PosiGPT:基于预训练的中文积极情感评论模型  

POSIGPT:A CHINESE POSITIVE EMOTIONAL TEXT GENERATION MODEL BASED ON PRE-TRAINING METHOD

作　　者：卢晨耀 李敏波[1,2] Lu Chenyao;Li Minbo(School of Software,Fudan University,Shanghai 200438,China;Shanghai Key Laboratory of Data Science,Shanghai 200438,China)

机构地区：[1]复旦大学软件学院,上海200438 [2]上海市数据科学重点实验室,上海200438

出　　处：《计算机应用与软件》2024年第9期33-40,共8页Computer Applications and Software

基　　金：国家自然科学基金项目(61671157)。

摘　　要：近年来,越来越多的人因为工作和生活的压力处于抑郁状态,许多人没有得到足够的鼓励和正确的指导。基于这个背景,提出中文评论模型PosiGPT。基于生成式预训练模型,使用中文微博数据进行训练,可以检测抑郁状态的博文,并产生积极回复。除了该模型外,还发布了PosiChat数据集,其包含源自新浪微博的抑郁文本及其积极评论。在PosiChat数据集上进行评估,结果表明模型生成的文本具有较强的流畅性和合理性,且在情感倾向上属于积极情感状态,初步达到了抑郁检测及积极回复的功能。In recent years,more and more people have been depressed due to the pressure of work and life,and many people have not received enough encouragement and correct guidance in time.Based on this background,this paper proposes a Chinese comment model named PosiGPT.PosiGPT used a generative pre-trained model and used Chinese Weibo data for training.It could detect blogs in depression and generate positive responses.In addition to this model,this paper released a dataset named PosiChat,which contained depression texts and their comments from Sina Weibo.PosiGPT was tested on PosiChat dataset.The results show that the comments generated by PosiGPT have strong fluency,rationality and are always positive,achieving the goal of depression detection and positive response.

关 键 词：文本生成 抑郁检测 积极情感评论 预训练 

分 类 号：TP391.43[自动化与计算机技术—计算机应用技术]