- We present demos of an ongoing project, domain informational vocabulary extraction (DIVE), which aims to enrich digital publications through entity and key informational words detection and by adding additional annotations. The system implements multiple strategies for biological entity detection, including using regular expression rules, ontologies, and a keyword dictionary. These extracted entities are then stored in a database and made accessible through an interactive web application for curation and evaluation by authors. Through the web interface, the user can make additional annotations and corrections to the current results. The updates can then be used to improve the entity detection in subsequent processed articles. Although the system is being developed in the context of annotating journal articles, it can also be beneficial to domain curators and researchers at large.
Keywords—component; Information systems applications; Information integration; Ontology; Text Mining
Due to its technical depth and rich informational content, a journal article often requires that readers, domain experts, and curators invest significant amounts of time and effort to fully comprehend and make intelligent use of its content. This can be especially true in emerging areas, where novel ideas and new terminologies may be presented without precedent. As new technologies accelerating scientific discovery and more content becomes available online, the number of new articles that must be read and understood continues to rise. Therefore, there is a pressing need to develop computational methods and tools that can enrich the information content of digital publications, improve its accessibility and utility, and facilitate the readers’ understanding by creating links between journal articles and relevant database entities during the article production process. To address this issue, we present software developments from an ongoing project, DIVE, which features auto extraction of informational vocabulary, web based access and curation tools, and integration into the digital publication process.
The framework implements several strategies in entity extraction, including using regular expression rules, ontology and a keyword dictionary. The results of the extracted biological entities are then stored in a database and made accessible through an interactive web application for curation and evaluation by authors and other domain experts. Through the web interface, a user can make additional annotations and corrections to the initial result set. The updates are stored and managed via the relational database for future improvements.
We present application demos to illustrate this framework using a sets of plant biology articles. We detail the design and implementation of the system, including entity detection, the extraction pipeline, and the web interface; we also present a use case demonstration. We would like to engage publishers and biology data curators in discussion and feedback.
There are three major steps in processing the documents:
text extraction, entity candidate extraction, and candidate
assessment. The input for the text extraction process is the
structured document tagged by Journal Article Tag Suite
(JATS) [
The web front end in our prototype is implemented using Django (v 1.8). Based on Python, the web front is easily programmable, extensible, and is pluggable with multiple popular databases. It forms the presentation layer of this system, relying on the back end code to run the entity extraction algorithms from the manuscript and to transfer the results in a JSON format.
The system can benefit the entire life cycle of the digital
publication, from initial manuscript submission to publishing
the article and presenting information to readers. At the initial
manuscript submission stage, the manuscript can be processed
to extract known key informational vocabulary, such as
biological entities, as well as to identify potential new technical
words. That information may be used by editors to identify
appropriate reviewers for the manuscript. After the article has
been accepted for publication, additional information about the
key informational words, such as links to external repositories
or reference sites, may also be embedded during the
prepublication production process to enrich the information
content and accessibility. Publication curators may also
leverage the information for curation. New information defined
and verified by experts may also be injected to other
information resources, such as Planteome [
Let us use an example to illustrate the features of our web interface, thereby displaying the various views, layouts, and functions available. Our prototype includes 609 manuscripts from the journal Plant Physiology.
The publication list view (Figure 1) is a paginated list of all the articles with an external DOI reference and the article title. The entity extraction phase can also generate phrases, while matching the ontology term alphabtetically, are not used for the purpose implied by the term. Those situations requires expert knolwedge and input.
Theorefore, each row also includes user control buttons for
editing the record. Figure 3 shows an example of the Entitiy
record editing interface. In this view, there are editable fields of
this record where a user may correct or enter new values. A
dynamic search box can be used to search and add new species
into the species menu, if the appropriate species was not
detected or inferred from the article. This search box uses an
online service from NCBI to provide a very comprehensive list
of options as the user dynamically types into it. Sentences of
occurrence of this entity are extracted from the manuscript with
the entity name highlighted in yellow. This again provides
better, almost complete context information for this entity, as
per the manuscript text.
Figure 2 shows example of exploring entities extracted
from a full article (i.e. 1002.xml in Figure 1). In the top of the
page, the title and abstract of the article [
It is important to note that some entities can be matched to multiple onotology terms. Such cases are currently resolved based on the general priorites we assinged to each extraction rule during the extraction phase. However, for a particular article, the result may not always be the most approrirate one. The prototype is still under development, and we welcome feedback from domain researchers and publishing professionals for future developments and improvements.
DIVE is partially supported by CyVerse (NSF award DBI0735191 and DBI-1265383) and the Gramene, a Comparative Plant Genomics Database (NSF award IOS-1127112).