19th July: Fluidity and Context
Summary of the Third Round Table. 19.07.2022:
The third session of the Round Table Meetings was held under the topic of fluidity and context. Firstly, an introduction to the round table session and its topic by Mirco Schönfeld was held again. In this he stated that ontologies and fluidity are normally not connected, but well-known concepts within the Cluster. The Round Table serves as a platform to collect critical perspectives that help to provide a platform with fluid ontologies.
The first topic was a repetition of the previous week’s topic of knowledge graphs. Knowledge graphs are data bases that build on RDF (Subject, Verb, Object) and can be extended into sematic network structures. Entities are connected though knowledge graphs. It is important to note that we cannot have one single ontology that can be represented within one knowledge graph. There is thus a multiplicity of ontologies. This means that often there are many words with the same meaning. This must then be taught to the computer as well. Knowledge graphs present a good tool to connect these different ontologies.
The next topic of discussion was that of context. There are two types of context of importance here. Firstly, there is the user context (the data about the user) and secondly the dataset context. Both, the user, and dataset context should be combined in the knowledge graph. A question one might ask themselves here is How can we use the context information to find the knowledge the researcher is looking for? For this the knowledge graph will show what other contexts might relate to the search term and will find a unique view on what might be interesting. The context information thus determines what the user will see on the system and will guide the user through the world of knowledge. When the user context is changed, the user view will change as well.
At this point the importance of context was highlighted by Oliver Baumann. For this he presented an example of two different researchers/scientists view on gender and they might use it to put out content. In relation to this he introduced the concept of word embedding, so how algorithms analyse the context of the input world in a document database and how different texts will exhibit different meanings of words, depending on their context. He illustrated this by presenting a case study of words and their connotations in Jane Austen’s work and on the website Reddit. A stark difference in understandings of words were shown using examples such as black or women. Using context in the knowledge graph might thus help lead people to new contexts that they might not have thought of before.
We need your consent to load content.In order to show videos we use the services of third-party providers. These providers can collect data about your activities. Further details can be found under “Learn more”.
Lastly in this session a discussion on the presented topics followed. This included discussions about issues that might arise from different words being used for the same context, as well as the issue that the concepts that the machines extract might not connect with our concepts. Another point raised what that the different profile information might lead to very different outcomes.
At the moment work is done however, to expand the key search terms and thus create a more facetted view on the search terms.
Wynand van der Walt then added that context has a broader meaning that what has been mentioned. Metadata is used to describe the search itself, thus presenting a new layer. Another point he raised was about language. Firstly, it being used to sort information, but also its importance in regard to accessibility. Things need to be done and translated into English and French for example.
Next up Oliver Baumann talked about the width of the windows, meaning that the broader the contexts will be the broader the results will become. Regarding the topic of language, he commented that the algorithm should be languages-agnostic if it can be identified as a token.
Another question that will be addressed another time was about the different data types and combining imagined and written data. The conclusion on knowledge graphs was that outside input is of high importance, since this is what helps them become richer.