Scholarlink

<p>In an era defined by digital commerce, the explosion of online consumer behaviour has generated vast reservoirs of opinionated data, shaping purchase decisions and brand perceptions alike. However, under such a competitive market regime, spam opinions are generated and manufactured to favour certain inferior products in an unethical manner. To mitigate this, the given paper introduces a bidirectional gated recurrent neural network (Bi-GRU) to detect spam in consumer reviews using glove-based word embeddings. The proposed model (Hybrid Bert Variant Bi-Directional GRU Spam Detection or HBVBGSD) is a combination of two different BERT (Bidirectional encoder representations from transformers) variants, namely DistilBERT and RoBERTa neural network (Robustly optimized BERT pretraining approach) along with Bi-Directional Gated-Recurrent Unit (Bi-GRU). The proposed hybrid transformer model allows the representation of the consumer reviews text into word vectors, which are then passed into the downstream Bidirectional-GRU architecture for further refinement and semantic extraction from given words, subsequently helping to identify spam or ham in the given consumer reviews text. Also, the application of the modified Ada-Belief optimizer has improved accuracy and minimized overfitting or underfitting constraints. The results show superior performance of the proposed HBVBGSD model in terms of accuracy and loss, thus outperforming existing pre-trained transformer models.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hybrid transformer Bi-GRU model with modified ada-belief optimizer for detecting spam in consumer reviews

  • 期刊论文

  • 发表日期：

    2025年9月11日

查看全文
  • Sourav Sinha,
  • Revathi Sathiya Narayanan,
  • Indrajit Mukherjee

摘要

In an era defined by digital commerce, the explosion of online consumer behaviour has generated vast reservoirs of opinionated data, shaping purchase decisions and brand perceptions alike. However, under such a competitive market regime, spam opinions are generated and manufactured to favour certain inferior products in an unethical manner. To mitigate this, the given paper introduces a bidirectional gated recurrent neural network (Bi-GRU) to detect spam in consumer reviews using glove-based word embeddings. The proposed model (Hybrid Bert Variant Bi-Directional GRU Spam Detection or HBVBGSD) is a combination of two different BERT (Bidirectional encoder representations from transformers) variants, namely DistilBERT and RoBERTa neural network (Robustly optimized BERT pretraining approach) along with Bi-Directional Gated-Recurrent Unit (Bi-GRU). The proposed hybrid transformer model allows the representation of the consumer reviews text into word vectors, which are then passed into the downstream Bidirectional-GRU architecture for further refinement and semantic extraction from given words, subsequently helping to identify spam or ham in the given consumer reviews text. Also, the application of the modified Ada-Belief optimizer has improved accuracy and minimized overfitting or underfitting constraints. The results show superior performance of the proposed HBVBGSD model in terms of accuracy and loss, thus outperforming existing pre-trained transformer models.