TL;DR

This paper introduces an Ontology-Based Annotator (OBA) web service designed to automatically annotate text with concepts from ontologies, a crucial step for the semantic web. The service addresses the lack of user-friendly, integrated annotation tools by offering a flexible, service-oriented architecture that leverages a vast collection of biomedical ontologies.

What Problem Does This Paper Address?

The semantic web vision relies on content being semantically described using ontologies. However, manually annotating web content is a significant challenge, as highlighted by the recurring "help will be provided" note in calls for papers for semantic web conferences. While ontologies are increasingly common, especially in domains like biomedicine, there's a scarcity of convenient, easily integrable tools for users to perform this annotation. This gap means much of the web's content remains unstructured and inaccessible to automated agents. The paper specifically targets the biomedical domain, where the sheer volume of data necessitates efficient annotation for effective data integration and knowledge discovery.

Methods

The OBA web service employs a two-step workflow:

  1. Concept Recognition: This initial step involves a concept recognizer (specifically, mgrep in their biomedical implementation) that identifies potential concepts within a given text. This is achieved through string matching and syntactic techniques like stemming and spelling correction, using a dictionary built from a comprehensive set of ontology terms (names and synonyms).
  2. Semantic Expansion: The direct annotations from the first step are then enhanced using semantic relationships within ontologies. The paper describes several types of expansion components that can be plugged into the service:
  3. is_a Transitive Closure: This component adds parent concepts to annotations, leveraging hierarchical relationships (e.g., if "melanoma" is annotated, "neoplasm" is also added if "melanoma is_a neoplasm").
  4. Semantic Distance: This component finds related concepts within a specified semantic distance in an ontology.
  5. Ontology Mapping: This component utilizes existing mappings between different ontologies to create new annotations.

The service is designed to be flexible, allowing users to select and parameterize these expansion components. The output is semantically rich, returned as an OWL ontology populated with annotations as instances.

Datasets

The paper does not explicitly mention specific training datasets for the concept recognizer. However, for evaluation and implementation, they utilized:

  • PubMed Citations: A dataset of 1,050,000 PubMed citations (including title, abstract, and metadata) was used to evaluate the annotator's utility.
  • Biomedical Ontologies: The service leverages a large collection of publicly available terminologies and ontologies, including those from the Unified Medical Language System (UMLS) and the National Center for Biomedical Ontology (NCBO) repository. Their biomedical prototype specifically uses 47 terminologies and ontologies.

Key Insights and Ideas

  • Service-Oriented Architecture for Annotation: Moving annotation from a manual, often tedious task to an automated web service is crucial for widespread adoption and integration into existing workflows.
  • Leveraging Biomedical Ontologies: The sheer volume and complexity of biomedical data make ontologies indispensable for semantic annotation. Accessing and integrating a broad range of these ontologies is key to comprehensive annotation.
  • Semantic Expansion is More Than Just Matching: Simply identifying concept names isn't enough. The ability to infer related concepts through semantic relationships (like is_a hierarchies) significantly enriches the annotations and their utility.
  • OWL as a Semantic Output Format: Returning annotations in OWL provides a structured, semantically rich representation that can be easily processed by other semantic web tools and agents.
  • Flexibility in Expansion: Allowing users to choose and combine semantic expansion components makes the service adaptable to different annotation needs and levels of detail.

Key Results

  • High Accuracy: The mgrep concept recognizer used in their biomedical implementation reported an accuracy of over 95% in recognizing disease names.
  • Utility Demonstrated: When annotating 1,050,000 PubMed citations, the service generated 174,840,027 annotations (18% direct, 82% ISA_CLOSURE). An average of 160 annotating concepts per citation was observed, with approximately 99% of the test set being annotated (with at least one concept).
  • Service Deployment: The OBA web service is deployed as both a SOAP and a RESTful web service, with a user-friendly web interface available.
  • Biomedical Indexing: The service was used to create an ontology-based index of open biomedical resources, demonstrating its practical application in enhancing search capabilities.

Limitations and Future Work

Limitations:

  • Concept Recognition Flexibility: While mgrep is effective, the paper notes that the choice of concept recognizer is not fixed, implying that current options might have limitations or that better ones could be integrated.
  • Dictionary Management: The current dictionary is built from a fixed set of ontologies. Users cannot currently customize the dictionary on demand.
  • Annotation Storage: The OBA service itself does not store annotations; users are responsible for managing this information.
  • Structured Document Processing: The current prototype primarily handles free text. Processing structured documents directly is a future goal.
  • Semantic Description of Service Functionality: While the service returns semantically described results (OWL), the service itself does not yet have a semantic description of its functionality using frameworks like OWL-S or WSMO.

Future Work:

  • Improved Concept Recognition: Exploring recognizers that can identify not just concepts but also relations between them.
  • On-Demand Ontology Selection: Allowing users to choose any ontology for annotation and providing "topic-oriented" dictionaries to simplify usage.
  • New Semantic Expansion Components: Developing and integrating more semantic expansion components, such as semantic distance and concept similarity based on data-driven ontology alignment.
  • Semantic Web Service Integration: Further developing the semantic description of the service's functionality to align with Semantic Web Service standards (OWL-S, WSMO).
  • Testbed for NCBO: The OBA service will serve as a testbed for evaluating semantic web services within the NCBO framework, potentially adding a semantic layer to existing NCBO services.

Summary updated

Abstract

Semantic annotation is part of the vision for the semantic web. Ontologies are required for this task, and although they are in common use, there is a lack of annotation tools for users that are convenient, simple to use and easily integrated into their processes. This paper presents an ontology-based annotator web service methodology that can annotate a piece of text with ontology concepts and return annotations in OWL. Currently, the annotation workflow is based on syntactic concept recognition (using concept names and synonyms) and on a set of semantic expansion algorithms that leverage the semantics in ontologies (e.g., is_a relations). The paper also describes an implementation of this service for life sciences and biomedicine. Our biomedical annotator service uses one of the largest available set of publicly available terminologies and ontologies. We used it to create an index of open biomedical resources. Both the deployed web service and a user interface can be accessed at
BibTeX
@misc{jonquet2012help,
  author = {Clément Jonquet and Mark A. Musen and Nigam H. Shah},
  title = {{Help will be provided for this task: Ontology-Based Annotator Web Service}},
  year = {2012},
}
Figure