Relevant commit: 4975ec6

index.php is loading OKM/standard_head.php and OKM/header.php on lines 4 and 10.

Currently, clicking on the back button will take the user back to the page that they had been on before Head Start. User testing shows, however, that users expect to return to the previous state of the visualization, e.g. that they would zoom out again after zooming in.

When zoomed in and only a single item is shown in the list, filtering the list brings back the other items in the list.

I am having problems with processing the output of Pubmed and I am pretty sure that they are encoding-related. When I check the encoding with Encoding(metadata$abstract), the result is "Unknown". Is it possible that everything that comes back from get_papers is UTF-8 encoded?

In some cases, the majority of papers gets assigned to one bubble. this leads to an overfull bubble, in which papers are overlapping. Example: http://openknowledgemaps.org/vis.php?id=9869332073d044946d607b7f6b54c9a8&query=maps

Currently, interaction with Head Start is limited to filtering, searching, and zooming.

It would be nice to give users the possibility to edit a map. This would allow users to correct perceived errors, and to extend maps with their own literature. The following features come to mind:

• the ability to move, resize, and rename areas,
• the ability to move papers to a different area,
• the ability to correct the metadata,
• the ability to introduce new papers and areas to the map.

These features only make sense in combination with an adequate persistence infrastructure where the changes are saved.

JavaScript console output: uncaught exception: event show inappropriate in current state visible

Right now, there are a number of style definitions in JavaScript that can be moved to CSS. Furthermore, there are a number of constants that are defined in the code and not upfront.

Steps to reproduce:

1. Load Head Start
2. Enlarge list
3. Click on any list item in list -> zoom in on related bubble
4. Next action triggers an inappropriate event (see console)

Highlighting areas when hovering over the area name in a list item would be useful.

Thanks to @mxphares for the idea.

While the current version of Head Start can be used on smartphones and tablets, there are a number of limitations with regards to usability and performance. Thus, a dedicated mobile version is needed that is optimized for small screens, touch displays and mobile processors.

It should be possible to export the visualization as a static SVG or PNG file.

In the timeline view, the first hover works as expected. The second mouseover results in all circles getting selected. See image below.

When in timeline view, the link back to normal view causes an error.

When the list is collapsed, you can't zoom out by clicking on the empty area of the list. Zooming out in the map area still works as expected.

Thanks to @mxphares for the bug report!

Currently we are using the D3 implementation of Force Layout on client side.

Not sure if there is a noticeable difference in performance if we run force layout on the server in R for ~100 nodes and deliver the final map. But I guess that the performance should at least improve for weaker devices?

But most importantly the user would no longer have the waiting phase until the layout is finished. This would also solve #31. As the map is ready to go from the start we no longer need to worry about the synchronization between paper list and map.

Steps to reproduce:

1. Load visualization
2. Resize window so that it becomes smaller than the visualization
3. Resize window to its original size

Now the representation will be different than after step 1 (Smaller bubbles, smaller fonts etc.).

When the user moves over a bubble, the bubble title is hidden to show the papers in the area. This breaks the user experience, especially with long titles and overlapping bubbles, as users want to see the title in some form at any time. A tooltip is shown after a few seconds, but this is not satisfactory. A possible solution could be jQuery tooltips as they can be immediately shown.

It would be awesome to extend the search functionalty to Pubmed. This would require to conduct a search in Pubmed and to retrieve basic metadata, abstract and categories for the first 100 papers. For display purposes in the visualization, we will also need to retrieve the full text PDF on demand (if it's available).

Currently the maps are not reproducible in the sense that the same papers as input will create differently rotated maps. By using PCA after NMDA, which would align the first NMDA axis with the direction of maximum variance.

Currently, the components of the visualization communicate directly with each other. When an event is triggered that effects two or more components (e.g. a zoom which updates both the chart and filters the list elements to show only papers of the zoomed area), the component which was notified about the event (in this case the chart), also directly updates the other components (in this case the list).

Therefore, there is high dependency between the different components and interaction is distributed and hard to manage. In the mediator pattern, components do not directly communicate with each other but through a mediator object. This would result in more loosely coupled components, and most possibly a better overview of the interaction between the components.

Otherwise, the file cannot be found on Linux systems using autoload.inc.php

It should be possbile to zoom out with a double click on a bubble. For many users this is an expected behaviour that stems from mobile devices.

The PDF URL has changed.

Exception: "uncaught exception: event zoom inappropriate in current state enlarged"

Steps to reproduce:

1. Click on bubble
2. Click on a paper title in the list
3. Enter search term
4. Click on the bubble
5. Click on the same paper in the list
6. Click on the bubble again

Feedback from @brembs: "In terms of then reading the literature, it would be good to be able to get a single click access to the full-text."

This can be achieved by adding an overlay when a user hovers over a paper, or an entry in the list.

When browsing through the list of search results, often individual articles are clicked on and focussed in order to see the detailed description. After zooming out I would expect to resume my search at the same article I clicked on

Zooming in and switching bubbles while zoomed in works perfectly, but zooming out is laggy.

Add a persistence layer that enables storing the state of a visualization in a database for later retrieval. This is intended to replace the static CSV files currently in use.

Even if show_dropdown and show_timelineare set to false, the respective elements appear in the visualisation after zooming in and out of a bubble.

PLOS queries

Is searchplos.r already parallelized? If not, would it improve the performance if we would use parallel searchplos(start=i, limit=10) calls with i = seq(0, len_papers, @10)?

Non-metric multidimensional scaling

Instead of using nmds {ecodist} (doc), we could use par.nmds {parfossil}(doc) which is the implementation of the ecodist function with multi-core foreach-loops instead of the standard for-loops.

At least on my machine this reduced the runtime for nmds from ~15s to ~5s

The interface looks much improved since I last found it. Zooming works very well and navigating the bubbles seems intuitive. I'll likely have some minor details to note upon extended usage, but for now this looks good to me.
In terms of then reading the literature, it would be good to be able to get a single click access to the full-text.

The default view would need to be set to expanded for #31 (show list while calculating the maps)

What do you think of that?

It should be possible to tweet the visualization using a tweet button. It should be possible to include a still image of the visualization in the tweet.

In zoomed out state, filtering works as expected. But once you zoom into an area, the filter isn't applied any more. I have attached images of the correct and the errneous behavior. In the example, the list and the bubble should be filtered for "social".

Correct

Erroneous

I don't yet know how representative my current search is so I'll expand on my searches, but one issue I noticed was that my technical term 'operant' was expanded to 'operative' and 'operating', neither of which are synonyms. Hence, the groupings are too often off-topic.
I realize that this is likely not completely trivial, but there likely needs to be some list of technical terms in the background. How about using the various ontologies? For instance, gene ontology is widely used and should be quite up to date. I haven't checked it in a while, but there are numerous behavior-related terms in GO and 'operant' might be there along with related concepts.
But perhaps these ontologies are already incorporated and just need to be corrected?

Hovering works only on top div, but mouse already changes into hand on bottom div but nothing happens when I click.

Add a tour with popups and highlights to explain the basic ideas and the functionality of the visualization (think new feature presentation in Google products).

Steps to reproduce:

1. Enter a search term
2. Click on bubble
3. Click on bubble again

Now all of the papers are shown in the list which should not be the case.

TF-IDF should perform better in finding distinctive cluster labels.

This way users can delete the search term with one click, without having to mark the term etc..

When zoomed in, the surrounding circles should not gain focus on mouse over. Also, the selected circle should not lose focus on mouse out. I have attached images of the erroneous and the correct behavior below.

Erroneous

Correct

Current requests are processed from start till end in one go and cause the known waiting times for users. In order to enhance the user experience we could split a request in:

1. Get data from PLOS

   Once the data is retrieved from PLOS the list containing the papers can already be displayed.

2. Calculate the map ...

   ... while the user explores the papers as a list.

Users should be able to set some options for their PLOS search: time range, article types and journals. That way, you can show e.g. just the research articles for a certain topic in the last month.

Users should be able to export data from Headstart without having to resort to looking at the network traffic to find the JSON. Suggested formats are JSON and CSV.

When a visualization context is available (e.g. the parameters of the PLOS search), then this context should be displayed in the visualization to improve the user's ability to interpret the visualization.

When clicking on the zoomed circle, the visualization is zoomed out. That should not happen. Instead, it should just stay in the zoomed position.

Before we progress with improvements on the frontend, we should complete the mediator implementation. The goal is that there are no cross-module calls any more. All logic between modules should be handled by the mediator.

Currently, width and height of the parent element are taken. It would be better, if users could explicitely state a width and height when calling Headstart.

Search allows you to filter using multiple keywords separated by space, even if they are not in a phrase, searching for e.g. "technology learning" finds entries with "technology enhanced learning". But the words currently don't get highlighted.