Content-based Features Research Articles

Interactive Guide Assignment System with Destination Recommendation and Built-in Chatbox

This proposed system provides a website called 'Safari Nepal', where users can search for destinations and check their location on a map. Users when registering on the website, can fill up the details about themselves and choose to either be a tour guide or a tourist. Based on the search and preferences of the user, similar destinations are recommended to the user via a recommendation system that uses a content-based recommendation feature. This feature works on the data obtained from the user, either explicitly or implicitly. The concept of K-Nearest Neighbours (KNN) and Cosine similarity makes the recommendation more accurate. KNN uses a distance algorithm that sorts from most liked destinations to least liked, based on the preferences of the user. This sorted list of destinations is further filtered by Cosine similarity, which is a measure of how similar two vectors in an inner product space are. It is calculated by taking the cosine of the angle between two vectors and determining whether two vectors are pointing towards the same general direction. Thus, combined KNN and Cosine similarity gives a better recommendation to the user. The map is integrated into the system using Mapbox API. Also, the system connects users with tour guides and gives them space to chat via a chatbox called ‘Travel Buddy’ where they can discuss further the destination, the amount charged by the guide, etc. The chatting feature on the system allows multiple users to connect and make conversations about the destination creating various chatrooms. In the system, the user can also publish their blogs describing their experiences and share their thoughts on particular destinations.

Attention-Based Personalized Compatibility Learning for Fashion Matching

The fashion industry has a critical need for fashion compatibility. Modeling compatibility is a challenging task that involves extracting (in)compatible features of pairs, obtaining compatible relationships between matching items, and applying them to personalized recommendation tasks. Measuring compatibility is a complex and subjective concept in general. The complexity is reflected in the fact that relationships between fashion items are determined by multiple matching rules, such as color, shape, and material. Each personal aesthetic style and fashion preference differs, adding subjectivity to the compatibility concept. As a result, personalized factors must be considered. Previous works mainly utilize a convolutional neural network to measure compatibility by extracting general features, but they ignore fine-grained compatibility features and only model overall compatibility. We propose a novel neural network framework called the Attention-based Personalized Compatibility Embedding Network (PCE-Net). It comprises two components: attention-based compatibility embedding modeling and attention-based personal preference modeling. In the second part, we utilize matrix factorization and content-based features to obtain user preferences. Both pieces are jointly trained using the BPR framework in an end-to-end method. Extensive experiments on the IQON3000 dataset demonstrate that PCE-Net significantly outperforms most baseline methods.

A deep learning approach to detect phishing websites using CNN for privacy protection

Nowadays, with the variety of internet frauds, every web user while browsing the net is vulnerable to being a target of various attacks. The phishing attack is one of the largest and most effective cyber threats; it is a sort of social engineering technique employed by web hackers, with the aim of deceiving users and stealing their credentials for financial gain. The continuous growth and the rising volume of phishing websites have led researchers to propose several anti-phishing solutions to fight against this cyber-attack such as visual similarity-based approaches, list-based approaches, machine learning, heuristics-based techniques … etc, moreover deep learning in recent years has gained increasing interest in several areas, especially in the phishing detection area. In this paper, we propose a deep learning approach to detect phishing websites using convolutional neural networks testing both 1D CNN & 2D CNN with three feature types, URL-based features, content-based features, and third-party services-based features. The experimental results show that 1D CNN is more adequate for phishing detection and it achieves a high accuracy value of 96.76%. Moreover, it reduces the training time compared to other deep learning-based works.

Retweet Prediction Based on User-Based, Content-Based, Time-Based Features Using ANN Classification Optimized with the Bat Algorithm

Twitter is one of the most popular social media platforms today for information dissemination. It is favored by the public due to its real-time information sharing capabilities. Twitter provides two important features for information dissemination: Tweets and Retweets. Tweets allow users to write messages that can be instantly shared. Each tweet can contain text, media such as images, videos, or URLs. Retweets allow users to repost someone else's tweet and distribute it to their own followers. The Retweet feature is considered an effective way to spread information, as a high number of retweets indicates that the information in the tweet is spreading quickly and widely. This research aims to predict retweets based on several features: User-Based Feature, Content-Based Feature, and Time-Based Feature. The classification method used is Artificial Neural Network, which is optimized using a Nature-Inspired Algorithm called Bat Algorithm. The evaluation results of this study show an accuracy of 86%, precision of 87.8%, recall of 93.6%, and F1-score of 90.6% without imbalance class handling. Under Undersampling condition, the accuracy is 80.8%, precision is 91.0%, recall is 81.4%, and F1-score is 85.9%. Under Oversampling condition, the accuracy is 82.4%, precision is 89.6%, recall is 85.6%, and F1-score is 87.5%. These results indicate that using user-based, content-based, and time-based features, applying Artificial Neural Network classification method, and optimizing hyperparameters using Bat Algorithm are effective in predicting retweets.

Segmentation of Medical Images with Adaptable Multifunctional Discretization Bayesian Neural Networks and Gaussian Operation

Bayesian statistics is incorporated into a neural network to create a Bayesian neural network (BNN) that adds posterior inference aims at preventing overfitting. BNNs are frequently used in medical image segmentation because they provide a stochastic viewpoint of segmentation approaches by producing a posterior probability with conventional limitations and allowing the depiction of uncertainty over following distributions. However, the actual efficacy of BNNs is constrained by the difficulty in selecting expressive discretization and accepting suitable following disseminations in a higher-order domain. Functional discretization BNN using Gaussian processes (GPs) that analyze medical image segmentation is proposed in this paper. Here, a discretization inference has been assumed in the functional domain by considering the former and dynamic consequent distributions to be GPs. An upsampling operator that utilizes a content-based feature extraction has been proposed. This is an adaptive method for extracting features after feature mapping is used in conjunction with the functional evidence lower bound and weights. This results in a loss-aware segmentation network that achieves an F1-score of 91.54%, accuracy of 90.24%, specificity of 88.54%, and precision of 80.24%.

Web-Informed-Augmented Fake News Detection Model Using Stacked Layers of Convolutional Neural Network and Deep Autoencoder

Today, fake news is a growing concern due to its devastating impacts on communities. The rise of social media, which many users consider the main source of news, has exacerbated this issue because individuals can easily disseminate fake news more quickly and inexpensive with fewer checks and filters than traditional news media. Numerous approaches have been explored to automate the detection and prevent the spread of fake news. However, achieving accurate detection requires addressing two crucial aspects: obtaining the representative features of effective news and designing an appropriate model. Most of the existing solutions rely solely on content-based features that are insufficient and overlapping. Moreover, most of the models used for classification are constructed with the concept of a dense features vector unsuitable for short news sentences. To address this problem, this study proposed a Web-Informed-Augmented Fake News Detection Model using Stacked Layers of Convolutional Neural Network and Deep Autoencoder called ICNN-AEN-DM. The augmented information is gathered from web searches from trusted sources to either support or reject the claims in the news content. Then staked layers of CNN with a deep autoencoder were constructed to train a probabilistic deep learning-base classifier. The probabilistic outputs of the stacked layers were used to train decision-making by staking multilayer perceptron (MLP) layers to the probabilistic deep learning layers. The results based on extensive experiments challenging datasets show that the proposed model performs better than the related work models. It achieves 26.6% and 8% improvement in detection accuracy and overall detection performance, respectively. Such achievements are promising for reducing the negative impacts of fake news on communities.

Malicious URL Detection Using Machine Learning

Throughout the years, internet usage has grown significantly. The internet continues to transform how we interact with people, organise the flow of goods, and communicate knowledge all across the world. The attackers have used this popularity to their advantage to participate in illegal activities that would lead to monetary advantage. There has been a rise in malicious websites that launch client-side attacks over time, which cannot be identified effectively by existing approaches such as blacklisting. As a result, an efficient solution to detect these malicious websites is required. In this study, we used the random forest method to develop a machine learning model while integrating lexical features, hostbased features, and content-based features. The model has an accuracy of 94.7%

Trimap-guided feature mining and fusion network for natural image matting

Mining interesting objects with trimap guidance and fusing multi-level features are two important issues for trimap-based matting. For feature mining, existing methods simply feed trimaps to an encoder network and mine features evenly in different regions with a decoding module, which lacks an efficient and effective way to mine interesting objects pointed out by trimaps. For emerging content-based feature fusion, most existing matting methods only focus on local features which lack the guidance of a global feature with strong semantic information related to the interesting object. In this paper, we propose a trimap-guided feature mining and fusion network consisting of our trimap-guided non-background multi-scale pooling (TMP) module and global-local context-aware fusion (GLF) modules. Considering that trimap provides strong semantic guidance, our TMP module focuses effective feature mining on interesting objects under the guidance of trimap without extra parameters. Furthermore, our GLF modules use global semantic information of interesting objects mined by our TMP module to guide an effective global-local context-aware multi-level feature fusion. In addition, we build a common interesting object matting (CIOM) dataset to advance high-quality image matting. Experimental results on the Composition-1k test set, Alphamatting benchmark, and our CIOM test set demonstrate that our method outperforms state-of-the-art approaches. Codes and models will be publicly available at https://github.com/Serge-weihao/TMF-Matting.

Generative Adversarial Networks Classifier Optimized with Water Strider Algorithm for Fake Tweets Detection

In this manuscript, a deep learning-based Generative Adversarial Networks (GAN) classifier optimized with the Water Strider Optimization Algorithm (WSOA) is proposed to solve the issue of fake tweets in Twitter. Initially, input Twitter data is taken from the “Twitter API” dataset. Then, pre-processing is done in the Twitter dataset. After that, the pre-processed tweet features are selected based on the Bag of Words model, mutual information, and the Chi-square method. Then the selected features are given into the Generative Adversarial Networks classifier for detecting fake tweets. Here, the Generative Adversarial Networks classifier is used to fake tweets detection. But it does not expose any optimization strategies adopted to determine the optimum parameters to confirm the accurate categorization of fake tweets. Therefore, Water Strider Optimization Algorithm is applied to optimize the parameter in Generative Adversarial Networks classifier (FT-GAN-WSOAC). The proposed method is activated in MATLAB and the performance is analyzed with performance metrics. The experimental results show that the proposed FT-GAN-WSOA classifier provides higher accuracy of 25.14%, 25.21%, 20.44%, and 34.38% than the existing methods, such as fake tweet detection with content-based features using a custom rule-based algorithm (FT-CBF-CRBA) method, fake tweet detection with long/short-term memory using a random forest algorithm (FT-LSTM-RFA) method, fake tweet detection with zero-shot learning using bidirectional encoder representations from transformers (FT-ZSL-BERT) method, and fake tweet detection with a hybrid convolution neural network and recurrent neural network with long/short-term memory (FT-Hyb CNN-RNN-LSTM) method, respectively.

“Give me happy pop songs in C major and with a fast tempo”: A vocal assistant for content-based queries to online music repositories

This paper presents an Internet of Musical Things system devised to support recreational music-making, improvisation, composition, and music learning via vocal queries to an online music repository. The system involves a commercial voice-based interface and the Jamendo cloud-based repository of Creative Commons music content. Thanks to the system the user can query the Jamendo music repository by six content-based features and each combination thereof: mood, genre, tempo, chords, key and tuning. Such queries differ from the conventional methods for music retrieval, which are based on the piece’s title and the artist’s name. These features were identified following a survey with 112 musicians, which preliminary validated the concept underlying the proposed system. A user study with 20 musicians showed that the system was deemed usable, able to provide a satisfactory user experience, and useful in a variety of musical activities. Differences in the participants’ needs were identified, which highlighted the need for personalization mechanisms based on the expertise level of the user. Importantly, the system was seen as a concrete solution to physical encumbrances that arise from the concurrent use of the instrument and devices providing interactive media resources. Finally, the system offers benefits to visually-impaired musicians.

Link and edge weight prediction in air transport networks — An RNN approach

Predicting the future structure of air transport networks is important for several stakeholders in terms of e.g., access to markets, prospects for economic integration and development of regions. Link and edge weight prediction aims to foretell whether two airports will be connected by a direct flight in a future stage of the development of a network and the frequency with which services will be provided. This work assesses the capacity of popular similarity-based algorithms to predict network evolution in air transport. It also proposes a supervised recurrent neural network-based learning framework (RNN) for link prediction. It draws on a set of topological, temporal and content-based features. Experimental results from network data that maps the European Air Transport Network in the period between 2010 and 2019 show that similarity-based algorithms are not able to predict future network stages well. Their performance in predicting newly emerging links remains below expectations formulated in the earlier link prediction literature. The proposed RNN framework outperforms traditional similarity algorithms by a substantial margin. However, the results suggest that link and edge weight prediction remain challenging in sparse air transport networks. Predictive performance must be optimised even further before forecasts can be used to inform concrete policy decisions.

Deep Neural Network for Detecting Fake Profiles in Social Networks

This paper proposes a deep neural network (DNN) approach for detecting fake profiles in social networks. The DNN model is trained on a large dataset of real and fake profiles and is designed to learn complex features and patterns that distinguish between the two types of profiles. In addition, the present research aims to determine the minimum set of profile data required for recognizing fake profiles on Facebook and propose the deep convolutional neural network method for fake accounts detection on social networks, which has been developed using 16 features based on content-based and profile-based features. The results demonstrated that the proposed method could detect fake profiles with an accuracy of 99.4%, equivalent to the achieved findings based on bigger data sets and more extensive profile information. The results were obtained with the minimum available profile data. In addition, in comparison with the other methods that use the same amount and kind of data, the proposed deep neural network gives an increase in accuracy of roughly 14%. The proposed model outperforms existing methods, achieving high accuracy and F1 score in identifying fake profiles. The associated findings indicate that the proposed model attained an average accuracy of 99% while considering two distinct scenarios: one with a single theme and another with a miscellaneous one. The results demonstrate the potential of DNNs in addressing the challenging problem of detecting fake profiles, which has significant implications for maintaining the authenticity and trustworthiness of online social networks.

Language Variety Prediction using Word Embeddings and Machine Leaning Algorithms

Abstract: Author Profiling is a technique of predicting demographic characteristics like gender, age, location, nativity language, educational background etc., of an author by analysing their written texts. Author profiling is used in several text processing applications like forensics analysis, marketing, security. The author profiling techniques identify the stylistic differences among the author writing styles to identify the demographics of authors. Researchers experimented with various stylistic features like lexical features, content-based features, syntactic features, semantic features, domain specific features, structural features, readability features etc., to identify the stylistic differences among different author’s texts. The dataset plays an important role to analyse the stylistic differences of authors. PAN is one competition organizes different types of tasks in every year to encourage the participants around the globe for providing solutions to different types of text classification problems like plagiarism detection, authorship attribution, authorship verification, authorship profiling, celebrity profiling, style change detection, fake news spreaders detection, hate speech spreaders detection etc. The author profiling task was introduced in 2013 by the organizers of PAN competition. The organizers carefully gather the datasets and make available to the researchers for providing solutions to the problems. Every year the organizers conduct competitions on different sub-tasks of author profiling and provides datasets in different languages and in different genres. In 2017 competition, PAN introduces a task of predicting the language variety of an author. They release the dataset in four languages. In this work, we proposed an approach for English language dataset of language variety prediction. The proposed approach used the word embeddings generated by the Word2Vec model and BERT (Bidirectional Encoder Representations from Transformers) model. The word embeddings are used for generating the document vectors by combining the word embeddings of words those contain in documents. The document vectors are trained with two machine learning algorithms such as support vector machine and random forest. The Random Forest attained best accuracy of 96.87 for language variety prediction when experiment conducted with BERT embeddings

A comparison of misinformation feature effectiveness across issues and time on Chinese social media

Misinformation on social media is a nonnegligible phenomenon that causes successive adverse impacts. Numerous scholarly efforts have been devoted to automatic misinformation detection to address this problem. The effective feature is the key to achieving high identification performance. However, the effectiveness of the feature may change in different issues and time considering the manifold social contextual reasons. Most extant literature on misinformation detection does not differentiate between topics, issues or domains. Although some research compares detection across domains, they concentrate on the model's overall performance, neglecting the effectiveness of individual features. Furthermore, the comparison studies mainly incorporate single-domain issues rather than issues that cover multiple domains. It is still difficult to determine which domain's misinformation characteristics will match those of multi-dimensional issues. Since the misinformation nowadays covers multiple domains, finding robust features in misinformation detection over issues and time is an urgent research agenda. In this study, we collected datasets of two issues, climate change and genetically modified organisms (GMOs), between January 1st, 2010 and December 31st, 2020 on Weibo, manually annotated the veracity status of the posts, and compared the performance of the proposed features in identifying misinformation by applying logistic regression. The results demonstrate that (1) the predicting power of content-based features, including topic and sentiment, is relatively robust compared to user-based and propagation-based features across issues and time. (2) The feature effectiveness varied at different time points. Our findings imply that future research could consider focusing more on content-based features, especially implicit features from the content in misinformation detection. Moreover, researchers should evaluate the feature effectiveness at different time stages to improve the efficiency of misinformation detection.

HELPHED: Hybrid Ensemble Learning PHishing Email Detection

Phishing email attack is a dominant cyber-criminal strategy for decades. Despite its longevity, it has evolved during the COVID-19 pandemic, indicating that adversaries exploit critical situations to lure victims. Plenty of detectors have been proposed over the years, which mainly focus on the contents or the textual information of emails; however, to cope with the evolution of phishing emails more sophisticated approaches should be introduced that will exploit all the emails’ traits to enhance the detection capability of Machine Learning/Deep Learning classifiers. To tackle the limitations of existing works, this paper proposes a phishing email detection methodology, named HELPHED that focuses on the detection of phishing emails by combining Ensemble Learning methods with hybrid features. The hybrid features provide an accurate representation of emails by fusing their content and textual traits. We propose two methods of HELPHED, the first one employs the Stacking Ensemble Learning method, while the second method utilizes the Soft Voting Ensemble Learning. Both methods deploy two different Machine Learning algorithms to handle the hybrid features separately, yet in parallel, minimizing the features’ complexity and improving the model’s performance. A thorough evaluation analysis is carried out considering innovative guidelines that aim to prevent partial and misleading results. Experimental tests verified that the combination of hybrid features with Ensemble Learning, overall, accomplishes better detection performance than when employing only content-based or text-based features. Numerical results on a rich imbalanced dataset (i.e., 32,051 benign and 3,460 phishing email samples) that considers the evolution of phishing emails show that Soft Voting Ensemble Learning outperforms other prominent Machine Learning/Deep Learning algorithms and existing works yielding F1-score equal to 0.9942.

The spike gating flow: A hierarchical structure-based spiking neural network for online gesture recognition.

Action recognition is an exciting research avenue for artificial intelligence since it may be a game changer in emerging industrial fields such as robotic visions and automobiles. However, current deep learning (DL) faces major challenges for such applications because of the huge computational cost and inefficient learning. Hence, we developed a novel brain-inspired spiking neural network (SNN) based system titled spiking gating flow (SGF) for online action learning. The developed system consists of multiple SGF units which are assembled in a hierarchical manner. A single SGF unit contains three layers: a feature extraction layer, an event-driven layer, and a histogram-based training layer. To demonstrate the capability of the developed system, we employed a standard dynamic vision sensor (DVS) gesture classification as a benchmark. The results indicated that we can achieve 87.5% of accuracy which is comparable with DL, but at a smaller training/inference data number ratio of 1.5:1. Only a single training epoch is required during the learning process. Meanwhile, to the best of our knowledge, this is the highest accuracy among the non-backpropagation based SNNs. Finally, we conclude the few-shot learning (FSL) paradigm of the developed network: 1) a hierarchical structure-based network design involves prior human knowledge; 2) SNNs for content-based global dynamic feature detection.

A Hybrid Phishing Detection System Using Deep Learning-based URL and Content Analysis

Phishing attacks are one of the most preferred types of attacks for cybercriminals, who can easily contact a large number of victims through the use of social networks, particularly through email messages. To protect end users, most of the security mechanisms control Uniform Resource Locator (URL) addresses because of their simplicity of implementation and execution speed. However, due to sophisticated attackers, this mechanism can miss some phishing attacks and has a relatively high false positive rate. In this research, a hybrid technique is proposed that uses not only URL features, but also content-based features as the second level of detection mechanism, thus improving the accuracy of the detection system while also minimizing the number of false positives. Additionally, most phishing detection algorithms use datasets that contain easily differentiated data pieces, either phishing or legitimate. However, in order to implement a more secure protection mechanism, we aimed to collect a larger and high-risk dataset. The proposed approaches were tested on this High-Risk URL and Content-Based Phishing Detection Dataset that only contains suspicious websites from PhishTank. According to experimental studies, an accuracy rate of 98.37 percent was achieved on a more realistic dataset for phishing detection.

Predictive modeling for suspicious content identification on Twitter.

The wide popularity of Twitter as a medium of exchanging activities, entertainment, and information is attracted spammers to discover it as a stage to spam clients and spread misinformation. It poses the challenge to the researchers to identify malicious content and user profiles over Twitter such that timely action can be taken. Many previous works have used different strategies to overcome this challenge and combat spammer activities on Twitter. In this work, we develop various models that utilize different features such as profile-based features, content-based features, and hybrid features to identify malicious content and classify it as spam or not-spam. In the first step, we collect and label a large dataset from Twitter to create a spam detection corpus. Then, we create a set of rich features by extracting various features from the collected dataset. Further, we apply different machine learning, ensemble, and deep learning techniques to build the prediction models. We performed a comprehensive evaluation of different techniques over the collected dataset and assessed the performance for accuracy, precision, recall, and f1-score measures. The results showed that the used different sets of learning techniques have achieved a higher performance for the tweet spam classification. In most cases, the values are above 90% for different performance measures. These results show that using profile, content, user, and hybrid features for suspicious tweets detection helps build better prediction models.

Prediction Retweet Using User-Based and Content-Based with ANN-GA Classification Method

Current technological advances have caused rapid dissemination of information, especially on social media, one of which is Twitter. Retweeting or reposting messages is considered an easily available information diffusion mechanism provided by Twitter. By finding out why a user retweets a tweet from another person and by making this prediction we can understand how information diffuses on Twitter. In this study, Artificial Neural Network – Genetic Algorithm is used in the classification process and uses user-based and Content-Based features. Evaluation result obtained in this study are 90% accuracy, 72% precision, 83% recall, and 65% F1-Score value on the model by Oversampling.

Detecting COVID-19 vaccine hesitancy in India: a multimodal transformer based approach.

COVID-19 has emerged as the greatest threat in recent times, causing extensive mortality and morbidity in the entire world. India is among the highly affected countries suffering severe disruptions due this pandemic. To overcome the adverse effects of COVID-19, vaccination has been identified as the most effective preventive measure globally. However, a growing amount of hesitancy has been observed among the general public regarding the efficacy and possible side-effects of vaccination. Such hesitancy may proved to be the greatest hindrance towards combating this deadly pandemic. This paper introduces a multimodal deep learning method for Indian Twitter user classification, leveraging both content-based and network-based features. To explore the fundamental features of different modalities, improvisations of transformer models, BERT and GraphBERT are utilized to encode the textual and network structure information. The proposed approach thus integrates multiple data representations, utilizing the advances in both transformer based deep learning as well as multimodal learning. Experimental results demonstrates the efficacy of proposed approach over state of the art approaches. Aggregated feature representations from multiple modalities embed additional information that improves the classification results. The findings of the proposed model has been further utilized to perform a study on the dynamics of COVID-19 vaccine hesitancy in India.