CN113919291A - Master-slave parallel operation current sharing method based on analog control - Google Patents

Abstract

The invention discloses a master-slave parallel operation current sharing method based on analog control, which comprises the following steps: step 1, preprocessing a table and a text; step 2, performing corpus annotation by adopting a brat annotation tool; directly marking the linguistic data in the excel file; step 3, according to the result of the brat corpus labeling, carrying out named entity identification by using a BilSTM + CRF model; classifying by using a BERT model according to the result of excel corpus labeling; and 4, fusing the extracted entity elements and the classified texts according to the corresponding relation, and analyzing the elements by carrying out named entity identification and text on the approval opinions of the bank loan so as to return related element information.

Description

Master-slave parallel operation current sharing method based on analog control

Technical Field

The invention relates to the technical field of network information, in particular to a master-slave parallel operation current sharing method based on analog control.

Background

The traditional issuing and auditing of public credit service is completed by manually auditing paper data and system information in credit operation, the credit operation intelligent auditing system aims to explore the system automation level of auditing by artificial intelligence related technology, the targeted pain point is mainly the substitute risk control capability (including reducing the influence of artificial subjective or objective factors and improving the auditing standard rate of loan data), the second aspect is to improve the efficiency and improve the user experience (especially the efficiency in a remote auditing mode), the third aspect is to relieve the situation of human resource tension (release the human resource of professional auditors), and meanwhile, the mechanism construction of a strong data accumulation sediment and an intelligent research and development platform can be enhanced.

Taking the implementation of the scene of the approval opinions as an example, the analysis necessity and value comprise three aspects:

1) in the aspect of human resources:

in 2019, in all years, the operation review of all banks of credit checks that the credit business is 193789 in total, and the average review time length of one approval opinion is calculated according to one business and 15 minutes: 193789 × 15/60 ═ 48447.25 hours, i.e., a business, and an intelligent audit of approval may release 48447.25 hours of work per year for a released underwriting audit.

2) The service handling efficiency is improved:

after the manual labor is replaced by the system automation, the business handling efficiency is directly improved, the market demand response time is greatly shortened, the customer experience is improved, and the market competitiveness of the client is further improved.

3) And (3) improving the risk management and control capability:

the operation completed by the system has the advantages of strong calculation ability, all weather and no fatigue factor influence, directly improves the operation standard rate and reduces the artificial subjective factor influence of the link.

Disclosure of Invention

The invention aims to provide a master-slave parallel operation current sharing method based on analog control, so as to solve the problems in the background technology.

In order to achieve the purpose, the invention provides the following technical scheme:

a master-slave parallel operation current sharing method based on analog control comprises the following steps:

step 1, preprocessing a table and a text;

step 2, performing corpus annotation by adopting a brat annotation tool; directly marking the linguistic data in the excel file;

step 3, according to the result of the brat corpus labeling, carrying out named entity identification by using a BilSTM + CRF model; classifying by using a BERT model according to the result of excel corpus labeling;

and 4, fusing the extracted entity elements and the classified texts according to the corresponding relation.

As a further technical solution of the present invention, step 1 specifically is: the preprocessing operation on the table text comprises the steps of extracting the lines where the opinions in the table are located, converting the extracted and combined opinion content into the text, and cleaning the characters.

As a further technical solution of the present invention, the step 2 includes the following substeps:

step 21, respectively storing the corpora into a txt file and a ann file by adopting a brat marking tool and using a BIO marking method;

and step 22, marking the rear example of the linguistic data cell in the excel file by using an integer label, and finally putting all the semantic data cell in the txt file.

As a further technical solution of the present invention, step 3 includes the following substeps:

step 31, segmenting training data through a jieba toolkit;

step 32, sequencing the data after word segmentation according to the sequence to obtain corresponding subscript indexes, and storing the subscript indexes into a word _ to _ index array;

step 33, converting the word segmentation data into corresponding index vectors through word _ to _ index, and performing truncation according to a fixed length;

step 34, loading the downloaded word2vec word vectors, and constructing a word vector matrix according to the stored indexes in the word _ to _ index;

step 35, inputting the knowledge of the word vector into the model, and encoding by using the word vector matrix obtained in the step 34;

step 36, inputting the encoded vector into the long-term and short-term memory network, wherein the module will also use the hidden node state of the previous time point as the input of the current neural network unit, and meanwhile, a gating mechanism is used, that is, part of information of the hidden node state of the previous time point is selected in a controllable manner, so that information fusion of the current time node is performed, and finally hidden layer information is obtained;

step 37, inputting the hidden layer information into a CRF layer, and calculating by using a Viterbi coding algorithm of the CRF layer to obtain an entity label result;

and step 38, inputting the hidden layer information into a softmax function to obtain a probability matrix of the classification labels, and finally obtaining the final classification label of each paragraph through an argmax function.

As a further technical solution of the present invention, step 4 includes the following substeps:

step 41, obtaining the elements of the number, the precondition, the management requirement and the risk prompt in the text by calling the single service for the single examination and approval opinions;

and step 42, obtaining the classification label of each sentence by calling the comprehensive approval opinions through the comprehensive service: 11. 0, 4, 5, 6, wherein 11 represents a sentence containing a subsidiary; 0 represents none; 4 represents a sentence containing a management requirement; 5 represents a sentence containing a precondition; and 6 represents a sentence containing a risk hint.

Compared with the prior art, the invention has the beneficial effects that: the invention carries out named entity recognition and text on the approval opinions of the bank loan to analyze the elements and return the related element information.

Drawings

FIG. 1 is a flow chart of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Referring to fig. 1, a master-slave parallel operation current sharing method based on analog control includes the following steps:

step 1, preprocessing a text;

the method comprises the following steps: the preprocessing operation on the table text comprises the steps of extracting the lines where the opinions in the table are located, converting the extracted and combined opinion content into the text, and cleaning the characters.

Step 2, performing corpus annotation by adopting a brat annotation tool; directly marking the linguistic data in the excel file;

wherein, the following substeps are included:

and step 21, storing the corpora into a txt file and a ann file respectively by adopting a brat marking tool and using a BIO marking method.

Step 22, marking the rear example of the linguistic data cell in the excel file by using an integer label, and finally putting all the semantic data cell in the txt file;

step 3, according to the result of the brat corpus labeling, carrying out named entity identification by using a BilSTM + CRF model; classifying by using a BERT model according to the result of excel corpus labeling;

wherein, the following substeps are included:

and 31, segmenting the training data through a jieba toolkit.

And 32, sequencing the data after word segmentation according to the sequence to obtain corresponding subscript indexes, and storing the subscript indexes in a word _ to _ index array.

And step 33, converting the word segmentation data into corresponding index vectors through word _ to _ index, and performing truncation according to a fixed length.

And step 34, loading the downloaded word2vec word vectors, and constructing a word vector matrix according to the stored indexes in the word _ to _ index process.

And 35, inputting the knowledge of the word vector into the model, and encoding by using the word vector matrix obtained in the step 34.

And step 36, inputting the coded vector into the long-term and short-term memory network, wherein the module also takes the hidden node state of the previous time point as the input of the current neural network unit, and meanwhile, a gating mechanism is utilized, namely, part of information of the hidden node state of the previous time point is selected in a controllable manner, the information fusion of the current time node is carried out, and finally hidden layer information is obtained.

And step 37, inputting the hidden layer information into a CRF layer, and calculating by using a Viterbi coding algorithm of the CRF layer to obtain an entity label result.

And 38, inputting the hidden layer information into a softmax function to obtain a probability matrix of the classification label, and finally obtaining the final classification label of each paragraph through an argmax function.

And 4, fusing the extracted entity elements and the classified texts according to the corresponding relation.

Wherein, the following substeps are included:

step 41, obtaining the elements of the number, the precondition, the management requirement and the risk prompt in the text by calling the single service for the single examination and approval opinions;

and step 42, obtaining the classification label of each sentence by calling the comprehensive approval opinions through the comprehensive service: 11. 0, 4, 5, 6, wherein 11 represents a sentence containing a subsidiary; 0 represents none; 4 represents a sentence containing a management requirement; 5 represents a sentence containing a precondition; and 6 represents a sentence containing a risk hint.

And 43, adding the management requirements, the preconditions and the risk prompts of the corresponding subsidiary companies to the single corresponding elements according to the credit clients.

It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.

Furthermore, it should be understood that although the present description refers to embodiments, not every embodiment may contain only a single embodiment, and such description is for clarity only, and those skilled in the art should integrate the description, and the embodiments may be combined as appropriate to form other embodiments understood by those skilled in the art.

Claims (5)

1. A master-slave parallel operation current sharing method based on analog control is characterized by comprising the following steps:

step 1, preprocessing a table and a text;

step 2, performing corpus annotation by adopting a brat annotation tool; directly marking the linguistic data in the excel file;

step 3, according to the result of the brat corpus labeling, carrying out named entity identification by using a BilSTM + CRF model; classifying by using a BERT model according to the result of excel corpus labeling;

and 4, fusing the extracted entity elements and the classified texts according to the corresponding relation.

2. The master-slave parallel operation current sharing method based on analog control according to claim 1, wherein the step 1 is specifically: the preprocessing operation on the table text comprises the steps of extracting the lines where the opinions in the table are located, converting the extracted and combined opinion content into the text, and cleaning the characters.

3. The master-slave parallel-operation current sharing method based on analog control according to claim 2, wherein the step 2 includes the following sub-steps:

step 21, respectively storing the corpora into a txt file and a ann file by adopting a brat marking tool and using a BIO marking method;

and step 22, marking the rear example of the linguistic data cell in the excel file by using an integer label, and finally putting all the semantic data cell in the txt file.

4. The master-slave parallel-operation current sharing method based on analog control according to claim 1, wherein the step 3 comprises the following sub-steps:

step 31, segmenting training data through a jieba toolkit;

step 32, sequencing the data after word segmentation according to the sequence to obtain corresponding subscript indexes, and storing the subscript indexes into a word _ to _ index array;

step 33, converting the word segmentation data into corresponding index vectors through word _ to _ index, and performing truncation according to a fixed length;

step 34, loading the downloaded word2vec word vectors, and constructing a word vector matrix according to the stored indexes in the word _ to _ index;

step 35, inputting the knowledge of the word vector into the model, and encoding by using the word vector matrix obtained in the step 34;

step 36, inputting the encoded vector into the long-term and short-term memory network, wherein the module will also use the hidden node state of the previous time point as the input of the current neural network unit, and meanwhile, a gating mechanism is used, that is, part of information of the hidden node state of the previous time point is selected in a controllable manner, so that information fusion of the current time node is performed, and finally hidden layer information is obtained;

step 37, inputting the hidden layer information into a CRF layer, and calculating by using a Viterbi coding algorithm of the CRF layer to obtain an entity label result;

and step 38, inputting the hidden layer information into a softmax function to obtain a probability matrix of the classification labels, and finally obtaining the final classification label of each paragraph through an argmax function.

5. The master-slave parallel-operation current sharing method based on analog control according to claim 1, wherein the step 4 comprises the following sub-steps:

step 41, obtaining the elements of the number, the precondition, the management requirement and the risk prompt in the text by calling the single service for the single examination and approval opinions;

and step 42, obtaining the classification label of each sentence by calling the comprehensive approval opinions through the comprehensive service: 11. 0, 4, 5, 6, wherein 11 represents a sentence containing a subsidiary; 0 represents none; 4 represents a sentence containing a management requirement; 5 represents a sentence containing a precondition; and 6 represents a sentence containing a risk hint.

Images