Scholarlink

<p>Deep learning leverages textual feedback and contextual cues to improve rating prediction accuracy through simulated cognitive reasoning. User reviews provide critical insights for accurate preference modeling. Sentiment analysis identifies intrinsic preferences and behavioral patterns, improving user requirement comprehension. An attention-based deep learning framework with sentiment-infused preference modeling (DeepSU) is proposed to emulate human cognition effectively. Multidimensional preferences are captured by extracting salient features from travel platforms. Sentiment consistency detection identifies anomalous reviews by comparing inferred sentiment with ratings to refine the dataset. A co-attention mechanism integrates user features and sentiment estimations through learned attention weights, generating embeddings that encode personalized characteristics and affective tendencies toward attractions. Experimental results demonstrate that DeepSU outperforms the strongest baseline by 15.2% and 13.3% on the Ctrip travel dataset. It achieves lower RMSE than baseline methods under data sparsity and maintains performance advantages as the input scale increases, while keeping training time minimal. Ablation studies indicate that the sentiment analysis module contributes more significantly to overall performance than the user preference module. Evaluations on datasets from two distinct travel platforms demonstrate consistent superiority over baseline techniques, confirming robust generalization and reliability.</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

Revisiting cognitive process-driven model with sentiment-infused user preference for recommendation

  • 期刊论文

  • 发表日期：

    2025年9月4日

查看全文
  • Tingting Chen,
  • Suqing Lin,
  • Jingheng Wu,
  • Shuhua Zhang

摘要

Deep learning leverages textual feedback and contextual cues to improve rating prediction accuracy through simulated cognitive reasoning. User reviews provide critical insights for accurate preference modeling. Sentiment analysis identifies intrinsic preferences and behavioral patterns, improving user requirement comprehension. An attention-based deep learning framework with sentiment-infused preference modeling (DeepSU) is proposed to emulate human cognition effectively. Multidimensional preferences are captured by extracting salient features from travel platforms. Sentiment consistency detection identifies anomalous reviews by comparing inferred sentiment with ratings to refine the dataset. A co-attention mechanism integrates user features and sentiment estimations through learned attention weights, generating embeddings that encode personalized characteristics and affective tendencies toward attractions. Experimental results demonstrate that DeepSU outperforms the strongest baseline by 15.2% and 13.3% on the Ctrip travel dataset. It achieves lower RMSE than baseline methods under data sparsity and maintains performance advantages as the input scale increases, while keeping training time minimal. Ablation studies indicate that the sentiment analysis module contributes more significantly to overall performance than the user preference module. Evaluations on datasets from two distinct travel platforms demonstrate consistent superiority over baseline techniques, confirming robust generalization and reliability.