Scholarlink

Now-days social media became popular source of communication. An event is an incident occurred in real-world. The Social media may be used to sense around the real world events as most of them are communicated on this platform. The sensed data are further used to detect significant events. This detected information will help the people and organization to act accordingly. Some events like natural disasters, accidents etc. require detection in real time as well as prompt action. In this work, we applied Distill-BERT model for event detection. Distill-BERT model is 60% faster and 40% smaller in size as compared to BERT-base model. This makes the event detection faster which is desired in real time processing. Also accuracy is not compromised as compared to base model (only ~3% compromised). This compromised is covered by applying N-gram model that improvise the system performance. The experimental results shows that F1-measure for BERT-base, Distill-BERT are 0.75 and 0.72 respectively. The proposed method improves the F1-measure to the value 0.76. Hence proposed approach work better for event detection especially in real time scenario.

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

Event Detection and Analysis from Social Media Data Using N-gram and Distil-BERT Model

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Dharmendra Mangal,
  • Hemant Makwana

摘要

Now-days social media became popular source of communication. An event is an incident occurred in real-world. The Social media may be used to sense around the real world events as most of them are communicated on this platform. The sensed data are further used to detect significant events. This detected information will help the people and organization to act accordingly. Some events like natural disasters, accidents etc. require detection in real time as well as prompt action. In this work, we applied Distill-BERT model for event detection. Distill-BERT model is 60% faster and 40% smaller in size as compared to BERT-base model. This makes the event detection faster which is desired in real time processing. Also accuracy is not compromised as compared to base model (only ~3% compromised). This compromised is covered by applying N-gram model that improvise the system performance. The experimental results shows that F1-measure for BERT-base, Distill-BERT are 0.75 and 0.72 respectively. The proposed method improves the F1-measure to the value 0.76. Hence proposed approach work better for event detection especially in real time scenario.