Given the immense volume of textual content being generated on the Internet, encompassing a wide array of materials such as news stories, legal documents, and academic papers, the process of document summarization is becoming increasingly crucial. With the overwhelming amount of text, manual summarization becomes impractical due to the substantial effort, time, and monetary resources it demands. Multi-document summarization (MDS) emerges as a potent solution for aggregating information, offering concise and informative summaries from clusters of interrelated documents. In this context, our pioneering survey systematically reviews the recent advancements in deep learning-based MDS models. We introduce a new taxonomy to efficiently classify the various approaches taken in neural network construction, thereby offering a survey of the present state of the art. At its core, our work consists of introducing a novel deep-learning-based method for extracting summarization of multi-document information with an emphasis on opinions. To measure semantic sentence similarities, we propose an innovative combined approach that leverages determinant point processes. Through the use of filters of varying sizes, a convolutional layer analyses pairs of sentences, capturing foundational attributes. To sift through complex details and eliminate extraneous data, we harness capsule networks, which transform spatial and orientation relationships into sophisticated representations. Additionally, we employ RNN-LSTM for reconstructing sentence pairs, introducing a reconstruction loss to enhance our overarching objective function. The culmination of our efforts is the Opinion Summarizer Model, which achieves remarkable performance, as indicated by the highest AR metric on the dataset (ARS: 0.2548). In essence, our technique significantly surpasses existing state-of-the-art summarization methods like RegManual, PriorSum, and more. This underscores the efficacy of our approach in generating superior summaries that encapsulate the essence of opinion-oriented multi-document content.