基于随机词汇迭代模型的POI分类检索

从特征选择、局部区域划分和词汇语义相似性计算入手,利用随机词汇迭代模型(random terms iterativemodel,RTIM)进行海量兴趣点(point of interest,POI)文本分类.通过词汇频度、集中度和离散度方法筛选出特征词汇;依据文本与各POI类别间的相似度进行局部区域划分;在每个局部区域内基于词汇在文本中的排列顺序构建词频向量,基于词频向量中词频的随机删除和重构,获取特征映射矩阵;通过特征映射矩阵将文本转为特征向量,并采用SVM分类器进行POI文本分类.实验证明,该方法有效提升了POI文本分类准确性和覆盖率.     

结合双粒子群和K-means的混合文本聚类算法

传统K-means算法对初始聚类中心选择较敏感, 结果有可能收敛于一般次优解, 为些提出一种结合双粒子群和K-means的混合文本聚类算法。设计了自调整惯性权值策略, 根据最优适应度值的变化率动态调整惯性权值。两子群分别采用基于不同惯性权值策略的粒子群算法进化, 子代间及子代与父代信息交流, 共享最优粒子, 替换最劣粒子, 完成进化, 该算法命名为双粒子群算法。将能平衡全局与局部搜索能力的双粒子群算法与高效的K-means算法结合, 每个粒子是一组聚类中心, 类内离散度之和的倒数是适应度函数, 用K-means算法优化新生粒子, 即为结合双粒子群和K-means的混合文本聚类算法。实验结果表明, 该算法相对于K-means、PSO等文本聚类算法具有更强鲁棒性, 聚类效果也有明显的改善。     

基于蚁群聚集信息素的半监督文本分类算法

半监督文本分类中已标记数据与未标记数据分布不一致,可能导致分类器性能较低。为此,提出一种利用蚁群聚集信息素浓度的半监督文本分类算法。将聚集信息素与传统的文本相似度计算相融合,利用Top-k策略选取出未标记蚂蚁可能归属的种群,依据判断规则判定未标记蚂蚁的置信度,采用随机选择策略,把置信度高的未标记蚂蚁加入到对其最有吸引力的训练种群中。在标准数据集上与朴素贝叶斯算法和EM算法进行对比实验,结果表明,该算法在精确率、召回率以及F1度量方面都取得了更好的效果。     

基于GPU的文本特征选择与特征加权

通过对GPU通用计算与文本分类的研究,提出了一种基于GPU的文本特征选择与加权方法。首先介绍了文本分类中常用的特征选择方法和特征加权方法,并在GPU上实现了其中的DF(文档频率)方法和TFIDF方法。实验结果显示,利用提出的并行特征选择与加权方法能能有效的提高特征选择与加权过程的速度。     

面向信息检索的藏文文本索引策略研究

互联网文本数量持续爆炸式增长,用户通过互联网查找信息变得更加困难,响应时间得不到满足。针对藏文本身的语言学特点,探讨一种面向信息搜索的藏文文本索引建立策略,建立一种高效的藏文文本索引,以提高藏文信息检索速度。     

基于LASVM-NC和TF.RF的文本分类方法

非凸在线支持向量机（LASVM-NC）具有抗噪能力强和训练速度快的优点,而词频相关频率积（tf.rf）则是一种自适应能力很强、分类性能非常好的文本特征。通过把非凸在线支持向量机和词频相关频率积相结合,提出了一种新的文本分类方法,即LASVM-NC＋tf.rf。实验结果表明,这种方法在LASVM-NC与多种其他特征的结合中性能是最好的,且与SVM＋tf.rf相比,不仅所产生的分类器具有泛化能力更强、模型表达更稀疏的优点,而且在处理含噪声的数据时具有更好的鲁棒性,在处理大规模数据时具有快得多的训练速度。     

Exploring the Interactions of Storylines from Informative News Events

Today＇s news readers can be easily overwhelmed by the numerous news articles online. To cope with information overload, online news media publishes timelines for continuously developing news topics. However, the timeline summary does not show the relationship of storylines, and is not intuitive for readers to comprehend the development of a complex news topic. In this paper, we study a novel problem of exploring the interactions of storylines in a news topic. An interaction of two storylines is signified by informative news events that play a key role in both storylines. Storyline interactions can indicate key phases of a news topic, and reveal the latent connections among various aspects of the story. We address the coherence between news articles which is not considered in traditional similarity-based methods, and discover salient storyline interactions to form a clear, global picture of the news topic. User preference can be naturally integrated into our method to generate query-specific results. Comprehensive experiments on ten news topics show the effectiveness of our method over alternative approaches.     

Inductive Model Generation for Text Classification Using a Bipartite Heterogeneous Network

Algorithms for numeric data classification have been applied for text classification. Usually the vector space model is used to represent text collections. The characteristics of this representation such as sparsity and high dimensionality sometimes impair the quality of general-purpose classifiers. Networks can be used to represent text collections, avoiding the high sparsity and allowing to model relationships among different objects that compose a text collection. Such network- based representations can improve the quality of the classification results. One of the simplest ways to represent textual collections by a network is through a bipartite heterogeneous network, which is composed of objects that represent the documents connected to objects that represent the terms. Heterogeneous bipartite networks do not require computation of similarities or relations among the objects and can be used to model any type of text collection. Due to the advantages of representing text collections through bipartite heterogeneous networks, in this article we present a text classifier which builds a classification model using the structure of a bipartite heterogeneous network. Such an algorithm, referred to as IMBHN （Inductive Model Based on Bipartite Heterogeneous Network）, induces a classification model assigning weights to objects that represent the terms for each class of the text collection. An empirical evaluation using a large amount of text collections from different domains shows that the proposed IMBHN algorithm produces significantly better results than k-NN, C4.5, SVM, and Naive Bayes algorithms.     

稀疏卷积非负矩阵分解的语音增强算法

非平稳噪声和低信噪比条件下提高增强语音质量一直以来都是语音增强研究的难题。近年来,卷积非负矩阵分解在语音增强算法中成功应用,本文进一步考虑语音信号在时频域的稀疏性,提出了稀疏卷积非负矩阵分解(Sparse Convolutive Nonnegative Matrix Factorization, SCNMF)的语音增强算法。该算法包括训练和增强两个阶段。训练阶段通过SCNMF算法分别对纯净语音和噪声的频谱进行训练,得到纯净语音和噪声字典,并将其作为增强阶段的先验信息。增强阶段首先通过SCNMF算法对带噪语音的频谱进行分解,然后利用纯净语音和噪声联合字典对语音编码矩阵进行估计,重构增强语音。本文通过实验仿真分析了稀疏因子对增强语音质量的影响。实验结果表明,在非平稳噪声和低信噪比条件下,本文算法增强效果均优于多带谱减、非负矩阵分解、卷积非负矩阵分解等传统的算法。     

用于文本分类的特征项权重算法改进

TF-IDF算法是文本分类中一种常用的权重计算方法,但是TF-IDF仅仅考虑了特征项在文本中出现的次数以及该特征项在训练集中的出现频率,没有考虑特征项在各个类间的分布情况及特征项的语义信息。因此针对TF-IDF的不足提出了一种改进的TF-IDF算法,此算法既考虑了特征项在类内的分布情况又考虑了特征项的位置及长度等语义因素,能更好地反映特征项的重要性。用朴素贝叶斯分类器验证其有效性,实验结果表明该算法优于TF-IDF算法,能较好地提高文本分类的准确率。