Scholarlink

<p>The booming online drug reviews have opened up the prospects of AI-based drug recommendation (DR) systems. Nevertheless, the majority of the current methods are directed at generic sentiment classification or popularity-based ranking that do not have the capacity to integrate patient-specific sentiment intelligence, and thus, cannot be used to make highly personalized suggestions. This disparity restricts their usefulness in helping patients and physicians with accurate medication decisions. To overcome this, the proposed MedRecommenderX, a scalable, sentiment-aware framework, which uses Natural Language Processing (NLP) methods to convert patient-generated drug reviews into DR outputs that are personalized. The framework starts with review preprocessing text cleaning, tokenization, stop-word removal, and lemmatization and then feature extraction by Bag of Words (BoW), Term Frequency Inverse Document Frequency (TF-IDF) and Word2Vec. These characteristics are used to train several sentiment classification models, such as Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Naive Bayes, Recurrent Neural Networks (RNNs), and Transformers to make predictions about sentiment labels (negative, neutral, positive). The results of sentiment analysis are then applied in the K-Means clustering in order to cluster drugs that have similar sentiments. Lastly, cosine similarity and sentiment scores are used to derive recommendations by calculating a personalized recommendation score per patient. The results of the experiments show that the MedRecommenderX model is more effective than the benchmark models like Stacked ANN, Logic-Operator Neural Network (LONN), Particle Swarm Optimization-Artificial Neural Network (PSO-ANN), and LSTM in terms of accuracy, 97.8 and 98.2% based on the evaluation measures, and thus effective in providing accurate and personalized drug recommendations based on sentiment analysis (SA). This is attributed to the hybrid nature of MedRecommenderX, which allows multiple NLP-based feature extraction techniques to be combined with different sentiment classifiers to represent sentiment richly and cluster more accurately than deep learning-only models such as LSTM.</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

MedRecommenderX: a scalable sentiment-aware framework for personalized drug recommendations

  • 期刊论文

  • 发表日期：

    2025年11月28日

查看全文
  • Deepika K.,
  • Harika Vanam

摘要

The booming online drug reviews have opened up the prospects of AI-based drug recommendation (DR) systems. Nevertheless, the majority of the current methods are directed at generic sentiment classification or popularity-based ranking that do not have the capacity to integrate patient-specific sentiment intelligence, and thus, cannot be used to make highly personalized suggestions. This disparity restricts their usefulness in helping patients and physicians with accurate medication decisions. To overcome this, the proposed MedRecommenderX, a scalable, sentiment-aware framework, which uses Natural Language Processing (NLP) methods to convert patient-generated drug reviews into DR outputs that are personalized. The framework starts with review preprocessing text cleaning, tokenization, stop-word removal, and lemmatization and then feature extraction by Bag of Words (BoW), Term Frequency Inverse Document Frequency (TF-IDF) and Word2Vec. These characteristics are used to train several sentiment classification models, such as Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Naive Bayes, Recurrent Neural Networks (RNNs), and Transformers to make predictions about sentiment labels (negative, neutral, positive). The results of sentiment analysis are then applied in the K-Means clustering in order to cluster drugs that have similar sentiments. Lastly, cosine similarity and sentiment scores are used to derive recommendations by calculating a personalized recommendation score per patient. The results of the experiments show that the MedRecommenderX model is more effective than the benchmark models like Stacked ANN, Logic-Operator Neural Network (LONN), Particle Swarm Optimization-Artificial Neural Network (PSO-ANN), and LSTM in terms of accuracy, 97.8 and 98.2% based on the evaluation measures, and thus effective in providing accurate and personalized drug recommendations based on sentiment analysis (SA). This is attributed to the hybrid nature of MedRecommenderX, which allows multiple NLP-based feature extraction techniques to be combined with different sentiment classifiers to represent sentiment richly and cluster more accurately than deep learning-only models such as LSTM.