Loan review business processing method, device, storage medium and electronic equipment

Abstract

The present application discloses a loan review business processing method, device, storage medium and electronic device. Relating to the field of artificial intelligence, the method includes: obtaining the identity information of a user who needs to handle a target loan business, and obtaining the business information of the target loan business; inputting the identity information into a first model to obtain a first recommended business, wherein the first recommended business is used to review the target loan business, and the first model is trained by multiple groups of first training samples, and each group of first training samples includes historical identity information and first historical recommended business; inputting the business information into a second model to obtain a second recommended business; extracting the target approval document from the business database, and handling the first recommended business and the second recommended business through the target approval document to obtain the handling result. Through this application, the problem of low loan review efficiency in the related art of executing loan review business is solved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and more specifically, to a loan review business processing method, device, storage medium and electronic equipment. Background Art

[0002] In related technologies, financial institutions primarily rely on expert experience to review loan applications, manually accessing the required information from their internal systems and external information platforms. However, due to the growing volume and increasing difficulty of review, the shortage of reviewers is becoming increasingly prominent. While the volume of review applications at financial institutions continues to increase, characterized by larger financing amounts, longer financing terms, a higher proportion of credit methods, an increase in projects in manufacturing segments, and the emergence of new business models, the number of reviewers has not been able to keep pace, increasing business pressure and highlighting the contradiction.

[0003] In related technologies, financial institutions' credit operations generate documents such as review reports, which contain extensive information and credit experience. However, these documents are stored as attachments and lack sufficient online and digitized functionality. The various dimensions of information required for review work are primarily obtained through manual queries. The integration of due diligence reports and review reports between the front-end and middle-end credit platforms is weak, and data sharing is limited. Reviewers primarily rely on personal experience to obtain information and make risk assessments, leading to errors in the acquisition and analysis of key information.

[0004] Throughout the entire credit approval process, business personnel primarily facilitated the credit process through offline document transfer. This process involved loopholes between pre-loan due diligence personnel and loan reviewers regarding project loan processes, hindering progress. In the post-loan phase, business personnel regularly conducted project reviews to proactively identify credit risks. The credit process underutilized document assets. For example, historical reports written by business personnel were scattered across various channels, lacking unified management and utilization. Document management was primarily manual, not online. The report review process relied heavily on manual verification, resulting in low efficiency.

[0005] Currently, no effective solution has been proposed to the problem of low efficiency in loan review in related technologies. Summary of the Invention

[0006] The main purpose of this application is to provide a loan review business processing method, device, storage medium and electronic equipment to solve the problem of low loan review efficiency in performing loan review business in related technologies.

[0007] To achieve the above-mentioned objectives, according to one aspect of the present application, a method for processing loan review services is provided. The method comprises: obtaining identity information of a user who needs to process a target loan service, and obtaining service information of the target loan service; inputting the identity information into a first model to obtain a first recommended service, wherein the first recommended service is used to review the target loan service, and the first model is trained using multiple sets of first training samples, each set of first training samples including historical identity information and a first historical recommended service; inputting the service information into a second model to obtain a second recommended service, wherein the second recommended service is used to review the target loan service, and the second model is trained using multiple sets of second training samples, each set of second training samples including historical service information and a second historical recommended service; extracting a target approval document from a service database, and processing the first recommended service and the second recommended service using the target approval document to obtain a processing result.

[0008] Optionally, before extracting the target approval document from the business database, the method also includes: receiving approval information entered by the business personnel for approving the target loan business, generating an approval document based on the approval information; and storing the approval document in the business database, wherein the business database contains approval documents for multiple types of loan businesses.

[0009] Optionally, before storing the approval document in the business database, the method also includes: extracting sensitive data from the approval document, wherein the sensitive data includes sensitive text and sensitive images, the sensitive text includes at least one of the following: legal person name, shareholder name, company name and holding company name, and the sensitive image includes at least one of the following: equity structure chart and corporate legal person holding ratio chart; performing a desensitizing operation on the approval document through a natural language processing model to obtain a desensitized approval document, and storing the desensitized approval document in the business database.

[0010] Optionally, performing a desensitizing operation on the approval document through a natural language processing model to obtain a desensitized approval document includes: performing a filtering operation on the approval document to obtain an updated approval document, wherein the filtering operation includes at least one of the following: stop word filtering and common word filtering; retrieving sensitive text from the updated approval document, and encrypting the sensitive text through a preset mask in the natural language processing model to obtain an approval document with the sensitive text masked; extracting all images from the approval document with the sensitive text masked, and inputting all images into an image recognition model to obtain a vector set of all images, wherein the vector set contains multiple vectors, and each vector corresponds to an image; retrieving a target vector from the vector set, wherein the target vector is a vector corresponding to the sensitive image; encrypting the image corresponding to the target vector through a preset mask to obtain an approval document with the sensitive image masked.

[0011] Optionally, extracting the target approval document from the business database includes: receiving an approval document query request, extracting keywords from the approval document query request, and retrieving a first approval document containing the keywords from the business database; inputting the text content in the approval document query request into a semantic recognition model to obtain a semantic vector, and retrieving a second approval document containing the semantic vector from the business database; merging the first approval document and the second approval document, and eliminating duplicate approval documents to obtain the target approval document.

[0012] Optionally, before retrieving the first approval document containing keywords from the business database, the method also includes: processing the approval document into a preset format, wherein the approval file in the preset format does not support downloading; parsing the file in the preset format to obtain word segments and sentences; constructing a keyword index based on the word segments, and constructing a semantic vector index based on the sentences; and constructing an inverted index through the keyword index and the semantic vector index.

[0013] Optionally, the first model is trained in the following manner: obtaining business processing records, extracting multiple review businesses from the business processing records; determining the historical identity information of the user who processed the review business, and determining the review business as the first historical recommended business; determining each first historical recommended business and the historical identity information corresponding to the first historical recommended business as a group of first training samples to obtain multiple groups of first training samples; training the preset neural network model through multiple groups of first training samples to obtain the first model.

[0014] Optionally, the second model is trained in the following manner: obtaining business processing records, extracting multiple review businesses from the business processing records; determining historical business information of loan businesses that require review business, and determining the review business as a second historical recommended business; determining each second historical recommended business and the historical business information corresponding to the second historical recommended business as a group of second training samples to obtain multiple groups of second training samples; training the preset neural network model through multiple groups of second training samples to obtain the second model.

[0015] To achieve the above-mentioned purpose, according to another aspect of the present application, a loan review business processing device is provided. The device includes: an acquisition unit, configured to acquire the identity information of a user who needs to process a target loan business, and to acquire business information of the target loan business; a first input unit, configured to input the identity information into a first model to obtain a first recommended business, wherein the first recommended business is used to review the target loan business, and the first model is trained by multiple sets of first training samples, each set of first training samples including historical identity information and a first historical recommended business; a second input unit, configured to input business information into a second model to obtain a second recommended business, wherein the second recommended business is used to review the target loan business, and the second model is trained by multiple sets of second training samples, each set of second training samples including historical business information and a second historical recommended business; and an extraction unit, configured to extract the target approval document from the business database, and process the first recommended business and the second recommended business using the target approval document to obtain a processing result.

[0016] The present application adopts the following steps: obtaining the identity information of a user who needs to apply for a target loan business and obtaining the business information of the target loan business; inputting the identity information into a first model to obtain a first recommended business, wherein the first recommended business is used to review the target loan business, the first model is trained by multiple sets of first training samples, each set of first training samples includes historical identity information and a first historical recommended business; inputting the business information into a second model to obtain a second recommended business, wherein the second recommended business is used to review the target loan business, the second model is trained by multiple sets of second training samples, each set of second training samples includes historical business information and a second historical recommended business; extracting the target approval document from a business database, and processing the first recommended business and the second recommended business based on the target approval document to obtain a processing result, thereby solving the problem of low loan review efficiency in the related art. By inputting the user's identity information into the first model to obtain the first recommended business, inputting the business information into the second model to obtain the second recommended business, and processing the first recommended business and the second recommended business based on the target approval document extracted from the business database to obtain a processing result for approving the target loan business, the loan review efficiency is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0018] Figure 1 This is a flowchart of a loan review business processing method provided in accordance with an embodiment of the present application;

[0019] Figure 2is a schematic diagram of storing an approval document in a business database according to an embodiment of the present application;

[0020] Figure 3 Schematic diagram of a desensitization method for approval documents provided in accordance with an embodiment of the present application;

[0021] Figure 4 This is a schematic diagram of extracting target approval documents according to an embodiment of the present application;

[0022] Figure 5 is a schematic diagram of constructing an inverted index according to an embodiment of the present application;

[0023] Figure 6 is a schematic diagram of training a first model and a second model according to an embodiment of the present application;

[0024] Figure 7 is a schematic diagram of a loan review business processing device provided according to an embodiment of the present application;

[0025] Figure 8 is a schematic diagram of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0027] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0028] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0029] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0030] It should be noted that the information collected is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data comply with the relevant laws, regulations and standards of the relevant regions, take necessary confidentiality measures, do not violate public order and good customs, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0031] For ease of description, some nouns or terms involved in the embodiments of the present application are explained below:

[0032] Bert: Bidirectional Encoder Representations from Transformers, a bidirectional text encoding pre-training model.

[0033] Sigmoid: Neural network activation function that maps input results to between (0, 1).

[0034] FC: Fully Connected Layers, fully connected layers in neural networks.

[0035] Embedding: word embedding layer, converting different modal data into dense vectors.

[0036] DNN: Deep Neural Network deep neural network layer.

[0037] Concate: Connect features in a certain dimension to expand the feature dimension.

[0038] WDL: Wide & Deep Learning for Recommender System, a deep push network consisting of wide and deep layers.

[0039] Bert-Span: An improved Bert model that improves model performance by expanding the mask range.

[0040] Bert-NER: Implements named entity recognition capabilities using the Bert model base.

[0041] Resnet: Deep residual network, which has been widely used in the field of computer vision.

[0042] The present invention will be described below in conjunction with preferred implementation steps. Figure 1 This is a flowchart of a loan review business processing method provided in accordance with an embodiment of the present application. Figure 1 As shown, the method includes the following steps:

[0043] Step S101: obtaining identity information of a user who needs to handle a target loan business, and obtaining business information of the target loan business.

[0044] Specifically, the user may be a user applying for a loan at a financial institution. The target loan business may be any of the various loan businesses offered by the financial institution. Identity information may include name, ID number, asset information, and contact information. Business information may include relevant information about the target loan business applied for by the user, such as loan type, loan amount, and loan application process. By obtaining the user's identity information and business information related to the target loan business, the corresponding loan approval business can be recommended based on the identity and business information.

[0045] Step S102: input the identity information into the first model to obtain a first recommended business, wherein the first recommended business is used to review the target loan business. The first model is trained by multiple groups of first training samples, and each group of first training samples includes historical identity information and a first historical recommended business.

[0046] Specifically, the first model can be a machine learning model, such as a Bert model. A machine learning model using a neural network is trained by multiple sets of first training samples to obtain a trained first model, and the first model outputs a first recommended business based on the input identity information. The first recommended business is a business that reviews the target loan business. For example, a business that reviews the user's loan qualifications and identity information. In order to efficiently handle the approval process for the target loan business, business recommendation models in both customer and industry dimensions (i.e., the first model and the second model) are used to support business personnel in viewing the first recommended business in the customer dimension and the second recommended business in the industry dimension.

[0047] Step S103: Input the business information into the second model to obtain a second recommended business, wherein the second recommended business is used to review the target loan business. The second model is trained by multiple groups of second training samples, and each group of second training samples includes historical business information and a second historical recommended business.

[0048] Specifically, the second model can be a machine learning model. The machine learning model using a neural network is trained using multiple sets of second training samples to obtain a trained second model. The second model then outputs a second recommended business based on the input business information. The second recommended business is a business that reviews the target loan business. For example, a business that reviews whether the loan process or business complies with the relevant regulations of the financial institution.

[0049] Step S104 , extracting the target approval document from the business database, and processing the first recommended business and the second recommended business according to the target approval document to obtain a processing result.

[0050] Specifically, the business database can be a database that stores various loan approval documents for various business representatives at a financial institution. The target approval documents can be the documents used to complete the approval processes for the first and second recommended businesses. Business representatives upload the relevant loan approval documents to the business database in advance. When a loan application, such as the first and second recommended business, is required, the target approval documents required for the loan application process are automatically retrieved from the business database. The first and second recommended business applications are processed using the target approval documents, resulting in a processing result, which is also the approval result for the target loan application.

[0051] The loan review business processing method provided in the embodiment of the present application obtains the identity information of a user who needs to apply for a target loan business and obtains the business information of the target loan business; inputs the identity information into a first model to obtain a first recommended business, wherein the first recommended business is used to review the target loan business, and the first model is trained by multiple groups of first training samples, each group of first training samples includes historical identity information and a first historical recommended business; inputs the business information into a second model to obtain a second recommended business, wherein the second recommended business is used to review the target loan business, and the second model is trained by multiple groups of second training samples, each group of second training samples includes historical business information and a second historical recommended business; extracts the target approval document from the business database, and processes the first recommended business and the second recommended business based on the target approval document to obtain a processing result, thereby solving the problem of low loan review efficiency in the related art. By inputting the user's identity information into the first model to obtain the first recommended business, inputting the business information into the second model to obtain the second recommended business, and processing the first recommended business and the second recommended business based on the target approval document extracted from the business database to obtain a processing result for approving the target loan business, thereby achieving the effect of improving loan review efficiency.

[0052] In order to improve the efficiency of loan review, the approval documents entered by the business personnel are uniformly stored in the business database. Optionally, in the loan review business processing method provided in the embodiment of the present application, before extracting the target approval document from the business database, the method also includes: receiving the approval information entered by the business personnel for approving the target loan business, generating an approval document based on the approval information; storing the approval document in the business database, wherein the business database contains approval documents for multiple types of loan businesses.

[0053] Specifically, the business personnel can be the relevant personnel in charge of the loan review process of the financial institution, such as the account manager who receives the user before the loan, the reviewer and re-reviewer during the loan, and the post-loan personnel, etc. The approval information can be the information entered by the business personnel into the digital credit system for the target loan business, and the business database can be the database of the digital credit system of the financial institution. Figure 2 Schematic diagram of storing approval documents into a business database according to an embodiment of the present application, such as Figure 2 As shown, the account manager logs into the digital credit system, obtains relevant approval information of the target loan business from the internal information sources and external information sources of the financial institution, enters the approval information into the digital credit system, generates an approval document and submits it to the digital credit system platform for review. If the review fails, the account manager writes a review opinion on the approval document and returns it to the account manager. If the review passes, the document is stored in the business database of the digital credit system.

[0054] The reviewer logs into the digital credit system, reviews the authenticity of the target loan information, checks whether there are any missing materials in the relevant documents of the target loan, writes an audit report and enters it into the digital credit system, generates an approval document and submits it to the digital credit system platform for review. If the review fails, the reviewer writes a review opinion on the approval document and returns it to the reviewer. If the review passes, it is stored in the business database of the digital credit system. The reviewer logs into the digital credit system, reviews the authenticity of the target loan information, checks whether there are any missing materials in the relevant documents of the target loan, writes a review report and enters it into the digital credit system, generates an approval document and stores it in the business database of the digital credit system. The post-loan staff logs into the digital credit system, enters the post-loan inspection report on the target loan, generates an approval document and stores it in the business database of the digital credit system.

[0055] The embodiment of the present application realizes the online, digital and intelligent transformation of the loan review business process under the premise of ensuring data compliance with regulations and security. By introducing online editors and component capabilities, business personnel are supported to introduce content fragments required for approval documents into the editor by dragging and dropping. At the same time, the source of the content fragments is recorded through the tracking technology. By running through the entire credit process, a one-stop document writing platform is provided to business personnel. It provides multiple capabilities such as online document editing, data traceability, search engine construction, content-level control, business recommendations, etc., which improves the efficiency of loan review.

[0056] In order to improve data security, a desensitizing operation is performed on the approval document. Optionally, in the loan review business processing method provided in the embodiment of the present application, before storing the approval document in the business database, the method further includes: extracting sensitive data from the approval document, wherein the sensitive data includes sensitive text and sensitive images, and the sensitive text includes at least one of the following: legal person name, shareholder name, company name and holding company name, and the sensitive image includes at least one of the following: equity structure chart and corporate legal person holding ratio chart; performing a desensitizing operation on the approval document through a natural language processing model to obtain a desensitized approval document, and storing the desensitized approval document in the business database.

[0057] Specifically, natural language processing models can be the Bert-Span model and the Resnet model. Because loan review involves sensitive business information that cannot be disclosed to users, content-level control involves desensitizing sensitive text based on the Bert-Span model, and desensitizing sensitive images based on the Resnet model. This embodiment ensures information security for loan review by desensitizing approval documents.

[0058] Optionally, in the loan review business processing method provided in the embodiment of the present application, a desensitizing operation is performed on the approval document through a natural language processing model to obtain the desensitized approval document, including: performing a filtering operation on the approval document to obtain an updated approval document, wherein the filtering operation includes at least one of the following: stop word filtering and common word filtering; retrieving sensitive text from the updated approval document, and encrypting the sensitive text through a preset mask in the natural language processing model to obtain an approval document with the sensitive text masked; extracting all images from the approval document with the sensitive text masked, and inputting all images into an image recognition model to obtain a vector set of all images, wherein the vector set contains multiple vectors, and each vector corresponds to an image; retrieving a target vector from the vector set, wherein the target vector is a vector corresponding to the sensitive image; encrypting the image corresponding to the target vector through a preset mask to obtain an approval document with the sensitive image masked.

[0059] Specifically, Figure 3 Schematic diagram of the desensitization method for approval documents provided in accordance with the embodiment of the present application. Figure 3As shown, the approval document is stored in the business database, and the approval document stored in Docx (a file format) format is exported from the business database. Desensitization operations are performed on the approval document, including text desensitization and image desensitization. During the text desensitization process, the approval document is input into the natural language processing model, that is, the Bert-Span model. By fine-tuning the data of the natural speech processing model, the extraction of relevant sensitive texts such as legal person names, shareholder names, company names, and holding company names is achieved, and mixed extraction of Chinese and English is supported. At the same time, regular expressions are introduced as a supplement to model extraction to achieve the extraction of relevant sensitive texts such as stock names and company abbreviations. After stop word filtering and common word filtering operations, the updated approval document is obtained. Sensitive text is extracted from the updated approval document through extraction methods such as regular extraction, and merged and deduplicated to obtain sensitive text that needs to be desensitized. The sensitive text is encrypted using the preset mask in the natural language processing model to obtain the approval document with the sensitive text masked.

[0060] During the image desensitization process, an image embedding model was trained based on the ResNet base to achieve vectorization of image data. Sensitive images in all approval documents in the business database are vectorized and stored in the image vector database. When performing image desensitization on approval documents, all images in the approval documents are vectorized, and the image vector database is retrieved to see whether there is a vector corresponding to the image in the approval document. If so, it means that there is a sensitive image in the approval document. The image classification task is converted into an image retrieval task. Sensitive image recognition is achieved. The image corresponding to the target vector (that is, the sensitive image) is encrypted using a preset mask to obtain the approval document with the sensitive image masked.

[0061] The embodiment of the present application avoids the leakage of sensitive text and sensitive images by performing a desensitization operation on the approval document.

[0062] In order to improve loan review efficiency, a retrieval function for approval documents is provided. Optionally, in the loan review business processing method provided in the embodiment of the present application, extracting a target approval document from the business database includes: receiving an approval document query request, extracting keywords from the approval document query request, and retrieving a first approval document containing the keywords from the business database; inputting the text content in the approval document query request into a semantic recognition model to obtain a semantic vector, and retrieving a second approval document containing the semantic vector from the business database; merging the first approval document and the second approval document, and eliminating duplicate approval documents to obtain the target approval document.

[0063] Specifically, in order to improve the efficiency of the first and second recommended services, the business database storing the approval documents provides a search function. By receiving the approval document query request from the loan examiner, the target approval document is retrieved from the business database based on the keywords and text content in the approval document query request. For example, Figure 4 Schematic diagram of extracting target approval documents according to an embodiment of the present application, such as Figure 4 As shown, the loan examiner initiates a query request, and according to the query request, the recall results based on ES (ElasticSearch, an open source search engine) retrieval and the recall results based on semantic retrieval are realized in the business database. After the two recall results are inverted indexed, the document sets are merged, and the contents are merged, the search results are sorted to obtain the target approval documents.

[0064] It should be noted that the search function of the business database can be implemented by introducing semantic search technology based on ES search, which can accurately feedback the search content even when the user's intention is unclear. It helps business personnel to quickly browse documents and save time in system searches. The embodiment of this application improves the efficiency of loan review business by providing a search function for approval documents.

[0065] Optionally, in the loan review business processing method provided in the embodiment of the present application, before retrieving the first approval document containing keywords from the business database, the method also includes: processing the approval document into a preset format, wherein the approval file in the preset format does not support downloading; parsing the file in the preset format to obtain word segments and sentences; constructing a keyword index based on the word segments, and constructing a semantic vector index based on the sentences; and constructing an inverted index through the keyword index and the semantic vector index.

[0066] Specifically, the preset format may be HTML format. In order to implement the indexing function of the business database, it is necessary to construct an inverted index for the approval documents in the business database. Figure 5 is a schematic diagram of constructing an inverted index according to an embodiment of the present application, such as Figure 5 As shown in the figure, when uploading approval documents to the business database, they must first be preprocessed into HTML format to prevent the front-end from downloading the original approval documents. The loan review system downloads the approval documents from the business database and parses the text within them to build an Elasticsearch index. Furthermore, the text within the approval documents is converted into semantic vectors using the offline-trained semantic model Sentence Bert. These vectors are then stored in the vector database. Finally, the Elasticsearch index and the semantic index are merged to form an inverted index file.

[0067] By constructing an inverted index of approval documents, this embodiment of the application improves the efficiency of obtaining approval documents, promotes the digital transformation of the entire credit business process, and maximizes the channels for business personnel to obtain information while ensuring business security. It narrows the knowledge gap between business personnel and improves business processing efficiency. It also provides a foundation for the accumulation of business knowledge.

[0068] In order to recommend the target loan business to the loan review personnel for approval, it is necessary to train the first model. Optionally, in the loan review business processing method provided in the embodiment of the present application, the first model is trained in the following manner: obtaining business processing records, and extracting multiple review businesses from the business processing records; determining the historical identity information of the user who handled the review business, and determining the review business as the first historical recommended business; determining each first historical recommended business and the historical identity information corresponding to the first historical recommended business as a group of first training samples to obtain multiple groups of first training samples; training the preset neural network model through multiple groups of first training samples to obtain the first model.

[0069] Specifically, Figure 6 Schematic diagram of training the first model and the second model according to an embodiment of the present application, such as Figure 6 As shown, business processing records are obtained from the business database, customer data (that is, historical identity information) and industry data (historical business information) are requested, and the audit business and the historical identity information corresponding to the audit business are used as the first training samples to train the preset neural network model to obtain the first model. The preset neural network model includes Embedding Model (a natural language processing model), DNN Model (deep learning model), WDL model, Concate, FC and Sigmoid.

[0070] It should be noted that in the service recommendation process, this embodiment feeds identity information and service information into the Deep and Wide layers of the WDL model, respectively, ultimately implementing the Sigmoid activation function to recommend the first and second recommended services. Backend processing categorizes the candidate services by those with similar industry and customer dimensions, enabling service recommendations across different dimensions.

[0071] In order to recommend the target loan business to the loan review personnel for approval, it is necessary to train a second model. Optionally, in the loan review business processing method provided in the embodiment of the present application, the second model is trained in the following manner: obtaining business processing records, and extracting multiple review businesses from the business processing records; determining the historical business information of the loan business that requires review business, and determining the review business as the second historical recommended business; determining each second historical recommended business and the historical business information corresponding to the second historical recommended business as a group of second training samples to obtain multiple groups of second training samples; training the preset neural network model through multiple groups of second training samples to obtain the second model.

[0072] Specifically, if Figure 6 As shown, business processing records are obtained from the business database, customer data (i.e., historical identity information) and industry data (historical business information) are requested, and the audit business and the historical business information corresponding to the audit business are used as the second training sample to train the preset neural network model to obtain the second model. This embodiment of the application realizes loan review business recommendations based on the business dimension by training the second model.

[0073] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0074] The present application also provides a loan review business processing device. It should be noted that the loan review business processing device of the present application embodiment can be used to execute the loan review business processing method provided in the present application embodiment. The loan review business processing device provided in the present application embodiment is introduced below.

[0075] Figure 7 Schematic diagram of a loan review business processing device according to an embodiment of the present application. Figure 7 As shown, the device includes:

[0076] The acquisition unit 701 is used to acquire the identity information of the user who needs to apply for the target loan business and acquire the business information of the target loan business;

[0077] A first input unit 702 is configured to input identity information into a first model to obtain a first recommended business, wherein the first recommended business is used to review a target loan business. The first model is trained using multiple sets of first training samples, each set of first training samples including historical identity information and a first historical recommended business.

[0078] A second input unit 703 is configured to input business information into a second model to obtain a second recommended business, wherein the second recommended business is used to review the target loan business. The second model is trained by multiple sets of second training samples, each set of second training samples including historical business information and a second historical recommended business.

[0079] The first extraction unit 704 is configured to extract a target approval document from the business database, and process the first recommended business and the second recommended business according to the target approval document to obtain a processing result.

[0080] The loan review business processing device provided in the embodiment of the present application obtains the identity information of the user who needs to handle the target loan business and obtains the business information of the target loan business through the acquisition unit 701; the first input unit 702 inputs the identity information into the first model to obtain a first recommended business, wherein the first recommended business is used to review the target loan business, and the first model is trained by multiple groups of first training samples, each group of first training samples includes historical identity information and first historical recommended business; the second input unit 703 inputs the business information into the second model to obtain a second recommended business, wherein the second recommended business is used to review the target loan business, and the second model is trained by multiple groups of second training samples. The sample training is obtained, and each group of second training samples includes historical business information and second historical recommended business; the first extraction unit 704 extracts the target approval document from the business database, and processes the first recommended business and the second recommended business through the target approval document to obtain the processing result, which solves the problem of low loan review efficiency in the related technology. The first recommended business is obtained by inputting the user's identity information into the first model, and the second recommended business is obtained by inputting the business information into the second model. The first recommended business and the second recommended business are processed according to the target approval document extracted from the business database to obtain the processing result of approving the target loan business, thereby achieving the effect of improving the loan review efficiency.

[0081] Optionally, in the loan review business processing device provided in the embodiment of the present application, the device also includes: a receiving unit, used to receive approval information entered by business personnel for approving the target loan business, and generate an approval document through the approval information; a storage unit, used to store the approval document in a business database, wherein the business database contains approval documents for multiple types of loan businesses.

[0082] Optionally, in the loan review business processing device provided in the embodiment of the present application, the device also includes: a second extraction unit, used to extract sensitive data from the approval document, wherein the sensitive data includes sensitive text and sensitive images, and the sensitive text includes at least one of the following: legal person name, shareholder name, company name and holding company name, and the sensitive image includes at least one of the following: equity structure chart and corporate legal person holding ratio chart; an execution unit, used to perform a desensitizing operation on the approval document through a natural language processing model to obtain a desensitized approval document, and store the desensitized approval document in a business database.

[0083] Optionally, in the loan review business processing device provided in the embodiment of the present application, the execution unit includes: an execution module, which is used to perform a filtering operation on the approval document to obtain an updated approval document, wherein the filtering operation includes at least one of the following: stop word filtering and common word filtering; a first retrieval module, which is used to retrieve sensitive text from the updated approval document, and encrypt the sensitive text through a preset mask in the natural language processing model to obtain an approval document with the sensitive text masked; an extraction module, which is used to extract all images from the approval document with the sensitive text masked, and input all images into the image recognition model to obtain a vector set of all images, wherein the vector set contains multiple vectors, and each vector corresponds to an image; a second retrieval module, which is used to retrieve a target vector from the vector set, wherein the target vector is a vector corresponding to the sensitive image; an encryption module, which is used to encrypt the image corresponding to the target vector through a preset mask to obtain an approval document with the sensitive image masked.

[0084] Optionally, in the loan review business processing device provided in the embodiment of the present application, the first extraction unit 704 includes: a receiving module, used to receive an approval document query request, extract keywords from the approval document query request, and retrieve a first approval document containing the keywords from the business database; an input module, used to input the text content in the approval document query request into a semantic recognition model to obtain a semantic vector, and retrieve a second approval file containing the semantic vector from the business database; a merging module, used to merge the first approval document and the second approval document, and eliminate duplicate approval documents to obtain a target approval document.

[0085] Optionally, in the loan review business processing device provided in the embodiment of the present application, the first extraction unit 704 also includes: a processing module, used to process the approval document into a preset format, wherein the approval document in the preset format does not support downloading; a parsing module, used to parse the file in the preset format to obtain word segments and sentences; a first construction module, used to construct a keyword index based on word segments and a semantic vector index based on sentences; a second construction module, used to construct an inverted index through the keyword index and the semantic vector index.

[0086] Optionally, in the loan review business processing device provided in the embodiment of the present application, the first model is trained in the following manner: obtaining business processing records, extracting multiple review businesses from the business processing records; determining the historical identity information of the user who processed the review business, and determining the review business as the first historical recommended business; determining each first historical recommended business and the historical identity information corresponding to the first historical recommended business as a group of first training samples to obtain multiple groups of first training samples; training the preset neural network model through multiple groups of first training samples to obtain the first model.

[0087] Optionally, in the loan review business processing device provided in the embodiment of the present application, the second model is trained in the following manner: obtaining business processing records, extracting multiple review businesses from the business processing records; determining historical business information of loan businesses that require review business, and determining the review business as a second historical recommended business; determining each second historical recommended business and the historical business information corresponding to the second historical recommended business as a group of second training samples to obtain multiple groups of second training samples; training the preset neural network model through multiple groups of second training samples to obtain the second model.

[0088] The loan review business processing device includes a processor and a memory. The above-mentioned acquisition unit 701, first input unit 702, second input unit 703 and first extraction unit 704 are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize corresponding functions.

[0089] The processor contains a kernel, which retrieves the corresponding program unit from the memory. You can have one or more kernels, and by adjusting kernel parameters, you can improve loan approval efficiency.

[0090] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0091] An embodiment of the present invention provides a computer-readable storage medium having a program stored thereon, which implements a loan review business processing method when the program is executed by a processor.

[0092] An embodiment of the present invention provides a processor, which is used to run a program, wherein the program executes a loan review business processing method when running.

[0093] Figure 8 Schematic diagram of an electronic device according to an embodiment of the present application. Figure 8As shown, electronic device 801 includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, the following steps are performed: obtaining identity information of a user who needs to apply for a target loan service and obtaining service information of the target loan service; inputting the identity information into a first model to obtain a first recommended service, wherein the first recommended service is used to review the target loan service, and the first model is trained using multiple sets of first training samples, each set of first training samples including historical identity information and a first historical recommended service; inputting service information into a second model to obtain a second recommended service, wherein the second recommended service is used to review the target loan service, and the second model is trained using multiple sets of second training samples, each set of second training samples including historical service information and a second historical recommended service; extracting target approval documents from a service database, and applying the first and second recommended services based on the target approval documents to obtain a processing result. The device herein may be a server, a PC, a PAD, a mobile phone, or the like.

[0094] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program that is initialized with the following method steps: obtaining identity information of a user who needs to handle a target loan business, and obtaining business information of the target loan business; inputting the identity information into a first model to obtain a first recommended business, wherein the first recommended business is used to review the target loan business, and the first model is trained by multiple groups of first training samples, and each group of first training samples includes historical identity information and a first historical recommended business; inputting business information into a second model to obtain a second recommended business, wherein the second recommended business is used to review the target loan business, and the second model is trained by multiple groups of second training samples, and each group of second training samples includes historical business information and a second historical recommended business; extracting a target approval document from a business database, and handling the first recommended business and the second recommended business through the target approval document to obtain a handling result.

[0095] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0096] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0097] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0098] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0099] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0100] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0101] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0102] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0103] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0104] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A loan review business processing method, characterized in that: include: Obtaining the identity information of the user who needs to apply for the target loan business and obtaining the business information of the target loan business; Inputting the identity information into a first model to obtain a first recommended business, wherein the first recommended business is used to review the target loan business, the first model is trained by multiple groups of first training samples, each group of first training samples includes historical identity information and first historical recommended businesses, and the first recommended business is used to review the loan qualifications of the user; Inputting the business information into a second model to obtain a second recommended business, wherein the second recommended business is used to review the target loan business, the second model is trained by multiple groups of second training samples, each group of second training samples includes historical business information and second historical recommended businesses, and the second recommended business is used to review the loan process; extracting a target approval document from a business database, and processing the first recommended business and the second recommended business using the target approval document to obtain a processing result; The step of extracting the target approval document from the business database includes: receiving an approval document query request, extracting keywords from the approval document query request, and retrieving a first approval document containing the keywords from the business database; inputting text content in the approval document query request into a semantic recognition model to obtain a semantic vector, and retrieving a second approval document containing the semantic vector from the business database; and merging the first approval document and the second approval document, and eliminating duplicate approval documents to obtain the target approval document.

2. The method according to claim 1, characterized in that Before extracting the target approval document from the business database, the method further includes: receiving approval information entered by the business personnel for the target loan business, and generating an approval document based on the approval information; The approval document is stored in the business database, wherein the business database contains approval documents for multiple types of loan businesses.

3. The method according to claim 2, characterized in that Before storing the approval document in the business database, the method further includes: Extracting sensitive data from the approval document, wherein the sensitive data includes sensitive text and sensitive images, wherein the sensitive text includes at least one of the following: legal person name, shareholder name, enterprise name, and holding enterprise name; and the sensitive image includes at least one of the following: equity structure chart and corporate legal person holding ratio chart; A desensitization operation is performed on the approval document through a natural language processing model to obtain a desensitized approval document, and the desensitized approval document is stored in the business database.

4. The method according to claim 3, characterized in that The approval document is desensitized using a natural language processing model, and the desensitized approval document includes: Performing a filtering operation on the approval document to obtain an updated approval document, wherein the filtering operation includes at least one of the following: stop word filtering and common word filtering; Retrieving the sensitive text from the updated approval document, and encrypting the sensitive text using a preset mask in the natural language processing model to obtain an approval document with the sensitive text masked; Extracting all images from the approval document with the sensitive text masked, and inputting all images into an image recognition model to obtain a vector set of all images, wherein the vector set includes multiple vectors, each vector corresponding to one image; Retrieving a target vector from the vector set, wherein the target vector is a vector corresponding to the sensitive image; The image corresponding to the target vector is encrypted using the preset mask to obtain an approval document with the sensitive image masked.

5. The method according to claim 1, wherein Before retrieving the first approval document containing the keyword from the business database, the method further includes: Processing the approval document into a preset format, wherein the approval document in the preset format does not support downloading; Parsing the file in the preset format to obtain word segmentation and sentence segmentation; Constructing a keyword index based on the word segmentation and constructing a semantic vector index based on the sentence segmentation; An inverted index is constructed using the keyword index and the semantic vector index.

6. The method according to claim 1, characterized in that The first model is trained in the following way: Obtaining business processing records, and extracting multiple audited businesses from the business processing records; Determining historical identity information of the user who handled the review service, and determining the review service as a first historically recommended service; Determine each first historical recommended service and the historical identity information corresponding to the first historical recommended service as a group of first training samples, to obtain multiple groups of first training samples; The preset neural network model is trained using the multiple groups of first training samples to obtain the first model.

7. The method according to claim 1, characterized in that The second model is trained in the following way: Obtaining business processing records, and extracting multiple audited businesses from the business processing records; Determining historical business information of the loan business that needs to undergo the review business, and determining the review business as a second historical recommended business; Determine each second historical recommended service and the historical service information corresponding to the second historical recommended service as a group of second training samples, to obtain multiple groups of second training samples; The preset neural network model is trained using the multiple groups of second training samples to obtain the second model.

8. A loan review business processing device, characterized in that: include: An acquisition unit, configured to acquire identity information of a user who needs to apply for a target loan business, and acquire business information of the target loan business; a first input unit, configured to input the identity information into a first model to obtain a first recommended business, wherein the first recommended business is used to review the target loan business, the first model being trained by multiple sets of first training samples, each set of first training samples including historical identity information and first historical recommended businesses, and the first recommended business being used to review the loan qualifications of the user; a second input unit, configured to input the business information into a second model to obtain a second recommended business, wherein the second recommended business is used to review the target loan business, the second model is trained by multiple sets of second training samples, each set of second training samples includes historical business information and second historical recommended businesses, and the second recommended business is used to review the loan process; An extraction unit is configured to extract a target approval document from a business database, and process the first recommended business and the second recommended business using the target approval document to obtain a processing result; The extraction unit includes: a receiving module for receiving an approval file query request, extracting keywords from the approval file query request, and retrieving a first approval file containing the keywords from the business database; an input module for inputting text content in the approval file query request into a semantic recognition model to obtain a semantic vector, and retrieving a second approval file containing the semantic vector from the business database; and a merging module for merging the first approval file and the second approval file, and eliminating duplicate approval files to obtain the target approval file.

9. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored program, wherein when the program is run, it controls the device where the non-volatile storage medium is located to execute the loan review business processing method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: It includes one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the loan review business processing method described in any one of claims 1 to 7.

Images