Scholarlink

As today’s world has become addicted to the internet, Recommendation Systems (RS) continuously revolve around us online. These online systems spend a major part of their resources to make the RS work efficiently and effectively so as to increase their profits. But, many RS are being evaluated though the traditional evaluation parameters like Precision, Recall and F1 Score. As the trend of language-based systems has arrived, recommendations in these systems should also be evaluated through the appropriate metrics. To assess the quality of the generated texts, many linguistic based metrics exist like BLEU, ROUGE, BERTScore and BARTScore. Out of these, BLEU, ROUGE and BARTScore don’t consider semantics of the generated text. So, to overcome this drawback, BERTScore is used which considers both syntactic and semantics of the generated text. Moreover, BARTScore has the advantage as it focuses on measuring the linguistic quality and coherence of generated text. So, in this paper, we propose a novel metric Semantic Recommendation Score (SRS) which takes the advantages of both BERTScore and BARTScore.

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

Bridging Gap Between Semantic Understanding and Linguistic Quality in Recommender Systems: A Semantic Recommendation Score (SRS) for Evaluation

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Kirti Jain,
  • Rajni Jindal

摘要

As today’s world has become addicted to the internet, Recommendation Systems (RS) continuously revolve around us online. These online systems spend a major part of their resources to make the RS work efficiently and effectively so as to increase their profits. But, many RS are being evaluated though the traditional evaluation parameters like Precision, Recall and F1 Score. As the trend of language-based systems has arrived, recommendations in these systems should also be evaluated through the appropriate metrics. To assess the quality of the generated texts, many linguistic based metrics exist like BLEU, ROUGE, BERTScore and BARTScore. Out of these, BLEU, ROUGE and BARTScore don’t consider semantics of the generated text. So, to overcome this drawback, BERTScore is used which considers both syntactic and semantics of the generated text. Moreover, BARTScore has the advantage as it focuses on measuring the linguistic quality and coherence of generated text. So, in this paper, we propose a novel metric Semantic Recommendation Score (SRS) which takes the advantages of both BERTScore and BARTScore.