Scholarlink

<p>This study proposes a modeling and feedback mechanism oriented to user behavior analysis to enhance the design optimization capability of cultural and creative products in terms of interactive experience. The study aims to construct an intelligent design decision-making system that can be used for interface iteration and experience improvement. First, a set of non-intrusive multimodal perception mechanisms is designed to synchronously collect user behaviors in the cultural and creative platform, such as mouse movement, click stream, eye movement path, and voice command. Second, the Transformer embedding layer is used to perform high-dimensional semantic compression on multimodal data. Concurrently, a long short-term memory (LSTM) model integrating the attention mechanism is constructed to complete user satisfaction regression and multi-label classification of potential design pain points. Finally, an interaction optimization feedback mechanism based on model output is built to realize abnormal behavior localization, interface heat map analysis, and automatic recommendation of personalized design suggestions. Experimental results demonstrate that the proposed Transformer-enhanced LSTM (T-LSTM) model achieves superior performance across all evaluation indicators. In the satisfaction prediction task, this model obtains a mean absolute error of 0.219 and root-mean-square error of 0.315. For pain point identification, the model reaches an F1-score of 0.889 and an area under the curve of 0.942. The comprehensive evaluation shows consistent outperformance over five benchmark models: gated recurrent unit, bidirectional long short-term memory, Informer, time-series transformer, and multilayer perceptron. Meanwhile, after interface optimization, the average user task completion time is shortened by 16.2 s, the number of clicks is reduced by 7.3 times, and the subjective satisfaction score is increased by 0.82 points. All of these are verified by t test significance analysis (<i>p</i> &lt; 0.01). The research conclusion indicates that the model has strong behavior recognition ability and design feedback value and can support intelligent interface optimization practices for complex interactive scenarios.</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

Design and improvement of cultural and creative products integrating the long short-term memory algorithm for human–computer interaction behavior analysis

  • 期刊论文

  • 发表日期：

    2026年1月7日

查看全文
  • Yanyan Xu,
  • Honghe Gao,
  • Yi Chen

摘要

This study proposes a modeling and feedback mechanism oriented to user behavior analysis to enhance the design optimization capability of cultural and creative products in terms of interactive experience. The study aims to construct an intelligent design decision-making system that can be used for interface iteration and experience improvement. First, a set of non-intrusive multimodal perception mechanisms is designed to synchronously collect user behaviors in the cultural and creative platform, such as mouse movement, click stream, eye movement path, and voice command. Second, the Transformer embedding layer is used to perform high-dimensional semantic compression on multimodal data. Concurrently, a long short-term memory (LSTM) model integrating the attention mechanism is constructed to complete user satisfaction regression and multi-label classification of potential design pain points. Finally, an interaction optimization feedback mechanism based on model output is built to realize abnormal behavior localization, interface heat map analysis, and automatic recommendation of personalized design suggestions. Experimental results demonstrate that the proposed Transformer-enhanced LSTM (T-LSTM) model achieves superior performance across all evaluation indicators. In the satisfaction prediction task, this model obtains a mean absolute error of 0.219 and root-mean-square error of 0.315. For pain point identification, the model reaches an F1-score of 0.889 and an area under the curve of 0.942. The comprehensive evaluation shows consistent outperformance over five benchmark models: gated recurrent unit, bidirectional long short-term memory, Informer, time-series transformer, and multilayer perceptron. Meanwhile, after interface optimization, the average user task completion time is shortened by 16.2 s, the number of clicks is reduced by 7.3 times, and the subjective satisfaction score is increased by 0.82 points. All of these are verified by t test significance analysis (p < 0.01). The research conclusion indicates that the model has strong behavior recognition ability and design feedback value and can support intelligent interface optimization practices for complex interactive scenarios.