In this study, we investigate the role of visual content in accurately predicting movie genres. By extracting keyframes from Hindi, Bengali, Malayalam, and Telugu language-based Indian movie trailers in the Flickscore dataset, we analyse visual elements using visual language model (VLM) and Large Language Models (LLMs). Our approach focuses on the FAMOS aspects (focus, action, mood, object, setting) to understand the movie’s theme, summary, and genre. This method eliminates the need to manually prepare textual metadata, thus reducing the time and effort required for genre identification. The visual features captured from the keyframes are leveraged to predict genres more efficiently, offering a reliable way to automate this process. Additionally, we integrate the visual information into a content-based filtering (CBF) system to predict user preferences. Our study highlights the effectiveness of using visual features, which significantly enhances the performance of recommendation systems (RS) by improving accuracy in predicting user preferences based on movie genres. We demonstrate that by analysing the implicit content in movie frames, we can achieve better insights than traditional metadata-driven approaches. Overall, our findings emphasise that visual content holds valuable information for understanding movie genres and can play a critical role in user preference understanding.

RNA plays a crucial role in regulating gene expression, thereby ensuring the maintenance of genetic integrity. RNA can fold into various structures based on alternative intra-molecular base-pairings and plays a critical role in the production of functional proteins. Many experimental and computational techniques have revealed the secondary and tertiary structures of RNA molecules. With the increase in RNA data, structure predictions have evolved from conventional to advanced computational methods. Various bioinformatics methods have been developed to predict RNA structures to understand underlying molecular mechanisms. This review summarizes multiple methodologies for RNA structure prediction, encompassing biophysical techniques, probing methods, and computational approaches. These computational approaches include free energy minimization, comparative sequence analysis, deep learning algorithms, and hybrid methods. Since the current era is dedicated to deep learning techniques, the present review highlights the significance of these methods to provide better insights into RNA structures that can be further explored to discover novel therapeutic drug targets for diseases.

Many people have been severely affected by the COVID-19 pandemic, which has caused intense anxiety, fear, and complex feelings or emotions. People’s emotions have changed and become more complicated since coronavirus vaccinations were introduced. Sentiment analysis of COVID-19 vaccination is critical for understanding public perception, vaccine hesitancy, monitoring vaccine impact, identifying adverse reactions, making policies, and allocating resources. Some artificial intelligence (AI)-based systems have been reported in the literature to analyze the sentiment of COVID-19 vaccination. However, most of them are end-to-end models that require explanation for their results in identifying COVID-19 vaccination sentiment. An explainable AI-based model can improve decision-making, transparency, and interpretability. It can enable users to comprehend how the model makes predictions and the factors that influence the outcome. Therefore, this study suggests a COVID-Twitter-BERT and LSTM (CT-BERT-LSTM) staking that is explicable for determining people’s views on the COVID-19 vaccination. The prediction of the proposed CT-BERT-LSTM model is then examined to determine where the suggested system successfully learned the context of the tweet and where it failed to do so. The proposed CT-BERT-LSTM model performed exceptionally well and outperformed the existing models with aF1-score of 0.88 in the sentiment identification of the COVID-19 vaccination.

In countries where there is a low availability of resources for language, businesses face the challenge of overcoming language barriers to reach their customers. One possible solution is to use collaborative filtering-based recommendation systems in their native languages. These systems employ algorithms that understand the customers’ preferences and suggest products or services in their native language. Collaborative filtering (CF) is a popular recommendation technique that simulates word-of-mouth phenomena. However, the accuracy of a CF recommendation can be affected by sparse data. In this research article, we present a novel hybrid weighted multi-deep ranking supervised hashing (HWMDRH) approach. Our method leverages both user-based and item-based CF by merging the item-based deep ranking weighted multi-hash recommender system prediction with the user-based deep ranking weighted multi-hash recommender system prediction to generate Top-N prediction. We conducted extensive experiments on the MovieLens 1M dataset, and our results show that the proposed HWMDRH model outperforms existing models and achieves state-of-the-art performance across recall, precision, RMSE, and F1-score metrics.

Twitter is increasingly being used during disasters to communicate with authorities, ascertain the ground reality, and coordinate real-time rescue and recovery activities. Geographical location information about users and events is critical in these scenarios. Geotagged tweets are extremely infrequent, and other location fields, such as user location and place name, are unreliable. The extraction of geographical information from tweet text is limited by the fact that individuals frequently publish multilingual tweets that contain numerous grammatical and spelling errors, as well as nonstandard acronyms. As a result, determining the geographical location of the tweet is a challenging problem. This article presents a technique based on deep neural networks for extracting geographical references mentioned in bilingual tweets. Several deep learning-based models, including convolutional neural networks (CNNs), long short-term memory (LSTM), bidirectional LSTM (Bi-LSTM), and attention-based Bi-LSTM, are implemented on real-world English and Hindi language tweets to determine their suitability for extracting location references. The proposed CNN, along with a conditional random field at the last layer, is found to perform better than other models, with an$F_{1}$-score of 0.858. The findings of this study can aid in early event detection, pinpointing the area of devastation and victims, real-time traffic management, and a number of other location-based applications. The suggested system’s code and trained model may be obtained athttps://github.com/Abhinavkmr/Bi-lingual-location-reference-identification.git.

Hate speech on social media has become a big problem, making regular users very upset and giving victims depression and suicidal thoughts. Early identification of the user spreading this type of hate speech may be a better solution, allowing hate speech to be stopped at source. In this article, we attempt to identify these hate speech spreaders by finding a representation for each user. Each user’s comments are aggregated and fed to an auto-encoder to train it. The encoder part of the auto-encoder is used to get an encoded vector for each user. The encoded vector is used with different machine learning (ML) classifiers to determine if a user is spreading hate speech. The proposed model was tested using the dataset released by PAN 2021 (https://pan.webis.de/data.html) hate speech spreader profiling competition in English and Spanish. The experimental results show that support vector machine (SVM) with encoded vectors as features outperforms existing models with an accuracy of 92% for both English and Spanish dataset. The proposed features extraction technique is found to be equally effective at identifying fake news spreaders on fake news datasets provided by PAN 2020 yielding accuracy values of 95% and 83% for English and Spanish, respectively.

Protein function prediction is a major challenge in the field of bioinformatics which aims at predicting the functions performed by a known protein. Many protein data forms like protein sequences, protein structures, protein-protein interaction networks, and micro-array data representations are being used to predict functions. During the past few decades, abundant protein sequence data has been generated using high throughput techniques making them a suitable candidate for predicting protein functions using deep learning techniques. Many such advanced techniques have been proposed so far. It becomes necessary to comprehend all these works in a survey to provide a systematic view of all the techniques along with the chronology in which the techniques have advanced. This survey provides comprehensive details of the latest methodologies, their pros and cons as well as predictive accuracy, and a new direction in terms of interpretability of the predictive models needed to be ventured by protein function prediction systems.

After the development of next-generation sequencing techniques, protein sequences are abundantly available. Determining the functional characteristics of these proteins is costly and time-consuming. The gap between the number of protein sequences and their corresponding functions is continuously increasing. Advanced machine-learning methods have stepped up to fill this gap. In this work, an advanced deep-learning-based approach is proposed for protein function prediction using protein sequences. A set of autoencoders is trained in a semi-supervised manner with protein sequences. Each autoencoder corresponds to a single protein function only. In particular, 932 autoencoders corresponding to 932 biological processes and 585 autoencoders corresponding to 585 molecular functions are trained separately. Reconstruction losses of each protein sample for every autoencoder are used as a feature to classify these sequences into their corresponding functions. The proposed model is tested on test protein samples and achieves promising results. This method can be easily extended to predict any number of functions having an ample amount of supporting protein sequences. All relevant codes, data and trained models are available at https://github.com/richadhanuka/PFP-Autoencoders.

Wireless sensor network (WSN) is formed by a large number of low-cost sensors. In order to exchange information, sensor nodes communicate in an ad hoc manner. The acquired information is useful only when the location of sensors is known. To use GPS-aided devices in each sensor makes sensors more costly and energy hungry. Hence, finding the location of nodes in WSNs becomes a major issue. In this paper, the authors propose a combination of range based and range-free localization scheme. In their scheme, for finding the distance, they use received signal strength indication (RSSI), which is a range based center of gravity technique. For finding the location of non-anchor nodes, the authors assign weights to anchor and non-anchor nodes based on received signal strength. The weight, which is assigned to anchor and non-anchor nodes, are designed by fuzzy logic system (FLS).