Spam Problem Research Articles

Detection of Fake Tweets Using Sentiment Analysis

Social platform is one of the most commonly used sites in the today’s world, and people from different places exchange information, express opinions, etc. Twitter is one of the most convenient platforms where Twitter data are frequently used for research and are one of the largest micro-blogging platforms. As there is an increase in number of spam accounts, there is a need to distinguish between genuine and fake user accounts. In Twitter, there is rise in the number of spam accounts, to spread malicious information among the users. So, to overcome the problem of spam in Twitter, an analysis of the opinions and the emotions of users in the form of tweets have been carried out. In this paper, we present a model which performs sentimental analysis on real-time data that are collected on a particular topic of interest for the users who recently posted on that topic. Using this model, we compute the sentiment score of each user with the content-based features to detect spam in Twitter. The proposed method makes use of a custom rule-based algorithm for detection of bots and compares it with few other algorithms like MLP, decision tree and random forest to see how efficient the model is in detection of spam accounts.

Twitter Spam Detection using Pre-trained Model

In the age of technology social media platform is becoming a great companion for expressing the thoughts, information, and opinion. It became the powerful tool for every person who wants to expand their networks of people beyond the physical boundation. We are living at that age where various categories of social media platform available according to work needed, it may be Facebook, LinkedIn, WhatsApp or Twitter. We are focusing our work on Twitter, It is also known as microblogging site which provides service to express the opinion in limited words. As the popularity of twitter is growing day by day users are joining the platform very fast, as it happens another side many spammers are also taking undue advantages of this platform, for any social media platform it is very important to maintain the secure, safer and trustworthy environment for their legitimate users. Twitter spams are more harmful than e-mail spam because of their higher clickthrough rate, as in the social network if someone trusted some spam a genuine post than it is higher chance that the persons in the network might also trust on that spam post and may click on it. There are plenty of methods available to handle the task of twitter spam detection problem, we are solving this problem of twitter spam at tweet level.Pre-trained models are some breakthrough in the journey of machine learning and natural language processing after their advancement they are of great help. Here we are using Bidirectional Encoder Representation from Transformer (BERT) model to solve the problem as our task is to solve the problem of imbalance dataset as well as the multilingual dataset, BERT makes a clear distinction in this type of task, the main advantage of this type’s model is that we don’t have to collect millions of data for better performance of the machine learning model.

Relevant SMS Spam Feature Selection Using Wrapper Approach and XGBoost Algorithm

In recent years with the widely usage of mobile devices, the problem of SMS Spam increased dramatically. Receiving those undesired messages continuously can cause frustration to users. And sometimes it can be harmful, by sending SMS messages containing fake web pages in order to steal users’ confidential information. Besides spasm number of hazardous actions, there is a limited number of spam filtering software. According to this paper, XGBoost algorithm used for handling SMS spam detection problem. Number of structural features was collected from previous studies. 15 structural features were extracted from Tiago’s dataset, which is the most frequently used dataset by researchers. For selecting the optimal relevant features, two different types of wrapper feature selection algorithms were used in order to reduce and select best relevant features. The accuracy and performance obtained by the selected features via sequential backward selection method was better comparing to sequential forward selection method. The extracted nine optimal features can be a good representation of a spam SMS message. Additionally, the classification accuracy obtained by the proposed method using nine optimal features with XGBoost algorithm is 98.64 using 10-fold cross validation.

An Anti-Spam Detection Model for Emails of Multi-Natural Language

The spam is one of the illegal and negative practices that involves the use of email services to send unsolicited emails such as phishing for the purpose of scamming which influences the reliability of email. Investigations have been conducted from various perspectives in order to examine this spam problem and how it affects society. In this regard, many studies have been carried out with the aim of studying the effect of spam activity on finance, economy, marketing, business and management, while other studies have focused on studying the influence of spam on security and privacy. Consequently, the literature affords various anti-spam methods that blocks or filters spam emails. This paper investigates the existing anti-spam methods, highlights some current problems and carries out an improved anti-spam model. In this regard, a new agent-based of Multi-Natural Language Anti-Spam (MNLAS) model is proposed. The MNLAS model process in the spam filtering process of an email both visual information such as images and texts in English and Arabic languages. The Jade agent platform and Java environments are employed in the implementation of MNLAS model. The MNLAS model is tested on a 200 emails’ dataset and the results show that it is able to detect and filter various kinds of spam emails with high accuracy.

Study on the Effect of Preprocessing Methods for Spam Email Detection

The use of email as a communication technology is now increasingly being exploited. Along with its progress, email spam problem becomes quite disturbing to email user. The resulting negative impacts make effective spam email detection techniques indispensable. A spam email detection algorithm or spam classifier will work effectively if supported by proper preprocessing steps (noise removal, stop words removal, stemming, lemmatization, term frequency). This research studies the effect of preprocessing steps on the performance of supervised spam classifier algorithms. Experiments were conducted on two widely used supervised spam classifier algorithms: Naïve Bayes and Support Vector Machine. The evaluation is performed on the Ling-spam corpus dataset and uses evaluation metrics: accuracy. The experimental results show that different preprocessing steps give different effects to different classifier.

Detecting Spam Review through Spammer’s Behavior Analysis

Online reviews about the purchase of a product or services provided have become the main source of user opinions. To gain profit or fame usually spam reviews are written to promote or demote some target products or services. This practice is known as review spamming. In the last few years, different methods have been suggested to solve the problem of review spamming but there is still a need to introduce new spam review detection method to improve accuracy results. In this work, researchers have studied six different spammer behavioral features and analyzed the proposed spam review detection method using weight method. An experimental evaluation was conducted on a benchmark dataset and achieved 84.5% accuracy.

A unified framework for detecting author spamicity by modeling review deviation

The success of e-commerce firms is highly dependent on the increasing number of customer reviews. However, to gain profit or fame, people may try to challenge the system by writing deceptive reviews that unjustly promote and/or demote target products or services. In this paper, a unified unsupervised framework is proposed to address the problem of opinion spamming. The rationale is that although not all outlier reviews are spam, spammers usually exhibit abnormities and deviations from normal users on certain dimensions concerning the same or even many products, thereby increasing their corresponding degrees of spamming (called “spamicity” in this paper). We introduce a set of abnormity signals from a review deviation angle and also present in detail an aspect-based review deviation dimension to model latent content deviation. Afterwards, a joint review deviation divergence is computed and ranked for detecting final opinion reviewer spamicity. Results of experiments conducted on a real-life Amazon review dataset demonstrate the effectiveness of the proposed approach.

Removing non-relevant links from top search results using feature score computation

To promote website in search engine rankings in order to get better visibility and more traffic, search engine optimizers have to follow legal ways but some search engine optimizers manipulate the web pages and boost the irrelevant pages in search results and this leads to practice the problem of web spam. These malicious pages reduce the performance of the search engine. We innovates a method to override the irrelevant search results. In this paper a new approach is designed and developed to move non-deserving pages downward from list of results returned by search engine. The first step was to determine the importance of different page features used for the ranking of a page in search results. Based on this information the devised system identified the features in a page and assigned some weight to each feature. The user query i.e. search keyword, precision and ranking factor are used to calculate precision score of a page. This score is used to predict ranking position for each page. A comparison was made between original ranking positions and predicted ranking position of irrelevant web pages. From analysis of results of Google, accuracy of search results improved from 88% to 99% for the corpus compiled now.

Detecting opinion spams and fake news using text classification

In recent years, deceptive content such as fake news and fake reviews, also known as opinion spams, have increasingly become a dangerous prospect for online users. Fake reviews have affected consumers and stores alike. Furthermore, the problem of fake news has gained attention in 2016, especially in the aftermath of the last U.S. presidential elections. Fake reviews and fake news are a closely related phenomenon as both consist of writing and spreading false information or beliefs. The opinion spam problem was formulated for the first time a few years ago, but it has quickly become a growing research area due to the abundance of user‐generated content. It is now easy for anyone to either write fake reviews or write fake news on the web. The biggest challenge is the lack of an efficient way to tell the difference between a real review and a fake one; even humans are often unable to tell the difference. In this paper, we introduce a new n‐gram model to detect automatically fake contents with a particular focus on fake reviews and fake news. We study and compare 2 different features extraction techniques and 6 machine learning classification techniques. Experimental evaluation using existing public datasets and a newly introduced fake news dataset indicate very encouraging and improved performances compared to the state‐of‐the‐art methods.

Design of a Machine Learning Based Predictive Analytics System for Spam Problem

Design of a Machine Learning Based Predictive Analytics System for Spam Problem

Behavioural account-based features for filtering out social spammers in large-scale twitter data collections

Online social networks (OSNs) have become an important source of information for a tremendous range of applications and researches. However, the high usability and accessibility of OSNs have exposed many information quality (IQ) problems which consequently decrease the performance of OSNs dependent applications. Social spammers are a particular kind of ill-intentioned users who degrade the quality of OSNs information through misusing all possible services provided by OSNs. Given the fact that Twitter is not immune towards the social spam problem, different researchers have designed various detection methods of a spam content. Ho-wever, the tweet-based detection methods are not effective for detecting a spam content because of the dynamicity and the fast evolution of spam. Moreover, the robust account-based features are costly for extraction because of the need for huge volume of data from Twitter’s servers, while most other account-based features don’t model the behavior of social spammers. Hence, in this paper, we introduce a design of new 10 robust behavioral account-based features for filte-ring out spam accounts existing in large-scale Twitter crawled data collections. Our features focus on modeling the behavior of social spammers, such as the time correlation among tweets. The experimental results show that our new behavioral features are able to correctly classify the majority of social spammers (spam accounts), outperforming 75 account-based features de-signed in the literature.

A novel ensemble decision tree based on under-sampling and clonal selection for web spam detection

Currently, web spamming is a serious problem for search engines. It not only degrades the quality of search results by intentionally boosting undesirable web pages to users, but also causes the search engine to waste a significant amount of computational and storage resources in manipulating useless information. In this paper, we present a novel ensemble classifier for web spam detection which combines the clonal selection algorithm for feature selection and under-sampling for data balancing. This web spam detection system is called USCS. The USCS ensemble classifiers can automatically sample and select sub-classifiers. First, the system will convert the imbalanced training dataset into several balanced datasets using the under-sampling method. Second, the system will automatically select several optimal feature subsets for each sub-classifier using a customized clonal selection algorithm. Third, the system will build several C4.5 decision tree sub-classifiers from these balanced datasets based on its specified features. Finally, these sub-classifiers will be used to construct an ensemble decision tree classifier which will be applied to classify the examples in the testing data. Experiments on WEBSPAM-UK2006 dataset on the web spam problem show that our proposed approach, the USCS ensemble web spam classifier, contributes significant classification performance compared to several baseline systems and state-of-the-art approaches.

A Topic-Based Hidden Markov Model for Real-Time Spam Tweets Filtering

Online social networks (OSNs) have become an important source of information for a tremendous range of applications and researches such as search engines, and summarization systems. However, the high usability and accessibility of OSNs have exposed many information quality (IQ) problems which consequently decrease the performance of the OSNs dependent applications. Social spammers are a particular kind of ill-intentioned users who degrade the quality of OSNs information through misusing all possible services provided by OSNs. Social spammers spread many intensive posts/tweets to lure legitimate users to malicious or commercial sites containing malware downloads, phishing, and drug sales. Given the fact that Twitter is not immune towards the social spam problem, different researchers have designed various detection methods which inspect individual tweets or accounts for the existence of spam contents. However, although of the high detection rates of the account-based spam detection methods, these methods are not suitable for filtering tweets in the real-time detection because of the need for information from Twitter’s servers. At tweet spam detection level, many light features have been proposed for real-time filtering; however, the existing classification models separately classify a tweet without considering the state of previous handled tweets associated with a topic. Also, these models periodically require retraining using a ground-truth data to make them up-to-date. Hence, in this paper, we formalize a Hidden Markov Model (HMM) as a time-dependent model for real-time topical spam tweets filtering. More precisely, our method only leverages the available and accessible meta-data in the tweet object to detect spam tweets exiting in a stream of tweets related to a topic (e.g., #Trump), with considering the state of previously handled tweets associated to the same topic. Compared to the classical time-independent classification methods such as Random Forest, the experimental evaluation demonstrates the efficiency of increasing the quality of topics in terms of precision, recall, and F-measure performance metrics.

Email spam origins: does the CAN SPAM act shift spam beyond United States jurisdiction?

Spam is the sending of bulk, commercial, unsolicited electronic marketing messages. Today email spam makes up the majority of all email sent worldwide. The CAN SPAM Act is U.S. federal legislation that was passed in response to the growing spam problem. However, cybercrime such as email spam does not respect national borders, and spam can be sent from anywhere in the world, so it is questioned whether the CAN SPAM Act simply shifted spam beyond United States jurisdiction. The present research sought to address this question by analyzing the impact that prosecutions of spammers has had on a sample of 5,490,905 spam emails sent between 1998 and 2013. Two measures of spam nation of origin, U.S. IP address and U.S. domain name among spam emails, were created to assess the possible impact that the CAN SPAM Act might have had on the nation of origin associated with spam email messages. Results suggest a possible impact, with more ongoing CAN SPAM Act trials and media attention predicting less spam associated with the United States. CAN SPAM activity appeared to have the largest impact on US domain name registrations. Implications of the findings and suggestions for future research are discussed.

CAPTCHA Cracker using MATLAB

is a lot of activity over the internet. Be it, posting a comment on someones blog or making a Gmail account or booking a ticket online. But with that also comes the problem of spamming. Completely Automated Public Turing test to tell Computers and Humans Apart (CAPTCHA) (8) which are the twisted words that block the entries of bots on website. CAPTHCAs can effectively test if the user is human or machine. Hence it is of great importance that CAPTCHA is well checked for its vulnerability against such attacks. So this paper presents this medium to check the strength of CAPTCHA against the written CAPTCHA cracking code. This can be used by the web developers implementing CAPTCHA, to check well in advance how secure is the CAPTCHA used in their software.

Efficient decision support for detecting content polluters on social networks: an approach based on automatic knowledge acquisition from behavioral patterns

With the rise of social networking services such as Facebook and Twitter, the problem of spam and content pollution has become more significant and intractable. Using social networking services, users are able to develop relationships and share messages with others in a very convenient manner; however, they are vulnerable to receiving spam messages. The automatic detection of spammers or content polluters on the network can effectively reduce the burden on the service provider in making a decision on appropriate counteractions. Content polluters can be automatically identified by using the supervised learning technique of artificial intelligence. To build a classification model with high accuracy automatically from the training data set, it is important to identify a set of useful features that can classify polluters and non-polluters. Moreover, because we deal with a huge amount of raw data in this process, the efficiency of data preparation and model creation are also critical issues that need to be addressed. In this paper, we present an efficient method for detecting content polluters on Twitter. Specifically, we propose a set of features that can be easily extracted from the messages and behaviors of Twitter users and construct a new breed of classifiers based on these features. The proposed approach requires only a minimal number of feature values per Twitter user and thus adds considerably less time to the overall mining process compared to other methods. Experiments confirm that the proposed approach outperforms previous approaches in both classification accuracy and processing time.

An Adaptive and Collaborative Server-Side SMS Spam Filtering Scheme Using Artificial Immune System

ABSTRACTSMS Spam, which is an unsolicited or unwanted message, is a major problem with Global System for Mobile Communication (GSM) subscribers. Existing Spam filters have not been able to stop the SMS Spam problem due to frequent drift in spammer’s words, limited bag of words for training, device portability, and high computational overhead of filters. This paper presents a collaborative and adaptive server-side SMS Spam filter using Artificial Immune System (coined ExAIS_SMS). The proposed scheme involves five modules: the innate mechanism, the user feedback, the quarantine, the tokenizer, and the immune engine. In this study, a new English corpus consisting of 5,240 SMS messages from 20 different users was collected for the study. A comprehensive experimental analysis on the SMS data set reveals the constant changes of Spam keywords and the impact of user feedback for system adaptability. In order to prove the efficiency of the proposed scheme, ExAIS_SMS was benchmarked with existing systems using the NUS corpus. The result gave an overall accuracy of 99% for ExAIS_SMS, 98% for Bayesian, and 97% for a client side AIS. The results showed that ExAIS_SMS is an efficient SMS Spam filtering technique, especially in resource constrained mobile phones.

A Comprehensive Study of Email Spam Botnet Detection

The problem of email spam has grown significantly over the past few years. It is not just a nuisance for users but also it is damaging for those who fall for scams and other attacks. This is due to the complexity intensification of email spamming techniques which are advancing from traditional spamming (direct spamming) techniques to a more scalable, elusive and indirect approach of botnets for distributing email spam messages. In this paper, we first discuss the sources and architectures used by the spamming botnets for sending massive amount of email spam. Then we present detailed chronicles of spamming botnets which systematically describes the timeline of events and notable occurrences in the advancement of these spamming botnets. This paper also aims to represent a comprehensive analysis of particular email spamming botnet detection techniques proposed in the literature. We attempt to categorize them according to both their nature of defense and method of detection, also revealing and comparing their advantages and disadvantages extensively. We also present a qualitative analysis of these techniques. Finally, we summarize the future trends and challenges in detecting email spamming botnets.

Evaluation of the CAN SPAM Act: Testing Deterrence and Other Influences of E-mail Spammer Legal Compliance Over Time

E-mail spam has been growing since its inception. The Controlling the Assault of Non-Solicited Pornography and Marketing Act (CAN SPAM Act) is U.S. federal legislation that was passed in response to the growing spam problem. A series of evaluations followed after the Act, most of which reported that compliance with the Act’s requirements among spammers had not been affected. However, none of these evaluations used methods that were sufficiently rigorous, failing to capture the continuous nature of CAN SPAM Act’s enforcement, using a limited number of measures of noncompliance, and ignoring a variety of possible spurious influences. This research addresses all of these limitations by analyzing a sample of 5,490,905 spam e-mails sent between 1998 and 2013. Ten measures of spammer compliance with the CAN SPAM Act were operationalized to test the impact the Act had. Thirteen measures of CAN SPAM Act enforcement were coded from news articles and included in a time-series regression. Findings suggest that the Act may in fact be a deterrent, but in such a way as to increase CAN SPAM Act violations of header forgery as a precaution against being caught.

Improved email spam detection model with negative selection algorithm and particle swarm optimization

The adaptive nature of unsolicited email by the use of huge mailing tools prompts the need for spam detection. Implementation of different spam detection methods based on machine learning techniques was proposed to solve the problem of numerous email spam ravaging the system. Previous algorithm used in email spam detection compares each email message with spam and non-spam data before generating detectors while our proposed system inspired by the artificial immune system model with the adaptive nature of negative selection algorithm uses special features to generate detectors to cover the spam space. To cope with the trend of email spam, a novel model that improves the random generation of a detector in negative selection algorithm (NSA) with the use of stochastic distribution to model the data point using particle swarm optimization (PSO) was implemented. Local outlier factor is introduced as the fitness function to determine the local best (Pbest) of the candidate detector that gives the optimum solution. Distance measure is employed to enhance the distinctiveness between the non-spam and spam candidate detector. The detector generation process was terminated when the expected spam coverage is reached. The theoretical analysis and the experimental result show that the detection rate of NSA–PSO is higher than the standard negative selection algorithm. Accuracy for 2000 generated detectors with threshold value of 0.4 was compared. Negative selection algorithm is 68.86% and the proposed hybrid negative selection algorithm with particle swarm optimization is 91.22%.