Text label mining method based on comparative learning
A text labeling and labeling technology, applied in neural learning methods, unstructured text data retrieval, text database browsing/visualization, etc., can solve the problems of uncontrollable text labels, unsatisfactory effects, and inability to generalize general labels.
Pending Publication Date: 2021-12-17
武汉众智数字技术有限公司
View PDF0 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
The extractive text label mining method extracts entities and text fragments in the text as text labels, so it cannot generalize generalized labels that do not appear in the text, which also leads to unsatisfactory results for upstream applications that rely on text label mining. question
Compared with the extraction method, the Transformer-based generative text label mining method can overcome the generalization problem of generalized labels that do not appear in the text, but the output text labels are uncontrollable, and even the generated labels do not conform to the lexical specification.
In addition, the training of supervised learning models relies on high-quality data marked manually, which requires a lot of labor costs
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment 1
[0031] This embodiment discloses a text label mining method based on contrastive learning, such as figure 1 with 2 ,include:
[0032] S100. Encode each label in the label set by a label expresser to obtain a label representation vector, and store all label representation vectors in a vector retrieval library;
[0033] Specifically, in S100, the encoding process of the label representation is as follows: each label in the label set is encoded through the representation layer, and then the output of the encoding layer is pooled through the pooling layer to obtain the deep semantic representation vector of the label , and finally the L2 norm normalization operation is performed on the pooled semantic vector through the L2Norm module.
[0034] S200. Encode the text of the label to be generated in the text set using a text representation to obtain a text representation vector;
[0035] Specifically, in S200, the encoding process of the text representation is as follows: encode e...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
A text label mining method based on comparative learning comprises the steps: coding each label in a label set through a label representation device, obtaining label representation vectors, and storing all the label representation vectors in a vector retrieval library; encoding the text to be subjected to tag generation in the text set by adopting a text representation device to obtain a text representation vector; retrieving N tags most similar to the text representation vector from the vector retrieval library, filtering through a set threshold value, and returning a final tag as the tag of the text. The problem that a supervised learning model depends on high-quality annotation data and consumes a large amount of labor cost in the prior art is solved.
Description
technical field [0001] The invention relates to the field of label mining, in particular to a text label mining method based on contrastive learning. Background technique [0002] Text tags are a streamlined refinement of text content, usually used to express text topics or key information, so that users can efficiently obtain the content they are interested in. The extractive text label mining method extracts entities and text fragments in the text as text labels, so it cannot generalize generalized labels that do not appear in the text, which also leads to unsatisfactory results for upstream applications that rely on text label mining. question. Compared with the extraction method, the Transformer-based generative text label mining method can overcome the generalization problem of generalized labels that do not appear in the text, but the output text labels are uncontrollable, and even the generated labels do not conform to the lexical specification. . In addition, the ...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More Patent Type & Authority Applications(China)
IPC IPC(8): G06F16/34G06F40/30G06K9/62G06N3/08
CPCG06F16/345G06F40/30G06N3/08G06F18/22
Inventor 张宇
Owner 武汉众智数字技术有限公司