基于MS-Cluster与Prompt-Learning话题检测与追踪技术  

Topic Detection and Tracking Technology Based on MS-Cluster and Prompt-Learning

作　　者：李崭 杜晓童 黄浩 任秋霖 

机构地区：[1]中国电子科技集团公司第十研究所,四川 成都

出　　处：《计算机科学与应用》2023年第10期1918-1927,共10页Computer Science and Application

摘　　要：话题检测与追踪技术随着信息处理技术以及人工智能技术的发展,已经取得了较好的发展,但在实际应用中,由于算法标注数据需求高、训练代价大,很难较好的落地应用。本文提出了基于MS-Cluster与Prompt-Learning的话题检测追踪技术,通过聚类分析过程初步进行话题聚合,在此基础上通过提示学习推理进行话题补偿,完成话题检测与追踪过程。该方法在包含13个话题的测试数据集上进行测试验证,证明该方法在零样本与低样本标注情况下有较好效果,同时相较于其他主流话题检测追踪技术在准确率与召回率上都有提升。Topic detection and tracking technology has been developing well with the development of information processing technology and artificial intelligence technology. However, in practical applications, it is difficult to achieve good deployment due to the high demand for algorithm annotated data and the large training cost. This article proposes a topic detection and tracking technology based on MS-Cluster and Prompt-Learning. The method performs topic aggregation through clustering analysis and topic supplementation through prompt learning reasoning to complete the topic de-tection and tracking process. The method was tested on a dataset of 13 topics, and it showed good results in the case of zero-shot learning and few-shot learning, and it outperformed other main-stream topic detection and tracking technologies in terms of accuracy and recall rate.

关 键 词：话题检测追踪技术 提示学习 小样本学习 聚类分析 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]