METHOD AND DEVICE FOR EXTRACTING WEB-PAGES SUBJECT-MATTER Russian patent published in 2020 - IPC G06F16/36 G06F16/951 

FIELD: information technology.

SUBSTANCE: invention relates to means of retrieving web-pages subject-matter. Obtaining possible web pages and pre-built machine learning model, wherein each possible web page comprises a plurality of pre-selected possible thematic sentences, wherein each possible thematic sentence comprises several verbal segments. Values of verbal characteristics are determined, which indicate levels of importance of verbal segments in each possible web page, respectively, and inputting said verbal characteristics values into machine learning model to obtain importance value for each verbal segment. For each possible web page determining the value of partial order for each possible thematic proposal in accordance with values of importance of verbal segments contained in a possible thematic proposal. For each possible web page, selecting one of a plurality of possible subject proposals associated with the partial order value, exceeding a predetermined threshold value as a target thematic sentence of a possible web page.

EFFECT: technical result consists in improvement of accuracy of subject proposals extracted from web pages.

20 cl, 6 dwg