Similarity-Preserving Hashing for Stock Analysis

In this paper, we present an efficient hash-based approach to address the problem of similarity search in stock market databases. Hashing is one of the standard indexing search schemes but the number of entries in the same location can become very large and make it unmanageable. Different from the standard hashing algorithm that avoids numerous collisions, the similarity-preserving hashing provides to maximize the probability of collision of close items and minimize the probability of collision of faraway items. An efficient k-nearest neighbors search based on the similarity-preserving hashing is also proposed. The experimental results based on the SET (The Stock Exchange of Thailand) and DJIA (Dow Jones Industrial Average) databases have shown that the average precisions of an efficient hash-based are greater than or equal to 0.80. Finally, we apply the similarity-preserving hashing for speculating the stock trading signal trajectory for the last segment query. The suggestion uses the frequency of occurrence for the query in the historical data.