LogosLink User's Manual · LogosLink version 2.0.0

Model Refinement

Some kinds of models (contexts, ontologies and question sets) can be refined. Refining a model means creating another model that is very similar to the first one, but incorporates some changes. This is useful in scenarios where you need to work with different levels of detail at different levels.

For example, you can have an ontology embedded in your corpus. Then, you can copy this ontology to each of the topics in the corpus, and add some extra elements to it to better express the situation represented by the topic. Then, you can copy each topic's ontology to each document in the topic, and add extra elements to it, to cater for the peculiarities of the document. In this scenario, all the ontologies involved stay connected, and analytics that operate on them will use these connections when computing results.

Details

Example

Imagine that you are working with a corpus about immigration. You can add an embedded ontology that captures the major concepts about this theme, such as immigrants, travel, origin and destination countries, discrimination, etc.

Then, you add a few topics to the corpus, in order to organise documents to study specific sub-themes withing immigration. For example, you could add topics such as "School" and "Labour market" to study how immigrants' experiences develop and are portrayed when going to school and when looking for a job. For each of these topics, you create an ontology by copying it from the corpus, and then add, modify or even delete elements so that each of these ontologies capture the specific theme they deal with. For example, you could add ontology elements to deal with bullying at school or academic performance to the ontology in the "School" topic.

Finally, when you start analysing documents, you can create an ontology for each one by copying that in the associated topic. Then, you can add extra elements to the document ontology to better reflect the issues mentioned in the document. For example, if you want to analyse a document titled "Grading Biases in School Teachers" within the "School" topic, you may want to add ontology elements to represent biases and grades.

In this manner, you end up with a "tree" of ontologies. The "root" of the tree is the corpus ontology, as it is the most abstract and covering. Then, the tree branches into each topic, adding details. Each topic ontology is a refined version of the corpus ontology. Finally, the tree's leaves are the document ontologies, as they contain the most details and are the most specific. Each document ontology is a refined version of the corresponding topic ontology.

Connections between refined models

Elements in the models that can be refined (contexts, ontologies and question sets) have identifiers. An identifier is a piece of text that distinguishes an element in a model from other elements in the same model or another models of the same kind. For example, an category with identifier "h5kq75c2" in the corpus ontology from the previus example and a category with the same identifier in one of the topic ontologies are considered to be the same category. This is so even if you change their names, properties or associations.

In other words, an identifier establishes the identity of an element. If you change an element's identifier, you change its identity.

LogosLink keeps track of refinement relationships between model by looking at common identifiers. In the example above, imagine that you have an "Immigrant" category in your corpus ontology, and give it the identifier "imm". When you refine this ontology to create the "School" topic ontology, the "Immigrant" category with the "imm" identifier will be copied along. You may then want to rename this category as "Immigrant Child" so that it better describes immigrant children at school. You can rename it. As long as you don't alter the "imm" identifier, the "Immigrant Child" category in the "School" topic ontology will be considered identical to the "Immigrant" category in the corpus ontology.

Uses and consequences of model refinement

Using identifiers as illustrated above is recommended. It provides you with semantic connections between elements across models, and allows LogosLink to make certain inferences when computing analytics results. The following sections provide some examples.

Working with an aggregated context

Imagine that you want to analyse documents about three case studies about a common theme, such as controversial heritage. You create a corpus, add a topic for each case study, and populate it with documents. Then, you create a context for each topic, each one with its own positions and agents as determined by the specific case study.

You could work with this, analysing each document in terms of the context of the associated topic. You would be able to run analytics at the topic level that show aggregated results for documents in the topic. But you wouldn't be able to compare results across topics and derive overall conclusions.

To do this, you would need to create an aggregated context at the corpus level. Using LogosLink Desktop, you can easily create a corpus-embedded context from the existing topic contexts. When you do this, LogosLink copies elements in each of the contexts and brings along their identifiers. You end up with a corpus context that contains everything in each of the topic contexts. This allows you to run corpus-wide analytics that rely on a reference context, such as position adherence analytics, and look for commonalities or differences across topics.

Using abstraction with ontologies

Refining ontologies allows LogosLink to use abstraction when reporting analytics results.

Imagine a scenario like the one described at the beginngin of this page, consisting of a corpus with a few topics and documents for each topic, and refined ontologies at each level. Imagine also that you record denotations in document ontologies or argumentation models. When, later on, you run a denotation analytics at the corpus level, the reference ontology will be the "root" one, that is, the abstract ontology embedded in the corpus. However, denotations have been recorded against the particular ontologies for each document, not the corpus one. Since identifiers allows LogosLink to keep track of identity across models, the analytics can still provide meaningful aggregated results across documents.

For example, imagine that you have recorded a denotation in a school-related document targetting the "GradingBias" category. Imagine that this category is a subtype of the more abstract "Discrimination" category in the corpus-level ontology. When displaying results, the denotation analytics will show this denotation as targetting a subtype of "Discrimination", because "Discrimination" exists in the reference ontology whereas "Grading Bias" doesn't. LogosLink carries out this abstraction process automatically, thus allowing you see results that aggregate different ontologies.