Due to recent advances in Semantic Web (SW), the amount of Linked Data
(LD) available particularly in Resource Description Framework (RDF) increases
rapidly (http://lod-cloud.net). However, LD is still used mostly by SW experts.
There are two main obstacles to making LD accessible for common Web users:
(1) the need to learn the query language, SPARQL, and (2) the need to know
the schemas underlying various datasets to be queried. Approaches to ease the
access to LD include graphical query interfaces [
Typically, a natural language question-answering (NLQA) system takes natural language queries as input. The queries are then converted in a structured query language, e.g. SPARQL, which will be used to consult a knowledge base (KB), e.g., a SPARQL endpoint, or KBs. While there are a number of relevant previous works, it is widely understood that the development of a NLQA system requires expertise in various technologies, e.g., natural language processing (NLP), database schema analysis, inference, and so on. There is thus a natural call for collaboration among interested parties.
With the goal to provide a platform of open collaboration for development of NLQA systems, the OKBQA framework has been developed. Recently, it has reached a milestone: (1) core module categories for NLQA systems are gured out and their APIs are documented, (2) a repository of OKBQA-compatible modules is implemented, and 24 modules are registered, and (3) a prototype demo system is implemented and two work ows for QA in English and Korean have been set up. This manuscript presents a summary of OKBQA Framework, and the demo presentation will show how the system works to support collaboration for development of NLQA systems. 2
An OKBQA repository is implemented and maintained to provide a venue for sharing information about modules developed for the OKBQA framework (http://repository.okbqa.org). The registration of modules is open to anybody. At the time of writing, there are 24 modules registered to the repository. 4
A prototype demo system is developed and maintained as a public service (http://ws.okbqa.org/wui-2016/), (1) to demonstrate how work ows in OKBQA actually work, and (2) to support development of modules for the framework. Currently, two work ows have been set-up for QA in English and Korean. Users can choose a work ow and try it with natural language queries.
Note that the performance of currently available work ows may not yet be competitive, because they are composed by connecting modules developed by di erent groups without much tuning for harmonization. To improve the performance, further development of the modules is required, and the interface of the OKBQA demo system is designed to support it.
Firstly, the interface allows users to modify the work ows. Note that a modules for the OKBQA framework is required to be a REST service, and a work ow is de ned as a sequence of URIs (of the REST services). This means that anyone can develop a module to replace one in a prede ned work ow. Suppose one has developed a new DM, either by improving an existing one or by newly implementing it. Once it is deployed as a REST service, the new DM can be tested in a work ow by simply specifying (the URL of) the DM as the DM component of the work ow.
Secondly, the interface allows users to inspect the input and output of each module during execution of a work ow. Figure 2 shows a screen-shot of the interface, which shows the execution of the English work ow with the example query Which river ows in Seoul?. The left pane shows the progress of the work ow, and the right pane shows the input and output of each module. With the design, the OKBQA framework may be thought as an SDK system for development of NLQA systems
The OKBQA framework is developed as a platform of open collaboration for the development of NLQA systems. The modules developed for the framework can be found at the OKBQA repository, and they are all freely available as opensource projects. The prototype demo system is maintained as a public service to support distributed, voluntary development of modules for the framework.
There is a large room for improvement in the framework. For example, the composition of a work ow is not yet su ciently exible, and the performance of current reference work ows is not yet competitive. Nevertheless, we believe it is a signi cant milestone that such a framework has begun to work to organize distributed contributions. We hope this presentation to be an opportunity to receive feedback from interested parties and also to invite potential collaborators.
The development of OKBQA Framework is supported by the ExoBrain project (http://exobrain.kr/). JDK is supported by the Life Science Database Integration Project funded by National Bioscience Database Center (NBDC) of Japan Science and Technology Agency (JST). ACNN and RU are supported by the H2020 project HOBBIT (GA no. 688227) and the EuroStars projects DIESEL (01QE1512C) and QAMEL (01QE1549C). The authors thank to all the participants in OKBQA hackathon so far for their contribution.