Dear all,
Sorry for being silent for so long. We are working on new topics, a new task description, and new evaluation metrics.
1. You find results from last year's evaluation here: http://trec-car.cs.unh.edu/results/
2. Block your calendars. The TREC CAR 2018 submission deadline will be on August 1.3. In addition to the retrieval tasks, we will offer a "passage
ordering" task.
passage ordering task: We will give you an outline (yes, I caved)
and a budget of paragraphs. Your task is to provide a sequence of
paragraphs that would represent the article in reading order.
Your article should cover all topics in the outline, and address
topics coherently (i.e., all paragraphs for heading 1 should be
right next to one another, and not interleaved with paragraphs for
other headings).
You will provide one sequence per outline (not per heading).
You will provide a sequence of paragraph ids (you don't get to
modify the paragraph).
We will give tell you how long the sequence should be (i.e., a budget $k$ of paragraphs).
The passage ordering task is different from the passage ranking task, is that a ranking is sorted by relevance. For the ordering, you should consider the flow of arguments, coherence of topics, and maximize coverage.
Alternatively, we offer a passage ranking task as in previous years, but it will be evaluated through the passage ordering task as follows. We will take the top ranked paragraphs for each heading in order of relevance.
You are allowed to submit a mix of results for the passage ordering and ranking task -- but in total only three passage runs will be considered.
For the evaluation, each passage will be assigned to the most
relevant top-level heading, i.e., query-facet, in the outline. We
will use this to assess relevance for the article, coverage of
query facets, topical coherence.
Please stay tuned for more details on the submission format,
training data, and test topics.
If you have any suggestion for evaluation metrics, please reach out to us.
Best,
Laura Dietz, Ben Gamari, John Foley
Ryan,
Thanks for your interest, and: Welcome!
You don't have to participate in previous years.
There is only one task this year, and it is passage ordering.
However: Part of the submission is to also select a subset of passages that will be re-ordered in the next step.
Nevertheless: We plan on providing a set of simple rankings from which you can construct your candidate set.
Let me know if you have any questions.
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/94155e6d-4d89-4e92-a64f-0bfdd0676acc%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
To unsubscribe from this group and stop receiving emails from it, send an email to trec-car+unsubscribe@googlegroups.com.
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/3412d818-036f-4e84-817c-ed17c071e89e%40googlegroups.com.
There is only one task this year, and it is passage ordering.
Alternatively, we offer a passage ranking task as in previous years, but it will be evaluated through the passage ordering task as follows. We will take the top ranked paragraphs for each heading in order of relevance.
You are allowed to submit a mix of results for the passage ordering and ranking task -- but in total only three passage runs will be considered.
Yutao,
I think you might have read information about the TREC CAR Y2 evaluation which ran in 2018. This information does apply to this year's evaluation.
This year we offer only the passage ordering task.
We will provide some basic code for converting passage ranking
data into a naive paragraph ordering. So if you only want to work
on passage ranking, this code will convert your ranking outputs
into a passage ordering submission.
We will offer several evaluation metrics, some of which measure
the quality of paragraphs in your submitted ordering/ranking.
We do not officially host an entity ranking evaluation. However,
we will likely provide an automatic entity benchmark after the
submission, to encourage research in that area.
I hope this helps --- please keep on asking more questions!
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/6c2d683c-9070-4a06-8303-3d3b4779d9d0%40googlegroups.com.