Towards a Novel User Satisfaction Modelling for Museum Visit Recommender Systems

George PavlidisCommunications 2018 · DOI · OpenAlex

Abstract

Modern recommender systems technology appeared in Cultural Heritage application relatively recently, particularly during the dawn of the 21st century. There is already a significant amount of relevant works in the bibliography, which has been primarily empowered by large-scale research and development projects. Various approaches have been adopted from the recommender systems technology, including collaborative filtering, content-based, knowledge-based and hybrid systems. In most of these approaches that focused on museum guidance, which is the focus of this paper, the museum has been assumed to be a form of a gallery and the visitor was treated primarily as a user in seek on engagement and enjoyment. The free museum roaming was the main form of visit that has been considered and targeted, while the educational factors and storytelling aspects have been markedly overlooked. In this paper a new framework for the user satisfaction modelling is being presented that quantifies user satisfaction based on a weighted combination of various probabilistic factors that are being estimated during a museum visit. The goal is to provide a model of user satisfaction that can be used for museum recommenders that could guide either free-roaming visits or guided-tour scenarios for visitors of various motivations and backgrounds.

Introduction

While recommender systems (RS) are ubiquitous in digital domains like e-commerce and streaming, applying them to physical cultural heritage sites presents unique challenges. Most existing museum recommenders treat the venue as a simple gallery for "free roaming," focusing almost exclusively on maximizing user enjoyment.

However, this approach often overlooks two critical elements: the "Museum’s Voice" (educational goals and featured items) and the physical/temporal constraints of a real-world visit. Visitors in a physical space experience fatigue, are distracted by nearby "shiny" objects, or become frustrated by crowds. This paper proposes a novel framework that shifts the focus from maximizing satisfaction to minimizing the probability of visitor dissatisfaction ($p_d$), acting as a "defensive" strategy to maintain engagement.

The Proposed Framework

The framework introduces a probabilistic model that estimates dissatisfaction as a composite of four distinct dynamics. By integrating these factors, the system can predict the moment a visitor is likely to lose interest or deviate from a meaningful path.

  1. Temporal Dynamics ($p_t$): The model employs a "two-clock" system. One clock tracks the visitor's remaining time, while the other tracks the "required" time for an exhibit. Dissatisfaction is modeled using a piecewise continuous function: an exponential decay if the user skips content too quickly, and an inverse quadratic function if they spend too much time (suggesting they might be stuck or over-exerting).
  2. Proximity Dynamics ($p_p$): This accounts for the "distraction" factor. If a highly popular or "featured" item is physically near the current exhibit, it may reduce the satisfaction of the current interaction by pulling the user’s attention away.
  3. Content-Based Dynamics ($p_c$): This models the user’s bias toward specific themes. It combines semantic similarity with physical distance, recognizing that a relevant item nearby is a more likely target than a highly relevant one on the other side of the building.
  4. Obtrusions ($p_o$): A normalized estimate of physical obstacles, such as overcrowding or closed galleries.
Figure 2: Conceptual representation of a museum visit path influenced by popularity, featured status, obtrusions, and neighborhood dynamics.

The final model utilizes a minimax approach: the system aims to minimize the maximum possible estimated dissatisfaction. This ensures the recommender prevents the worst-case scenario of total visitor disengagement, whether in free-roaming or structured storytelling scenarios.

Evaluation and Key Insights

The framework was evaluated using a large-scale simulation involving 1,000 museum exhibits and 10,000 simulated visitors. Despite a high sparsity (1% rating density), the simulation demonstrated several key advantages:

  • Modeling the "Negative": In a physical environment, it is often more impactful to predict what will cause a user to quit than to predict what will provide "maximal" joy.
  • Stakeholder Influence: The model explicitly includes "featured item scores" provided by the museum, balancing user preferences with curatorial intent.
  • Visualizing Disengagement: The model can generate "dissatisfaction profiles" for a tour, identifying exactly which points in a sequence are likely to lose the visitor's interest. This can help curators identify "weak points" in exhibition design.
  • Weighted Flexibility: By adjusting the weights of the four factors, the system can be tuned for different visitor types (e.g., a "power user" focused on content vs. a "casual visitor" more influenced by proximity).

Limitations and Future Work

Currently, the weights for the various factors are predefined or heuristic. Future work will focus on learning these weights via machine learning from real-world visitor data. Additionally, while the simulation is large-scale, it cannot fully capture the unpredictability of human behavior in a physical space. The author suggests applying this framework to "storytelling" scenarios, where the recommender must maintain a narrative thread while adapting to real-time physical constraints.

Summary updated

Abstract

Modern recommender systems technology appeared in Cultural Heritage application relatively recently, particularly during the dawn of the 21st century. There is already a significant amount of relevant works in the bibliography, which has been primarily empowered by large-scale research and development projects. Various approaches have been adopted from the recommender systems technology, including collaborative filtering, content-based, knowledge-based and hybrid systems. In most of these approaches that focused on museum guidance, which is the focus of this paper, the museum has been assumed to be a form of a gallery and the visitor was treated primarily as a user in seek on engagement and enjoyment. The free museum roaming was the main form of visit that has been considered and targeted, while the educational factors and storytelling aspects have been markedly overlooked. In this paper a new framework for the user satisfaction modelling is being presented that quantifies user satisfaction based on a weighted combination of various probabilistic factors that are being estimated during a museum visit. The goal is to provide a model of user satisfaction that can be used for museum recommenders that could guide either free-roaming visits or guided-tour scenarios for visitors of various motivations and backgrounds.
BibTeX
@inproceedings{pavlidis2018towards,
  author = {George Pavlidis},
  title = {{Towards a Novel User Satisfaction Modelling for Museum Visit Recommender Systems}},
  booktitle = {Communications},
  year = {2018},
  doi = {10.1007/978-3-030-05819-7\_6},
}
Figure