基于改进机器学习的定向运动推荐及系统构建

Directional motion recommendation and system construction based on improved machine learning

为更好地刻画用户向量和活动向量的映射关系,进而提高城市穿越与定向运动推荐的准确率,提出一种基于DSSM-FM的个性化推荐模型。其中,针对传统One-Hot编码存在高维、稀疏和向量间关联性差的问题,运用FM算法替代嵌入层,以此更好地表征特征之间的联系;引入DSSM(Deep Structured Semantic Models,深度语义匹配模型),在FM预训练的基础上,更深入地挖掘特征间的高阶关系,从而更好地挖掘向量隐藏特征。结果表明,与其他算法对比,本研究提出的推荐方案的准确率达89.62%;同时系统应用表明,构建的推荐模型可实现定向运动的推荐。由此得出,本方案可行,可用于定向运动的推荐中。

To better characterize the mapping relationship between user vectors and activity vectors,and improve the accuracy of urban travel and directional motion recommendations,a personalized recommendation model based on DSSM-FM is proposed.Among them,in response to the problems of high dimension,sparsity,and poor correlation between vectors in traditional One-Hot encoding,FM algorithm is used to replace the embedding layer to better represent the connections between features;Based on FM pre-training,DSSM(Deep Structured Semantic Models)is introduced to dig deeper into the higher-order relationships between features,so as to better mine the hidden features of vectors.The results show that compared with other algorithms,the accuracy of the recommendation scheme proposed in this study reaches 89.62%;At the same time,the system application indicates that the recommendation model constructed in this study can achieve directional motion recommendation.From this,it can be concluded that this scheme is feasible and can be used in the recommendation of directional motions.