Meta information for each run

87 views
Skip to first unread message

Laura Dietz

unread,
Aug 3, 2018, 12:27:01 PM8/3/18
to trec...@googlegroups.com

Dear all,

I encourage you to take a look at the TREC CAR submission form before you prepare your runs. You will have to fill in some meta data about each of your runs, such as which external resources or methods you used. We will use this information in the overview report to study which kinds of methods/data have a positive impact.

Each team can submit up to 3 automatic entity rankings  and up to 3 automatic passage rankings. I anticipate most of your runs will be automatic.

TREC's concept of a manunal run is one where the ranking is created by a human, or with the help of a human. If you have a run which would "cheat" in some way or another but you would like to see how it fares in comparison, please upload it as a manual run.  We will probably not be able to include manual runs in the assessment (but we will try our best).


As always, feel free to ask questions.


Laura


Here the submission form:

https://ir.nist.gov/trecsubmit/car.html   (The login/password was sent to you by Ellen Voorhees a couple of months ago)

It looks like this:


REPLY EMAIL ADDRESS:
na...@internet.address
ORGANIZATION:
RUN IDENTIFICATION:
(case sensitive)
SUBMISSION FILE: File can be compressed using either gzip or bzip2, but you cannot use archive files (e.g., zip or tar)
RUN DETAILS
TASK  Passage Ranking
 Entity Ranking
RUN TYPE:   Automatic
  Manual
RUN FEATURES: Please mark any of the following features as they apply to this run. Please clarify any checked feature in the External Resources field below."
  method uses a linguistic knowledge graph such as Wordnet
  method uses a knowledge graph such as DBpedia, Freebase, etc.
  method uses entity linking
  method uses a dump of Wikipedia
  method uses data v2.0, v1.5 or v1.4 data provided as "unprocessedAllButBenchmark", unprocessedtrain",or "halfwiki"
  method uses data v2.1 data provided as "unprocessedAllButBenchmark"
  method uses another external document corpus (e.g., ClueWeb, CommonCrawl, ...)
  method uses pre-trained word embeddings
  method uses neural network technology
  method uses learning-to-rank
  method uses clustering, topic models, or summarization
  method uses other supervised machine learning method
  method does not make use of any training data (unsupervised)
  method is trained with data provided as train-v2.0
  method is trained with data provided as train-v1.5
  method is trained with data provided as train-v1.4
  method is trained with data provided as test200
  method is trained with data provided as benchmarkY1train
  method is trained with data provided as benchmarkY1test automatic qrels
  method is trained with data provided as benchmarkY1test manual qrels
  method uses other training data
EXTERNAL RESOURCES Please describe the features checked above here.
RANKING Please list the retrieval models used as part of the pipeline (e.g., BM25, query likelihood).
DESCRIPTION Please give a short description of the techniques used to produce the run.
CITATION If this method is already published, please give the citation to the publication here.
PRIORITY FOR ASSESSMENT:

(If the same precedence is given to different runs, NIST will arbitrarily choose.)

Rodrigo Nogueira

unread,
Aug 14, 2018, 2:23:47 PM8/14/18
to TREC Car
Hello,

What is "RUN TYPE" in the form? Shouldn't all submissions be "Automatic"?

Thanks,
Rodrigo

Laura Dietz

unread,
Aug 15, 2018, 3:04:10 AM8/15/18
to trec...@googlegroups.com
I suspect most submitted runs are automatic.

If you would want to submit a run created with a human in the loop, you could do this by submitting a "manual" run.

Also if you want to submit a run that is not 100% in line with the tasks requirements (i.e you looked up the page on Wikipedia) you may submit it as manual.

Best,
Laura
--
You received this message because you are subscribed to the Google Groups "TREC Car" group.
To unsubscribe from this group and stop receiving emails from it, send an email to trec-car+u...@googlegroups.com.
To post to this group, send email to trec...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/trec-car/abb3726d-e542-4b35-b673-879ce855fc2f%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.


Laura Dietz

unread,
Aug 15, 2018, 12:59:13 PM8/15/18
to trec...@googlegroups.com
Dear all,

I wish you all a Happy TREC CAR submission day!

To be safe, submit by tonight midnight --- but practically the submission form will be open all night until Ellen Voorhees comes in the office tomorrow morning and closes it.

Looking forward to your submitted runs!

Best,
Laura

xuesiy...@gmail.com

unread,
Aug 16, 2018, 5:58:11 AM8/16/18
to TREC Car
The file submitted for run `TUA1TEAM2' contains a different run tag (`Diptera‚y '). The run-tag in the results file and the run-tag given on the Web submission form must match. Please resubmit the run with consistent run tags. No record of this submission will remain at NIST.

I am still be in this problem, does anyone know what is the tag of web submission form??


Best
Reply all
Reply to author
Forward
0 new messages