Knowledge entity recommendation method and system based on heterogeneous network embedding

Abstract

The invention discloses a knowledge entity recommendation method and system based on heterogeneous network embedding. The knowledge entity recommendation method specifically comprises the steps: designing a knowledge entity association model, and constructing a heterogeneous network through the aggregation of multiple types of knowledge entities and the associated data; generating node feature vectors of different dimensions based on different random walk element path frameworks through a heterogeneous network mapping algorithm; based on the node feature vectors, performing similarity calculation between the nodes by utilizing cosine similarity and a linear weighting method; and developing recommendation services from three types of knowledge entity recommendations based on types, association-based knowledge entity recommendations, or structure-based knowledge entity recommendations. According to the knowledge entity recommendation method, global feature learning is carried out on theheterogeneous network through a network embedding algorithm, and effective recommendation of all knowledge entities is realized.

Description

technical field [0001] The invention belongs to the technical field of data mining, and in particular relates to a method and system for recommending knowledge entities based on heterogeneous network embedding. Background technique [0002] The recommendation system is an important mechanism to solve information overload, and it is the basis for the good operation of the information service platform. A large number of recommendation methods and systems have been proposed, including content-based recommendation, collaborative filtering, analysis based on graph mining, and so on. [0003] However, due to the "power-law distribution" characteristics of knowledge entities, some problems often arise when developing recommendation services: (1) High computational complexity: data shows sparseness, and computational complexity shows exponential growth with scale expansion Trend; (2) Single strategy: recommendation services often ignore the use of long-tail knowledge entities, but ...

AI Technical Summary

Problems solved by technology

[0003] However, due to the "power-law distribution" characteristics of knowledge entities, some problems often arise when developing recommendation services: (1) High computational complexity: data shows sparseness, and computational complexity shows exponential growth with scale expansion Trend; (2) Single strategy: recommendation services often ignore the use of long-tail knowledge entities, but are influenced by popular knowledge entities, and this differentiation is further strengthened in the recommendation process

Existing studies have introduced network embedding methods into recommendation systems, which can perform recommendation calculations in low dimensions while retaining the global characteristics of the network, but the research on heterogeneous networks has not been fully explored

Images