Query Suggestion Using Anchor Text

Many query suggestion techniques have been proposed to better capture user intent and to improve search effectiveness. However, most of these methods make use of query logs which are not readily available to the research community. Anchor text, on the other hand, is widely available and has proven useful for many tasks. In this paper, we investigate the problem of query suggestion via random walk and demonstrate that anchor text can be an excellent substitute for a query log. In particular, our results suggest that anchor text is as effective as a real log for this technique and it even outperforms the real log in some cases. In addition, we propose a simple but efficient implementation of the random walk procedure under the MapReduce framework. Each iteration of the walk can be done in a matter of minutes on graphs with million of vertices.