發(fā)布時(shí)間：2018-04-16 09:37

  本文選題：廣告 + 回歸模型��； 參考：《南京大學(xué)》2012年碩士論文

【摘要】：廣告是眾多互聯(lián)網(wǎng)站點(diǎn)的主要盈利模式之一,隨著互聯(lián)網(wǎng)廣告行業(yè)的發(fā)展,有越來(lái)越多的網(wǎng)站將廣告位交由廣告聯(lián)盟托管。廣告聯(lián)盟和多個(gè)廣告主簽訂廣告投放計(jì)劃,通過(guò)一系列匹配算法,將廣告展現(xiàn)在最合適的網(wǎng)站上。廣告普遍按照點(diǎn)擊次數(shù)計(jì)費(fèi),但廣告的點(diǎn)擊率是不確定的,不同廣告的點(diǎn)擊單價(jià)也不同。對(duì)于網(wǎng)站主來(lái)說(shuō),一個(gè)廣告位未來(lái)能獲取的收益是未知的,這在一定程度上影響了網(wǎng)站主的積極性。在這種情況下,如何預(yù)測(cè)一個(gè)廣告位所能獲取的收入就成為一個(gè)亟待解決的問(wèn)題。 本文介紹了一種預(yù)測(cè)廣告位收益的方法,主要思想是利用大量歷史數(shù)據(jù),通過(guò)支持向量機(jī)來(lái)建立一個(gè)回歸模型。主要工作分為三個(gè)部分。第一,環(huán)境特征抽取,環(huán)境特征通過(guò)抓取并分析廣告位所在網(wǎng)頁(yè)的內(nèi)容得到。第二,數(shù)據(jù)處理,分析所有特征數(shù)據(jù)的有效性和特征之間的相關(guān)性,之后根據(jù)模型訓(xùn)練的需要,對(duì)數(shù)據(jù)進(jìn)行篩選和轉(zhuǎn)換。第三,模型訓(xùn)練與優(yōu)化,模型訓(xùn)練基于已有的支持向量機(jī)算法庫(kù)完成,在此基礎(chǔ)上通過(guò)參數(shù)尋優(yōu)、特征選取、特征綁定等方法做了大量?jī)?yōu)化。本文詳細(xì)闡述了這三個(gè)階段的工作和成果,并在最后進(jìn)行了簡(jiǎn)單總結(jié),提出了未來(lái)可以進(jìn)行的改進(jìn)。
[Abstract]:Advertising is one of the main profit models for many Internet sites. With the development of the Internet advertising industry, more and more websites are handing over advertising positions to ad federations.The Advertising Alliance has signed up with multiple advertisers to display ads on the most appropriate site through a series of matching algorithms.Ads generally charge according to the number of clicks, but the click rate is uncertain, and the click price varies from ad to ad.For website owners, the future revenue of an advertising site is unknown, which to some extent affects the enthusiasm of site owners.In this case, how to predict the revenue a advertising position can achieve becomes a problem to be solved.This paper introduces a method to predict the revenue of advertising bit. The main idea is to build a regression model by using a large amount of historical data and support vector machine.The main work is divided into three parts.First, environmental feature extraction, which is obtained by grabbing and analyzing the content of the page in which the advertisement is located.Secondly, data processing analyzes the validity of all feature data and the correlation between features, and then filters and transforms the data according to the needs of model training.Third, model training and optimization, model training based on the existing support vector machine algorithm library, on this basis through parameter optimization, feature selection, feature binding and other methods to do a lot of optimization.In this paper, the work and achievements of these three stages are described in detail, and a brief summary is made at the end of this paper, and some possible improvements in the future are put forward.
【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)】：TP393.092

【參考文獻(xiàn)】

相關(guān)期刊論文 前1條

1 張浩然,韓正之;回歸支持向量機(jī)的改進(jìn)序列最小優(yōu)化學(xué)習(xí)算法[J];軟件學(xué)報(bào);2003年12期

，

本文編號(hào)：1758361