| 王晓慧,覃京燕.基于深度循环神经网络的社交网络用户情感研究[J].包装工程,2021,42(4):77-82. |
| 基于深度循环神经网络的社交网络用户情感研究 |
| Affective Modeling of Social Networks Based on Deep Recurrent Neural Networks |
| 投稿时间：2020-11-20 |
| DOI：10.19554/j.cnki.1001-3563.2021.04.009 |
| 中文关键词: 社交网络 情感计算 循环神经网络 深度学习 推荐系统 |
| 英文关键词: social network affective computing recurrent neural networks deep learning recommendation system |
| 基金项目:中央高校基本科研业务费(FRF-TP-18-007A3，FRF-IDRY-19-030);佛山市人民政府科技创新专项资金项目(BK20AF002);佛山市促进高校科技成果服务产业发展扶持项目(2020DZXX05) |
|
| 摘要点击次数: |
| 全文下载次数: |
| 中文摘要: |
| 目的 将深度学习与社交网络、情感计算相结合，探索利用深度神经网络进行社交网络用户情感研究的新方法和新技术，探索模型在用户需求分析和推荐上的应用。方法 自动筛选和挖掘海量社交网络数据，研究具有长时记忆的非先验情感预测方法，对网络中海量的用户数据、人与人之间关系进行建模，为关联时间序列创建LSTM模型，并结合其相互关系融入统一的大型深度循环网络中。具体包括:基于注意力模型的社交网络异构数据处理;基于深度LSTM的长时记忆建模，研究子网络选取、深度LSTM设计，以及针对社交网络的大型网络结构设计;基于社交网络情感模型和强化学习的推荐算法。结果 提高了分析的准确度，降低了对先验假设的依赖，减轻了人工情感模型的工作量和偏差，增强了对不同网络数据的普适性;供深度模型使用。结论 研究成果促进了深度学习与情感计算的结合，可推动网络用户行为分析和预测的研究，可用于个性化推荐、定向广告等领域，具有广泛的学术意义和应用前景。 |
| 英文摘要: |
| The work aims to combine deep learning with social networks and affective computing and explore new methods and new technologies to establish the affective model of social networks by deep neural networks, to explore the application of the model in user demand analysis and recommendation. The automatic filtering and mining of social network data were carried out. Affective prediction technique with long-term memory and no prior knowledge was studied to model vast amounts of user data and interpersonal relationship data, to establish the LSTM model for time series and integrate them into a uniform large deep recurrent network in combination with their interrelation. The main contents included:heterogeneous data processing of social networks based on attention model; long-term memory model based on deep LSTM, which studied the sub-network selection, deep LSTM structure and large network structure of social networks in allusion to social networks; recommendation algorithm based on the established affective model and reinforcement learning. This research reduced dependence on priori assumptions, improved the analysis accuracy, lightened the workload and bias of the artificial affective model and enhanced the universality of all kinds of different network data. It can be used for deep model. The research results contribute the combination of deep learning and affective computing and promote the research on user behavior analysis and prediction, can be used in personalized recommendation and targeted advertising. It has the wide academic meaning and application prospect. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
