Abstract: Web database users often use the keywords that are familiar to them for expressing their query intentions and this may lead to unsatisfactory results due to the limitation of the users’ knowledge. Providing top-k diverse and relevant queries can broaden user knowledge scope and thus can help them to formulate more efficient queries. To address this problem, this paper proposes a top-k diverse keyword query suggestion approach. It first leverages frequency of co-occurrence and correlations between different keywords in query history to measure the intra-and inter-keyword couplings. And then, a semantic matrix, which reserves the coupling relationships between keywords, is generated. Based on the semantic matrix, the semantic similarities between keyword queries can be measured by using a kernel function. To quickly provide the top-k diverse and semantically related queries, this approach first finds the typical queries from query history by using the probabilistic density estimation method. After this, it finds the representative queries from the set of typical queries and then creates the orders for each representative query according to the similarities of remaining queries in the set of typical queries to the representative query. When a new query coming, the similarities between the given query and representative queries are computed, and then the top-k diverse and semantically related queries can be selected by using threshold algorithm (TA) over the orders of representative queries. The experimental results demonstrate that both the keyword coupling relationship and query semantic similarity measuring methods can achieve the high accuracy, and the effectiveness of top-k diverse query selection method is also demonstrated.