Version 4.11.5.1 is a minor upgrade in terms of visible changes, but is actually substantial, as it includes the first reasonably complete version of the Microsoft Academic Graph (MAG) features in EPPI-Reviewer Web. These are still on invitation only (more information here), but making them available in the Web version is a major requirement for opening them up to a wider audience. Aside from this, this release includes a handful of bugfixes and small improvements.

Features based on the Microsoft Academic Graph

In our roadmap to test, evaluate and validate this vast set of new features, a very important milestone is reached with this release. Most functionalities (and all essential ones) are now implemented in EPPI-Reviewer Web. These include a brand new "Boolean search" against all 240 million records in MAG, as well as all the Machine Learning powered searches. The EPPI-Reviewer Web implementation also includes a number of user interface improvements (compared to EPPI-Reviewer 4), based on what we have learned from the real-world experience of the "invited" users, as well as our own in-house usage.
The next steps in our plan are to further improve these features are "presented", so as to make their purposes clearer, and produce extensive documentation/guidance. Meanwhile, our work on formal evaluation is progressing steadily.

Enhancements

In the Item Details page (EPPI-Reviewer Web), whenever an item already contains one or more PDF documents, a small "PDF" icon will now appear on the top-left area where the item details are shown. Clicking on the icon will automatically load the first PDF uploaded into the current item.

In the Coding Only interface, the Item Details page allowed users to change the "complete/incomplete" and "locked" statuses of coding (for any coding tool). We received some feedback suggesting that it is not appropriate to allow users with the Coding Only role. Upon reflection/discussions, we agreed and implemented the change. Starting from this release, changing the "complete/incomplete" and "locked" statuses of coding is not possible via the Coding Only interface - this applies to both versions (4 and Web). Nothing changed in the full user interfaces.

Bugfixes

Both versions: the paging of items lists could produce inconsistent results. This problem was rare and mostly visible when dealing with items lists produced from search results, if and when the search did not include scores/ranking. We discovered that, when done at different times, the Database could sort lists of items in slightly different ways; since the sorting that happens at the database level determines what items appear in a given page, this could mean that, while moving through the pages, some items would be missed, and some other items would appear in two pages instead.
The problem applied to both versions and has always been present - it has been unnoticed until now because of its extreme rarity. Outside of non-ranked search results, we found that it was practically impossible to make the problem happen deliberately, while for search results, in our tests we could only observe this behaviour when comparing the paging produced by changing pages after waiting for fairly long amounts of time (usually hours).
This problem is now fixed for all lists (we could observe it to occur only on very long search lists, but the same issue could potentially affect any item list), in the process, we could also make the data-collection phase for these lists marginally faster.

EPPI-Reviewer Web, PDF (line-by-line) coding. When coding some selected text, if the selection spanned two or more pages, the text-to-code association would either not be saved at all, or be saved incorrectly. If the selected text was being added to a code that had not been applied to the current item, the operation would appear to complete successfully, but in fact, no data would be saved. If the selected text was being added to a code that had already been applied to the current item, the selection from the first page would be saved (almost) correctly, while the parts in the subsequent pages would not. This bug was not fully solvable (not in the ideal way), therefore from now on, EPPI-Reviewer Web will refuse to save "multi-page selection/coding" and show an explanation instead.

Update: version 4.11.5.2

This release does not include new features, which is why we're adding the details in here.
BugFix: importing items from Microsoft Academic. Items imported from MAG had a problem with Authors (first and last names were inverted) and didn't import the most useful URL. They now import authors correctly and the imported URL points to the actual MAG record, which in turn will list all the known URLs for the reference in question.

BugFix/Enhancement: deduplication algorithm. When calculating the similiraty score, a high score (0.97) could be assigned to references with similar (not identical) titles, if other information (DOI, journal, pages, issue) was missing. This could increase the chances of not identifying false positives, especially for the case of "Reports" (typically, these are research reports that are self-published, are not "journal articles" and therefore lack additional information such as pages, journal title, etc.). The problem is now sloved.

BugFix: importing items. Importing new references from a tagged file (RIS or otherwise) could fail if one of the references contained more than 256 authors and author names were repeated. This is an extremely rare occurrence and usually happened only for cases where the tagged file was the result of some pre-proccessing error occured before reaching EPPI-Reviewer; however, the problem is now solved and such imports will not fail anymore.

Future plans

We are hoping that we'll soon feel ready to open up the MAG-powered features to wider audiences, but of course, much will depend on the ongoing evaluations and feedback from early adopters. We expect the next "missing" EPPI-Reviewer Web feature to be implemented will be "setting up configurable reports", we will also write new fixes/enhancements, as usual.