无关语获取与语料聚类方法研究

剔除无关语及语料聚类对提高自然语言理解的质量具有重要意义,也是自然语言理解的预处理关键技术.鉴于无关语在语料中存在明显的特性,本文通过种子无关语推导出强无关语,并依据强无关语识别并导出新的无关语;然后,基于2-gram构建句子之间的相似性,利用层次法对语料进行聚类对QA语料进行问题相似的聚类.最后,通过识别的新无关语实验及语料聚类实验,验证本文提出方法的有效性.

用户短文本无关语自动识别方法研究

在用户短文本中,意思相同的句子有多种表述方式,这些句子中存在很多与句意无关的信息,称为无关语。针对一般方法无关语识别准确度不高的问题,论文提出了一种通过二阶隐马尔科夫模型来自动识别用户短文本中无关语的方法。本方法在建模过程中将词本身、词性以及词的相对位置作为特征来对隐马尔科夫模型进行扩充。实验结果表明,论文给出的用户短文本无关语识别方法可以避免对训练文本进行手工编写规则的限制,且在准确率和召回率方面均有一定程度的提高。

Equation Discovery for Financial Forcasting in Context of Islamic Banking

This paper describes an equation discovery approach based on machine learning using LAGRAMGE as an equation discovery tool, with two sources of input, a dataset and model presented in context-free grammar. The approach is searching a large range of po- tential equations by a specific inodel. The parameters of the equation are fitted to find the best equations. The experiments are illustratedwith commodity prices from the London Metal Exchange for the period of January-October 2009. The outputs of the experiments are a large mumber of equations; some of the equations display that the predicted prices are following the market trends in perfect patterns.