透過一致矩陣分解跨富含內容之社群網路進行目標使用者預測

Target Node Prediction across Content-Rich Social Networks via Consistent Incidence Co-Factorization

簡浩恆(Hao-Heng Chien)

指導教授 ： 吳尚鴻

國立清華大學/電機資訊學院/資訊工程學系/碩士(2013年)

https://doi.org/10.6843/NTHU.2013.00679

摘要

隨著社群服務的成長，愈來愈多公司擁有多個社群網路，其中有一個重要的問題為「目標使用者預測」，目標是找到現有網路中最有可能加入另一個網路的使用者，如此廣告預算能夠更精確的置於這些使用者上面。
雖然這個問題能用跨領域學習來解，但在現實的情況中卻會因為下列原因導致效果並不好，首先是邊上的內容很常見且包含很有價值的資訊，但現有的作法並沒有將其納入考量，第二點是兩個社群網路可能基於不同原因形成而且各自發展，因此造成彼此之間的異質，導致現有的跨領域學習演算法不能成功傳遞資訊。
在論文中，我們提出「一致矩陣分解」來做到資訊的傳遞，「一致矩陣分解」考量了邊上的資訊來找到較有意義的隱藏因素，且利用找到行為一致的使用者處理異質性的問題，在實驗部分，我們同時展示「一致矩陣分解」在直接進行預測以及配合已存在的分類器時，都能很好的解決目標使用者預測的問題。

關鍵字

使用者預測 ； 社群網路

並列摘要

With the growth of social services, more and more social networks are owned by the same company. An important problem, called the target node prediction problem, to the owner of multiple social networks is to identify those users from one network (called the source network) who are likely to join another (called the target network) to become the target users, so that the advertisements can be placed more precisely and economically.
Although this problem can be solved using existing techniques in the field of cross domain learning, we observe that in many real-world situations the cross-domain classifiers perform sub-optimally due to the following reasons. First, most of the existing works do not take into account the contents of edges, which are common in practice and encode valuable information. Second, since the target and source networks may be formed by different reasons and evolve distinctively, they may be heterogeneous, preventing the existing cross domain classifiers from transferring the knowledge from the target network to source network.
In this paper, we propose the Consistent Incidence Co-Factorization (CICF) that helps the knowledge transfer between social networks. The CICF uses the edge-specific information to find better latent factors, and copes with the heterogeneity by transferring knowledge only through those common users that behave consistently in the two networks. Extensive simulations are conducted and the results demonstrate the effectiveness of CICF, either when it is used directly to predict the target users or indirectly as a preprocessing tool for existing classifiers.

並列關鍵字

node prediction ； social network

參考文獻

[7] Jingrui He, Yan Liu, and Richard Lawrence. Graph-based transfer learning. In Proc. of CIKM, pages 937–946, 2009.

[3] A. Blum and S. Chawla. Learning from labeled and unlabeled data using graph mincuts. In Proc. of ICML, 2001.

[4] Andrzej Cichocki and Anh-Huy Phan. Fast local algorithms for large scale nonnegative matrix and tensor factorizations. IEICE

Trans on Fundamentals of Electronics, Communications and Computer Science, 92(3):708–721, 2009.

volume 12, pages 210–219, 2007.