Research On Person Event Extraction Method For Computer Domain

Nowadays,the development of artificial intelligence technology is becoming more and more rapid,and knowledge graph has gradually received increasingly extensive attention.Under this situation,knowledge graph of discipline domain has emerged,and their applications in the field of education have become ever more important.Meanwhile,the knowledge graph of computer discipline domain plays a critical role in computer education,and the person event is an important part of the discipline knowledge graph.Adding person events to the graph will make the discipline knowledge graph richer and more interesting,and enhance enthusiasm of students for learning.The most important step in building the person event knowledge graph is event extraction.Therefore,this project takes building a person event knowledge graph in the field of computer discipline as an example.We build a corpus of person event extraction in the field of computer discipline,and design a person event extraction method based on BiGRU and CNN models,as well as obtain person event trigger words and event arguments,so as to build the person event knowledge graph in the field of computer discipline.The main research work of this project is as follows:1)A corpus of person event extraction for the computer domain is constructed.According to the characteristics of person texts in the computer field,this paper constructs the corpus of person events.The entities of person are extracted according to the curriculum concepts in the computer field,mainly to correspond the curriculum concepts with the person one by one.At the same time,according to the extracted person entities,the texts related to the persons are crawled from Wikipedia and other related websites as the original corpus,so as to construct the dataset for the person event extraction of the computer field.After that,we define the event trigger word types and event argument types according to the characteristics of the person events in the computer domain and the needs of extraction,and label the dataset for person event extraction in the computer domain according to the predefined event schema.2)This paper designs a computer domain person event detection method(SC-BiGRU-CNN)based on the serial fusion of BiGRU and CNN.In this paper,we extract the global semantic features of the person text by using a bidirectional GRU model,and input the extracted global semantic features into the CNN model to extract the local semantic features of sentences,in order to further extract a more fine-grained semantic feature representation.Meanwhile,the SC-BiGRU-CNN method is used to input the extracted features into the Softmax classifier to complete the classification of person event trigger words.The experimental results show that the proposed method has a significant improvement in F1 indicators compared with the baseline method.3)This paper designs a BiGRU-Attention-CNN-based method for extracting person event arguments in computer domain(Att-BiGRU-CNN*),which uses BiGRU to obtain long-range semantic dependency information of the input vector.Afterwards,the Attention layer is introduced to process the features outputted through the BiGRU layer to obtain more accurate global features.Then,the extracted global semantic features are passed through a convolutional layer and a dynamic multi-pooling layer to extract local semantic features in order to obtain the final features.Finally,the extracted final features are used to classify the person event arguments.Based on the experimental results,it can be concluded that the Att-BiGRU-CNN* method has improved the F1 indicators compared with the baseline method.