Text Datasets Research Articles

CoLAL: Co-learning Active Learning for Text Classification

In the machine learning field, the challenge of effectively learning with limited data has become increasingly crucial. Active Learning (AL) algorithms play a significant role in this by enhancing model performance. We introduce a novel AL algorithm, termed Co-learning (CoLAL), designed to select the most diverse and representative samples within a training dataset. This approach utilizes noisy labels and predictions made by the primary model on unlabeled data. By leveraging a probabilistic graphical model, we combine two multi-class classifiers into a binary one. This classifier determines if both the main and the peer models agree on a prediction. If they do, the unlabeled sample is assumed to be easy to classify and is thus not beneficial to increase the target model's performance. We prioritize data that represents the unlabeled set without overlapping decision boundaries. The discrepancies between these boundaries can be estimated by the probability that two models result in the same prediction. Through theoretical analysis and experimental validation, we reveal that the integration of noisy labels into the peer model effectively identifies target model's potential inaccuracies. We evaluated the CoLAL method across seven benchmark datasets: four text datasets (AGNews, DBPedia, PubMed, SST-2) and text-based state-of-the-art (SOTA) baselines, and three image datasets (CIFAR100, MNIST, OpenML-155) and computer vision SOTA baselines. The results show that our CoLAL method significantly outperforms existing SOTA in text-based AL, and is competitive with SOTA image-based AL techniques.

Research on blockchain abnormal transaction detection technology combining CNN and transformer structure

With the rapid development of blockchain and digital currencies, cybercrimes targeting blockchain transactions are proliferating. Various abnormal transaction behaviors caused by malicious nodes and malicious transactions have emerged in blockchain systems. Therefore, blockchain transaction security research is a hot issue in blockchain research. Deep learning has made remarkable achievements in image and audio. However, deep learning is less effective when processing text-like data than tree-based models. Therefore, for the highly unbalanced textual financial dataset, this paper designs a deep learning anomaly detection method called 1D SA-Inception by combining CNN with Transformer structure on the basis of traditional one-dimensional CNN (1D-CNN), introduces the Inception structure to learn the information of different scales of tabular data, and introduces the Transformer (self-attention) self-attention mechanism for the research of blockchain anomaly transaction detection technology, effectively improving the prediction effect. Finally, by conducting experiments on the extreme class imbalance dataset, the results show that the 1D SA-Inception designed in this paper outperforms the traditional CNN with AUC, G-mean, and F1 indexes of 91.05 %, 84.95 %, and 83.52 %, respectively, on the textual dataset. This method can more efficiently achieve the detection of abnormal transaction behavior, safeguard consumer rights and interests, and ensure the safe development of the financial sector.

TSIC-CLIP: Traffic Scene Image Captioning Model Based on Clip

Image captioning in traffic scenes presents several challenges, including imprecise caption generation, lack of personalization, and an unwieldy number of model parameters. We propose a new image captioning model for traffic scenes to address these issues. The model incorporates an adapter-based fine-tuned feature extraction part to enhance personalization and a caption generation module using global weighted attention pooling to reduce model parameters and improve accuracy. The proposed model consists of four main stages. In the first stage, the Image-Encoder extracts the global features of the input image and divides it into nine sub-regions, encoding each sub-region separately. In the second stage, the Text-Encoder encodes the text dataset to obtain text features. It then calculates the similarity between the image sub-region features and encoded text features, selecting the text features with the highest similarity. Subsequently, the pre-trained Faster RCNN model extracts local image features. The model then splices together the text features, global image features, and local image features to fuse the multimodal information. In the final stage, the extracted features are fed into the Captioning model, which effectively fuses the different features using a novel global weighted attention pooling layer. The Captioning model then generates natural language image captions. The proposed model is evaluated on the MS-COCO dataset, Flickr 30K dataset, and BUUISE-Image dataset, using mainstream evaluation metrics. Experiments demonstrate significant improvements across all evaluation metrics on the public datasets and strong performance on the BUUISE-Image traffic scene dataset.

A novel approach to texture recognition combining deep learning orthogonal convolution with regional input features.

Textures provide a powerful segmentation and object detection cue. Recent research has shown that deep convolutional nets like Visual Geometry Group (VGG) and ResNet perform well in non-stationary texture datasets. Non-stationary textures have local structures that change from one region of the image to the other. This is consistent with the view that deep convolutional networks are good at detecting local microstructures disguised as textures. However, stationary textures are textures that have statistical properties that are constant or slow varying over the entire region are not well detected by deep convolutional networks. This research demonstrates that simple seven-layer convolutional networks can obtain better results than deep networks using a novel convolutional technique called orthogonal convolution with pre-calculated regional features using grey level co-occurrence matrix. We obtained an average of 8.5% improvement in accuracy in texture recognition on the Outex dataset over GoogleNet, ResNet, VGG and AlexNet.

Short text classification using semantically enriched topic model

Modelling short text is challenging due to the small number of word co-occurrence and insufficient semantic information that affects downstream Natural Language Processing (NLP) tasks, for example, text classification. Gathering information from external sources is expensive and may increase noise. For efficient short text classification without depending on external knowledge sources, we propose Expressive Short text Classification (EStC). EStC consists of a novel document context-aware semantically enriched topic model called the Short text Topic Model (StTM) that captures words, topics and documents semantics in a joint learning framework. In StTM, the probability of predicting a context word involves the topic distribution of word embeddings and the document vector as the global context, which obtains by weighted averaging of word embeddings on the fly simultaneously with the topic distribution of words without requiring an additional inference method for the document embedding. EStC represents documents in an expressive (number of topics × number of word embedding features) embedding space and uses a linear support vector machine (SVM) classifier for their classification. Experimental results demonstrate that EStC outperforms many state-of-the-art language models in short text classification using several publicly available short text data sets.

Neural collapse inspired semi-supervised learning with fixed classifier

Pseudo-labeling-based approaches are gaining prominence in Semi-Supervised Learning (SSL). Recent studies have identified that the key bottleneck in this methodology is addressing insufficient, incorrect, and imbalanced pseudo-labels. In this paper, we argue that the intrinsic problem behind this bottleneck is classifier bias, i.e., the classifier's prototypes suffer from poor uniformity. Further, inspired by neural collapse that reveals an optimal structure under supervised training scenarios, we address this classifier bias by utilizing an offline simplex Equiangular Tight Frame (ETF) classifier with maximally and equally separated prototypes. During the training phase, we maintain the prototypes of the classifier as fixed and concentrate on refining the feature encoder. Specifically, we integrate a straightforward clustering-based pseudo-labeling strategy with information maximization for feature learning. In practice, the fixed ETF classifier prevents the model from falling into a detrimental cycle, where a biased classifier exacerbates misaligned features, further perpetuating this bias. Furthermore, the clustering-based pseudo-labeling strategy reduces the dependency on complex threshold-adjusting mechanisms and effectively navigates the quantity-quality trade-off that plagues existing SSL methods. Leveraging these methodologies, we develop a simple yet powerful approach, termed ETF-SSL. Extensive experiments across Image, Text, and Audio datasets demonstrate that ETF-SSL can achieve competitive or superior performance compared to existing approaches. This success highlights the benefits of using a fixed ETF classifier in SSL and points to promising directions for future research in this area. The code is available at: https://github.com/yichenwang231/ETFSSL.

Enhancing Document Clustering with Hybrid Recurrent Neural Networks and Autoencoders: A Robust Approach for Effective Semantic Organization of Large Textual Datasets

This research presents an innovative document clustering method that uses recurrent neural networks (RNNs) and autoencoders. RNNs capture sequential dependencies while autoencoders improve feature representation. The hybrid model, tested on different datasets (20-Newsgroup, Reuters, BBC Sports), outperforms traditional clustering, revealing semantic relationships and robustness to noise. Preprocessing includes denoising techniques (stemming, lemmatization, tokenization, stopword removal) to ensure a refined data set. Evaluation metrics (adjusted randomness evaluation, normalized mutual information evaluation, completeness evaluation, homogeneity evaluation, V-measure, accuracy) validate the effectiveness of the model and provide a powerful solution for organizing and understanding large text datasets.

The Language of Responsibility in the United Nations Security Council, 1946–2020

Abstract The United Nations Security Council (UNSC) is the United Nations' most powerful institutional body, charged with the “primary responsibility for the maintenance of international peace and security.” The main instrument through which the Council asserts this power is its resolutions, specifically by using resolution text to attribute responsibility. The UNSC uses responsibility language to assign tasks, identify accountability under international law, or reflect the Council's normative interpretation of political principles. Yet we lack a comprehensive empirical description of responsibility attributions in UNSC resolutions. We address this gap by providing an original dataset of the full text of all UNSC resolutions between 1946 and 2020. We use this data to show that the Council has significantly increased responsibility attributions since the end of the Cold War, but only for a very specific subset of terms, targeting predominantly states and individuals. We further demonstrate how the data can inform debates about the timing and status of the “responsibility to protect” as an international norm. The data and findings provide a helpful starting point for many future research endeavors, including the role of member states in the UNSC, quantitative and qualitative research on UNSC decision-making processes, or topic development of the UNSC agenda in general.

Proposed Machine learning model for predicting Egyptian Parliament Election Results

Political life and election have become one of the most important comments on social media sites. Governments have shown a keen interest in predicting the results of elections, whether presidential or parliamentary. The purpose of this study is to predict the results of the Egyptian Parliament elections using sentiment analysis, specifically Support Vector Machines (SVM), Naive Bayes, Decision Trees, and Random Forests in the context of machine learning. In this study, a sentiment analysis approach is employed to analyze public sentiment towards political parties and candidates leading up to Parliament elections. The sentiment analysis techniques are utilized to classify sentiment from textual data collected from Tweeter; Data were obtained in November 2020 before and during election days. The study utilizes a machine learning framework to train and test the models using a labeled dataset of sentiment-labeled political texts. The findings of this study reveal that sentiment analysis using machine learning can effectively predict the results of Parliament elections. The accuracy and performance of each technique are evaluated and compared to determine the most accurate and reliable predictor of election outcomes. This study contributes to the existing literature by applying sentiment analysis techniques to predict Parliament election results. The use of machine learning algorithms in combination with sentiment analysis, offers a novel approach to election forecasting. The findings of this study can be valuable for political analysts, election strategists, and policymakers seeking to understand public sentiment and predict election outcomes accurately.

Machine Learning Based Stroke Predictor Application

When blood flow to the brain stops or slows down, brain cells die because they don't get enough oxygen and nutrients. This condition is known as an ischemic stroke. It is now the biggest cause of death in the whole planet. Examining the afflicted people has shown a number of risk variables that are thought to be connected to the stroke's origin. Numerous studies have been conducted to predict the illnesses associated with stroke using these risk variables. The prompt identification of various warning symptoms associated with stroke has the potential to mitigate the severity of the stroke. The utilization of machine learning techniques yields prompt and precise predictive outcomes. Although its uses in healthcare are expanding, certain research domains have a stronger need for more study. We think that machine learning algorithms may aid in a deeper comprehension of illnesses and make an excellent healthcare partner. The textual dataset of numerous patients, which includes many medical variables, is gathered for this study. The missing values in the dataset are located and dealt with during processing. The dataset is used to train machine learning algorithms including Random Forest, Decision Tree classifier, and SVM. The method that delivers the greatest accuracy for our dataset is then selected once the accuracy of the algorithms has been determined. This aids patients in determining the likelihood of a brain stroke and ensuring they get the right medical attention.

The performance of large language models in intercollegiate Membership of the Royal College of Surgeons examination.

Large language models (LLM), such as Chat Generative Pre-trained Transformer (ChatGPT) and Bard utilise deep learning algorithms that have been trained on a massive data set of text and code to generate human-like responses. Several studies have demonstrated satisfactory performance on postgraduate examinations, including the United States Medical Licensing Examination. We aimed to evaluate artificial intelligence performance in Part A of the intercollegiate Membership of the Royal College of Surgeons (MRCS) examination. The MRCS mock examination from Pastest, a commonly used question bank for examinees, was used to assess the performance of three LLMs: GPT-3.5, GPT 4.0 and Bard. Three hundred mock questions were input into the three LLMs, and the responses provided by the LLMs were recorded and analysed. The pass mark was set at 70%. The overall accuracies for GPT-3.5, GPT 4.0 and Bard were 67.33%, 71.67% and 65.67%, respectively (p = 0.27). The performances of GPT-3.5, GPT 4.0 and Bard in Applied Basic Sciences were 68.89%, 72.78% and 63.33% (p = 0.15), respectively. Furthermore, the three LLMs obtained correct answers in 65.00%, 70.00% and 69.17% of the Principles of Surgery in General questions (p = 0.67). There were no differences in performance in the overall and subcategories among the three LLMs. Our findings demonstrated satisfactory performance for all three LLMs in the MRCS Part A examination, with GPT 4.0 the only LLM that achieved the pass mark set.

Automatic analysis of public health service text based on character level convolutional neural network

The text information processing technology of public health service is one of the hot research topics at present. To improve the defects of public health service texts, such as inaccurate word segmentation, spelling errors and professional vocabulary understanding, this study designed a character-level deep neural network model on the characteristics of public health service texts. In this model, the bidirectional short and short time memory and the attention pooling operation layer are introduced to make the model better classify the text according to the context. In addition, counter perturbation is introduced in this study to improve the robustness and generalization ability of the model, thus improving its classification effect. The performance verification results show that the proposed model has better classification performance on the public health service text data set. The anti-disturbance samples generated by the model are all in the range of 0–0.2 when WMD deviation degree is measured, while most of the other methods are in the range of 0.4–0.6. The experimental object of this study is ultrasonic examination data. The experimental results show that the automatic analysis model of public health service text based on character level convolutional neural network constructed in this study has excellent accuracy and convergence speed, and has excellent performance in the classification of public health service text in different subject areas.

Human-like cognition: visual features grouping for hard-to-group text dataset

Human-like cognition: visual features grouping for hard-to-group text dataset

Text Attribute Control via Closed-Loop Disentanglement

Abstract Changing an attribute of a text without changing the content usually requires first disentangling the text into irrelevant attributes and content representations. After that, in the inference phase, the representation of one attribute is tuned to a different value, expecting that the corresponding attribute of the text can also be changed accordingly. The usual way of disentanglement is to add some constraints on the latent space of an encoder-decoder architecture, including adversarial-based constraints and mutual-information-based constraints. However, previous semi-supervised processes of attribute change are usually not enough to guarantee the success of attribute change and content preservation. In this paper, we propose a novel approach to achieve a robust control of attributes while enhancing content preservation. In this approach, we use a semi-supervised contrastive learning method to encourage the disentanglement of attributes in latent spaces. Differently from previous works, we re-disentangle the reconstructed sentence and compare the re-disentangled latent space with the original latent space, which makes a closed-loop disentanglement process. This also helps content preservation. In addition, the contrastive learning method is also able to replace the role of minimizing mutual information and adversarial training in the disentanglement process, which alleviates the computation cost. We conducted experiments on three text datasets, including the Yelp Service review dataset, the Amazon Product review dataset, and the GoEmotions dataset. The experimental results show the effectiveness of our model.

Improving Medical Speech-to-Text Accuracy using Vision-Language Pre-training Models.

Automatic Speech Recognition (ASR) is a technology that converts spoken words into text, facilitating interaction between humans and machines. One of the most common applications of ASR is Speech-To-Text (STT) technology, which simplifies user workflows by transcribing spoken words into text. In the medical field, STT has the potential to significantly reduce the workload of clinicians who rely on typists to transcribe their voice recordings. However, developing an STT model for the medical domain is challenging due to the lack of sufficient speech and text datasets. To address this issue, we propose a medical-domain text correction method that modifies the output text of a general STT system using the Vision Language Pre-training (VLP) method. VLP combines textual and visual information to correct text based on image knowledge. Our extensive experiments demonstrate that the proposed method offers quantitatively and clinically significant improvements in STT performance in the medical field. We further show that multi-modal understanding of image and text information outperforms single-modal understanding using only text information.

Taking the Road Less Travelled: How Corpus‐Assisted Discourse Studies Can Enrich Qualitative Explorations of Large Textual Datasets

AbstractHow might interpretivist qualitative researchers tackle large data sets consisting of millions or even billions of words? Corpus‐assisted discourse studies (CADS) is the approach we explore here. Specifically designed for the analysis of voluminous textual data, it offers a recognized empirical approach for making sense of such data. But it does so within an epistemology that understands language to be central in shaping our understanding of the world around us, so that CADS can assist researchers in revealing the social dynamics of the text – including the ideology and power that is latent in many such corpora. Bringing together the training of applied linguists and a management scholar, we discuss the background to CADS and its differences from text‐mining approaches such as topic modelling, which have been more widely used in management studies to date. Focusing on the needs of people who are new to the approach, we then offer a worked example to show CADS’ potential in exploring a management‐related corpus. Our paper concludes with a discussion of the strengths and weaknesses of the approach and its potential for future discursively orientated management research – especially in the context of the rise of ‘big data’.

No time to lie: Examining the identity of pro-vaccination and anti-vaccination supporters through user-generated content

ObjectiveThis study delves into the social identity of pro-vaccination and anti-vaccination supporters, emphasizing an understanding of the values that shape these distinct identities. Furthermore, the research highlights that user-generated content pertaining to vaccines offers valuable insights into the underlying personal values of both pro-vaccination and anti-vaccination groups. MethodWe constructed a textual dataset based on 142,596 tweets. This data was analyzed in three steps. First, the linguistic characteristics of the textual data, together with the underlying personal values of the text creators, were identified using LIWC software. Second, the identified personal values were used as an input for the moderation analysis, which examined the relationship between personal values and social identity for pro- and anti-vaccination groups. Finally, an automated, in-depth text analysis was conducted in Mathematica to understand the narratives created by both groups. ResultsThe study findings indicate that both pro-vaccination and anti-vaccination supporters display characteristics of subcultures with distinct group identities. Consequently, based on the results, there is a need for more tailored public health communication strategies that address these two groups separately. ConclusionsUnderstanding how users create health-related content based on their personal values is crucial. Acknowledging and appreciating the diverse personal values and identities within different groups in the vaccination discourse can inform health communication efforts, aligning these efforts with the specific values of each group. This targeted communication is vital for effectively conveying relevant peer-reviewed health information amid the abundance of health-related user-generated content.

ALGORITHM FOR DETECTION OF ABUSIVE CONTENT IN AUDIO CONTENT FOR IMPLEMENTATION IN OBJECT-ORIENTED INFORMATION SYSTEM

The paper proposes the basic principles of developing an object-oriented information system for detecting abusive content in Ukrainian-language audio content based on a new algorithm that uses statistical and neural network approaches to detect abusive content. Detecting offensive content in text and audio content is an urgent task, as it helps to create a safe and healthy environment for communication, especially in online platforms. Offensive content can harm the people who hear or read it and violate their rights. It can also have a negative impact on society, contributing to the spread of hatred and violence. To detect abusive speech in audio content, the proposed approach uses two key components: the use of dictionary methods and the analysis of the emotional tonality of utterances. A set of reviews was used as a dataset to determine the abusive component of the content, which was expanded by the authors by adding words of abuse. An object-oriented information system architecture written in the Python programming language in the PyCharm programming environment is proposed. The information system consists of a software module for training recurrent neural network models and further saving trained instances, and a software module for detecting abusive content in Ukrainian-language audio content using trained RNN models. Since the recurrent neural network is trained on a short text data set, the system is less efficient at identifying texts that have a larger number of words. In the example of the proposed approach, the accuracy of detecting offensive content is more than 90%. This means that the algorithm works correctly in the absence of offending highlights in the test data set. The results of the analysis of the effectiveness of the proposed approach show that in the vast majority of cases the conclusions regarding the acceptability of audio content based on the level of abuse are correct.

Exploring the landscape of AI-assisted decision-making in head and neck cancer treatment: a comparative analysis of NCCN guidelines and ChatGPT responses.

Recent breakthroughs in natural language processing and machine learning, exemplified by ChatGPT, have spurred a paradigm shift in healthcare. Released by OpenAI in November 2022, ChatGPT rapidly gained global attention. Trained on massive text datasets, this large language model holds immense potential to revolutionize healthcare. However, existing literature often overlooks the need for rigorous validation and real-world applicability. This head-to-head comparative study assesses ChatGPT's capabilities in providing therapeutic recommendations for head and neck cancers. Simulating every NCCN Guidelines scenarios. ChatGPT is queried on primary treatments, adjuvant treatment, and follow-up, with responses compared to the NCCN Guidelines. Performance metrics, including sensitivity, specificity, and F1 score, are employed for assessment. The study includes 68 hypothetical cases and 204 clinical scenarios. ChatGPT exhibits promising capabilities in addressing NCCN-related queries, achieving high sensitivity and overall accuracy across primary treatment, adjuvant treatment, and follow-up. The study's metrics showcase robustness in providing relevant suggestions. However, a few inaccuracies are noted, especially in primary treatment scenarios. Our study highlights the proficiency of ChatGPT in providing treatment suggestions. The model's alignment with the NCCN Guidelines sets the stage for a nuanced exploration of AI's evolving role in oncological decision support. However, challenges related to the interpretability of AI in clinical decision-making and the importance of clinicians understanding the underlying principles of AI models remain unexplored. As AI continues to advance, collaborative efforts between models and medical experts are deemed essential for unlocking new frontiers in personalized cancer care.

Cross‐modal knowledge learning with scene text for fine‐grained image classification

AbstractScene text in natural images carries additional semantic information to aid in image classification. Existing methods lack full consideration of the deep understanding of the text and the visual text relationship, which results in the difficult to judge the semantic accuracy and the relevance of the visual text. This paper proposes image classification based on Cross modal Knowledge Learning of Scene Text (CKLST) method. CKLST consists of three stages: cross‐modal scene text recognition, text semantic enhancement, and visual‐text feature alignment. In the first stage, multi‐attention is used to extract features layer by layer, and a self‐mask‐based iterative correction strategy is utilized to improve the scene text recognition accuracy. In the second stage, knowledge features are extracted using external knowledge and are fused with text features to enhance text semantic information. In the third stage, CKLST realizes visual‐text feature alignment across attention mechanisms with a similarity matrix, thus the correlation between images and text can be captured to improve the accuracy of the image classification tasks. On Con‐Text dataset, Crowd Activity dataset, Drink Bottle dataset, and Synth Text dataset, CKLST can perform significantly better than other baselines on fine‐grained image classification, with improvements of 3.54%, 5.37%, 3.28%, and 2.81% over the best baseline in mAP, respectively.