随着个性化服务需求的增长,通过分析人类行为模式来预测人们的生活习惯、需求和偏好变得尤为重要。本文探讨了基于生活日志数据进行用户行为分类预测的重要性,提出一种融合卷积神经网络(CNN)和双向长短期记忆网络(Bi LSTM)的用户行为预...随着个性化服务需求的增长,通过分析人类行为模式来预测人们的生活习惯、需求和偏好变得尤为重要。本文探讨了基于生活日志数据进行用户行为分类预测的重要性,提出一种融合卷积神经网络(CNN)和双向长短期记忆网络(Bi LSTM)的用户行为预测模型TS-CNN-Bi LSTM。使用Liu-Life Log收集十二年的用户生活日志数据作为样本数据。考虑生活日志数据存在不公开位置信息的用户,基于同一数据集构建了两种不同特征组合的数据集DCP-Life Log和TS-Life Log。使用不同数据集评估模型预测性能。实验结果表明,所提模型在两个数据集上的宏平均精确率、召回率和F1值均优于基准模型,分别提升2.23%、1.30%和1.27%。With the growing demand for personalized services, it has become essential to predict people’s habits, needs, and preferences by analyzing human behavior patterns. In this paper, we explore the importance of classifying and predicting user behavior based on life log data and propose a user behavior prediction model TS-CNN-Bi LSTM that integrates a convolutional neural network (CNN) and bidirectional long short-term memory network (Bi LSTM). Twelve years of user life log data collected by Liu-Life Log is used as sample data. Considering the existence of users who do not disclose their location information in the life log data, two datasets with different feature combinations, DCP-Life Log and TS-Life Log, are constructed based on the same dataset. The model prediction performance is evaluated using different datasets. The experimental results show that the proposed model outperforms the benchmark model on both datasets regarding the macro-mean precision, recall, and F1 value, with an improvement of 2.23%, 1.30%, and 1.27%, respectively.展开更多
为实现对灰度不均匀医学图像分割的同时进行有偏场估计并校正,改进了基于局部高斯分布拟合(Local Gaussian Distribution Fitting,LGDF)能量的活动轮廓模型。通过分析图像有偏场模型的局部特性,将有偏场乘性因子引入图像局部灰度均值的...为实现对灰度不均匀医学图像分割的同时进行有偏场估计并校正,改进了基于局部高斯分布拟合(Local Gaussian Distribution Fitting,LGDF)能量的活动轮廓模型。通过分析图像有偏场模型的局部特性,将有偏场乘性因子引入图像局部灰度均值的表达中,从而使有偏场乘性因子成为新的能量函数的变量。能量函数的迭代最小化既实现了目标组织分割,又有效估计了有偏场。合成图像和真实医学图像实验表明该方法比现有多种方法分割性能更好,且利用估计的有偏场校正后的图像具有更好的视觉效果。展开更多
文摘随着个性化服务需求的增长,通过分析人类行为模式来预测人们的生活习惯、需求和偏好变得尤为重要。本文探讨了基于生活日志数据进行用户行为分类预测的重要性,提出一种融合卷积神经网络(CNN)和双向长短期记忆网络(Bi LSTM)的用户行为预测模型TS-CNN-Bi LSTM。使用Liu-Life Log收集十二年的用户生活日志数据作为样本数据。考虑生活日志数据存在不公开位置信息的用户,基于同一数据集构建了两种不同特征组合的数据集DCP-Life Log和TS-Life Log。使用不同数据集评估模型预测性能。实验结果表明,所提模型在两个数据集上的宏平均精确率、召回率和F1值均优于基准模型,分别提升2.23%、1.30%和1.27%。With the growing demand for personalized services, it has become essential to predict people’s habits, needs, and preferences by analyzing human behavior patterns. In this paper, we explore the importance of classifying and predicting user behavior based on life log data and propose a user behavior prediction model TS-CNN-Bi LSTM that integrates a convolutional neural network (CNN) and bidirectional long short-term memory network (Bi LSTM). Twelve years of user life log data collected by Liu-Life Log is used as sample data. Considering the existence of users who do not disclose their location information in the life log data, two datasets with different feature combinations, DCP-Life Log and TS-Life Log, are constructed based on the same dataset. The model prediction performance is evaluated using different datasets. The experimental results show that the proposed model outperforms the benchmark model on both datasets regarding the macro-mean precision, recall, and F1 value, with an improvement of 2.23%, 1.30%, and 1.27%, respectively.
文摘为实现对灰度不均匀医学图像分割的同时进行有偏场估计并校正,改进了基于局部高斯分布拟合(Local Gaussian Distribution Fitting,LGDF)能量的活动轮廓模型。通过分析图像有偏场模型的局部特性,将有偏场乘性因子引入图像局部灰度均值的表达中,从而使有偏场乘性因子成为新的能量函数的变量。能量函数的迭代最小化既实现了目标组织分割,又有效估计了有偏场。合成图像和真实医学图像实验表明该方法比现有多种方法分割性能更好,且利用估计的有偏场校正后的图像具有更好的视觉效果。
