|
|
Can I sit-in/audit CS 6604?
I do not "recommend" it :-). I strongly
believe that this type of course would have the
maximum benefit and utility if credited,
the project undertaken in full earnest, and the discussions
approached in the spirit of a regular course. If you insist
on auditing CS 6604, the requirements would be to conduct
a (literature) survey of a particular sub-area of recommender systems
and submit a sizable document (at least 40 pages, single spacing,
11 point or less, Non-Microsoft WORD) which will be
graded by the instructor
for polish, presentation, and critical analysis and
evaluation of the topic. Alternatively, a tutorial of some area
could also be submitted, one where the presentation itself is novel
and which brings out the key aspects of the area in an interesting
fashion.
For students who are crediting the course, a final report will
still be expected, although not necessarily of the same size
and polish as stated above.
Grading policies explained
As indicated on the web page and mentioned numerous
times in class, 25% of the grade will come from class
participation (talking, shouting matches, contributing
constructively to listserv discussions, volunteering as scribes
and note-takers, leading discussions etc.).
Recall also Slide 6, from our Introduction on Jan 15; I had originally
intended that everybody write a critical 2-3 page review of some paper. I
am willing to waive this requirement for those of you who are "regular
contributors" to class (see me in person if you have doubts that you
qualify under this category :-)). Of course, you could still turnin a
review if you like.
For others, you might want to avail of this opportunity. Notice that
reviewing a paper involves making fairly non-shallow, non-obvious remarks
and if you are reviewing a paper that was already discussed in
class, significantly more insight and novelty of arguments (than what was
enumerated in the lecture sessions) will be expected in your exposition.
See more detailed guidelines below.
The remaining 75% of the grade depends on your final project
report, expected to be of publish-worthy quality (again see guidelines
below).
Reviewing papers: suggested guidelines
You are welcome to review any published paper (conference/journal) of your
choice, so long as it pertains to recommender systems, broadly defined.
All reviews are expected to be 2-3 pages long and have the following approximate
format: (i) short (1-2 paragraph) summary/overview of the paper, (ii)
your opinion of the paper's relevance and technical
merit (a summary, again), (iii) detailed technical comments supporting
your observations in (ii), (iv) how the research can be improved/followed upon.
You are expected to present sound technical arguments and provide supporting
evidence for your review, if necessary. This could involve citing other
papers (that maybe the authors were not aware of), scribbling equations and
calculations (to show why their assumptions don't apply, for instance),
and/or providing statistics or any other supporting information that would
be appropriate. And yes, you are welcome to exceed the 3 page figure
above.
Preparing your final project report
Before starting your report (6-8 pages, see detailed guidelines
last), take a deep breath and read this rather
interesting and useful
writeup by Jonathan Shewchuck, UC Berkeley. Please also remember our
thrashings of papers in class and thus do not write your report
in ways that will
invite such abuse.
Some personal statements about how the report should be organized: I refer
to it as a "report" but you should prepare it so that it can be submitted
to a conference/journal as-is (i.e., without any changes). This
is my way of saying that unpolished work, spelling mistakes, formatting
blunders, incoherent writing is not acceptable.
The authors for the report should be only you and your project
partners (no, please do not include my name; if you do wish to acknowledge
some contribution I made, you can just mention it in an
"Acknowledgements/Thanks" section). If you received substantial assistance from
somebody (perhaps a system admin, some other colleague etc.), you are expected
to acknowledge them in your report. Also,
send me a separate email to clarify the
role of these parties in the research. Also, if you have had problems
working with any of your project members (ethical,work-related, behavioral),
please bring them to my attention if you think I should take these factors
into account.
Provide a brief and crisp abstract of your work, something that tells
the reader immediately what to expect in your report.
Then provide an introduction, stating the problem, what it is that you are doing
and why it is an important problem (the motivation). Then include a section
on related research, their shortcomings and how your work differs.
This should be followed by a description of your proposed
approach/solution/technique/methodology; perhaps theoretical, perhaps descriptive, but higly technical
nonetheless. Support your theories/statements with experiments, results,
statistics, observations. Interpret the results of your experiments carefully;
do not make sweeping statements, do not make "broad" and "anything goes"
statements (e.g., "the recommendations from our Rec2001 system were consistently
better than other techniques"; what does "better" mean? how is this defined,
etc.). Then, provide concluding remarks and a statement of future research
directions (yeah, yeah, you are probably waiting to run away from all of this,
but you should talk about this all the same). Acknowledgements and references
should be last.
Of course, you are welcome to reorganize the above suggested sequence
as you deem fit, as long as the "flow is smooth."
Do not insult the reader by repeating your own statements verbatim at several
portions in the text. It is quite irritating when this happens, and
gives the impression that you took some time yourself
to digest the importance of what you are talking about. If you must mention
some aspect at different points in the story, make sure you point out the
special "positional significance" of the aspect, as applicable to the
issue(s) under discussion.
The report is expected to cite all relevant research pertaining to the topic.
It is considered a serious and inexcusable blunder to not cite some work that
is pertinent. Even if you think that this other work was of low quality, you
are expected to include it and mention the connections to your own
research. Thus,
I am not expected to be able to find some (relevant) references/thread of
research that are not cited in your report.
If you have indeed submitted your report/planning to submit it for consideration
in a conf./journal, send me a separate email indicating this. If your work
involves a web site that we can try out etc, mention it somewhere in
the report.
An important thing to note:
- Straightforward application of technique X to dataset Y or application Z
does not constitute research. So, if you applied LSI on some interesting
music market data, and presented results, that doesn't qualify for a report (definitely not at the 6000 level).
This is again keeping with our view of producing "publish-quality" work.
I am willing to bet that such type of work is unpublishable nowadays in any
decent, respectable forum.
If you had to extend technique X in some innovative way or the application
of it to the problem was non-trivial (in some representational or implementation way), then that would make for a meaningful report.
Some good resources are available:
The report is expected to be 6-8 pages long, in double column format. Papers
shorter than 6 pages will not be accepted, although you are welcome to
stray beyond the 8 page upper bound.
I had to choose some consistent styling, so I settled on the IJCAI template.
You can find files for both LaTeX and Word at this
IJCAI'99 site. For LaTeX
users, notice that there is a style file, a bst file, as well as a sample
.tex file for how to format the paper. You
are required to use the above styles. Do not change the font size
declaration, line spacing, or such any tricks! All submitted final reports
have to look alike (except of course, for the content).
If you have problems or serious
difficulties working with these files, let me know ASAP. These files also
contain word limits on the abstract, formats for references etc., so most
of your questions about such things will be answered. Here's an example
paper formatted according to this style.
When you are all done, please send me a PDF version of the paper or give
me a hard copy. The report is due on the last day of class, before
class starts.
How the report will be graded
Your paper will be evaluated for:
- Relevance (10%)
- Strong technical content (40%)
- Originality and creativity (20%)
- Presentation (5%)
for a total of 75%. I will provide a detailed review of your paper along
each of the above categories and will use the IJCAI-provided form for
this purpose.