<p>Identifying the relevant information for a user’s query is a difficult task and is considered complex because contextual information may be spread across the document. Approaches, where documents are represented as a series of passages, have been used in the past to attempt to improve the performance of ad-hoc retrieval systems. In this paper, we present our approach of generating a graph from the top returned passages where each node in the graph corresponds to a passage (against a given query) and an edge weight represent a similarity between them. To form the graph, we generated only the edges between the nodes (passages) that belongs to the same document. The core idea of this strategy is to determine some indicator of the documents’ cohesiveness. A graph-based approaches is adopted to calculate the document cohesion. We discuss two different approaches of generating the cohesion score. At first, we only considered the passages within the same document (i.e. inter-passage connectivity) to formulate the graph and to measure the cohesion score. Later, we extend our approach by generating a bipartite graph where we also include query-passage similarity along with the inter-passage similarity to calculate the document cohesion. We compared the cohesion score of relevant (<i>R</i>) and non-relevant (<i>NR</i>) documents and illustrate that our passage-based graph approach produce different results for <i>R</i> and <i>NR</i> documents. Additionally, we re-ranked the search results and assessed the effectiveness of the system by combining the cohesion score with a document similarity score.</p>