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。     

Cryptographic Lightweight Encryption Algorithm with Dimensionality Reduction in Edge Computing

Edge Computing is one of the radically evolving systems through generations as it is able to effectively meet the data saving standards of consumers, providers and the workers. Requisition for Edge Computing based items have been increasing tremendously. Apart from the advantages it holds, there remain lots of objections and restrictions, which hinders it from accomplishing the need of consumers all around the world. Some of the limitations are constraints on computing and hardware, functions and accessibility, remote administration and connectivity. There is also a backlog in security due to its inability to create a trust between devices involved in encryption and decryption. This is because security of data greatly depends upon faster encryption and decryption in order to transfer it. In addition, its devices are considerably exposed to side channel attacks, including Power Analysis attacks that are capable of overturning the process. Constrained space and the ability of it is one of the most challenging tasks. To prevail over from this issue we are proposing a Cryptographic Lightweight Encryption Algorithm with Dimensionality Reduction in Edge Computing. The t-Distributed Stochastic Neighbor Embedding is one of the efficient dimensionality reduction technique that greatly decreases the size of the non-linear data. The three dimensional image data obtained from the system, which are connected with it, are dimensionally reduced, and then lightweight encryption algorithm is employed. Hence, the security backlog can be solved effectively using this method.     

基于Word2Vec模型特征扩展的Web搜索结果聚类性能的改进

对于用户泛化和模糊的查询，将Web搜索引擎返回的列表内容聚类处理，便于用户有效查找感兴趣的内容。由于返回的列表由称为片段（snippet）的短文本组成，而传统的单词频率-逆文档频率（TF-IDF）特征选择模型不能适用于稀疏的短文本，使得聚类性能下降。一个有效的方法就是通过一个外部的知识库对短文本进行扩展。受到基于神经网络词表示方法的启发，提出了通过词嵌入技术的Word2Vec模型对短文本扩展，即采用Word2Vec模型的TopN个最相似的单词用于对片段（snippet）的扩展，扩展文档使得TF-IDF模型特征选择得到聚类性能的提高。同时考虑到通用性单词造成的噪声引入，对扩展文档的TF-IDF矩阵进行了词频权重修正。实验在两个公开数据集ODP239和SearchSnippets上完成，将所提方法和纯snippet无扩展的方法、基于Wordnet的特征扩展方法和基于Wikipedia的特征扩展方法进行了对比。实验结果表明，所提方法在聚类性能方面优于对比方法。     

一种基于小字典不对等语料的跨语言词嵌入方法

双语词嵌入通常采用从源语言空间到目标语言空间映射,通过源语言映射嵌入到目标语言空间的最小距离线性变换实现跨语言词嵌入。然而大型的平行语料难以获得,词嵌入的准确率难以提高。针对语料数量不对等、双语语料稀缺情况下的跨语言词嵌入问题,该文提出一种基于小字典不对等语料的跨语言词嵌入方法,首先对单语词向量进行归一化,对小字典词对正交最优线性变换求得梯度下降初始值,然后通过对大型源语言(英语)语料进行聚类,借助小字典找到与每一聚类簇相对应的源语言词,取聚类得到的每一簇词向量均值和源语言与目标语言对应的词向量均值,建立新的双语词向量对应关系,将新建立的双语词向量扩展到小字典中,使得小字典得以泛化和扩展。最后,利用泛化扩展后的字典对跨语言词嵌入映射模型进行梯度下降求得最优值。在英语—意大利语、德语和芬兰语上进行了实验验证,实验结果证明该文方法可以在跨语言词嵌入中减少梯度下降迭代次数,减少训练时间,同时在跨语言词嵌入上表现出较好的正确率。     

面向智能客服的句子相似度计算方法

针对金融领域中智能客服的句子相似度计算方法进行了研究。利用基于词性的分词纠正模型减少中文歧义词、金融相关词汇的分词错误；通过词向量方法和循环神经网络分别提取词语级和句子级的语义特征，并且得到句子向量；用融合层计算出句子向量间的差异特征；对差异特征进行降维和归一化得到句子相似度计算结果。实验结果表明，该方法具有较高的准确率和[F1]值。     

A proposed expert system for word sense disambiguation: deductive ambiguity resolution based on data mining and forward chaining

One of the major issues in the process of machine translation is the problem of choosing the proper translation for a multi‐sense word referred to as word sense disambiguation (WSD). Two commonly used approaches to this problem are statistical and example‐based methods. In statistical methods, ambiguity resolution is mostly carried out by making use of some statistics extracted from previously translated documents or dual corpora of source and target languages. Example‐based methods follow a similar approach as they also make use of bilingual corpora. However, they perform the task of matching at run‐time (i.e. online matching). In this paper, by looking at the WSD problem from a different viewpoint, we propose a system, which consists of two main parts. The first part includes a data mining algorithm, which runs offline and extracts some useful knowledge about the co‐occurrences of the words. In this algorithm, each sentence is imagined as a transaction in Market Basket Data Analysis problem, and the words included in a sentence play the role of purchased items. The second part of the system is an expert system whose knowledge base consists of the set of association rules generated by the first part. Moreover, in order to deduce the correct senses of the words, we introduce an efficient algorithm based on forward chaining in order to be used in the inference engine of the proposed expert system. The encouraging performance of the system in terms of precision and recall as well as its efficiency will be analysed and discussed through a set of experiments.     

基于特征融合的中文简历解析方法研究

针对基于规则和统计的传统中文简历解析方法效率低、成本高、泛化能力差的缺点，提出一种基于特征融合的中文简历解析方法，即级联Word2Vec生成的词向量和用BLSTM（Bidirectional Long Short-Term Memory）建模字序列生成的词向量，然后再结合BLSTM和CRF（Conditional Random Fields）对中文简历进行解析（BLSTM-CRF）。为了提高中文简历解析的效率，级联包含字序列信息的词向量和用Word2Vec生成的词向量，融合成一个新的词向量表示；再由BLSTM强大的学习能力融合词的上下文信息，输出所有可能标签序列的分值给CRF层；再由CRF引入标签之间约束关系求解最优序列。利用梯度下降算法训练神经网络，使用预先训练的词向量和Dropout优化神经网络，最终完成对中文简历的解析工作。实验结果表明，所提的特征融合方法优于传统的简历解析方法。     

融合情感与语义信息的情感分析方法

在使用词嵌入法进行词转向量时，两个反义词会转换成相近的向量。如果这两个词是情感词，将会导致词的情感信息的丢失，这在情感分析任务中是不合理的。为了解决这个问题，提出了一种在词嵌入的基础上增加情感向量来获取情感信息的方法。首先利用情感词典资源构建情感向量，将其与词嵌入法得到的词向量融合在一起；然后采用双向长短期记忆（BiLSTM）网络获取文本的特征；最后对文本的情感进行分类。在4个数据集上分别对该方法与未融合情感向量的方法进行了实验。实验结果表明所提方法分类准确度与F1值都高于未融合方法，说明了加入情感向量有助于提高情感分析的性能。     

DEM: Deep Entity Matching Across Heterogeneous Information Networks

Heterogeneous information networks, which consist of multi-typed vertices representing objects and multi-typed edges representing relations between objects, are ubiquitous in the real world. In this paper, we study the problem of entity matching for heterogeneous information networks based on distributed network embedding and multi-layer perceptron with a highway network, and we propose a new method named DEM short for Deep Entity Matching. In contrast to the traditional entity matching methods, DEM utilizes the multi-layer perceptron with a highway network to explore the hidden relations to improve the performance of matching. Importantly, we incorporate DEM with the network embedding methodology, enabling highly efficient computing in a vectorized manner. DEM’s generic modeling of both the network structure and the entity attributes enables it to model various heterogeneous information networks flexibly. To illustrate its functionality, we apply the DEM algorithm to two real-world entity matching applications: user linkage under the social network analysis scenario that predicts the same or matched users in different social platforms and record linkage that predicts the same or matched records in different citation networks. Extensive experiments on real-world datasets demonstrate DEM’s effectiveness and rationality.

     

改进的多层感知机在客户流失预测中的应用

针对在传统的客户流失预测数据预处理中，使用one-hot编码处理离散属性导致数据维度增加及数据过于稀疏的问题，提出了两种基于多层感知机的改进后的客户流失预测模型。其主要思想是分别使用堆叠自编码器和实体嵌入两种方法对多层感知机进行改进，通过将离散属性的高维编码数据向低维空间映射，有效地减少了one-hot编码产生的稀疏数据，增加了离散属性值之间的关联度。在对两份公开的数据集进行交叉验证后的实验结果表明，改进后的模型既有效地提高了预测的准确度，又维持了传统多层感知机模型在并行化计算方面的优势。