Research On Industrial Product Relationship Extraction Method For Business Intelligence

With the rapid development of economy and society and the arrival of the era of big data,the acquisition of business intelligence is particularly important for enterprises to make scientific decisions,rationally optimize industrial resources and enhance the competitiveness of enterprises.There are huge amounts of information about the industrial products on the Internet,such as product launch site,product evaluation and electricity class web site.They all have various types of industrial product information.The industrial product information that exists on the Internet is used to construct a product knowledge graph.The method of intelligence mining and obtaining business intelligence has become the main research direction.The Internet is filled with a lot of redundant and even false information.To build a high-quality product knowledge graph,relationship extraction is an indispensable and key step.The existing methods have many shortcomings,and they are still worthy of further research and improvement.In this paper,the product knowledge graph is applied in the analysis and acquisition of business intelligence.Based on the mobile phone industry,the key technologies of relation extraction involved in the construction of the knowledge graph are deeply studied,and focusing on the improvement of relation extraction methods for text information and multimodal information.This paper's work is as follows:(1)For the existing text information relationship extraction work,only a single word vector model is used,which cannot fully capture all the semantic information of the sentence and thus limits the total information input of the model;the existing single neural network model cannot be better comprehensively utilized Contextual information and full grasp of local features.This paper proposes a framework that uses multiple word vector models to map the same corpus to form a multi-channel framework;and each channel uses convolutional neural networks and bidirectional long-and short-term memory networks.Combined method.By comparing with other models on the data set,it proves that the framework has significant effects.(2)Aiming at the phenomenon of network degradation caused by the increase of network model depth in the process of image feature extraction in the extraction of multimodal information relations;the incomplete and orderly expression caused by direct splicing of multi-modal information feature vectors For the problem of multi-modal information feature vector,this paper proposes a cross-modal relationship extraction model for multi-modal information,uses deep convolutional neural network combined with residual network to complete the feature extraction of image information,and then uses neural network model to analyze the cross-modal relationship.The eigenvectors of the modal are represented collaboratively.Experiments on data sets prove that the proposed cross-modal relation extraction model has higher accuracy.(3)Based on the mobile phone industry,this paper constructs a product knowledge graph platform for business intelligence,and provid the corresponding interface business intelligence analysis tools.Through the encapsulated intelligence analysis method center,it conducts intelligence mining and analysis to obtain the intelligence knowledge base.Finally,Human-computer interaction is carried out through the intelligence service module to obtain useful business intelligence for the enterprise.