基于Gibbs-LDA和最小二乘支持向量机的物联网安全预测方法

针对物联网中各类用户的网络行为出现复杂化、多样化和恶意化的特征和趋势,提出了一种基于Gibbs—LDA和最小二乘支持向量机的物联网安全预测方法;首先,提取通信时间、地址和内容等文中信息作为多维的通信记录样本,然后基于LDA模型,将安全事件建模为主题,获取样本特征并得到主题模型,通过Gibbs算法来估算LDA模型中的参数,从而建立了基于LDA的物联网安全多维预模型,最后,在LDA特征空间上建立了特征与安全事件分布的权重,并将此权重用于初始化各个支持向量机的预测结果,将权值最大的最小二乘支持向量的预测结果作为最终的结果;仿真实验证明了文中方法能有效地实现物联网安全预测,在NIPS和VAST数据集上进行仿真实验,结果表明了文中方法较其他方法具有预测精度高和预测时间短的优点,具有较大的优越性。     

基于Sentence-LDA主题模型的短文本分类

短文本特征稀疏、上下文依赖性强的特点,导致传统长文本分类技术不能有效地被直接应用。为了解决短文本特征稀疏的问题,提出基于Sentence-LDA主题模型进行特征扩展的短文本分类方法。该主题模型是隐含狄利克雷分布模型（Latent Dirichlet Allocation, LDA）的扩展,假设一个句子只产生一个主题分布。利用训练好的Sentence-LDA主题模型预测原始短文本的主题分布,从而将得到的主题词扩展到原始短文本特征中,完成短文本特征扩展。对扩展后的短文本使用支持向量机（Support Vector Machine, SVM）进行最后的分类。实验显示,与传统的基于向量空间模型（Vector Space Model,VSM）直接表示短文本的方法比较,本文提出的方法可以有效地提高短文本分类的准确率。     

基于信息内容和拓扑关系的社会媒体用户兴趣分类

随着社会的发展,信息已经成为社会发展越来越重要的部分,人类的信息传播活动越来越明显地展示出分众特征,对用户的分类成为人类信息活动的一个重要研究课题.从这一目标出发,分别基于信息内容、拓扑关系和两者综合的方法,按兴趣主题对社会媒体用户进行分类.对于基于信息内容的用户分类,采用LDA主题模型从用户所发布的内容中提取其主题分布,基于这一分布,采用支持向量机、决策树、贝叶斯等多种模型按兴趣主题对用户进行分类.对于基于拓扑关系的分类,依据相同兴趣主题的用户倾向于拥有共同的粉丝这一发现,构建分类模型来按兴趣主题对用户进行分类.然后提出综合信息内容和拓扑关系的分类方法来对用户进行分类.最后基于大规模Twitter数据的实验发现,采用综合方法对用户进行的兴趣分类性能明显高于采用单一信息内容或粉丝拓扑方法的性能.     

基于多特征的垃圾微博检测方法

随着微博平台的快速发展,垃圾信息检测与过滤也面临着巨大的考验,实时精确地识别垃圾信息对于提高用户的体验以及微博平台的可持续发展意义重大.本文根据新浪微博的真实数据,提出了一种基于多特征的垃圾微博检测方法.首先,提取微博的显式特征（用户特征、内容特征）;然后利用文档主题生成模型（LDA）提取微博中的隐含主题特征;最后根据所提取的微博特征利用支持向量机（SVM）构建分类器.实验结果表明,该方法相比于现有方法在准确率和F1值方面都有一定的提升.     

LDA-RR:一种基于评分和评论的推荐方法

推荐系统是解决互联网信息过载问题的有效途径之一,其中具有代表性的是协同过滤推荐。传统的协同过滤推荐方法只考虑评分信息,而评论信息则包含了用户和物品更具体的特征信息。使用主题模型LDA并结合评分信息和评论信息,提出了一种基于用户改进的LDA算法。假设每个用户下隐含着主题分布,主题下隐含着物品分布,同时 词语的分布由主题和物品共同决定,该算法根据潜在主题分布挖掘用户兴趣进而完成推荐。实验结果表明,改进的算法有效提升了推荐质量。     

基于BERT和LDA模型的酒店评论文本挖掘

通过主题模型对酒店评论文本进行文本挖掘,有利于引导酒店管理者和客户对评论信息做出合适的鉴别和判断。提出一种基于预训练的BERT语言模型与LDA主题聚类相结合的方法;利用中文维基百科语料库训练BERT模型并从中获取文本向量,基于深度学习算法对评论文本进行情感分类;通过LDA模型对分类后的文本进行主题聚类,分别获取不同情感极性文本的特征主题词,进而挖掘出酒店客户最为关注的问题,并对酒店管理者提出具有参考价值的建议。实验结果表明,通过BERT模型获取的文本向量在情感分类任务中表现较好,且BERT-LDA文本挖掘方法能使酒店评论文本的主题更具表达性。     

用户评论的分类获取

对网上获取的用户评论进行标注,并提取出与用户评论内容相关的特征,使用χ2统计提取不同类型评论进行特征选择,使用支持向量机分类方法进行学习,获得分类器,以此对网上时时更新的用户评论进行分类,挖掘出优秀的评论。实验结果显示该方法具有很高的召回率和准确率。     

基于LDA和Word2vec的微博标签生成算法

针对基于传统LDA主题模型的标签生成算法对用户兴趣主题描述不完整的问题,提出一种基于主题嵌入表示的微博用户标签生成算法TopicERP.该算法在LDA模型的基础上,通过引入Word2vec词嵌入模型,对用户兴趣主题进行全面描述,并对匹配度计算方法进行改进.首先利用LDA主题模型对用户微博进行主题分析,生成用户兴趣主题;然后利用Word2vec词嵌入模型将主题文本转换为主题向量,用于匹配度计算;最后,利用余弦相似度和主题在文档中的条件概率,计算主题向量与候选标签匹配度,选取Top-Q的候选标签作为目标用户标签.本文在公开微博数据集microPCU上进行实验,实验结果表明,该算法在总体性能上高于基于传统LDA主题模型的微博标签生成算法,生成的用户标签能够较为准确地描述用户的兴趣偏好.     

基于数据增强的地质文本主题模型

直接利用主题模型对地质文本进行聚类时会出现主题准确性低、主题关键词连续性差等问题, 本文采取了相关改进方法. 首先在分词阶段采用基于词频统计的重复词串提取算法, 保留地质专业名词以准确提取文本主题, 同时减少冗余词串数量节约内存花销, 提升保留词的提取效率. 另外, 使用基于TF-IDF和词向量的文本数据增强算法, 对原始分词语料进行处理以强化文本主题特征. 之后该算法与主题模型相结合在处理后的语料上提取语料主题. 由于模型的先验信息得到增强, 故性能得以提高. 实验结果表明本文算法与LDA模型相结合的方法表现较好, 在相关指标及输出结果上均优于其他方法.     

教学视频的文本语义镜头分割和标注

为了对教学视频这一专门类别视频进行自动标注,本文首先提取视频中的字幕信息,通过文本预处理后,使用视频中的字幕文本信息内容结合潜在狄利克雷分布（Latent Dirichlet allocation,LDA）主题模型方法获得视频镜头在主题上的概率分布,通过计算主题概率分布差异,进行语义层面镜头分割。然后以镜头为样本,使用安全的半监督支持向量机（Safe semi-supervised support vector machine,S4VM）方法,通过少量的标注镜头样本,完成对未标注镜头的自动标注。实验结果表明,本文方法利用字幕文本信息和LDA模型,有效完成了视频的语义镜头分割,不仅可以对镜头完成标注,而且可以对整个视频进行关键词标注。     

CSLabel: An Approach for Labelling Mobile App Reviews

Mobile apps (applications) have become a popular form of software, and the app reviews by users have become an important feedback resource. Users may raise some issues in their reviews when they use apps, such as a functional bug, a network lag, or a request for a feature. Understanding these issues can help developers to focus on users’ concerns, and help users to evaluate similar apps for download or purchase. However, we do not know which types of issues are raised in a review. Moreover, the amount of user reviews is huge and the nature of the reviews’ text is unstructured and informal. In this paper, we analyze 3 902 user reviews from 11 mobile apps in a Chinese app store — 360 Mobile Assistant, and uncover 17 issue types. Then, we propose an approach CSLabel that can label user reviews based on the raised issue types. CSLabel uses a cost-sensitive learning method to mitigate the effects of the imbalanced data, and optimizes the setting of the support vector machine (SVM) classifier’s kernel function. Results show that CSLabel can correctly label reviews with the precision of 66.5%, the recall of 69.8%, and the F₁ measure of 69.8%. In comparison with the state-of-the-art approach, CSLabel improves the precision by 14%, the recall by 30%, the F₁ measure by 22%. Finally, we apply our approach to two real scenarios: 1) we provide an overview of 1 076 786 user reviews from 1 100 apps in the 360 Mobile Assistant and 2) we find that some issue types have a negative correlation with users’ evaluation of apps.     

数据驱动的移动应用用户接受度建模与预测

应用市场（app market）已经成为互联网环境下软件应用开发和交付的一种主流模式.相对于传统模式,应用市场模式下,软件的交付周期更短,用户的反馈更快,最终用户和开发者之间的联系更加紧密和直接.为应对激烈的竞争和动态演变的用户需求,移动应用开发者必须以快速迭代的方式不断更新应用,修复错误缺陷,完善应用质量,提升用户体验.因此,如何正确和综合理解用户对软件的接受程度（简称用户接受度）,是应用市场模式下软件开发需考量的重要因素.近年来兴起的软件解析学（software analytics）关注大数据分析技术在软件行业中的具体应用,对软件生命周期中大规模、多种类的相关数据进行挖掘和分析,被认为是帮助开发者提取有效信息、作出正确决策的有效途径.从软件解析学的角度,首先论证了为移动应用构建综合的用户接受度指标模型的必要性和可行性,并从用户评价数据、操作数据、交互行为数据这3个维度给出基本的用户接受度指标.在此基础上,使用大规模真实数据集,在目标用户群体预测、用户规模预测和更新效果预测等典型的用户接受度指标预测问题中,结合具体指标,提取移动应用生命周期不同阶段的重要特征,以协同过滤、回归融合、概率模型等方法验证用户接受度的可预测性,并讨论了预测结果与特征在移动应用开发过程中可能提供的指导.     

App store mining is not enough for app improvement

The rise in popularity of mobile devices has led to a parallel growth in the size of the app store market, intriguing several research studies and commercial platforms on mining app stores. App store reviews are used to analyze different aspects of app development and evolution. However, app users’ feedback does not only exist on the app store. In fact, despite the large quantity of posts that are made daily on social media, the importance and value that these discussions provide remain mostly unused in the context of mobile app development. In this paper, we study how Twitter can provide complementary information to support mobile app development. By analyzing a total of 30,793 apps over a period of six weeks, we found strong correlations between the number of reviews and tweets for most apps. Moreover, through applying machine learning classifiers, topic modeling and subsequent crowd-sourcing, we successfully mined 22.4% additional feature requests and 12.89% additional bug reports from Twitter. We also found that 52.1% of all feature requests and bug reports were discussed on both tweets and reviews. In addition to finding common and unique information from Twitter and the app store, sentiment and content analysis were also performed for 70 randomly selected apps. From this, we found that tweets provided more critical and objective views on apps than reviews from the app store. These results show that app store review mining is indeed not enough; other information sources ultimately provide added value and information for app developers.     

Studying the consistency of star ratings and the complaints in 1 & 2-star user reviews for top free cross-platform Android and iOS apps

How users rate a mobile app via star ratings and user reviews is of utmost importance for the success of an app. Recent studies and surveys show that users rely heavily on star ratings and user reviews that are provided by other users, for deciding which app to download. However, understanding star ratings and user reviews is a complicated matter, since they are influenced by many factors such as the actual quality of the app and how the user perceives such quality relative to their expectations, which are in turn influenced by their prior experiences and expectations relative to other apps on the platform (e.g., iOS versus Android). Nevertheless, star ratings and user reviews provide developers with valuable information for improving the overall impression of their app. In an effort to expand their revenue and reach more users, app developers commonly build cross-platform apps, i.e., apps that are available on multiple platforms. As star ratings and user reviews are of such importance in the mobile app industry, it is essential for developers of cross-platform apps to maintain a consistent level of star ratings and user reviews for their apps across the various platforms on which they are available. In this paper, we investigate whether cross-platform apps achieve a consistent level of star ratings and user reviews. We manually identify 19 cross-platform apps and conduct an empirical study on their star ratings and user reviews. By manually tagging 9,902 1 & 2-star reviews of the studied cross-platform apps, we discover that the distribution of the frequency of complaint types varies across platforms. Finally, we study the negative impact ratio of complaint types and find that for some apps, users have higher expectations on one platform. All our proposed techniques and our methodologies are generic and can be used for any app. Our findings show that at least 79% of the studied cross-platform apps do not have consistent star ratings, which suggests that different quality assurance efforts need to be considered by developers for the different platforms that they wish to support.     

融合主题模型和协同过滤的多样化移动应用推荐

随着移动应用的急速增长,手机助手等移动应用获取平台也面临着信息过载的问题.面对大量的移动应用,用户很难找到想到的或适合的应用,而另一方面长尾应用淹没在资源池中不被人所知.已有推荐方法多注重推荐准确率,忽视多样性,推荐结果中多是下载量高的应用,使得推荐系统的数据积累越来越偏向于热门应用,导致长期的推荐效果越来越差.针对此问题,本文首先改进了两个推荐方法,提出了将用户的主题模型和应用的主题模型与MF相结合的LDA_MF模型,以及将应用的标签信息和用户行为数据同时加以考虑的LDA_CF算法.为了结合不同算法的优点,在保证推荐准确率的条件下提升推荐结果的多样性,我们提出了融合LDA_MF、LDA_CF以及经典的基于物品的协同过滤模型的混合推荐算法.文章使用真实的大数据评测所提推荐算法,结果显示所提推荐方法能够得到推荐多样性更好且准确率高的结果.     

Comparison of different machine learning techniques on location extraction by utilizing geo-tagged tweets: A case study

In emergencies, Twitter is an important platform to get situational awareness simultaneously. Therefore, information about Twitter users’ location is a fundamental aspect to understand the disaster effects. But location extraction is a challenging task. Most of the Twitter users do not share their locations in their tweets. In that respect, there are different methods proposed for location extraction which cover different fields such as statistics, machine learning, etc. This study is a sample study that utilizes geo-tagged tweets to demonstrate the importance of the location in disaster management by taking three cases into consideration. In our study, tweets are obtained by utilizing the “earthquake” keyword to determine the location of Twitter users. Tweets are evaluated by utilizing the Latent Dirichlet Allocation (LDA) topic model and sentiment analysis through machine learning classification algorithms including the Multinomial and Gaussian Naïve Bayes, Support Vector Machine (SVM), Decision Tree, Random Forest, Extra Trees, Neural Network, k Nearest Neighbor (kNN), Stochastic Gradient Descent (SGD), and Adaptive Boosting (AdaBoost) classifications. Therefore, 10 different machine learning algorithms are applied in our study by utilizing sentiment analysis based on location-specific disaster-related tweets by aiming fast and correct response in a disaster situation. In addition, the effectiveness of each algorithm is evaluated in order to gather the right machine learning algorithm. Moreover, topic extraction via LDA is provided to comprehend the situation after a disaster. The gathered results from the application of three cases indicate that Multinomial Naïve Bayes and Extra Trees machine learning algorithms give the best results with an F-measure value over 80%. The study aims to provide a quick response to earthquakes by applying the aforementioned techniques.     

LDA主题抽取模型在互联网旅游评论的应用

面对网络上日益丰富的评论信息资源,如何在海量的客户评论中快速有效的获取并使用其中的有效信息,成为人们日益关注的问题。研究目标是互联网上的旅游评论,通过使用数据挖掘算法分析获取评论中关于商品或服务的主题词,并提取所有评论中包含主题词的句子。使用主题抽取模型(LDA模型)进行半监督的聚类处理,建立景点评论的主题模型,实现了互联网旅游评论个性化的设置和查询。     

基于地理标签的LBSN链接预测模型

为更深入挖掘用户位置信息,本文从位置语义相似性角度挖掘用户特征.利用LDA算法对用户签到信息进行位置主题建模,采用Gibbs采样算法计算LDA模型中的分布函数,并根据这些分布提出了基于签到地点语义的用户相似性特征向量.利用有监督的机器学习算法,综合LBSN的网络结构信息、签到地点信息、地点语义信息得到多维相似性特征向量来进行链接预测.在Gowalla数据集上的实验结果表明,相较于传统的链接预测算法,将基于签到信息的多个相似性特征作为辅助信息的链接预测算法显著提高了LBSN链接预测的性能.     

A joint model for analyzing topic and sentiment dynamics from large-scale online news

Many of today’s online news websites and aggregator apps have enabled users to publish their opinions without respect to time and place. Existing works on topic-based sentiment analysis of product reviews cannot be applied to online news directly because of the following two reasons: (1) The dynamic nature of news streams require the topic and sentiment analysis model also to be dynamically updated. (2) The user interactions among news comments can easily lead to inaccurate topic extraction and sentiment classification. In this paper, we propose a novel probabilistic generative model (DTSA) to extract topics and the specified sentiments from news streams and analyze their evolution over time simultaneously. In DTSA, three different timescale models are studied to account for the historical dependencies of sentiment-topic word distributions at current epoch, continuous, skip and multiple timescale models. Additionally, we further consider the links among news comments to avoid the error caused by user interactions. In order to mine more interpretable topics, a Conditional Random Fields (CRF) model is adopted to label a set of meaningful phrases for augmenting the bag-of-word features. Finally, we derive distributed online inference procedures to update the model with newly arrived data and show the effectiveness of our proposed model on real-world data sets.