View Item 
        •   Utrecht University Student Theses Repository Home
        • UU Theses Repository
        • Theses
        • View Item
        •   Utrecht University Student Theses Repository Home
        • UU Theses Repository
        • Theses
        • View Item
        JavaScript is disabled for your browser. Some features of this site may not work without it.

        Browse

        All of UU Student Theses RepositoryBy Issue DateAuthorsTitlesSubjectsThis CollectionBy Issue DateAuthorsTitlesSubjects

        Machine Learning Analysis of Inner Experiences in Reports of Psychoactive Substances

        Thumbnail
        View/Open
        Bachelor_Thesis_AI_Apers_2020_final.pdf (1.521Mb)
        Publication date
        2020
        Author
        Apers, A.P.
        Metadata
        Show full item record
        Summary
        A better understanding of psychedelic experiences becomes increasingly important as research shows their therapeutic potential (Carhart-Harris and Goodwin, 2017). One possible approach to investigate these experiences is the analysis of written reports that document individuals’ experiences from the first-person point of view. An attempt at analysing such reports, described in Coyle et al. (2012), tries to predict which psychedelic substance is described based on a Bag of Words representation of the reports. However, this approach doesn’t provide enough meaningful information about the subjective inner experiences that users have. To overcome this limitation, the present research proposes an approach that predicts dimensions of the experiences that express information about the subjective inner state. A small subset of 120 reports in the Erowid Experience Vaults was annotated using a 5 point Likert scale for each of 8 dimensions that were identified as relevant based on Altered States of Consciousness literature. To predict these scores, seven different report representation methods and two types of regression techniques were used to create a total of 14 different models. Reports were represented using Sentiment Analysis, Bag of Words, Word2Vec and Doc2Vec approaches. The latter 3 approaches were also supplemented with Sentiment Analysis information. The most important finding is that all 14 regression models performed better compared to baseline null models in predicting dimensions relating to inner experiences. Furthermore, the models that adhered to word order and used Word2Vec/Doc2Vec techniques made better predictions compared to simple Bag of Words models. Finally, models that were supplemented with Sentiment Analysis information performed better compared to the counterpart models without this information. Additionally, challenges and suggested improvements for this new approach and some observations regarding the data are discussed. Future research could focus on investigating the influence of more variables, e.g. set and setting, on the inner experience. A more in-depth understanding of psychoactive experiences could help pharmacological and neuroscience research form hypotheses for new investigations.
        URI
        https://studenttheses.uu.nl/handle/20.500.12932/37104
        Collections
        • Theses
        Utrecht university logo