Benchmarked: **P@5 49.1%, R@5 97.9%** on a 240-page Opus-generated rich-prose corpus, **+31.4 points P@5** over its graph-disabled variant and over ripgrep-BM25 + vector-only RAG by a similar margin.
P@5提升31.4个百分点是一个实质性的检索质量跃升，而且是在消融测试（graph-disabled variant）中量化的——这意味着提升直接归因于知识图谱，而非其他变量。R@5 97.9%的高召回率意味着相关内容几乎不会被遗漏，P@5 49.1%则显示精确度仍有优化空间。从RAG基准来看，这个提升幅度意味着图谱结构在个人知识库中的效用可能远超大规模通用RAG系统的预期。
Every page write extracts entity refs and creates typed edges (attended, works_at, invested_in, founded, advises) with zero LLM calls.
「零LLM调用」建图是一个关键的工程决策：它意味着知识图谱的构建成本接近零、延迟极低，并且完全可以在每次写入时同步执行。相比之下，依赖LLM提取实体关系的系统必须在延迟、成本和图谱完整性之间做出妥协。这个设计选择也有其代价——纯模式匹配会遗漏需要语义理解才能识别的关系——但对于结构化的[[wiki/people/bob]]风格引用来说，这是正确的权衡。
bias: towards property graph / LPG
How to Spot Emerging Note Clusters Without Alphanumeric Note Numbering? by [[Ton Zijlstra]] in Interdependent Thoughts
I recall Bob Doto had a video at some point in which he used the local graph to show relationships to find bunches of notes for potentially writing pieces or articles as indicated in Tons' article.
One of the biggest issues with digital note taking tools is that they don't make it easy to see and identify chains of notes which might make for articles, chapters, or books.
Surely there must be some way to calculate neighborhoods of notes from a topological perspective? Perhaps if one imposed a measure on the space to create relative distances of notes?
If all human data were structured in one massive knowledge-graph (a global knowledge graph), we could unlock this potential. Fortunately, Solid is that knowledge graph.
I do expect new social platforms to emerge that focus on privacy and ‘fake-free’ information, or at least they will claim to be so. Proving that to a jaded public will be a challenge. Resisting the temptation to exploit all that data will be extremely hard. And how to pay for it all? If it is subscriber-paid, then only the wealthy will be able to afford it.
two-way links
🌟 Highlight words as they are spoken (karaoke anybody?). 🌟 Navigate video by clicking on words. 🌟 Share snippets of text (with video attached!). 🌟 Repurpose by remixing using the text as a base and reference.
If I understand it correctly, with hyperaudio, one can also create transcription to somebody else's video or audio when embedded.
In that case, if you add to hyperaudio the annotation capablity of hypothes.is or docdrop, the vision outlined in the article on Global Knowledge Graph is already a reality.
Paper by Gyuri Lajos and Andras Benedek. Gyuri's context was recommended by @wfinck. Looks like it pertains to knowledge graphs. Gyuri's own annotation calls it a "meta-knowledge graph"
A filing system is indefinitely expandable, rhizomatic (at any point of timeor space, one can always insert a new card); in contradistinction with the sequen-tial irreversibility of the pages of the notebook and of the book, its interiormobility allows for permanent reordering (for, even if there is no narrative conclu-sion of a diary, there is a last page of the notebook on which it is written: its pagesare numbered, like days on a calendar).
Most writing systems and forms force a beginning and an end, they force a particular structure that is both finite and limiting. The card index (zettelkasten) may have a beginning—there's always a first note or card, but it never has to have an end unless one's ownership is so absolute it ends with the life of its author. There are an ever-increasing number of ways to order a card index, though some try to get around this to create some artificial stability by numbering or specifically ordering their cards. New ideas can be accepted into the index at a multitude of places and are always internally mobile and re-orderable.
link to Luhmann's works on describing this sort of rhizomatic behavior of his zettelkasten
Within a network model framing for a zettelkasten, one might define thinking as traversing a graph of idea nodes in a particular order. Alternately it might also include randomly juxtaposing cards and creating links between ones which have similarities. Which of these modes of thinking has a higher order? Which creates more value? Which requires more work?
The latest advances in machine learning — namely transformers and self-supervised learning in natural language processing — will also make it possible to build personalized discovery engines that organize and surface information that is timely, relevant, and impactful
And possibly summarize it. And connect it to your own knowledge graph.
Readwise's spaced repetition and integration with Roam/Obsidian is doing some cool stuff with surfacing connections and serendipitous reminders. Here is a Twitter thread I wrote to myself when I started playing with it: https://twitter.com/alexbowe/status/1476817961897783296
In this proposed system, tags are full-fledged standalone files. They can be published and discussed just like any traditional heavyweight file can.
Tags are first class
The solution is to create a tag file that points to the original and edited photo, like DerivedWork(original=(some hash), derived=(some hash)).
Relational tags
Great teams have a plan to win when -- surprise, surprise -- they learn that a dozen other teams are pursuing their previously-thought-to-be-unique idea. They persevere when others (including us) tell them that ideas are cheap until they are brought to life. They both see themselves as unique and list many companies as their competitors.
We believe more work in the future will look like that of software developers today (automating away tasks and harnessing the flexible power of computers to get work done)
Soon we will see a one-person billion-dollar company, as many of the most talented individuals choose to work for themselves — as founders, in the creator economy, as freelancers, or in some other way.
In 1963, Ted Nelson coined the terms 'hypertext' and 'hypermedia' as part of a model he developed for creating and using linked content (first published reference 1965).[7] He later worked with Andries van Dam to develop the Hypertext Editing System (text editing) in 1967 at Brown University.
In fact, such small effectively closed scientific communities built on interpersonal relationships already exist to some extent
so the weights in the reputation graph are personal knowledge, not citations or whatever.
Knowledge graphs combine characteristics of several data management paradigms: Database, because the data can be explored via structured queries; Graph, because they can be analyzed as any other network data structure; Knowledge base, because they bear formal semantics, which can be used to interpret the data and infer new facts.
Characteristics / benefits of a knowledge graph
The ontology data model can be applied to a set of individual facts to create a knowledge graph – a collection of entities, where the types and the relationships between them are expressed by nodes and edges between these nodes, By describing the structure of the knowledge in a domain, the ontology sets the stage for the knowledge graph to capture the data in it.
How ontologies and knowledge graphs relate.
An ontology is as a formal, explicit specification of a sharedconceptualization that is characterized by high semantic ex-pressiveness required for increased complexity [9]. Ontolog-ical representations allow semantic modeling of knowledge,and are therefore commonly used as knowledge bases in artifi-cial intelligence (AI) applications, for example, in the contextof knowledge-based systems. Application of an ontology asknowledge base facilitates validation of semantic relationshipsand derivation of conclusions from known facts for inference(i.e., reasoning) [9]
Definition of an ontology
A knowledge graph acquires and integrates infor-mation into an ontology and applies a reasonerto derive new knowledge.
Definition of a Knowledge Graph
the name of something and when you press the button to go to the link if it wasn't there it made the card
This is a phenomenally important UX insight and affordance that has become a foundation of how all modern wiki-linking knowledge graph tools work today. Kudos to Ward for this!
The problem with the annotation notion is that it's the first time that we consider a piece of data which is not merely a projection of data already present in the message store: it is out-of-band data that needs to be stored somewhere.
could be same, schemaless datastore?
many of the searches we want to do could be accomplished with a database that was nothing but a glorified set of hash tables
Hello sql and cloure.set ns! ;P
There are objects, sets of objects, and presentation tools. There is a presentation tool for each kind of object; and one for each kind of object set.
very clojure-y mood, makes me think of clojure REBL (browser) which in turn is inspired by the smalltalk browser and was taken out of datomic (which is inspired by RDF, mentioned above!)
After the success of MORE, he went on to develop a scripting language whose syntax (for both code and data) was an outline. Kind of like Lisp with open/close triangles instead of parens! It had one of the most comprehensive implementation of Apple Events client and server support of any Mac application, and was really useful for automating other Mac apps, earlier and in many ways better than AppleScript.
Yes, lisp!
This is my thinking as well i.e. if you could (a) keep parentheses but render them differently. But not going over board in basic view so it's still editable like text. AND also have a more graphical view.
After the success of MORE, he went on to develop a scripting language whose syntax (for both code and data) was an outline.
Lisp! ;P
More was great because it had a well designed user interface and feature set with fluid "fahrvergnügen" that made it really easy to use with the keyboard as well as the mouse. It could also render your outlines as all kinds of nicely formatted and stylized charts and presentations. And it had a lot of powerful features you usually don't see in today's generic outliners.
fahrvergnügen German for "driving-pleasure. Yes! ALSO This is kind of central, in two ways.
A. you need to have good story for mouse only and keyboard only B. you need to have multi-modal rendering
Engelbart also showed how to embed lists and outlines in maps:https://www.youtube.com/watch?v=yJDv-zdhzMY&t=15m39s
Now this is interesting. Instead of normal map here they've had to use this simple sketch/graph. Just arrows etc. BUT There maybe an actual value in that kind of simplicity!
Question worth asking here is why we have to see all the detail on the map always? Google may have different incentives than just showing you only essential data.