請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/61591
標題: | 異質性社群網路之擴增性意涵導向個體檢索 Semantic-Driven Explorative Entity Search in Heterogeneous Social Networks |
作者: | Perng-Hwa Kung 龔鵬驊 |
指導教授: | 林守德(Shou-De Lin) |
關鍵字: | 社群網路分析,異質性網路,個體檢索,擴增性取樣, Social Network Analysis,Heterogeneous Network,Entity Search,Sample by Exploration, |
出版年 : | 2013 |
學位: | 碩士 |
摘要: | 由於網際網路的社群網路服務崛起,異質性社群網路分析近年來在產業界及學界得到高度的關注。此類網路涵括Twitter, Facebook, 甚或是資料網路如Google的Knowledge Graph等。這些社群網路通常具有資料量大、及網路中個體有獨特的身分或特殊的互動狀況等特性。亦即,網路中的點(個體)及邊(互動狀況)代表特定的種類。在設計特定的應用中,要有效地從這些資料中挖掘出特定的隱藏資訊,工程師通常需要利用像網路爬蟲的技巧一筆一筆從伺服器中讀取資訊。此類型的應用的一大挑戰在於,使用者通常無法直接擷取所有網路的相關資訊。在受權限限制的環境下,如何有效在最小成本中抓取型態相似的個體成為非常重要的課題。
本篇研究主要探討並定義在異質性網路中的擴增性意涵導向個體檢索:給定意涵路徑(meta-path)跟量度方式,將一步步從完整的網路中抓取相似的個體。此研究提出一個一般化的解決方案,在每個取樣的步驟時,賦予感興趣的點一個意涵路徑相似度的期望值,並進行分數權衡過的取樣。此外,將提出MetaRank,一個網頁排名的變形,進行期望值的估算。實驗結果於數個現實生活中的社群網路顯示提出的方法可有效在有限的搜索成本中估算意涵路徑度量的相似度。實驗亦將探討不同參數的設定。最後,使用附屬屬性來輔助預測模型的建立及擴增性檢索的準確度驗證此方法的延伸性。 Heterogeneous social network has seen a rapid rise in research and industry interest in the widely popularizing online social or information networks, such as Twitter, Facebook, or Google Knowledge Graph. Such networks are characterized by large-scale of data volume, and the varying multitude of roles that an individual (or entity) plays and interacts with other members of the network. Oftentimes, engineers that design applications to exploit the wealth of information hidden within the networks need to extract parts of the network and semantically similar entities to the target of interest via techniques such as crawling. This process faces the challenge that one very frequently does not have the permission to access the fully observed graph for network services at large. This study defines and examines the problem of exploratively searching semantically related nodes in heterogeneous social network, under the context of specific meta-path semantics dictated by the graph schema of the network. In particular, the paper proposes a framework to sequentially crawl entities from the full network, where at each stage, the process calculates the expected scores for the candidate nodes using metrics that measure meta-path similarity. Moreover, we propose score propagation heuristics to facilitate the estimation of such expected scores. Experiments on several real world networks reveal that the proposed methods can estimate meta-path semantic metrics using little exploration costs across various meta-path semantics. In addition, effects on different parameter settings are tested. Lastly, the study explores applying sampled nodes to reflect ability to identify group membership and train ranking models on attributes alone for metric score prediction. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/61591 |
全文授權: | 有償授權 |
顯示於系所單位: | 資訊工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-102-1.pdf 目前未授權公開取用 | 2.47 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。