Advancing Explainable AI in Text Summarization Through Integration With Topic Modelling

Abstract

The thesis investigates the potential of using Explainable Artificial Intelligence
(XAI) to efficiently develop a text summarization technique that can help resolve problems of accuracy and transparency in the automated text summarization task. The general concept is to create a method that can not only produce brief and informative summaries but is also interpretable in terms of how it works.
For this purpose, a research study proposes a hybrid architecture, by integrating Bidirectional Long Short-Term Memory (BiLSTM) with Gated Recurrent Units (GRU) neural network with a Multi-head Attention Mechanism. This approach enables the system to correctly identify complex semantic relations or text dependencies. In addition to the architectural approach, topic modeling techniques are employed to identify and extract the main themes from lengthy texts. This is an essential step in the summarization process to ensure that the produced summaries are coherent and reflect the main themes of the concept.
The innovations of attention mechanism are additionally deepened by employing Contextual Emphasis Index (CEI) and Enhanced Significance Spotlight (ESS) which enable it to assess the significance of the words or phrases more precisely. The act of summarizing being given more weight in the process of decision, thus making the system more focused also more explainable.
The performance of the proposed system has been quantitatively measured using the standard metrics like ROUGE to determine the degree of its capability for content extraction. The incorporation of XAI techniques with topic modeling has been proven to be significantly effective in generating top-quality summaries as well as making the whole process clear.
Offering a user, disoriented text summarization solution that effectively balances the employment of complicated computational methods and interpretability, it enhances the still-developing concept of explainable AI. It is of tremendous practical value in areas that handle a massive number of documents, such as the legal and academic sectors, where accountability, precision, and simplicity are basically the prerequisites for making good and well-informed decisions.