Text Quality Research Articles

Learning to write syntheses: the effect of process feedback and of observing models on performance and process behaviors

Writing a synthesis text involves interacting reading and writing processes, serving the comprehension of source information, and its integration into a reader-friendly and accurate synthesis text. Mastering these processes requires insight into process’ orchestrations. A way of achieving this is via process feedback in which students compare their process orchestration with examples. Access to such examples of enacted process orchestration models might have an additional learning effect. In the present study we replicated and extended the study of Vandermeulen et al. (Written Communication, 40(1), 90–144, 2023) on the effect of keystroke logging data-based process feedback with feed-forward exemplars when compared to national baseline performances. In addition, we report the effect of a brief extension in which learners had the opportunity to observe an enacted model of their choice, showing one of three orchestrations of the initial stage of writing a synthesis task. A total of 173 10th—grade students were randomly assigned to a process feedback condition with or without added models. A baseline, consisting of a nationally representative sample of upper-secondary students’ texts and processes, served as an alternative control group. Results showed that the process feedback, both with and without observation, had a significant effect on text quality. Regarding the process data, students in the feedback condition had a more prominent focus on the sources as they spent more time in them and switched more often between text and sources, compared to the baseline. The observation task magnified this effect.

Assessment of the quality of the text of safety standards with industrial semantic technologies

Most safety-critical systems are subject to rigorous assurance processes to justify that the systems are dependable. These processes are typically conducted in compliance with safety standards, e.g., DO-178C for software in aerospace. This can be a prerequisite so that a system is allowed to operate. However, following these standards can be challenging in practice because of issues in their text such as imprecision, ambiguity, and inconsistency. These issues can hinder compliance, delaying it and making it more expensive, or even preventing it. As a solution, we aim to define means that aid in the identification of the issues and thus facilitate their resolution. We have developed an approach for assessment of the quality of the text of safety standards with RQA - Quality Studio, an industrial tool for requirements quality analysis with semantic technologies. The approach is based on the extraction of analysis units from a standard, on the specification and exploitation of ontologies, and on the reuse of metrics provided by RQA - Quality Studio to evaluate text quality. The approach has been applied on the DO-178C standard, assessing its text as a whole and its different main individual parts. The quality of most of the text of the standard can be regarded as high. The most frequent issues in DO-178C are the use of passive voice, of synonyms, and of imprecise modal verbs. To the best of our knowledge, this is the first study that provides a means for a broad and detailed assessment of the quality of the text of safety standards, leading to the identification of specific aspects that could be improved in the text and indicating the extent to which quality issues affect it.

How do health authorities inform about general practitioner selection? Analysis of Italian Websites

Abstract Background The Next Generation EU and Italian Recovery and Resilience Plan share the common goal of supporting digital transformation. In health, adequate digitalization and web communication strategies are essential to achieve adequate population health outcomes. This work describes the level of digitalization and information of a fundamental primary care service: the choice or change of the General Practitioner (GP). Methods The analysis was conducted by consulting websites of Italian Local Health Authorities (LHAs). First, we explored digitalization levels of 105 websites through the Primary Care Digital Information (PCDI) composite index. It comprises four dimensions: informativeness, accessibility, inclusiveness, and adaptability, scoring from 0-5 (low-high digitalization). Second, we conducted a readability analysis of websites. Text quality was measured using the Gulpease and READ-IT indexes and New Basic Italian Vocabulary (NBIV) was used for lexical complexity. Results We found a national average of 2,48 for the PCDI (Min 0 - Max 4,38). Nationally, the best performing dimension was adaptability, while the worst was inclusiveness. Half of the LHAs provided several digital alternatives to choose or change the GP, while the remaining provided limited or no options. Most LHAs (70%) displayed directly on their home page a reference on how citizens can choose or change GP. However, 70% lacked specific information for non-residents citizens. Regarding readability, the national average was 48,78 for READ-IT (Min 32 - Max 62), on a scale from 0-100 (good-bad readability). Conclusions Overall, our findings depict variable and fragmented digitalization and information levels of a key health service. Despite best practices, several areas require monitoring and intervention. Moreover, some barriers characterize Italian health communication strategies, notably the variability of information across and within regions and on average low website readability. Key messages • There are significant digitalization discrepancies in Italy with respect to the primary care service observed. The PCDI scores highlight a fragmented scenario and opportunities for improvement. • The aspects related to the readability of information remain to be monitored. This study highlights the need to invest more in fostering efficient service comprehension and clarity.

From process to product: writing engagement and performance of EFL learners under computer-generated feedback instruction

Artificial intelligence (AI) technology is gradually penetrating the domain of education, opening up many possibilities for teaching and learning. Many educators, faced with the burden of commenting on substantial student essays, have introduced automated writing evaluation (AWE) into second language (L2) writing considering its affordance of immediate scores and diagnostic information. However, students’ processing strategies and perceptions of such computer-generated feedback and its impact on student writing quality, particularly as mediated by language proficiency, remain under-explored. This study examines the impact of Pigai, a Chinese AWE system, on revision processes and writing products of 42 English as a foreign language (EFL) learners with varying language levels by analyzing feedback points, feedback uptake, text quality in complexity, accuracy, and fluency (CAF), and perceptions. The findings are as follows. When confronted with AWE instruction, the majority of student work focuses on correcting errors, but higher-level students exhibit an emphasis on language improvement beyond the surface level compared to lower-level students. According to CAF measures, automated feedback exerts greater effects on accuracy for unskilled learners and lexical complexity for skilled learners in the development of interlanguage. Learner profiles and perceptions of students at different levels are explored along four dimensions: writing quality, cognitive engagement, behavioral engagement, and affective engagement. Finally, the potential issues of such technology-based writing instruction are pointed out.

Reactivity of the triple task on writing processes and product in adults with dyslexia.

The triple task (TT) is a method for assessing the dynamics of writing processes. It involves three tasks in one: writing a text, responding to a sound, and reporting the process. Previous research has mostly shown that the TT does not affect the writing process or the product. However, individuals with dyslexia often show difficulties in tasks that require organization, automation, integration of multiple processes, inhibition, and shifting/cognitive flexibility. The aim of this study was therefore to investigate whether TT affects the writing process and written product differently in adults with dyslexia compared to a control group of adults with typical reading skills. Two groups of adult native Croatian speakers were included in this study: 20 adults with developmental dyslexia and 20 adults with typical reading skills; evenly distributed by: age (18-38 years), gender (13 males, 7 females per group), educational level, and nonverbal cognitive abilities. All participants wrote one text with a TT and another without. The writing of the text was tracked with a keystroke logging program - Inputlog. The two texts were compared at process and product level. The results showed that measures of writing processes and text quality in the groups of adults with dyslexia and adults with typical reading skills were unlikely to be differentially affected by TT. However, in the condition without TT, the total number of characters per minute was higher, more keys were typed per minute and more words were deleted. As expected, adults with dyslexia produced shorter texts of lower quality and with more errors; they also produced fewer characters per minute, used fewer keystrokes and typed fewer. The study suggests that TT is unlikely to have a different impact on the writing process or written product in adults with dyslexia compared to adults with typical reading skills.

Generative Artificial Intelligence for Chest Radiograph Interpretation in the Emergency Department

Multimodal generative artificial intelligence (AI) methodologies have the potential to optimize emergency department care by producing draft radiology reports from input images. To evaluate the accuracy and quality of AI-generated chest radiograph interpretations in the emergency department setting. This was a retrospective diagnostic study of 500 randomly sampled emergency department encounters at a tertiary care institution including chest radiographs interpreted by both a teleradiology service and on-site attending radiologist from January 2022 to January 2023. An AI interpretation was generated for each radiograph. The 3 radiograph interpretations were each rated in duplicate by 6 emergency department physicians using a 5-point Likert scale. The primary outcome was any difference in Likert scores between radiologist, AI, and teleradiology reports, using a cumulative link mixed model. Secondary analyses compared the probability of each report type containing no clinically significant discrepancy with further stratification by finding presence, using a logistic mixed-effects model. Physician comments on discrepancies were recorded. A total of 500 ED studies were included from 500 unique patients with a mean (SD) age of 53.3 (21.6) years; 282 patients (56.4%) were female. There was a significant association of report type with ratings, with post hoc tests revealing significantly greater scores for AI (mean [SE] score, 3.22 [0.34]; P < .001) and radiologist (mean [SE] score, 3.34 [0.34]; P < .001) reports compared with teleradiology (mean [SE] score, 2.74 [0.34]) reports. AI and radiologist reports were not significantly different. On secondary analysis, there was no difference in the probability of no clinically significant discrepancy between the 3 report types. Further stratification of reports by presence of cardiomegaly, pulmonary edema, pleural effusion, infiltrate, pneumothorax, and support devices also yielded no difference in the probability of containing no clinically significant discrepancy between the report types. In a representative sample of emergency department chest radiographs, results suggest that the generative AI model produced reports of similar clinical accuracy and textual quality to radiologist reports while providing higher textual quality than teleradiologist reports. Implementation of the model in the clinical workflow could enable timely alerts to life-threatening pathology while aiding imaging interpretation and documentation.

Comparison of milling methods on the properties of common buckwheat flour and the quality of wantuan, a traditional Chinese buckwheat food.

The microstructural and techno-functional properties of buckwheat flour and its processability for making wantuan, as affected by different milling methods, were investigated. Results showed that the particle sizes (d(0.5)) of the flours made by stone-milling (SM), hammer-milling (HM), laboratory grinding with steaming pretreatment for 5min (LG-5) and 10min (LG-10) were 95.5, 111.5, 35.4 and 41.1μm, respectively. Moreover, SM and HM flours had less liberated starch granules and 20.84%-24.32% higher relative crystallinity than LG-10 flour. Slurries of laboratory-grinded flours showed excellent suspension stability. LG-10 flour had lowest pasting viscosities but greatest storage modulus and loss modulus. Color differences among the wantuan made from different flours were not visibly perceived (ΔE<5). Wantuan made from LG-5 flour exhibited highest textual quality due to its greatest resilience (0.376), good springiness (0.933) and accepted chewiness (1093.31). Concluding, steaming prior to grinding could improve the qualities of buckwheat flour for wantuan making.

Instruction in creative and argumentative writing: transfer and crossover effects on writing process and text quality

AbstractTo investigate whether a creative writing unit in upper secondary education would improve students’ creative as well as argumentative text quality and to examine whether it would change students’ writing behavior, we tested a creative writing unit based on encouraging writing in flow by using divergent thinking tasks. Four classes (Grade 10) participated in a switching replications design. Students received either creative writing instruction (CWI) or argumentative writing instruction (AWI). Key stroke logging software recorded students’ writing processes, their Creative Self-Concept (CSC) was measured, and text quality was rated holistically. Students were positive about the design of the creative writing unit and the lessons. The effects varied per panel. The first panel showed that CWI had an effect on creative text quality compared to AWI, while AWI had no effect on argumentative text quality, compared to CWI. This pattern indicates a transfer effect of creative writing instruction on argumentative text quality. The transfer effect was moderated by CSC, with larger effects for relatively high CSC-participants. The second panel did not replicate this pattern. Instead, a crossover effect was observed of CWI in panel 1 on the effect of participating in the unit on argumentative writing in panel 2, most pronounced in high CSC-participants. Students’ creative writing speed decreased in the first panel, except for students with a relatively high Creative Self-Concept, and then increased in the second panel. Our findings may guide decisions on incorporating creative writing in the curriculum.

Self-efficacy for writing and written text quality of upper secondary students with and without reading difficulties.

Self-efficacy for writing (SEW) and reading ability are some of several factors that may be related to the quality of written text that students produce. The aim of the current study was (1) to explore the variation in SEW and written text quality in L1-Swedish and L2-English among upper secondary students with different reading profiles in L1 (typical reading vs. reading difficulties) and with different study backgrounds (SB1year or SB2years = one or two years of studies of Swedish and English, respectively), and in the next step (2) to explore if individual variations in L1-reading and SEW may explain variation in written text quality. Participants were 100 upper secondary students (aged 17-18) with different reading profiles operationalized as typical reading and reading difficulties. Data consisted of screening for word recognition and reading comprehension, text quality results from argumentative L1- and L2-writing tasks, school information on study background in Swedish/English, and students' responses from an online survey about SEW. As to SEW results, an ANOVA revealed significant main effects for reading profile and study background in L1, but in L2 there was only a significant main effect for reading profile. Written text quality results indicated that there was a significant interaction effect between reading profile and study background in L1, indicating that the significant main effect for reading profile on written text quality was influenced by the group of students with reading difficulties and SB1year. There was a significant main effect for reading profile and study background on written text quality in L2. Students with reading difficulties and SB1year were the most vulnerable group, and they had the lowest scores in L1/L2 SEW and written text quality in L1 and L2. Multiple regression results indicated that word recognition and SEW contributed significantly to L1-text quality, and word recognition, reading comprehension, and SEW contributed significantly to L2-text quality. Thus, this study sheds light on the under-researched area of L1/L2 SEW and text quality of students with reading difficulties at the level of upper secondary school. Pedagogical implications are discussed and highlight the need for writing instruction across subjects in upper secondary school and for extra writing support/scaffolding for students with reading difficulties and shorter study background in the language subjects L1 (Swedish) and L2 (English).

The Future of Natural History Transcription: Navigating AI advancements with VoucherVision and the Specimen Label Transcription Project (SLTP)

Natural history collections are critical reservoirs of biodiversity information but collections staff are constantly grappling with substantial backlogs and limited resources. The task of transcribing specimen label text into searchable databases requires a significant amount of time, manual labor, and funding. To address this challenge, we introduce VoucherVision, a tool harnessing the capabilities of several Large Language Models (LLMs; Naveed et al. 2023) to augment specimen label transcription. The VoucherVision tool automates laborious components of the transcription process, leveraging an Optical Character Recognition (OCR) system and LLMs to convert unstructured label text into appropriate data formats compatible with database ingestion. VoucherVision uses a combination of structured output parsers and recursive re-prompting strategies to ensure consistency and quality of the LLM-formatted text, significantly reducing errors. Integration of VoucherVision with the University of Michigan Herbarium’s transcription workflow resulted in a significant reduction in per-image transcription time, suggesting significant potential advantages for collections workflows. VoucherVision offers promising strides towards efficient digitization, with curatorial staff playing critical roles in data quality assurance and process oversight. Emphasizing the importance of knowledge sharing, the University of Michigan Herbarium is backing the Specimen Label Transcription Project (SLTP), which will provide open access to benchmarking datasets, fine-tuned models, and validation tools to rank the performance of different methodologies, LLMs, and prompting strategies. In the rapidly evolving landscape of Artificial Intelligence (AI) development, we recognize the profound potential of diverse contributions and innovative methodologies to redefine and advance the transformation of curatorial practices, catalyzing an era of accelerated digitization in natural history collections. An early, public version of VoucherVision is available to try here: https://vouchervision.azurewebsites.net/

The effect of pre-task planning on computer-based second language writing

Drawing on models of first language writing and models of second language (L2) oral task performance, this quasi-experimental study investigated the effect of pre-task planning (PTP) on the product of computer-based L2 writing against a no-planning (NP) condition with the total time-on-task controlled for. It also examined the possible influence of writers’ strategy preferences on the effect of PTP. The study adopted a within-group design. Forty-three participants were required to complete two argumentative essays on a computer with one produced under the PTP condition and the other in the NP condition. A post-task questionnaire was used to collect participants’ strategy preferences. Statistical analyses showed that PTP affected only writing fluency and had no effect on complexity, accuracy, lexis, idea density and coherence and cohesion. Considerable individual differences were found on how learners were affected by the planning conditions but no mediating effect of strategy preferences was evident. Based on the results, we argue that teachers can require students to pre-plan when they want to encourage fluent writing but should not expect PTP to improve text quality in computer-based writing. Theoretical implications are discussed in relation to the different theoretical positions on the effect of PTP.

The proof of the translation process is in the reading of the target text: An eyetracking reception study

This article is an attempt to bridge the divide between translation process research (TPR) which has investigated how translators as specialised bilingual professionals use their expertise to translate texts and translation reception which explores how the texts are read and received by the target language readers. Over the last thirty years, TPR has provided empirically grounded findings to demonstrate the complexity of the cognitive processes in the translator's mind but much less empirical interest has been paid to how translated texts are read and processed by the readers. To redress this imbalance, we hypothesise that the cognitive effort invested in reading a translated text can be taken as proof of how successful the translation process has been. We report on an exploratory study in which two groups of participants read a high-quality and a low-quality translation of the same text while their eye movements were recorded by an eyetracker. We compare the readers' cognitive effort indexed by character-adjusted dwell time, number of runs and re-reading in the second and third run with the translators' character-adjusted cognitive effort invested in producing the target texts. The results show that the relationship between the translation process and the reading experience is not straightforward and depends on the quality of the target text.

‘Up to you’: Self-help books, depression and the reconstruction of reading

Like other consumers of ‘low-brow’ genres, self-help readers elicit polarized views in the literature. While little research to date has focused specifically on self-help readers with a history of mental illness, existing commentary reveals a particular tension: on the one hand, clinical researchers report positive outcomes for depressed readers engaged in bibliotherapy programs using self-help books, similar or superior to medication or talk therapy; on the other, scholars of media and culture express misgivings about the quality of self-help texts and highlight the negative potential of therapeutic discourse for individual readers and audiences more generally. By asking what actual readers do with self-help books, however, my research suggests an altogether more complex interaction between readers and the books they choose and use – especially as they navigate experiences of mental illness. Leveraging a reader-response heuristic in which I interviewed a cohort of Australian readers, this paper details some of the ways in which habitual consumers of self-help books describe their own interpretive activities, problematizing previous research that either emphasizes or downplays the significant expertise of vernacular audiences.

Neural Data-to-Text Generation Based on Small Datasets: Comparing the Added Value of Two Semi-Supervised Learning Approaches on Top of a Large Language Model

AbstractThis study discusses the effect of semi-supervised learning in combination with pretrained language models for data-to-text generation. It is not known whether semi-supervised learning is still helpful when a large-scale language model is also supplemented. This study aims to answer this question by comparing a data-to-text system only supplemented with a language model, to two data-to-text systems that are additionally enriched by a data augmentation or a pseudo-labeling semi-supervised learning approach.Results show that semi-supervised learning results in higher scores on diversity metrics. In terms of output quality, extending the training set of a data-to-text system with a language model using the pseudo-labeling approach did increase text quality scores, but the data augmentation approach yielded similar scores to the system without training set extension. These results indicate that semi-supervised learning approaches can bolster output quality and diversity, even when a language model is also present.

Write, talk and rewrite: the effectiveness of a dialogic writing intervention in upper elementary education

In this research, we developed and empirically tested a dialogic writing intervention, an integrated language approach in which grade 5/6 students learn how to write, talk about their writing with peers, and rewrite. The effectiveness of this intervention was experimentally tested in ten classes from eight schools, using a pretest–posttest control group design. Classes were randomly assigned to the intervention group (5 classes; 95 students) or control group (5 classes; 115 students). Both groups followed the same eight lessons in which students wrote four argumentative texts about sustainability. For each text, students wrote a draft version, which they discussed in groups of three students. Based on these peer conversations, students revised their text. The intervention group received additional support to foster dialogic peer conversations, including a conversation chart for students and a practice-based professional development program for teachers. Improvements in writing were measured by an argumentative writing task (same genre, but different topic; near transfer) and an instructional writing task (different genre and topic; far transfer). Text quality was holistically assessed using a benchmark rating procedure. Results show that our dialogic writing intervention with support for dialogic talk significantly improved students’ argumentative writing skills (ES = 1.09), but that the effects were not automatically transferable to another genre. Based on these results we conclude that a dialogic writing intervention is a promising approach to teach students how to talk about their texts and to write texts that are more persuasive to readers.

Correlates of Early Handwriting: Differential Patterns for Girls and Boys

ABSTRACT Research Findings: Fluent and legible handwriting is associated with longer texts and higher text quality and is therefore an important goal in the primary school curriculum. While girls’ handwriting tends to be more proficient than boys’, potential differences in the correlates of girls’ and boys’ early handwriting are poorly understood. In this exploratory study, we investigated early handwriting (legibility, fluency, and time) and potential motor (fine motor skills, visuomotor integration) and cognitive (executive functions: inhibition, shifting, working memory) correlates in a sample of 118 first-grade children (46% girls, M age: 7 years, 5 months). We tested gender differences and explored correlates of early handwriting skills of girls and boys. Extending previous findings, fine motor skills and visuomotor integration were related to early handwriting legibility, fluency, and time. Furthermore, shifting and working memory, but not inhibition, were related to specific handwriting skills. Moreover, girls outperformed boys regarding fine motor skills, visuomotor integration, and handwriting legibility. Interestingly, while the correlates of handwriting were diverse and strong for girls, only a few weak correlates were identified for boys. Practice or Policy: These results lay the ground for tailoring adequate interventions to support girls and boys (differently) in acquiring fluent and legible handwriting.

Semantic Coherence Analysis of English Texts Based on Sentence Semantic Graphs

With the reform of China's education industry, more and more universities are using computers to conduct examinations. For the automatic correction of essays as subjective questions, existing automatic English text scoring systems suffer from insufficient extraction of coherence information and low accuracy when analysing text coherence. Therefore, this paper proposes an unsupervised semantic coherence analysis model for English texts based on sentence semantic graphs, taking Chinese students' English compositions as the research context. Guided by the semantic coherence theory, the English text is represented as a sentence semantic graph, and an improved VF2 subgraph matching algorithm is used to mine the frequently occurring subgraph patterns in the sentence semantic graph. After that, the set of frequent subgraphs is generated by filtering the subgraph patterns according to their frequencies, and the subgraph frequency of each frequent subgraph is calculated separately. Finally, the distribution characteristics of frequent subgraphs and the semantic values of subgraphs in the sentence semantic graphs are extracted to quantify the overall coherence quality of English texts. The experimental results show that the model proposed in this paper has higher accuracy and practical value compared with the current methods of coherence analysis.

Comparative approaches to the assessment of writing: Reliability and validity of benchmark rating and comparative judgement

In the past years, comparative assessment approaches have gained ground as a viable method to assess text quality. Instead of providing absolute scores to a text as in holistic or analytic scoring methods, raters in comparative assessments rate text quality by comparing texts either to pre-selected benchmarks representing different levels of writing quality (i.e., benchmark rating method) or by a series of pairwise comparisons to other texts in the sample (i.e., comparative judgement; CJ). In the present study, text quality scores from the benchmarking method and CJ are compared in terms of their reliability, convergent validity and scoring distribution. Results show that benchmark ratings and CJ-ratings were highly consistent and converged to the same construct of text quality. However, the distribution of benchmark ratings showed a central tendency. It is discussed how both methods can be integrated and used such that writing can be assessed reliably, validly, but also efficiently in both writing research and practice.

The Role of Phonological Errors in Evaluation Metrics

In recent years, Natural Language Processing (NLP) has seen a surge in research, particularly in the &#x0D; areas of text summarization and machine translation. Evaluation metrics like ROUGE and BLEU have been &#x0D; widely used to assess the quality of texts using N-gram based approaches. However, these metrics often struggle &#x0D; when applied to data sourced from the internet, such as social media platforms, due to the prevalence of &#x0D; phonological errors. This study focuses on identifying the sources and frequency of phonological errors while &#x0D; addressing the question of whether they should be considered or not. Data from Twitter, a platform known for &#x0D; phonological errors, was collected, and studied, along with existing literature on the subject. The article proposes &#x0D; enhancing existing metrics by integrating edit distance algorithms like Levenshtein or Damerau-Levenshtein. By &#x0D; considering phonological errors in evaluations, this approach aims to improve accuracy and reliability in the NLP &#x0D; and machine translation domains. The ultimate goal of this study is to contribute to more sensitive and reliable &#x0D; evaluation metrics in these fields.

Arabic ChatGPT Tweets Classification Using RoBERTa and BERT Ensemble Model

ChatGPT OpenAI, a large-language chatbot model, has gained a lot of attention due to its popularity and impressive performance in many natural language processing tasks. ChatGPT produces superior answers to a wide range of real-world human questions and generates human-like text. The new OpenAI ChatGPT technology may have some strengths and weaknesses at this early stage. Users have reported early opinions about the ChatGPT features, and their feedback is essential to recognize and fix its shortcomings and issues. This study uses the ChatGPT tweets Arabic dataset to automatically find user opinions and sentiments about ChatGPT technology. The dataset is preprocessed and labeled using the TextBlob Arabic Python library into positive, negative, and neutral tweets. Despite extensive works for the English language, languages like Arabic are less studied regarding tweet analysis. Existing literature about Arabic tweet sentiment analysis has mainly focused on machine learning and deep learning models. We collected a total of 27,780 unstructured tweets from Twitter using the Tweepy SNscrape Python library using various hash-tags such as # Chat-GPT, #OpenAI, #Chatbot, Chat-GPT3, and so on. To enhance the model’s performance and reduce computational complexity, unstructured tweets are converted into structured and normalized forms. Tweets contain missing values, URL and HTML tags, stop words, punctuation, diacritics, elongations, and numeric values that have no impact on the model performance; hence, these increase the computational cost. So, these steps are removed with the help of Python preprocessing libraries to enhance text quality and consistency. This study adopts Transformer-based models such as RoBERTa, XLNet, and DistilBERT that automatically classify the tweets. Additionally, a hybrid transformer-based model is proposed to obtain better results. The proposed hybrid model is developed by combining the hidden outputs of the RoBERTA and BERT models using a concatenation layer, then adding dense layers with “Relu” activation employed as a hidden layer to create non-linearity and a “softmax” activation function for multiclass classification. They differ from existing state-of-the-art models due to the enhanced capabilities of both models in text classification. Hybrid models combine the different models to make accurate predictions and reduce bias and enhanced the overall results, while state-of-the-art models are incapable of making accurate predictions. Experiments show that the proposed hybrid model achieves 96.02% accuracy, 100% precision on negative tweets, and 99% recall for neutral tweets. The performance of the proposed model is far better than existing state-of-the-art models.