NER is the abbreviation of Named Entity Recognition.
OFBiz NER plugin provides a set of services to analyze product contents which may contain up to 6 classes, such as brand (B), color (C), garment type/style (G), garment number (N), size (S) and other (O).
OFBiz NER plugin uses IKAnalyzer to token sentences and Stanford CoreNLP (NER) to recognize entities in sentences.
IKAnalyzer uses its own dictionary to token Chinese words and Apache Lucene Analyer to token English words. In our test result, it is the best in open source (Apache License V2.0).
Why Stanford CoreNLP
Our requirement is similar to the situation described in this eBay paper. From the results of this paper we can see, CRF algorithm is the best.
Stanford CoreNLP uses CRF and open source (GLP v3+).
This plugin contains a folder in plugins and a folder in runtime, deploy them in OFBiz and start OFBiz.
Please note, this plugin depends on OFBiz Lucene and PriCat plugins, when training a new model, the log messages will be output to web page by htmlreport.
To avoid Chinese charaters encoding problem, please set file.encoding to UTF-8 in build.gradle, and expand jvm memory to a larger number:
def jvmArguments = ['-Xms1024M', '-Xmx4096M', '-Dfile.encoding=UTF-8']
Thanks Menghan Sun built the first version of this plugin in the summer of 2016. Wish her doctoral studies in CUHK happy and fun.