Prismatic: Interactive Multi-View Cluster Analysis of Concept Stocks

TL;DR

Prismatic tackles the challenge of identifying concept stocks by uniting data-driven correlations with knowledge-driven relationships. It introduces a three-stage clustering framework—dynamic cluster generation, knowledge-based cluster exploration, and correlation-based cluster validation—implemented via a multi-view clustering approach over a three-layer business-relational network. The system provides four coordinated visual views (financial-correlation network, correlation matrix, knowledge graph, and Prism time series) to enable end-to-end interactive clustering and concept-stock construction. Case studies on medicine and media stocks, together with expert interviews, demonstrate Prismatic’s ability to reveal nuanced cluster dynamics and support informed investment decisions in volatile markets.

Abstract

Financial cluster analysis allows investors to discover investment alternatives and avoid undertaking excessive risks. However, this analytical task faces substantial challenges arising from many pairwise comparisons, the dynamic correlations across time spans, and the ambiguity in deriving implications from business relational knowledge. We propose Prismatic, a visual analytics system that integrates quantitative analysis of historical performance and qualitative analysis of business relational knowledge to cluster correlated businesses interactively. Prismatic features three clustering processes: dynamic cluster generation, knowledge-based cluster exploration, and correlation-based cluster validation. Utilizing a multi-view clustering approach, it enriches data-driven clusters with knowledge-driven similarity, providing a nuanced understanding of business correlations. Through well-coordinated visual views, Prismatic facilitates a comprehensive interpretation of intertwined quantitative and qualitative features, demonstrating its usefulness and effectiveness via case studies on formulating concept stocks and extensive interviews with domain experts.

Figures (7)

• Figure 1: A simplified example illustrating different financial clusters. Sectors and industries are related hierarchically, while concept stocks can be constructed arbitrarily to label specific business relational knowledge.
• Figure 2: Overview of the Prismatic framework. The example follows from \ref{['fig:business-taxonomy']}. The three core modules of Prismatic enhance user exploration of concept stocks through a visual interface. (A) The data-driven perspective constructs a financial correlation network from financial time series to support multi-scale exploration of correlations. This involves (A1) checking for monotonic price co-movements, (A2) filtering out weak correlations to focus on significant relationships, and (A3) identifying key companies using betweenness centrality. (B) The knowledge-driven perspective constructs a multi-layer network from business relational knowledge and produces multi-view clusters. This network contains factual information about the business relationships, while the clusters assess business proximity through various business domain perspectives. (C) Prismatic interface offers an integrated workflow for interactive visual cluster analysis with cluster generation, summary, exploration, and validation, facilitating a seamless integration of data-driven and knowledge-driven insights.
• Figure 3: The visual interface of Prismatic. (A) provides the model configuration and illustrates the dynamic structure of financial correlation. (B) visualizes the data-driven community and presents a highly correlated cluster after validation and exploration. (C) features the Prism view with interaction making use of the prism metaphor. (D) incorporates business relational knowledge into the system to support cluster generation.
• Figure 4: The alternative design for the knowledge graph view: (A) Sankey diagram and (B) circular dendrogram.
• Figure 5: The Prism view design summarizes statistical aggregates across all time scales in a heatmap, with interactive features allowing users to explore and identify key trends between two time series efficiently.