{"paper_id":"2508.09935","title":"Language of Persuasion and Misrepresentation in Business Communication: A Textual Detection Approach","authors":["Sayem Hossen","Monalisa Moon Joti","Md. Golam Rashed"],"paper_date":"2025-08-13","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2508.09935v1","page_url":"https://thequant.space/flowcharts/language-of-persuasion-and-misrepresentation-in-business-com/","scores":{"empirical_rigor":4,"math_complexity":2,"quadrant":"Philosophers"},"proposes":{"status":"available","statement":"The paper proposes deep‑learning architectures for detecting persuasive and deceptive language in business communication and provides a new annotated dataset.","quote":"The study fills a gap in the present research with the creation of deep‑learning architectures that differentiate between persuasion and deception.","summary":"Business communication digitisation has reorganised the process of persuasive discourse, which allows not only greater transparency but also advanced deception. This inquiry synthesises classical rhetoric and communication psychology with linguistic theory and empirical studies in the financial…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"The analysis assumes that English‑only texts are sufficient for model development.","type":"data","stated":"explicit","quote":"Also, only using English reinforces the English‑before‑everything bias present in (Bochkay et al., 2023)[3]"},{"assumption":"The dataset is assumed to be representative despite its small size (4,848 instances).","type":"data","stated":"explicit","quote":"Several limitations come out of the present study. To begin with, the data (number of instances 4,848) is relatively small"},{"assumption":"Class distributions are assumed to be preserved by stratified sampling to avoid over‑fitting.","type":"model","stated":"explicit","quote":"The division was performed using stratified sampling, a published approach to alleviating the problem of overfitting and enhancing generalisability"},{"assumption":"Multi‑class label encoding assumes that the three categories (Factual, Misleading, Persuasive) are distinct enough for classification.","type":"model","stated":"explicit","quote":"Multi‑class label encoding was done by numerically mapping the categorical target classes: Factual, Misleading, and Persuasive"},{"assumption":"The models are evaluated assuming that training/validation accuracy reflects real‑world generalisation.","type":"statistical","stated":"implied","quote":"The slight difference between the training and validation metrics shows that it might have a good generalization with minimal overfitting."}]},"datasets":{"status":"available","items":[{"name":"Annotated business communications","provider":null,"kind":"text-news","access":"unknown","period":null,"universe":null,"quote":"The 4,848 annotated business communications form a rich, standardised dataset to study deceptive behaviours in business"}],"known":[{"id":"twitter","name":"Twitter / X","provider":"X Corp.","kind":"text-news","access":"public","quote":"The information was drawn from 13 heterogeneous modalities that include marketing emails, social-media advertising on Twitter/X Facebook, financial statements, lies in corporate statements, newspaper articles, LinkedIn postings, YouTube advertising, international-business reports, and marketing-criteria reports.","mentions":1,"used_in":1}],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"none"},"transaction_costs":{"status":"not_applicable","modeled":"n/a","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R3","candidates":{"front_matter":"n/a","evidence_card":"n/a","evidence_card_quoted":false},"meaning":"the paper reports no trading or execution result"},"results":{"status":"available","evaluation_type":"forecast-accuracy","findings":[{"finding":"In controlled settings, detection accuracies greater than 99% were achieved using computational textual analysis and transformer models.","kind":"predictive-accuracy","quote":"In controlled settings, detection accuracies of greater than 99% were achieved by using computational textual analysis as well as personalised transformer models."},{"finding":"The Custom Attention model attained the highest accuracy (≈0.99 training, 0.976 validation) among all tested architectures.","kind":"performance","quote":"The Custom Attention model, in particular, has the highest accuracy level of all the tried settings."},{"finding":"Simple Bidirectional LSTM reached validation accuracy of approximately 0.975 after early epochs.","kind":"performance","quote":"the validation accuracy (orange line) plateaus at an approximate of 0.975."},{"finding":"Advanced Bidirectional LSTM and Transformer models showed small gaps between training and validation metrics, indicating good generalisation and low over‑fitting risk.","kind":"performance","quote":"considered architectures as it has the smallest gap between the training and validation measures."},{"finding":"Misclassification mainly occurs between Persuasive and Misleading classes, reflecting their conceptual and lexical overlap.","kind":"statistical","quote":"Second, incorrect classification of Persuasive and Misleading datasets is the primary source of error for all the models"}],"reported_performance":null,"out_of_sample":{"method":"unclear","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":42,"citations":{"cited_by_in_corpus":0,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":0},"independent_papers":[],"scope":"Citations between papers in this corpus only; citations from outside it are not counted."}},"implementation":{"status":"available","public":false,"code_url":null,"note":"No repository presented by the authors as their own was found in the paper. Absence here is not proof that no code exists."},"replication":{"status":"none_found","card":null,"note":"No replication of this paper has been run by thequant.space yet."},"extraction":{"text_source":"fulltext","model":"openai/gpt-oss-120b","version":1,"extracted_at":"2026-11-14T17:30:00Z"},"provenance":{"claims":"Everything here is what the authors state in the paper. Nothing was re-run or reproduced by the platform.","quotes":"Assumptions, datasets, cost detail and findings were extracted by a language model. Every quote is verbatim paper text. Where an item has `support`, \"full\" means every figure in it appears in the quoted sentences and \"partial\" means a figure is stated elsewhere in the paper; items with a figure found nowhere in the paper were discarded.","datasets_known":"datasets.known lists well-known vendors and public datasets by a stable id, found by a dictionary over the full text where the sentence says the data is used; it is independent of the model and covers only names in the dictionary.","flags":"transaction_costs.modeled, results.evaluation_type and results.out_of_sample.method are platform classifications (structured-decision models); see https://thequant.space/score-guide/#penalty-flags","comparisons":"supports / contradicts / qualifies are a language model's reading of two papers' quoted findings, restricted to closely related papers that share no author. They are pointers to read, not verified replications, and coverage is partial.","implementation":"code_url is a repository the paper itself presents as its own (rule-scored, ambiguous cases judged by a model), checked for liveness when it was recorded."},"docs":"https://thequant.space/agents/"}