This paper presents FCAWarehouse, a prototype online data repository for FCA. The paper explains the motivation behind the development of FCAWarehouse and the features available, such as the ability to donate datasets and their respective formal contexts, the ability to generate arti cial formal contexts on-the- y, and how these features are also available through a set of web-services. The paper concludes by suggesting future work in order to enhance it's usability.
In recent years, di erent means of archiving empirical data have been considered and implemented to build a space where data collection could be exploited and analyzed accordingly, if required. The di erent archiving techniques not only introduced communities an alternative approach of maintaining the quality and accessibility of data, but also improved the indexing of available data collections with proper categorization. An example of such an archiving technique are online data repositories.
Online data repositories are digital libraries which allow the comprehensive
collection, management and preservation of digital content and the ability to
o er it to targeted user communities [
The idea of an FCA online data repository, to provide FCA practitioners with
a resource of public datasets in FCA formats, was proposed in [
The \Data-to-FCA" converter of the proposed repository was the most
complex component to implement with regards to the e ort and time needed to
develop it. However, since the publication of the aforementioned paper, the
\Datato-FCA" converter has been developed as a standalone desktop application called
FcaBedrock2, a formal context creator for FCA with the ability of converting
datasets in various formats to formal contexts in the Burmeister (.cxt) or FIMI
(.dat) formats [
FCAWarehouse4 is a prototype data repository for FCA developed as part of a
BSc Computing nal year project [
All of the les and metadata are provided by the donator during the donation process. In cases where the formal contexts are created using FcaBedrock, a
5 Representational State Transfer (ReST). http://www.ibm.com/developerworks/ webservices/library/ws-restful/ Bedrock le (.bed) is also provided which contains the metadata of the conversion so that users can load the original data and the Bedrock le in FcaBedrock and have a look at, or modify, the parameters and conversion criteria set for each attribute in the original dataset (Figure 4).
The les donated for each dataset are stored physically on the server, with the database only holding the URL to each dataset. As the prototype is currently hosted on a server with limited storage capabilities, all les uploaded for a dataset during the donation process are automatically compressed as .zip les.
After a user has donated a dataset, the librarians are noti ed about the new submission and have the ability of accepting or declining the submitted dataset. Consequently, only datasets marked as `Accepted' by the librarians are visible to end-users. FCAWarehouse provides the ability of generating arti cial formal contexts in the Burmeister (.cxt) format by prede ning the number of formal objects, number of formal attributes and the density. Developers of tools such as the formal concept miners In-Close6 and FCbO7 can use this feature to generate formal contexts of size of their choice in order to test the limits and e ciency of their algorithms.
The generated arti cial formal contexts are randomly generated by prede ning the number of objects, number of attributes and density of the formal context. The density is de ned as a percentage indicating the minimum amount of crosses (formal attributes) that each row (formal object) should contain in the formal context. Setting for example number of objects to 10, number of attributes to 10 and density percentage to 70% will result in a formal context with 10 formal objects and 10 formal attributes where at least 70% of each row is consisted of crosses (Figure 5). Setting the density to 0 will generate a random arti cial formal context with no density criteria.
FCAWarehouse implements a set of web services for the interoperability of FCAWarehouse with third-party FCA applications. At the moment there are a total of four web services, namely: { GenerateContext: Accepts as input the number of objects, number of attributes, density percentage and outputs a corresponding arti cial Burmeister formal context in XML. { GetAllDataSets: Returns all datasets, along with their metadata, in XML.
Only datasets marked as `Accepted' by the librarians are returned. { Get10MostRecentDatasets: Returns the 10 most recent datasets along with their metadata, in XML. Only datasets marked as `Accepted' by the librarians are returned. { SearchDataset: Accepts a string as input and returns any datasets which contain the given string in their name, along with their metadata, in XML.
Only datasets marked as `Accepted' by the librarians are returned. 3
While useful in it's current state, FCAWarehouse is still in a prototypical state; a number of improvements can be implemented to enhance it's usability. For example, the feature of arti cially generating formal contexts can be extended to simulate formal contexts that real datasets would produce. This can be achieved by de ning sets of adjacent columns in the formal context to represent a single attribute. Assuming, for example, a categorical attribute (with each of it's values being a formal attribute in the formal context) and each formal object only being able to have one of it's values (quite common in real datasets), will result in a formal context where no more than one cross will exist in the columns representing that attribute. The same logic can be applied to various type of attribute: boolean attributes could be interpreted as single formal attributes in the formal context and continuous attributes could be grouped using ranges rather than one formal attribute for each numerical value. In this way, FCAWarehouse can act as a benchmarker for the comparison of tools and algorithms by providing citable random data as well as converted real datasets.
A feature which could prove quite useful in the future, with some modications and adjustments, are FCAWarehouse's web services. Interesting usecases can emerge from this feature; for example, the formal context creator FcaBedrock could feature a \Create formal context from FCAWarehouse" option. By using the provided web services, a list of the available datasets could appear in FcaBedrock. Selecting one of the datasets could automatically download the dataset and start auto detecting its values to create a formal context. Creating a formal context in FcaBedrock requires inputting a dataset, de ning its metadata and then creating the formal context; considering, however, the \Generate Context" web service, the ability of creating arti cial formal contexts in FcaBedrock without requiring initial data as input could be made possible with minimal e ort.
Further work also includes the incorporation of the tool FcaStone, to convert data between FCA and non-FCA formats, in FCAWarehouse to o er the power of FCA to those currently outside of the FCA community.
The nal vision of FCAWarehouse is of an online FCA data repository which facilitates the creation, conversion and donation of datasets for FCA, providing a useful collection of real and arti cial datasets in a wide variety of FCA and non-FCA formats to open the way for the wider use of FCA.