Scientific Library Lookup Assistant system for PuBMed
LiBLA is a tool designed to assist users find articles of interest more objectively by analyzing commonalities in the abstracts of registered articles.
LiBLA objectively replaces the need for you to manually review papers, find commonalities, and perform repeated PubMed searches, generating weekly recommendations.
System configuration and processing pipeline
LiBLA is composed of a section for creating keywords, a section for suggesting recommended papers every week, a section for improving the accuracy of the recommended papers, and a section for managing papers.
Creating keywords using strong ontology dictionary
The keywords are created using an ontology dictionary and proprietary text mining technology.
The ontology dictionary consists of seven categories and contains 1.3 million genes and approximately 220,000 diseases, phenotypes, tissues, biological species, functions, compounds, and their synonyms, and is optimized for analyzing papers on basic research such as genomics, microorganisms, viruses, and drug discovery.
The combination of text mining and ontology dictionary is the strongest
Using these ontology dictionaries and synonym terms, text mining technology is used to accurately detect keywords, and the system is adjusted to select navigation keywords with high commonality.
LibLA is a fully white-box in-house AI model
based on TF-IDF.
In the section that suggests recommended papers each week, text mining is performed on the XML data of PubMed Abstract downloaded every week, and papers with high commonality are selected based on the TD-IDF value and Euclidean distance of each keyword.
LiBLA learning is a white box
In order to accumulate registered papers and improve accuracy, LiBLA has a function that allows users to "like" recommended papers. Papers that are "liked" are added to the papers used to generate keywords, further improving the accuracy of the keywords.
Independent management for each folder
In the paper management section, folders are created for each topic to manage papers and keywords. Adding comments to registered and recommended papers makes it easier to organize papers.
