Philgoo project status report

• What dataset will you be using? What does it look like?
  • I am using MUC-7 as in (Borthwick, 1998)
  • The dataset is devided to training set, dryrun set and formal set.
  • A set has multiple <DOC></DOC>s but I am not using features from position in document so I merged the whole.

• What did you do to get acquainted with the data?
  • I built a parser for the training set and test set

• Do you plan on looking at the same problem, or have you changed your plans?
  • I am looking for the same problem. However I am realizing getting comparable score to published models is not trivial.
  • Deep analyzing model structure, training method and data character will require stable output. I may consider using off-the-shelf code later but not now.
  • Other jobs than implementing and runnding classifiers such as morphological processing, feature selecting, gathering feature data etc are much bigger than expected.

• If you plan on writing code, what have you written so far, in what languages
  • Preprocessor for MUC7 NE data: rubi
  • Parser for MUC7 NE data: C++
  • HMM with joint likelihood: C++
  • HMM with conditional likelihood: C++ (in progress)

• What do you still need to do?
  • Implement more features
  • CRF with joint likelihood
  • CRF with conditional likelihood

• If you've run a baseline system on the data and gotten some results, what are they?
  • HMM with joint likelihood will function as the baseline.
  • The results accuracy is near 30% so much to improve.
  • Process on morphological level is required. CMU and CMUs should be the same organization which is not now.