Background

Whilst a growing number of studies in positive psychology have used qualitative approaches, collecting and analysing such data is time-consuming and labour-intensive. Natural language processing (NLP) from the computer sciences use machine learning algorithms to extract themes from massive qualitative data and generate insights, enabling the rapid encoding and analysis of massive amounts of data. Yet few studies have directly compared NLP with other qualitative analysis methods in wellbeing science.

Aims

This proof-of-concept compares similarities and differences between themes generated by NLP from conventional qualitative analysis methods.

Method

Drawing on existing data that explored conceptualisations and experiences of wellbeing, we compare existing results of the data produced by conventional qualitative analyses (prototype analysis and interpretive phenomenological analysis) with an NLP analysis adapted from Leeson and colleagues (2019).  The NLP algorithm was implemented using Python. The program runs on the qualitative data collected and generates a set of themes, which are compared with results from and judged based on conceptual similarities.

Results

Themes from NLP were comparable with results yielded from traditional approaches, suggesting that NLP might be useful in conjunction with conventional qualitative analysis methods, although supervision of the NLP models is necessary to ensure the results align with conventional wellbeing theories. Among NLP techniques, topic modelling, a technique that can summarise a large amount of textual data into recurring themes, is a promising approach worth exploring. Notably, results indicate that the choice of a specific NLP technique should be informed by the purpose of research, the volume of data and the technical infrastructure, and expertise available.

Conclusion

This project aligns with the emerging trend of bringing computational social science into wellbeing research, adding to the methodological tools available within the wellbeing sciences fuelling the development of data-driven methodologies in the field.