【摘要】：近年來,社交網(wǎng)絡(luò)由于其便利性和及時性,成為人們分享和交流的一個主要平臺,也帶來了在線媒體信息的爆炸性增長。挖掘在線媒體中熱點(diǎn)信息成為一個備受關(guān)注的研究方向,其中預(yù)測在線社交網(wǎng)絡(luò)中內(nèi)容的流行趨勢對營銷、流量控制都具有重要意義。本文分析了國內(nèi)外社交網(wǎng)絡(luò)流行趨勢的研究現(xiàn)狀,通過分析社交網(wǎng)絡(luò)中社交關(guān)系和社會影響,提出了預(yù)測流行趨勢和用戶行為的解決方案。本文主要工作如下:1.本文通過對典型社交網(wǎng)站數(shù)據(jù)的分析,發(fā)現(xiàn)內(nèi)容發(fā)布后,轉(zhuǎn)發(fā)過程中不同時間段的轉(zhuǎn)發(fā)用戶對內(nèi)容流行度有重要作用,并且發(fā)現(xiàn)活躍度高但是相互關(guān)注數(shù)不高的用戶對其朋友的影響更大;部分內(nèi)容是潛在流行內(nèi)容,他們在前期不流行,但是隨著時間推移反而變得十分流行�；谶@些發(fā)現(xiàn),本文提出一種發(fā)現(xiàn)社交網(wǎng)絡(luò)內(nèi)容傳播過程中的關(guān)鍵節(jié)點(diǎn)的方法,用于預(yù)測潛在流行內(nèi)容和用戶轉(zhuǎn)發(fā)行為。2.本文根據(jù)社交網(wǎng)絡(luò)內(nèi)容轉(zhuǎn)發(fā)序列中用戶的關(guān)鍵性提出了一種基于關(guān)鍵節(jié)點(diǎn)來預(yù)測內(nèi)容流行趨勢的算法框架�；谵D(zhuǎn)發(fā)用戶的流行預(yù)測算法首先將內(nèi)容轉(zhuǎn)發(fā)序列劃分成T個時間窗,然后提取每個時間片段內(nèi)轉(zhuǎn)發(fā)用戶的關(guān)鍵性作為T維特征,并用回歸算法對最終流行度進(jìn)行預(yù)測。在典型社交網(wǎng)絡(luò)(微博)數(shù)據(jù)集上進(jìn)行實(shí)驗(yàn),發(fā)現(xiàn)相比現(xiàn)有流行預(yù)測算法,本文提出的算法在預(yù)測準(zhǔn)確度和排序準(zhǔn)確度上都有明顯提升(MAE提升36.8%,tau提升2.9%),并且基于轉(zhuǎn)發(fā)用戶的趨勢預(yù)測方法在預(yù)測流行度較高的內(nèi)容更加準(zhǔn)確,也能更早地發(fā)現(xiàn)流行內(nèi)容。3.本文分析社交網(wǎng)絡(luò)中的社會影響,用局部子網(wǎng)絡(luò)來描述全局網(wǎng)絡(luò)對用戶的社會影響,提出根據(jù)社會影響和節(jié)點(diǎn)關(guān)鍵性預(yù)測用戶行為的模型�；诰植可鐣绊懙内厔蓊A(yù)測方法首先構(gòu)建目標(biāo)用戶的局部子網(wǎng)絡(luò),然后根據(jù)鄰居節(jié)點(diǎn)關(guān)鍵性和節(jié)點(diǎn)之間相關(guān)性衡量局部網(wǎng)絡(luò)的社會影響。在典型社交網(wǎng)絡(luò)(微博)數(shù)據(jù)集上進(jìn)行實(shí)驗(yàn),對比現(xiàn)有預(yù)測用戶轉(zhuǎn)發(fā)行為的算法,發(fā)現(xiàn)分類效果有明顯提升(相比基準(zhǔn)算法分類準(zhǔn)確率提升20.6%)。這說明局部網(wǎng)絡(luò)的社會影響確實(shí)存在并影響了用戶的行為。本文還研究了局部網(wǎng)絡(luò)影響聚合的不同方法,發(fā)現(xiàn)社會影響是與時間密切相關(guān)的,隨著鄰居節(jié)點(diǎn)轉(zhuǎn)發(fā)時間越久,其累積的影響也越大。
[Abstract]:In recent years, because of its convenience and timeliness, social network has become a main platform for people to share and communicate, and has also brought the explosive growth of online media information. Mining hot information in online media has become a hot research direction, in which predicting the popular trend of online social network content is of great significance to marketing and traffic control. This paper analyzes the current situation of social networking trends at home and abroad, and puts forward solutions to predict popular trends and user behaviors by analyzing social relations and social impacts in social networks. The main work of this paper is as follows: 1. By analyzing the data of typical social networking sites, we find that after the content is released, the forwarding users in different time periods play an important role in the popularity of the content. It was also found that users with high activity but low mutual attention had greater influence on their friends. Part of the content is potential popular content, they were not popular in the early days, but became very popular over time. Based on these findings, this paper proposes a method to discover key nodes in the process of content propagation on social networks, which is used to predict potential popular content and user forwarding behavior. 2. This paper presents an algorithm framework to predict the trend of social network content based on key nodes according to the key of users in the social network content forwarding sequence. The popular prediction algorithm based on forwarding users first divides the content forwarding sequence into T time windows, then extracts the key of forwarding users within each time segment as T-dimension feature, and predicts the final popularity with regression algorithm. Experiments on typical social network (Weibo) data set show that the proposed algorithm has a significant improvement in prediction accuracy and sorting accuracy (MAE increase is 36.8% and tau increase 2.9%). And the trend forecasting method based on forwarding users is more accurate in predicting the content with higher popularity, and can also find the popular content earlier. This paper analyzes the social impact in social networks, describes the social impact of global networks on users by using local subnetworks, and puts forward a model to predict user behavior based on social impact and node key. The trend prediction method based on the local social impact firstly constructs the local subnetwork of the target user and then measures the social impact of the local network according to the key of the neighbor node and the correlation between the nodes. Experiments on typical social network (Weibo) data sets show that the classification effect is significantly improved (20.6% higher than the classification accuracy of the benchmark algorithm) by comparing the existing algorithms for predicting user forwarding behavior. This shows that the social impact of local networks does exist and affect the behavior of users. This paper also studies different methods of local network influencing aggregation. It is found that social impact is closely related to time. The longer the neighbor node forwards, the greater the cumulative impact is.
【學(xué)位授予單位】：電子科技大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：TP393.09

【參考文獻(xiàn)】

相關(guān)期刊論文 前1條

1 陶卿;姚穗;范勁松;方廷健;;一種新的機(jī)器學(xué)習(xí)算法:Support Vector Machines[J];模式識別與人工智能;2000年03期

相關(guān)碩士學(xué)位論文 前1條

1 陳威;在線社會媒體的流行性預(yù)測研究[D];電子科技大學(xué);2016年

，

本文編號：2306864