Developing a Lightweight Summarization Algorithm for Arabic Text Using Federated Learning

Despite the significant advancements in Large Language Models (LLMs) and their ability to summarize, translate, and accurately answer questions, they all rely on powerful servers, as they must be trained on millions of parameters and large datasets. However, in cases where privacy and performance are critical, these systems cannot always be relied upon. In these situations, a lightweight algorithm is needed to provide certain services, such as automatic summarization. This research introduces a fast algorithm for document summarization without requiring extensive training on large datasets. These services help in reducing the required time and effort to extract key information from texts, articles, and research papers. Moreover, the research proposes a collaboration framework to enhance the accuracy without affecting privacy based on federated learning. The implementation demonstrates the efficiency of the proposed algorithm, achieving accuracy close to 95%. A higher rate can be attained in future if the proposed framework of the federated learning is applied.