Use of Transfer Learning for Affordable In-Context Fake Review Generation
Gespeichert in:
| Titel: | Use of Transfer Learning for Affordable In-Context Fake Review Generation |
|---|---|
| Autoren: | Luis Ibañez-Lissen, Lorena González-Manzano, José M. de Fuentes, Manuel Goyanes |
| Weitere Verfasser: | Comunidad de Madrid, Ministerio de Universidades (España), Universidad Carlos III de Madrid, European Commission |
| Quelle: | e-Archivo. Repositorio Institucional de la Universidad Carlos III de Madrid Universidad Carlos III de Madrid (UC3M) |
| Verlagsinformationen: | Institute of Electrical and Electronics Engineers (IEEE), 2025. |
| Publikationsjahr: | 2025 |
| Schlagwörter: | fake reviews, Informática, transfer-learning in-context, natural language processing (NLP), transformers |
| Beschreibung: | Fake reviews are a threat to the trust of online shopping platforms. To produce them, attackers may use assorted techniques based on machine learning. Transfer learning may enable them to leverage already trained models, thus reducing the training requirements. However, the feasibility of these techniques for generating in-context targeted fake reviews has not been explored yet. To address this issue, this paper analyses the suitability of transfer learning using existing models (TS and BART) on different domains (restaurants and technological products). Our results show that: (1) Our work reaches better realism and diversity than previous proposals using artificial intelligence techniques; (2) Our reviews produced with TS can be spot by an automatic detector with a precision of 49% at the highest; (3) Human detection is only 6% better than random guessing, at the highest; and (4) only 1 hour of training and Sk real reviews are needed to produce realistic fake reviews. |
| Publikationsart: | Article |
| Dateibeschreibung: | application/pdf |
| ISSN: | 2372-2096 |
| DOI: | 10.1109/tbdata.2025.3536927 |
| Zugangs-URL: | https://hdl.handle.net/10016/39250 |
| Rights: | CC BY |
| Dokumentencode: | edsair.doi.dedup.....22d9db50eee980f5e3131c509da0c7f1 |
| Datenbank: | OpenAIRE |
Nájsť tento článok vo Web of Science | Abstract: | Fake reviews are a threat to the trust of online shopping platforms. To produce them, attackers may use assorted techniques based on machine learning. Transfer learning may enable them to leverage already trained models, thus reducing the training requirements. However, the feasibility of these techniques for generating in-context targeted fake reviews has not been explored yet. To address this issue, this paper analyses the suitability of transfer learning using existing models (TS and BART) on different domains (restaurants and technological products). Our results show that: (1) Our work reaches better realism and diversity than previous proposals using artificial intelligence techniques; (2) Our reviews produced with TS can be spot by an automatic detector with a precision of 49% at the highest; (3) Human detection is only 6% better than random guessing, at the highest; and (4) only 1 hour of training and Sk real reviews are needed to produce realistic fake reviews. |
|---|---|
| ISSN: | 23722096 |
| DOI: | 10.1109/tbdata.2025.3536927 |