Deep Learning Based Transformers for Keyword Extraction

The task of Keyword extraction is a natural language processing technique that involves automatically identifying the most relevant words or phrases in a piece of text such as web pages, emails, social media posts, instant messaging conversations…. In order to identify the main topics or themes in the text, this can be useful for tasks such as document classification, information retrieval, and summarization. In this paper, we present a novel deep learning approach to extract keywords from Arabic documents based on transformers. In Our approach we used prompt engineering; we guide the model towards generating the desired output. We demonstrate how to better interpret the inner workings of these models, which can provide insights into how to control their output more effectively. To evaluate the proposed approach, an experiment was conducted by comparing the extracted keywords with those manually selected. The results showed that the proposed approach achieved an average precision of 0.69. The proposed approach shows promise, but further research may be needed to improve its precision.