基于注意力机制的深度学习态势信息推荐模型

Attention-Based Deep Learning Situation Information Recommendation Model

随着信息技术的迅猛发展,战场态势数据呈现出体量大、类型多、增长快、价值密度低等“4v”特点,能够增强态势感知的同时,大量冗余数据也会严重干扰指挥员对有用信息的提取和有效利用,影响指挥员快速、准确地决策。针对态势信息给指挥员带来的信息过载问题,提出一种基于注意力机制的深度学习态势信息推荐模型,引入双层注意力机制,利用多层神经网络分别学习项目级和组件级的注意力权重,深究指挥员与态势信息之间的潜在关系,构建指挥员偏好预测模型,提高推荐的准确性。

With the rapid development of IT,battlefield situation data presents the characteristics of"4v":volume,variety,velocity and value.While it can enhance situational awareness,a large number of redundant data will seriously interfere with the commander’s extraction and effective use of useful information,and affect the commander’s rapid and accurate decision-making.To address the problem of information overload,we propose a model of deep learning situation information recommendation based on attention mechanism,which introduces a double-layer attention mechanism to assign item and components attentive weights respectively.We explore the latent relationship between commanders and situation information by using multi-layer neural network to build the commander preference prediction model and improve the accuracy of recommendation.