Skip to content

Latest commit

 

History

History
101 lines (87 loc) · 5.47 KB

README.md

File metadata and controls

101 lines (87 loc) · 5.47 KB
Raw

Spell Corrector Readme

You can use this tool to generate noisy channel model parameters, test the performance and prune the model. Then you can use the model to do spell correction task.

  • Project name
spelling-corrector (This is a java project)
  • Project orgnization
project |-src |-lib |-bin |-out |-data |-build.xml

Build project

In the project SpellCorrectorBuild root directory, use ant command to build the project, and ouput a jar file SpellCorrectorBuild.jar in directory ./out/ .

How to use

  1. Prepare the dictionary file words.txt , original spell data file final.out and parameter file parameter , and putSpellCorrectorBuild.jar in the same directory.
  2. Use command java -jar SpellCorrectorBuild.jar to run the project.
  3. You can get a test_result.txt file to save the test infomation, and channle_data.txt to save the noisy channel model parameter.

File descriptions

  • channel_model.txt The file is like the format: (word_slice key_slice log_probability)
  • parameter A json format file:
{

	"model_file": "chnnel_data.txt",
	"train_file": "final.out",
	"dic_file": "words.txt",
	"equal_prob": 0.9,
	"most_dis": 2,
	"context_num": 2,
	"transfer_freq": "loglog",
	"top_num": 3,
	"train": "yes",
	"test": "yes",
	"prune": "no"
}
  1. equal_prob is the probability of p(x|y) where x=y, you can tune equal_prob by a validation method.
  2. most_dis is the most edit distance in your application, usually you can set as 2.
  3. context_num is the context window you use in the model, the larger context you use the more precise model you will get, but the model parameters will be increased. Usually you can set as 2.
  4. transfer_freq is the type of approache you use to smooth your frequency. You can set as loglog, log and no. Using loglog soothing approache will get the highest precision. Using no means that you will use the original frequency.
  5. top_num is the number of ranked candidates you will get.
  6. train, test and prune are the parameters that if you want to train the model, test the model or prune the model again. If you want, you can set it as yes, otherwise you can set it as no.
  7. Another parameter smooth value is the value you will use when the score of a pair is not in your model due to the sparseness of the model. We calculate the value by the average of the 10 smallest data in channle_data.txt file.

Example data

In the data directory, there is some example data. You can refer to it.

  • final.out is the original train file, it's of json format. A line represents a case of miss spelling. The format is as follows.
  1. word is the word user wants to input.
  2. key is the word user actually inputs.
  3. cnt is the number of this case in the input method engine (IME) user data.
  4. match_type is wether this input is precise or predict by the IME.
  5. cor_type is wether this input is a "corrector" type or "spell check" type (we only consider spell check type, because IME can correct missing input by keyboard position),
  6. If cor_type is "spell check", then spell_info is the information of this spelling. It includes the follows. spell_in is the word user actually inputs. spell_out is the word user wants to input. spell_type is the missing input type, including __ins__ (user inserts an additional letter), __del__ (user deletes a letter), __tra__ (user transposes two letters). spell_pos is the position of this missing input. predict_type is wether this input is precise or predict by the IME. evidence_len is the length of the word user wants to input.
  • words.txt is the vocabulary file.
  • test_data.txt is the test file, and each line is a missing input case. The first column is the word user wants to input, and the second column is the word user actually inputs.
  • test_result.txt is the test results of the model. prune_data.txt is the pruned model file. channel_data.txt and channel_data_loglog.txt are the output model file.

Presentation

For more details about the spell correction techniques, you can follow this tutorial here. Reference

License

spelling-corrector is published under MIT License

Copyright (c) 2015 Yu Gong (@pangolulu)

Permission is hereby granted, free of charge, to any person obtaining a copy of
this software and associated documentation files (the "Software"), to deal in
the Software without restriction, including without limitation the rights to use,
copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the
Software, and to permit persons to whom the Software is furnished to do so,
subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.