Key node identification across multiple social nets using a weighted multi-layer and ensemble based approach /
Social networks have become an essential hub of personal information and individual relationship due to their heavy reliance on these media. Analysis of their interaction can provide a wealth of information about social behavior. Multiple platforms of social networks make this problem interesting, a...
Saved in:
| 主要作者: | |
|---|---|
| 格式: | Thesis |
| 語言: | English |
| 主題: | |
| 在線閱讀: | http://studentrepo.iium.edu.my/handle/123456789/9369 |
| 標簽: |
添加標簽
沒有標簽, 成為第一個標記此記錄!
|
| 總結: | Social networks have become an essential hub of personal information and individual relationship due to their heavy reliance on these media. Analysis of their interaction can provide a wealth of information about social behavior. Multiple platforms of social networks make this problem interesting, as the information of individual social interaction is spread over multiple social media platform. The identification of important players in these real-world social networks has been a matter of great emphasis due to its effectiveness in multiple disciplines. The current research proposes to employ techniques for the analysis of multiple social networks/ relationships to evaluate their applicability in Key Individual Identification. Research deals with each social dimension as a layer in the multiple-layer network model. The modeling of networks under a multilayer umbrella allows representation of each individual from multiple relationships perspectives and multilayer visualization of these networks helps in revealing interesting information during analysis. The dissertation studies the different architectures used to analyze the multiple networks and proposes an aligned approach to analyze multiple networks and compares the proposed approach with the commonly used other approaches. Most algorithms proposed to identify key nodes emphasize a single objective of interest and thus perform well only in limited, focused domains. However, in real life domains, multiple objectives of interest should be considered. Consequently, the proposed research studies and evaluates different aspects of an individual's interaction across social networks by applying different categories of features to cover multiple different objectives of concern and formulate the global feature matrix. This global space represents the overall node behavior in all networks which participate in the decision making collectively. The framework applies the ensemble-based approach using majority voting to predict key nodes in networks. The thesis also studies feature types by evaluating the combination of different features types and compares their results to test model correctness and performance. Furthermore, in multiple social relationship networks, the strength or importance of a relationship varies and cannot be assumed equal. This research studies the role of relative importance of social dimensions to enhance the performance of frameworks developed and handles it by using optimization strategy and measure of relevance to weigh the layers according to their impact on the key node roles in the model. The model is tested and evaluated on multiple datasets from real-world and online social networks using performance parameters such as accuracy, recall, and specificity. The proposed ensemble outperformed other classifiers and existing approaches using the proposed multi-layer analysis approach. The performance improvement of our framework demonstrates that proposed technique can be effectively used for the analysis of multi-layer based networks however the current research does not capture time-based changes and requires the annotations of network nodes. |
|---|---|
| Item Description: | Abstracts in English and Arabic. |
| 實物描述: | xvi, 276 leaves : colour illustrations ; 30cm. |
| 參考書目: | Includes bibliographical references (leaves 174-188). |