Fusing Image and Text Features for Scene Sentiment Analysis Using Whale-Honey Badger Optimization Algorithm (WHBOA)

Prem Shanker Yadav, Dinesh Kumar Tyagi, Santosh Kumar Vipparthi

Abstract: Developing a real-time sentiment analysis application that relies solely on features extracted from images or textual content falls short of capturing human emotions’ nuanced and multifaceted nature. The unlabeled dataset, though useful, has limitations for sentiment analysis due to its general image descriptions, which lack emotional depth and do not include direct sentiment labels. Finding scene sentiment is a challenging task. To address this, combining textual descriptions with visual features is crucial. Important parameters include entropy, bag of words, and parts of speech (nouns, adjectives, and verbs) for textual analysis, alongside visual features like SIFT, SURF, and color histograms. These features are integrated to capture a comprehensive range of sentiment cues, enhancing the accuracy and depth of sentiment insights. This paper proposes an optimized adaptive neuro-fuzzy inference system for a compelling feature enhancement using the Whale-Honey Badger Optimization Algorithm (WHBOA). The proposed method identifies the most relevant and effective features from both textual and visual data. It captures visual-specific attributes to provide a richer and more detailed representation of visual content, addressing the limitations of general image descriptions and paving the way for the development of predictive models. Additionally, text pre-processing cleans and normalizes the textual data. We conducted an extensive comparative performance evaluation to assess the effectiveness and accuracy of the proposed model. The model is compared with the Nearest Neighbor, Support Vector Machine (SVM), and Decision Tree classification algorithms for the performance assessments.The results demonstrate that the optimized model performs better, achieving an accuracy of approximately 91.2%, compared to the other models.