基于深度学习的Stack Overflow问题帖分类方法

针对基于正则表达式和传统机器学习的分类方法分别存在模式手工提取困难和性能瓶颈的问题, 提出一种基于深度学习的问题帖分类方法, 采用深度文本挖掘模型TextCNN和融合注意力机制的TextRNN构建分类模型. 实验结果表明, 基于深度学习的方法在多数问题目的类别上的分类性能优于已有基准方法, 且使用的Adam优化器优于SGD优化器, 使用Glove预训练的词向量优于使用随机生成的词向量. 该方法以提问目的对帖子进行分类, 可为分析Stack Overflow(SO)上的帖子讨论主题增加新维度.     

一种基于类不平衡学习的情感分析方法

针对网络评论中普遍存在的负面评论较少而影响力却较大的类不平衡问题, 提出一种基于类不平衡学习的情感分析方法. 该方法利用深度学习训练过程中的概率输出, 以计算样例的信息熵作为影响因子构建交叉信息熵损失函数. 在IMDB公开数据集上进行实验验证的结果表明, 基于集成信息熵损失函数的双向长短期记忆网络能处理类不平衡问题; 对数据的统计分析结果表明, 该策略能提升基于双向长短期记忆网络的评论情感极性分类性能. 针对AUC(area under curve)指标, 使用集成信息熵损失函数的双向长短期记忆网络模型比未考虑类不平衡的深度学习模型在中位数上最多提升15.3%.     

The role of orientation experiments in discovering mechanisms

Many types of experiments have been recognized in the literature. One important type we discuss in this article is the orientation experiment. While orientation experiments are like other types of experiments in that they are tests for causal relevance, they also have other qualities. One important (but not the only) goal of these experiments is to offer a rough, qualitative characterization of the mechanism responsible for a capacity of interest, effectively constraining future research. This makes them particularly useful during the early stages of investigation, when an explanandum-phenomenon has just been identified and several (often competing) hypotheses as to the qualitative character of the mechanism responsible for it are proposed. We illustrate our claims, and explicate a number of additional aims that orientation experiments can sometimes serve, by considering three case studies from different era's, namely the discovery of the mechanisms responsible for i) the capacity of eels to produce numbing sensations (17th and 18th century), ii) puerperal fever in Semmelweis' Vienna Maternity Hospital (19th century), and iii) the capacity of pigeons to home (20th century).     

Forecasting economic indicators using a consumer sentiment index: Survey‐based versus text‐based data

Given the confirmed effectiveness of the survey‐based consumer sentiment index (CSI) as a leading indicator of real economic conditions, the CSI is actively used in making policy judgments and decisions in many countries. However, although the CSI offers qualitative information for presenting current conditions and predicting a household's future economic activity, the survey‐based method has several limitations. In this context, we extracted sentiment information from online economic news articles and demonstrated that the Korean cases are a good illustration of applying a text mining technique when generating a CSI using sentiment analysis. By applying a simple sentiment analysis based on the lexicon approach, this paper confirmed that news articles can be an effective source for generating an economic indicator in Korea. Even though cross‐national comparative research results are suited better than national‐level data to generalize and verify the method used in this study, international comparisons are quite challenging to draw due to the necessary linguistic preprocessing. We hope to encourage further cross‐national comparative research to apply the approach proposed in this study.     

论科学解释的实用特征和生物学解释的语境相关性

科学解释具有实用特征(pragmatic aspect).这一特征与"科学解释的标准形式"相比,具有解释主体的相关性,即因解释主体的差异寻求多样化的解释形式.本文通过生物学解释的特征的分析,指出生物学解释具有因解释主体不同而形成的语境相关性,而这一特性决定了生物学解释与物理学解释的本质差异.     

认知语言学对隐喻的诠释

世界在变,人对世界的认知也是在不断深化；对语言研究从规范、描写发展到解释认知语言学在客观主义和主观主义之间架起了一座桥梁,提出了主客观相结合的体验主义认识论。隐喻是一种重要的认知模式,是新的语言意义产生的根源。本文通过认知语言学理论阐述研究隐喻的理论基础,并解释隐喻与认知、隐喻与本义的关系,说明对隐喻的研究应注意解释的范围与意义,应从文化与思维入手,探讨语言的哲学实质,从而对语言进行深层次的研究。     

对一道习题结论的思考与拓展

通过对一道习题的解答,借助图象,给结论一个直观易记的几何解释,并对这一结论进行了拓展。     

解释学习在机器人过程控制中的应用

机器人过程控制是智能机器人系统的一个特别重要而又十分复杂的研究课题。本文将解释学习应用于机器人过程控制，为之开发了一种新的方法。仿真结果表明，该方法可行、有效。同时，还尝试了将Ｐｒｏｌｏｇ语言作为机器人过程控制的运行机制的形式化表达工具的可能性。     

系统功能语法与语篇分析

本文首先讨论了系统功能语法语篇分析的基本理论,包括语言的三大纯理功能及其各自实现的子系统、语篇分析的两个层次以及"语境→语篇→评价("联系语境变量)的分析步骤,并用该模式分析了一个完整的语篇。分析表明用系统功能语法框架对语篇进行分析可以揭示语篇深层含义,可以解释语篇的语义和文体特点。     

决策树在文本分类中的应用

介绍了文本分类的思想,探讨了通过决策树学习来对文本进行有效分类的过程。