Methods and apparatus for creating domain-specific intended-meaning natural language processing pipelines
US11681878B2 · kind B2 · utility
Assignee
Inventors
Key dates
| Filing date | Nov 8, 2022 |
| Grant date | Jun 20, 2023 |
| Priority date | — |
| Expiry date | Nov 8, 2042 |
Classification
- Technology area (CPC G)Physics
- CPC primaryG06F40/253
- WIPO fieldComputer technology
- WIPO sectorElectrical engineering
Abstract
A method includes receiving a dataset that includes a plurality of input texts. Each input text from the plurality of texts is associated with a content category from a plurality of content categories based on a comparison between that input text and an intended meaning that is common for each comparison. For each model in a plurality of models, and for each content category from the plurality of content categories, that model is executed on each input text from the plurality of input texts to generate an average similarity/dissimilarity score for that content category. At least one model from the plurality of models is selected, based on the average similarity score for each content category from the plurality of content categories for each model in the plurality of models, to determine whether an input text is similar/dissimilar to the intended meaning.
Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.