A label semantic learning method based on a network structure and semantic relevance measurement

Abstract

The invention relates to a label semantic learning method based on a network structure and semantic relevance measurement, the method comprising the steps of initializing a label network to obtain a fact label network G based on the fact of user behavior; constructing a label network GR after the protocol stipulation according to the fact label network G; applying an improved random walk strategyto label network GR, and constructing a label network GC based on the random walk strategy; constructing a label network GT based on label-related text information; normalizing the label network GC and the label network GT, and learning the label semantic vector representation by the random walk strategy and the word vector learning method. The method of the invention is reasonable; the method makes full use of the network topology, takes into account the relevant text information contained in the node, can learn label semantic vectors, which are easy to operate, high in confidence, adequatein expression and low in noise from the topology and text expression in a relatively short period of time, can be widely used in the label network learning and label semantic learning of the text setcontaining labels.

Description

technical field [0001] The invention belongs to the technical field of network representation learning, in particular to a tag semantic learning method based on network structure and semantic correlation measurement. Background technique [0002] In the network representation learning technology, text semantic learning is mainly to represent the features of the text, that is, to represent the target text object (word, sentence, paragraph, article) as a numerical value (single value, vector or matrix). In view of the needs of computing and semantic modeling applications, the currently commonly used models are Latent Semantic Analysis (LSA) based on matrix singular value decomposition, latent Dirichlet Allocation (LDA) based on probability model, and neural network-based solution The word vector representation model NNLM (Neural Network Language Model) and word2vec, etc., these models are mainly used for long text. In the face of short texts with high sparseness and high nois...

AI Technical Summary

Problems solved by technology

[0004] To sum up, the existing algorithms are difficult to solve the problems that the original label network comes from the unconstrained collaborative facts of users, the information is noisy and changeable, the topic boundary is not clear or the topic drifts, and the semantic model fails.

Images