Version 1.4

Released on Nov 3, 2025

• Improved clustering for topics by incorporating paper similarity based on LLM-embeddings into the bibliographic graph and joint embeddings computation
• Reuse the embeddings DB for fetching text embeddings
• Use parquet format for Faiss index storage

Version 1.3

Released on June 1, 2025

• Reworked text embeddings usage for global similarity network construction
• Supported Sentence Transformer models for whole-text embeddings
• Implemented Questions for the paper feature using improved embeddings
• Preview of Semantic search built with Sentence Transformer for embeddings, Postgresql with vector extension for embeddings storage, and Faiss for fast embedding lookups

Version 1.2

Released on April 3, 2025

Changes:

• Improved lookup for frequent words, e.g. cell, human, cancer, etc.
• Introduced analysis for group of papers

Version 1.1

Released on March 14, 2025

Changes:

• Use weighted sum of graph and text embeddings instead of concatenation
• Improved graph visualization - limit the number of shown connected papers
• Updated databases
• Updated libraries
• Docker base image updated to Ubuntu 24.04 LTS

Version 1.0

Changes:

• All search results are saved permanently
• Use different sparse graphs for embeddings and visualization
• Use kNN induced graph to create sparse graph
• Tweaked analysis parameters for better clusters extraction
• Updated libraries

Version 0.21

Version 0.21

Released on Jan 25, 2022

• Updated PubMed database to the version 2022
• Updated pubtrends.net config to use nginx
• Improved single paper analysis - pay more attention to paper own keywords/citations while expanding by references
• Show papers graph for single-paper analysis
• Fixed issues with missing references for fresh papers