Result integrating is a key component for keyword querying across heterogeneous databases. Once the results from various search engines are collected, the search engine merges them into a single ranked list. In this paper, firstly, we present a novel model of searching, which the database is an undirected graph and results are modeled as answer trees that match individual keywords in the query. Secondly, we propose the query translator and algorithm to process most queries with complicate Boolean semantics including AND, OR, NOT and their combinations. Thirdly, we propose a multi-level caching method, which integrated results are assigned to store cache to form top-k sorting, and the final results are generated. Experiments show that the proposed integrating strategy provides significant performance benefits.