Construction And System Development Of Knowledge Graph In Lithium Mining

With the relatively mature development of the general knowledge graph,the domain knowledge graph of different research objects is in urgent need of professionals to supplement the gaps in the research field.Moreover,as a new energy metal,lithium ore is an indispensable research object in the current stage of rapid development of new energy automobile industry.Therefore,this thesis selects the field of lithium ore to construct its knowledge graph,which plays a vital role in scholars' understanding of the field,provides knowledge support for experts in the field,and further enriches the content of the domain knowledge graph.In this thesis,11,748 corpus data were crawled from CNKI and related websites in the field of lithium to form a corpus in the field of lithium.At the same time,triples of entity attributes were crawled from the dictionary in the field of lithium.Secondly,according to the characteristics of the close context of Chinese language and the consideration of contextual information by Bert,Bert-Bi LSTM-CRF is innovatively applied to the data in the field of lithium ore.Then,in relation extraction,this thesis proposes a method to consider the position features of entities in statements and combine them with word vectors to form feature vectors,and applies the Attention mechanism combined with Bi LSTM in relation extraction model in the field of lithium ore.Then the knowledge is merged and stored.Finally,the domain knowledge graph is applied and the domain knowledge graph system is developed.To sum up,this thesis mainly carries out research from three aspects: knowledge acquisition,construction of domain knowledge graph of lithium mining and development of domain knowledge graph system of lithium mining:1.Knowledge acquisition: It can be divided into entity identification in the field of lithium ore and relationship extraction in the field of lithium ore.In the aspect of entity recognition in this field,we replace the static word vector technique word2 vec in the Bi LSTM-CRF model with the Bert pre-training model.Compared with the Bi LSTM-CRF model and the Bi LSTM model,the F1 value is increased by 1.53% and 23.37% respectively in the experiment.In terms of relation extraction in this field,the position feature vector is considered to enhance the data,and the attention mechanism with low model complexity is added to the Bi LSTM neural network model to extract the relation type information in statements,and 73.9% of the F1 value is obtained.2.Lithium domain knowledge graph: knowledge extraction is carried out to obtain triad information based on a large number of language data,and knowledge merging is conducted with encyclopedia attribute triad obtained by crawler.Finally,about 1600 obtained triad data are imported into Neo4 j graph database for recording,storage and updating.3.Lithium mining domain knowledge graph system: Springboot and Vue technology are used to conduct front-end and back-end separate development of the system,and the system finally has the functions of knowledge visualization,knowledge query,knowledge management and user login management,so as to realize the operation of knowledge addition,deletion,modification and check,which is beneficial to the update and maintenance of knowledge graph.