Backlinking Is Not Very Useful -- Often Even Harmful
Backlinking Is Not Very Useful -- Often Even Harmful
A Zettelkasten is a personal tool for thinking and writing that creates an interconnected web of thought. Its emphasis is on connection and not mere collection of ideas. Lists of backlinks, manually made or automatically generated, don't reveal the context. That's terrible for a growing archive.
Howdy, Stranger!
Comments
I think I disagree with this piece, and I suspect that it sums up a lot of why I've been finding the zettelkasten.de approach to note-taking uncomfortably prescriptive of late.
My notes are a hypertext, and since they're a hypertext that lives in a system I control, linking within them has a richer set of affordances than allowed for by something like the web. Meaningfully bi-directional links are a big perk here.
Certainly it's a good idea to give context where possible, but automatic backlinking facilities allow me to:
grepThere's much to be said for simple formats and relatively simple tooling, but sometimes rejecting basic automation on the grounds of distraction or triviality can obscure other types of choice and context that it offers a user.
I am a Zettler
I think that software trying to help us to analyse our notes corpus is not a bad thing by itself. But, as the example with automatically generated backlinks shows, software often cannot do better, since it simply doesn't have the means to do so. I.e., there's usually no way for a machine to understand our notes.
To have the software actually provide features that are truly useful to us, software would need to be able to parse the notes. I.e., ideally, a note would be human readable as well as machine parsable.
These are very important points! A link to another note gets much more useful if we also state why we're linking to it. I.e., we should not only place a link but also strive to describe the type of the link relationship.
Using a free-text sentence to describe the context & link relationship type of a linked note is a good first step. However, this is likely only human readable and not machine parsable.
In my opinion, the next further improvement would be to offer a formalized & standard way to encode the link relationship type with the link itself. Besides many other benefits, this would also allow software to provide a much better backlinks feature. E.g., if we could mark certain note links as "supporting" (or "opposing") the thoughts in our current note, the software could group the autogenerated backlinks by link relationship type. That way, the software could visually separate links to "supporting" notes from "opposing" ones. IMO, such "qualified links" would be truly useful since they'd better preserve the knowledge aspect of a link.
For bibliographic citations, there are already tools available which support "qualified links" (aka "semantic citations"). For example, Pandoc Scholar allows to prepend a bibliographic citekey with a property that describes why you are actually citing it. The format is
@<property>:citekey. So, in case of Pandoc Markdown, such a semantic citation would look like@agrees_with:Miller+Johnson2020or@disputes:McGee2019.The used properties stem from the CiTO ontology which already offers a comprehensive set of relationship types, ready to be used. And software could offer these relationship types via autocompletion when composing a link.
Similar to semantic citations, Markdown links could incorporate a relationship type like this:
[[type:link]][[link description|type:link]][link description](type:link)Examples:
The ideas in this note are further extendend in [[is_extended_by:20201113]].These statements question the conclusion that [backlinks are useful](critiques:Why_backlinks_are_useful.md)The use of qualified links would also allow software to label connections in graph visualizations, or to filter a graph by a certain relationship type. These are just a few examples, but IMO qualified links have a huge potential to further advance what knowledge management software can do.
I plan to implement support for qualified links in my own app and I hope that other software will do the same (and that we can come up with a common syntax for it).
While I agree with the sentiment of your argument @Sascha, I don't think it is comprehensive enough.
All you wrote is certainly true for the simple backlink example you gave. However, backlinks can provide more context than that. If you look at Andy Matuschak's notes, you'll notice that he does provide more context and these are automatically generated backlinks. As an outsider, his backlinks helped me massively to explore his notes collection as they turn the directed graph of his notes into an undirected graph.
I wonder what your opinion is on these kinds of more advanced backlinks.
I think that @msteffens made some great points as regards to enriching backlinks. It also shows how the practice of good linking can be used to enhance automatic backlinking capabilities. It also shows how blurry the line can get, when semantically enhancing links in a way that is machine readable.
I furthermore think that a naked backlink can be better than no link, once in a while. I think it speaks of a growing collection of immature notes in an area. I'd say: It's a fine option, if it's not the only option, but it can increase discoverability - which might be prefered over avoiding the paradox of choice. Not all notes are perfect on the first try and so aren't links (or backlinks).
I think though that manually placed "good" backlinks enhance your engagement with what one is working on - but that doesn't mean that having the option to look at auto-backlinks is a bad thing - if used mindfully.
The mindfulness part is important here, though: It all comes back to training your mind to deal with the abundant choices you create when working on your notes system! I'd say avoiding this is not necessarily good or helpful.
I must say that I also disagree with the main point of the article if it's not restricted to specific kinds of backlinks. At the top of every one of my zettels there is a link to the zettel itself in the form
[[UID]]. If you click on it, the note list in The Archive shows all the zettels that have a link to this zettel, so according to the definition in the article it is a backlink.I find this kind of backlink very useful – mostly for two reasons:
Aren't links just links? From one perspective, outbound from another perspective inbound. Both want to be meaningful to my future selves. @msteffens and @matti are suggesting mature linking hygiene.
I agree mindfulness is important here and everywhere. More training can't hurt either.
Will Simpson
My zettelkasten is for my ideas, not the ideas of others. I don’t want to waste my time tinkering with my ZK; I’d rather dive into the work itself. My peak cognition is behind me. One day soon, I will read my last book, write my last note, eat my last meal, and kiss my sweetie for the last time.
kestrelcreek.com
This seems like a fairly clear "$thing is bad, don't do $thing or you'll fail" sort of article.
A count of individual entries within my VimWiki and links found therein, which I guess is a reasonable proxy for size:
Probably around 300k words, although a decent chunk of that is program output, fragments of chats, etc. Earliest entry in my VimWiki diary is from July 2016. There's a backlog of paper notebooks, blog content, and index cards going back to the late 1990s that I'm slowly working on integrating.
This is an important thing to mention, I think.
A bit of background: Recently, when @fbb posted about his software, Sems, I stumbled upon his Bachelor's Thesis and looked at some of the references there. -- This led me to a comparison of user interface interaction speed for research tasks (find some information in a nested table of contents. (The source: Richard Chimera and Ben Shneiderman (1994): An Exploratory Evaluation of Three Interfaces for Browsing Large Hierarchical Tables of Contents, ACMTrans. Inf. Syst. 4, 1994, Vol. 12, S. 383--406.) The interactions they measured were about finding stuff in a large table-of-contents-style outline, like 'What chapter, section, or subsection number, if any, has the title, "Aerial Refueling Fire Emergencies"?', and some of the study's conclusions outright mention that some interfaces apparently enhance exploration.
Andy's notes are very explor-able, too, and part of that are the backlinks that you can use to get from a detail note to a more general topic overview. Same with @nickmilo22's IMF demonstration. As an outsider, you don't know anything about the structure, so you enter the web somewhere and then click around.
Apart from casually browsing through my notes, which I don't often do, I don't think there's much actual exploration happening in my own archive. (Of course I forget where I put things in the past and have to search and follow links a bit. That's different from looking at other people's hypertexts: I'm looking for something in particular with expectations about how I might get there in mind.)
An anecdote from the Chimera/Shneiderman study: when people expand a tree view of a content section that is really long, some chuckled, some physically moved backwards; at the very least, they were surprised by what they saw, maybe even overwhelmed a bit. It's not their own content, it's something unknown they have to plow through to complete tasks. -- Now when I think about my own large outline files (e.g. my emacs org-mode file for The Archive development is ~2000 lines of text in total and >16k words), I don't get overwhelmed when I open the file. I don't know every nook and cranny inside out, but I feel rather confident getting around.
So I'd conclude that the study might be revealing interesting things about exploring stuff you don't know, and how user interfaces can help there, but it doesn't tell you much about which mode of interaction and which interface is "better" for things you created yourself.
With that in mind, I'd argue that the stuff that's useful for exploration of unknown material is not automatically as useful for working with material you created yourself.
(I would consider showing backlinks in an online publication of my Zettelkasten, if I ever create one, with the hopes of aiding people to discover e.g. the structure notes that point to a note they're viewing, but so far I can live without this features most of the time when I produce and work with my stuff.)
Author at Zettelkasten.de • https://christiantietze.de/
I think the Internet analogy is misleading. Back-reference search engines for scientific literature are widely used exactly because they offer even more connections. Sometimes, the biggest question on my mind is truly "in what other context is this relevant?" Has anybody used these results? Applied these methods? Challenged the conclusions?
Claiming that backlinks are "bad" is bizarre, since, I will argue, they are obviously providing context and information about the relationship between notes. As a thought experiment, lets order context relevance between notes. We have a note "Z" and these related notes:
This list is not ordered, but if I were to assign an expected "relevance" to these notes, then "A note linking to Z" is quite important. I'd argue, that it is the emergent structure of a Zettelkasten that gives interesting and new insight -- not the path that has already been trodden.
Comment from a simple user - I started with the concepts of links and backlinks, but quickly realized that concept was just tied to the order in which I created zettels. Links are just links; they shouldn't be a vector and shouldn't have a direction. On one occasion, I might step from zettel A to zettel M in my ZK; at another time it might be from M to A. I don't care if originally I thought A was the parent idea and M the child; in the fullness of time and with a more mature ZK, that distinction loses meaning. So I like to make all links connect both ways so that they are not directional.
@Sascha I wonder if you boinked your head one of your bad backlinks and became disoriented and hit publish while in a daze.
I am not a fan of naysaying!
Add me to the disagree list.
P.S. to my previous comment - I do all my linking "explicitly", that is, I insert the UID for the linked zettel (#2) at the bottom of the existing zettel (#1), along with an explanation of why the link is important, so that I know later whether or not to follow that link. And then I go to the zettel #2 and insert a UID for zettel #1, again with an explanation of why the link makes sense in that direction (not necessarily the same explanation as from #1 to #2). So, I don't know if you call the second one a "back link" or not, I just call them both "links" and the rationale for following from #1 to #2 or from #2 to #1 is explicitly stated.
I used to follow the method described by @Vinho to "find backlinks" but soon realized it was unsatisfactory for determining why I should check any of the backlinked zettels. That's why I settled on explicitly (and manually) establishing all links.
I think that, in some cases, link direction can have a very important meaning and thus great value. Esp. in the light of my above comment on qualified links (semantic links), I think it's very useful if we're able to distinguish between link directions. For example, my notes "A" and "C" have a very defined relationship to each other and I want my links to convey this info:
"the statement in note C disproves my thoughts in note A"
"note A is disproved by note C"
Being able to distinguish between links and their backlinks is useful IMO. But by manually placing the backlinks, they become just links and software cannot identify any link direction anymore. However, if instead software automatically generates backlinks and presents them in a separate view or in a graph (or if you simply search for them), the info about original link direction can be better maintained. Thus, I personally feel, that manually placing backlinks is inferior since I loose the option of indicating direction with my link.
If, however, I could label my links with a link type, then manually placed backlinks could transport this info as well and it would become a non-issue. Right now, the best option is to describe this as part of the text that accompanies the link.
@msteffens From a technical perspective, I find your labeling idea intriguing -- I find it hard to imagine how to apply this to less formalized contents, and if at the end of the day the label-able things are outweighed by contents that are not label-able that easily. Have you tried this with your notes and looked how it went? (If I recall correctly, the way you present your notes it looks like you do a lot of analysis of (empirical) sources.)
Author at Zettelkasten.de • https://christiantietze.de/
@ctietze yes, semantic links likely won't make sense in all cases, and should only be used where it makes sense. But I feel that link relationship types have great potential for some areas, like e.g. when analyzing/comparing scientific literature or possibly argument chains.
I haven't tried this yet with a larger corpus of real-world data, so I can't yet report about whether it would work in practice for me. And, as you point out, it's usefulness may depend heavily on one's notes corpus and its topics. But link types would be entirely optional and wouldn't add any additional complexity if not used.
Semantic links are currently discussed at various places for various tools (e.g., Obsidian, Org-roam, Roam, Foam). So there seems to be some common interest. These discussions also show that being able to attach a property to a link (like
[[property::link]]) may be also generally useful – apart from the use cases I'm envisioning (think CiTO).what is the difference between
citationandcitesfor CiTO properties?I find these naming conventions (or those suggested in other platforms) too limited for the greater part of my notes to be actually of any use.
my first Zettel uid: 202008120915
The description for the CiTO citation class mentions the difference :
In CiTO (and using my layman terms), a "citation" describes the link from A to B.
To further describe the actual type of this link relationship, you'd use a property (like e.g. "cites as data source" etc). The property "cites as data source" is a sub-property of "cites" which is the main (i.e. parent / most generic) property that all other (more specialized) sub-properties are derived from.
Example of a citation:
The CiTO ontology just proposes a standardized set of properties that can be used to describe the type of relationship for a link. This has been developed with academic citations in mind and is mostly useful for that area.
However, IMO that doesn't mean that we can't define our own terms. I've given this mostly as an example of what would be possible. But the general idea is more important: Namely that you can add a property to a link (like
[[property::link]]) in order to define its type of relationship.I imagine that a user could enter whatever property name he/she finds useful. And when autocompleting link properties, the software could simply suggest all properties that you've previously used.
And, of course, you can always ignore this feature, and simply link via
[[link]].Thanks for pointing out the value of the link format of
[[property::link]]. I can see how I could work with this if thepropertykey were optional and completely unrestricted, not tied to "link direction". Using a property of say,argument,support, or maybe a pairmethod/practice. This could be property optional added other types of links: citekeys, links to images and pdfs, or other appication links.I'm curious, @ctietze might this be something available when scripting comes to The Archive?
This seems like a down-vote for seeding responsibility for creating "backlinks" to a software algorithm and an up-vote for manually creating "backlinks".
Will Simpson
My zettelkasten is for my ideas, not the ideas of others. I don’t want to waste my time tinkering with my ZK; I’d rather dive into the work itself. My peak cognition is behind me. One day soon, I will read my last book, write my last note, eat my last meal, and kiss my sweetie for the last time.
kestrelcreek.com
Everything is prescriptive if you are looking hard enough. I think that reading text is more of a responsibility of the reader. There isn't much that can be done as an author in this regards. Once enough context has been provided fairly clear "$thing is bad, don't do $thing or you'll fail" sort of statements are unavoidable. Without it would make a publication of such text pointless, except for the cases when you want to empathize the non-conclusiveness of your statements. Remember that we are talking about a blog article here, not a tweet.
my first Zettel uid: 202008120915
this is very important. @Sascha correctly identifies the shortcomings of such links for knowledge connection. While the use of semantic links seem beneficial it emphasises even more the connection of notes. CiTO is dedicated to that use. I don't believe that this language can be used for knowledge connection simply by adding some custom properties to it. When link placement should remain human readable different semantics are needed for different concepts to prevent confusion.
my first Zettel uid: 202008120915
@msteffens I feel that some of the discussion in this particular thread is a bit too rarified for me (not that I don't understand it, but rather that I have a hard time relating). Clearly some of the contributors are heavily researching particular areas of knowledge, with the end in mind of a thesis or advanced technical paper, and they are driven by those short and mid-term goals. I have been in that position and understand the drivers. That's not a criticism, just a statement of what dominates their lives and their thinking right now.
I do have a comment, though, offering a different perspective. I make it from the point of view of a semi-retired person who is interested in creating a ZK that reflects a wide range of things I have learned and at the same time is looking for new insights. So, my working goals are different. Also, I offer this not to be argumentative (honestly) but just to give a counter-thought for consideration.
My thought is this - as I have developed a ZK, I find myself removing hierarchy, organization and "direction", not adding it. It gets in there by itself, almost, or at least with very little effort by me. While it may reflect my current way of thinking, have been quite helpful in the past, and in some cases, be quite logical and sensible, I find it also tends to mask or even obscure new insights. If I am not careful, my ZK reflects too accurately the well-trodden pathways of my (or someone else's) brain. I see this as undesirable and I want to shake it up.
That is why I purposefully write my zettels as the first step, as opposed to starting them with a link from a structure note, for example. They start off essentially floating in my ZK. Then, after the zettel is reasonably mature, I look for connections that "make sense" to me (I understand that sounds vague) - not because they fit into a system of organization or knowledge, but based on their own merits. This actually takes a lot of time and searching through other zettels, some not clearly related. It usually takes a lot more time than writing the zettel in the first place.
This method of creating each zettel applies no matter what the source of the idea. It could be one of several ideas that I have encountered in a book or article or TED talk. Each idea is in a sense first isolated, then connected.
Whether or not these efforts to eschew the existing patterns in my brain bear fruit, we will see.
And maybe I am agreeing with @Sascha when he says that backlinks are not useful and may be harmful, for my own reasons. I hadn't thought I agreed when I first read his article.
(1) is currently my ZK method. (2) I can not agree with. Links and backlinks are breadcrumbs in my knowledge path. I’m happy to take either route later in pursuing (1). Can’t see how they could be harmful. YMMV!
DITTO........ not to steal your thoughts, but to better explain my just previous comment. Somehow, independently, we seem to have arrived at the same destination,
@GeoEng51 Many thanks for your detailed explanations of your way of thinking and approach! It‘s great that this forum is so diverse and carries so many people from different disciplines and all with their own use cases.
I didn‘t mean to add too much of an edge case and/or to dilute the main topic of this article. I admittedly tend to view the topic of personal info+knowledge management from my own background (which is natural sciences). So it greatly helps to hear different view points!
I guess my point was mostly to suggest that improvements in our linking behavior could enable software to support us better, like with backlinks that would better maintain context and semantics.
Yes, that makes sense. It's not a matter of one right (or wrong) way of doing things, but of supporting as many users with their own ways and preferences, as possible.
I too enjoy the lively discussion on this forum. @Sascha picked a good topic for this particular thread, as it seems to have some controversy attached to it - always useful for generating discussion
.
I think this is both promising and set up to fail.
I think that is the core difference. If you open his note "Evergreen note-writing as fundamental unit of knowledge work" you get the backlink on "Evergreen Note" which is very useful for us as outsiders. However, if you worked properly you should neither link definitions for explanations in your own Zettelkasten nor your system should provide you with any presentation of this concept. This is a good example on how to write for other people, but a bad example on how to write to create knowledge. A good link would be towards a note on a discussion on the statement
as knowledge work is based on flow of truth and not like a legal text which is more concept orientated.
So, to me these backlinks provide an additional tool to make sense of Matuschak's notes. I like them. However, they provide little value for knowledge work (understanding or creation).
I'd never say that there is no possible use of backlinks. The first case you are presenting is a perfect example of using backlinks for going steps back and forth. That is not what I am talking about.
I'd say the second use case will slowly disappear (or at least) should disappear if you have more practice. For example, one should always create the link to a note first before creating it. The level of proper integration of the note depends on your note and your Zettelkasten. It is neither good or bad to have a lot or view links. It depends on the specific use case (and user case).
That being said, the second use of backlinks might work for you, at least for now. However, I'd say that the focus is too much on note linking and to little on knowledge connection.
I am not arguing but just telling you from what I experienced: The interesting and new insights don't happen if you shake your Zettelkasten (roam, browse etc.) and hope that existing but unacknowledged ideas pop up.
I personally did dedicated sessions to search for those interesting and new insights. The experience was overwhelmingly disappointing. Actually, my productivity took off as I ditched those practices and focused on knowledge creation itself. Then the eureca moments did increase quite a lot.
---> @MikeBraddock said:
No. Just the result of a decade of experience and teaching.
I am a Zettler
@Sascha Thanks for your comments! I understand your concerns and think they are valid. I esp. agree on the problem of not being able to formalize link context.
Still, I think I’d like to try link types and see how it goes. Esp in the area of linking between literature notes and for bibliographic citations I feel they could be very useful.
Reflecting on the reactions over the past couple of days, one thing bubbled up to the top again and again. I wonder why we see so many emotional reactions for such a technical detail
Of course @Sascha and I have talked about some comments and points that are being made from the discussion, but from a perspective of "what to do to get outcome X", it was hard to discuss anything substantial. E.g. @msteffens provided a lot of interesting ideas to tackle the problem of meaningless links, and @Vinho chimed in with a (totally spot-on) report from his experience and how backlinks saved the day. -- And for me, getting to know someone's observation of regular interactions is also a useful to figure out how to improve an app. Double win!
I'd like to encourage y'all to spend a couple of minutes from time to time observing recurring ways you engages with your Zettelkasten, and then try to leave out a feature of your app of choice. (E.g. hide the note list/search results, so all you have is links.) Can you live without it? If not, why? What could be missing in the content of your notes to make things work regardless? -- Well, that's a suggestion at least to help us all to have a more substantial discussion of actually observable things in the world and in our lives, and to help us all improve the ways we work in the process.
Author at Zettelkasten.de • https://christiantietze.de/
@Sascha wrote:
I'm arguing from my own experience -- but that is semantics
Anyhow, what I want to say is that backlinks is an important part of understanding what is actually going on a zettelkasten -- at least for me. The backlinks are a key part of my workflow for exploring and understand the context of my notes. And yes, I have done without them for a while, since I, at first, didn't know there was a button to access them (in TiddlyWiki.)
But I will keep an eye on my use over the next few days, and if I find something worth sharing, I will write a post (probably also about graphs/networks since I use those a lot, too.)