PLOMS AI

Deep Neural Network for Arabic Tweets Sentiment Analysis Related to COVD-19

Jawad Hassan Alkhateeb — 2022-01-28

Aside from the Coronavirus pandemic, several other major crises erupted worldwide. Various industries have been irreparably harmed, and many organizations have succumbed to the calamity. There is an unavoidable need to examine various patterns on social media platforms in order to reduce public anxiety and misconceptions. The study examines the emotional orientations of Arabic persons who use social media, specifically the Twitter platform. From November 2020 to January 2021, we gathered data from Twitter. Tweets have been sent from a number of Arab cities. Natural Language Processing (NLP) and Machine Learning (ML) methods are used to identify whether an opinion's sentiment is favorable, negative, or neutral. This study gathers Arabic tweets and then manually annotates them to classify them as negative, positive, neutral, etc. In this work, word embedding and TFIDF are employed as vector features, with Long Short-Term Memory (LSTM) and Naive Bayes used for classification. This paper offers a learned LSTM model and a Naive Bayes model based on the collected tweets, leveraging two powerful ML methods. The LSTM model is more effective than the Nave Bayes model since its superior performance. This is noted by achieving an accuracy of 99% for the LSTM model. The analysis of this study aids various governments and corporate entities in better understanding public attitude and behavior in the face of the pandemic and making strategic decisions in response. Furthermore, this study focuses on data visualization by exhibiting an emotion plot and a word cloud. This study is an extended version of our work "the Sentiment Analysis of Arabic Tweets Related to COVID-19 Using Deep Neural Network".

A Study of Crowd Abnormal Events Understanding in Surveillance Videos

Mousumi Yeasmin Benzir — 2022-09-09

Crowd abnormal events detection in surveillance videos is a common topic in computer vision. For better security and safety, automatic video surveillance systems can detect and record abnormal activities at public and private places. However, traditional methods based on optical flow or segmentation cannot show good detection performance. On the other hand, deep learning based solutions for crowd unusual events detection showed better performance than those of conventional machine learning. This paper includes the latest deep learning models for crowd abnormal events detection in surveillance videos and their overall performance study.

Internet of Things (IoT) Security Vulnerabilities: A Review

Rabie Ramadan — 2021-10-20

The Internet of Things (IoT) collects and processes data from remote locations, substantially increasing the productivity of dispersed systems or individuals. Due to the restricted budget available for power consumption, IoT devices usually lack advanced data encryption and device authentication. The hardware components used in IoT devices are not high-end, and therefore, the integrity and security of the majority of IoT devices are in question. For instance, an adversary may include a Hardware Trojan (HT) during the manufacturing phase of IoT hardware devices to trigger data leakage or device failures in addition to other security issues. Here, we examine security risks to IoT in this paper, with a particular focus on attacks aimed at compromising the software, hardware, communication, and chip.

Intrusion Detection Based Learning in Wireless Sensor Networks

Rabie Ramadan — 2021-08-11

Wireless Sensor Networks (WSNs) have different limitations, including storage and processing capabilities. Besides, sensors' communication ranges are limited. Besides, those limitations raise the issue of the network intrusion and sensors' capabilities in detecting intruders. This paper introduces two different algorithms to detect intrusions in wireless sensor-based systems, which are more vulnerable to attacks. The first proposed algorithm is supervised learning-based classification. On the other hand, the second algorithm is unsupervised learning-based clustering. Both algorithms try to detect intrusions using a set of detection rules that are structured in the form of decision trees. The algorithms are detailed and extensively examined on a well-known dataset. They also are tested against two different architectures, two and three levels networks. The three-level architecture represents the sensor node, sink node, and base station level, while the two-level architecture represents the levels of sensor and sink nodes. An enhancement for decision-tree-based classification algorithms is also proposed by changing the decision tree to a binary tree. Such change made a significant enhancement in the complexity of reaching a decision. The produced decision trees use a similar decision tree node structure as the one used in Classification and Regression Trees (CART). The performance of our proposed algorithms and techniques are measured and extensively examined.