A semantic-based model with a hybrid feature engineering process for accurate spam detection

Email communication has been stated to be the most affordable, economical, and quickest communication way nowadays. Although there are significant perks of emails, sadly, its practice has been baffled by the vast amount of unsolicited and often deceitful emails. And these extreme amounts of spam are decreasing the essence of information present on the Internet and causing concern among users. Many spam detection models have been suggested and experimented with within the literature; however, the listed accuracy showed that more could be done in this area to improve accuracy. This work describes how to detect spam emails using machine learning based on words, numbers, and characters in the emails’ content. We have used some prevalent machine learning models (Naive Bayes’, Neural Network, K-NN, Tree, Logistic Regression) and compared them to classify emails. This paper’s primary focus is obtaining optimal accuracy with the help of limited features out of 57.KeywordsMachine learning in emailSpam emailsSpam emails detection

Spam emails have recently become a concern on the Internet. Machine learning techniques such as Neural Networks, Naïve Bayes, and Decision Trees have frequently been used to combat these spam emails. Despite their efficiency, time complexity in high-dimensional datasets remains a significant challenge. Due to a large number of features in high-dimensional datasets, the intricacy of this problem grows exponentially. The existing approaches suffer from a computational burden when thousands of features are used (high-time complexity). To reduce time complexity and improve accuracy in high-dimensional datasets, extra steps of feature selection and parameter tuning are necessary. This work recommends the use of a hybrid logistic regression model with a feature selection approach and parameter tuning that could effectively handle a big dimensional dataset. The model employs the Term Frequency-Inverse Document Frequency (TF-IDF) feature extraction method to mitigate the drawbacks of Term Frequency (TF) to obtain an equal feature weight. Using publicly available datasets (Enron and Lingspam), we compared the model’s performance to that of other contemporary models. The proposed model achieved a low level of time complexity while maintaining a high level of spam detection rate of 99.1%.

Spam email detection is a critical component of securing and maintaining reliable digital communication systems. This study explores the effectiveness of various machine learning algorithms in classifying spam, with an emphasis on enhancing accuracy and precision through systematic preprocessing, advanced feature engineering, and text preprocessing. Six models were evaluated: Logistic Regression, Support Vector Classifier, Multinomial Naïve Bayes, K-Nearest Neighbors, AdaBoost, and Bagging Classifier using a comprehensive preprocessing pipeline that included Term Frequency–Inverse Document Frequency vectorization, feature scaling, and the incorporation of engineered features such as character counts. Experimental results reveal that Multinomial Naïve Bayes consistently achieved the highest precision 1.00 and strong accuracy 0.979 when paired with feature scaling, while Logistic Regression delivered robust and stable performance across multiple configurations with precision exceeding 0.96, making it a reliable choice for real-world deployment. Although Support Vector Classifier and AdaBoost exhibited competitive baseline performance, Support Vector Classifier showed limitations when handling numeric features, whereas AdaBoost maintained consistent results across scenarios. These findings underscore the critical role of tailored preprocessing and ensemble learning in improving classification outcomes and highlight the comparative strengths of different algorithms in real-world spam detection. In particular, Multinomial Naïve Bayes proved highly effective for precision-critical tasks, while Logistic Regression emerged as a dependable solution for environments requiring consistent reliability. Overall, this work advances machine learning-based spam filtering by identifying models that successfully balance precision, adaptability, and computational efficiency.

Text representation in a vector, known as embedding, is crucial for various classification tasks including sentiment analysis. It helps to process and understand natural language text more effectively. It has evolved from static approaches, such as bag-of-words and n-grams, to more dynamic approaches that consider the context and meaning of words, such as word embeddings and contextualized embeddings. Word embeddings use neural networks to learn vector representations of words based on their co-occurrence patterns in large text corpora. On the other hand, contextualized embeddings, such as BERT, consider the context of each word within a sentence or document to generate more nuanced representations. Numerous researchers have suggested modifying the original Word2Vec and BERT embeddings to include sentiment information. This paper provides a comprehensive overview of these methods by including a detailed discussion of various evaluation techniques. The paper also outlines several challenges related to embeddings that can be addressed in order to improve the results of sentiment analysis.

This Spam emails have become a severe challenge that irritates and consumes recipients' time. On the one hand, existing spam detection techniques have low detection rates and cannot tolerate high-dimensional data. Moreover, due to the machine learning algorithm's effectiveness in identifying mail as solicited or unsolicited, their approaches have become common in spam detection systems. This paper proposes a lightweight machine learning-based spam detection model based on Random Forest (RF) algorithm. According to the empirical results, the proposed model achieved a 97% accuracy on the spambase dataset. Furthermore, the performance of the proposed model was evaluated using standard classification metrics such as Fscore, Recall, Precision, and Accuracy. The comparison of Our model with state-of-the-art works investigated in this paper showed the model performs better, with an improvement of 6% for all metrics.

Logistic regression is a popular binary classification approach, but like any machine learning algorithms, it has its limitations and possible concerns such as class imbalance, large datasets, and overfitting, which reduce its accuracy and efficiency. This study enhanced the Logistic Regression algorithm's performance for email spam detection by addressing these problems using the techniques of Term Frequency-Inverse Document Frequency for class imbalance, Recursive Feature Elimination for large datasets, and Principal Component Analysis for overfitting concerns. TF-IDF improves feature representation, highlighting key terms that differentiate spam from non-spam. RFE systematically eliminates irrelevant features, reducing computational complexity and enhancing efficiency, particularly for large datasets. PCA mitigates overfitting by reducing the dimensionality of feature spaces, ensuring the model generalizes effectively to unseen data. The enhanced Logistic Regression model demonstrated a significant improvement in spam detection accuracy, achieving up to 98% accuracy with TF-IDF. RFE reduced training time while maintaining robust performance on large datasets, and PCA improved model generalization, reducing overfitting risks. The proposed enhancements successfully address the key limitations of traditional Logistic Regression models in spam detection. This refined approach improves predictive accuracy, computational efficiency, and robustness, making it highly applicable to real-world email security systems.

Skip-Gram word embeddings, estimated from large text corpora, have been shown to improve many NLP tasks through their high-quality features. However, little is known about their robustness against parameter perturbations and about their efficiency in preserving word similarities under memory constraints. In this paper, we investigate three post-processing methods for word embeddings to study their robustness and memory efficiency. We employ a dimensionality-based, a parameter-based and a resolution-based method to obtain parameter-reduced embeddings and we provide a concept that connects the three approaches. We contrast these methods with the relative accuracy loss on six intrinsic evaluation tasks and compare them with regard to the memory efficiency of the reduced embeddings. The evaluation shows that low Bit-resolution embeddings offer great potential for memory savings by alleviating the risk of accuracy loss. The results indicate that post-processed word embeddings could also enhance applications on resource limited devices with valuable word features.

We consider the problem of joint modeling of videos and their corresponding textual descriptions (e.g. sentences or phrases). Our approach consists of three components: the video representation, the textual representation, and a joint model that links videos and text. Our video representation uses the state-of-the-art deep 3D ConvNet to capture the semantic information in the video. Our textual representation uses the recent advancement in learning word and sentence vectors from large text corpus. The joint model is learned to score the correct (video, text) pairs higher than the incorrect ones. We demonstrate our approach in several applications: 1) retrieving sentences given a video; 2) retrieving videos given a sentence; 3) zero-shot action recognition in videos.

In the digital world a lot of emails are received every day, and most of them are not of any relevance to us, some are containing suspicious links which can cause harm to our system in some way or other. This can be overcome by using spam detection. It is the process of classifying whether the email is a genuine one or if it is some kind of spam. The purpose of spam detection is to deliver relevant emails to the person and separate spam emails. Already every email service provider has spam detection but still, its accuracy is not that much, sometimes they classify useful emails as spam. This paper focuses on the comparative analysis approach, where various Machine Learning models are applied to the same dataset. The different machine learning models were compared based on accuracy and Precision. Support vector machine results in 98.09% accuracy.

In this paper, we analyze word embeddings for a morphologically rich language, Tigrinya. Tigrinya is a Semitic language spoken natively in Eritrea and Ethiopia by over seven million people. The unique and complex morphology of Semitic languages, which includes Arabic, Amharic, and Hebrew, is commonly known as 'root and template pattern' morphology. This morphology generates a large number of inflected forms that often cause out-of-vocabulary (OOV) challenges in language processing. This problem is more challenging for low resource languages, such as Tigrinya, that offers very little support of annotated resources. Word embedding methods, given a large raw text corpus, form semantic and syntactic vector representation of words. Therefore, we construct a new text corpus and investigate the optimal settings for generating word vectors for Tigrinya. We also utilize word embeddings to improve the performance of a Tigrinya part-of-speech tagger created from a small tagged corpus.

The rapid advancement of information and communication technology has transformed the way humans interact and exchange information. Among various digital communication tools, email remains one of the most widely used; however, it is often exploited to send spam messages. Spam emails can contain phishing links, malware, or unsolicited advertisements, posing significant risks to individuals and organizations. Therefore, developing accurate and efficient spam detection methods is becoming increasingly important. This study proposes a lightweight and efficient spam email classification approach using the naïve Bayes algorithm combined with TF-IDF feature extraction and the synthetic minority oversampling technique (SMOTE) to address class imbalance. A series of preprocessing steps tokenization, lemmatization, stopword removal, and term frequency-inverse document frequency (TF-IDF) transformation were applied to normalize and vectorize email text data. The SMOTE technique was applied precisely to the training dataset to balance the class distribution and avoid data leakage during evaluation. Experimental results showed that the naïve Bayes model initially achieved 88% accuracy, 86% recall, 100% precision, and 92% F1 score. After proper application of SMOTE, the model achieved 100% accuracy, precision, recall, and F1 score, indicating perfect classification of spam and non-spam (ham) emails. These results confirm that proper class balancing significantly improves the model’s ability to detect spam emails. Overall, this study highlights the effectiveness of combining TF-IDF, naïve Bayes, and SMOTE as a robust yet computationally efficient solution for modern spam detection, particularly suited to real-time and resource-constrained environments.

Generative text summary is an important branch of natural language processing. Aiming at the problems of insufficient use of semantic information, insufficient summary precision and the problem of semantics-loss in the current generated text summary method, an enhanced semantic model is proposed based on dual-encoder, which can provide richer semantic information for sequence-to-sequence architecture through dual-encoder. The enhanced attention architecture with dual-channel semantics is optimized, and the empirical distribution and Gain-Benefit gate are built for decoding. In addition, the position embedding and word embedding are merged into the word embedding technology, and the TF-IDF(term frequency-inverse document frequency), part of speech, key score are added to word’s feature. Meanwhile, the optimal dimension of word embedding is optimized. This paper aims to optimize the traditional sequence mapping and word feature representation, enhance the model’s semantic understanding, and improve the quality of the summary. The LCSTS and SOGOU datasets are used to validate proposed method. The experimental results show that the proposed method can improve the performance of the ROUGE evaluation system by 10–13 percentage points compared with other listed algorithms. We can observe that the semantic understanding of the text summaries is more accurate and the generation effect is better, which has a better application prospect.

Neural machine translation has significantly pushed forward the quality of the field. However, there are remaining big issues with the output translations and one of them is fairness. Neural models are trained on large text corpora which contain biases and stereotypes. As a consequence, models inherit these social biases. Recent methods have shown results in reducing gender bias in other natural language processing tools such as word embeddings. We take advantage of the fact that word embeddings are used in neural machine translation to propose a method to equalize gender biases in neural machine translation using these representations. Specifically, we propose, experiment and analyze the integration of two debiasing techniques over GloVe embeddings in the Transformer translation architecture. We evaluate our proposed system on the WMT English-Spanish benchmark task, showing gains up to one BLEU point. As for the gender bias evaluation, we generate a test set of occupations and we show that our proposed system learns to equalize existing biases from the baseline system.

Background: Text classification techniques are increasingly important with the exponential growth of textual data on the internet. Term Frequency-Inverse Document Frequency (TF-IDF) and Count Vectorizer(CV) are commonly used methods for feature extraction. TF-IDF assigning weights to terms based on their frequency. CV simply counts the occurrences of terms. The performance of CV as well as TF-IDF are evaluated and compared with KNN and DT classifiers across text datasets. Methodology: The investigation begins with preprocessing. The feature vectors are created using both TF-IDF and CV. Feature vectors are passed into the KNN and DT classifiers at in training stage. Experiments are executed the usage of Kaggle's public database Ukraine 10K tweets sentiment_analysis dataset and the Womens ecommerce clothing reviews dataset. Findings: The average of precision, recall, f1 score and accuracy of KNN with TF-IDF were 84.5%, 87%, 83%, 87% respectively and KNN with CV were 83.5%, 87%, 83.5%, 87% respectively. Similarly, average of precision, recall, f1 score and accuracy of DT with TF-IDF were 89%, 89%, 89%, 89% respectively and DT with CV were 89%, 89.5%, 89.5%, 89.5% respectively. The results obtained in this research is consistent with previous similar research result. Conclusions: The performance of TF-IDF is almost similar as CV for a particular dataset and a particular classifier in this study. Novelty: The experiment performed using these classifiers and feature extraction methods on the datasets is a novelty and contribution of this research.

For information retrieval and text-mining, a robust scalable framework is required to represent the information extracted from documents and enable visualization and query of such information. One very widely used model is the vector space model which is based on the bag-of-words approach. However, it suffers from the fact that it loses important information about the original text, such as information about the order of the terms in the text or about the frontiers between sentences or paragraphs. In this paper, we propose a graph-based text representation, which is capable of capturing (i) Term order (ii) Term frequency (iii) Term co-occurrence (iv) Term context in documents. We also apply the graph model into our text mining task, which is to discover unapparent associations between two and more concepts (e.g. individuals) from a large text corpus. Counterterrorism corpus is used to evaluate the performance of various retrieval models, which demonstrates feasibility and effectiveness of graphic text representation in information retrieval and text mining.