基于深度学习的资源损毁智能推荐算法

Intelligent Recommendation Algorithm for Resource Damage Based on Deep Learning

战场态势经常变化,资源损毁故障时有发生,需要提供替换建议。然而人工筛选效率低下,基于规则筛选制定规则难,规则繁杂,且样本扰动对规则影响大,针对上述需求和不足,本文基于深度学习算法构造了一套资源替换推荐算法。算法采用BERT和多层感知机的组合作为编码器,分别从实体名称、邻接关系、属性集合三个大方面进行了相似度的计算,最后使用多层感知机生成资源替换评分。在构造的战场资源知识图谱上的实验显示,所构造的算法能够实现93%的TOP1准确率以及99%的TOP10准确率,推理时间在毫秒级,满足了战场态势多变的情况下对资源推荐算法的准确率与运算时间的要求。

The battlefield situation often changes,and resource damage and faults occur frequently,requiring replacement suggestions.However,manual screening is inefficient,making it difficult to develop rules based on rule screening.The rules are complex,and sample perturbations have a significant impact on the rules.In response to these needs and shortcomings,this paper constructs a set of resource replacement recommendation algorithms based on deep learning algorithms.The algorithm uses a combination of BERT and multi-layer perceptron as the encoder,and calculates similarity from three major aspects:entity name,adjacency relationship,and attribute set.Finally,a multi-layer perceptron is used to generate resource replacement scores.The experiment on the constructed battlefield resource knowledge graph shows that the constructed algorithm can achieve a TOP1 accuracy of 93%and a TOP10 accuracy of 99%,with inference time in milliseconds,meeting the requirements for accuracy and computation time of resource recommendation algorithms in the context of changing battlefield situations.