Improvement of Automatic Extraction of Inventive Information with Patent Claims Structure Recognition

Résumé

Our recent research finding produces methods for automatic extraction of inventive information out of patents thanks to the use NLP; notably the automatic text processing. However, these methods have drawbacks due to a high amount of noise (duplicates, errors) in the output result that prevent the further use of TRIZ methodology. In the mean-time, we observed that patent claims are the most important source for inventive information. These text paragraphs have nevertheless a dual nature (combining legal and technical vocabulary) and this nature engender part of the observed noise. We postulate that taking into consideration claims hierarchical structure and its structural information can reduce the time for extraction and refine the final output quality, which is the principal aim of the paper. In this paper, we report on the methodology we have employed based on the patent claim structure recognition as a way to address our objectives.

BibTeX

@inproceedings{berdyugina:hal-04247035,
  TITLE = {{Improvement of Automatic Extraction of Inventive Information with Patent Claims Structure Recognition}},
  AUTHOR = {Berdyugina, Daria and Cavallucci, Denis},
  URL = {https://hal.science/hal-04247035},
  BOOKTITLE = {{Intelligent Computing. SAI 2020}},
  ADDRESS = {en ligne, France},
  PUBLISHER = {{Springer Verlag (Germany)}},
  SERIES = {Advances in Intelligent Systems and Computing},
  VOLUME = {1229},
  PAGES = {625-637},
  YEAR = {2020},
  MONTH = Jul,
  DOI = {10.1007/978-3-030-52246-9\_46},
  KEYWORDS = {Natural Language Processing (NLP) ; Text mining ; TRIZ},
  HAL_ID = {hal-04247035},
  HAL_VERSION = {v1},
}