Cross-media scientific research resource feature extraction model training method and device and cross-media scientific research resource feature extraction method and device

Abstract

The invention provides a cross-media scientific research resource feature extraction model training method, a cross-media scientific research resource feature extraction method and a cross-media scientific research resource feature extraction device. The training method comprises the following steps: acquiring scientific research resource data of different media sources based on a cross-media scientific research resource data real-time acquisition system and performing data structured processing; performing model training and optimization on the first deep learning model according to the scientific research resource text data to form a scientific research resource text semantic feature extraction model; and performing model training and optimization on the second deep learning model according to the scientific research resource image data to form a scientific research resource image semantic feature extraction model. The method can be specially suitable for semantic feature learning of the scientific research resource data, improves the reliability, convenience and effectiveness of cross-media scientific research resource data extraction, and can improve the characterization capability of a deep learning model trained by adopting the scientific research resource data during feature extraction for the scientific research resource data. And the comprehensiveness, diversity, accuracy and reliability of the extraction result are improved.

Description

technical field [0001] The present application relates to the technical field of semantic feature learning, and in particular, to a feature extraction model training method, feature extraction method and apparatus for cross-media scientific research resources. Background technique [0002] In order to improve the retrieval convenience of scientific research resource data and realize the automatic matching between scientific research resource data and social needs, it is necessary to carry out data collection and semantic feature learning training on scientific research resource data to form a database for matching scientific research resource data and retrieval keywords. Then, the deep learning model is applied to extract semantic features with a high degree of automation for scientific research resource data. [0003] At present, in order to achieve semantic feature extraction for scientific research resource data, the existing method is usually to use a deep learning model...

AI Technical Summary

Problems solved by technology

[0003] At present, in order to realize the semantic feature extraction of scientific research resource data, the existing methods usually use deep learning models to extract data features. However, although deep learning models have general feature Some methods usually only use general-purpose data to directly train the deep learning model, so that the deep learning model cannot well reflect its representation ability when faced with scientific research resource data that requires accurate semantic expression and strong pertinence, that is, the influence The reliability and comprehensiveness of the subsequent application of the model for semantic feature extraction

At the same time, due to the wide distribution of scientific research resource data, most of the data presents a cross-media distribution and the form of expression includes not only text, etc., the existing methods cannot directly obtain these data and use them as training data

Images