Scholarlink

<p>Early identification of psychometric conditions facilitate better support to academics hence to offer precise multi-modal psychometric analysis this research gathered student data in the form of both texts and audio. For textual data analysis, an enhanced Word2vec model is used to represent text data into fixed vector representation. Further, a novel hybridized Circular Position Embedding enabled Large Language Model Meta Artificial Intelligence (LLaMA-2) 13B with Low-Rank Adaptation (LoRA) model (CiPE-LoRA) is proposed for text classification. This hybridized approach fine-tune small subset of parameters and reducing computational costs for maintaining the better performance during the execution. For audio data analysis, 2D Mel Spectrogram method is developed that effectively extract the features, while Gated Residual Autoencoder (GRAE) is utilized to find the psychometric level. Experimental findings demonstrated that, proposed CiPE-LoRA and GRAE approach attained better performance improvement in analyzing both text and audio data, with an accuracy of 98.95% and 97.92% respectively. These outcomes proved the efficiency of the proposed models for analysing and classifying the psychometric level.</p>

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

Psychometric analysis using LLaMA based circular position embedding and gated residual auto-encoder

  • 期刊论文

  • 发表日期：

    2026年1月10日

查看全文
  • Shruti Agrawal,
  • Rajiv Iyer

摘要

Early identification of psychometric conditions facilitate better support to academics hence to offer precise multi-modal psychometric analysis this research gathered student data in the form of both texts and audio. For textual data analysis, an enhanced Word2vec model is used to represent text data into fixed vector representation. Further, a novel hybridized Circular Position Embedding enabled Large Language Model Meta Artificial Intelligence (LLaMA-2) 13B with Low-Rank Adaptation (LoRA) model (CiPE-LoRA) is proposed for text classification. This hybridized approach fine-tune small subset of parameters and reducing computational costs for maintaining the better performance during the execution. For audio data analysis, 2D Mel Spectrogram method is developed that effectively extract the features, while Gated Residual Autoencoder (GRAE) is utilized to find the psychometric level. Experimental findings demonstrated that, proposed CiPE-LoRA and GRAE approach attained better performance improvement in analyzing both text and audio data, with an accuracy of 98.95% and 97.92% respectively. These outcomes proved the efficiency of the proposed models for analysing and classifying the psychometric level.