The Impact of Input Types on Smart Contract Vulnerability Detection Performance Based on Deep Learning: A Preliminary Study

This paper aims to explore the application of deep learning in smart contract vulnerabilities detection. Smart contracts are an essential part of blockchain technology and are crucial for developing decentralized applications. However, smart contract vulnerabilities can cause financial losses and system crashes. Static analysis tools are frequently used to detect vulnerabilities in smart contracts, but they often result in false positives and false negatives because of their high reliance on predefined rules and lack of semantic analysis capabilities. Furthermore, these predefined rules quickly become obsolete and fail to adapt or generalize to new data. In contrast, deep learning methods do not require predefined detection rules and can learn the features of vulnerabilities during the training process. In this paper, we introduce a solution called Lightning Cat which is based on deep learning techniques. We train three deep learning models for detecting vulnerabilities in smart contract: Optimized-CodeBERT, Optimized-LSTM, and Optimized-CNN. Experimental results show that, in the Lightning Cat we propose, Optimized-CodeBERT model surpasses other methods, achieving an f1-score of 93.53%. To precisely extract vulnerability features, we acquire segments of vulnerable code functions to retain critical vulnerability features. Using the CodeBERT pre-training model for data preprocessing, we could capture the syntax and semantics of the code more accurately. To demonstrate the feasibility of our proposed solution, we evaluate its performance using the SolidiFI-benchmark dataset, which consists of 9369 vulnerable contracts injected with vulnerabilities from seven different types.

Sentiment analysis of news headlines is an important factor that investors consider when making investing decisions. We claim that the sentiment analysis of financial news headlines impacts stock market values. Hence financial news headline data are collected along with the stock market investment data for a period of time. Using Valence Aware Dictionary and Sentiment Reasoning (VADER) for sentiment analysis, the correlation between the stock market values and sentiments in news headlines is established. In our experiments, the data on stock market prices are collected from Yahoo Finance and Kaggle. Financial news headlines are collected from the Wall Street Journal, Washington Post, and Business-Standard website. To cope with such a massive volume of data and extract useful information, various embedding methods, such as Bag-of-words (BoW) and Term Frequency-Inverse Document Frequency (TF-IDF), are employed. These are then fed into machine learning models such as Naive Bayes and XGBoost as well as deep learning models such as Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM). Various natural language processing, andmachine and deep learning algorithms are considered in our study to achieve the desired outcomes and to attain superior accuracy than the current state-of-the-art. Our experimental study has shown that CNN (80.86%) and LSTM (84%) are the best performing models in relation to machine learning models, such as Support Vector Machine (SVM) (50.3%), Random Forest (67.93%), and Naive Bayes (59.79%). Moreover, two novel methods, BERT and RoBERTa, were applied with the expectation of better performance than all the other models, and they did exceptionally well by achieving an accuracy of 90% and 88%, respectively.

Information and communication technology based inter-organizational systems enable companies to integrate information and conduct business electronically across different parts of the organization. For organizations embracing blockchain, smart contracts provide automation and operational efficiency for inter-organizational systems. Initially utilised for financial transactions, smart contract are extended beyond banking and deployed in wide number of organizations. Smart contracts are regarded as self-executing type of contract consisting of agreement’s terms embedded directly into the code which plays a vital role in operability for inter-organizational systems, however, smart contract vulnerabilities can arise due to programming errors, leading to security issues. The effects of smart contract vulnerabilities can be significant, including loss of funds, unauthorized access to sensitive information, manipulation of data, and loss of trust in the application leading to catastrophic financial losses followed by legal implications for an organization based on blockchain technology. The goal of smart contracts exploiting vulnerabilities is to discover and eliminate potential security vulnerabilities in smart contract code prior to it being deployed. Detecting vulnerabilities in a timely manner helps to prevent financial losses, unauthorized access, and data manipulation. In order to provide a robust solution to detect vulnerabilities in smart contracts, the proposed methodology presents a novel approach for rapid detection of vulnerabilities by integrating genetic algorithm with isolation forest. Furthermore, enhancing smart contract vulnerability identification with higher accuracy and false-positive rate provides a reliable gateway for organizations to adopt blockchain.

The blockchain 2.0 age, marked by smart contract and Ethereum, has arrived couple years ago. Its technologies have expanded the application scenarios of blockchain technology and driven the boom of decentralized Finance. However, smart contract vulnerabilities and security issues are also emerging one after another. Hackers have exploited these vulnerabilities to cause huge economic losses. In recent years, a large amount of research on the analysis and detection of smart contract vulnerabilities has emerged, but there has been no common detection tool and corresponding test dataset. In this paper, we build GSVD dataset (Generalized Smart Contract Vulnerability Dataset) consisting four offline datasets using smart contracts on two chains, Polygon and BSC: two small Solidity datasets consisting of 153 labeled smart contract source codes, which can be used to test the performance of vulnerability mining tools; two large Solidity datasets consisting of 52,202 un labeled real smart contract source codes that can be used to verify the correctness of various theories and tools under a large number of real data conditions. At the same time, this paper integrates the scripting framework accompanying the GSVD dataset, which can execute a variety of popular automated vulnerability detection tools on top of these datasets and generate analysis results of contracts and potential vulnerabilities. We tested the Minor dataset under GSVD using three tools (Slither, Manticore, Mythril) that are kept up to date and found that the combined use of all tools detected 61.1% of labeled vulnerabilities, of which Mythril has the highest detection rate of 42.6%. It is not difficult to conclude that there`re still ample room for advancement for current smart contract vulnerability mining tools because of their underlying methods. Besides, our dataset can contribute to the ultimate target greatly by providing mining tools plenty real contracts information.

Smart contracts are programs running on blockchain. In recent years, due to the persistent occurrence of security-related accidents in smart contracts, the effective detection of vulnerabilities in smart contracts has received extensive attention from researchers and engineers. Machine learning-based vulnerability detection techniques have the advantage that they do not need expert rules for determining vulnerabilities. However, existing approaches cannot identify vulnerabilities when the versions of smart contract compilers are updated. In this paper, we propose OC-Detector (Opcode Clustering Detector), a smart contract vulnerability detection approach based on clustering opcode instructions. OC-Detector learns the characteristics of opcode instructions to cluster them and replaces opcode instructions belonging to the same cluster with the ID of the cluster. After that, the similarity between the contract under analysis and contracts in the vulnerability database is calculated to identify vulnerabilities. The experimental results demonstrate that OC-Detector improves the F1 value of detecting vulnerabilities from 0.04 to 0.40 compared to DC-Hunter, Securify, SmartCheck and Osiris. Additionally, compared to DC-Hunter, the F1 value is improved by 0.27 when detecting vulnerabilities in smart contracts compiled by different versions of compilers.

Smart contracts are decentralized applications running on the blockchain to meet various practical scenario demands. The increasing number of security events regarding smart contracts have led to huge pecuniary losses and destroyed the ecological stability of contract layer on the blockchain. Faced with the increasing quantity of contracts, it is an emerging issue to effectively and efficiently detect vulnerabilities in smart contracts. Existing methods of detecting vulnerabilities in smart contracts like Oyente mainly employ symbolic execution. This method is very time-consuming, as the symbolic execution requires the exploration of all executable paths in a contract. In this work, we propose an efficient model for the detection of vulnerabilities in Ethereum smart contracts with machine learning techniques. The model is able to effectively and fast detect vulnerabilities based on the patterns learned from training samples. Our model is evaluated on 49502 real-world smart contracts and the results verify its effectiveness and efficiency.

With the increasing popularity of Decentralized Applications (DApps) in blockchain, securing smart contracts has been a long-term, high-priority subject in the domain. Among the various research directions for vulnerability detection, fuzzing has received extensive attention because of its high effectiveness. However, with the increasing complexity of smart contracts, existing fuzzers may waste substantial time exploring locations irrelevant to smart contract vulnerabilities. In this article, we present FunFuzz, a function-oriented fuzzer, which is dedicatedly tailored for detecting smart contract vulnerability with high effectiveness and efficiency. The key observation in our research is that most smart contract vulnerabilities exist in specific functions rather than randomly distributed in all program code like other traditional software. To this end, unlike traditional fuzzers which mainly target code coverage, FunFuzz identifies risky functions while pruning non-risky ones in smart contracts. In this way, it significantly narrows down the exploration scope during the fuzzing process. In addition, FunFuzz employs three unique strategies to direct itself toward effectively discovering vulnerabilities specific to smart contracts (e.g., reentrancy, block dependency, and gasless send). Extensive experiments on 170 real-world contracts demonstrate that FunFuzz outperforms state-of-the-art fuzzers in terms of effectiveness and efficiency.

Smart contracts are frequently targeted by hackers because they hold large amounts of money and cannot be modified once they are published. Existing detection methods mainly focus on known vulnerabilities with clear features and cannot deal with unknown vulnerabilities. As a consequence, proposing a method for detecting unknown vulnerabilities in smart contracts represents a significant advancement in the field of smart contract security. Aiming at this problem, based on the idea that the opcode sequences of transactions containing unknown vulnerabilities have similarities to the opcode sequences of transactions containing known vulnerabilities, we propose a novel approach for unknown vulnerability detection in smart contracts using a CNN-BiLSTM model. Our model determines whether a vulnerability is unknown by detecting the opcode sequence representing the entire execution process of a transaction. Experimental results with the opcode sequences of transactions show that the model can achieve 82.86% accuracy and 83.63% F1-score.KeywordsSmart contractsUnknown vulnerabilitiesVulnerability detectionCNN-BiLSTMOpcode sequences

Sentiment analysis (SA) detects people’s opinions from text engaging natural language processing (NLP) techniques. Recent research has shown that deep learning models, i.e., Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Transformer-based provide promising results for recognizing sentiment. Nonetheless, CNN has the advantage of extracting high-level features by using convolutional and max-pooling layers; it cannot efficiently learn a sequence of correlations. At the same time, Bidirectional RNN uses two RNN directions to improve extracting long-term dependencies. However, it cannot extract local features in parallel, and Transformer-based like Bidirectional Encoder Representations from Transformers (BERT) are the computational resources needed to fine-tune, facing an overfitting problem on small datasets. This paper proposes a novel attention-based model that utilizes CNNs with LSTM (named ACL-SA). First, it applies a preprocessor to enhance the data quality and employ term frequency-inverse document frequency (TF-IDF) feature weighting and pre-trained Glove word embedding approaches to extract meaningful information from textual data. In addition, it utilizes CNN’s max-pooling to extract contextual features and reduce feature dimensionality. Moreover, it uses an integrated bidirectional LSTM to capture long-term dependencies. Furthermore, it applies the attention mechanism at the CNN’s output layer to emphasize each word’s attention level. To avoid overfitting, the Guasiannoise and GuasianDroupout are adopted as regularization. The model’s robustness is evaluated on four English standard datasets, i.e., Sentiment140, US-airline, Sentiment140-MV, SA4A with various performance matrices, and compared efficiency with existing baseline models and approaches. The experiment results show that the proposed method significantly outperforms the state-of-the-art models.

Recently, fake news has been widely spread through the Internet due to the increased use of social media for communication. Fake news has become a significant concern due to its harmful impact on individual attitudes and the community’s behavior. Researchers and social media service providers have commonly utilized artificial intelligence techniques in the recent few years to rein in fake news propagation. However, fake news detection is challenging due to the use of political language and the high linguistic similarities between real and fake news. In addition, most news sentences are short, therefore finding valuable representative features that machine learning classifiers can use to distinguish between fake and authentic news is difficult because both false and legitimate news have comparable language traits. Existing fake news solutions suffer from low detection performance due to improper representation and model design. This study aims at improving the detection accuracy by proposing a deep ensemble fake news detection model using the sequential deep learning technique. The proposed model was constructed in three phases. In the first phase, features were extracted from news contents, preprocessed using natural language processing techniques, enriched using n-gram, and represented using the term frequency–inverse term frequency technique. In the second phase, an ensemble model based on deep learning was constructed as follows. Multiple binary classifiers were trained using sequential deep learning networks to extract the representative hidden features that could accurately classify news types. In the third phase, a multi-class classifier was constructed based on multilayer perceptron (MLP) and trained using the features extracted from the aggregated outputs of the deep learning-based binary classifiers for final classification. The two popular and well-known datasets (LIAR and ISOT) were used with different classifiers to benchmark the proposed model. Compared with the state-of-the-art models, which use deep contextualized representation with convolutional neural network (CNN), the proposed model shows significant improvements (2.41%) in the overall performance in terms of the F1score for the LIAR dataset, which is more challenging than other datasets. Meanwhile, the proposed model achieves 100% accuracy with ISOT. The study demonstrates that traditional features extracted from news content with proper model design outperform the existing models that were constructed based on text embedding techniques.

As a trusted decentralized application, smart contracts manage a large number of digital assets on the blockchain. Vulnerability detection of smart contracts is an important part of ensuring the security of digital assets. At present, many researchers extract features of smart contract source code for vulnerability detection based on deep learning methods. However, the current research mainly focuses on the single representation form of the source code, which cannot fully obtain the rich semantic and structural information contained in the source code, so it is not conducive to the detection of various and complex smart contract vulnerabilities. Aiming at this problem, this paper proposes a vulnerability detection model based on the fusion of syntax and semantic features. The syntactic and semantic representation of the source code is obtained from the abstract syntax tree and control flow graph of the smart contract through TextCNN and Graph Neural Network. The syntactic and semantic features are fused, and the fused features are used to detect vulnerabilities. Experiments show that the detection accuracy and recall rate of this model have been improved on the detection tasks of five types of vulnerabilities, with an average precision of 96% and a recall rate of 90%, which can effectively identify smart contract vulnerabilities.

Detecting cheating in online exams is a major challenge, not least to guarantee the originality and independence of answers. This paper presents a comparative analysis of three feature extraction methods for cheating detection based on similarity detection: Term frequencyinverse document frequency (TF-IDF), Bag of Words (BoW), and a new approach combining K-Shingling, MinHashing, and Locality Sensitive Hashing (LSH). We evaluate these methods in terms of their ability to accurately and efficiently identify similarities between student responses. Experimental results show that the K-Shingling, MinHashing, and LSH pipelines consistently outperform or match traditional approaches. Logistic regression and random forest classifiers with MinHashing + LSH achieve perfect scores of 1.00 in terms of precision, recall, F1 score, and accuracy, demonstrating the robustness and effectiveness of the method. In comparison, TF-IDF and BoW show mixed performance between classifiers, with notable limitations in terms of scalability and sensitivity to text variations. This study highlights the scalability and computational efficiency of the K-shingling, MinHashing, and LSH approaches, making them particularly suitable for large-scale online examination environments. By offering a detailed performance comparison, we demonstrate that K-shingling, MinHashing, and LSH provide a more reliable and efficient solution for detecting cheating in online exams, paving the way for greater academic integrity in digital education.

This study presents the development of a realistic and comprehensive SDN-ATK dataset designed to evaluate the effectiveness of machine learning (ML) and deep learning (DL) approaches for attack detection in Software-Defined Networking (SDN) environments. Unlike existing datasets, SDN-ATK explicitly includes attacks targeting key SDN components such as SDN controllers and OpenFlow switches, addressing a critical gap in current research. We evaluated three ML (XGBoost, Random Forest, and Decision Tree) and three DL (Convolutional Neural Network (CNN), Feed-forward Neural Network (FNN), and Long Short-Term Memory (LSTM)) algorithms across binary and multiclass classification tasks to assess detection performance. Our results demonstrate that DL models, particularly FNN and CNN outperform ML counterparts, achieving 98–99% accuracy, precision, and recall in binary classification. Explainability analyses were conducted using SHAP (SHapley Additive explanations) on the XGBoost model, offering valuable insights into the importance of feature and improving transparency in ML-based attack detection. The study’s findings provide critical guidance for both academia and industry, highlighting that within our Ryu-based SDN testbed, DL models demonstrated more reliable and balanced performance for large-scale attack detection. This work lays a solid foundation for future research, including developing real-time, intelligent, and explainable intrusion detection systems for SDN environments.

Prediction of the stage of cancer plays an important role in planning the course of treatment and has been largely reliant on imaging tools which do not capture molecular events that cause cancer progression. Gene-expression data-based analyses are able to identify these events, allowing RNA-sequence and microarray cancer data to be used for cancer analyses. Breast cancer is the most common cancer worldwide, and is classified into four stages - stages 1, 2, 3, and 4 [2]. While machine learning models have previously been explored to perform stage classification with limited success, multi-class stage classification has not had significant progress. There is a need for improved multi-class classification models, such as by investigating deep learning models. Gene-expression-based cancer data is characterised by the small size of available datasets, class imbalance, and high dimensionality. Class balancing methods must be applied to the dataset. Since all the genes are not necessary for stage prediction, retaining only the necessary genes can improve classification accuracy. The breast cancer samples are to be classified into 4 classes of stages 1 to 4. Invasive ductal carcinoma breast cancer samples are obtained from The Cancer Genome Atlas (TCGA) and Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) datasets and combined. Two class balancing techniques are explored, synthetic minority oversampling technique (SMOTE) and SMOTE followed by random undersampling. A hybrid feature selection pipeline is proposed, with three pipelines explored involving combinations of filter and embedded feature selection methods: Pipeline 1 - minimum-redundancy maximum-relevancy (mRMR) and correlation feature selection (CFS), Pipeline 2 - mRMR, mutual information (MI) and CFS, and Pipeline 3 - mRMR and support vector machine-recursive feature elimination (SVM-RFE). The classification is done using deep learning models, namely deep neural network, convolutional neural network, recurrent neural network, a modified deep neural network, and an AutoKeras generated model. Classification performance post class-balancing and various feature selection techniques show marked improvement over classification prior to feature selection. The best multiclass classification was found to be by a deep neural network post SMOTE and random undersampling, and feature selection using mRMR and recursive feature elimination, with a Cohen-Kappa score of 0.303 and a classification accuracy of 53.1%. For binary classification into early and late-stage cancer, the best performance is obtained by a modified deep neural network (DNN) post SMOTE and random undersampling, and feature selection using mRMR and recursive feature elimination, with an accuracy of 81.0% and a Cohen-Kappa score (CKS) of 0.280. This pipeline also showed improved multiclass classification performance on neuroblastoma cancer data, with a best area under the receiver operating characteristic (auROC) curve score of 0.872, as compared to 0.71 obtained in previous work, an improvement of 22.81%. The results and analysis reveal that feature selection techniques play a vital role in gene-expression data-based classification, and the proposed hybrid feature selection pipeline improves classification performance. Multi-class classification is possible using deep learning models, though further improvement particularly in late-stage classification is necessary and should be explored further.

Security attacks in smart contracts have drawn extensive attention due to the financial loss and erosion of trust caused by vulnerabilities. Even worse, smart contract is a tamper proof digital agreement and fixing bugs in it is difficult, so it is necessary for developers to detect security vulnerabilities in smart contract before deployment. Researchers have proposed several methods on smart contract vulnerability detection. However, despite the numerous vulnerability-finding tools, few of them have ideal detection performance because most of them rely on fixed rules, which is inefficient. In this paper, a modularized and systematic Deep Learning-based framework is proposed to automatically detect smart contracts vulnerability, called DeeSCVHunter. Particularly, we focus on two types of smart contract vulnerabilities: reentrancy and time dependence. And we propose a novel notion of Vulnerability Candidate Slice (VCS) to help models capture the key point of vulnerability. We conduct experiments on real-world dataset and the results prove the effectiveness of VCS, which leads to 25.76% improvement in F1-score at most. And extensive experiments also show that our proposed framework significantly outperforms state-of-the-art methods.