چكيده به لاتين
The detection system of named entities is a tool for processing natural language. It extract named entities or the same specific names in the text, such as individuals, organizations, and locations from non-structured texts. today, despite the limited work done in this regard for persian language, the existing methods have not been able to achieve desirable effects on european languages such as english language.
One of the main reasons for this problem is the lack of proper text corpora that have been studied in some way in recent years. another reason for this issue is the lack of adequate attention researchers in persian language processing field using newest approaches, especially approaches based on deep neural networks. While the use of deep neural networks in domains such as image processing, signal, natural language, and even many other areas has led to significant improvements.
for this reason, several architecture based on deep neural network have been proposed to identify named entities in persian language. In the development of these architectures, two approaches are based on specific language clues and context-based methode. the first approach tries to use the features of syntactic, structural, gazetteers and semantic features in detecting named entities in order to further enhance the system. in fact, this approach is based on the embedding of information derived from these specific language clues including information based on the dependency tree, the part of speech, gazetteers, and the meaning of words. This approach has managed to increase the accuracy of the knowledge boundaries by 2.53 %, reaching 79.98 percent in the scale.
While the second approach attempts to use transfer learning from the embedding of the words constructed for the language model in the BERT model to use information based on the context words that has managed to improve the accuracy of 80.17 percent in the scale.
