Query-Based Improvement Procedure and Self-Adaptive Graph Construction Algorithm for Approximate Nearest Neighbor Search
The nearest neighbor search problem is well known since 60s. Many approaches have been proposed. One is to build a graph over the set of objects from a given database and use a greedy walk as a basis for a search algorithm. If the greedy walk has an ability to find the nearest neighbor in the graph starting from any vertex with a small number of steps, such a graph is called a navigable small world. In this paper we propose a new algorithm for building graphs with navigable small world properties. The main advantage of the proposed algorithm is that it is free from input parameters and has an ability to adapt on the fly to any changes in the distribution of data. The algorithm is based on the idea of removing local minimums by adding new edges. We realize this idea to improve search properties of the structure by using the set of queries in the execution stage. An empirical study of the proposed algorithm and comparison with previous works are reported in the paper.