Random Walk with Wait and Restart on Document Co-citation Network for Similar Document Search

One of the latest algorithms for computing similarities between nodes in a graph is Random Walk with Restart (RWR). However, on a document co-citation network for similar document search, computing transition probabilities remains difficult. To solve the problem, this paper proposes a Random Walk with Wait and Restart (RWWR) algorithm, which contains a new technique for adjusting the transition probability by incorporating a “selfreturning” edge into the normalization. To evaluate its effectiveness empirically, the search performance of two retrieval methods using RWWR was compared to a method using the standard RWR; the performance was measured by average precision and nDCG. The experiment was conducted on a test collection created from the Open Access Subset of PubMed Central, and the results indicated that the RWWR methods tend to outperform the standard RWR method.