Research On Text Entity Extraction Combining Related Image Information

With the rapid development of the Internet and the explosive growth of digital information,precious data's wealth has been brought.However,most of the network data is redundant data,which is not of great value,redundant data brings a great challenge to the task of information extraction.As the most important step in information extraction,entity extraction needs to be further improved.In this thesis,the existing method of extracting entities from text modal data using neural network is improved.In the first mock exam,the data types in network become richer with the development of network.The traditional entity extraction method can only start from a single mode,and it can't be considered the connection between multiple modalities simultaneously.The traditional entity extraction method is not well in cross-modality data.In order to complete the cross-modality entity extraction,this thesis designs the cross-modality entity extraction method combined with convolution neural network and other methods.Based on the existing research,this thesis proposes the lifting method and cross-modality entity extraction method of text modal entity extraction.In addition to the research of entity extraction method,this thesis also applies the entity extraction method.The main work of this thesis is as follows:(1)Research on text modal entity extraction.This thesis improves the existing text modal entity extraction method,and improves the entity extraction model by adding attention mechanism on the basis of Bi-directional Long Short-Term Memory neural network.The added attention mechanism can show the strong and weak relationship between different neurons by giving different words training weights in the network,and improve the ability of the network to process long text.Experimental results show that the improved method of text modal entity extraction is more effective than the original method.(2)Research on cross-modality entity extraction.In order to effectively extract entities from the cross-modality data including the image and corresponding text.A cross-modality entity extraction method is designed.Firstly,the image is transformed into a feature vector,and then the feature vector of the image is taken as the initial step of the Bi-directional Long Short-Term Memory network,which is fused with the subsequent input text information to extract entities from the fused data.Finally,the cross-modality entity extraction model is realized by this method.The additional image modal information and the original text modal information can complement each other in semantic expression,and improve the effect of entity extraction in cross-modality data.The experimental results show that the performance of cross-modality entity extraction model in cross-modality data is better than that of the method without image information fusion.(3)Cross-modality entity extraction system.Entity extraction is the basis of many artificial intelligence tasks.In order to realize the application of entity extraction method,this paper constructs a cross-modality entity extraction system.The system is a website system,through which entities can be extracted from text modal data or cross-modality data.Users can select entity extraction methods according to their own needs.The system also uses the technology of target detection,web crawler and text similarity calculation to expand the application of cross-modality entity extraction,which shows the importance of image modal information for cross-modality entity extraction.