A joint clustering method for large-scale heterogeneous data

Abstract

The invention discloses a large-scale heterogeneous data oriented co-clustering method. The method comprises the following steps that entities and a heterogeneous relation between the entities are extracted from the heterogeneous data to obtain a heterogeneous relation matrix; the entity X2 of the small scale is selected from two corresponding entities in the heterogeneous relation matrix, and an incidence matrix is set according to an incidence relation of the entity X2; a symmetric matrix sparse decomposition method is adopted to decompose the incidence matrix to obtain a clustering instruction matrix B corresponding to the entity X2; the matrix B is used as an input, tri-decomposition is carried out on heterogeneous relation moment R to obtain a clustering instruction matrix corresponding to an entity X1, and entity type division is achieved through the clustering instruction matrix corresponding to the entity X1 and the clustering instruction matrix corresponding to the entity X2. According to the method, sparsity of the matrixes can be reduced, and the accuracy of the co-clustering method is improved.

Description

technical field [0001] The invention belongs to the field of Internet information mining, and in particular relates to a joint clustering method for large-scale heterogeneous data, which can reduce the sparsity of large-scale heterogeneous data. Background technique [0002] With the rise of heterogeneous information networks such as Weibo and social networks, heterogeneous information mining has become a research hotspot in the field of data mining. A heterogeneous network contains many types of entities, and there are complex interactions among entities. For example, Weibo contains entities such as users, messages, tags, words, etc. When a user publishes a message, the message is composed of words, and the message also includes tags. By extracting the relationship data between entities and performing joint clustering analysis, the potential structural relationship between different entities in a heterogeneous network can be mined. [0003] Non-negative matrix factorizati...

AI Technical Summary

Problems solved by technology

[0004] (1) Unbalanced problem: When the scale of heterogeneous data to be analyzed increases, the scale of different types of entities in the heterogeneous data does not show a uniform growth pattern

The time complexity of the traditional non-negative matrix factorization method is related to the row and column scale of the matrix, so the computational time complexity is high when dealing with large-scale data

[0005] (2) Sparsity problem: relational data in a real heterogeneous network is relatively sparse, and as the scale of heterogeneous data to be analyzed further increases, relational data becomes extremely sparse

The traditional non-negative matrix factorization method does not work well for the abnormally sparse relationship matrix

Images