An Integrated Deep Learning Approach for Enhanced Fake Review Detection on E Commerce Platforms Using Multimodal and Fine-Grained Features
Fuente:
Zenodo
Saved in:
| Main Author: | |
|---|---|
| Format: | Recurso digital |
| Published: |
Zenodo
2026
|
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866902269820141568 |
|---|---|
| author | Deshan Sachintha Kannangara, Prof. Chunyong Yin, Sachini Amani Henda Vitharana |
| author_facet | Deshan Sachintha Kannangara, Prof. Chunyong Yin, Sachini Amani Henda Vitharana |
| contents | <div class="box"> <div class="abs-text">Fake reviews are one of the greatest problems in e-commerce sites, which has a great impact on consumer confidence and buying behaviors. As the AI generated and synthetic content is expanding at a rapid pace, it has become harder to distinguish between fake review patterns. In this paper, the authors suggest a trimodal deep learning model to identify fake reviews, through the textual analysis, visual analysis, and identification of 22 handcrafted numeric features. A fine-tuned Bidirectional Encoder Representations from Transformers (BERT) model is used to represent the textual modality, whereas the visual modality is represented by image features obtained through the use of a ResNet-50-based convolutional neural network. Further, the architecture includes a specific neural sub network that processes 22 numeric variables such as readability-related variables and image quality variables. The model, rather than basic early fusion, utilizes an attention-based fusion mechanism which learns dynamic weights of importance of the three modalities to use the complementary signals more effectively in making a classification. The given framework was tested on one Kaggle multimodal fake review dataset which included 20,144 restaurant reviews. The results of the experiment indicate that the model was more effective than the baseline classifiers, such as Logistic Regression and Random Forest with accuracy of 99.95, F1-score of 99.95, and ROC-AUC of 99.99. These results show the usefulness of an attention based trimodal learning in identifying advanced fake reviews in online markets.</div> </div> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_20350656 |
| institution | Zenodo |
| language | |
| publishDate | 2026 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | An Integrated Deep Learning Approach for Enhanced Fake Review Detection on E Commerce Platforms Using Multimodal and Fine-Grained Features Deshan Sachintha Kannangara, Prof. Chunyong Yin, Sachini Amani Henda Vitharana <div class="box"> <div class="abs-text">Fake reviews are one of the greatest problems in e-commerce sites, which has a great impact on consumer confidence and buying behaviors. As the AI generated and synthetic content is expanding at a rapid pace, it has become harder to distinguish between fake review patterns. In this paper, the authors suggest a trimodal deep learning model to identify fake reviews, through the textual analysis, visual analysis, and identification of 22 handcrafted numeric features. A fine-tuned Bidirectional Encoder Representations from Transformers (BERT) model is used to represent the textual modality, whereas the visual modality is represented by image features obtained through the use of a ResNet-50-based convolutional neural network. Further, the architecture includes a specific neural sub network that processes 22 numeric variables such as readability-related variables and image quality variables. The model, rather than basic early fusion, utilizes an attention-based fusion mechanism which learns dynamic weights of importance of the three modalities to use the complementary signals more effectively in making a classification. The given framework was tested on one Kaggle multimodal fake review dataset which included 20,144 restaurant reviews. The results of the experiment indicate that the model was more effective than the baseline classifiers, such as Logistic Regression and Random Forest with accuracy of 99.95, F1-score of 99.95, and ROC-AUC of 99.99. These results show the usefulness of an attention based trimodal learning in identifying advanced fake reviews in online markets.</div> </div> |
| title | An Integrated Deep Learning Approach for Enhanced Fake Review Detection on E Commerce Platforms Using Multimodal and Fine-Grained Features |
| url | https://doi.org/10.5281/zenodo.20350656 |