In the field of sentiment analysis, understanding the complex range of human emotions from product reviews presents a formidable challenge, especially when considering the multimodal nature of contemporary datasets. This paper introduces a new approach to emotion detection by using the synergistic potential of text and images through an attention-based multimodal system. Our method employs a Bidirectional Long Short-Term Memory (BiLSTM) model combined with an attention mechanism to intricately analyze textual data alongside multiple Convolutional Neural Networks (CNNs) to process image data, effectively capturing the emotional undertones conveyed through visual content. LoRA (Low-Rank Adaptation) is used to minimize the huge parameter computations on the image and text concatenation. Uniquely, this study focuses on a dataset curated from Amazon reviews, a domain where little or no prior research has been done to detect emotion from multimedia data. By thoroughly compiling this dataset, a significant resource gap in this field could be lessened. Our results demonstrate that the proposed attention-based BiLSTM and CNN framework significantly outperforms existing models, offering deeper insights into consumer emotions beyond the traditional binary sentiment classification. This advancement underscores the importance of integrating multiple data modalities for a comprehensive understanding of consumer feedback and opens new avenues for research in emotion detection within product reviews.

As internet use in communication networks has grown, fake news has become a big problem. The misleading heading of the news loses the trust of the reader. Many techniques have emerged, but they fail because fraudsters or exploiters find new ways to deceive them. Semantic analysis and machine learning techniques play a significant role in fake news detection. We must semantically assess terms used in the headline and main content before filtration because words can have different meanings in different contexts. In the paper, a method for determining fake news is introduced by calculating the dissimilarity between the title and content of the news. Vector distance calculators are used to extract semantic dissimilarities, which were then utilized as an additional feature. Initially, Term frequency-inverse document frequency (Tf-Idf) and Mutual Information (MI) are employed for text Feature Extraction (FE) on the ‘title’ and ’content’ of the news articles. Subsequently, four different vector distance calculators are used to extract vector distance-based features. The resulting distance values are used to train various machine learning classifiers, achieving the highest accuracy of 99%. Our method provides a comprehensive analysis by capturing diverse aspects of semantic dissimilarity from various distance calculators. The proposed method is then compared with previous techniques to demonstrate its effectiveness.

The rise of fake news presents a critical challenge to societal stability, emphasizing the urgent need for efficient detection systems. This study introduces an innovative approach to identifying fake news by exploiting the semantic discrepancies between the titles and content of news articles. Our method involves first summarizing article contents using a feature selection (FS) technique, followed by calculating vector distances between the titles and summarized contents using Cosine similarity, Jaccard distance and Euclidean distance. These methods enable us to identify multiple semantic dissimilarity characteristics, resulting in a more thorough examination. The obtained distance values are combined and used as features in our fake news detection models. These composite features are used to train various machine learning (ML) and deep learning (DL) models across three distinct news article datasets. Our approach achieves an accuracy rate of nearly 99.9%. The results underscore the effectiveness of leveraging semantic differences between article titles and content, offering a robust alternative to methods that focus solely on individual textual components. This technique not only improves accuracy but also provides a scalable solution to combating fake news in digital media.

Suicide is a significant public health issue that devastates individuals and society. Early warning systems are crucial in preventing suicide. The purpose of this research is to create a deep learning model to identify suicide risk using a combination of bidirectional encoder representations from transformers (BERT) and an ensemble of multiple convolutional neural networks (CNN). BERT is used to encode the text data into numerical representations that capture the context-aware meaning of words and phrases, while ensemble CNN is used to analyze the encoded text data to identify patterns and relationships relevant to suicide risk. The model is trained on a large corpus of text data from social media and suicide notes and evaluated on a validation set of labeled data. The results of the proposed model perform better than the recent approaches in detecting suicide risk and can potentially be a valuable tool for suicide prevention efforts.

With the exponential growth of e-commerce, understanding consumer sentiments from online product reviews has become crucial for businesses. This study presents a comprehensive approach to online product review sentiment analysis, encompassing text and image data. To facilitate our research, we compiled a novel dataset from the Amazon e-commerce platform, comprising paired textual reviews, corresponding images and sentiment. The work employs a multimodal deep-learning approach to capture the nuanced sentiment expressed through both text and images. Specifically, we integrate the Bidirectional Encoder Representations from Transformers (BERT) model, enhanced with a low-rank adaptation to refine its self-attention mechanism, for textual sentiment analysis and the VGG16 neural network for image feature extraction. Our experiments demonstrate that this multimodal approach surpasses unimodal techniques, achieving convincing results in sentiment classification.

As Internet technology progresses, users are increasingly opting for online shopping. Users can now not only purchase items but also offer feedback on their satisfaction with them. In terms of enhancing product quality, these reviews are incredibly valuable to both the user and the vendor. It becomes challenging to manually process a large amount of review data that is increasing in volume, subjectivity, and heterogeneity. To address this issue, machine learning (ML) technologies have been used in real-world applications, although there are still many hurdles. In order to achieve a more accurate result, this research study has used a mix of many fundamental machine learning techniques. Two-phase feature Extraction is used with chi square-based feature selection (FS) to reduce the dimension of the data. Three Ensamble classifiers were utilised, which merged several basic classifiers to choose the best outcome. The suggested model is tested using a dataset of 74000 product reviews from two well-known e-commerce sites, Flipkart and Amazon. And the model produces the most accurate results, with a 98% accuracy rate.