A Domain-Knowledge and Data Corporately Driven Based Dynamic Cache Prefetching Method for High Frequency Behavior Prediction

Cache prefetching methods have been extensively studied and applied due to their effectiveness in enhancing the computational efficiency of software systems. However, traditional cache prefetching approaches are limited in performance because they fail to leverage domain-specific knowledge. To address this issue, this paper proposes a domain-knowledge and data corporately driven dynamic cache prefetching method. The proposed method first utilizes an autoencoder to autonomously learn implicit features from historical sales data, thereby obtaining their high-dimensional representations. Subsequently, a domain knowledge graph is constructed to integrate various sales behaviors with the aforementioned representations, enabling the learning of a behavioral influence model. Finally, the influence of each sales behavior is jointly learned with the sales representations to train a behavior classifier, which predicts the characteristics of each sales behavior and generates or updates the cache accordingly. Furthermore, we instantiate this method to a real-time scenario in the domain of civil aviation e-commerce ticketing and design the core data structures and functions of this prototype.