Hello Derya,

Termine is a text mining webservice from the National Centre for Text Mining (NaCTeM) at the Univeristy of Manchester. EPPI-Reviewer accesses this webservice as part of its text mining functionality.

There was a problem with the Termine web service a couple of week ago but it looked to be back running properly. I just tested myself and it didn't return any results for me either. We will look into this to find out what is happening.

Best regards,

Jeff

Dear Derya,

Yes, TerMine is now working properly. I have tested the screening prioritisation functions as described on page 71 of the user manual: it all works as expected on one of my test reviews, making it difficult for me to find out what doesn't work in your case. Text mining techniques can be tricky, for example, I would be surprised if they worked with the example review, as there are too few references to work with. Also: it's sometimes a good idea to "clean" the list of terms, so to make sure that all corresponding concepts are actually relevant.

In other words, in order to provide more specific help I will need more information:

1) how are you selecting the items that are used to create the list of terms? Have you tried with a shorter list?
2) is the corresponding list very long? Do all the terms extracted seem relevant?
3) assuming that you are not finding any similar references after clicking "Search on Terms", have you tried eliminating the terms with the lower score?

I guess the questions above already suggest what I think may be the problem: it's difficult to evaluate this technology without working on a fully featured "real world" example. Text mining can be very useful for big reviews (50,000 references or more) where the only possible way to handle the screening phase within the expected time requires some sort of prioritisation, but smaller scale tests may simply lack the numbers and data-diversity required.

Please do let us know if this helps and how your work is proceeding.

Sergio

Dear Derya,

I confess I'm puzzled now. I tried to re-create the issue you report and I couldn't. First, I've tried on one of our test reviews, and it all worked as expected (even when working with 10k+ list searchable items), this made me suspect that something may be going wrong in your peculiar case, so I gave a look at your review (ID=1306), I hope this is all right.

Within the aforementioned review, I've tried to see if I could re-create the error in the following (arbitrary) way:

1) in the "Search" tab, I've searched for a specific expression, so to retrieve a sample of references (to use as my source of data). Got 24 documents and listed them.
2) in the main document tabs, I've used the "Find more items like these" dialog to get the TerMine terms.
3) I've cleaned the terms list, so to end up with about 12 terms that looked significant (disregarding the score), I was (arbitrarily) selecting terms that were likely to be present in other documents as well, but that seemed to be specific enough to provide some kind of meaningful result. I'm mentioning this to clarify that even if arbitrary, this test is not entirely meaningless.
4) I clicked "search on Terms" and after some 10 seconds I got a new list of 173 references.

Once again, it was all successful, so I still don't know what went wrong. If you'll look at the "search" tab, both of my searches (from point 1 and 4 above) are there for you to have a look.
Looking at your "search" record, I can only speculate about what the error might have been: it is entirely possible that you stumbled on a "timeout" error. This happens sometimes (especially with searches) and is triggered when creating a new search takes more than 30 seconds. It may be due to bad luck (the search happened in a very busy moment) or it may be structural: if it is repeatable, please let us know! The "term search" functionality is a pretty complex one from the computational point of view, this means two things: it is difficult to find a way that ensures that no searches will time out, as this will be influenced by the general traffic, but if you are able to produce a search that invariably fails, then I would really like to know. In this second case, I may be able to use the failing search analyse why it is taking too long and optimise the procedure so to speed every future term-search.

I'm sorry that I don't have a definite answer: I will appreciate your collaboration to pursue this issue. Please give a close look to the error message you may see and report back: I would like to know exactly how you triggered it (you may want to use the "export terms" button to save the current list of terms). Also, does the error message mention a "Timeout" somewhere?

Many thanks,
Sergio

Hi Derya,

yes, feel free to send screenshots to eppisupport, we'll continue our conversation via email.

Sergio