CN114118817A - Bank sunshine loan-handling loan examination and dispatching method, device and system - Google Patents

Bank sunshine loan-handling loan examination and dispatching method, device and system Download PDF

Info

Publication number
CN114118817A
CN114118817A CN202111439938.XA CN202111439938A CN114118817A CN 114118817 A CN114118817 A CN 114118817A CN 202111439938 A CN202111439938 A CN 202111439938A CN 114118817 A CN114118817 A CN 114118817A
Authority
CN
China
Prior art keywords
loan
representing
classification
pending
loan service
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202111439938.XA
Other languages
Chinese (zh)
Other versions
CN114118817B (en
Inventor
董兴磊
赵涛
班风宝
陆权
许延波
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jinan Rural Commercial Bank Co ltd
Original Assignee
Jinan Rural Commercial Bank Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Jinan Rural Commercial Bank Co ltd filed Critical Jinan Rural Commercial Bank Co ltd
Priority to CN202111439938.XA priority Critical patent/CN114118817B/en
Publication of CN114118817A publication Critical patent/CN114118817A/en
Application granted granted Critical
Publication of CN114118817B publication Critical patent/CN114118817B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
    • G06Q10/06311—Scheduling, planning or task assignment for a person or group
    • G06Q10/063112—Skill-based matching of a person or a group to a task
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/35—Clustering; Classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/10—Office automation; Time management
    • G06Q10/103—Workflow collaboration or project management
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/03—Credit; Loans; Processing thereof

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Human Resources & Organizations (AREA)
  • General Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Health & Medical Sciences (AREA)
  • Economics (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • General Business, Economics & Management (AREA)
  • Molecular Biology (AREA)
  • Marketing (AREA)
  • Biophysics (AREA)
  • Development Economics (AREA)
  • Educational Administration (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Technology Law (AREA)
  • Databases & Information Systems (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Game Theory and Decision Science (AREA)
  • Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)

Abstract

The invention discloses a method, a device and a system for bank sunshine loan processing examination and dispatching, wherein the method comprises the following steps: obtaining a document of the pending loan service, and obtaining a plurality of participles of the document of the pending loan service by carrying out image recognition processing and participle processing on the document of the pending loan service; performing primary classification on the to-be-processed loan service through a first classification model; determining whether manual dispatching is needed or not based on the primary classification result, if so, manually dispatching, and otherwise, performing secondary classification on the to-be-handled loan service through a second classification model; and based on the classification result of the secondary classification, the to-be-processed loan service is assigned to the examiner with the highest conformity degree of the to-be-processed loan service for examination. The invention can effectively improve the accuracy, the reasonability and the safety of the order dispatching and has better practical application value.

Description

Bank sunshine loan-handling loan examination and dispatching method, device and system
Technical Field
The invention relates to the technical field of internet communication, in particular to a method, a device and a system for bank sunshine loan processing and loan examination and dispatching.
Background
Commercial banks are basically perfect in an examination and approval system, but some problems generally exist in the operation of examination, approval and loan handling links, and are mainly reflected in the dispatching side of bank sunshine loan handling examination.
There are many types of loans in banks, and generally, they are classified by loan terms, including short-term loans, mid-term loans, and long-term loans; classified according to repayment modes, including live loan and periodic loan; classified according to loan purposes, including consumption loans, securities loans, and the like; according to loan guarantee classification, including bill cash-in-place loan, bill mortgage loan, commodity mortgage loan, credit loan and the like; classified by interest rate, including fixed interest rate loans and floating interest rate loans.
However, in the actual loan examination, in addition to the above general loan types, there are some loans that require field investigation, that is, investigation on the field on the validity, safety, and profitability of the borrower loan (hereinafter, referred to as a loan transaction requiring field investigation, also referred to as an embedded loan investigation transaction), that is, the work of the examiner is mainly divided into two parts: on-site loan surveys and routine loan reviews.
Before dispatching the loan documents, the loan documents entering the loan examination process are classified, and then are distributed to examiners for examination according to loan categories.
At present, there are two kinds of group's formulas of manual group's and the automatic group's of system in the prior art mainly, and current bank loan group's technique is for overcoming the problem that artifical group's efficiency is low, generally adopts automatic group's formula, and automatic group has improved the efficiency of group greatly, however, still has the problem in following several respects:
(1) the existing bank automatic dispatching technology utilizes a machine learning model to classify loans, but the machine learning model is not improved aiming at bank loan business due to the adoption of the simple machine learning model, so that the problem of inaccurate classification exists.
(2) The prior bank automatic dispatching technology puts all loan businesses into a dispatching queue for automatic dispatching, does not consider the situations of urgent and important loan businesses and the like, only considers the general bank loan classification during loan classification, does not consider the classification of on-site investigation loans, namely, does not organically combine manual dispatching and automatic dispatching, and cannot meet the requirement of accurate dispatching of banks.
(3) The existing bank automatic order dispatching technology does not consider the conformity degree of the examiner and the loan service to be examined and approved during order dispatching, or simply matches the work load of the examiner and the work load of the loan service, does not make objective and detailed evaluation on various work indexes of the examiner, and has the problems of unreasonable and inaccurate order dispatching.
(4) The existing bank automatic dispatching technology is classified in a machine learning mode, a large amount of data operation is needed, the data processing capacity of a bank terminal is limited, data processing of the bank needs to be backed up, data processing and storage are only carried out on the bank terminal, and the risk of data loss exists.
If the loan order is not timely and accurate enough, the efficiency and the accuracy of the subsequent loan processing business of the bank can be greatly influenced, so that the benefits of the bank and the loan are damaged. Therefore, there is a need to develop a method, device and system for dispatching orders to improve the accuracy, rationality and safety of dispatching orders for examining and approving bank loans, and solve the above problems in the present invention, so as to better satisfy the demand of dispatching orders for loan handling in the sun of banks.
Disclosure of Invention
In view of the above problems, the present invention provides a method, an apparatus and a system for bank sunshine loan examination and dispatching, which can substantially eliminate the above problems of low dispatching accuracy, unreasonable dispatching and existing safety due to the limitations and disadvantages of the prior art. In order to solve the above problems, the technical solution proposed by the present invention is as follows:
in one aspect, the invention provides a method for checking and dispatching loan in the sun of a bank, which comprises the following steps:
s1, obtaining a document of the pending loan service, and obtaining a plurality of participles of the document of the pending loan service by carrying out image recognition processing and participle processing on the document of the pending loan service;
s2, inputting a plurality of word segments of the bill of the pending loan service into a first classification model trained in advance, and carrying out primary classification on the pending loan service through the first classification model;
s3, determining whether manual order dispatching is needed or not based on the primary classification result, if so, turning to the step S6, otherwise, turning to the step S4;
s4, inputting a plurality of word segments of the bill of the pending loan service into a pre-trained second classification model so as to perform secondary classification on the pending loan service through the second classification model;
s5, based on the classification result of the secondary classification, assigning the to-be-processed loan service to an examiner with the highest conformity degree of the to-be-processed loan service for examination;
s6, manually dispatching the order by a dispatching person with manual dispatching authority;
wherein the first classification model and the second classification model use the same neural network model, and the neural network model includes:
an input layer for receiving the plurality of participles;
the embedding layer is used for converting the multiple participles into Word vectors through a Word2vec model and weighting the Word vectors to form weighted Word vectors;
the bidirectional LSTM layer is used for extracting semantic features from the weighted word vector to form a first feature vector, and inputting the first feature vector into the TextCNN layer;
the TextCNN layer is used for receiving the first feature vector and extracting local position features from the first feature vector to form a second feature vector;
and the output layer is used for combining the first feature vector and the second feature vector into a third feature vector so that the classifier can classify by using the third feature vector.
Preferably, step S1 specifically includes the following sub-steps:
s11, carrying out image recognition on the bill of the pending loan service to extract text information in the bill of the pending loan service;
s12, preprocessing the text information to remove irrelevant information to obtain preprocessed text information;
and S13, performing word segmentation on the preprocessed text information by using a word segmentation tool to obtain a plurality of words.
Preferably, the Word vector after Word2vec model conversion is weighted, and the weighting formula is as follows:
Figure RE-GDA0003422559770000041
Figure RE-GDA0003422559770000042
wherein, W (t)iAnd d) represents the word tiThe weight in the text d, tf (t)iAnd d) represents the word tiThe frequency of occurrence in the text d, N represents the total number of texts,
Figure RE-GDA0003422559770000043
indicating the presence of a word tiTotal number of texts.
Preferably, the calculation formula of the bidirectional LSTM layer is as follows:
Figure RE-GDA0003422559770000044
Figure RE-GDA0003422559770000045
Figure RE-GDA0003422559770000046
wherein the content of the first and second substances,
Figure RE-GDA0003422559770000047
an output state vector representing the direction before time t,
Figure RE-GDA0003422559770000048
a state output vector representing the direction after time t,
Figure RE-GDA0003422559770000049
representing the output state vector, xtAn input representing the time of the t-instant,
Figure RE-GDA00034225597700000410
the output state vector representing the direction before time t-1,
Figure RE-GDA00034225597700000411
the state output vector representing the direction after time t +1,
Figure RE-GDA00034225597700000412
representing a forward input-hidden weight matrix,
Figure RE-GDA0003422559770000051
to indicate the frontTowards the concealment-concealment weight matrices,
Figure RE-GDA0003422559770000052
representing a front-wise concealment-output weight matrix,
Figure RE-GDA0003422559770000053
representing a rear directional input-hidden weight matrix,
Figure RE-GDA0003422559770000054
representing a rear direction concealment-concealment weight matrix,
Figure RE-GDA0003422559770000055
representing a rear-direction concealment-output weight matrix,
Figure RE-GDA0003422559770000056
byrespectively representing the forward direction, the backward direction and the final output offset vector.
Preferably, in the TextCNN layer, a plurality of convolutional layers are used to perform convolution operation on the first feature vector, where the input of the current convolutional layer is the convolution result of the previous convolutional kernel and the first feature vector, and the convolution result of the last convolutional layer is the second feature vector.
Preferably, step S5 includes the following sub-steps:
s51, obtaining the workload of the to-be-checked loan service of each examiner, determining pre-distribution examiners based on the workload of the to-be-checked loan service of each examiner, and calculating the score of each pre-distribution examiner for the to-be-processed loan service;
s52, sequencing the scores of the to-be-processed loan businesses of each pre-distribution examiner, and assigning the to-be-processed loan businesses to the examiners with the highest scores.
Preferably, in step S51, the score of each pre-assigned reviewer for the pending loan transaction is calculated according to the following equation (1):
Figure RE-GDA0003422559770000057
wherein, Scorei,jScore representing the ith reviewer for the pending loan transaction with transaction class j, ctWeight, k, representing the ith reviewer's review of the t-th type of loan transactiontIndicates the number of times that the ith reviewer has processed the t-th type of loan transaction, njRepresenting the total number of times the ith reviewer reviews the jth type of loan transaction, and n represents the total number of times the ith reviewer reviews all types of loan transactions.
Preferably, the method further comprises the steps of: and uploading the data related to the loan audit to a cloud server for backup.
The invention also provides a system for checking and dispatching the loan in the sun of the bank, which comprises: a processor, a memory and a program stored on the memory and executable on the processor, the program when executed by the processor implementing the bank sun loan audit assignment method as claimed in any one of claims 1-8.
The present invention also provides a computer-readable storage medium having stored thereon a program which, when executed by a processor, implements the bank sunshine loan examination and assignment method according to any one of claims 1 to 8.
Compared with the prior art, the method, the device and the system for bank sunshine loan processing examination and dispatching can effectively improve the accuracy, the reasonability and the safety of dispatching, and have better practical application value.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used for describing the embodiments will be briefly described below. It is understood that these drawings are merely exemplary and that variations can be made in those drawings without the exercise of inventive faculty, which modifications are intended to be included within the scope of the invention.
Fig. 1 is a flowchart of a method for bank sun loan-handling loan examination and assignment according to an embodiment of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer and enable those skilled in the art to better understand the technical solutions of the present invention, embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It is to be understood that the described embodiments are merely exemplary of some, and not necessarily all, embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the specific embodiments given in the description of the invention without inventive step, shall fall within the scope of protection of the invention.
As shown in fig. 1, the sun loan audit policy issuing method for banks according to the embodiment of the present invention includes the following steps:
s1, obtaining a bill of the pending loan service, and obtaining a plurality of participles of the bill of the pending loan service by carrying out image recognition processing and participle processing on the bill of the pending loan service.
S2, inputting the multiple word segments of the bill of the pending loan service into a first classification model trained in advance, and performing primary classification on the pending loan service through the first classification model.
And S3, determining whether manual order dispatching is needed or not based on the primary classification result, if so, turning to the step S6, and otherwise, turning to the step S4.
And S4, inputting a plurality of word segments of the bill of the pending loan service into a pre-trained second classification model, and performing secondary classification on the pending loan service through the second classification model.
And S5, based on the classification result of the secondary classification, assigning the pending loan service to an examiner with the highest compliance degree of the pending loan service for examination.
And S6, manually dispatching the order by a dispatching person with manual dispatching authority.
The first classification model and the second classification model adopt the same neural network model, the neural network model comprises an input layer, an embedding layer, a bidirectional LSTM layer, a TextCNN layer and an output layer, the embedding layer adopts a Wordevc model, the plurality of participles are converted into word vectors through the Wordevc model, and the word vectors are weighted to form weighted word vectors.
The above steps S1-S6 are explained in detail below.
In step S1, text information extraction and word segmentation of documents of pending loan services are mainly implemented, wherein step S1 specifically includes the following sub-steps:
s11, carrying out image recognition on the bill of the pending loan service to extract text information in the bill of the pending loan service.
The traditional bank dispatching system needs manual input of keyword information in the loan transaction document, which consumes labor cost. After the receipt of the to-be-processed loan service is obtained, the text information in the receipt image, such as a borrower, loan amount, loan application, repayment mode, repayment deadline and the like, is identified through an image identification technology.
And S12, preprocessing the text information to remove irrelevant information to obtain preprocessed text information.
The bill of the loan transaction contains the key information, and also contains irrelevant information such as punctuation, special symbols, etc. Step S12 is to remove extraneous information by preprocessing the text to facilitate subsequent data processing of the loan transaction.
And S13, performing word segmentation on the preprocessed text information by using a word segmentation tool to obtain a plurality of words.
Optionally, the word segmentation tools used in the present invention are, for example, discotic word segmentation, Paoding word segmentation, etc. In order to better adapt to the word segmentation processing of the bank data list, an auxiliary database is added on the basis of a word segmentation library carried by a word segmentation tool, wherein the auxiliary database comprises a financial database and a custom database. In actual work, common bank terms are supplemented into a custom database so as to continuously improve the database used for word segmentation and improve the word segmentation effect.
The invention directly utilizes the existing word segmentation tool and effectively improves the word segmentation efficiency and the word segmentation accuracy aiming at the bank loan service by supplementing the financial database and self-defining data.
Next, in step S2, the pending loan transaction document is primarily classified by using the first classification model. Specifically, the first classification model extracts the information of the borrower (including credit, loan record, famous property and the like of the borrower), the loan amount, the loan application and other characteristic information, and classifies the loan service to be examined according to the information, and the primary classification result comprises: normal loan transactions and important loan transactions.
For example, the important loan service includes a loan service requiring a site survey, a loan service requiring preferential processing (for example, identified as a borrower vip client by borrower information), an emergency loan service requiring accelerated processing, and the like.
The present invention performs step S2 to classify loan transactions into general loan transactions and important loan transactions, where the important loan transactions are manually assigned, and the general loan transactions are automatically assigned by the system. It can be seen that the present invention combines the separate processing of important loan transactions and general loan transactions, manual assignment and automatic assignment, which ensures the efficiency of the order assignment, while ensuring the assignment of important loan transactions to the most appropriate reviewers through manual assignment. For example, the dispatching personnel may designate experienced, work-intensive reviewers to review critical loan transactions, to prioritize them, and to ensure quality of service. For the loan transaction needing on-site investigation, the dispatcher can specify two suitable examiners to accompany the investigators to the on-site to participate in the investigation according to the regulations.
Because the manager who executes manual dispatching is more familiar with the working conditions of the staff under his hands, compared with the mechanical distribution of the machine, the invention is manually distributed for a small amount of important loan businesses, and automatically distributed by the machine for a large amount of common loan businesses, namely, the invention combines the automatic distribution of the machine and the manual distribution when dispatching the loan businesses, thereby taking the efficiency and the accuracy into consideration.
In step S3, it is determined whether manual dispatching is required based on the primary classification result of step S2, and if so, the process proceeds to step S6, and manual dispatching is performed by a dispatching person having the authority of manual dispatching. In general, a bank manager is set as a dispatching person having a manual dispatching authority. If manual dispatching is not needed, the procedure goes to step S4 for automatic dispatching.
In step S4, the multiple participles of the document of the pending loan service are input into a second classification model trained in advance, so as to perform secondary classification on the pending loan service through the second classification model. As can be seen from the above description, the classification is divided into the normal loan transaction and the important loan transaction through the primary classification of step S2, and if it is determined that the pending loan transaction belongs to the normal loan transaction, the secondary classification is continued through the second classification model, so that the participles input into the second classification model in step S4 are the same as the participles input into the first classification model in step S2, that is, after the classification in step S2, the participles of the pending loan transaction are continuously input into the second classification model in step S4 for secondary classification.
The secondary classification performed in step S4, wherein the category of the secondary classification is the general classification mentioned in the background of the invention, that is, the secondary classification includes live loan, periodic loan, consumption loan, securities loan, bill cash loan, bill mortgage loan, merchandise mortgage loan, credit loan, fixed interest rate loan, floating interest rate loan, and so on. It should be noted that the above loan types are merely examples, and the present invention is not limited to a specific loan type.
The first classification model and the second classification model are explained in detail below.
The first classification model and the second classification model are the core part of the invention, which relates to the accuracy of classification. The accuracy of the dispatching can be improved only by improving the accuracy of the service classification, thereby providing guarantee for high-quality and high-efficiency examination and loan service. Based on this, the invention focuses on improving the classification model.
In the prior art, generally, a word vector conversion model is used for converting word segments into word vectors, and then a classification model is used for extracting features of the word vectors, so that classification is realized. In addition, when extracting features of a long word sequence, the bidirectional LSTM discards important word information due to model capacity problems, while TextCNN is good at extracting local features of the current word.
The invention provides an improved classification model based on the existing neural network model. The classification model of the invention is a mixed neural network model, which integrates Word2vec, bidirectional LSTM and TextCNN, fully utilizes the advantages of the models, and can effectively improve the classification precision.
It should be noted that, before the classification model of the present invention is used, historical data is used for training, and the specific training process is not described again.
According to a preferred embodiment of the present invention, the first classification model and the second classification model use the same neural network model, which includes:
an input layer for receiving the plurality of participles;
the embedding layer is used for converting the multiple participles into Word vectors through a Word2vec model and weighting the Word vectors to form weighted Word vectors;
the bidirectional LSTM layer is used for extracting semantic features from the weighted word vector to form a first feature vector, and inputting the first feature vector into the TextCNN layer;
the TextCNN layer is used for receiving the first feature vector and extracting local position features from the first feature vector to form a second feature vector; wherein the local position feature may include a relationship between words and position information of the words in the text.
And the output layer is used for combining the first feature vector and the second feature vector into a third feature vector, classifying by using the third feature vector and outputting a classification result.
According to the preferred embodiment of the invention, the embedding layer adopts a trained Word2vec model, and the Word vectors are converted from the Word segmentations through the wordeve model. It should be noted that Word2vec used in the present invention is only an example, and any other suitable model may be used by those skilled in the art to convert the participle into the Word vector.
The Wordevec model comprises a CBOW model and a Skip-gram model, and the CBOW model is adopted in the invention.
In order to extract characteristic words which are beneficial to classification in the subsequent characteristic extraction step, the word vector converted by the Wordevec model is weighted by the method, and the weighting formula is as follows:
Figure RE-GDA0003422559770000111
Figure RE-GDA0003422559770000112
wherein, W (t)iAnd d) represents the word tiThe weight in the text d, tf (t)iAnd d) represents the word tiThe frequency of occurrence in the text d, N represents the total number of texts,
Figure RE-GDA0003422559770000113
indicating the presence of a word tiTotal number of texts.
The weight of each word in the text is obtained through formulas (1) and (2), a weight matrix is formed, and the weight matrix is multiplied by the word vector matrix, so that a weighted word vector can be obtained.
After obtaining weighted word vectors through the embedding layer, the weighted word vectors are input into the bidirectional LSTM layer, the bidirectional LSTM layer adopts a bidirectional LSTM model, and in order to simplify a network structure and reduce the number of network parameters, the bidirectional LSTM model does not include an attention layer. In addition, the bi-directional LSTM model of the present invention preserves the position-order relationship, thereby allowing words of different distances in a sentence to be connected.
Specifically, the calculation formula of the bidirectional LSTM model of the invention is as follows:
Figure RE-GDA0003422559770000121
Figure RE-GDA0003422559770000122
Figure RE-GDA0003422559770000123
wherein the content of the first and second substances,
Figure RE-GDA0003422559770000124
an output state vector representing the direction before time t,
Figure RE-GDA0003422559770000125
a state output vector representing the direction after time t,
Figure RE-GDA0003422559770000126
representing the output state vector (i.e. the first eigenvector, described earlier), xtAn input representing the time of the t-instant,
Figure RE-GDA0003422559770000127
the output state vector representing the direction before time t-1,
Figure RE-GDA0003422559770000128
the state output vector representing the direction after time t +1,
Figure RE-GDA0003422559770000129
representing a forward input-hidden weight matrix,
Figure RE-GDA00034225597700001210
representing a front-facing concealment-concealment weight matrix,
Figure RE-GDA00034225597700001211
representing a front-wise concealment-output weight matrix,
Figure RE-GDA00034225597700001212
representing a rear directional input-hidden weight matrix,
Figure RE-GDA00034225597700001213
representing a rear direction concealment-concealment weight matrix,
Figure RE-GDA00034225597700001214
representing a rear-direction concealment-output weight matrix,
Figure RE-GDA00034225597700001215
byrespectively representing the forward direction, the backward direction and the final output offset vector.
Next, the vectors output by the bi-directional LSTM layer are input to the TextCNN layer to further extract local position features. The TextCNN layer performs convolution calculation on the input vector, and according to a preferred embodiment of the present invention, performs convolution operation on the first feature vector by using N convolution layers (preferably, N ═ 5), where the input of the current convolution layer is the convolution result of the previous convolution kernel and the first feature vector, and the convolution result of the last convolution layer is the second feature vector. Note that, for the first convolutional layer, since there is no convolutional result of the previous convolutional layer, the input of the first convolutional layer is only the first feature vector.
The invention further improves the precision of feature extraction by improving the convolution operation of the TextCNN layer and adopting a continuous convolution mode.
According to the preferred embodiment of the invention, the output layer calculates the classification probability by adopting a softmax function, and finally outputs the class with the maximum probability as the classification result.
According to the preferred embodiment of the present invention, the present invention further updates the first classification model and the second classification model by using the classification result data of the first classification model and the second classification model, so as to continuously optimize the first classification model and the second classification model, and improve the classification accuracy.
After the secondary classification result is obtained in the step S4, the step S5 is executed, and the pending loan service is assigned to the examiner who has the highest compliance degree with the pending loan service for examination based on the secondary classification result.
Specifically, step S5 includes the following substeps:
s51, obtaining the workload of the to-be-checked loan service of each examiner, determining pre-distribution examiners based on the workload of the to-be-checked loan service of each examiner, and calculating the score of each pre-distribution examiner for the to-be-processed loan service;
s52, sequencing the scores of the to-be-processed loan businesses of each pre-distribution examiner, and assigning the to-be-processed loan businesses to the examiners with the highest scores.
In step S5, to avoid overstock of loan transaction examination work and improve examination efficiency, the present invention first obtains the workload of the loan transaction to be examined of each examiner, and determines the examiner whose workload of the loan transaction to be examined is lower than a preset threshold as a pre-assigned examiner, so as to avoid assigning the loan transaction to be examined to the examiner who has multiple unfinished loan transactions on hand, wherein the preset threshold can be set by the manager, for example, set to 2 pieces.
In the specific implementation of step S51, the present invention presets a censorship staff table, in which each censorship staff and the corresponding pending loan transaction workload are located. That is, the step S51 of "obtaining the workload of the pending loan transaction for each reviewer, and determining the pre-allocated reviewer based on the workload of the pending loan transaction for each reviewer" may specifically include: and obtaining the workload of the loan transaction to be checked of each examiner through the examiner workload table, comparing the workload of the loan transaction to be checked of each examiner with a preset threshold value, and determining the examiners with the workload of the loan transaction to be checked lower than the preset threshold value as pre-allocated examiners.
In step S51, in order to assign the loan transaction to the reviewer best suited to handle the loan transaction, the present invention scores each pre-assigned reviewer, and in particular, calculates each pre-assigned reviewer' S score for the pending loan transaction according to the following equation (6) based on the discovery utilization policy:
Figure RE-GDA0003422559770000141
wherein, Scorei,jScore representing the ith reviewer for the pending loan transaction with transaction class j, ctWeight, k, representing the ith reviewer's review of the t-th type of loan transactiontIndicates the number of times that the ith reviewer has processed the t-th type of loan transaction, njRepresenting the total number of times the ith reviewer reviews the jth type of loan transaction, and n represents the total number of times the ith reviewer reviews all types of loan transactions.
It will be appreciated that the weight ctThe overall quality condition of the t-th type loan transaction processed by the examiner is reflected, and the overall quality condition can be comprehensively set according to the working time of the examiner for processing the t-th type loan transaction, the feedback evaluation of the client, the working age of the examiner, the working performance of the examiner and the like.
Next, step S52 is executed to sort the scores of each pre-assigned reviewer for the pending loan transaction, and assign the pending loan transaction to the reviewer with the highest score.
According to a preferred embodiment of the present invention, in order to relieve the pressure of the bank terminal, the first classification model and the second classification model are stored in a first cloud terminal server and a second cloud terminal server, respectively. In addition, in order to ensure the security of the audit data, the invention also comprises: and uploading the data related to the loan examination to a third cloud terminal server for backup.
In addition, in order to realize the customs loan handling, namely, to enable the borrower to know the loan handling condition in time, the invention also comprises: the loan audit progress is sent to the borrower's intelligent terminal, e.g., cell phone, in time.
As can be seen from the above description, the present invention fully considers the problem of limited data processing capability of the bank terminal, performs model training and classification by using the cloud server, and distributes different models on different cloud terminal servers, so as to further improve the efficiency and security of data processing.
The invention also provides a system for checking and dispatching the loan in the sun of the bank, which comprises: a processor, a memory and a program stored on the memory and executable on the processor, the program when executed by the processor implementing the method steps of the bank sun loan audit issuance method embodiment as described above.
The present invention also provides a computer readable storage medium having stored thereon a program which, when executed by a processor, performs the method steps of the above-described embodiments of the bank sun loan audit issuance method. The computer-readable storage medium may be a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk.
Furthermore, it is to be noted that in the device and method of the invention, it is obvious that the individual components or steps can be decomposed and/or recombined. These decompositions and/or recombinations are to be regarded as equivalents of the present invention. Also, the steps of performing the above-described series of processes may naturally be performed in the order described or in chronological order, but need not necessarily be performed in chronological order, and some steps may be performed in parallel or independently of each other. It will be understood by those of ordinary skill in the art that all or any of the steps or elements of the methods and apparatus of the present invention may be implemented in any computing device (including processors, storage media, etc.) or network of computing devices, in hardware, firmware, software, or any combination thereof.
The above description is only a preferred embodiment of the present invention, and for those skilled in the art, the present invention should not be limited by the description of the present invention, which should be interpreted as a limitation.

Claims (10)

1. A bank sun loan processing loan examination and order dispatching method comprises the following steps:
s1, obtaining a document of the pending loan service, and obtaining a plurality of participles of the document of the pending loan service by carrying out image recognition processing and participle processing on the document of the pending loan service;
s2, inputting a plurality of word segments of the bill of the pending loan service into a first classification model trained in advance, and carrying out primary classification on the pending loan service through the first classification model;
s3, determining whether manual order dispatching is needed or not based on the primary classification result, if so, turning to the step S6, otherwise, turning to the step S4;
s4, inputting a plurality of word segments of the bill of the pending loan service into a pre-trained second classification model so as to perform secondary classification on the pending loan service through the second classification model;
s5, based on the classification result of the secondary classification, assigning the to-be-processed loan service to an examiner with the highest conformity degree of the to-be-processed loan service for examination;
s6, manually dispatching the order by a dispatching person with manual dispatching authority;
wherein the first classification model and the second classification model use the same neural network model, and the neural network model includes:
an input layer for receiving the plurality of participles;
the embedding layer is used for converting the multiple participles into Word vectors through a Word2vec model and weighting the Word vectors to form weighted Word vectors;
the bidirectional LSTM layer is used for extracting semantic features from the weighted word vector to form a first feature vector, and inputting the first feature vector into the TextCNN layer;
the TextCNN layer is used for receiving the first feature vector and extracting local position features from the first feature vector to form a second feature vector;
and the output layer is used for combining the first feature vector and the second feature vector into a third feature vector so that the classifier can classify by using the third feature vector.
2. The method according to claim 1, wherein step S1 comprises the following sub-steps:
s11, carrying out image recognition on the bill of the pending loan service to extract text information in the bill of the pending loan service;
s12, preprocessing the text information to remove irrelevant information to obtain preprocessed text information;
and S13, performing word segmentation on the preprocessed text information by using a word segmentation tool to obtain a plurality of words.
3. The method according to claim 1 or 2, characterized in that for WoWeighting the word vectors converted by the rd2vec model, wherein the weighting formula is as follows:
Figure RE-FDA0003422559760000021
Figure RE-FDA0003422559760000022
wherein, W (t)iAnd d) represents the word tiThe weight in the text d, tf (t)iAnd d) represents the word tiThe frequency of occurrence in the text d, N represents the total number of texts,
Figure RE-FDA0003422559760000023
indicating the presence of a word tiTotal number of texts.
4. The method of claim 1 or 2, wherein the calculation formula of the bi-directional LSTM layer is as follows:
Figure RE-FDA0003422559760000024
Figure RE-FDA0003422559760000025
Figure RE-FDA0003422559760000026
wherein the content of the first and second substances,
Figure RE-FDA0003422559760000027
an output state vector representing the direction before time t,
Figure RE-FDA0003422559760000028
a state output vector representing the direction after time t,
Figure RE-FDA0003422559760000029
representing the output state vector, xtAn input representing the time of the t-instant,
Figure RE-FDA00034225597600000210
the output state vector representing the direction before time t-1,
Figure RE-FDA00034225597600000211
the state output vector representing the direction after time t +1,
Figure RE-FDA00034225597600000212
representing a forward input-hidden weight matrix,
Figure RE-FDA00034225597600000213
representing a front-facing concealment-concealment weight matrix,
Figure RE-FDA00034225597600000214
representing a front-wise concealment-output weight matrix,
Figure RE-FDA0003422559760000031
representing a rear directional input-hidden weight matrix,
Figure RE-FDA0003422559760000032
representing a rear direction concealment-concealment weight matrix,
Figure RE-FDA0003422559760000033
representing a rear-direction concealment-output weight matrix,
Figure RE-FDA0003422559760000034
byrespectively representing the forward direction, the backward direction and the final output offset vector.
5. The method according to claim 1 or 2, wherein at the TextCNN layer, the first feature vector is convolved with a plurality of convolutional layers, wherein the input of the current convolutional layer is the convolution result of the previous convolutional kernel and the first feature vector, and the convolution result of the last convolutional layer is the second feature vector.
6. The method according to claim 1, wherein step S5 includes the sub-steps of:
s51, obtaining the workload of the to-be-checked loan service of each examiner, determining pre-distribution examiners based on the workload of the to-be-checked loan service of each examiner, and calculating the score of each pre-distribution examiner for the to-be-processed loan service;
s52, sequencing the scores of the to-be-processed loan businesses of each pre-distribution examiner, and assigning the to-be-processed loan businesses to the examiners with the highest scores.
7. The method of claim 6, wherein in step S51, the score of each pre-assigned reviewer for the pending loan transaction is calculated according to the following equation (1):
Figure RE-FDA0003422559760000035
wherein, Scorei，jScore representing the ith reviewer for the pending loan transaction with transaction class j, ctWeight, k, representing the ith reviewer's review of the t-th type of loan transactiontIndicates the number of times that the ith reviewer has processed the t-th type of loan transaction, njRepresenting the total number of times the ith reviewer reviews the jth type of loan transaction, and n represents the total number of times the ith reviewer reviews all types of loan transactions.
8. The method according to any one of claims 1-7, characterized in that the method further comprises the steps of: and uploading the data related to the loan audit to a cloud server for backup.
9. A bank sun loan-handling loan review worksheet system, the system comprising: a processor, a memory and a program stored on the memory and executable on the processor, the program when executed by the processor implementing the bank sun loan audit assignment method as claimed in any one of claims 1-8.
10. A computer-readable storage medium having stored thereon a program which, when executed by a processor, implements the bank sun-loan review assignment method according to any one of claims 1 to 8.
CN202111439938.XA 2021-11-30 2021-11-30 Bank loan examination order dispatching method, device and system Active CN114118817B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111439938.XA CN114118817B (en) 2021-11-30 2021-11-30 Bank loan examination order dispatching method, device and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111439938.XA CN114118817B (en) 2021-11-30 2021-11-30 Bank loan examination order dispatching method, device and system

Publications (2)

Publication Number Publication Date
CN114118817A true CN114118817A (en) 2022-03-01
CN114118817B CN114118817B (en) 2022-08-05

Family

ID=80368215

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111439938.XA Active CN114118817B (en) 2021-11-30 2021-11-30 Bank loan examination order dispatching method, device and system

Country Status (1)

Country Link
CN (1) CN114118817B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107832400A (en) * 2017-11-01 2018-03-23 山东大学 A kind of method that location-based LSTM and CNN conjunctive models carry out relation classification
CN109102157A (en) * 2018-07-11 2018-12-28 交通银行股份有限公司 A kind of bank's work order worksheet processing method and system based on deep learning
US20200184540A1 (en) * 2018-12-10 2020-06-11 Accenture Global Solutions Limited Artificial intelligence and machine learning based conversational agent
CN112800765A (en) * 2021-01-22 2021-05-14 南京亚派软件技术有限公司 Automatic work order generation method
CN113535960A (en) * 2021-08-02 2021-10-22 中国工商银行股份有限公司 Text classification method, device and equipment

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107832400A (en) * 2017-11-01 2018-03-23 山东大学 A kind of method that location-based LSTM and CNN conjunctive models carry out relation classification
CN109102157A (en) * 2018-07-11 2018-12-28 交通银行股份有限公司 A kind of bank's work order worksheet processing method and system based on deep learning
US20200184540A1 (en) * 2018-12-10 2020-06-11 Accenture Global Solutions Limited Artificial intelligence and machine learning based conversational agent
CN112800765A (en) * 2021-01-22 2021-05-14 南京亚派软件技术有限公司 Automatic work order generation method
CN113535960A (en) * 2021-08-02 2021-10-22 中国工商银行股份有限公司 Text classification method, device and equipment

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
PAN DONG-HONG ET.AL.: "A Chinese implicit sentiment classification model combining contextual features", 《COMPUTER ENGINEERING AND SCIENCE》 *
刘佳: "基于改进词嵌入表示和卷积神经网络的银行智能派单***的研究", 《工业控制计算机》 *
王立荣: "Word2vec-CNN-Bilstm短文本情感分类", 《福建电脑》 *

Also Published As

Publication number Publication date
CN114118817B (en) 2022-08-05

Similar Documents

Publication Publication Date Title
Mai et al. Deep learning models for bankruptcy prediction using textual disclosures
Raghupathi et al. A neural network application for bankruptcy prediction
Wang et al. A new random subspace method incorporating sentiment and textual information for financial distress prediction
Jiang et al. Mining semantic features in current reports for financial distress prediction: Empirical evidence from unlisted public firms in China
Lu et al. Credit rating change modeling using news and financial ratios
Chen et al. Bankruptcy prediction using machine learning models with the text-based communicative value of annual reports
Shi et al. A novel imbalanced data classification approach based on logistic regression and Fisher discriminant
WO2023071120A1 (en) Method for recognizing proportion of green assets in digital assets and related product
Haryono et al. Aspect-based sentiment analysis of financial headlines and microblogs using semantic similarity and bidirectional long short-term memory
Dasari et al. Prediction of bank loan status using machine learning algorithms
CN111143533B (en) Customer service method and system based on user behavior data
CN114118817B (en) Bank loan examination order dispatching method, device and system
Holowczak et al. Testing market response to auditor change filings: A comparison of machine learning classifiers
CN116662546A (en) Complaint text labeling method, device, equipment and medium
Park et al. Automatic construction of context-aware sentiment lexicon in the financial domain using direction-dependent words
CN109635289A (en) Entry classification method and audit information abstracting method
Lee et al. Application of machine learning in credit risk scorecard
Islam et al. Stock market prediction of Bangladesh using multivariate long short-term memory with sentiment identification.
Hristova et al. RatingBot: A Text Mining Based Rating Approach.
Lee et al. An Integral Predictive Model of Financial Distress
Kotsiantis et al. Financial Application of Neural Networks: two case studies in Greece
Makatjane Deep learning for sentiment analysis to predict the probability of bank loan default
Li et al. Parametric and non-parametric combination model to enhance overall performance on default prediction
Melnyk et al. Improving the quality of credit activity by using scoring model
Baranidharan et al. Prediction and Analysis of Financial Crises Using Machine Learning

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant