NN-tree: A technology for accelerating the query process of a high-dimensional data tree index structure by using a neural network
Li Yu, Lingli Li · 2021
The k-nearest neighbors (k NN) search is widely used in queries. There are many fields used in the practical world, such as image matching, voice matching, and information retrieval. The tree index structure is a well-known and efficient method, but a significant disadvantage of this method is that it is difficult to balance query time, index space, and query accuracy. Recently, researchers have proposed using graph structures and neural networks to solve the k NN search problem. However, a large amount of space is required to store the graph, which makes it difficult to use graph-based methods in the real world. In this article, we combine the traditional tree-based index structure with neural network technology and propose a new index mechanism NN-tree. First, NN-tree is a tree structure that requires less space. Second, it can optimize the index structure based on past query history to improve query accuracy. Third, NN-tree can use neural networks to replace the tree index structure and use the parallel computing capabilities of the device to speed up the query process. Finally, we propose a discriminator that can adjust the query speed and accuracy. Experiments on ten diverse datasets show that our method has strong versatility and can be compared with state-of-the-art methods in terms of performance.