QuoteQuest: A Novel Approach for Image-Based Quote Recommendations

Abstract

The combination of text and images in online communication has become essential in the digital age, especially on social media platforms where user engagement is dominated by images. Though automated image captioning has advanced, there is still a lack of emotionally charged and contextually rich quotes to accompany images and improve their communicative impact. Through the automation of the quote recommendation process for images, this thesis aims to enhance the quality of digital communication by closing the gap between visual material and textual significance. In order to address this problem, the research presents a novel hybrid model that incorporates Natural Language Processing and Computer Vision in a synergistic way. The model decodes the context and visual content of images using sophisticated CV techniques. Simultaneously, cutting edge natural language processing algorithms produce and suggest quotes that are both emotionally and contextually relevant to the images. With the help of user feedback and expert consultation, prototypes are refined through iterative development, guaranteeing the model's applicability and efficacy in actual situations. With 99.95% accuracy, 99.42% precision, 99.47% recall, and a 99.44% F1 score, the keyword extraction model demonstrated its remarkable performance in identifying relevant keywords. Similar to this, the quote recommendation system performed well, receiving high marks for appropriateness (90.18%), creativity (94.18%), and relevance (89.09%), demonstrating its ability to provide interesting and social media-friendly quotes.