Questions needed to be answered

• Problem definition

  • What is the value of the topic? How will this be evaluated?
  • What are the challenges?
  • What ambiguities do you expect to encounter and have to resolve?
  • What kind of features would you use?
    • The words? Perhaps other features as well?
  • What is the starting point?
  • What improvements would be made, and how can they be achieved?
  • Problem specific questions
    • Are there two distinct tasks, summarization and emotion detection?
    • What are the 8 sentiment classes?
    • Would it identify specific parts of the writing that lead to different reactions? How?
    • What exactly is meant by "text pertaining to the character"?

• Methodology (== How to do)

  • What technique will be used?
  • What models/procedures would be used for the task?
  • How would the task be done?
    • How will you identify the characters in the story?
    • How would inferences be drawn from the corpus?
    • How would meanings be extracted from the question?
      • How would they be represented?
      • How would that be used to query the knowledge base?
    • How would the question be transformed into a geographical database query?
    • How would the prediction work?

• Dataset

  • What language would the data be in?
  • What corpus will be used?
  • Is there any particular reason you want to use a certain language?
  • Will you use spoken language or written? (If written, how is it relevant to dialogue systems?)

• Evaluation

  • How will the result be evaluated?
  • What defines a good data for a certain task?
  • Does the content come with explicit reviews (such as a star rating)?
  • If not, then perhaps a site can be found (like yelp) with such ratings? That can then be crawled to get data for training the models.
  • If the task is performed on unlabeled data, how will performance be evaluated?

• Advice

  • Creating an application is great, but it shouldn't consume too much of the effort; the majority of the effort should be spent on language processing.
  • You should look at the particulars of your corpus, and also do a small literature review to see what kind of techniques have been used in the field.
  • It would be good to do a short review of existing methods and what they can accomplish.