Theresa:

In this message and the next one are your comprehensive exam
questions.  Feel free to ask questions of any of us.

Please answer each question in 7-10 double-spaced (not-too-small font)
pages.

Typically, ISP students have nine days to complete the written portion
of their comprehensive exams.   You are free to hand it in earlier
than that, if you wish.  
===================================================================
Professor Ashley's question is in Word and will be sent in the next
message.
===================================================================
Professor Hwa's question:

Consider applying the learning algorithms discussed in the following 
readings to the problem of summarizing customer product reviews (as
the task is conceived of in Hu and Liu (2004):
  * Blum & Mitchell,
  * Mitchell, chapter 3
  * Mitchell, chapter 6
  * Mitchell, chapter 8
  * Scholkopf

Determine whether the algorithms might be well-suited to the problem. 
Compare the relative advantages and disadvantages of the different 
approaches.  Discuss in terms of
* the availability of required resources,
* the choice of problem representations,
* computational complexity considerations

What are the biggest challenges for each approach?  Are they insurmountable?

Because summarization is not on your reading list, don't focus on the
summarization step per se, but rather on determining ingredients to be
included in such summaries.
===================================================================
Professor Wiebe's question:

Compare and contrast:

  * Different approaches to creating affective and subjectivity
    lexicons.  Consider the following dimensions:

        * criteria for including entries in the lexicon (i.e., what
          types of lexical items does the approach aim to include?)
          Which seem more appropriate for supporting NLP applications?

        * What types of information should be included about the
          dictionary  entries, to support NLP applications?  Sketch how
          you envision the information being used in NLP applications
          (consider representative NLP applications -- you don't need
           to be exhaustive).

        * How could the dictionary be populated (entries plus
          information about them)?   How feasible are the approaches?
          What resources would be needed? What performance might we expect?

===================================================================