Unsupervised Text Summarization for Abstract-Based Retrieval

Abstract

The efficient retrieval of pertinent content from expansive textual data stores has become increasingly indispensable due to the widespread availability of information. The goal of this paper is to propose a text summarization framework that is focused on enhancing retrieval-based document summarization techniques. Unstructured data is rapidly increasing in the online environment, and label-free summarization methods are needed to enhance retrieval. Deep learning It uses document preprocessing, which consists of segmenting sentences into tokens, embedding using models such as SBERT, and semantic modeling of sentences. These inter-sentence relations are then mapped as a similarity graph, and the graph-based ranking algorithm is used to rank sentences by their significance. Salient sentences are then chosen, and extracted sentences are organized to create a summary that provides key information about the document. Using this technique the retrieve engine obtains user queries based on abstracted summaries rather than full documents, thereby cutting costs of processing power but enhancing exactness. In addition, the model training does not consist of training datasets, which renders the approach domain-agnostic. The experiments carried out demonstrated that abstract-based retrieval through unsupervised multi-document summaries had better relevance and ranking in comparison with the conventional methods. The framework offers a realistic and scalable information retrieval method to intelligent active searching of a text-rich environment when limited labeled information is available.