Nybom, Jussi Waltteri (2024) Data-Driven Approaches to Narrative Personalisation through Psychologically Motivated Models. PhD thesis, University of the Arts London.
Type of Research: | Thesis |
---|---|
Creators: | Nybom, Jussi Waltteri |
Description: | AI-driven personalisation offers a clear opportunity for creative industries to engage audiences more effectively. This project seeks to understand how such personalisation can be effectively and ethically exploited in story experiences to generate greater audience engagement. More specifically, this thesis addresses personalisation of narratives to accommodate the user’s preferences, with an aim to understand and accommodate them better. For this, three studies are conducted. In the first study, an interactive narrative is created with the purpose of incorporating the user’s choices to create a user profile featuring the Five-Factor Model (FFM) and the Need for Affect (NFA), with the questions designed to understand the user’s preferences within the narrative and therefore possibly indicating their personality in general. Next, a narrative is personalised to fit with the user’s estimated personality. It is hypothesised that the choices the user makes within the narrative would have at least some correlation with the choices they would make in real life, meaning the narrative could be used as a personality test. Even if little or no correlation with real-life preferences can be found, the choices could indicate preferences within fictional narratives, such as how complex, imaginative or painful they prefer narratives to be, and what the protagonist should be like. Nevertheless, it did turn out the interactive narrative could be used to measure at least Extraversion and Emotional Stability. The effectiveness of personalising the story is then tested, seeing how effective it is to change the style of language according to Extraversion levels, the protagonist personality according to other FFM factors, and the ending according to the NFA. The results were strikingly strong, with personalisation appearing to improve the experience across all traits and with both personality test and interactive narrative results. In the second study, we attempt to use Natural Language Processing (NLP) for modifying the language, so that the personalisation could be done automatically and not by hand. For this, a number of language models were trained and used to create different version of a short story. Different versions of the ending were also created. The results were then tested on participants, and their opinions on the story and its language were compared with their FFM personality scores, their reading skills and their age and gender. The results presented a complicated picture featuring some surprises. In the third study, the Myers–Briggs Type Indicator (MBTI) was used instead, with several machine learning algorithms tried for classifying users by their MBTI type based on text they have written on social media. It was suggested this approach could be used for an MBTI-based recommender system that identifies novels with authors, characters or narrators similar to the reader. The results suggested that the approach could indeed work, with results in the extraversion dimension particularly promising, but better data would be needed to gain strong enough results for a good recommender system. There are several potential impacts relating to this work. It could lead to the creation of new measures for testing people’s preferences in art. User profiles using the measures could then be used to personalise narratives to fit with the user’s preferences in various ways. By integrating personality frameworks, recommender algorithms can suggest novels, films and other works that not only align with users' personality traits but also cater to their broader preferences, offering a more tailored and enriching experience, with little usage data needed. Similar approaches with NLP can also be used to alter pre-existing works. An extensive literature review is also conducted, giving a wide introduction to the topic and related fields. This is then used as a background for suggesting more possible future pathways for personalising narratives. |
Your affiliations with UAL: | Research Centres/Networks > Institute for Creative Computing |
Date: | February 2024 |
Funders: | Arts and Humanities Research Council |
Date Deposited: | 27 Jan 2025 14:39 |
Last Modified: | 27 Jan 2025 14:39 |
Item ID: | 23240 |
URI: | https://ualresearchonline.arts.ac.uk/id/eprint/23240 |
Repository Staff Only: item control page | University Staff: Request a correction