Scholarlink

Platforms like Quora, Reddit, and Stack Overflow are widely used by users to share and respond to a large volume of questions and answers. Though these platforms provide user-generated content, they struggle with repetitive or similar queries. The same question is frequently asked by several users in different ways, which leads to duplication. It is ineffective to manually detect and handle duplicate questions. To address this issue, an automatic system for detecting duplicate questions is essential. In addition to detecting duplicate questions using a pre-trained Sentence-BERT model (Galli et al. in Information 15(2):68, 2024) and cosine similarity, this system can also summarize answers. To maintain diversity in the summary, we apply K-Means clustering on the embeddings of matching answers. A representative answer from each cluster is selected to form a summary, which is further refined by integrating important sentences from TextRank and selected answers based on the BM25 ranking and the final summary is presented to the user (Liu et al. in Knowl-Based Syst 287:111447, 2024).

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

Leveraging ML to Measure Similarity Between Questions and Answers

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Vasavi Ravuri,
  • Jayasree Gondipalle,
  • Shalini Muskula,
  • Sai Jahnavi Rallapalli,
  • Kondisetty Venkata Jayasree

摘要

Platforms like Quora, Reddit, and Stack Overflow are widely used by users to share and respond to a large volume of questions and answers. Though these platforms provide user-generated content, they struggle with repetitive or similar queries. The same question is frequently asked by several users in different ways, which leads to duplication. It is ineffective to manually detect and handle duplicate questions. To address this issue, an automatic system for detecting duplicate questions is essential. In addition to detecting duplicate questions using a pre-trained Sentence-BERT model (Galli et al. in Information 15(2):68, 2024) and cosine similarity, this system can also summarize answers. To maintain diversity in the summary, we apply K-Means clustering on the embeddings of matching answers. A representative answer from each cluster is selected to form a summary, which is further refined by integrating important sentences from TextRank and selected answers based on the BM25 ranking and the final summary is presented to the user (Liu et al. in Knowl-Based Syst 287:111447, 2024).