Human-Centered Financial Signal Processing: A Case Study on Stock Chart Analysis

In this paper, we explore the “human-centered” financial model. To illustrate this idea, we conducted a case study on the stock chart, referring to stock price chart plus stock volume chart. We first construct the stock chart with professional stock traders’ visual attention (SPSTV) dataset, which contains 150 stock charts images associated with eye-movement data from 10 professional stock traders. Based on the SPSTV dataset, the transfer learning and human attention inspired morphological operation are leveraged to develop the stock chart attention model. In validation experiments, compared to other models, SamVgg optimized by transfer learning and human visual attention function performs best with the AUC_Judd, CC, SIM, and NSS of 96.11%, 82.74%, 69.84%, and 2.84, respectively. Through visual comparative analysis, we can find that the visual attention map area after the double optimization strategy is more focused overall and has less excess attention at the edges. This visual optimization will enhance people’s observation experience. The proposed model has great potential for two application scenarios: (1) instruct amateur traders how to observe stock charts; and (2) evaluate stock analysis ability of investors. In the future, we will continue to iterate the model and try to apply it in real economic activities to generate benefits.