During my data modeling process of historical occupations, I must go through various meta model transformations. Starting with Qualitative Data Analysis (QDA) in XML to structure the important segments of historical texts, I’m going on with that information to create normalised and enriched data in tabular form, which in turn is imported to a graph database to build an ontology. Surfing between the meta models may cause headaches, because there is a lot of terms to mix up and it is sometimes hard to decide which data from one point of model belongs to which point of another model. Attributes exist in XML, but also in relational databases; do they describe the same information? Do they contain the same data? What is the relationship between an individual in RDF/OWL and a tuple in a relational database? In this blog series I will walk through the meta models of Qualitative Data Analysis (QDA), XML, relational databases, simple graphs, the labeled property graph database neo4j and the ontology language RDF/OWL to focus on the terminology of each meta model and intersections and differences between the individual terms.
Let’s begin with the terminology of QDA. The main objective of QDA is to structure media-represented information with “codes” (a brief introduction here). Codes classify a segment of information regarding to theoretically interesting characteristics. (Früh 2004, S. 42) A code can classify objects, topics or analytical constructs and thus moves between a broad understanding in the sense of a “concept” and a narrow understanding in the sense of a “variable”. (Kuckartz 2016, S. 34) Classify a segment with a code creates a “coding”. “Coding units” define the size of the unit to be segmented in order to make codings understandable and uniform. In most cases their length varies from one word to one sentence – depending on the complexity of the assigned code.
The second level of structuring are the examined “cases”. Since I’m interested in occupational profiles, every description of an occupational title is one “case” and I define codes like productive activity with a length of one sentence, which could contain subcodes with more detailed information about “tasks”, “products” and “materials” with a coding unit length of one word. Annotating the sentence: ”The goldsmith casts gold into rings.” with the codes could colour it up to:
There are powerful software solutions like MAXQDA or Atlas.ti that can help to code and analyse text in a very comfortable way, but they reach their limits when it comes to enriching codings with more information. As we see later, it is for example important to assign unique identifiers to words to distinguish their meaning from their orthographical form; otherwise algorithms can’t discern if “cast” means to “form by pouring (e.g., wax or hot metal) into a cast or mold” or “engaging someone for a film”.
A meta model like XML together with the data model from the Text Encoding Initiative (TEI) offers the possibility to keep our original text and to enrich it with further information at the same time. This approach has the advantage that information is not lost through irreversible transformation processes and it is easier to implement the data in data bases, but it will cost more time and require skills in XML and XLST. If it is enough for you to focus on the codes for (statistical) analysis with some back links to the codings for a better understanding of the case? QDA Software will do it. Does your main interest lie in the traceable processing of the codings and you may want to transfer your data to another format or import it into a database? XML may be your friend.