Photothermal-Seek: Literature-Grounded Knowledge Graphs for Discovering High-Performance Solar Evaporators

Solar interfacial evaporation (SIE) offers a low-energy route for desalination and water purification, but high-performance evaporator design still relies heavily on experience. Here, we report Photothermal-Seek, a literature-grounded and evidence-anchored knowledge framework that connects synthesis, structure, and performance information in SIE materials. Built from 2771 publications, Photothermal-Seek integrates large language model (LLM)-assisted extraction, graph-based reasoning, and full-text retrieval of source evidence, generating a knowledge base with 57 707 nodes, 126 649 relationships, and 1535 curated material nodes. Repeated object-level audits showed that most sampled extracted objects were supported by source evidence, while benchmark evaluation demonstrated improved domain-specific reasoning relative to general-purpose language models. Literature-scale semantic analysis shows that conjugated materials had the largest representation in the corpus-defined upper evaporation-rate tertile, while multicomponent systems near the conjugated-carbonaceous/fibrous semantic boundary emerged as candidates for further investigation. Guided by this rule, we built and tested a polypyrrole-coated delignified wood evaporator with a cellulose nanofiber/poly(vinyl alcohol) capillary-regulation layer (PPy@DW/CP), which achieved 2.275 kg·m^{- 2}·h^{- 1}. A held-out Dyson sphere-like evaporator (DSE-V) case further supports performance estimation for unseen evaporator designs. These results show that literature-grounded knowledge graphs can connect reported SIE knowledge with real material design and support the early assessment of new evaporator systems.