tuem / resembla

Resembla: Word-based Japanese similar sentence search library

Features

• Candidate elimination using N-gram index and bit-parallel edit distance computation
• Word, kana and romaji-based edit distance variables and their ensemble
• Support vector regression with linguistic features

Requirements

• MeCab
• LIBSVM
• ICU
• C++11 compiler

Other included libraries

• Catch
• cmdline
• JSON for Modern C++
• SimString
• paramset

Running example

• install MeCab, LIBSVM, ICU and a C++11 compiler

  • if you use CentOS 7, see Wiki

• clone, build and install Resembla

cd /var/tmp
git clone https://github.com/tuem/resembla.git
cd resembla/src
make
sudo make install
cd executable
make
sudo make install
#optional
cd /var/tmp/resembla/misc/mecab_dic/unidic/
./install-unidic.sh
cd /var/tmp/resembla/misc/mecab_dic/mecab-unidic-neologd/
./install-mecab-unidic-neologd.sh

• run with example files

# on src/executable
./resembla_index -c ../../example/conf/name.json
./resembla_cli -c ../../example/conf/name.json
# input some names like 'タケダ'
./resembla_index -c ../../example/conf/address.json
./resembla_cli -c ../../example/conf/address.json
# input some addresses like '京都北区'
# you may need to run install-unidic.sh or edit configuration file
./resembla_index -c ../../example/conf/apple.json
./resembla_cli -c ../../example/conf/apple.json
# input some sentences like 'りんごおいしくねえ'