Combining query ambiguity and query-URL strength for log-based query suggestion

Abstract

The ambiguity of query may have potential impact on the perfor-mance of Query Suggestion. For getting better candidates adapting to query’s ambiguity, we propose an efficient log-based Query Suggestion method. Firstly we construct a Query-URL graph from logs and calculate the bidirectional transition probabilities between queries and URLs. Then, by taking URL’s rank and order into consideration, we make a strength metric of the Query-URL edge. Besides, we conduct random walk with the edge strength and transition prob-ability to measure the closeness among queries. To reflect the influence of query ambiguity, we exploit an entropy-based method to calculate the entropy of each query as a quantitative indicator for ambiguity, making a notion of ambiguity similarity as an available factor in relevance estimation. Finally we incorporate ambiguity similarity with closeness to derive a comprehensive relevance mea-surement. Experimental results show that our approach can achieve a good effect.