Real-estate recommendation research has fallen behind modern AI: only 15% of studies use deep learning and none adopt state-of-the-art explainability or transfer-learning tools. Eighty percent rely on proprietary data, most work targets Asian residential markets, and fairness audits are rare—prompting authors to propose a trust-focused evaluation framework to steer future systems.

Main Finding

A PRISMA-compliant systematic review of 59 peer‑reviewed studies (2005–2025) finds that real estate recommendation systems (RERSs) lag substantially behind current AI best practices. Adoption of deep learning and advanced methods for interpretability, transfer learning, and fairness is limited, while research practices exhibit strong data and geographic biases. The authors propose a trust-based evaluation (T‑EVAL) framework and a nine‑layer conceptual architecture to guide the development of more transparent, fair, and data‑efficient RERS.

Key Points

• Scope: 59 peer‑reviewed studies published 2005–2025; systematic PRISMA review of algorithms, data modalities, evaluation practices, and ethical considerations.
• Low adoption of state‑of‑the‑art AI:
  • Deep learning used in 15% of studies.
  • No reviewed work implements state‑of‑the‑art post‑hoc explainable AI (XAI) or transfer learning (TL) frameworks.
• Dataset and domain biases:
  • 80% reliance on proprietary datasets.
  • 56% of studies concentrated in Asia.
  • 91% focus on residential properties.
• Evaluation and ethics gaps:
  • Limited fairness auditing despite known discrimination risks in housing markets.
  • Evaluation practices emphasize predictive performance over interpretability, user trust, and market‑level outcomes.
• Proposed solutions:
  • T‑EVAL — a trust‑based evaluation integrating predictive accuracy, user trust, fairness, and market efficiency.
  • A nine‑layer conceptual architecture for transparent, ethical, and data‑efficient next‑generation RERS.
• Contribution: Establishes an empirical benchmark of technology adoption gaps and outlines a research agenda for responsible AI in real estate decision support.

Data & Methods

• Methodology: PRISMA‑compliant systematic literature review.
• Corpus: 59 peer‑reviewed articles spanning 2005–2025 selected and coded for:
  • Algorithmic approaches (e.g., classical ML, deep learning),
  • Use of XAI and transfer learning methods,
  • Data modalities and provenance (proprietary vs. open),
  • Geographic and property‑type coverage,
  • Evaluation metrics and ethical analyses (fairness, transparency, user studies).
• Synthesis: Quantitative summary of adoption rates (e.g., % using deep learning, % proprietary datasets) and qualitative synthesis identifying methodological and normative gaps.
• Limitations noted by the authors (implicit): reliance on published literature may undercount proprietary or industry deployments; review limited to peer‑reviewed sources through 2025.

Implications for AI Economics

• Information frictions and market efficiency:
  • Limited use of XAI and market‑level evaluation risks suboptimal price discovery and persistent search frictions; improvements could increase allocative efficiency in housing markets.
  • T‑EVAL’s inclusion of market efficiency as an objective aligns algorithm design with welfare‑oriented outcomes (e.g., lower search costs, timelier matches).
• Competition and platform dynamics:
  • Proprietary datasets and opaque models may create entry barriers and concentration effects among dominant platforms; openness and transfer learning could lower switching costs and promote competition.
• Distributional effects and fairness:
  • Sparse fairness auditing increases the risk of algorithmic reinforcement of segregation, biased price recommendations, or discriminatory steering—outcomes with clear welfare and equity implications.
  • Incorporating fairness metrics into RERS evaluation is necessary to assess redistributive and exclusionary consequences.
• Data scarcity and productivity:
  • Low adoption of TL and data‑efficient methods suggests missed opportunities to scale recommendations across geographies and submarkets, limiting productivity gains from AI in thin markets.
  • Economically valuable techniques to consider: transfer learning, few‑shot learning, federated learning, and synthetic data—each with trade‑offs for privacy, generalization, and external validity.
• Policy and regulation:
  • Findings motivate policy interventions: standards for transparency, mandatory fairness audits for marketplace algorithms, data‑sharing incentives, and benchmarks for market‑level impact assessment.
  • Regulators and platform designers should consider requiring multi‑dimensional evaluation (accuracy, trust, fairness, market outcomes) akin to T‑EVAL.
• Research agenda for AI economics:
  • Develop open, geographically diverse benchmarks and datasets to reduce proprietary‑data bias.
  • Perform causal and field experiments to measure RERS impacts on prices, search behavior, and market segmentation.
  • Formalize economic models linking algorithmic recommendations to market equilibrium, welfare, and distributional outcomes.
  • Evaluate trade‑offs between transparency, competition, and privacy; design incentive mechanisms that align platform objectives with social welfare.
  • Integrate XAI and TL in RERS research and measure their economic effects (e.g., on adoption, trust, and market efficiency).

Overall, the review signals that advancing RERS along XAI, TL, and fairness dimensions is not only a technical necessity but also an economic priority: doing so can improve market efficiency, reduce inequities, and shape competitive dynamics in online real estate markets.