Eigenexpression Extraction and Keyphrase Extraction

• [series labeling
  • Eigen-Expression Extraction labels the “beginning of the eigenexpression” and the “range of eigenexpressions”.
  • The RAKE stop list generation algorithm in keyphrase extraction counts the number of times a word is “in” or “next to” a keyphrase.
  • Mapping to series labeling would be labeled “keyphrase range” and “keyphrase adjacency”.
  • Personally, I think it would be better to distinguish between “right neighbor” and “left neighbor”.

Which labeling is better?

• As for keyphrase extraction, I think it’s more straightforward to label around the latter keywords.
• What do humans do when they explicitly state that a sentence is indistinguishable from a ground sentence and “it’s a keyword”?
  • For example, enclose in brackets
  • Conversely, even if other conditions are the same, the probability of a key phrase being a key phrase is naturally increased in the area enclosed by brackets.
  • With the latter labeling, the ” ” token naturally corresponds to “the label to the left of the keyword.
• What is the reason why eigenexpression extraction often uses the former labeling?
  • Eigenexpression may be contiguous.
  • The method of labeling the perimeter is impossible.

PS

• I was comparing 2-1 and 3, but there are more detailed steps.

• 4 can identify consecutive keywords.

• 5 can distinguish “words that do not appear at the end of keywords but often appear within keywords” such as “of

• end labeling

---

This page is auto-translated from /nishio/固有表現抽出とキーフレーズ抽出 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.