Abstract
With the development of modern information technology and global information network, more and more knowledge service providers are transforming to digital platform service mode. As the intermediary of the market, the enterprise gathers the supply and demand information of knowledge demanders and knowledge suppliers together through the virtual network platform, so as to achieve the maximum matching of the interests of both sides. For the actual knowledge service supply and demand matching problem, considering that the cardinality of matching objects may be large. Firstly, we propose a multi-attribute ordered clustering method based on probabilistic linguistic information, which makes the two-sided matching problem in the same ranked class. Secondly, considering that the preference information is affected by the subject’s psychology, based on the disappointment theory, the disappoint-satisfaction function is constructed to modify the preference information. In addition, the expected preference value is defined to ensure that the matching parties are mutually satisfied. An example shows that the method is effective and practical. The method proposed in this paper extends the application of knowledge service supply and demand matching method, and provides a theoretical basis for improving the efficiency and personal satisfaction of knowledge service supply and demand matching.
Introduction
Digital platforms increasingly act as intermediaries connecting "knowledge demanders" (e.g., companies needing custom software) with "knowledge suppliers" (e.g., freelance development teams). This creates a classic two-sided matching problem. However, applying standard matching algorithms to large-scale knowledge platforms presents three major challenges: scale, human psychology, and the conflict between individual and group optimization. When thousands of users are on a platform, it is computationally heavy for users to rank every potential partner. Furthermore, traditional models assume user preferences are static, whereas human satisfaction is relative and influenced by expectations and potential missed opportunities (disappointment/regret). Finally, algorithms that maximize overall group satisfaction often force individuals into unsatisfactory matches, leading to platform churn.
Methodology
The proposed framework addresses these challenges through a three-step process involving clustering, psychological preference modification, and iterative matching.
1. Probabilistic Linguistic Ordered Clustering
To solve the scale problem, the platform evaluates users on various attributes using probabilistic linguistic terms (e.g., an 80% chance a supplier is "good" and a 20% chance they are "excellent"). The algorithm calculates the proximity of each user to an "ideal" profile and groups them into ranked tiers (e.g., Tier 1, Tier 2). Matching is restricted to participants within the same tier, which drastically reduces the computational load and ensures comparable quality levels.
2. Preference Modification via Disappointment Theory
User preferences are not absolute. Based on disappointment theory, the authors construct a "disappointment-satisfaction" function to modify raw preference scores. This function factors in multi-reference dependence, recalculating scores based on how a user would feel comparing a specific match to all other possible matches in their tier. This ensures the model accounts for the psychological impact of perceived "missed opportunities."
3. Multi-Stage Feedback Mechanism
The system employs a dual-objective optimization model to identify optimal pairs. It introduces a negotiation-like feedback loop: if a user's minimum expectations are not met, the system asks if they are willing to lower their standards for the next round. This iterative process continues until matches are finalized or users refuse further compromise.
Results and Discussion
The method was demonstrated using a synthetic illustrative example involving 25 software companies (demanders) and 20 university development teams (suppliers) evaluated across attributes like technical complexity and cultural fit.
The clustering algorithm successfully grouped participants into four distinct, manageable ranked classes. Following the application of psychological preference modifications, the initial optimization and subsequent feedback loop—where users gradually adjusted their minimum expected values—resulted in 14 highly satisfactory matching pairs. Sensitivity analysis revealed that as users become more sensitive to disappointment (adjusting $\alpha$ and $\beta$ parameters), the psychological impact of potential missed opportunities dominates their baseline preference scores, highlighting the importance of psychological modeling in matching efficiency.
Conclusion
This research provides a theoretical basis for improving the efficiency and personal satisfaction of knowledge service matching. By incorporating "disappointment aversion" and iterative compromise, the model behaves more like real-world human decision-making than traditional static models. Future work will focus on calibrating these psychological parameters using empirical data from real-world digital platforms to further refine the model's accuracy.
Summary updated