Toward Explainable Trading Analytics Features: A Systematic Literature Review and Proposed AI-Augmented Design for Broker Flow Data in Mobile Stock Trading Applications

This paper presents a systematic literature review (SLR) examining the application of explainable AI (XAI), recommender systems, and large language models (LLMs) to trading analytics, with a specific focus on broker transaction disclosure data ("broker flow") in the Indonesian capital market context. Following PRISMA and Kitchenham and Charters guidelines, the review searched seven academic databases (IEEE Xplore, Science Direct, Emerald Insight, Springer Nature, Wiley Online Library, ACM Digital Library, and Taylor & Francis), identifying 103 records, of which 26 studies were selected for full analysis. The review synthesizes seven categories of trading analytics features from the selected literature: explainable visualization, explainable recommender systems, reliable LLM architectures, hybrid AI with human oversight, bias and sentiment detection, prediction models without explainability, and regulatory-driven XAI. A feature-to-literature matching analysis reveals a significant gap: despite the commercial availability of broker flow features in Indonesian securities applications (Stockbit, IPOT, Ajaib, and the third-party platform NeoBDM), none of the 26 selected studies apply XAI, recommender systems, or LLM techniques specifically to broker transaction or institutional flow data. Building on this identified gap, the study proposes an Explainable Broker Flow Assistant, a feature design integrating four components: (1) a visual explainability layer extending existing broker flow charts with Grad-CAM-style highlighting, (2) a grounded LLM query interface based on a three-layer Prevention-Detection-Mitigation architecture to mitigate hallucination risk, (3) granular investor classification distinguishing institutional, retail, and inactive brokers, and (4) a behavioral bias warning system to flag trading patterns driven by social sentiment rather than underlying broker flow signals. This study contributes a technically grounded taxonomy of trading analytics features and provides empirical evidence of an academic-industry gap specific to broker flow analytics in an emerging-market trading context, offering both theoretical grounding and a practical feature design pathway for future implementation and evaluation. Keywords: explainable AI, trading analytics, recommender systems, broker flow, systematic literature review, financial technology, mobile stock trading