From Music Information Retrieval to Stock Market Analysis: Theoretical Discussion on Feature Extraction Transfer

Finding similar objects and patterns based on the similarity score is one of the fundamental and useful tasks in Data Mining. Different applications may introduce different features that need to be extract and analyzed. However, if two applications share some similar core concepts, it is possible to transfer some features that will be beneficial to transfer learning. Thus, this paper uses some features from Music Information Retrieval and Stock Market Analysis to theoretically illustrate the possibility of Feature Extraction Transfer. We use a 3-tuple or 6-tuple vector to record the music fundamental melody whereas a 5-tuple vector to record the daily behavior of the stock market from the candlestick chart. Hence, the flow of one music melody and the flow of one stock market can be treat as a time series vector sequence. Using this linkage, we have computed some feature exaction from Music Information Retrieval onto Stock Market Analysis and obtained some positive results. For example, the similarity between Activision Blizzard Inc and Zynga Inc have achieved a similarity score of 0.6250. Moreover, these positive results gave some ideas on implementing a self-supervised learning based system to manage your stock market and the potential of implementing a transfer learning between these two applications.