本文選題：輿情分析　切入點：影響評價　出處：《天津大學(xué)》2015年博士論文　論文類型：學(xué)位論文

【摘要】：當(dāng)今網(wǎng)絡(luò)化和大數(shù)據(jù)時代,社會媒體已成為輿情事件發(fā)生、發(fā)展與演化的重要陣地。研發(fā)基于社會媒體網(wǎng)上—網(wǎng)下實時交互的輿情態(tài)勢精準(zhǔn)感知與態(tài)勢研判技術(shù),提升突發(fā)事件應(yīng)對處置的能力,既是學(xué)術(shù)研究的熱點與難點,也是商業(yè)、社會乃至國家安全面臨的急迫需求與重大任務(wù)。在此背景下,本論文基于社會媒體輿情事件的多源異構(gòu)大數(shù)據(jù)信息,主要工作包括如下三個方面:(1)融合多源異構(gòu)信息進(jìn)行輿情事件影響分析。針對網(wǎng)絡(luò)輿情數(shù)據(jù)多維度、多來源、信息異構(gòu)化等內(nèi)在特征,及現(xiàn)有輿情指標(biāo)體系覆蓋度不夠、指標(biāo)項選取難以多維量化和跨源融合等問題,提出了融合多源和異構(gòu)信息的輿情事件影響量化評價體系。該體系融合了新聞網(wǎng)站、論壇網(wǎng)站及微博網(wǎng)站等多種數(shù)據(jù)源,設(shè)計了基于情感計算、文本挖掘和用戶社會網(wǎng)絡(luò)分析的評價指標(biāo)項,并采用層次模糊評價方法進(jìn)行多維指標(biāo)權(quán)重計算,實現(xiàn)了數(shù)據(jù)與專家領(lǐng)域知識的有機(jī)集成。實驗表明,本論文提出的量化評價體系,能夠較為準(zhǔn)確和全面地刻畫網(wǎng)絡(luò)輿情事件的影響,與事件發(fā)展過程有較好的符合度,優(yōu)于基于單一來源的事件評價方法。(2)研究了面向輿情事件的實體詞抽取與擴(kuò)展演變挖掘。針對輿情事件發(fā)展過程中事件描述實體詞動態(tài)變化的問題,提出了基于雙層模型的事件描述實體詞抽取方法,采用融合用戶關(guān)聯(lián)關(guān)系網(wǎng)絡(luò)的查詢詞擴(kuò)展方法對實體詞進(jìn)行自動擴(kuò)展,實現(xiàn)了對事件演化過程中中文變體詞的識別。并將所提方法應(yīng)用于輿情事件的數(shù)據(jù)采集查詢詞擴(kuò)展,在事件數(shù)據(jù)采集中得到了實驗驗證。實驗表明,本論文方法提供了良好的實體詞抽取效果,以及實體詞擴(kuò)展詞和變體詞識別結(jié)果,應(yīng)用于輿情事件數(shù)據(jù)采集可提升數(shù)據(jù)采集的召回率,尤其是在事件描述關(guān)鍵詞包含“屏蔽敏感詞”的情況下效果明顯。(3)研究了面向輿情事件的基于異質(zhì)信息網(wǎng)絡(luò)的社區(qū)和話題動態(tài)演化分析方法。針對輿情事件發(fā)展過程中用戶與話題社區(qū)數(shù)量不確定、結(jié)構(gòu)不連續(xù)及兩者不能協(xié)同演化的問題,構(gòu)建了基于異質(zhì)信息網(wǎng)絡(luò)理論的輿情事件用戶社區(qū)與話題演化分析模型,實現(xiàn)了基于狄利克雷過程混合模型的異質(zhì)社區(qū)的協(xié)同發(fā)現(xiàn)及演化。實驗表明,論文構(gòu)建的模型能準(zhǔn)確地刻畫輿情事件的動態(tài)社區(qū)演化過程,較好地解決輿情事件演化過程中社區(qū)數(shù)量自動學(xué)習(xí)、社區(qū)結(jié)構(gòu)平滑演化、社區(qū)與話題協(xié)同發(fā)現(xiàn)等難題。
[Abstract]:In the era of network and big data, social media has become an important position for the occurrence, development and evolution of public opinion events. It is not only the hot and difficult point of academic research, but also the urgent need and important task for business, society and even national security to improve the ability of dealing with emergencies. This paper is based on the social media public opinion event multi-source heterogeneous big data information, the main work includes the following three aspects: 1) fusion of multi-source heterogeneous information for public opinion event impact analysis. Such internal characteristics as information isomerization, insufficient coverage of the existing public opinion index system, difficulty in selecting index items for multidimensional quantification and cross-source fusion, etc. This paper proposes a quantitative evaluation system of the influence of public opinion events, which integrates multi-source and heterogeneous information. The system integrates various data sources, such as news website, forum website and Weibo website, and designs a calculation based on emotion. Text mining and user social network analysis of the evaluation index items, and the use of hierarchical fuzzy evaluation method for multi-dimensional index weight calculation, to achieve the organic integration of data and expert domain knowledge. The quantitative evaluation system proposed in this paper can accurately and comprehensively depict the influence of network public opinion events and has a good consistency with the event development process. It is better than the event evaluation method based on single source to study the entity word extraction and extended evolution mining for public opinion event. Aiming at the problem of the event describing the dynamic change of entity word during the development of public opinion event, An event description entity word extraction method based on two-layer model is proposed, and the query word extension method based on user relational network is used to extend entity word automatically. This paper realizes the recognition of Chinese variant words in the process of event evolution, and applies the proposed method to extend the query words of public opinion event data collection, and gets the experimental verification in the event data collection. The experimental results show that the proposed method can be applied to the data acquisition of public opinion events. This paper provides a good result of entity word extraction, and the result of entity word extension and variant word recognition, which can improve the recall rate of public opinion event data collection. In particular, when the event description keyword contains "masking sensitive words", the effect is obvious. (3) the analysis method of community and topic dynamic evolution based on heterogeneous information network for public opinion event is studied. Aiming at the development of public opinion event, this paper studies the analysis method of community and topic dynamic evolution based on heterogeneous information network. The number of users and topic communities in the process is uncertain, Based on heterogeneous information network theory, the analysis model of user community and topic evolution of public opinion events is constructed. The collaborative discovery and evolution of heterogeneous communities based on the Delikley process hybrid model are realized. Experiments show that the model constructed in this paper can accurately depict the dynamic community evolution process of public opinion events. It can solve the problems such as automatic learning of community quantity, smooth evolution of community structure and cooperative discovery of community and topic in the process of evolution of public opinion events.
【學(xué)位授予單位】：天津大學(xué)
【學(xué)位級別】：博士
【學(xué)位授予年份】：2015
【分類號】：G206;G254

【相似文獻(xiàn)】

相關(guān)期刊論文 前10條

1 許鑫;章成志;;互聯(lián)網(wǎng)輿情分析及應(yīng)用研究[J];情報科學(xué);2008年08期

2 魏麗萍;;互聯(lián)網(wǎng)輿情形成機(jī)制探析[J];濰坊學(xué)院學(xué)報;2010年01期

3 陳永剛;孫卉W，

本文編號：1640439