Project News

2026.10.21 🎯 Refined Journal Selection and Filtering: We have completed a major overhaul of our article selection pipeline to ensure BiblioAssistant remains sharply focused on high-quality, high-relevance research. Rather than monitoring broad journal catalogs, the system now tracks a curated selection of leading peer-reviewed journals in physical hydrology, hydrometeorology, climate science, and remote sensing. Major multidisciplinary journals (such as Nature and Science) are now tracked through targeted topic queries, while high-volume open-access mega-journals and peripheral disciplines (such as purely agronomic studies, urban micro-drainage, or plant biology) have been excluded. In parallel, our relevance filter now evaluates papers using complete journal and topic context alongside stricter scoring criteria. We have also audited the entire archive, removing over 6,000 off-topic articles to keep the catalog clean, concise, and genuinely useful.
2026.09.26 ⚙️ Pipeline Architecture Upgrade & Historical Backfill Restart: We have completed a comprehensive operational review of the BiblioAssistant processing pipeline. Candidate relevance filtering is now executed locally with Ollama (Llama 3.1 8B), eliminating external API quota limits and ensuring high-throughput screening without bottlenecks. Deep paper synthesis is now powered by Google Gemini 2.5 Flash, providing near-instant structured extraction with a 1M token context window. Because previous runs were constrained by third-party API rate limits during historical backfilling, we have reset the backfill cursor to September 2026 to systematically re-scan backwards week by week, recovering any relevant literature that was previously missed.
2026.03.17 🌊 Relevance Filter Update: We have refined our filtering criteria to better focus on terrestrial hydrology and water resources. Submarine Groundwater Discharge (SGD) and purely marine/coastal studies focused on nutrient fluxes have been explicitly excluded to ensure our selection remains strictly centered on the active water cycle and freshwater management.
2026.03.17 📊 Scientific Quality Upgrade: Integrated Journal Metrics. We have implemented a new filtering stage based on journal quality metrics (OpenAlex). The system now utilizes the 2-year Mean Citedness (Impact Factor equivalent) and h-index to automatically discard works from sources with lower academic impact (IF < 2.0 or h-index < 50, with topic-specific adjustments). As part of this upgrade, we performed a retroactive purge of 275 low-quality articles, ensuring our repository remains focused on high-impact, rigorous research.
2026.03.16 🐛 Critical Bug Fix & Backfill Restart: We identified and resolved a major structural bug in our discovery module where the OpenAlex API was truncating results for high-volume searches (like monitoring core journals such as HESS, WRR, and JHM). By implementing cursor-based pagination, we now ensure 100% retrieval of relevant literature. Due to the significant increase in correctly discovered articles, we have restarted our historical backfill process from today, moving backwards in smaller, 7-day batches to manage the new volume efficiently.
2026.03.12 🛠️ System Architecture & New Sections: Over the last two days, we significantly improved our platform's data integrity by migrating all author and journal tracking to use unique OpenAlex IDs. Thanks to this, we have launched two entirely new sections on the website: Authors and Journals, where you can explore dedicated pages and statistics for each entity. We also resolved a critical database bug that had temporarily halted the daily pipeline, successfully restoring both real-time discovery and our historical backfill process.
2026.03.10 📊 New Feature: Filter Audit Page. We have launched a dedicated Filter section in the menu to provide transparency on our automated screening process. This audit tool displays the last 7 days of processed papers, showing whether they passed our relevance filter and the specific reasoning behind the decision. To ensure full transparency, we also publish the active AI Filter Prompt used by the system. This enhancement helps us continuously refine our selection criteria for better accuracy.
2026.02.26 🔍 Efficiency upgrade: We have introduced a multi-stage pre-filtering system. New articles now pass through a 'funnel' starting with a Journal Blacklist and an OpenAlex Topic Pre-screening. Only relevant candidates proceed to the LLM analysis, significantly reducing processing time and focusing only on core hydrological and climate science.
2026.02.25 🛡️ System enhancement: We have officially transitioned to a peer-review-only policy. All preprints and non-journal sources (e.g., Zenodo, Figshare) have been excluded to ensure the highest academic quality. Meanwhile, our backfilling process continues at a steady pace, currently processing October 2025, and we continue refining our relevance filters.
2026.02.16 🎯 We are working on fine-tuning the filtering prompt. The goal is to reduce false positives from articles that lack a clear focus on water resources, hydrology, droughts, ...
2026.02.15 🔄 A backfilling script has been created and implemented. This tool will allow us to automatically retrieve and process relevant articles from previous months and years to complete the project's historical database.
2026.02.09 🚀 We officially launch the BiblioAssistant website. This portal was born with the mission to automate the monitoring of scientific literature in hydrology and climate change, facilitating the synthesis work for researchers.