Roelofsen, Renée (2025) The Influence of ACT-R’s Declarative Memory System on the HUMAT Agent-Based Model. Master's Thesis / Essay, Computational Cognitive Science.
|
Text
mCCS2025RoelofsenRL.pdf Download (1MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
Toestemming.pdf Restricted to Registered users only Download (192kB) |
Abstract
Agent-based models (ABMs) are used to study and predict behaviour across many scientific domains, making it crucial that they reflect real-world behavioural outcomes as closely as possible. Given that memory plays a significant role in shaping behaviour, this preliminary study explores if ACT-R’s declarative memory system can help produce more realistic micro- and macro-level behaviour in the HUMAT ABM. HUMAT uses social scientific theories to model agent cognition, decision-making and social interactions driven by a set of three core needs which agents strive to fulfil when making a decision. To assess the impact of ACT-R’s memory system, a test case is run on both HUMAT’s basic integrated framework and a HUMAT model integrated with ACT-R’s declarative memory system. In the test case, residents of a neighbourhood decide whether to collectively adopt a heating network, forming preferences through social interactions involving either persuasion (signalling) or seeking persuasion (inquiring). Conditions were determined based on different populations and average social network sizes. Results indicate that the ACT-R integrated model produces realistic micro-level behaviour, like signalling and inquiring alters in a more varied manner. Additionally, the ACT-R integrated model has the ability for preferences to fade over time, although this effect may be too strong in some cases. On a macro level, Humats contacted around 80% of their social network in the ACT-R integrated model instead of 99%. This was more realistic for most conditions but not for the conditions with the smallest average social network size. After running both models for 100 days, the ACT-R integrated model produced less social satisfaction and more non-social satisfaction compared to the control model. This caused a higher level of cognitive dissonance in the ACT-R integrated model, and no noteworthy difference in overall satisfaction between the two models except in the conditions with a social networks size of 12. In these conditions, overall satisfaction was slightly higher in the control model. Despite the more realistic aspects of decision making behaviour and substantial differences in some evaluation metrics, the final preference divisions of both models was similar. Therefore, the simpler control model is the best model for the test case of this study. However, the overall results indicate that ACT-R’s declarative memory system is best utilised in more complex environments. Future studies are necessary to investigate this further.
| Item Type: | Thesis (Master's Thesis / Essay) |
|---|---|
| Supervisor name: | Taatgen, N.A. and Jager, W. |
| Degree programme: | Computational Cognitive Science |
| Thesis type: | Master's Thesis / Essay |
| Language: | English |
| Date Deposited: | 06 Jun 2025 07:14 |
| Last Modified: | 13 Jun 2025 11:06 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/35255 |
Actions (login required)
 |
View Item |