Ontologies and their use in Behavioural Science

What are ontologies and why do they matter?

Behavioural science is a broad field that encompasses psychology, sociology, anthropology, neuroscience, and other disciplines. Given the complexity of human behaviour, the myriad factors that influence it, and the significant consequences it can have for individuals and society, researchers, practitioners, and policymakers are constantly seeking ways to organise, analyse, and communicate information about human behaviour.  A powerful tool for organising such knowledge is the use of ontologies. But what exactly are ontologies, why are they important, and how is BR-UK leveraging ontologies to advance behavioural science? 

The term "ontology" originally comes from philosophy, where it refers to the study of existence and reality. However, in the context of science and information technology, an ontology is a formal representation of knowledge within a domain. It is a structured way of organising concepts within a field, and the relationships between them, using a shared language. You can think of an ontology as a sophisticated and interconnected dictionary or taxonomy for a specific field of study. For example, if you were to use an ontology to capture knowledge and data related to the behaviour swimming, then you might represent it something like the figure below: 

The ontology displays a number of classes within a network. These classes are arranged hierarchically into “child” and “parent” classes. For example, different types of swimming strokes, such as “backstroke” and “breaststroke”, are the child classes of the parent class, “swimming stroke”. The ontology also allows us to specify the different locations where swimming may take place (e.g., in a “swimming pool” or “lake”), whether the behaviour involves the use of any equipment (e.g., using “armbands” or “water weights”), and how the behaviour can be measured (e.g., the time spent swimming or the distance a person swims). Each concept within the ontology has a label that names the concept, a definition for the concept, a unique identifier (uniform resource identifier; URI), which is a sequence of characters to identify an abstract or physical resource. 

Ontologies provide a shared vocabulary, facilitating clearer communication of the phenomena being studied. By defining relationships between concepts, ontologies help integrate knowledge and data from different disciplines and research areas, and enable different databases and systems to "talk" to each other by providing a common framework to classify knowledge and data. Once a body of knowledge has been represented by an ontology, we can use computer-based tools to automate data extraction from scientific reports, develop more accurate and efficient predictive models of behaviour, and design more personalised and effective interventions to help people change behaviour. 

Challenges in Developing Ontologies for the Behavioural Sciences

Human behaviour is complex and capturing this complexity in a structured ontology is a significant challenge, as is maintaining ontologies over time as understanding of behaviour develops. Because behavioural science spans multiple disciplines, each with its own terminology and conceptual frameworks, integrating these diverse perspectives into a cohesive ontology can be challenging. As such, there is often a tension between creating highly detailed, comprehensive ontologies, and maintaining usability for researchers and practitioners. 

BR-UK: The Future of Ontologies in Behavioural Science

The BR-UK team has been exploring how we can improve the use of ontologies in the social and behavioural sciences in three lines of research:

1. Making Ontologies Interoperable (DEMO-INTER). Ontologies that are interoperable (i.e., can be used together without causing inconsistency or conflict), provide the means to foster a shared language, reduce fragmentation of theories, and support the use of data in ways that are more replicable, machine-readable, and integrable across different disciplines. However, behavioural researchers do not currently have a tried-and-tested method for evaluating or enhancing the interoperability of ontologies. This is important because researchers may want to compare how different ontologies represent concepts, align them where possible and useful, and facilitate collaborative working and integration of data. The DEMO-INTER project has developed a method to identify identical and conceptually similar classes within ontologies, enabling unintended overlaps that hinder interoperability to be detected and addressed. This method has been informed by an empirical investigation and comparison of both AI and human approaches and provides a step-by-step approach to assessing the interoperability of existing ontologies (you can read more about this work here).
2. Ontologising Theories of Behaviour Change (DEMO-THEORY). Identifying relevant behaviour change theories for developing interventions and comparing and integrating theories of behaviour change is hampered by the plethora of theories and by inconsistent naming and defining of theory constructs. To advance theory development, it is important to develop a consistent, coherent, and comprehensive way of representing constructs from behavioral theories. The DEMO-THEORY project has conducted a comprehensive mapping of constructs from behaviour change theories to ontology classes so that the theories can be easily searched, compared, and integrated. To date, 1379 constructs have been identified across 87 theories, which have been mapped to 495 ontology classes. This mapping enables a common set of core constructs to be identified across theories, and ontology-based reasoning together with machine learning enables the identification of canonical, integrative theory elements (you can read more about this work here).
3. Ontologising Datasets (DEMO-DATA). Ontologies can be used to improve the interoperability of datasets, making it easier to search within and across different datasets, and combine datasets to promote evidence synthesis and model development. The DEMO-DATA project has developed and applied a method for annotating datasets in the social and behavioural sciences, using data on smoking and e-cigarette trends as an example. The aim is to demonstrate how ontologies can simplify working with datasets, making them searchable, compatible, and easier to combine in a way that ensures consistency and enables new insights that would not be possible if analysed separately (you can read more about this work here).

Conclusion

Additional input: Fatima Sabiu Maikore, Janna Hastings, Robert West, Suvodeep Mazumdar, Susan Michie, Thomas Webb and Vita Lanfranchi.