Abstract
Searching complex electronic catalogs often leads to "zero results" dead ends, causing user frustration and a lack of trust in the system. This paper introduces Explicit Passive Analysis, a methodology designed to bridge the gap between user-controlled query building and automated preference-based systems. By identifying minimal conflicting sets of constraints and providing concise relaxation suggestions, the proposed framework transforms search failures into diagnostic learning opportunities. Preliminary results from a pilot study indicate that this approach significantly increases user confidence and decision-making efficiency without increasing cognitive load.
Introduction
In the realm of Multi-Attribute Decision Problems (MADP)—such as searching for flights, real estate, or automobiles—users rarely possess a complete understanding of their preferences at the outset. Instead, preferences are constructed dynamically through interaction with the system. However, traditional search interfaces often fail to support this constructive process. When a user’s criteria are too restrictive, they are met with a blank results page, offering no insight into which specific constraints caused the failure or how to rectify the search. This "transparency gap" prevents users from understanding the boundaries of the database, ultimately undermining their confidence that they have found the best possible option.
The Problem: Control vs. Transparency
Existing electronic catalog interfaces generally fall into two categories, both of which present significant drawbacks:
- Query-Building Systems: These allow users to manually apply hard constraints. While providing high levels of control, they are fragile. If a query is overconstrained, the system returns nothing, leaving the user to guess which attribute (e.g., price, mileage, or brand) is the dealbreaker.
- Preference-Based Systems: These systems use weighted attributes to rank items, often showing "partial matches" to avoid the zero-results problem. However, the ranking logic is frequently opaque, leading users to feel they have lost control over the search process and making it difficult to verify why certain items are prioritized over others.
Proposed Methodology: Explicit Passive Analysis
To address these issues, we propose Explicit Passive Analysis, a framework that provides diagnostic feedback while maintaining user control. The system operates through two primary mechanisms:
1. Minimal Conflicting Sets
When a search returns no results, the system identifies the smallest group of constraints that are mutually exclusive. Instead of a generic error, the user receives a specific explanation: "There are no Audi cars that are both cheaper than €6,000 and manufactured after the year 2000." This allows the user to immediately understand the market trade-offs and the limitations of the current dataset.
2. Concise Relaxation Suggestions
The system supplements conflict identification with actionable "what-if" scenarios. These suggestions guide the user toward successful queries with minimal adjustments, such as: "You could find results if you increase your maximum price to €6,500" or "if you remove the requirement for electric windows."
To prevent information overload, the prototype was configured to compute conflicting sets for combinations of up to four constraints and limited relaxation suggestions to a maximum of three attributes.
Experimental Setup
The framework was evaluated using a high-dimensional dataset of 7,000 second-hand cars provided by Comparis. Each entry featured 35 distinct attributes (e.g., make, price, year, engine type), representing the complexity typical of modern electronic catalogs. A pilot study was conducted with four users across four interface types, varying the presence of query-building, preference-based ranking, and explicit analysis tools.
Key Insights and Results
Trust through Characterization
The study suggests that user trust is not merely a product of seeing a "best" result, but of understanding the domain. Explicit analysis helps users "map" the available data. By seeing why certain items do not exist, users become convinced that the results they do find are truly the best available options.
The "Dead End" as a Learning Tool
In this framework, a "zero hits" result is no longer a failure but a diagnostic event. It teaches the user about market realities (e.g., the correlation between luxury brands and price points), allowing them to adjust their expectations based on real-world data.
Pilot Study Findings
- Increased Confidence: Participants using explicit passive analysis reported significantly higher confidence in their final selections compared to those using traditional query-building or preference-based systems.
- Efficiency: Despite the addition of analytical text and suggestions, total interaction time did not increase. The explanations appeared to streamline the decision-making process, helping users navigate to a final choice faster.
- User Preference: Users expressed dissatisfaction with pure preference-based weighting (finding it hard to understand) and pure query-building (finding it frustrating when it failed). The hybrid approach of explicit analysis provided a "safety net" without removing the user from the steering wheel.
Limitations and Future Work
The primary limitation of this research is the small sample size of the initial pilot study. While qualitative trends were strong, they lack statistical significance. The authors are currently preparing a larger-scale study with 40 participants to validate these findings. Furthermore, future research will investigate the "sweet spot" for information density—determining how many constraints and suggestions can be presented before causing cognitive fatigue in the user.
Summary updated