TWI672597B - Automatic text labeling method and system - Google Patents

Abstract

The invention discloses an automatic text labeling method and system. The system comprises an artificial intelligence language processing module, an artificial intelligence text classification module, a label library and an artificial intelligence label output module. The artificial intelligence language processing module is configured to receive an input text and convert the input text into at least one specific dimension of the quantized feature; the artificial intelligence text classification module is configured to convert the at least one specific dimension of the quantized feature into at least one a characteristic matrix of a specific confidence level; the tag library is configured to store at least one tag that needs to be output; and the artificial intelligence tag output module is configured to convert the at least one specific confidence level feature matrix into at least one text tag and a corresponding Quantitative confidence level.

Description

Automatic text labeling method and system

The invention relates to an automatic text labeling method and system; in particular to an automatic text labeling method and system for artificial intelligence to parse text content text, which is used to give the text content text a possible label, so that the classified text can accelerate humanity through artificial intelligence. Classification efficiency, or replace human completion of text classification.

In today's data technology era, a variety of texts emerge in an endless stream, such as doctor's diagnosis, diversified news, and a large number of original articles from the media. Faced with such a rich variety of information, people urgently need some automated tools to help them find the key information they need accurately and quickly from the vast information ocean. Therefore, how to quickly and accurately analyze the text content and give it possible Labeling, which allows classified texts to accelerate human classification efficiency through artificial intelligence, is one of the important topics worthy of consideration by those with ordinary knowledge in the field.

The invention provides an automatic text labeling method and system for artificially analysing text content text, which is used to give the text content text to possible labels, so that the classified text can accelerate the efficiency of human classification through artificial intelligence, or replace human beings to complete text classification. The present invention provides an automatic text labeling system. The system comprises an artificial intelligence language processing module, an artificial intelligence text classification module, a label library and an artificial intelligence label output module. The artificial intelligence language processing module is configured to receive an input text and convert the input text into at least one specific dimension of the quantized feature; the artificial intelligence text classification module is configured to convert the at least one specific dimension of the quantized feature into at least one a characteristic matrix of a specific confidence level; the tag library is configured to store at least one tag that needs to be output; and the artificial intelligence tag output module is configured to convert the at least one specific confidence level feature matrix into at least one text tag and a corresponding Quantitative confidence level. The present invention further provides an automatic text labeling method comprising the steps of: first receiving an input text and converting the input text into at least one specific dimension of the quantized feature. Then, the quantized feature of the at least one specific dimension is converted into a feature matrix of at least one specific confidence level. After that, at least one tag that needs to be output is stored. Then, the at least one specific confidence level feature matrix is converted into at least one text label and a corresponding quantization confidence level.

The invention is best understood by reference to the detailed description and the accompanying drawings. Various embodiments are discussed below with reference to the drawings. However, those skilled in the art will readily appreciate that the detailed description of the drawings herein is for the purpose of explanation and description For example, the teachings presented and the needs of a particular application may result in a variety of alternative and suitable methods for implementing the functionality of any of the details described herein. Thus, any method may extend beyond the specific implementation selections in the following embodiments described and illustrated. Certain terms are used throughout the description and following claims to refer to particular elements. It should be understood by those of ordinary skill in the art that hardware manufacturers may refer to the same elements by different nouns. The scope of this specification and the subsequent patent application do not use the difference of the names as the means for distinguishing the elements, but the difference in function of the elements as the criterion for distinguishing. The term "including" as used throughout the specification and subsequent claims is an open term and should be interpreted as "including but not limited to". In addition, the term "coupled" is used herein to include any direct and indirect electrical connection. Therefore, if a first device is coupled to a second device, it means that the first device can be directly electrically connected to the second device or indirectly electrically connected to the second device through other devices or connection means. Please refer to FIG. 1. FIG. 1 is a block diagram of an automatic text labeling system 100 of the present invention. As shown in FIG. 1 , the automatic text labeling system 100 includes an artificial intelligence language processing module 110 , an artificial intelligence text classification module 120 , a label library 130 , and an artificial intelligence label output module 140 . The artificial intelligence language processing module 110 is configured to receive an input text d and convert the input text d into at least one specific dimension of the quantized feature. That is, the artificial intelligence language processing module 110 is used to quantify the unstructured text d. The artificial intelligence language processing module 110 is electrically connected to a scanning device or a digital camera. After the scanning device or the digital camera acquires the document image, the optical character recognition (Optical Character Recognition) technology is used to scan or The text on the image of the document after the photo is digitized. Regarding the quantized feature of converting the input text d into at least one specific dimension, for example, the artificial intelligence language processing module 110 converts the input text d into a two-dimensional matrix (TxW) by using a technique of word embedding. The characteristic matrix, where T is the length of the time axis, that is, the sequence length of the input text, and W is the self-designed parameter of the core of the text classification. The size of this parameter is positively related to the complexity of the label system. Please note that in the implementation of the present invention In the example, the feature matrix of the two-dimensional matrix (TxW) can also be regarded as a one-dimensional matrix arranged in chronological order. The artificial intelligence text classification module 120 is configured to convert the at least one specific dimension of the quantized feature into at least one specific confidence level feature matrix. That is, the artificial intelligence text classification module 120 is configured to receive the digitized character feature matrix and the characteristic matrix of the corresponding output confidence level of the given label. For example, after the artificial intelligence text classification module 120 receives the feature matrix of the two-dimensional matrix (TxW), the artificial intelligence text classification module 120 processes the two-dimensional matrix (Neural Networks) to the two-dimensional matrix. The matrix of the matrix (TxW) is transformed into a one-dimensional matrix, where the array size is as large as the lexicon size of the tag library. The neural network can be a Recurrent Neural Network, a Long Short-Term Memory, or a Convolutional Neural Network. Please note that this is only Embodiments of the invention are not limiting of the invention. As shown in FIG. 1, the tag library 130 is configured to store at least one tag that needs to be output, that is, a tag of a core word in the text is stored in the tag library 130, and the artificial intelligence tag output module 140 will The feature matrix of the at least one specific confidence level is converted into at least one text label and a corresponding quantitative confidence level. That is, the artificial intelligence tag output module 140 is configured to convert a numerically-conformed feature matrix into a corresponding tag. For example, the present invention defines a loss function, which may be However, it is not limited to cross-entropy, focal loss, and mean square error MSE to calculate the difference between the confidence level matrix of the core output and the real result. The larger the difference, the larger the loss function will be. value. In addition, the artificial intelligence tag output module 140 uses a deep learning algorithm such as Stochastic gradient descent (SGD), Adagrad, AdaDelta, Adam, RMSProp to iteratively update the parameters of the core to accurately predict the level of confidence. It should be noted that these deep learning algorithms are only embodiments of the present invention, and are not limitations of the present invention. Any algorithm that can optimize parameters is in accordance with the spirit of the present invention and falls within the scope of the present invention. In an embodiment of the present invention, after the artificial intelligence language processing module 110 is configured to receive the diagnostic text content, the artificial intelligence language processing module 110 extracts various information of the diagnostic text content, performs semantic representation, and The diagnostic text content is converted into a feature matrix of the two-dimensional matrix (TxW). Thereafter, the artificial intelligence text classification module 120 can calculate the confidence level corresponding to each label, such as the result of how much confidence level the "leg trauma" has. The artificial intelligence label output module 140 converts the digitized confidence level matrix into a corresponding label. For example, if the confidence level is set to 0.5, the "leg trauma" label in the label library 130 corresponds to a confidence level of 0.5. The above will be judged to correspond to the "leg trauma" label. According to the above description, those skilled in the art can easily understand that, in other embodiments of the present invention, if the content of the diagnostic text received by the artificial intelligence language processing module 110 is "left foot contusion, the left foot fourth toe distal end The phalanx fracture, the artificial intelligence label output module 140 will be converted into a label: (1) lower limb contusion (2) trunk compression bruise. If the content of the diagnostic text received by the artificial intelligence language processing module 110 is “left upper limb laceration (outpatient surgery stitching 3 stitches)”, the artificial intelligence label output module 140 is converted into a label: (1) hand Open wounds, except for fingers. Alternatively, if the content of the diagnostic text received by the artificial intelligence language processing module 110 is “left ureteral calculi with obstructive renal edema”, the artificial intelligence label output module 140 is converted into a label: (1) renal edema (2) Kidney and ureteral stones. It should be noted that the above-described embodiments of the labels are merely illustrative of the invention and are not limiting of the invention. By the same token, in another embodiment of the present invention, the method and system of the present invention can be used to advance the post-mortem verification of the medical record to the event reminder. When the doctor writes the diagnosis, he can remind him of the non-compliant content in real time, and eliminate the occurrence of non-standard medical records from the source. The system can also automatically identify the doctor's diagnosis according to the natural language understanding and medical knowledge, and provide medical insurance for the artificial intelligence. Please refer to FIG. 2. FIG. 2 is a flowchart of an automatic text labeling method according to the present invention, including but not limited to the following steps (note that if substantially the same result is obtained, these steps are not necessarily To be executed in accordance with the execution order shown in FIG. 2): Step S200: Start. Step S210: Receive an input text and convert the input text into a quantized feature of at least one specific dimension. Step S220: Convert the quantized feature of the at least one specific dimension into a feature matrix of at least one specific confidence level. Step S230: Store at least one tag that needs to be output. Step S240: Convert the at least one specific confidence level feature matrix into at least one text label and a corresponding quantization confidence level. Please refer to the steps shown in FIG. 2 and the components shown in FIG. 1 to know how each component operates. For the sake of brevity, it will not be repeated here. In this embodiment, an automatic text labeling method and system is proposed. The implementation of the method may depend on a computer program, which may be an application for automatically diagnosing a medical certificate based on a diagnostic book management system. The computer system may be a terminal device such as a smart phone, a tablet computer, or a personal computer that runs the above computer program. In addition, each module in the automatic text labeling system 100 of the present invention, that is, the artificial intelligence language processing module 110, the artificial intelligence text classification module 120, the label library 130, and the artificial intelligence label output module 140, can use a computer. The program can be implemented in a hardware manner (such as directly forming a wafer). The invention has the advantages that the automatic text labeling method and system can not only replace the manual, but also the classified text can speed up the manual classification efficiency through artificial intelligence, and the medical authority and the insurance company can pass the medical record by using the method and system of the invention. Natural language analysis, big data analysis of medical conditions and disease trends, thereby improving medical management and insurance services. The above description is only for the various embodiments of the present invention, and is not intended to limit the scope of the present invention. Therefore, all modifications and equivalent structural changes that are made by the present specification and the drawings are included in the present invention. Covers the scope of patents.

100‧‧‧Automatic text labeling system

110‧‧‧Artificial Wisdom Language Processing Module

120‧‧‧Artificial Wisdom Text Classification Module

130‧‧‧ tag library

140‧‧‧Artificial Wisdom Label Output Module

S200~S240‧‧‧ flowchart symbol

The various embodiments are described below with reference to the accompanying drawings, in which FIG. A block diagram of the labeling system. 2 is a flow chart of an automatic text labeling method of the present invention.

Claims (10)

An automatic text labeling system, comprising: an artificial intelligence language processing module, configured to receive an input text and convert the input text into at least one specific dimension of the quantized feature; an artificial intelligence text classification module coupled to the artificial a smart language processing module, configured to convert the at least one specific dimension of the quantized feature into at least one specific confidence level feature matrix; a tag library for storing at least one tag to be output; and an artificial intelligence tag output module And coupled to the artificial intelligence text classification module and the label library, configured to convert the at least one specific confidence level feature matrix into at least one text label and a corresponding quantization confidence level. According to the automatic text labeling system of claim 1, wherein the artificial intelligence language processing module is additionally used to quantify unstructured text. According to the automatic text labeling system of claim 1, wherein the artificial intelligence text classification module is further configured to receive the digitized character feature matrix and the corresponding output confidence level characteristic matrix of the label. According to the automatic text labeling system of claim 1, wherein the artificial intelligence text classification module is further used to convert the quantized feature into a one-dimensional matrix, wherein the matrix array size is larger than the font size of the tag library. . According to the automatic text labeling system of claim 1, wherein the artificial intelligence label output module is further used to convert a numerically confident level feature matrix into a corresponding label. An automatic text labeling method, comprising: receiving an input text and converting the input text into at least one specific dimension of the quantized feature; converting the at least one specific dimension of the quantized feature into at least one specific confidence level feature matrix; storing required output At least one tag; and converting the at least one specific confidence level feature matrix into at least one text tag and a corresponding quantified confidence level. The automatic text labeling method according to item 6 of the patent application scope further includes numerically unstructured text. The automatic text labeling method according to item 6 of the patent application scope further includes receiving the digitized character feature matrix and the characteristic matrix of the corresponding output confidence level of the given label. According to the automatic text labeling method of claim 6, the method further comprises converting the quantized feature into a one-dimensional matrix, wherein the matrix array size is larger than the lexicon size of the tag library. The automatic text labeling method according to item 6 of the patent application scope further includes converting the characteristic matrix of the numerical confidence level into a corresponding label.