Auto Bracketing
2019-03-21
Auto-bracketing is not a good idea because if you generate a lot of low-quality links, they will be suggested only by garbage and "making a house into a garbage dump" will occur. It is true that keyword extraction on each page and making it into a link would be a disaster. But that is a problem of "low quality," not of automation. What is quality of links?
Consider a bad example
For example, let's say each page of a book is a Scrapbox page.
If a certain keyword X appears 100 times, it's not appropriate to make it a link all the time.
https://gyazo.com/04ed3f1e6635b34079512d12170502d5
If you see a hundred links, it's just "Wow, that's a lot of links.
Then it would still be better to present only the information, "You will find that keyword in book A.
This is Examples of unexpected discoveries in searches.
A keyword is unexpectedly presented as being included in a certain book and feels valuable
You can open the book and search again to see which specific page in the book it is on.
There is no straightforward way to achieve this on Scrapbox.
For example, if each page has a "book ID (name or ISBN)", you can use the ID and a set of keywords to narrow down the search within a book.
If we try to achieve this with a link, we would need a link that expresses "the occurrence of keyword X in book A".
Narrow links are useful
In other words, "links that appear infrequently in other text" are useful
On the other hand, my most linked in Scrapbox is the KJ method and the Engineer's Intellectual Production Method.
Words related to the content of interest are naturally frequent.
The dilemma of choosing something that simply appears infrequently is to choose something that is "uninteresting".
Similar concept to [IDF
If all links are equivalent, then it is a function of IDF
In fact, the links are not equivalent.
The same "five times" but these are different in usefulness.
https://gyazo.com/0a94b11c8751580c06fe7914c76756f8
Links between distant objects are beneficial.
So there is "near" and "far" as a relationship between pages
How is this defined?
Page proximity
What is Page proximity?
Pages belonging to the same chapter are close pages
This should give the chapter structure as metadata
Pages belonging to the same author's book are close pages
Book > Chapter > Section ... Hierarchical Structure
When the lower structure contains keywords, the upper structure also contains keywords.
This has a pooling feel to it.
Hierarchical structure is a good fit for Given
A keyword that appears five times in the same book is different from a keyword that appears five times in the same chapter or five times across chapters.
pooling Then the former would be one time and the latter five times.
Higher IDFs in higher tiers are more valuable.
What if the hierarchical structure is not Given?
Assuming a book, adjacent pages are close pages
Pages with similar content (= many keywords in common) are close pages
Hierarchical structures can be created by Agglomerative Hierarchical Clustering.
https://gyazo.com/b60143d3354814ea817ad015c51b3cc9
Proximity is determined by search Can't you?
nodal point of thought 2019-02-18#5c6a9ccfaff09e00004ee473 as it relates to
Bonus if the link already exists in Scrapbox
I want it to be able to run incrementally.
I hope it can be improved incrementally.
Nodes of Thought2019-03-21
---
This page is auto-translated from /nishio/自動ブラケティング. 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.