Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Hybrid Personalization Using Declarative and Procedural Memory Modules of the Cognitive Architecture ACT-R

Kevin Innerebner, Dominik Kowald, Markus Schedl, Elisabeth Lex

TL;DR

The paper addresses the opacity of sub-symbolic recommender systems by proposing a hybrid ACT-R architecture that combines declarative memory and procedural memory via production rules to enable psychology-informed, transparent personalization. It outlines how item interactions are stored as chunks with activations and how production rules adjust activations to drive top-k recommendations, providing explanations grounded in symbolic reasoning. Key contributions include a hybrid retrieval-and-reasoning framework, rule-classification and adaptation mechanisms (individual, group, global), explainability interfaces, and plans to study cognitive biases and counterfactuals. If realized, this approach could yield human-centered recommender systems that reflect cognitive processes and biases, with mechanisms for counterfactual analysis and bias mitigation.

Abstract

Recommender systems often rely on sub-symbolic machine learning approaches that operate as opaque black boxes. These approaches typically fail to account for the cognitive processes that shape user preferences and decision-making. In this vision paper, we propose a hybrid user modeling framework based on the cognitive architecture ACT-R that integrates symbolic and sub-symbolic representations of human memory. Our goal is to combine ACT-R's declarative memory, which is responsible for storing symbolic chunks along sub-symbolic activations, with its procedural memory, which contains symbolic production rules. This integration will help simulate how users retrieve past experiences and apply decision-making strategies. With this approach, we aim to provide more transparent recommendations, enable rule-based explanations, and facilitate the modeling of cognitive biases. We argue that our approach has the potential to inform the design of a new generation of human-centered, psychology-informed recommender systems.

Hybrid Personalization Using Declarative and Procedural Memory Modules of the Cognitive Architecture ACT-R

TL;DR

The paper addresses the opacity of sub-symbolic recommender systems by proposing a hybrid ACT-R architecture that combines declarative memory and procedural memory via production rules to enable psychology-informed, transparent personalization. It outlines how item interactions are stored as chunks with activations and how production rules adjust activations to drive top-k recommendations, providing explanations grounded in symbolic reasoning. Key contributions include a hybrid retrieval-and-reasoning framework, rule-classification and adaptation mechanisms (individual, group, global), explainability interfaces, and plans to study cognitive biases and counterfactuals. If realized, this approach could yield human-centered recommender systems that reflect cognitive processes and biases, with mechanisms for counterfactual analysis and bias mitigation.

Abstract

Recommender systems often rely on sub-symbolic machine learning approaches that operate as opaque black boxes. These approaches typically fail to account for the cognitive processes that shape user preferences and decision-making. In this vision paper, we propose a hybrid user modeling framework based on the cognitive architecture ACT-R that integrates symbolic and sub-symbolic representations of human memory. Our goal is to combine ACT-R's declarative memory, which is responsible for storing symbolic chunks along sub-symbolic activations, with its procedural memory, which contains symbolic production rules. This integration will help simulate how users retrieve past experiences and apply decision-making strategies. With this approach, we aim to provide more transparent recommendations, enable rule-based explanations, and facilitate the modeling of cognitive biases. We argue that our approach has the potential to inform the design of a new generation of human-centered, psychology-informed recommender systems.
Paper Structure (9 sections, 3 equations, 1 figure)

This paper contains 9 sections, 3 equations, 1 figure.

Table of Contents

  1. Introduction
  2. Background: ACT-R Architecture
  3. Our Vision: Hybrid Modeling Approach
  4. Production Rules
  5. Classes of Rule Conditions
  6. Rule Adaptation
  7. Explainability
  8. Cognitive Biases
  9. Conclusions and Future Work

Figures (1)

  • Figure 1: Visualization of our envisioned hybrid modeling approach based on ACT-R. The declarative memory stores item interactions as chunks and computes initial activations. The IF conditions of the production rules in the procedural memory are checked against these activations and could lead to reweighted activations. The items associated with the highest reweighted activations form our top-$k$ recommendations, for which explanations are provided based on the applied production rules.