Text Processing Research Articles

Automatic Image Captioning using Deep Learning

Image captioning aims to detect this information by describing the image content through image and text processing techniques. Now-a-days image captioning is one of the recent and growing research problem. Day by day various solutions are being introduced for solving the problem. Even though, many solutions are already available, a lot of attention is still required for getting better and precise results. So, we came up with the idea of developing a image captioning model using different combinations of Convolutional Neural Network architecture along with Long Short Term Memory to get better results. We have used three combination of CNN and LSTM for developing the model. The proposed model is trained with three Convolutional Neural Network architecture such as Inception-v3,Xception, ResNet50 for feature extraction and Long ShortTerm Memory for generating the relevant captions. Among these models, the best combination based on the accuracy of the model. It is trained using the Flicker8k dataset

Railway accident causation prediction with improved transformer model based on lexical information and contextual relationships

The railway system is a prime example of a safety-critical system. Predicting the causes of railway accidents holds immense significance in enhancing railway transportation safety. Previous approaches to railway causation analysis have encountered huge challenges regarding data processing and analytical capabilities. To address this concern, this paper proposes an innovative deep model framework based on the Transformer architecture that utilizes historical data on railway equipment accidents to predict the causes behind such incidents. Firstly, this paper proposes the utilization of Convolutional Block Attention in the domain of text processing, serving as a lexical encoder to augment word semantics acquisition in accident texts. Subsequently, in order to address the deficiency of traditional Transformers that lack positional representation information, we propose incorporating a BiGRU (Bidirectional Gated Recurrent Unit) as a contextual positional information encoder to capture contextual positional information in railway accident data effectively. Finally, considering that accident data reports are discrete tabular data, this study suggests employing cue word techniques for preprocessing accident data to alleviate the model's learning burden. We applied the proposed model to the FRA (Federal Railroad Administration) dataset. The results demonstrate that our model surpasses the current state-of-the-art language models, exhibiting superior performance compared to the optimal model with a notable improvement of 3.56%, 0.42%, and 0.76% in Precision, Recall, and F1-score, respectively. Furthermore, our model accurately predicts accident categories prone to misjudgment even when trained on limited data, outperforming existing language models. The study findings will contribute to the prevention and management of railway accidents.

CrisisTransformers: Pre-trained language models and sentence encoders for crisis-related social media texts

Social media platforms play an essential role in crisis communication, but analysing crisis-related social media texts is challenging due to their informal nature. Transformer-based pre-trained models like BERT and RoBERTa have shown success in various NLP tasks, but they are not tailored for crisis-related texts. Furthermore, general-purpose sentence encoders are used to generate sentence embeddings, regardless of the textual complexities in crisis-related texts. Advances in applications like text classification, semantic search, and clustering contribute to the effective processing of crisis-related texts, which is essential for emergency responders to gain a comprehensive view of a crisis event, whether historical or real-time. To address these gaps in crisis informatics literature, this study introduces CrisisTransformers, an ensemble of pre-trained language models and sentence encoders trained on an extensive corpus of over 15 billion word tokens from tweets associated with more than 30 crisis events, including disease outbreaks, natural disasters, conflicts, and other critical incidents. We evaluate existing models and CrisisTransformers on 18 crisis-specific public datasets. Our pre-trained models outperform strong baselines across all datasets in classification tasks, and our best-performing sentence encoder improves the state-of-the-art by 17.43% in sentence encoding tasks. Additionally, we investigate the impact of model initialization on convergence and evaluate the significance of domain-specific models in generating semantically meaningful sentence embeddings. The models are publicly available at: https://huggingface.co/crisistransformers.

Computer Feed based Text Processing Using NLP

The goal of this research project is to create a comprehensive multimedia and text processing framework that can be used in a variety of applications. The objective is to combine a collection of libraries and instruments for the purpose of extracting, analyzing, and modifying textual information obtained from multimedia files, including images, videos, and audio files. The strategy involves applying cutting-edge techniques such as computer vision, audio processing, natural language processing (NLP), optical character recognition (OCR), and computer vision. Important findings include accurate text extraction from multimedia inputs, accurate textual data analysis using natural language processing techniques, and the production of insightful information via sentiment analysis, summarization, and translation. Its main contributions are related to the development of a flexible system that can process a wide range of multimedia inputs and provide useful text processing features. This project provides a foundation for future research and useful applications in the fields of text processing and multimedia.

Exploration of an Open Vocabulary Model on Semantic Segmentation for Street Scene Imagery

This study investigates the efficacy of an open vocabulary, multi-modal, foundation model for the semantic segmentation of images from complex urban street scenes. Unlike traditional models reliant on predefined category sets, Grounded SAM uses arbitrary textual inputs for category definition, offering enhanced flexibility and adaptability. The model’s performance was evaluated across single and multiple category tasks using the benchmark datasets Cityscapes, BDD100K, GTA5, and KITTI. The study focused on the impact of textual input refinement and the challenges of classifying visually similar categories. Results indicate strong performance in single-category segmentation but highlighted difficulties in multi-category scenarios, particularly with categories bearing close textual or visual resemblances. Adjustments in textual prompts significantly improved detection accuracy, though challenges persisted in distinguishing between visually similar objects such as buses and trains. Comparative analysis with state-of-the-art models revealed Grounded SAM’s competitive performance, particularly notable given its direct inference capability without extensive dataset-specific training. This feature is advantageous for resource-limited applications. The study concludes that while open vocabulary models such as Grounded SAM mark a significant advancement in semantic segmentation, further improvements in integrating image and text processing are essential for better performance in complex scenarios.

Exploring Stemming Techniques in Ambon Malay Languages: A Systematic Literature Review

Stemming in Ambonese posed a significant challenge due to its extensive lexicon, encompassing approximately 127,000 base words as recorded in the Kamus Besar Bahasa Indonesia (Indonesian Dictionary). This complexity arises from the task of extracting base words from those with affixes, necessitating the removal of various affixes such as prefixes, infixes, suffixes, and their combinations. This process greatly influences analytical outcomes. To address this linguistic complexity, several stemming algorithms were developed. These include Nazief Adriani, Enhanced Confix Stripping, Sastrawi, and Tala, each offering unique techniques to handle stemming complexities in Indonesian. The selection of the appropriate algorithm is crucial for ensuring the accuracy and reliability of the stemming process within the analytical framework. In conducted stemming research, there were variations in methods used. The most frequently used algorithm was Nazief Adriani, with 17 recorded cases, followed by Enhanced Confix Stripping with 12 cases. Sastrawi, although less frequent, was used in 4 cases, while Tala appeared in 1 case. This diversity reflects the available choices in selecting a fitting stemming method. However, this may relate to factors such as ongoing research projects, funding availability, or other external conditions affecting research production during that period. Consequently, stemming research remains an interesting and relevant topic, with the potential for continued growth and significant contributions to text processing and linguistic research in the future.

Text Classification on Customer Feedback: A Systematic Literatures Review

User feedback in text classification serves multiple purposes, including refining models, enhancing datasets, adapting to user preferences, identifying emerging topics, and evaluating system performance, all of which contribute to the creation of more effective and user-centric classification systems. Many text classification techniques, including data mining, machine learning, and deep learning approaches, have been employed in previous literature, each making significant contributions to the field. This paper aims to contribute by guiding researchers seeking commonly used classification techniques and evaluation metrics in text processing. Additionally, it identifies the classification technique that generates higher accuracy and works as a basis for researchers to synthesize studies within their respective fields. PRISMA methodology is adapted to systematically review 28 current literatures on text classification on user feedback. The results obtained are guided by four research questions; paper distribution year, dataset source and size; evaluation metric and model accuracy. The review has shown that support vector machines (SVM) are frequently employed and consistently achieve high levels of accuracy as high as 97.17% with various datasets used. The future direction of this work could explore models that integrate sentiment analysis and natural language understanding to more accurately capture nuanced user opinions and preferences.

Self-Esteem, Self-Satisfaction with Voice, and Self-Assessment with Descriptive Terms of One's Own Voice for Voice Among Brazilian LGBTQIAPN+ People

ObjectivesTo analyze self-esteem, self-satisfaction with voice, and self-assessment with descriptive terms of one's own voice among Brazilian LGBTQIAPN+ people. Study DesignCross-sectional, observational, qualitative-quantitative study. MethodsThe study included any person who identified as Brazilian LGBTQIAPN+ and excluded those under 18 years old. Participants answered a sociodemographic questionnaire, the Rosenberg Self-Esteem Scale, and the Descriptive terms of one's own voice (which were analyzed with text processing techniques). The 10 terms most listed as positive or negative were selected for simple logistic regression associated with the Rosenberg Self-Esteem Scale. The alpha level of significance used in all analyses was 5%. ResultsThe study considered the responses from 411 participants. It is noteworthy that 92.45% (n=380) of the studied population identified as cisgender. Among the participants, 28 were classified as having unsatisfactory self-esteem, 208 as having average self-esteem, and 178 as having satisfactory self-esteem. Individuals with unsatisfactory scores described themselves using terms such as "high," "boring," "ugly," "unfriendly," and "masculine," while those with average scores used terms like "insecure," "nasal," "out of tune," and "high." Those with satisfactory scores associated themselves with terms such as "expressive," "sympathetic," "kind," "pleasant," and "stable." The terms "cheerful," "adequate," and "boring" significantly associated with the Rosenberg Self-Esteem Scale.score. ConclusionLGBTQIAPN+ people with unsatisfactory self-esteem referred to their voices with negative descriptive terms, similar to people with average self-esteem, who often did not select any positive term. On the other hand, people with satisfactory self-esteem assessed their voices with positive descriptive terms and few negative adjectives. People with satisfactory scores are more likely to think that their voice is “cheerful” and less likely to consider it “adequate” or “boring” than people with unsatisfactory scores on the Rosenberg Self-Esteem Scale. The majority of participants were cisgender homosexual males, thus caution should be exercised in generalizing the data to the entire community.

Challenges Related to Grammatical and Morphological Processing of Arabic Texts by Means of Artificial Intelligence

This research aims to study the challenges facing artificial intelligence in processing Arabic texts grammatically and morphologically, by conducting a survey on a random sample of Microsoft Word 2010 users. The research provided a theoretical framework for the subject of the study, as it addressed the concept of artificial intelligence, as well as the general challenges facing artificial intelligence and some challenges specific to processing Arabic texts. The research relied on the descriptive approach, and this helped highlight the importance of the study and provide a comprehensive vision of its subject. The study concluded a set of results, the most important of which are: - There are many challenges facing users in processing Arabic texts grammatically and morphologically, such as derivation, morphology, and the breadth and comprehensiveness of the Arabic language. - The inability of artificial intelligence so far to reach a high degree of quality and efficiency in processing Arabic texts grammatically and morphologically. - The possibility of processing the Arabic language in the future by adding some software operations that help develop the machine’s capabilities to understand the Arabic language and the ability to express it effectively, through the use of new artificial intelligence methods, such as deep learning and machine learning. The study recommended a number of important recommendations, which are: - Microsoft Word is constantly updated by Microsoft, in order to ensure the quality and efficiency of the program in processing Arabic texts grammatically and morphologically. - Encouraging students and researchers to conduct more scientific research in the field of automated processing of linguistic texts, in order to enhance the progress and development of this vital field.

Methods of Automatic Analysis of Information Presentation Dynamics in Texts Based on Adaptable Dictionaries of Scientific Terms

The paper proposes a method for determining the dynamics of aspectual content of scientific texts. The problems arising in the automatic processing of scientific texts are discussed, approaches to the creation of a combined method of aspect-oriented analysis of scientific texts are proposed. The results of experiments of aspect analysis of scientific publications in the field of mathematics are given. The algorithm of aspect analysis is built on the basis of processing both the semantic domain graph formed as a result of automatic extraction of information from the text of the mathematical encyclopedia and the weighted semantic graph of the publication. Weighting functions for calculating the dependence of links between professional terms of the semantic domain graph are proposed. During the processing the text of a scientific publication, the semantic domain graph is transformed into the semantic graph of the publication, the two-phase processing algorithm of which forms semantic dependencies between text fragments.

Human Language Data Processing using NLTK

Natural Language Toolkit (NLTK) is a comprehensive Python library designed to facilitate the exploration, analysis, and processing of human language data. With its extensive collection of tools, NLTK provides researchers, developers, and educators with a powerful platform for tasks ranging from basic text processing to advanced natural language understanding and machine learning. The toolkit includes modules for tokenization, stemming, lemmatization, part-of-speech tagging, named entity recognition, syntactic parsing, semantic analysis, and more. Furthermore, NLTK offers access to numerous linguistic resources such as corpora, lexicons, and treebanks, making it an invaluable resource for both learning and research in the field of natural language processing (NLP). NLTK serves as an indispensable tool for unlocking the complexities of human language

Non-symmetric over-time pooling using pseudo-grouping functions for convolutional neural networks

Convolutional Neural Networks (CNNs) are a family of networks that have become state-of-the-art in several fields of artificial intelligence due to their ability to extract spatial features. In the context of natural language processing, they can be used to build text classification models based on textual features between words. These networks fuse local features to generate global features in their over-time pooling layers. These layers have been traditionally built using the maximum function or other symmetric functions such as the arithmetic mean. It is important to note that the order of input local features is significant (i.e. the symmetry is not an inherent characteristic of the model). While this characteristic is appropriate for image-oriented CNNs, where symmetry might make the network robust to image rigid transformations, it seems counter-productive for text processing, where the order of the words is certainly important. Our proposal is, hence, to use non-symmetric pooling operators to replace the maximum or average functions. Specifically, we propose to perform over-time pooling using pseudo-grouping functions, a family of non-symmetric aggregation operators that generalize the maximum function. We present a construction method for pseudo-grouping functions and apply different examples of this family to over-time pooling layers in text-oriented CNNs. Our proposal is tested on seven different models and six different datasets in the context of engineering applications, e.g. text classification. The results show an overall improvement of the models when using non-symmetric pseudo-grouping functions over the traditional pooling function.

Development and validation of a pulmonary function test data extraction tool for the US department of veterans affairs electronic health record

ObjectivePulmonary function test (PFT) results are recorded variably across hospitals in the Department of Veterans Affairs (VA) electronic health record (EHR), using both unstructured and semi-structured notes. We developed and validated a hospital-specific code to extract pre-bronchodilator measures of obstruction (ratio of forced expiratory volume in one second [FEV1] to forced vital capacity [FVC]) and severity of obstruction (percent predicted of FEV1).ResultsAmong 36 VA facilities with the most PFTs completed between 2018 and 2022 from a parent cohort of veterans receiving long-acting controller inhalers, 12 had a consistent syntactical convention or template for reporting PFT data in the EHR. Of the 42,718 PFTs identified from these 12 facilities, the hospital-specific text processing pipeline yielded 24,860 values for the FEV1:FVC ratio and 23,729 values for FEV1. A ratio of FEV1:FVC less than 0.7 was identified in 17,615 of 24,922 studies (70.7%); 8864 of 24,922 (35.6%) had a severe or very severe reduction in FEV1 (< 50% of the predicted value). Among 100 randomly selected PFT reports reviewed by two pulmonary physicians, the coding solution correctly identified the presence of obstruction in 99 out of 100 studies and the degree of obstruction in 96 out of 100 studies.

Leveraging Bidirectionl LSTM with CRFs for Pashto Tagging

Part-of-speech tagging plays a vital role in text processing and natural language understanding. Very few attempts have been made in the past for tagging Pashto Part-of-Speech. In this work, we present a Long Short-term Memory–based approach for Pashto part-of-speech tagging with special focus on ambiguity resolution. Initially, we created a corpus of Pashto sentences having words with multiple meanings and their tags. We introduce a powerful sentences representation and new architecture for Pashto text processing. The accuracy of the proposed approach is compared with state-of-the-art Hidden Markov Model. Our Model shows 87.60% accuracy for all words excluding punctuation and 95.45% for ambiguous words; however, Hidden Markov Model shows 78.37% and 44.72% accuracy, respectively. Results show that our approach outperforms Hidden Markov Model in Part-of-Speech tagging for Pashto text.

Large Language Models for Rapid Simplification of Quality Assurance Data Input: Field Trial with Real Data in the Context of Tumour Documentation in Urology

Large Language Models (LLMs) such as ChatGPT have rapidly brought the application of artificial intelligence into widespread use. Among many different use cases for text generation and processing, one application is the extraction of data from existing documents and conversations for simplified and automated form-filling. In the field of quality assurance and documentation of cancer diseases, there is currently a significant workload involved in transferring data under various aspects into slightly varying formats and applying interpretations such as the TNM classification of tumours. However, there is a lack of trials with real data to assess the applicability of LLM-supported processes in this area, which would enable an evaluation of efficiency and practicality. This study aims to implement and assess such a trial. A trial was conducted with N=153 privacy-compliant and ethics committee-cleared medical reports from 25 patients. Using the publicly available version of ChatGPT 4.0, an automated script was used to extract the date of initial diagnosis and common tumor classifications. The results were then individually checked for accuracy. Based on this, the utility of a simple system for guided support in tasks related to tumour documentation was assessed. Additionally, the approach was evaluated in terms of operational costs for the model and its applicability. In summary, the study concludes that the use of generative AI in this field is promising and suitable as a tool even in an untrained state. In a simplified calculation, costs of 35 cents are offset by a value creation of 61,54 euros. However, it also becomes clear that AI can only act in a supportive role, and the correct integration with pre-made specific prompts and tools into the workflow is crucial for a relevant performance. The use of generative AI in the context of search, transfer, and interpretation tasks in the creation of tumor documentation is a promising approach. However, its implementation in practical applications must be closely monitored, and the optimal interaction between man and machine should continue to be evaluated and must be accompanied by tools and task-specific prompts.

E-Commerce Spam Review Detection using Machine Learning

With the proliferation of e-commerce platforms, the authenticity of online reviews has become increasingly crucial. In this research, we present a method for detecting spam reviews in e-commerce platforms, employing the Random Forest algorithm. Leveraging a dataset comprising reviews from Amazon Yelp dataset, we employ a combination of Natural Language Processing (NLP) techniques and text processing methods to uncover underlying patterns distinguishing between genuine and fake reviews. Real-time data scraping from the Amazon website facilitated the acquisition of a diverse range of reviews, subsequently stored in a CSV file for analysis. These reviews stored in CSV file is fed to model for prediction. Our model effectively discerns between authentic and spam reviews, offering a valuable tool for maintaining the integrity of e-commerce platforms and ensuring informed consumer decision-making. Key Words: Web Scrapping, Text Mining, Sentiment Analysis.

Extending Context Window in Large Language Models with Segmented Base Adjustment for Rotary Position Embeddings

In the realm of large language models (LLMs), extending the context window for long text processing is crucial for enhancing performance. This paper introduces SBA-RoPE (Segmented Base Adjustment for Rotary Position Embeddings), a novel approach designed to efficiently extend the context window by segmentally adjusting the base of rotary position embeddings (RoPE). Unlike existing methods, such as Position Interpolation (PI), NTK, and YaRN, SBA-RoPE modifies the base of RoPE across different dimensions, optimizing the encoding of positional information for extended sequences. Through experiments on the Pythia model, we demonstrate the effectiveness of SBA-RoPE in extending context windows, particularly for texts exceeding the original training lengths. We fine-tuned the Pythia-2.8B model on the PG-19 dataset and conducted passkey retrieval and perplexity (PPL) experiments on the Proof-pile dataset to evaluate model performance. Results show that SBA-RoPE maintains or improves model performance when extending the context window, especially on longer text sequences. Compared to other methods, SBA-RoPE exhibits superior or comparable performance across various lengths and tasks, highlighting its potential as an effective technique for context window extension in LLMs.

The effects of question previewing on response accuracy and text processing: An eye-movement study

The present study investigated the impact of manipulating reading strategies (i.e., reading the questions first [QF] or reading the passage first [PF]) during a reading comprehension test where we explored how reading strategy was related to student characteristics (i.e., reading achievement and working memory capacity). Participants' eye movements were monitored as they read 12 passages and answered multiple-choice questions. We examined differences in (a) response accuracy, (b) average total time on words in the text, (c) total task reading time, and (d) time reading text relevant to questions as a function of PF and QF strategies. Analyses were conducted to examine whether findings varied as a function of student characteristics (i.e., reading achievement and working memory capacity) and grade level (Grades 3, 5, and 8). Several interesting findings emerged from our study, including a limited effect of reading strategy use on response accuracy, with only eighth graders demonstrating better accuracy in the QF condition, and several demonstrations of PF leading to more efficient test-taking processes, including (a) longer average total reading times on words in the passage in the PF condition that could be associated with creating a better mental model of the text, (b) often being associated with less total-task time, and (c) being associated with more successful search strategies. Implications for providing teachers and students with strategies are discussed.

ICON: Instagram Profile Classification Using Image and Natural Language Processing Methods

The use of social media has grown significantly, and businesses are now using these platforms to promote their products and services. To do this, companies have created business accounts on social networks. However, social media platforms can also be a breeding ground for unwanted behaviors such as cyberbullying, sexual content, and promotional comments. To address this issue, a study was conducted to create a system that could classify public accounts on Instagram by analyzing comments, profile pictures, bios, and posts shared by users with business accounts. First, a crawler was developed, and data were collected using this crawler and then anonymized. Next, the collected data were processed using natural language processing (NLP) techniques for text and image processing methods for images to extract features and create a dataset. Nearly 10 000 profiles and 30 000 comments from public accounts were manually tagged to create the classification model. The final model had an accuracy rate of 95% on the dataset, allowing for the effective identification of different types of business accounts on Instagram.

Exploratory analysis on the natural language processing models for task specific purposes

Natural language processing (NLP) is a technology that has become widespread in the area of human language understanding and analysis. A range of text processing tasks such as summarisation, semantic analysis, classification, question-answering, and natural language inference are commonly performed using it. The dilemma of picking a model to help us in our task is still there. It’s becoming an impediment. This is where we are trying to determine which modern NLP models are better suited for the tasks set out above in order to compare them with datasets like SQuAD and GLUE. For comparison, BERT, RoBERTa, distilBERT, BART, ALBERT, and text-to-text transfer transformer (T5) models have been used in this study. The aim is to understand the underlying architecture, its effects on the use case and also to understand where it falls short. Thus, we were able to observe that RoBERTa was more effective against the models ALBERT, distilBERT, and BERT in terms of tasks related to semantic analysis, natural language inference, and question-answering. The reason is due to the dynamic masking present in RoBERTa. For summarisation, even though BART and T5 models have very similar architecture the BART model has performed slightly better than the T5 model.