Knowledge graph-based generative zero sample prediction method

Abstract

The invention discloses a knowledge graph-based generative zero sample prediction method, which comprises the following steps: constructing a knowledge graph fusing various semantic information by taking a hierarchically structured category as a category node and taking category connection attribute description, text description and external knowledge as additional nodes; encoding semantic information of the knowledge graph by adopting a graph neural network algorithm to generate category vector representation; and using the generated category vector representation as an input to a generationmodel to generate a sample of the category for learning and prediction of a zero sample learning algorithm. According to the method, the knowledge graph fusing various semantic information is constructed, and the samples with richer features and higher inter-class discrimination are generated for each invisible class based on the knowledge graph, so that the prediction problem of the invisible class samples is well solved.

Description

technical field [0001] The invention relates to the field of generative zero-sample learning, in particular to a generative zero-sample prediction method based on knowledge graphs. Background technique [0002] Zero-shot Learning (ZSL) is an important branch in the field of transfer learning, which is mainly used to deal with the problem of missing samples in supervised learning. Typical supervised learning requires manual labeling of training samples to guide the machine learning model to extract features, and the labeling of samples often requires a huge amount of manpower and financial resources, especially in classification problems, when some new categories appear, it is necessary to manually Label hundreds or thousands of training samples. The heavy sample labeling work makes the model difficult to generalize. [0003] Zero-sample learning technology can deal with the problem of model learning and prediction in the absence of training samples, that is, using the sema...

AI Technical Summary

Problems solved by technology

However, there are many noises in these descriptions, and it is difficult to extract useful information

Images