應用混合式隨機漫步模型在以社群為主的推薦上

Abstract

Random Walk的觀念近年來已經廣泛地被運用在推薦系統的設計上。相對於collaborative filtering，Random Walk在計算locality的問題以及處理額外資訊表現更佳，例如項目間的額外資訊。 然而傳統的random walk為基礎的方法在處理兩種不同方向的意見有很大的限制。換句話說，同時運用random walk來傳播正面和負面意見並不是有效的。在此我們提出一個正式並有效率的random walk為基礎的方法來解決這個問題，並且有理論上的證明來確保這個過程將會收斂。 本論文第二個貢獻是我們認為一個好的解釋系統不單只有提供解釋並且能合理地解釋系統的決定。因此我們提出一個方法以追蹤影響力的路徑及子圖的方法來產生解釋。 我們將我們的方法在MovieLens以及Netflix上進行實驗，實驗顯示我們的方法贏過傳統的random walk以及其他衍生的方法，並且能與其他架構的方法互相比較，像是CF，supervised learning methods.

The concept of random walk (RW) has been widely applied in the design of recommendation systems. Compared to the popular collaborative-filtering approach, RW-based approaches are more effective in handling the locality problem and taking extra information, such as the relationships between items or users, into consideration. However, the traditional RW-based approach has a serious limitation in handling bi-directional opinions. In other words, the propagation of positive and negative information simultaneously in a graph is non-trivial using random walk. To address the problem, this thesis presents a novel and efficient RW-based model that can handle both positive and negative comments with the theoretical guarantee of convergence. Furthermore, we argue that a good recommendation system should provide users not only a list of recommended items but also reasonable explanations for the decisions. Therefore, we propose a technique that generates explanations by back-tracking the influential paths and subgraphs. The results of experiments on the MovieLens and Netflix datasets show that our model outperforms state-of-the-art RW-based algorithms, and is comparable to that of other types of methods such as the CF and supervised learning methods.