Synthetic crafting of training and test data for named entity recognition by utilizing a rule-based library
View Patent ↗A method and system for extracting and labeling Named-Entity Recognition (NER) data in a target language for use in a multi-lingual software module has been developed. First, a textual sentence is translated to the target language using a translation module. A named entity is identified and extracted within the translated sentence. The named entity is identified by either: exact mapping; a semantically similar translated named entity that meets a predetermined minimum threshold of similarity; or utilizing a rule-based library for the target language. Once identified, the named entity is labeled with a pre-determined category and stored in a retrievable electronic database.
1. A method for extracting and labeling Named-Entity Recognition (NER) data in a target language for use in a multi-lingual software module, comprising:
pre-processing a textual sentence to replace textual organizations and textual persons in the source language using a static list of organizations and names previously stored in a retrievable electronic database for the target language;
translating the textual sentence to the target language using an open source translation module;
identifying a named entity within the translated textual sentence by:
(i) if an exact mapping of a translated named entity is available, using the exact mapping of the translated named entity,
(ii) if the exact mapping is not available and identifying a semantically similar translated named entity that meets a predetermined minimum threshold of similarity as determined by a confidence score in embedded similarity is possible, identifying the semantically similar translated named entity that meets the pre-determined minimum threshold of similarity, and
(iii) if the exact mapping is not available and identifying the semantically similar translated named entity that meets the predetermined minimum threshold of similarity as determined by a confidence score in embedded similarity is not possible, utilizing a rule-based library for the target language;
labeling the identified named entity with a pre-determined category; and
storing the labeled named entity in a retrievable electronic database for later retrieval by the multi-lingual software module.
2. The method of claim 1 , where the textual sentence from the target language is pre-processed to include target language specific features.
3. The method of claim 2 , where the target language specific features are names of organizations.
4. The method of claim 2 , where the target language specific features are names of persons.
5. The method of claim 2 , where the target language specific features are locations.
6. The method of claim 2 , where the target language specific features are date/times.
7. The method of claim 2 , where the target language specific features are monetary amounts.
8. The method of claim 2 , where the target language specific features are numbers.
9. An apparatus comprising:
a processor;
a non-transitory machine-readable storage medium that provides instructions that, if executed by the processor, are configurable to cause the apparatus to perform operations comprising,
pre-processing a textual sentence to replace textual organizations and textual persons in the source language using a static list of organizations and names previously stored in a retrievable electronic database for the target language;
translating a textual sentence to a target language using an open source translation module;
identifying a named entity within the translated textual sentence by:
(i) if an exact mapping of the translated named entity is available, using the exact mapping of the translated named entity,
(ii) if an exact mapping is not available and identifying a semantically similar translated named entity that meets a predetermined minimum threshold of similarity as determined by a confidence score in embedded similarity is possible, identifying a semantically similar translated named entity that meets a pre-determined minimum threshold of similarity, and
(iii) if an exact mapping is not available and identifying a semantically similar translated named entity that meets a predetermined minimum threshold of similarity as determined by a confidence score in embedded similarity, is not possible, utilizing a rule-based library for the target language;
labeling the identified named entity with a pre-determined category; and
storing the labeled named entity in a retrievable electronic database for later retrieval by the multi-lingual software module.
10. The apparatus of claim 9 , where the textual sentence from the target language is pre-processed to include target language specific features.
11. The apparatus of claim 10 , where the target language specific features are names of organizations.
12. The apparatus of claim 10 , where the target language specific features are names of persons.
13. The apparatus of claim 10 , where the target language specific features are locations.
14. The apparatus of claim 10 , where the target language specific features are date/times.
15. The apparatus of claim 10 , where the target language specific features are numbers.