arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2006.01222cs.CLcs.SI
收藏

基于BERT的会话式社交媒体文本信息披露和支持的集成建模

BERT-based Ensembles for Modeling Disclosure and Support in Conversational Social Media Text

Netaji Subhas Institute of Technology(内塔吉·苏巴斯理工学院);International Institute of Information Technology, Hyderabad(海得拉巴国际信息技术学院)

Tanvi Dadu, Kartikey Pant, Radhika Mamidi

摘要:

人们越来越有兴趣了解人类如何发起和进行对话。会话的情感理解关注的是说话者如何利用情感对情景和彼此做出反应的问题。在CL-Aff共享任务中,组织者发布了Get it #OffMyChest数据集,其中包含来自休闲和忏悔对话的Reddit评论,标记为披露和保密特征。在本文中,我们介绍了一个预测集成模型,利用微调的上下文词嵌入,罗伯特和阿尔伯特。我们表明,我们的模型在所有考虑的指标优于基础模型,实现了$3\%$的F1分数的改善。我们进一步进行统计分析,并概述对给定数据集的更深入见解,同时为数据集提供新的影响特征。

英文摘要:

There is a growing interest in understanding how humans initiate and hold conversations. The affective understanding of conversations focuses on the problem of how speakers use emotions to react to a situation and to each other. In the CL-Aff Shared Task, the organizers released Get it #OffMyChest dataset, which contains Reddit comments from casual and confessional conversations, labeled for their disclosure and supportiveness characteristics. In this paper, we introduce a predictive ensemble model exploiting the finetuned contextualized word embeddings, RoBERTa and ALBERT. We show that our model outperforms the base models in all considered metrics, achieving an improvement of $3\%$ in the F1 score. We further conduct statistical analysis and outline deeper insights into the given dataset while providing a new characterization of impact for the dataset.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑