Spam Filtering

Spam has become a major issue in computer security because it is a channel for threats such as computer viruses, worms and phishing. Many solutions feature machine-learning algorithms trained using statistical representations of the terms that usually appear in the e-mails. Still, these methods require a training step with labelled data. Dealing with the situation where the availability of labelled training instances is limited slows down the progress of filtering systems and offers advantages to spammers. Currently, many approaches direct their efforts into Semi-Supervised Learning (SSL). SSL is a halfway method between supervised and unsupervised learning, which, in addition to unlabelled data, receives some supervision information such as the association of the targets with some of the examples. Collective Classification for Text Classification poses as an interesting method for optimising the classification of partially-labelled data. In this way, we propose here, for the first time, Collective Classification algorithms for spam filtering to overcome the amount of unclassified e-mails that are sent every day.

This paper presents the use of corpus linguistics techniques on supposedly "clean" corpora and identifies potential pitfalls. Our work relates to the task of filtering sensitive content, in which data security is strategically important for the protection of government and military information, and of growing importance in combating identity fraud. A naïve keyword filtering approach produces a large proportion of false positives, and the need for more fine-grained approaches, suggests the consideration of using corpus linguistics for such content filtering. We present work undertaken on the Enron corpus, a collection of emails that has had various tasks undertaken on various portions and versions of it by other researchers. We have made some efforts to reconcile differences between the different versions by considering what impact some of these versions have on our results. Our anticipated efforts in using automatic ontology learning Gillam and Ahmad 2005], and local grammars to discover points of interest within the Enron corpus have been reoriented to the problem of discovering "confidentiality banners" with a view to removing them from the collection. Our work to date makes strong use of collocation patterns [Smada 1993] to identify the signatures of these banners, and to use the banners themselves to better understand the different versions of the Enron corpus. We further consider the use of extended collocation patterns to identify text "zones", following , and the subsequent potential for sentiment analysis ; 1

In this paper we describe an approach to information assurance in which we can prevent breach of confidentiality. Specifically, we examine aspects of the propagation of confidential information via email. Email provides one simple mechanism by which naive users can rapidly breach confidence, and effectively demonstrates the difficulty in enforcing "best practice". We propose and demonstrate an intelligent filtering system aimed at prevention of disclosure and enforcement of "best practice". We show how work in corpus linguistics and data mining can contribute significantly to information assurance and report on experiments on a relatively large email collection that demonstrates the value of such work. Our experiments further demonstrate the potential for reducing false positives. We give due consideration to the danger of missed messages that should have been prevented from propagation.

The growing popularity of YouTube video-sharing platforms requires organizations to analyze viewer comments for public opinion assessment and content development. A web application powered by machine learning techniques analyzes sentiment expressions present in YouTube comments as the main subject of this research paper. Natural Language Processing techniques function within the system to obtain video comments automatically which leads to positive and negative and neutral classification of sentiments. The application obtains essential video information which includes the title alongside view count and like count statistics and a full description allowing users to understand the video performance better. The system provides users with seven features including visual charts that display results as well as PDF report downloads with built-in capabilities for accessing past analyses through a history directory. This system functions as an important asset for viewers interested in video evaluation alongside creators who want to grasp user reactions and feedback toward their content. The tool relies on YouTube alongside sentiments and machine learning and Natural Language Processing (NLP) methods to analyze viewer feedback through comments alongside a web application to determine public sentiment about videos and provides graphical displays accompanied by PDF report generation for audience analysis and content strategy evaluation.

Electronic Mail (E-mail) has established a significant place in information user’s life. Mails are used as a major and important mode of information sharing because emails are faster and effective way of communication. Email plays its important role of communication in both personal and professional aspects of one’s life. The rapid increase in the number of account holders from last few decades and the increase in the volume of mails have generated various serious issues too. The content base mail classification can be classified into four ways namely Private, Public, Newsletter, and Anonymous. Every user has the right to choose their keyword (a semi-private password). Those contacts who know the user’s keyword will be classified as private contacts and those users who are unknown them classified anonymous contacts. A contact can be classified as public or private, upon verification of an anonymous contact. Any newsletter or group mails are classified into newsletter contacts. It is...

Access control in multi-tenant cloud environments faces significant challenges due to encrypted communications, protocol diversity, and dynamic tenant behavior. Traditional access control methods, such as static-ruled-based and signaturebased access control, fail to handle encrypted traffic effectively, while machine learning-based approaches are computationally expensive and require extensive training datasets. To address these limitations, we propose BLAC (Bit-Level Access Control)a lightweight, protocol-agnostic framework that integrates bitlevel traffic signatures with real-time policy enforcement. It ensures scalability through distributed classification, strict tenant isolation, and policy-driven access decisions. In this article, Conceptual validation demonstrates proposed Framework is protocol independence and resilience against security threats such as encrypted payloads, replay attacks, and malicious traffic injection, paving the way for future empirical deployment and testing.

Abstract. In the modern digital era, the demand for highly available and resilient
systems is constantly increasing, especially in cloud environments and data centers
that provide critical services. Xen virtualization is one of the most widely
used open source technologies for virtual machine management and enables efficient
resource allocation and high flexibility in the provision of services. The
Service Level Agreement (SLA) serves as a key indicator of the reliability and
quality of these services, which is influenced by various factors such as downtime,
CPU performance, bandwidth and data loss. Although SLA is a critical
metric for ensuring system performance, its relationship to other system factors
remains a challenge in advanced optimization and management.
The aim of this study is to investigate the effects of downtime, bandwidth, CPU
performance and data loss on SLA using a linear regression analysis. For this
purpose, experimental data was collected from a virtualized environment with
Xen. The analysis aims to determine the specific impact of each factor on the
SLA and create a clear model of their interdependencies.
The results of this research will contribute to the development of optimized strategies
for system administrators and cloud service providers, enabling them to
minimize downtime, reduce data loss and maximize network performance. These improvements will ensure higher service availability, especially in scenarios
where virtual machine failures pose a risk to the stable operation of the system.

The exponential growth of mobile communication has intensified the threat of SMS spam, compromising user security and trust in messaging platforms. This study addresses this challenge by designing and deploying a robust spam detection system using machine learning. We analyze a publicly available SMS dataset through rigorous pre-processing, including text normalization, tokenization, and feature engineering, followed by TF-IDF vectorization. A comparative evaluation of 11 classifiers-spanning probabilistic models, ensemble methods, and linear classifiers-reveals that ensemble techniques outperform traditional algorithms. The Extra Trees Classifier and XGBoost achieve state-of-the-art results, with 97.9% accuracy and 97.5% precision, demonstrating their efficacy in distinguishing spam from legitimate messages. To bridge the gap between research and practical application, we develop an interactive Streamlit web application that enables real-time spam classification with a user-friendly interface. This work underscores the potential of ensemble learning for text classification tasks and provides a scalable framework for combating SMS spam in real-world scenarios.

Spam consists of varieties of contents like text, image, embedded HTML, MIME attachments and also the volume of spam mails sent per day is massive. To handle this high volume, high velocity and large varieties of spam, a scalable spam filtering solution is required. Scalable solutions available for machine learning and statistical studies can be used to implement a scalable solution for spam filtering also. From Big data Analytics domain, Mahout is an open source library from Apache for building scalable solutions in machine learning. This paper uses mahout framework to analyse the time and accuracy efficiencies of the results of two Naive Bayes classification algorithms. Keywords: Apache Mahout, big data, scalable algorithms, Naive Bayes algorithms

The Autism Spectrum Disorder (ASD) is a neurological disease, which affects the mental, social and physical state of a person. A person of any age group can be found infected by it. It is very difficult to identify, if a person is the victim of this disorder. Classical approaches that find the occurrence of autism in a person is time consuming and expensive. Machine learning approaches, on the other hand, have paved the way for intelligent diagnostics. This paper focusses on identification of specific traits that helps to automate the diagnosis process and further evaluate and perform a comparative analysis of various machine learning algorithms namely K-Nearest Neighbour, Logistic Regression, SVM and Naïve Bayes, to predict the occurrence of autism disorder. Experimental analysis shows that the Naïve Bayes algorithm provides a better accuracy of 99.6% compared to other algorithms.

An intelligent system that uses Natural Language Processing (NLP) and Machine Learning (ML) to automate resume classification is presented in this paper. Key resume features, such as education, skills, and job titles, are extracted and used to train models like Logistic Regression, SVM, Random Forest, and BERT. NLP preprocessing techniques, such as tokenization, stop word removal, and vectorization, prepare the text for analysis. The results of experiments show that these models improve classification accuracy and decrease the time needed for hiring. The system assists recruiters by ranking qualified candidates and eliminating applications that are not relevant, making the hiring process quicker and more efficient.

In the modern era, mobile phones have become ubiquitous, and Short Message Service (SMS) has grown to become a multi-million-dollar service due to the widespread adoption of mobile devices and the millions of people who use SMS daily. However, SMS spam has also become a pervasive problem that endangers users' privacy and security through phishing and fraud. Despite numerous spam filtering techniques, there is still a need for a more effective solution to address this problem [1]. This research addresses the pervasive issue of SMS spam, which poses threats to users' privacy and security. Despite existing spam filtering techniques, the high false-positive rate persists as a challenge. The study introduces a novel approach utilizing Natural Language Processing (NLP) and machine learning models, particularly BERT (Bidirectional Encoder Representations from Transformers), for SMS spam detection and classification. Data preprocessing techniques, such as stop word removal and tokenization, are applied, along with feature extraction using BERT. Machine learning models, including SVM, Logistic Regression, Naive Bayes, Gradient Boosting, and Random Forest, are integrated with BERT for differentiating spam from ham messages. Evaluation results

With the growth of networking the usage of mails are also enhanced. Due to rapid growth of internet, dependency of communication is mostly based on electronics mails for both commercial and business purposes. According to today's scenarios electronics mails are also plays vital role in marketing or production advertisement. Hence numerous marketing firms used e- mails as a tool for promoting their products and services. These types of mails are generally called spam mails. Sometimes it is quite difficult to identify important mails among such group of spam mails. Because of this identification of such types of mails are essential so that they can keep away from important mails. In this paper we are presenting various techniques that are used spam filtering.

This paper describes our approach towards the ECML/PKDD Discovery Challenge 2010. The challenge consists of three tasks: (1) a Web genre and facet classification task for English hosts, (2) an English quality task, and (3) a multilingual quality task (German and French). In our approach, we create an ensemble of three classifiers to predict unseen Web hosts whereas each classifier is trained on a different feature set. Our final NDCG on the whole test set is 0.537 for Task 1, 0.844 for Task 2, and 0.823 (French) and 0.793 (German) for Task 3, which ranks fourth place in the ECML/PKDD Discovery Challenge 2010.

The widespread use of email as a primary communication medium has led to an increase in spam messages, which pose significant threats to privacy, productivity, and cybersecurity. Spam emails, often disguised as legitimate messages, can carry malicious links, phishing scams, and fraudulent content. This paper presents a machine learning-based approach for identifying spam emails with high accuracy. By employing natural language processing (NLP) techniques and the Naïve Bayes classifier, we preprocess a labeled dataset of email messages, extract relevant features, and train a classification model. The model's effectiveness is evaluated using performance metrics such as accuracy, precision, recall, and F1-score. The results demonstrate the reliability and practicality of machine learning in mitigating email spam, offering a scalable and adaptive solution to an ongoing digital challenge

Phishing remains one of the most prevalent and evolving cybersecurity threats, exploiting humanvulnerabilities through deceptive digital communication. This study proposes a dynamic, Windows-specific phishing detection model leveraging Random Forest machine learning techniques. By integrating Term Frequency-Inverse Document Frequency (TF-IDF) vectorization with structured email features, the model classifies phishing and legitimate emails with high accuracy. Using secondary data and publicly available datasets, the model achieved a classification accuracy of 98.31% and demonstrated balanced performance across precision, recall, and F1-score metrics. This research underscores the effectiveness of hybrid feature strategies and ensemble learning for phishing detection while outlining key limitations and future directions, including model generalization and real-world deployment readiness.

The expressive power of regular expressions has been often exploited in network intrusion detection systems, virus scanners, and spam filtering applications. However, the flexible pattern matching functionality of regular expressions in these systems comes with significant overheads in terms of both memory and CPU cycles, since every byte of the inspected input needs to be processed and compared against a large set of regular expressions. In this paper we present the design, implementation and evaluation of a regular expression matching engine running on graphics processing units (GPUs). The significant spare computational power and data parallelism capabilities of modern GPUs permits the efficient matching of multiple inputs at the same time against a large set of regular expressions. Our evaluation shows that regular expression matching on graphics hardware can result to a 48 times speedup over traditional CPU implementations and up to 16 Gbit/s in processing throughput. We demonstrate the feasibility of GPU regular expression matching by implementing it in the popular Snort intrusion detection system, which results to a 60% increase in the packet processing throughput.

Classifier performance optimization in machine learning can be stated as a multi-objective optimization problem. In this context, recent works have shown the utility of simple evolutionary multi-objective algorithms (NSGA-II, SPEA2) to conveniently optimize the global performance of different anti-spam filters. The present work extends existing contributions in the spam filtering domain by using three novel indicator-based (SMS-EMOA, CH-EMOA) and decomposition-based (MOEA/D) evolutionary multi-objective algorithms. The proposed approaches are used to optimize the performance of a heterogeneous ensemble of classifiers into two different but complementary scenarios: parsimony maximization and e-mail classification under low confidence level. Experimental results using a publicly available standard corpus allowed us to identify interesting conclusions regarding both the utility of rule-based classification filters and the appropriateness of a three-way classification system in the spam filtering domain.

Question and Answering system is one of the widely used Mechanism in student Community for learning. This paper mainly focuses on Question and answering system based on Paragraphs. Datasets such as the Stanford Question-Answering Dataset (SQuAD) have improved performance on machine reading comprehension tasks but this is applicable mostly for English. In India, the third most spoken regional language is Telugu. Telugu belongs to the Dravidian language family and is characterized by a rich morphological system resulting in long and complex word forms. The work in this paper concentrates on developing a question answering system for 10th class biology Textbook for Andhra Pradesh. The Telugu biology Question answering and system (TelBioQA) which aims to generate answers and that explain ambiguous questions and solve problems with phrases and coreferences. The system processes the queries and generates respective suitable responses using natural language processing (NLP). A deepset xlm roberta base-squad2 is used in Telugu question answering system. While testing with the metrics it shows the empirical results which are optimistic compared to other models. The F1 Score is 86%, and the Extract Match Score is 65%.

Question and Answering system is one of the widely used Mechanism in student Community for learning. This paper mainly focuses on Question and answering system based on Paragraphs. Datasets such as the Stanford Question-Answering Dataset (SQuAD) have improved performance on machine reading comprehension tasks but this is applicable mostly for English. In India, the third most spoken regional language is Telugu. Telugu belongs to the Dravidian language family and is characterized by a rich morphological system resulting in long and complex word forms. The work in this paper concentrates on developing a question answering system for 10th class biology Textbook for Andhra Pradesh. The Telugu biology Question answering and system (TelBioQA) which aims to generate answers and that explain ambiguous questions and solve problems with phrases and coreferences. The system processes the queries and generates respective suitable responses using natural language processing (NLP). A deepset xlm roberta base-squad2 is used in Telugu question answering system. While testing with the metrics it shows the empirical results which are optimistic compared to other models. The F1 Score is 86%, and the Extract Match Score is 65%.

There is a critical need for organizations to share data within and across infospheres and form coalitions so that analysts could examine the data, mine the data, and make effective decisions. Each organization could share information within its infosphere. An infosphere may consist of the data, applications and services that are needed for its operation. Organizations may share data with one another across what is called a global infosphere that spans multiple infospheres. It is critical that the war fighters get timely information. Furthermore, secure data and information sharing is an important requirement. The challenge is for data processing techniques to meet timing constraints and at the same time ensure that security is maintained. This proposal addresses information operations across infospheres. We first describe secure timely data sharing across infospheres and then focus on Role-based access control and Usage control in such an environment. Our goal is to send timely inf...

The advanced architecture of Large Language Models (LLMs) has revolutionised natural language processing, enabling the creation of text that convincingly mimics legitimate human communication, including phishing emails. As AI-generated phishing emails become increasingly sophisticated, a critical question arises: How effectively can current email systems and detection mechanisms identify these threats? This study addresses this issue by analysing 63 AI-generated phishing emails created using GPT-4o. It evaluates the effectiveness of major email services, Gmail, Outlook, and Yahoo, in filtering these malicious communications. The findings reveal that Gmail and Outlook allowed more AI-generated phishing emails to bypass their filters compared to Yahoo, highlighting vulnerabilities in existing email filtering systems. To mitigate these challenges, we applied 60 stylometric features across four machine learning models: Logistic Regression, Support Vector Machine, Random Forest, and XGBoost. Among these, XGBoost demonstrated superior performance, achieving 96% accuracy and an AUC score of 99%. Key features such as imperative verb count, clause density, and firstperson pronoun usage were instrumental to the model's success. The dataset of AI-generated phishing emails is publicly available on Kaggle to foster further research.

In this paper, we present a new spam filter which acts as an additional layer in the spam filtering process. This filter is based on what we call a representative vocabulary. Spam e-mails are divided into categories in which each category is represented by a set of tokens which form a Representative Text (RT). Tokens are strings of characters (words, sentences, or some times meaningless strings of characters). This RT is used to compute a resemblance ratio with incoming e-mails. With this ratio we decide whether the incoming e-mail is a spam. This filter was implemented and integrated to Spamihilator software. Some experimental and interesting results will be presented.

Summary The attribute-oriented induction (AOI) method is a useful tool for data capable of extracting generalized knowledge from relational data and the user's background knowledge. However, a potential weakness of AOI is that it only provides a snapshot of generalized knowledge, not a global picture. In addition, the method only mined knowledge from positive facts in databases. Rare but important negative generalized knowledge can be missed. Hence, the aim of this study is to proposal two novel mining approaches to generate multiple-level positive and negative generalized knowledge. The approaches discussed in this paper are more flexible and powerful than currently utilized methods and can be expected to have wide applications in diverse areas including e- commerce, e-learning, library science, and so on.

During the past thirty years, the world of computing has evolved from large centralised computing centres to an increasingly distributed computing environment, where computation and communication capabilities are being embedded in artefacts of everyday life. Billions of computational entities will interact in systems with ever changing configurations determined by local and global context, for example, the location of the user. In such dynamic environments, users would be overwhelmed if involved in computing-related decisions every time the context changes. Due to the number of decisions required to sustain continuous service, most decisions will have to be made by the computing entities themselves. Moreover, due to the global scale of the environment and the potential risk of disconnected operations, the computing entities may have to make these decisions autonomously, without relying on a given fixed infrastructure. Knowledge, especially about the context of the interaction, is vital for the accuracy of these decisions. However, keeping information on a global scale is unfeasible for resource-constrained entities, so some degree of uncertainty must be assumed. This peer-to-peer type of interaction in an uncertain world where interactions are needed to go forward resembles what occurs in human social networks. The notion of trust has emerged in human society to allow humans to make decisions under such circumstances. It has been proposed that computing entities can make decisions based on a computational model of trust. The trust engine run by each entity distributes and gathers pieces of evidence, that is, knowledge about the interacting entities: direct observations, recommendations or reputation. Since the trust engines collaborate and malicious collaborating entities exist, security through collaboration must be considered. As the real world does not have a unique legitimate authority, computing entities are owned by multiple authorities and operated from multiple jurisdictions. As in real life, no administrator can be perpetually present to manage the interactions. The trust engine can adapt security in a peer-to-peer way. A crucial element for the use of trust is to know with whom the entities interact, which corresponds to authentication in traditional computer security. However, this element has been disregarded in computational trust: this is ill-fated given that virtual identities are the means for a number of attacks that are less possible in face-to-face settings. This thesis sets up a framework, called entification, which encompasses both computational trust and identity aspects, and whose goal is to be applicable to global computing. For this purpose, this thesis

Combatting email spam has remained a very daunting task. Despite the over 99% accuracy in most non-image-based spam email detection, studies on image-based spam hardly attain such a high level of accuracy as new email spamming techniques that defeat existing spam filters emerges from time to time. The number of email spams sent out daily has remained a key factor in the continued use of spam. In this paper, a simple convolutional neural network model, 123DNet was developed and trained with 28,929 images drawn from 2 public datasets and a Personally Generated dataset. The model was optimized to the least set of layers to have 1 input layer, 2 embedded Convolutional layers as a hidden layer, and 3 neural network layers. The model was tested with a total of 4,339 images of the three dataset samples and then with a separate set of 1,200 images to test performance on never-seen-before images. A Classification Performance analysis was carried out using the confusion matrix. Performance metrics including Accuracy, Precision, True Negative Accuracy, Sensitivity, Specificity, and F1 Measure were computed to ascertain the model's performance. The Model returned an F1 Score of 97% on a public dataset's test sample and 88% on Never-seen-before test samples outperforming some pre-existing models while performing significantly well on the newly generated image test samples. It is recommended that a model that performed so well with new never-seen-before spam images be integrated into spam filtering systems.

Digital materials can be protected from failures by replicating them at multiple autonomous, distributed sites. A Peer-to-peer Information Preservation and Exchange (PIPE) network is a good way to build a distributed replication system. A significant challenge in such networks is ensuring that documents are replicated and accessible despite malicious sites. Such sites may hinder the replication of documents in a variety of ways, including agreeing to store a copy but erasing it instead, refusing to serve a document, or serving an altered version of ...

This paper presents EmailValet, a system that learns users' emailreading preferences on email-capable wireless platforms -specifically, on two-way pagers with small "qwerty" keyboards and an 8-line 30-character display. In use by the authors for about three months, it has gathered data on email-reading preferences over more than 8900 email messages received by the authors during this period. The paper presents results comparing the ability of different learning methods to form models that can predict whether a given message should be forwarded to the user's wireless device. Our results show that the best performance of one method, over a range of established learning methods developed on the information retrieval and machine learning communities, was able to achieve a break-even point of over 53% for one user that had received over 5000 messages. We also find that, in general, all methods are able to achieve better performance than what would be achieved by a baseline of simply forwarding all messages to the wireless device, and that many methods are able to find procedures that, although they forward only a small fraction of the messages that a user would want, achieve 100% precision on those messages that it does actually choose to forward. able to achieve a break-even 1 point of 53%. We also present some more detailed analyses of the various methods, including how different ways of encoding the email for the learner affect learning performance. We conclude the paper with a discussion of future work and final remarks.

This paper explores the fundamental principles of internal linking and PageRank optimization. It provides a detailed overview of how internal links influence a website’s SEO, helping distribute link equity and improving user navigation. The paper also examines PageRank’s origins, its calculation model, and how it remains a critical ranking factor today.

Malicious websites are those that are created to harm visitors or exploit their information for illegal purposes. These websites are commonly utilized in attacks, such as phishing, malware distribution, and scams. Clicking on a malicious URL can result in catastrophic outcomes, such as data breaches, financial losses, and identity theft. Detecting and blocking these websites is essential for protecting individuals and organizations from online threats, preserving data security, and sustaining confidence in online platforms. Researchers presented several methods for detecting malicious websites. Due to the threat's evolution, the problem remains unsolved. This paper presents a machine-learning model for malicious website identification. To process the textual contents, the experiment uses four different text vectorization techniques such as Count, TF-IDF, Hashing and word embedding (average Word2Vec). Eight machine learning models are tested to assess performance. The outcome demonstrates that the average word2vec embedding with extreme gradient boosting and random forest obtains 94.76% and 94.70% accuracy, respectively.

In virtual machine environments each application is often run in its own virtual machine (VM), isolating it from other applications running on the same physical machine. Contention for memory, disk space, and network bandwidth among virtual machines, coupled with an inability to share due to the isolation virtual machines provide, leads to heavy resource utilization. Additionally, VMs increase management overhead as each is essentially a separate system. Stork is a package management tool for virtual machine environments that is designed to alleviate these problems. Stork securely and efficiently downloads packages to physical machines and shares packages between VMs. Disk space and memory requirements are reduced because shared files, such as libraries and binaries, require only one persistent copy per physical machine. Experiments show that Stork reduces the disk space required to install additional copies of a package by over an order of magnitude, and memory by about 50%. Stork downloads each package once per physical machine no matter how many VMs install it. The transfer protocols used during download improve elapsed time by 7X and reduce repository traffic by an order of magnitude. Stork users can manage groups of VMs with the ease of managing a single machineeven groups that consist of machines distributed around the world. Stork is a real service that has run on PlanetLab for over four years and has managed thousands of VMs. Stork: Package Management for Distributed VM Environments Cappos, et al. efficient and reducing the load on the repository. Stork uses fail-over mechanisms to improve the reliability of downloads, even if the underlying content distribution systems fail. Stork also makes use of publish/subscribe technology to ensure that VMs are notified of package updates in a timely fashion. Repository Client Mgmt Embedded User Private Key

Email phishing is a manipulative technique aimed at compromising information security and user privacy. To overcome the limitations of traditional detection methods, such as blacklists, this research proposes a phishing detection model that leverages natural language processing (NLP) and deep learning technologies to analyze Indonesian email headers. The primary objective is to more efficiently detect zero-day phishing attacks by focusing on the unique linguistic and cultural context of the Indonesian language. This enables the development of models capable of recognizing phishing attack patterns that differ from those in other language contexts. Four models are tested, combining Indonesian bidirectional encoder representation of transformers (IndoBERT) and FastText feature extraction techniques with convolutional neural network (CNN) and long short-term memory (LSTM) deep learning algorithms. The results indicate that the combination of FastText and CNN achieved the highest performance in accuracy, precision, and F1-score metrics, each at 98.4375%. Meanwhile, the FastText model with LSTM showed the best performance in recall, with a score of 98.9583%. The research suggests exploring deeper into email content or integrating analysis between headers and email content in future studies to further improve accuracy and effectiveness in phishing email detection.

In this paper, we examine the application of various grouping techniques to help improve the efficiency and reduce the costs involved in an electronic discovery process. Specifically, we create coherent groups of email documents which characterize either a syntactic theme, a semantic theme or an email thread. All such grouped documents can be reviewed together leading to a faster and more consistent review of documents. Syntactic grouping of emails is based on near duplicate detection whereas semantic grouping is based on identifying concepts in the email content using information extraction. Email thread detection is achieved using a combination of segmentation and near duplicate detection. We present experimental results on the Enron corpus that suggest that these approaches can significantly reduce the review time and show that high precision and recall in identifying the groups can be achieved. We also describe how these techniques are integrated into the IBM eDiscovery Analyzer product offering.

Abstract—In recent times, the problem of Unsolicited Bulk Email (UBE) or commonly known as Spam Email, has increased at a tremendous growth rate. We present an analysis of survey based on classifications of UBE in various research works. There are many research instances for classification between spam and non-spam emails but very few research instances are available for classification of spam emails, per se. This paper does not intend to assert some UBE classification to be better than the others nor does it propose any new classification but it bemoans the lack of harmony on number and definition of categories proposed by different researchers. The paper also elaborates on factors like intent of spammer, content of UBE and ambiguity in different categories as proposed in related research works of classifications of UBE.

Email has become a fast and cheap means of online communication. The main threat to email is Unsolicited Bulk Email (UBE), commonly called spam email. The current work aims at identification of unigrams in more than 2700 UBE that advertise body-enhancement drugs. The identification is based on the requirement that the unigram is neither present in dictionary, nor is a slang term. The motives of the paper are many fold. This is an attempt to analyze spamming behaviour and employment of wordmutation technique. On the side-lines of the paper, we have attempted to better understand the spam, the slang and their interplay. The problem has been addressed by employing Tokenization technique and Unigram BOW model. We found that the non-lexicon words constitute nearly 66% of total number of lexis of corpus whereas non-slang words constitute nearly 2.4% of non-lexicon words. Further, non-lexicon non-slang unigrams composed of 2 lexicon words, form more than 71% of the total number of such uni...

In this paper, we introduced a statistical rule-based method to create rules for SpamAssassin to detect spams in different languages. The theoretical framework of generating and maintaining multilingual rules were also illustrated. The experiments were conducted against the dataset of three languages including Chinese, Vietnamese and English. The detecting achievement of multilingual rule was 89.5% for the true detection and only 3.8% for the failed alarm at the threshold of 2 while the true detection rate of single language rule was not over 61% and the failed alarm rate was up to 4.9%.

Advanced machine learning and natural language techniques enable attackers to launch sophisticated and targeted social engineering based attacks. To counter the active attacker issue, researchers have since resorted to proactive methods of detection. Email masquerading using targeted emails to fool the victim is an advanced attack method. However automatic text generation requires controlling the context and coherency of the generated content, which has been identified as an increasingly difficult problem. The method used leverages a hierarchical deep neural model which uses a learned representation of the sentences in the input document to generate structured written emails. We demonstrate the generation of short and targeted text messages using the deep model. The global coherency of the synthesized text is evaluated using a qualitative study as well as multiple quantitative measures.

Spam which is one of the most popular and also the most relevant topic that needs to be understood in the current scenario. Everyone whether it may be a small child or an old person are using emails everyday all around the world. The scenario which we are seeing is that almost no one is aware or in simple sentence they do not know what actually the spam is and what they will do in their systems. Spam in general means unsolicited or unwanted mails. Botnets are considered one of the main source of the spam. Botnet means the group of software’s called bots and the function of these bots is to run on several compromised computers autonomously and automatically. The main objective of this paper is to detect such a bot or spambots for the Bulletin Board System (BBS). BBS is a computer that is running software that allows users to leave a message and access information of general interest. Originally BBSes were accessed only over a phone line using a modem, but nowadays some BBSes allowed ...

In this paper we describe an online/incremental linear binary classifier based on an inter-esting approach to estimate the Fisher subspace. The proposed method allows to deal with datasets having high cardinality, being dynamically supplied, and it efficiently copes ...

Search Engine spam is a web page or a portion of a web page which has been created with the intention of increasing its ranking in search engines. Web spamming refers to actions intended to mislead search engines and give some pages higher ranking than they deserve. Anyone who uses a search engine frequently has most likely encountered a high ranking page that consists of nothing more than a bunch of query keywords. These pages detract both from the user experience and from the quality of the search engine. Search engine spam is a webpage that has been designed to artificially inflating its search engine ranking. Recently this search engine spam has been increased dramatically and creates problem to the search engine and the web surfer. It degrades the search engine’s results, occupies more memory and consumes more time for creating indexes, and frustrates the user by giving irrelevant results. Search engines have tried many techniques to filter out these spam pages before they can ...

Nowadays, Search Engines have made progress lately and the number of the pages of web sites increases every days. The Search Engines ‫‬ ‫‬ ‫‬ the most common search systems are for meeting the needs of the users in searching the information in web sites and web pages. Search Engine Optimization (SEO) is a suitable strategy for increasing the ranking of the web pages and also increasing the organic traffic of the web sites in Search Engines. By means ‫‬ ‫‬ ‫‬ of ‫‬ ‫‬ ‫‬ Search Engines, ‫‬ ‫‬ ‫‬ Search ‫‬ ‫‬ ‫‬ Engine ‫‬ ‫‬ ‫‬ Result ‫‬ ‫‬ ‫‬ Pages (SERPs) in web based on ‫‬ ‫‬ ‫‬ keyword ‫‬ ‫‬ ‫‬ indexing is done. ‫‬ ‫‬ ‫‬ This cause the SEO for searching the information in web sites take place by the ranking. So, in this paper, we tried to review and evaluation the SEO ‫‬ ‫‬ ‫‬ and increasing the ranking of web sites in Search Engines.

Email spam is a kind of electronic spam, which tends to be a more difficult problem nowadays among all internet challenges. Spam mails are mostly sent in commercial purpose, some of them may contain malware links that lead to phishing websites. The aim of this study is to classify into ham and spam emails with an optimized and well efficient classification technique. Ham holds emails that are legitimate or legally valid message can get accepted by users. Spam emails are unwanted emails that a user doesn't want and to get rid of it. This study emphasizes on the improvement in classifying all mails into these two groups with minimal requirement of training and with an accuracy of hundred percent. Here in this study, Modified Naïve Bayes (MNB) classifier ensured the requirements with very low percentage of training and produces accurate results than existing Naïve Bayes (NB) or Supporting Vector Machine (SVM) classifier.

The series "Lecture Notes in Networks and Systems" publishes the latest developments in Networks and Systems-quickly, informally and with high quality. Original research reported in proceedings and post-proceedings represents the core of LNNS. Volumes published in LNNS embrace all aspects and subfields of, as well as new challenges in, Networks and Systems. The series contains proceedings and edited volumes in systems and networks, spanning the areas of Cyber

This study proposes a new method that utilizes the correlation structure between the number of words in the mail and the Bayesian score. Spam mails usually do not have a stable style and features. Spammers who send such mails, go on changing the features. The most often used statistical filter for email filtering is the Naive Bayesian filter. However, the training data and the word corpus that the filter designer utilized will determine the filter's design. A new mail with unknown nature is classified into spam (unsolicited mail) or ham (legitimate mail) basing on a score by combining conditional probabilities of tokens in the mail. The statistical behavior of this score indicates some interesting features, which can be explored to improve performance of the filter. We report the results of an experiment using Enron data set and highlight the advantages of the new filter. We also propose a new method of testing the model using random data sets.

The use of email has grown exponentially over the past decade, making it one of the most widely used forms of electronic communication. Recently, spam emails have become a major issue for email users. A spammer is someone who sends out unsolicited, mass emails. It causes problems for email servers, wastes time, and uses up bandwidth. There are a variety of spam filtering techniques that can be used to divide incoming email into spam and legitimate communications. The average user is negatively impacted due of the proliferation of spam emails, which makes it more difficult to find the truly useful messages and can even drive some to ignore crucial ones. Therefore, there is a requirement for an efficient email spam detector that can block the maximum number of spam emails with the highest degree of precision, ensuring that no legitimate emails are mistakenly deleted as spam. This work proposes a hybrid strategy for detecting spam emails by using Modified Particle Swarm Optimization for Anti-Spam Filtering System (MPSO_ASFS). The data employed in this work is linearly separable, and the Proposed MGA ASFS approach achieved higher accuracy than competing classifiers on the Ling Spam Corpus dataset (92.23%), the PU dataset (94.56%), the Enron Corpus dataset (91.34%), and the Spam Assassin dataset (91.35%).