Term and Vector-based Approach

• Documents are represented by "bags of words" - essentially these are the conflated word stems plus statistical information of the words appearing in the document.

• Problems:

  • "bus accidents over the holidays", "car accidents during New Year's Eve", "nuclear accidents in Ukraine"
  • The 1st and 2nd `accidents' should be more similar than the 1st and the 3rd.
  • "New Year's Eve" should be treated as one single word.

• Structure information is needed for higher precision.