Research On Feature Fusion Method For Low Quality Cultural Data

In recent years,the digital trend of traditional cultural works of art is becoming more and more obvious.In the face of massive cultural digital resources,how to quickly and accurately mine the cultural theme of these data,so as to explore its rich cultural content to the greatest extent,has become one of the urgent problems to be solved in the field of cultural analysis research.Using computer,artificial intelligence and other methods for cultural analysis is an effective method,but different from other data,cultural data has the characteristics of low quality,which is mainly reflected in heterogeneity,disorder and incompleteness.Therefore,the traditional data fusion method is difficult to be directly applied to cultural data.To solve the problems existing in the feature fusion of low-quality cultural data,according to the data characteristics,this thesis analyzes the shortcomings of existing research methods,and puts forward targeted solutions.Firstly,aiming at the problems of heterogeneous and disordered cultural data,a heterogeneous cultural data clustering method based on BP neural network is proposed;Secondly,for the cultural data with incomplete characteristics,a subject classification method of incomplete cultural data based on CR-LDA is proposed.The main work of this thesis is as follows:(1)Starting with the characteristic information of heterogeneous cultural data,its cultural characteristics are classified from multiple dimensions to form a heterogeneous cultural characteristic information matrix containing all kinds of characteristic information.The correlation degree of geography,time,art and subject characters is calculated from each dimension of the matrix by using their own correlation degree calculation formula.The data set is substituted into BP neural network for training,and the measurement standard of correlation degree between heterogeneous cultural data is formed.Taking the correlation degree as the clustering basis,the fuzzy C-means clustering algorithm is used to cluster all heterogeneous cultural data.The fuzzy C-means is more sensitive to the selection of the initial clustering center.Here,the membership degree between the initial clustering centers is limited,and the membership degree between each other is required to be greater than the constraint threshold,so that the clustering results not only meet the closeness between the same categories,but also ensure the separation of various categories.(2)Starting from the characteristics of incomplete cultural data,according to the characteristic information contained in it,find the associated heterogeneous cultural data from the heterogeneous cultural data set,and expand its corresponding characteristic information.In LDA,the correlation parameters between cultural theme layer and cultural characteristics are added.The model is established through the classification results of the data in the complete heterogeneous cultural data set and the characteristic information contained therein.Using the existing and expanded characteristic information of incomplete cultural data,the model is deduced by Gibbs sampling method to mine cultural themes.(3)Build an experimental platform to analyze the performance of the algorithm proposed in this thesis.The experimental results based on heterogeneous cultural data sets show that the improved fuzzy C-means clustering strategy based on BP neural network can gather highly correlated heterogeneous cultural data and solve the problem of disorder of heterogeneous cultural data.Experiments on incomplete cultural data sets show that CR-LDA model has achieved good results in cultural topic mining for incomplete cultural data.The feature fusion method for low quality cultural data proposed in this thesis has good performance after experimental verification,and has certain theoretical and engineering reference value for the research in the field of cultural data analysis.