Overview of Nearest Neighbor Subtree Search Methods

In many scientific areas there is a frequent need to extract a common pattern from multiple data. In most cases, however, an approximate but low cost solution is preferred to a high cost exact match. To establish a fast search engine an efficient heuristic method should be implemented. Our investigation is devoted to the approximate nearest neighbor search (ANN) for unordered labeled trees. The proposed modified best-first algorithm provides a O((Nq+Nb)·M + K·Nq·Nb/M) cost function with simple implementation details. According to our test results, realized with smaller trees where the brute-force algorithm could be tested, the yielded results are a good approximation of the global optimum values. Based on the results of the tests, the execution cost for the base best-first algorithm is about one order of magnitude larger than the cost for the porposed modified best-first approximation method.