【摘要】：在網(wǎng)絡(luò)信息爆炸的年代,網(wǎng)絡(luò)輿情對(duì)人們?nèi)粘Ｉ畹挠绊懺絹?lái)越大。網(wǎng)絡(luò)輿情中的公共危機(jī)事件常常會(huì)成為人們關(guān)注的焦點(diǎn),其中的惡性發(fā)展的公共危機(jī)事件對(duì)社會(huì)穩(wěn)定和人們的生命財(cái)產(chǎn)安全都有著嚴(yán)重的危害。因此,及時(shí)識(shí)別公共危機(jī)事件,分析事件演化,對(duì)把握輿論導(dǎo)向,具有重要意義。本文分別研究了網(wǎng)絡(luò)新聞中的公共危機(jī)事件識(shí)別和微博中的危機(jī)事件演化。 在公共危機(jī)事件識(shí)別方面,因?yàn)橛|發(fā)詞是事件的核心標(biāo)識(shí),本文研究以觸發(fā)詞為核心的事件識(shí)別方法。首先采用擴(kuò)展觸發(fā)詞的方式彌補(bǔ)單純使用觸發(fā)詞方法召回率較低的缺陷,但是這樣就造成了準(zhǔn)確率的降低。因此使用基于規(guī)則的方法對(duì)擴(kuò)展觸發(fā)詞方法進(jìn)行改造。又提出了改進(jìn)的Apriori算法用以挖掘規(guī)則關(guān)鍵項(xiàng),節(jié)省了很多人工定義規(guī)則的時(shí)間。另外,本文對(duì)基于統(tǒng)計(jì)的分類方法進(jìn)行了一系列研究,提出了分層標(biāo)記的特征選擇方法,并以之優(yōu)化最大熵分類器來(lái)完成事件識(shí)別任務(wù)。 在危機(jī)事件演化方面,首先結(jié)合已有的情感詞典NTUSD和微博常用表情的中文描述構(gòu)建微博情感詞典；然后研究微博中公共危機(jī)事件的生命周期,并結(jié)合已有研究指出情感分布符合泊松分布規(guī)律,并提出了計(jì)算泊松分布擬合參數(shù)的算法IIA,通過(guò)該算法可以計(jì)算出微博中的熱點(diǎn)事件是否可以擬合泊松分布及其擬合參數(shù)；最后通過(guò)事件情感詞的權(quán)重差來(lái)判斷熱點(diǎn)事件發(fā)展趨勢(shì),若情感極性呈負(fù)向則認(rèn)為事件有可能發(fā)展成為公共危機(jī)事件。 除此以外,本文設(shè)計(jì)并實(shí)現(xiàn)了公共危機(jī)事件識(shí)別系統(tǒng),該系統(tǒng)是對(duì)理論知識(shí)的實(shí)際應(yīng)用,可以用來(lái)對(duì)網(wǎng)絡(luò)新聞進(jìn)行公共危機(jī)事件識(shí)別,及時(shí)發(fā)現(xiàn)公共危機(jī)事件。 本文的創(chuàng)新點(diǎn)有以下兩個(gè)方面： 1.提出分層標(biāo)記的特征選擇方法進(jìn)行特征簡(jiǎn)約和包裝,提高分類性能； 2.提出增量迭代接近算法,用于估計(jì)泊松分布的擬合參數(shù)。
[Abstract]:In the era of network information explosion, network public opinion has more and more influence on people's daily life. The public crisis event in the network public opinion often becomes the focus that people pay close attention to, the public crisis event of the malignant development among them has serious harm to the social stability and people's life and property safety. Therefore, it is of great significance to identify public crisis events and analyze the evolution of public crisis events in time to grasp the guidance of public opinion. This paper studies the identification of public crisis events in network news and the evolution of crisis events in Weibo. In the aspect of public crisis event recognition, because trigger word is the core of event identification, this paper studies the method of event recognition with trigger word as the core. Firstly, the method of extended trigger word is used to compensate for the low recall rate of the simple trigger word method, but this results in the reduction of the accuracy rate. Therefore, the extended trigger word method is modified by rule-based method. An improved Apriori algorithm is proposed to mine the key items of the rules, which saves a lot of time of manually defining the rules. In addition, this paper makes a series of researches on the classification method based on statistics, proposes a method of feature selection based on hierarchical markers, and uses it to optimize the maximum entropy classifier to complete the task of event recognition. In the aspect of crisis event evolution, we first combine the existing emotion dictionary (NTUSD) with the Chinese description of Weibo's commonly used expressions, and then study the life cycle of public crisis events in Weibo. Combined with the previous studies, it is pointed out that the distribution of emotion accords with the law of Poisson distribution, and the algorithm IIA, for calculating the fitting parameters of Poisson distribution can be used to calculate whether the hot events in Weibo can fit the Poisson distribution and its fitting parameters. Finally, the development trend of hot events is judged by the weight difference of event affective words. If the emotional polarity is negative, the event may develop into a public crisis event. In addition, this paper designs and implements a public crisis event identification system, which is a practical application of theoretical knowledge and can be used to identify public crisis events in network news and to discover public crisis events in time. The innovation of this paper has the following two aspects: 1. In order to improve the classification performance, a feature selection method based on stratified marking is proposed to simplify and package the features. 2. An incremental iterative approach algorithm is proposed to estimate the fitting parameters of Poisson distribution.
【學(xué)位授予單位】：北方工業(yè)大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2015
【分類號(hào)】：TP393.092;D035.2

【參考文獻(xiàn)】

相關(guān)期刊論文 前10條

1 付劍鋒;劉宗田;付雪峰;周文;仲兆滿;;基于依存分析的事件識(shí)別[J];計(jì)算機(jī)科學(xué);2009年11期

2 陳磊;陳世鴻;劉宇;王云華;;一種非常規(guī)突發(fā)事件演化的可計(jì)算模型[J];計(jì)算機(jī)工程與科學(xué);2011年09期

3 夏華林;張仰森;;基于規(guī)則與統(tǒng)計(jì)的Web突發(fā)事件新聞多層次分類[J];計(jì)算機(jī)應(yīng)用;2012年02期

4 丁菊玲;勒中堅(jiān);;基于觀點(diǎn)樹的網(wǎng)絡(luò)輿情危機(jī)預(yù)警方法[J];計(jì)算機(jī)應(yīng)用研究;2011年09期

5 周茜,趙明生,扈e，

本文編號(hào)：2224568