Scholarlink

<p>Artificial Intelligence (AI) has the potential to revolutionize the medical and pharmaceutical sectors. AI and related technologies can significantly address some supply and demand challenges in the pharmaceutical system, such as pharmaceutical AI assistants, chatbot technology, and PharmaRobots. Generative AI (GenAI) is a type of AI that aims to produce new content rather than merely recognizing it. One of the most significant advances in natural-language processing (NLP) in recent years is the development of Large Language Models (LLMs). In this study, we propose a chatbot system to support pharmacists through the development of a pharmaceutical chatbot assistant, called PharmaBot, which is proficient in delivering accurate and contextually relevant responses concerning medications. To this end, we developed a general architectural design that focuses on tailoring LLMs, utilizing retrieval-augmented generation (RAG) and Knowledge Graphs (KGs) to evaluate their performance with specific pharmaceutical resources. A comprehensive knowledge base was constructed by meticulously preprocessing 18,698 pharmaceutical files from the Vidal Group. A key architectural innovation is a dual-embedding strategy that captures both semantic and structural information, facilitating nuanced and context-aware similarity searches. By adopting sophisticated evaluation measures, such as ROUGE, BERTScore, METEOR, and Cosine Similarity, the effectiveness of the algorithms used in producing precise and cohesive summaries was evaluated. The empirical findings conclusively identified that the Google Gemma 2 27B model is the optimal configuration, as it achieved the highest performance across all quality metrics (BERTScore F1-score: 0.834, ROUGE-L F1-score: 0.561, and METEOR score: 0.616), and demonstrates exceptional efficiency with the shortest average response time. Based on the aggregated proposals and findings in the existing literature, this paper concludes with a set of challenges and research recommendations, hopefully contributing to guiding research in the extremely active pharmaceutical domain.</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

Pharmaceutical chatbot assistant using generative AI and knowledge graph with specific pharmaceutical resources

  • 期刊论文

  • 发表日期：

    2026年4月13日

查看全文
  • Brahami Menaouer,
  • Chalabi Younes,
  • Elouissi Elmehdi Mokhtar,
  • Sabri Mohammed

摘要

Artificial Intelligence (AI) has the potential to revolutionize the medical and pharmaceutical sectors. AI and related technologies can significantly address some supply and demand challenges in the pharmaceutical system, such as pharmaceutical AI assistants, chatbot technology, and PharmaRobots. Generative AI (GenAI) is a type of AI that aims to produce new content rather than merely recognizing it. One of the most significant advances in natural-language processing (NLP) in recent years is the development of Large Language Models (LLMs). In this study, we propose a chatbot system to support pharmacists through the development of a pharmaceutical chatbot assistant, called PharmaBot, which is proficient in delivering accurate and contextually relevant responses concerning medications. To this end, we developed a general architectural design that focuses on tailoring LLMs, utilizing retrieval-augmented generation (RAG) and Knowledge Graphs (KGs) to evaluate their performance with specific pharmaceutical resources. A comprehensive knowledge base was constructed by meticulously preprocessing 18,698 pharmaceutical files from the Vidal Group. A key architectural innovation is a dual-embedding strategy that captures both semantic and structural information, facilitating nuanced and context-aware similarity searches. By adopting sophisticated evaluation measures, such as ROUGE, BERTScore, METEOR, and Cosine Similarity, the effectiveness of the algorithms used in producing precise and cohesive summaries was evaluated. The empirical findings conclusively identified that the Google Gemma 2 27B model is the optimal configuration, as it achieved the highest performance across all quality metrics (BERTScore F1-score: 0.834, ROUGE-L F1-score: 0.561, and METEOR score: 0.616), and demonstrates exceptional efficiency with the shortest average response time. Based on the aggregated proposals and findings in the existing literature, this paper concludes with a set of challenges and research recommendations, hopefully contributing to guiding research in the extremely active pharmaceutical domain.