Multi-document summarization involves extracting key information from a collection of documents and discarding unnecessary details to generate a short and coherent summary. This method provides users with a streamlined representation of large volumes of information, significantly saving time and effort. This paper presents a BERT and rank-based bidirectional long short-term memory (BiLSTM) model to achieve multi-document summarization. A key contribution of this work is the introduction of step optimization, a technique that significantly improves the efficiency of the classifier by minimizing computational overhead and fine-tuning hyperparameters in a more targeted way. This approach processes multiple documents and generates a single coherent summary using techniques such as Term Frequency-Inverse Document Frequency (TF-IDF) and weighted graph embeddings. By applying these methods, the model captures the most relevant information, reduces redundancy, and improves the quality of summaries generated from diverse document collections.