本文選題：復(fù)雜網(wǎng)絡(luò) + 社區(qū)挖掘　； 參考：《吉林大學(xué)》2012年博士論文

【摘要】：復(fù)雜網(wǎng)絡(luò)社區(qū)挖掘是近十年最前沿的多學(xué)科交叉研究熱點(diǎn)之一，已被廣泛應(yīng)用于恐怖組織識(shí)別、蛋白質(zhì)功能預(yù)測(cè)、搜索引擎等諸多領(lǐng)域。本文基于蟻群算法、遺傳算法、馬爾科夫動(dòng)力學(xué)方法，對(duì)社區(qū)挖掘問(wèn)題進(jìn)行研究。 提出基于蟻群算法的社區(qū)挖掘方法RWACO。它結(jié)合了馬爾科夫隨機(jī)游走模型及集成學(xué)習(xí)思想，通過(guò)“強(qiáng)化社區(qū)內(nèi)連接、弱化社區(qū)間連接”這一進(jìn)化策略使社區(qū)結(jié)構(gòu)逐漸呈現(xiàn)。實(shí)驗(yàn)表明，RWACO較一些代表性算法具有更高的聚類(lèi)精度。 提出基于遺傳算法的社區(qū)挖掘方法GALS。它采用了基于圖的編碼策略L(fǎng)AR，以模塊性函數(shù)Q作為目標(biāo)函數(shù)。針對(duì)傳統(tǒng)變異方法之不足，，我們面向LAR編碼給出邊緣基因的概念；推導(dǎo)出模塊性函數(shù)Q之局部單調(diào)性；在上述兩點(diǎn)的基礎(chǔ)上提出了一個(gè)快速有效的局部搜索變異算法。在人工網(wǎng)絡(luò)及真實(shí)網(wǎng)絡(luò)上進(jìn)行測(cè)試，并與當(dāng)前代表性算法進(jìn)行比較，實(shí)驗(yàn)表明了GALS的有效性。 提出基于馬爾科夫動(dòng)力學(xué)的重疊社區(qū)挖掘算法UEOC。首先將原始網(wǎng)絡(luò)與相應(yīng)的退火網(wǎng)絡(luò)融合為一個(gè)集成網(wǎng)絡(luò)，在集成網(wǎng)絡(luò)上給出一個(gè)基于約束的馬爾科夫動(dòng)力學(xué)新模型，以逐步呈現(xiàn)網(wǎng)絡(luò)中的每個(gè)社區(qū)。然后基于局部社區(qū)函數(shù)“導(dǎo)電率”，設(shè)計(jì)一個(gè)有效的截方法，將已呈現(xiàn)出的社區(qū)抽取出來(lái)。如果網(wǎng)絡(luò)具有重疊結(jié)構(gòu)，被抽取出的社區(qū)則天然呈現(xiàn)重疊現(xiàn)象。實(shí)驗(yàn)表明，UEOC可快速有效的發(fā)現(xiàn)重疊社區(qū)結(jié)構(gòu)。
[Abstract]:The mining of complex online communities is one of the most advanced interdisciplinary research hotspots in the past decade. It has been widely used in terrorist tissue identification, protein function prediction, search engine and many other fields. Based on ant colony algorithm, genetic algorithm and Markov dynamics, community mining problem is studied in this paper. A community mining method based on ant colony algorithm (RWACO) is proposed. It combines Markov random walk model with the idea of integrated learning, and makes the community structure appear gradually through the evolutionary strategy of "strengthen the connection within the community and weaken the connection between the communities". Experiments show that RWACO has higher clustering accuracy than some representative algorithms. A community mining method based on genetic algorithm (GALS-based) is proposed. It adopts the graph-based coding strategy LAR and takes the modular function Q as the objective function. In view of the shortcomings of traditional mutation methods, we give the concept of edge gene for LAR coding, deduce the local monotonicity of modular function Q, and propose a fast and effective local search mutation algorithm based on the above two points. The experiments on artificial network and real network show the effectiveness of GALS. An overlapping community mining algorithm based on Markov dynamics is proposed. Firstly, the original network and the corresponding annealing network are merged into an integrated network, and a new constrained Markov dynamics model is presented in the integrated network to gradually present each community in the network. Then, based on the local community function "conductivity", an effective truncation method is designed to extract the existing community. If the network has overlapping structure, the extracted community is naturally overlapped. Experiments show that UUOC can quickly and effectively find overlapping community structures.
【學(xué)位授予單位】：吉林大學(xué)
【學(xué)位級(jí)別】：博士
【學(xué)位授予年份】：2012
【分類(lèi)號(hào)】：TP311.13;O157.5

【參考文獻(xiàn)】

相關(guān)期刊論文 前6條

1 金弟;劉大有;楊博;劉杰;何東曉;田野;;基于局部探測(cè)的快速?gòu)?fù)雜網(wǎng)絡(luò)聚類(lèi)算法[J];電子學(xué)報(bào);2011年11期

2 何東曉;周栩;王佐;周春光;王U

本文編號(hào)：1873363