There are three steps:
git clone [HTTP-project-url]- Run
yarn. It will install necessary packages - Run
yarn dev. will compile the project and when it succeed, pop up the app
On the canvas layout, the shortcut pattern is different for different platforms.
- For Mac users,
fn+ctrl+[key]is the shortcut pattern. For example, to connect a source node (claim/pdf, etc.) with one or some arbitrary target nodes, please select the source nodes, andfn+ctrl+[right click]on the target nodes. - For Windows users,
fn+ctrl+[key]is the shortcut pattern.
The right panel of the app has 3 views. Each view has different roles.
- Canvas layout allows you to write and organize grounded claims
- Autograb list view shows a list of relevant sentences throughout the paper that fit your interest. The name "autograb" means these sentences are automatically grabbed by the text classification model.
- DocEditor view is an wizard-of-oz interface to synthesis claims. As you writing, the interface will pops up a similar grounded claim that you have wrote before.
For mac users, ctrl+alt+[view_ID] is the shortcut to switch between different views.
- Auto-grab: the feature that automatically grabs detailed information from the text content of the pdf.
To build the project into a standalone deskop application, run yarn dist:mac.
By default, it will take 3-5 mins and once finished, will save a .zip file in the path dist/compressor-0.0.1-mac.zip.
Study 1: Do grounded claims improve sensemaking?
- Subjects: ~40. undergrads? class credit?
- Task Constraints: under 1.5 hours. topics subjects can handle.
- Task: Given our design and the default
-
ctrl-f
-
transitions
- text: anything
- table: object attributes
- network: relationships encoded with proximity, angle, containment, lines
- map: has named sequential axis
-
clustering
- initialize canvas coords with tsne
-
chrome extension + firebase
- select text, right click send to firebase list of {url, time, query}
- firebase onlist add, knn search, list view
-
Fine-tuned, conditional semantic vectors for KNN search
- Fine-tuned by machines
- fastai + gorbid our pdfs?
- conditional
- paper/website title, authors, venue, section header
- knowledge graph links
- Fine-tuned by machines
-
document editor
- insert one doc in another
- transcullison?
- paraphrasing something imported
- create node from selected text
- create nodes/links from outline heirarchy
- insert viewbox into doc
- change font size of any character
- change font color
- drag lines to create links between items in the editor
-
queries -> search for anything, traverse the graph, large UI
- query builder
- a modal that take up entire window
- autocomplete for search parameters
- consider neo4j queries
- query builder
-
autocomplete -> small UI
- text matching algorithm
- could be ghost text instead of popup
-
multi-instance customizable layouts
- takes a query as input
- canvas
- list view
- drag order
- drag increase height
- show links in margin?
- add/remove/edit any node
- drag links between
- grid view
- drag increase col/row size
- show links in gutters?
-
pdf viewer
- manually fix autograb
- good ole' fashion text highlighting
- text highlighting from auto-grab
- autocomplete for phrases in selected area
- perf
- zoom
- prevent rerender
-
graph viewer
- select all left/right/up/down
- move many with arrows
- force directed
- collision
- alignment tools
- html-to-image?
- perf: react spring + transform + opacity
-
Bug reports
- text editor id can be undefined. maybe fixed?
** Can't focus multiple instances of slate editors, e.g. in a list: ** Cause: conflicting keys used to find editor in the dom Solution: editorUtils.tsx/initKeySafeSlate()
This would be either field deployment inspiration or walk path for a demo video.
process: random youtube rec -> watch 30 sec vid -> skim paper -> find interesting bit -> download into pdf dir -> open up electron -> grab bits to share -> screenshot -> paste into slack -> add context message in slack (edited)
- Predictive Interactions
- easily dismissible, correctness, alternate possibilities, refinements
- Read Eric Horvitz Principles of Mixed-Initiative User Interfaces
ngrok http 3000 -host-header="localhost:3000"
https://github.com/levelgraph/levelgraph 1200 # easy save to disk 3x slower than graphology for neighbors https://github.com/graphology/graphology#readme 200 # 3x faster than levelgraph for neighbors http://js.cytoscape.org/
https://ialab.it.monash.edu/webcola/index.html https://github.com/d3/d3-force
https://www.infoq.com/presentations/electron-pitfalls
https://github.com/CharlesStover/reactn https://github.com/diegohaz/constate
https://indicative.adonisjs.com/docs/syntax-guide
https://github.com/neo/react-use-scroll-position https://github.com/LeetCode-OpenSource/rxjs-hooks#apis
https://github.com/bestiejs/platform.js#readme https://benchmarkjs.com/
https://github.com/staltz/callbag-basics # push and pull? https://mostcore.readthedocs.io/en/latest/api.html # fastest https://github.com/ReactiveX/rxjs
level of detail/precision/specificity + attrs (esp. causality + utility) broadcasting, static docs, cognitive tutors
what's the expected outcome of some scenario (what's y given x) what scenario is optimal for creating some outcome (what predicts y) what does the author mean by x (use expands to & same as ) how confident should I be in statement X