Predicting Click Rates by Consistent Bipartite Spectral Graph Model

Search advertising click-through rate (CTR) is one of the major contributions to search ads' revenues. Predicting the CTR for new ads put a direct impact on the ads' quality. Traditional predicting methods limited to Vector Space Model fail to sufficiently consider the search ads' characteristics of heterogeneous data, and therefore have limited effect. This paper presents consistent bipartite graph model to describe ads, adopting spectral co-clustering method in data mining. In order to solve the balance partition of the map in clustering, divide-and-merge algorithm is introduced into consistent bipartite graph's co-partition, a more effective heuristic algorithm is established. Experiments on real ads dataset shows that our approach worked effectively and efficiently.