The SemanticXBRL Dataset, Semantic Financial Data from XBRL Filings

This manuscript was submitted as 'Data Description' and is reviewed along the following dimensions: (1) Quality of the dataset. (2) Usefulness (or potential usefulness) of the dataset. (3) Clarity and completeness of the descriptions (see review criteria on the SWJ website).

(1) Quality of the dataset
(2) Usefulness of the dataset

I question whether a direct translation of XBRL to RDF is ideal, especially the XBRL notion of context.
The XBRL data suffers from using XML (linkbases) as underlying data model; RDF typically favours an entity-centric data model.
Thus, a more intuitive representation in the Semantic Web context is to follow the modelling constructs that established vocabularies use, such as Data Cube (QB) or SKOS.
However, the presented dataset does not use established vocabularies (such as the QB vocabulary which is applicable to XBRL and widely used for modelling numerical datasets).
Why the authors did not follow the QB model for representing numerical data is not explained.

Further, the authors do not provide their dataset as Linked Data, i.e., an HTTP GET on Semantic XBRL URIs do not return the corresponding RDF.
In an updated version, the URI schemes/templates for constructing entity URIs should be included (e.g., the URI template for identifying entities identified with a CIK).

XBRL's calculationArcs that describe how various measures related to each other are not mentioned in the paper.
The SIC taxonomy for classification of EDGAR entities is not mentioned.
The paper does not include licensing information.
Also, there is no reported usage of the data.

(3) Quality and completeness of description

My colleagues and I have done work in the XBRL space which is missing from the description (see, for example, the Linked EDGAR dataset, http://datahub.io/dataset/linked-edgar, which even provides links to Freebase and DBpedia, and Kaempgen et al., "Accepting the XBRL Challenge with Linked Data for Financial Data Integration", http://2014.eswc-conferences.org/sites/default/files/papers/paper_224.pdf).

In sum, the modeling decisions are not sufficiently explained, the dataset is not available as Linked Data, and the contribution over existing work is unclear.

This manuscript was submitted as 'Data Description' and should be reviewed along the following dimensions: (1) Quality of the dataset. (2) Usefulness (or potential usefulness) of the dataset. (3) Clarity and completeness of the descriptions.

(1) Quality of the dataset.

This paper describes a dataset (named as "SemanticXBRL") converted from XBRL fillings published by the US SEC. According to the authors, the goal is to enhance interoperability and linking with the rest of the Web of Data. The "Evaluation" section aims to assess the quality of the approach and resulting datasets along these two dimensions. Two concrete examples are provided to show how to relate to comparable facts and how to perform complex aggregations.

However, there are three issues with the current results.

First, this paper did not explicitly define metrics to assess the quality of the resulting datasets used in this approach. These metrics should be based on typical usages patterns of the resulting datasets. Experiments should be designed accordingly and these metrics should be measurable.

Second, the current "Evaluation" section did not assess the quality of SemanticXBRL compared to original XBRL datasets.

Third, there is not enough demonstration about the improvement about linking to other datasets on the Web. Section 3.1 described how "linking to external data" was done but did not provide much quality assessment about the linking. Although it was mentioned that 63 links about companies were generated, it is not clear what it means in terms of usabilities of these links.

(2) Usefulness (or potential usefulness) of the dataset.

By mapping XML schema to OWL and XML data to RDF data, the SemanticXBRL datasets improves integrations within the same data source as is shown in Section 4. However, major usages of the SEC XBRL datasets involve reasoning over domain-specific concepts defined in the US-GAAP taxonomies, constraints checking, reconciling commons rules with exceptions, etc. These typical usage patterns are difficult to be achieved with the current mapping mechanisms described in this paper.

(3) Clarity and completeness of the descriptions.
This paper provides clear and comprehensive descriptions about the approach and the datasets.