ABSTRACT In today's data‐intensive client‐server systems, traditional caching methods often fail to meet the demands of modern applications, especially in mobile environments with unstable network conditions. This research addresses the challenge of improving data delivery by proposing an advanced prefetching framework that utilizes various embedding techniques. We explore how to model user navigation using graph‐based, autoencoder‐based, and sequence‐to‐sequence‐based embedding methods and assess their impact on prefetching accuracy and efficiency. Our study shows that utilizing these embedding techniques with supervised learning models improves prefetching performance. We also present a software architecture that blends supervised and unsupervised learning approaches, along with user‐specific and collective learning models, to create a robust prefetching mechanism. The contributions of this study include developing a scalable prefetching solution using machine learning/deep learning algorithms and providing an open‐source prototype of the proposed architecture. This paper offers a significant improvement over previous research and provides valuable insights for enhancing the performance of data‐intensive applications.

A key challenge faced in client–server systems that heavily rely on data is the fast delivery of data to end-users. To address this difficulty, this study presents a novel approach for modeling and forecasting user navigational browsing behavior, to establish an efficient prefetching mechanism. Regarding the approach for representing page visit data, we employ the Word2Vec embedding technique to encode each user’s page visit as a numerical vector. Regarding the encoding of browsing activity data, we utilize aggregation on the embedding vectors. These vectors correspond to page visits occurring sequentially and are used to describe each user’s browsing behavior using a numerical vector. In the proposed method, machine learning algorithms are employed to analyze and model the browsing behavior of all users. Machine learning models are employed to forecast the next user action during the navigation of data-intensive web and mobile application web pages. Subsequently, we employ this forecast to establish an intelligent prefetching method, which provides the capability of acquiring predicted web page data in proxy servers before it is requested. An experimental study was conducted using a large-scale open-source dataset derived from a mobile application used in a coffee shop, containing several hundred thousand sessions from tens of thousands of users over a 10-day period. The evaluation employed metrics such as prediction accuracy which can be called prefetching accuracy and cache hit/miss rates. The machine learning algorithms applied include K-Nearest Neighbor, AdaBoost, Decision Tree, Support Vector Machine, Multi-layer Perceptron, Random Forest, LightGBM, Long Short-Term Memory and Bidirectional Long Short-Term Memory. The purpose of this experimental study was to examine the efficacy of the proposed approach. The findings of the empirical investigation suggest that the proposed method has the potential to provide an efficient prefetching methodology when sufficient user navigational data is available. Hence, the method enhances the performance of data-intensive client–server-based systems.

Data-intensive Web Applications built using client–server architectures usually provide prefetching mechanisms to enhance data accessibility. Prefetching is a strategy of retrieving data before it is requested so that it can be ready when the user requests it. Prefetching reduces the load on the web server by making data available before the user requests it. Prefetching can be used for static content, such as images and web pages, as well as dynamic content, such as search results. Prefetching can also be used to improve the performance of web applications, as the data is available quickly. There are several scheduling methods, such as time-based scheduling, event-based scheduling and priority-based scheduling, for prefetching to ensure that essential data is always ready when the user requests it. In this study, we focus on time-based scheduling for prefetching. We introduce time-based scheduling methodologies using sequential pattern mining techniques and long-term short memory-based deep learning strategies. To show the usefulness of these strategies, we develop a prototype application. We conduct an extensive experimental study to evaluate the performance of the proposed time-based scheduling methodologies using both performance and accuracy metrics. Based on the computed metrics, using proposed prefetching methods provided a promising cache hit rate when using the optimal cache size. The results show that the proposed prefetching methodologies are useful in data-intensive web applications for enhancing data accessibility. Work remains to investigate the use of attention-based sequence-to-sequence models in the web prefetching domain.

Fast data delivery to users is a big challenge in client-server architecture-based data-intensive systems. Here, prefetching is a widely used technique to increase data-intensive mobile or web applications’ operational performance and data delivery speed. This study proposes a methodology that can address this challenge using deep-learning-enabled prefetching approaches. The proposed methodology minimizes the data-access latency by adopting an approach that models the customer navigational browsing data and predicts the following user actions. The proposed approach utilizes two different recurrent neural network methods to model the clickstream data. These methods include LSTM and bi-directional LSTM. To show the usability of the proposed methodology, we provide a prototype implementation. To this end, we use a public dataset obtained from log files of a coffee store mobile application. In the prototype implementation, because the small number of users accounted for the majority of the load on the system, we segmented users as active and cold users. To facilitate testing of the prototype implementation, we conducted an experimental study. Here, we investigated whether the system can predict users’ near-future requests. In the experimental study, we record the cache hit rates. The results show that the proposed prefetching is promising and can be utilized in client-server-based data-intensive applications.