發(fā)布時(shí)間：2018-02-16 19:13

  本文關(guān)鍵詞： 內(nèi)容廣告 廣告相關(guān)性 點(diǎn)擊率預(yù)估 廣告排序 邏輯回歸　出處：《南京大學(xué)》2012年碩士論文　論文類型：學(xué)位論文

【摘要】：內(nèi)容廣告是互聯(lián)網(wǎng)廣告的一種,基于分析頁(yè)面內(nèi)容和用戶信息將高相關(guān)性廣告展現(xiàn)在網(wǎng)頁(yè)上。內(nèi)容廣告系統(tǒng)與傳統(tǒng)互聯(lián)網(wǎng)廣告系統(tǒng)有著很大的不同,內(nèi)容廣告系統(tǒng)主要將小廣告主的廣告展現(xiàn)在長(zhǎng)尾流量上,因此,內(nèi)容廣告系統(tǒng)的廣告庫(kù)更大,流量也更多。在內(nèi)容廣告系統(tǒng)中,每次廣告檢索都是從百萬(wàn)級(jí)的廣告庫(kù)中挑選與頁(yè)面、用戶信息最相關(guān)的部分廣告,由于性能原因,無(wú)法運(yùn)用復(fù)雜的技術(shù)逐一計(jì)算每條廣告的相關(guān)性,所以,內(nèi)容廣告系統(tǒng)按照相關(guān)性計(jì)算的復(fù)雜度將檢索過(guò)程分成兩個(gè)部分：廣告粗選和廣告排序。廣告粗選階段采用計(jì)算量較小的技術(shù)挑選部分廣告,然后在廣告排序階段運(yùn)用復(fù)雜的分析技術(shù)對(duì)這部分廣告進(jìn)行排序。本文主要關(guān)注廣告排序階段,即點(diǎn)擊率預(yù)估。 傳統(tǒng)計(jì)算相關(guān)性的方法是提取廣告和頁(yè)面的關(guān)鍵詞向量,計(jì)算兩個(gè)向量的相似度,這種方法最大的缺點(diǎn)是忽略了廣告展示和點(diǎn)擊的歷史日志。本文介紹的點(diǎn)擊率預(yù)估系統(tǒng)通過(guò)提取廣告、用戶和頁(yè)面信息的特征,運(yùn)用邏輯回歸模型預(yù)估廣告點(diǎn)擊率,并基于此對(duì)廣告進(jìn)行排序,邏輯回歸模型從線下廣告歷史日志中訓(xùn)練得出。相對(duì)于傳統(tǒng)方法,點(diǎn)擊率預(yù)估技術(shù)利用的信息更加全面,從歷史日志中挖掘信息訓(xùn)練模型也使得相關(guān)性計(jì)算更加準(zhǔn)確。 本文主要介紹了點(diǎn)擊率預(yù)估系統(tǒng)的設(shè)計(jì)與實(shí)現(xiàn)。首先介紹了國(guó)內(nèi)外計(jì)算廣告相關(guān)性的各種方法,引出了點(diǎn)擊率預(yù)估,然后介紹了點(diǎn)擊率預(yù)估的算法原理和在實(shí)現(xiàn)點(diǎn)擊率預(yù)估系統(tǒng)的過(guò)程中使用的主要技術(shù)。在后續(xù)章節(jié)中,通過(guò)對(duì)內(nèi)容廣告系統(tǒng)的整體架構(gòu)以及設(shè)計(jì)思想的分析,引出了點(diǎn)擊率預(yù)估的需求包括功能、性能和內(nèi)外部接口。圍繞著需求展開了對(duì)點(diǎn)擊率預(yù)估系統(tǒng)的設(shè)計(jì)與實(shí)現(xiàn)的介紹,并著重在性能和算法實(shí)驗(yàn)的便捷性兩個(gè)方面做了重點(diǎn)的分析優(yōu)化。最后詳細(xì)分析了點(diǎn)擊率預(yù)估系統(tǒng)對(duì)整個(gè)內(nèi)容廣告系統(tǒng)帶來(lái)的效果提升。論文的最后,通過(guò)總結(jié)與展望,對(duì)技術(shù)的改進(jìn)方向以及應(yīng)用前景做了進(jìn)一步的分析。
[Abstract]:Content advertising is a kind of Internet advertising, which is based on analyzing page content and user information to display highly relevant ads on web pages. Content advertising systems are very different from traditional Internet advertising systems. The content advertising system mainly displays the small advertisers' advertisements on the long tail flow, so the content advertising system has a larger advertising base and more traffic. In the content advertising system, Each advertising retrieval is a selection of pages from the millions of ad libraries, the most relevant part of user information, because of performance reasons, can not use complex technology to calculate the relevance of each ad, so, The content advertising system divides the retrieval process into two parts according to the complexity of correlation calculation: ad selection and advertisement sorting. Then we use the complex analysis technology to sort this part of advertisements in the advertising sequencing stage. This paper mainly focuses on the advertising sequencing stage, that is, the prediction of click rate. The traditional method to calculate the correlation is to extract the keyword vector of the advertisement and the page, and calculate the similarity between the two vectors. The biggest drawback of this method is that it ignores the historical log of advertising display and click. The click rate prediction system introduced in this paper uses the logical regression model to estimate the ad click rate by extracting the features of advertisement, user and page information. The logical regression model is trained from the offline advertising history log. Compared with the traditional method, the information used by the click rate estimation technology is more comprehensive. Mining information training model from history log also makes correlation calculation more accurate. This paper mainly introduces the design and realization of the prediction system of click rate. Firstly, it introduces various methods of calculating the correlation of advertisement at home and abroad, and leads to the prediction of click rate. Then it introduces the algorithm principle and main technology used in the process of realizing the prediction system of click rate. In the following chapters, through the analysis of the whole structure and design idea of the content advertising system, The requirements for the prediction of click rate include function, performance and internal and external interfaces. The design and implementation of the system are introduced around the demand. The performance and the convenience of algorithm experiment are analyzed and optimized in detail. Finally, the effect of the click rate prediction system on the whole content advertising system is analyzed in detail. The improvement direction and application prospect of the technology are further analyzed.
【學(xué)位授予單位】：南京大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2012
【分類號(hào)】：TP311.52

【相似文獻(xiàn)】

相關(guān)期刊論文 前10條

1 鄭曉霞;劉超;鄒鈺;;基于邏輯回歸模型的中文垃圾短信過(guò)濾[J];黑龍江工程學(xué)院學(xué)報(bào)(自然科學(xué)版);2010年04期

2 俞慶生;;基于云平臺(tái)的邏輯回歸模型構(gòu)建算法的設(shè)計(jì)與實(shí)現(xiàn)[J];科技通報(bào);2013年06期

3 劉遵雄;黃志強(qiáng);鄭淑娟;張恒;;基于內(nèi)點(diǎn)法的稀疏邏輯回歸財(cái)務(wù)預(yù)警模型[J];計(jì)算機(jī)工程與設(shè)計(jì);2013年06期

4 邢秋菊,趙純勇,高克昌,郭躍;基于GIS的滑坡危險(xiǎn)性邏輯回歸評(píng)價(jià)研究[J];地理與地理信息科學(xué);2004年03期

5 吳敏;何瓏;;融合多特征的產(chǎn)品垃圾評(píng)論識(shí)別[J];微型機(jī)與應(yīng)用;2012年22期

6 王賢;張自力;;基于邏輯回歸的案件關(guān)聯(lián)[J];西南師范大學(xué)學(xué)報(bào)(自然科學(xué)版);2009年03期

7 齊超;陳鴻昶;于巖;;基于行為分析的微博信息傳播效果[J];計(jì)算機(jī)應(yīng)用;2014年08期

8 ;[J];;年期

9 ;[J];;年期

10 ;[J];;年期

相關(guān)碩士學(xué)位論文 前4條

1 祁全昌;基于內(nèi)容廣告平臺(tái)的點(diǎn)擊率預(yù)估系統(tǒng)的設(shè)計(jì)與實(shí)現(xiàn)[D];南京大學(xué);2012年

2 劉力銀;基于邏輯回歸的推薦技術(shù)研究及應(yīng)用[D];電子科技大學(xué);2013年

3 王兵;一種基于邏輯回歸模型的搜索廣告點(diǎn)擊率預(yù)估方法的研究[D];浙江大學(xué);2013年

4 劉建明;高管變更對(duì)民營(yíng)企業(yè)進(jìn)入高壁壘行業(yè)的影響研究[D];南開大學(xué);2011年

，

本文編號(hào)：1516239