基于特征选择和深度信念网络的文本情感分类算法

由于人类语言的复杂性，文本情感分类算法大多都存在因为冗余而造成的词汇量过大的问题。深度信念网络（DBN）通过学习输入语料中的有用信息以及它的几个隐藏层来解决这个问题。然而对于大型应用程序来说，DBN是一个耗时且计算代价昂贵的算法。针对这个问题，提出了一种半监督的情感分类算法，即基于特征选择和深度信念网络的文本情感分类算法（FSDBN）。首先使用特征选择方法（文档频率（DF）、信息增益（IG）、卡方统计（CHI）、互信息（MI））过滤掉一些不相关的特征从而使词汇表的复杂性降低；然后将特征选择的结果输入到DBN中，使得DBN的学习阶段更加高效。将所提算法应用到中文以及维吾尔语中，实验结果表明在酒店评论数据集上，FSDBN在准确率方面比DBN提高了1.6%，在训练时间上比DBN缩短一半。     

Novel data structure and visualization tool for studying technology evolution based on patent information: The DTFootprint and the TechSpectrogram

We introduce a new, bespoke data structure to analyze and visualize the evolution of a technology. The technology under analysis is defined by a set of patents corresponding to a technical field, owned by a company or invented by a team of research. Our data structure, the Dynamic Technology Footprint –DTFootprint–, facilitates the analysis and visualization of trends and dynamics of a given technology, and therefore the evolution of a technical field, a company, or a team of people. A graphical tool based on our data structure is defined, it is named Technology Spectrogram –TechSpectrogram–, because it is inspired by the acoustic frequency spectrograms: as the acoustic frequency spectrograms visualize the dynamics of an acoustic wave showing the evolution of its frequency components our tool shows the dynamics of a technology showing the evolution of its technological components, which are represented by the whole set of IPC-codes. Our graphical tool, the TechSpectrogram is shown for some study cases, and its application to the history of technology and technology management are disclosed.     

结合句法依存信息的方面级情感分类

引入句法依存信息到原方面术语,提出一种新的方面术语表示方法,利用Glove词向量表示单词以及单词与单词之间的依存关系,构造出包含句法依存信息的依存关系邻接矩阵和依存关系表示矩阵,利用图卷积神经网络和多头注意力机制将句法依存信息融入到方面术语中,使得方面术语表达与上下文结构高度相关。将改进后的方面词术语表示替换到现有模型后,模型泛化能力得到有效提升。对比试验和分析结果表明:该方法具有有效性和泛化性。     

明代盔甲种类及制作工艺研究

本文通过盔甲的制作及材料,区分了非金属材料的皮甲、瓠子作甲、藤甲、绵甲、纸甲以及金属材料组成的齐腰甲、齐腰明甲、缉甲、布甲、铁甲、钢甲、铜甲、锡甲等类型,并对其制作方式等进行分析。结合不同材质盔甲的使用地区、战争方式进行研究,以对明代盔甲有一个整体性的认知。     

An E-Business Event Stream Mechanism for Improving User Tracing Processes

With the rapid development in business transactions, especially in recent years, it has become necessary to develop different mechanisms to trace business user records in web server log in an efficient way. Online business transactions have increased, especially when the user or customer cannot obtain the required service. For example, with the spread of the epidemic Coronavirus (COVID-19) throughout the world, there is a dire need to rely more on online business processes. In order to improve the efficiency and performance of E-business structure, a web server log must be well utilized to have the ability to trace and record infinite user transactions. This paper proposes an event stream mechanism based on formula patterns to enhance business processes and record all user activities in a structured log file. Each user activity is recorded with a set of tracing parameters that can predict the behavior of the user in business operations. The experimental results are conducted by applying clustering-based classification algorithms on two different datasets; namely, Online Shoppers Purchasing Intention and Instacart Market Basket Analysis. The clustering process is used to group related objects into the same cluster, then the classification process measures the predicted classes of clustered objects. The experimental results record provable accuracy in predicting user preferences on both datasets.     

Deep learning based pipeline with multichannel inputs for patent classification

This work introduces a deep learning pipeline for automatic patent classification with multichannel inputs based on LSTM and word vector embeddings. Sophisticated text mining methods are used to extract the most important segments from patent texts, and a domain-specific pre-trained word embeddings model for the patent domain is developed; it was trained on a very large dataset of more than five million patents. The deep learning pipeline is using multiple parallel LSTM networks that read the source patent document using different input dimensions namely embeddings of different segments of patent texts, and sparse linear input of different metadata. Classifying patents into corresponding technical fields is selected as a use case. In this use case, a series of patent classification experiments are conducted on different patent datasets, and the experimental results indicate that using the segments of patent texts as well as the metadata as multichannel inputs for a deep neural network model, achieves better performance than one input channel.