文献阅读记录(八)2023-07-24

1.张量神经网络模型

张量神经网络模型(neural tensor network,NTN)的基本思想是,用双线性张量取代传统神经网络中的线性变换层,在不同的维度下将头、尾实体向量联系起来。

NTN为每个三元组(h,r,t)定义了如下评分函数,评价2个实体之间存在某个特定关系r的可能性:

 其中urT是一个与关系相关的线性层,g()是tanh函数,Mr∈Rd×d×k是一个三阶张量,Mr,1,Mr,2∈Rd×k是与关系r有关的投影矩阵。

NTN中的实体向量是该实体中所有单词向量的平均值,好处是,实体中的单词数量远小于实体数量,可以充分重复利用单词向量构建实体表示,降低实体表示学习的稀疏性问题,增强不同实体的语义联系。

虽然NTN引入了张量操作,能更精确地刻画实体和关系的复杂语义联系。但计算复杂度非常高,需要大量三元组样例才能得到充分学习。

2.矩阵分解模型

矩阵分解是得到低维向量表示的重要途径,代表方法是RESACL模型。

在该模型中,知识库三元组构成一个大的张量X,如果三元组(h,r,t)存在,则Xhrt=1,否则为0。张量分解旨在将每个三元组(h,r,t)对应的张量值Xhrt分解为实体和关系表示,使得Xhrt尽量地接近于lhMrlt。

它的思想和前面看到的LFM类似,不同的是RESACL会优化张量中的所有位置,包括值为0的位置,而LFM只会优化知识库中存在的三元组。

3.翻译模型Trans

通过word2vec词表示学习模型和工具包,受到启发,发现词向量空间存在有趣的平移不变现象。通过类比实验(C(king)-C(queen)≈C(man)-C(woman)),发现这种平移不变现象普遍存在于词汇的语义关系和句法关系中。

受此启发,提出了TransE模型。将知识库中的关系看作实体间的某种平移向量.对于每个三元组(h,r,t),TransE用关系r的向量lr作为头实体向量lh和尾实体向量lt之间的平移。也可以将lr看作从lh到lt的翻译,因此TransE也被称为翻译模型。

4.其他模型

在TransE提出之后,大部分知识表示学习模型是以TransE为基础的扩展。在TransE扩展模型以外,还有很多,例如全息表示模型(holographic embeddings,Hole)。

Hole提出使用头、尾实体向量的“循环相关”操作来表示该实体对。循环相关操作可以看作张量乘法特殊形式,具有较强的表达能力。优点如下:1)不可交换性.循环相关是不可交换的,即lh*lt≠lt*lh.而知识库中很多关系是不可交换的,因此该特点具有重要意义。2)相关性.循环相关操作得到的向量每一维都衡量了向量lh和lt的某种相似性。

  • 0
    点赞
  • 0
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
以下是一篇关于超市管理系统的外文文献,供您参考: Title: Design and Implementation of Supermarket Management System Based on Android Author: Zhang, Y. & Liu, Y. Journal: Journal of Physics: Conference Series Year: 2021 Abstract: With the development of mobile devices and the popularity of Android operating system, more and more businesses are choosing to develop management systems based on Android. In this paper, we present the design and implementation of a supermarket management system based on Android platform. The system is designed to provide a convenient and efficient way for supermarket managers to manage their inventory, sales, and customer data. The system consists of two main parts: the server-side software and the Android client. The server-side software is responsible for managing the database, processing requests from clients, and providing the necessary data to the clients. The Android client provides the user interface for the supermarket managers to interact with the system. The client can be installed on any Android device, such as a smartphone or a tablet. The system provides a range of features, including inventory management, sales management, customer management, and reporting. The inventory management module allows the managers to add, modify, and delete products in the inventory. They can also view the current stock level and set up alerts to notify them when the stock level is low. The sales management module allows the managers to record sales transactions, view sales reports, and analyze sales trends. The customer management module allows the managers to store customer information, such as name, address, and phone number. They can also view the purchase history of each customer. The reporting module provides various reports, such as sales report, inventory report, and customer report. The system has been implemented and tested in a real-world supermarket. The results show that the system is reliable, efficient, and easy to use. The managers are able to manage their supermarket more effectively and make better decisions based on the data provided by the system. In conclusion, the supermarket management system based on Android platform is a promising solution for businesses that need to manage their inventory, sales, and customer data. The system provides a convenient and efficient way for managers to access the data they need and make informed decisions. With the continued development of mobile devices and the Android operating system, we expect to see more businesses adopting this technology in the future.

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值