基于大语言模型智能体的跨域立场检测

Cross-Domain Stance Detection Based on Large Language Model Agent

随着网民参与互联网话题讨论规模的不断增大,针对跨域的文本立场检测需求日渐迫切。在跨域立场检测过程中,基于传统机器学习或单一智能体的立场检测模型通常存在因标注数据少、涉及话题领域多、文本复杂等而导致准确率低的问题,因此设计了一种基于大语言模型多智能体协同的立场检测模型。采用零样本学习策略,通过多个语义感知智能体与对抗智能体的协作对文本进行多维分析和辩证分析,能够理解和推理文本表达的立场和原因。通过实验对比,该方法提升了跨领域文本立场分析的准确性,同时具有良好的泛化性。

With the increasing number of netizens participating in the discussion of Internet topics,the need for cross-domain text stance detection is becoming more and more urgent.In the process of crossdomain stance detection,stance detection models based on conventional machine learning or single agents usually suffer from low accuracy due to the lack of labeled data,the involvement of many topic domains,and the complexity of text semantics,etc.Therefore,a multi-agent collaborative stance detection model based on large language model is designed.Zero-shot learning strategy is employed to carry out multidimensional analysis and dialectical analysis of text through the cooperation of multiple semantic sensing agents and adversarial agents,which can understand and reason about the stance expressed in the text.Experimental comparisons indicate that the method improves the accuracy of cross-domain text stance analysis,and it also has good generalizability.