Research And Implementation On Mongolian Relationship Extraction For Tourism Field

Nowadays,people usually choose the travel service software to search for information about the destination before traveling.Relationship Extraction for Tourism is an important foundation for the construction of Tourism knowledge graphs and the development of intelligent recommendation systems.At present,the relevant research methods for Tourism Relationship Extraction are more mature in Chinese,English and other major languages,but the research on Mongolian Relationship Extraction is still in its infancy.In order to promote the intelligent development of Tourism in minority areas,it is important to carry out the research on Mongolian Relationship Extraction for the Tourism field.This dissertation focuses on the Relationship Extraction of Mongolian Tourism,and the main works are as follows:(1)A corpus of Mongolian Relationship Extraction for the Tourism field is established.Because of a small public corpus in the Mongolian Tourism field,this dissertation uses crawler technology to collect relevant textual information for the Tourism field,translate and correct the Chinese text and merge it with Mongolian text information into a new dataset.After pre-processing and manual annotation,a corpus of 28 types of relations and 74,699 utterances in the Mongolian Tourism field was finally constructed.(2)According to the characteristics of Mongolian word formation,Mongolian Relationship Extraction model based on attention mechanism is constructed.The model uses Bidirectional Long Short-Term Memory Network(BiLSTM)for feature extraction and introduces an attention mechanism to optimise the feature vector on the basis of this model.The experimental results show that the F1 value of the Mongolian Relationship Extraction model is improved 4.9% compared with the baseline model in this dissertation.(3)To solve the problems of multiple meanings of words in the corpus,this dissertation proposes a Relationship Extraction method that combines pre-trained language model.In this method,the Mongolian BERT(Bidirectional Encoder Representations from Transformers)model is pre-trained according to the Mongolian word formation,on this basis,two Mongolian Relational Extraction models,BMBiLSTM-Attention and BM-BiGRU-Attention were constructed.The experimental results show that it was achieved a better result by using the method of Mongolian BERT of vector represention,and the F1 values of the above two models in the Relationship Extraction task are more than 70%.Based on the above researches,this dissertation designs and builds a Mongolian Relationship Extraction system,which realizes the Mongolian Relationship Extraction service for the Tourism field.