微型电脑应用

用户画像特征的提取主要是对特征词汇进行识别。由于特征词汇存在多重语义情况，容易出现特征提取结果偏差，导致用户画像特征提取结果的查准率不高。因此，提出一种基于自回归预训练语言模型的用户画像特征提取方法。选取基本特征，计算用户画像的显式特征，并进行聚类占比分析。根据关系矩阵挖掘隐式特征，完善用户画像特征。构建预训练语言模型，生成关键词预训练任务，添加自回归训练进行关键词的放缩训练，完成多重语义分类识别训练。标准化处理特征关键词，可视化处理提取结果，即可完成用户画像特征的提取。通过实验对4895名用户进行不同类型特征关键词的提取，提取结果的查准率显著提高，最高值达到0.88,平均值达到0.83,验证了所提出的方法的查准率符合实践要求。

The user profile feature extraction mainly involves the recognition of feature words. Due to the multiple semantic situations of feature words, it is easy to encounter the problem of feature extraction result deviation, resulting in a low accuracy rate of user profile feature extraction results. Therefore, a user profile feature extraction method based on autoregressive pre-trained language model is proposed. Basic feature is selected, explicit feature of user profile is calculated, and clustering proportion analysis is performed. Implicit feature based on the relationship matrix is mined to improve user profile feature. A pre-trained language model is built, keyword pre-training tasks are generated, autoregressive training for keyword scaling training is added, and multiple semantic classification recognition training is completed. By standardizing the processing of feature keywords and visualizing the extraction results, the extraction of user profile feature can be completed. Through experiments, different types of feature keywords are extracted from 4895 users. The accuracy rate of the extraction results is significantly improved, with a maximum value of 0.88 and an average value of 0.83. This verifies that the accuracy rate of the proposed method meets the practical requirements.