UETsegmenter

UETsegmenter is a toolkit for Vietnamese word segmentation. It uses a hybrid approach that is based on longest matching with logistic regression.

UETsegmenter is written in Java and developed in Esclipse IDE.

Note

UETsegmenter was inherited in UETnlp. UETnlp is a toolkit for Vietnamese text processing which can be used for word segmentation and POS tagging. UETnlp is much easier to use than UETsegmenter.

Overview

• src : folder of java source code

• uetsegmenter.jar : an executable jar file (see How to use)

• models : a pre-trained model for Vietnamese word segmentation

• dictionary : necessary dictionaries for word segmentation

How to use

The following command is used to run this toolkit, your PC needs JDK 1.8 or newer:

java -jar uetsegmenter.jar -r <what_to_execute> {additional arguments}

	-r	:	the method you want to execute (required: seg|train|test)

Additional arguments for each method:

• -r seg : Method for word segmentation. Needed arguments:

-m <models_path> -i <input_path> [-ie <input_extension>] -o <output_path> [-oe <output_extension>]

	-m	:	path to the folder of segmenter model (required)
	-i	:	path to the input text (file/folder) (required)
	-ie	:	input extension, only use when input_path is a folder (default: *)
	-o	:	path to the output text (file/folder) (required)
	-oe	:	output extension, only use when output_path is a folder (default: seg)

• -r train : Method for training a new model. Needed arguments:

-i <training_data> [-e <file_extension>] -m <models_path>

	-i	:	path to the training data (file/folder) (required)
	-e	:	file extension, only use when training_data is a folder (default: *)
	-m	:	path to the folder you want to save model after training (required)

After training, the models_path folder will contain 2 files: model and features.

• -r test : Method for testing a model. Needed arguments:

-m <models_path> -t <test_file>

	-m	:	path to the folder of segmenter model (required)
	-t	:	path to the test file (required)

APIs

3 APIs for Vietnames word segmentation are provided:

• Segment a raw text:

	String modelsPath = "models"; // path to the model folder. This folder must contain two files: model, features
	UETSegmenter segmenter = new UETSegmenter(modelsPath); // construct the segmenter
	String raw_text_1 = "Tốc độ truyền thông tin ngày càng cao.";
	String raw_text_2 = "Tôi yêu Việt Nam!";

	String seg_text_1 = segmenter.segment(raw_text_1); // Tốc_độ truyền thông_tin ngày_càng cao .
	String seg_text_2 = segmenter.segment(raw_text_2); // Tôi yêu Việt_Nam !

	// ... You only need to construct the segmenter one time, then you can segment any number of texts.

• Segment a tokenized text:

	// ...
	// ... construct the segmenter

	String tokenized = "Tôi , bạn tôi yêu Việt Nam !";
	String segmented = segmenter.segmentTokenizedText(raw_text_2); // Tôi , bạn tôi yêu Việt_Nam !

• Segment a raw text and return list of segmented sentences:

	// ...
	// ... construct the segmenter

	String text = "Tốc độ truyền thông tin ngày càng cao. Tôi, bạn tôi yêu Việt Nam!";
	List<String> segmented_sents = segmenter.segmentSentences(text); // [0] : Tốc_độ truyền thông_tin ngày_càng cao .
																	// [1] : Tôi , bạn tôi yêu Việt_Nam !

Citation

If you use the toolkit for academic work, please cite:

@INPROCEEDINGS{7800279, 
	author={T. P. Nguyen and A. C. Le}, 
	booktitle={2016 IEEE RIVF International Conference on Computing Communication Technologies, Research, Innovation, and Vision for the Future (RIVF)}, 
	title={A hybrid approach to Vietnamese word segmentation}, 
	year={2016}, 
	pages={114-119},
	doi={10.1109/RIVF.2016.7800279}, 
	month={Nov},
}

The approach used in the toolkit is also explained in the paper.