tathagatachakraborti / survey-visualizer Goto Github PK

For the 3 deployments:

• VAM-HRI [link]
• MACQ [link]
• XAIP [link]

Originated from discussions with Alberto Camacho during ICAPS 2022 demo session.

A feature to summarize a set of papers as related work. There exist tools for standalone paper summarization but the proposed form requires a summary comparing and contrasting among a selection of papers to give a good overview of a set of papers (that can, for example, seed the related work section of a new paper).

Is your feature request related to a problem? Please describe.

It's hard to tell what each taxonomic class does right now, there is no description attached to the nodes.

Describe the solution you'd like

1. Configuration entry to tell the compiler to look for descriptions.
2. Alternative rows of descriptions and labels in the master spreadsheet.

3. Descriptions go into the modals that open up when the node is clicked.

@sarathsreedharan

We want to have a single compilation run to produce the starting point for the Insights tab. This avoids having two separate spreadsheet reads and nightmares with what structures are supported and what is not + no having to map between IDs and slugs from separate reads.

• Produce starting YAML file for configuring rules @TathagataChakraborti
• Remove slugs and replace them with automated slugs @TathagataChakraborti
• Modify the encoding script to read from the YAML and compiled JSON instead. @haz
• Remove vestigeal flask stuff from the encoding script @haz
• Streamline the preference specification process 👀

This is the current compiled JSON schema; and here is an example JSON. Do you need me to modify or add things to it that make your job easier?

We have two different schemas right now with overlapping contents -- one for the server and one for the compiler. Need to merge the overlaps.

This is a summary of suggestions from IEEE VIS reviewers. If anyone wants to explore these
further, please open a new discussion and we can proceed further with more specific issues.

I liked the four visual components. The treemap and the bubble chart visualization
seem redundant, though. I think it would be better if they showed different but
complementary information. The second one, e.g., could show the number of most-
cited authors (above a threshold) per topic. They could also be linked to one
another (i.e., interactions in one would be reproduced in the another).

The network visualization could be improved by considering paper metadata. One
example would be to color nodes according to the paper publication year, a feature
that would help users to find time intervals of interest. Please see [1].

I should notice that such recommendations would be limited by the volume and accuracy
of the survey data included by the survey authors, i.e., if the authors only categorize
50–100 papers for a topic/field with thousands of papers, the recommendation might
be sound with respect to the list/set of categories, but eventually useless (as
such work could already exist outside of the set of collected papers). Perhaps,
additional factors such as number of papers on a particular category over time (or
over the course of the last N years), citation counts (if available), centralities
(incl. PageRank, for instance) in the citation and/or co-authorship networks, and
so on, could be eventually added to this recommendation approach.

For the 3 deployments:

• VAM-HRI [link]
• MACQ [link]
• XAIP [link]

Originated from discussions with William Yeoh and Alberto Camacho during ICAPS 2022 demo session.

Replay the timeline of papers in survey data and analyze to what extent the new paper generator feature matches with work actually done in the past.

For the frontend application, using checked in VAM-HRI files here.

The Install OS Packages step fails for all 3 deployment scripts with the following error:

Collecting sklearn
  Downloading sklearn-0.0.post5.tar.gz (3.7 kB)
  Preparing metadata (setup.py): started
  Preparing metadata (setup.py): finished with status 'error'
  error: subprocess-exited-with-error
  
  × python setup.py egg_info did not run successfully.
  │ exit code: 1
  ╰─> [18 lines of output]
      The 'sklearn' PyPI package is deprecated, use 'scikit-learn'
      rather than 'sklearn' for pip commands.

I am not sure if this is a dependency we actually need. Also, just re-running the failed jobs somehow works. 😕❓

Add whitepaper on the tool to arXiv.

The taxonomy and both paradigm tabs have individual category count values reading from the prior categories' totals. Ex: 'Body Extension Alterations' should have the count value assigned to 'Special Effect Alterations', 'Special Effect Alterations' should have the count value assigned to 'Cosmetic Alterations', etc.

The hierarchy is not expanding into space for smaller depth. Ideally, it should take up the horizontal space so that the node labels are not cut off unless absolutely required.

Example: http://ibm.biz/xaipviz

Is your feature request related to a problem? Please describe.

Currently, the page title (as seen on the browser tab) is quite generic.

Describe the solution you'd like

Use information from the configuration to populate this.

1. A name entry in the metadata

---
metadata:
  acronym: VAMHRI
  name: VAM-HRI Survey
  title_text: 'Virtual, Augmented, and Mixed Reality for Human-Robot Interaction'

2. Compile a new JSON file that index.html can read from (in the compilation script).

Is your feature request related to a problem? Please describe.

Currently, the configuration file only admits absolute paths.

1. This becomes a bit ugly for CI pipelines.
2. We can't have working examples contained in the repo using the sample files.

Describe the solution you'd like

A relative tag to the configuration entry for paths to spreadsheets and PDFs.

    input_file:
      filename: data/slug.xlsx
      relative: true
....
- name: Network
  files_directory: 
    location: data/pdfs
    relative: true

Dummy issue to host screenshots used in documentation.

The old papers checked in have caused a lot of bloat in the commit history. It's safe to reset it and make the first clone more friendly.

Time to migrate to carbon v11. This should also fix the sass dependency workarounds organically.

instructions   walkthrough

Is your feature request related to a problem? Please describe.

We lack basic insights into the feature space outside of the new paper generator.

Describe the solution you'd like

1. Tags with no papers.
2. Tags with the most number of papers.
3. Tags with the least number of papers.

Is your feature request related to a problem? Please describe.

Currently, the configuration file requires specifying which one is the active worksheet in the spreadsheet. Most often, this is just the first sheet and should not require specification.

    input_file:
      filename: path/to/spreadsheet
      active_worksheet: main

Describe the solution you'd like

If no active_worksheet is specified, use the first one.

• Change the CI pipeline for MACQ to deploy to Heroku instead.
• Map CNAME from macq.planning.domains

The starting index for rows and columns for taxonomy and paper configuration starts at 0 but the key_map for the papers starts at 0. This is very confusing. 😒 [example]

Going to convert all to start to 1 instead.

Describe the bug

k-paper suggestion is broken. Getting the following error in the server logs:

172.17.0.1 - - [05/May/2022 01:12:21] "POST /imagine HTTP/1.1" 500 -
Traceback (most recent call last):
  File "/usr/local/lib/python3.10/site-packages/flask/app.py", line 2095, in __call__
    return self.wsgi_app(environ, start_response)
  File "/usr/local/lib/python3.10/site-packages/flask/app.py", line 2080, in wsgi_app
    response = self.handle_exception(e)
  File "/usr/local/lib/python3.10/site-packages/flask_cors/extension.py", line 165, in wrapped_function
    return cors_after_request(app.make_response(f(*args, **kwargs)))
  File "/usr/local/lib/python3.10/site-packages/flask/app.py", line 2077, in wsgi_app
    response = self.full_dispatch_request()
  File "/usr/local/lib/python3.10/site-packages/flask/app.py", line 1525, in full_dispatch_request
    rv = self.handle_user_exception(e)
  File "/usr/local/lib/python3.10/site-packages/flask_cors/extension.py", line 165, in wrapped_function
    return cors_after_request(app.make_response(f(*args, **kwargs)))
  File "/usr/local/lib/python3.10/site-packages/flask/app.py", line 1523, in full_dispatch_request
    rv = self.dispatch_request()
  File "/usr/local/lib/python3.10/site-packages/flask/app.py", line 1509, in dispatch_request
    return self.ensure_sync(self.view_functions[rule.endpoint])(**req.view_args)
  File "/PROJECT/server.py", line 59, in imagine
    result = imagine.find_k_new_papers(data["num_papers"], caller)
  File "/PROJECT/vamhri_encoded.py", line 435, in find_k_new_papers
    for _ in range(k):
TypeError: 'str' object cannot be interpreted as an integer

To Reproduce

1. Set interactive to true in the config

- name: Insights
  disabled: false
  interactive: true

2. Bring up the client and the server.
3. Go to the Insights tab, and hit "what's next".
4. See the error above.

Is your feature request related to a problem? Please describe.

Currently, the new paper generator on the Insights tab has no modifiers -- especially for larger datasets, it's difficult to focus the search based on the user's needs. We want more interactive elements!

This will help out with a better user experience as well as being able to finish the computations faster.

Describe the solution you'd like

Currently, enabling the interactive key on the interface allows the user to pick

• One or more taxonomic classes of interest
• One or more papers of interest (which internally translate to a selection of taxonomic classes that those papers belong to, in addition to the identity of the paper itself)

How we interpret this is left to be determined, but it's a start. Possible interpretations:

1. Consider only this subset of taxonomic classes.
2. Consider the selected papers as neighbors.
3. Consider the selected papers as producing topics of interest, feeding into (1) above.

One problem with 1-3 is that with a low number of selections, there won't be any solutions. But we can start with everything selected and adjust to deselections instead.

Is your feature request related to a problem? Please describe.

The visualization in embedding space is kind of meh. Ideally, we would like for the hierarchy and cluster encoded in the underlying taxonomic classes to come out in the visualization.

Describe the solution you'd like

Exploring possible integration with HOTT encoding. 😍

paper