The purpose of this paper is to present a set of algorithmic improvements upon an extension of the Set-Based Model (Kalogeropoulos et al (2020)) that focuses on the dependence among document terms employing graph representations. A graph-based approach to document representation can positively affect the information retrieval process due to the ability of graphs to capture the syntactic notion and, in some cases, the semantic relationship among document terms. The aforementioned model generates complete graphs; thus, each document term will be interdependent with the rest. Consequently, an interdependent segment or multiple parts of a document called a window is defined, in which the graph creation algorithms are applied. The proposed methods aim to approximate the window size by exploiting the document length. Moreover, an attempt to create multiple windows is made, considering the relationship between a sentence and a paragraph, which is reflected in the semantic importance of nodes and edges. An attempt to tackle the stop-word detection problem on bridge nodes is made by implementing algorithmic schemes that use core decomposition (Seid-man (1983a)) to identify the importance of such nodes in a sample of the corpus collection. Finally, a simple reranking scheme and an ensemble voting technique are implemented to enhance model performance on queries where the proposed approach lacks performance. The experimental analysis made on multiple document collections exhibits performance improvements and in some queries, our 1 approach outperforms even state-of-the-art models such as BM25 (Robertson et al (2004)) and ColBERT (Khattab and Zaharia (2020)).