Tax auditing method, device and equipment and storage medium
By configuring multiple partitions for language models and loading language models, the problems of high development costs and cumbersome auditing processes in the prior art when updating tax audit requirements are solved, and more efficient tax auditing is achieved.
Patent Information
- Application Number
- CN202411752956.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-02
- Publication Date
- 2025-03-25
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
When the existing technology updates the tax audit requirements, the development cost is high, and the calling process of multiple audit models is cumbersome, resulting in low tax audit efficiency.
By configuring multiple partitions for the language model in memory, configuring task queues and tax audit specification information in each partition, loading the language model, receiving tax audit requests, extracting tax materials, calling the language model for auditing, and summarizing the audit results.
It reduces the workload of updating tax audit specification information, reduces development costs, simplifies the calling process of multiple audit models, and improves the efficiency of tax audits.
Smart Images
Figure CN119693161A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of natural language processing, and in particular to a tax audit method, device, equipment and storage medium. Background Art
[0002] With the digital upgrade and intelligent transformation of taxation, many places have begun to implement online tax processing, build smart taxation, steadily implement electronic reforms, and strive to reduce costs.
[0003] Online tax processing mainly uses artificial intelligence to assist in audits. At present, when updating audit requirements, such as adding new audit requirements or modifying audit requirements, a new audit model will be built, and the development cost is high. In addition, multiple audit models have been accumulated over a long period of time, and multiple audit models are called for auditing the same tax document. The process of calling multiple audit models is relatively cumbersome, which makes the efficiency of tax audits low. Summary of the invention
[0004] In view of this, the present invention provides a tax audit method, apparatus, device and storage medium to improve the efficiency of tax audit.
[0005] A first aspect of the present invention provides a tax audit method, comprising:
[0006] Determine a language model trained based on tax data; configure multiple partitions for the language model in memory; each partition is configured with a task queue and tax audit specification information;
[0007] Loading the language model in each of the partitions;
[0008] A tax audit request sent by a client is received; the tax audit request includes a tax declaration document;
[0009] Extracting multiple tax materials from the tax declaration documents;
[0010] Writing the tax material into the task queue of the partition adapted to the tax material;
[0011] In each of the partitions, calling the language model to review the tax materials in the task queue according to the tax review specification information to obtain a tax review sub-result;
[0012] The tax audit sub-results are aggregated into a total tax audit result of the tax declaration document.
[0013] Optionally, loading the language model in each partition includes:
[0014] loading at least one instance for the language model in the partition;
[0015] The read permission of the task queue and the tax audit specification information in the partition is configured for at least one of the instances.
[0016] Optionally, loading at least one instance for the language model in the partition includes:
[0017] counting a first quantity of the tax materials in the task queue in the partition;
[0018] The maximum value between 1 and the task ratio is taken as a reference value; the task ratio is the ratio between the first number and the second number configured for an instance of the language model;
[0019] Rounding up the square root of the reference value to obtain a third quantity;
[0020] A third number of instances is maintained for the language model in the partition.
[0021] Optionally, maintaining instances that match the third number for the language model in the partition comprises:
[0022] counting a fourth number of instances generated for the language model in the partition;
[0023] If the fourth number is less than the third number, generating a new instance for the language model in the partition according to a fifth number; the fifth number is a difference between the third number and the fourth number;
[0024] If the fourth number is greater than the third number, the instances generated for the language model are deleted in the partition according to a sixth number; the sixth number is a difference between the fourth number and the third number.
[0025] Optionally, in each of the partitions, calling the language model to review the tax materials in the task queue according to the tax review specification information to obtain a tax review sub-result includes:
[0026] In each of the partitions, calling the language model to read the tax materials from the task queue, and reading the tax audit specification information;
[0027] Calling the language model to construct the tax materials and the tax audit specification information into a guideline;
[0028] The language model is called to review the tax materials according to the instructions to obtain a tax review sub-result.
[0029] Optionally, calling the language model to construct the tax materials and the tax audit specification information into a guideline includes:
[0030] If the tax declaration document is not reviewed for the first time, the historical tax review results are queried; the historical tax review results are tax review sub-results obtained from the historical review of the same tax material in the tax declaration document;
[0031] The language model is called to construct the tax materials, the tax audit specification information and the historical tax audit results into guidance language.
[0032] Optionally, calling the language model to construct the tax materials and the tax audit specification information into a guideline includes:
[0033] Querying the business information of the enterprise to which the client belongs;
[0034] The language model is called to construct the tax materials, the tax audit specification information and the business information into a guideline.
[0035] A second aspect of the present invention provides a tax audit device, comprising:
[0036] A language model determination module is used to determine a language model trained based on tax data; multiple partitions are configured in the memory for the language model; each partition is configured with a task queue and tax audit specification information;
[0037] A language model loading module, used for loading the language model in each of the partitions;
[0038] A tax audit request receiving module, used for receiving a tax audit request sent by a client; the tax audit request includes a tax declaration document;
[0039] A tax material extraction module, used for extracting multiple tax materials from the tax declaration document;
[0040] A tax material writing module, used for writing the tax material into the task queue of the partition adapted to the tax material;
[0041] A tax material review module, used for, in each of the partitions, calling the language model to review the tax materials in the task queue according to the tax review specification information to obtain a tax review sub-result;
[0042] The tax audit total result generating module is used to aggregate the tax audit sub-results into the tax audit total result of the tax declaration document.
[0043] Optionally, the language model loading module includes:
[0044] an instance loading module, configured to load at least one instance for the language model in the partition;
[0045] The read permission configuration module is used to configure the read permission of the task queue and the tax audit specification information in the partition for at least one of the instances.
[0046] Optionally, the instance loading module includes:
[0047] A queue quantity statistics module, used for counting the first quantity of the tax materials in the task queue in the partition;
[0048] A reference value calculation module, configured to take the maximum value between 1 and a task ratio as a reference value; the task ratio is a ratio between the first number and a second number configured for an instance of the language model;
[0049] generating a quantity calculation module, configured to round up the square root of the reference value to an integer to obtain a third quantity;
[0050] An instance maintenance module is used to maintain instances that meet the third number for the language model in the partition.
[0051] Optionally, the instance maintenance module includes:
[0052] a partition quantity counting module, configured to count a fourth quantity of instances generated for the language model in the partition;
[0053] an expansion module, configured to generate a new instance for the language model in the partition according to a fifth number if the fourth number is less than the third number; the fifth number being a difference between the third number and the fourth number;
[0054] A shrinking module is used to delete the instances generated for the language model in the partition according to a sixth number if the fourth number is greater than the third number; the sixth number is a difference between the fourth number and the third number.
[0055] Optionally, the tax material review module includes:
[0056] A partition reading module, used for calling the language model in each partition to read the tax materials from the task queue, and to read the tax audit specification information;
[0057] An instruction constructing module, used for calling the language model to construct the tax materials and the tax audit specification information into an instruction;
[0058] The tax audit sub-result generation module is used to call the language model to audit the tax materials according to the instructions to obtain the tax audit sub-result.
[0059] Optionally, the instruction construction module includes:
[0060] A historical tax audit result query module is used to query historical tax audit results if the tax declaration document is not audited for the first time; the historical tax audit results are tax audit sub-results obtained from historical audits of the same tax material in the tax declaration document;
[0061] The context construction module is used to call the language model to construct the tax materials, the tax audit specification information and the historical tax audit results into guidance language.
[0062] Optionally, the instruction construction module includes:
[0063] A business information query module, used to query the business information of the enterprise to which the client belongs;
[0064] The environment construction module is used to call the language model to construct the tax materials, the tax audit specification information and the business information into guidance language.
[0065] A third aspect of the present invention provides an electronic device, the electronic device comprising:
[0066] at least one processor; and
[0067] a memory communicatively connected to the at least one processor; wherein,
[0068] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can perform the tax audit method as described in the first aspect above.
[0069] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the tax audit method as described in the first aspect above.
[0070] A fifth aspect of the present invention provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the tax audit method as described in the first aspect is implemented.
[0071] In this embodiment, a language model trained based on tax data is determined; multiple partitions are configured for the language model in the memory; task queues and tax audit specification information are configured in each partition; the language model is loaded in each partition; a tax audit request sent by a client is received; the tax audit request has a tax declaration document; multiple tax materials are extracted from the tax declaration document; the tax materials are written into the task queue of the partition adapted to the tax materials; in each partition, the language model is called to audit the tax materials in the task queue according to the tax audit specification information to obtain the tax audit sub-result; the tax audit sub-results are summarized as the tax audit total result of the tax declaration document. This embodiment provides a common language model in the tax field, and configures tax audit specification information for various tax materials, so that tax semantic understanding is decoupled from audit logic. When updating audit requirements, the tax audit specification information is mainly updated and the language model is updated as a supplement. The workload of updating the tax audit specification information is low, which can effectively reduce development costs. Moreover, under the framework of one language model and multiple tax audit specification information, the framework is clear and easy to manage, and has strong extensibility. Tax audit can be realized by loading tax audit specification information on demand, which effectively improves the efficiency of tax audit.
[0072] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS
[0073] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0074] Figure 1 This is a flow chart of a tax audit method provided in Example 1 of the present invention.
[0075] Figure 2 It is a structural diagram of a tax audit device provided in Embodiment 2 of the present invention.
[0076] Figure 3 It is a structural schematic diagram of an electronic device provided in Embodiment 3 of the present invention. DETAILED DESCRIPTION
[0077] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below 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 of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.
[0078] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can cover sequential implementations other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof 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 that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0079] Embodiment 1
[0080] See also Figure 1 , showing a flow chart of a tax audit method provided in Embodiment 1 of the present invention, the method can be executed by a tax audit device, the tax audit device can be implemented in the form of hardware and / or software, and the tax audit device can be configured in an electronic device.
[0081] like Figure 1 As shown, the method includes:
[0082] Step 101: Determine a language model trained based on tax data.
[0083] In this embodiment, a language model (LM), especially a large language model (LLM), is constructed and trained based on deep learning.
[0084] The structure of the language model is not limited to artificially designed neural networks, but can also be a neural network optimized by a model quantization method, a neural network searched for tax characteristics by a NAS (Neural Architecture Search) method, and so on. This embodiment does not impose any restrictions on this.
[0085] In practical applications, general tax data can be accumulated as corpus through various channels, and the tax data can be used to train the language model, so that the language model is a special language model in the tax field.
[0086] Furthermore, multiple partitions are configured for the language model in the memory of the computer cluster; each partition is configured with a task queue and tax audit specification information.
[0087] The task queue is used to store tax materials extracted from various tax declaration documents, and the tax audit specification information is the specification for auditing certain tax materials.
[0088] Step 102: Load the language model in each partition.
[0089] In actual applications, the language model can be loaded into each partition for running.
[0090] In one embodiment of the present invention, step 102 may include the following steps:
[0091] Step 1021: Load at least one instance for the language model in the partition.
[0092] In each partition, at least one instance may be loaded for the language model, so that the language model is put into operation.
[0093] In one design, the number of instances loaded for the language model in each partition is a default value, which can be set based on the amount of tasks undertaken by the task queues in each partition historically.
[0094] In another design, the number of instances loaded for the language model in each partition is a dynamically changing value, which can be adjusted according to the amount of tasks undertaken by the task queues in each partition in history to achieve load balancing and improve resource utilization.
[0095] In this design, the first quantity of tax materials in the task queue can be counted in the partition.
[0096] The maximum value between 1 and the task ratio is taken as the reference value; wherein the task ratio is the ratio of the first number to the second number configured for an instance of a language model, which is used to measure the load of a single language model.
[0097] The square root of the reference value is rounded up to obtain the third quantity, thereby achieving nonlinear adjustment.
[0098] Then, the third quantity can be expressed as:
[0099]
[0100] Among them, Num1 is the first number, Num2 is the second number, Num3 is the third number, max is the function of taking the maximum, and Ceil is the function of rounding up.
[0101] At this time, the scaling technology is used in the partition to maintain instances that meet the third number for the language model.
[0102] The so-called scaling technology may refer to counting a fourth number of instances currently generated for the language model in the partition, and comparing the fourth number with the third number.
[0103] If the fourth number is less than the third number, a new instance is generated for the language model in the partition according to the fifth number; wherein the fifth number is the difference between the third number and the fourth number, thereby achieving capacity expansion.
[0104] If the fourth number is greater than the third number, the instances currently generated for the language model are deleted in the partition according to the sixth number; wherein the sixth number is the difference between the fourth number and the third number, thereby achieving capacity reduction.
[0105] At this time, the sixth number of language model instances can be screened in the partition by random or other methods, and the instance is prohibited from reading new tax materials from the task queue. When the instance completes the tax materials being processed, the instance is released.
[0106] Step 1022: configure the read permission of the task queue and tax audit specification information in the partition of at least one instance.
[0107] In the same partition, at least one instance is configured with read permissions for the task queue and the tax audit specification information, so that the instance can read the tax materials from the task queue and read the tax audit specification information.
[0108] Step 103: Receive the tax audit request sent by the client.
[0109] In this embodiment, each enterprise can log in to the client in the government affairs system and submit a tax audit request through the client. The tax audit request contains the tax declaration documents submitted by the relevant enterprise during tax processing, and the tax declaration documents record various tax information of the enterprise during tax processing.
[0110] Step 104: extract multiple tax materials from the tax declaration document.
[0111] In this embodiment, in response to the tax audit request of the client, the tax declaration document can be read from the tax audit request, and a plurality of corresponding tax materials can be extracted from the tax declaration document according to the preset field meanings.
[0112] Step 105: Write the tax materials into the task queue of the partition that is compatible with the tax materials.
[0113] In this embodiment, the tax material may be written into a task queue of a partition adapted to the tax material according to the fields.
[0114] Adaptation means that the tax audit specification information in this partition can be used for the audit of the tax materials.
[0115] Step 106: In each partition, the language model is called to review the tax materials in the task queue according to the tax review specification information to obtain the tax review sub-result.
[0116] In each partition, when a language model is idle, the language model can be called to read tax materials from the task queue, and the tax materials can be reviewed according to the requirements of the tax review specification information to obtain the tax review sub-result.
[0117] In one embodiment of the present invention, step 106 may include the following steps:
[0118] Step 1061: In each partition, call the language model to read tax materials from the task queue, and read tax audit specification information.
[0119] In each partition, the task queue is a FIFO (First Input First Output) queue, and the tax materials are written into the task queue in chronological order. Then, the language model (ie, instance) can be called to read the tax materials from the task queue according to time.
[0120] In addition, the language model (ie, instance) may be called to read tax audit specification information at a specified address.
[0121] Step 1062: Call the language model to construct the tax materials and tax audit specification information into guidance language.
[0122] In this embodiment, the template of the instruction prompt can be pre-loaded in the memory, and the language model (ie, the instance) is called to write the tax materials and the tax audit specification information into the corresponding positions in the template, thereby constructing the instruction prompt.
[0123] Furthermore, it can be determined whether it is the first time to review the tax declaration documents filed by the enterprise based on information such as the ID of the tax declaration documents.
[0124] If the tax declaration document is not reviewed for the first time, you can query the historical tax audit results; the historical tax audit results are the tax audit sub-results obtained from the historical review of the same tax material in the tax declaration document.
[0125] The language model (i.e., instance) is called to write tax materials, tax audit specification information, and historical tax audit results into the template and construct them into a prompt. In this way, the language model (i.e., instance) can obtain contextual information and improve the accuracy of tax audits.
[0126] In addition, the business information of the enterprise to which the client belongs can be queried. For example, if the enterprise enjoys a certain tax preferential policy, in addition to the supporting documents, the relevant business information can be provided as an auxiliary document to confirm from multiple dimensions that the enterprise meets the conditions for enjoying the preferential policy.
[0127] The language model (i.e., instance) is called to write tax materials, tax audit specification information, and business information into the template and construct it as a prompt. In this way, the language model (i.e., instance) can obtain environmental information and improve the accuracy of tax audits.
[0128] It should be noted that the historical tax audit results and business information can be used separately or in combination, and this embodiment does not limit this.
[0129] Step 1063: Call the language model to review the tax materials according to the instructions to obtain the tax review sub-result.
[0130] In the language model (ie, instance), the tax materials can be audited using the tax audit specification information in accordance with the requirements of the prompt, with reference to historical tax audit results and business information, to obtain tax audit sub-results.
[0131] Step 107: Aggregate the tax audit sub-results into the tax audit total result of the tax declaration document.
[0132] In this embodiment, the tax audit sub-results of multiple tax materials in the same financial declaration document can be aggregated together and integrated into the tax audit total result of the tax declaration document using a fixed format, which is provided to the staff as a reference for tax audit.
[0133] In this embodiment, a language model trained based on tax data is determined; multiple partitions are configured for the language model in the memory; task queues and tax audit specification information are configured in each partition; the language model is loaded in each partition; a tax audit request sent by a client is received; the tax audit request has a tax declaration document; multiple tax materials are extracted from the tax declaration document; the tax materials are written into the task queue of the partition adapted to the tax materials; in each partition, the language model is called to audit the tax materials in the task queue according to the tax audit specification information to obtain the tax audit sub-result; the tax audit sub-results are summarized as the tax audit total result of the tax declaration document. This embodiment provides a common language model in the tax field, and configures tax audit specification information for various tax materials, so that tax semantic understanding is decoupled from audit logic. When updating audit requirements, the tax audit specification information is mainly updated and the language model is updated as a supplement. The workload of updating the tax audit specification information is low, which can effectively reduce development costs. Moreover, under the framework of one language model and multiple tax audit specification information, the framework is clear and easy to manage, and has strong extensibility. Tax audit can be realized by loading tax audit specification information on demand, which effectively improves the efficiency of tax audit.
[0134] Embodiment 2
[0135] See also Figure 2 , shows a schematic diagram of the structure of a tax audit device provided by Embodiment 2 of the present invention. Figure 2 As shown, the device comprises:
[0136] The language model determination module 201 is used to determine the language model trained based on the tax data; configure multiple partitions for the language model in the memory; each of the partitions is configured with a task queue and tax audit specification information;
[0137] A language model loading module 202, used to load the language model in each of the partitions;
[0138] The tax audit request receiving module 203 is used to receive a tax audit request sent by a client; the tax audit request includes a tax declaration document;
[0139] A tax material extraction module 204 is used to extract multiple tax materials from the tax declaration document;
[0140] A tax material writing module 205, used for writing the tax material into the task queue of the partition adapted to the tax material;
[0141] A tax material review module 206 is used to call the language model in each partition to review the tax materials in the task queue according to the tax review specification information to obtain a tax review sub-result;
[0142] The tax audit total result generating module 207 is used to aggregate the tax audit sub-results into the tax audit total result of the tax declaration document.
[0143] In one embodiment of the present invention, the language model loading module 202 includes:
[0144] an instance loading module, configured to load at least one instance for the language model in the partition;
[0145] The read permission configuration module is used to configure the read permission of the task queue and the tax audit specification information in the partition for at least one of the instances.
[0146] In one embodiment of the present invention, the instance loading module includes:
[0147] A queue quantity statistics module, used for counting the first quantity of the tax materials in the task queue in the partition;
[0148] A reference value calculation module, configured to take the maximum value between 1 and a task ratio as a reference value; the task ratio is a ratio between the first number and a second number configured for an instance of the language model;
[0149] generating a quantity calculation module, configured to round up the square root of the reference value to an integer to obtain a third quantity;
[0150] An instance maintenance module is used to maintain instances that meet the third number for the language model in the partition.
[0151] In one embodiment of the present invention, the instance maintenance module includes:
[0152] a partition quantity counting module, configured to count a fourth quantity of instances generated for the language model in the partition;
[0153] an expansion module, configured to generate a new instance for the language model in the partition according to a fifth number if the fourth number is less than the third number; the fifth number being a difference between the third number and the fourth number;
[0154] A shrinking module is used to delete the instances generated for the language model in the partition according to a sixth number if the fourth number is greater than the third number; the sixth number is a difference between the fourth number and the third number.
[0155] In one embodiment of the present invention, the tax material review module 206 includes:
[0156] A partition reading module, used for calling the language model in each partition to read the tax materials from the task queue, and to read the tax audit specification information;
[0157] An instruction constructing module, used for calling the language model to construct the tax materials and the tax audit specification information into an instruction;
[0158] The tax audit sub-result generation module is used to call the language model to audit the tax materials according to the instructions to obtain the tax audit sub-result.
[0159] In one embodiment of the present invention, the instruction building module includes:
[0160] A historical tax audit result query module is used to query historical tax audit results if the tax declaration document is not audited for the first time; the historical tax audit results are tax audit sub-results obtained from historical audits of the same tax material in the tax declaration document;
[0161] The context construction module is used to call the language model to construct the tax materials, the tax audit specification information and the historical tax audit results into guidance language.
[0162] In one embodiment of the present invention, the instruction building module includes:
[0163] A business information query module, used to query the business information of the enterprise to which the client belongs;
[0164] The environment construction module is used to call the language model to construct the tax materials, the tax audit specification information and the business information into guidance language.
[0165] The tax audit device provided in the embodiment of the present invention can execute the tax audit method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the tax audit method.
[0166] Embodiment 3
[0167] See also Figure 3 , shows a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, blade servers, mainframe computers, and other suitable computers. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or claimed herein.
[0168] like Figure 3As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.
[0169] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.
[0170] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the tax audit method.
[0171] In some embodiments, the tax audit method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the tax audit method described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to execute the tax audit method in any other appropriate manner (e.g., by means of firmware).
[0172] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.
[0173] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.
[0174] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
[0175] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).
[0176] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.
[0177] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.
[0178] Embodiment 4
[0179] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the tax audit method provided by any embodiment of the present invention is implemented.
[0180] In the process of implementation, the computer program product can be written in one or more programming languages or a combination thereof to perform the computer program code of the present invention, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).
[0181] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.
[0182] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A tax audit method, characterized in that: include: Identify the language model trained on the tax data; Configuring multiple partitions for the language model in the memory; each of the partitions is configured with a task queue and tax audit specification information; Loading the language model in each of the partitions; Receive tax audit request sent by client; The tax audit request contains tax declaration documents; Extracting multiple tax materials from the tax declaration documents; Writing the tax material into the task queue of the partition adapted to the tax material; In each of the partitions, calling the language model to review the tax materials in the task queue according to the tax review specification information to obtain a tax review sub-result; The tax audit sub-results are aggregated into a total tax audit result of the tax declaration document.
2. The method according to claim 1, characterized in that The loading of the language model in each partition includes: loading at least one instance for the language model in the partition; The read permission of the task queue and the tax audit specification information in the partition is configured for at least one of the instances.
3. The method according to claim 2, characterized in that The step of loading at least one instance for the language model in the partition comprises: counting a first quantity of the tax materials in the task queue in the partition; The maximum value between 1 and the task ratio is taken as a reference value; the task ratio is the ratio between the first number and the second number configured for an instance of the language model; Rounding up the square root of the reference value to obtain a third quantity; A third number of instances is maintained for the language model in the partition.
4. The method according to claim 3, characterized in that The maintaining, in the partition, instances matching the third number for the language model comprises: counting a fourth number of instances generated for the language model in the partition; If the fourth number is less than the third number, generating a new instance for the language model in the partition according to a fifth number; the fifth number is a difference between the third number and the fourth number; If the fourth number is greater than the third number, instances generated for the language model are deleted in the partition according to a sixth number; the sixth number is a difference between the fourth number and the third number.
5. The method according to any one of claims 1 to 4, characterized in that In each of the partitions, the language model is called to review the tax materials in the task queue according to the tax review specification information to obtain a tax review sub-result, including: In each of the partitions, calling the language model to read the tax materials from the task queue, and reading the tax audit specification information; Calling the language model to construct the tax materials and the tax audit specification information into a guideline; The language model is called to review the tax materials according to the instructions to obtain a tax review sub-result.
6. The method according to claim 5, characterized in that The calling of the language model to construct the tax materials and the tax audit specification information into a guideline includes: If the tax declaration document is not reviewed for the first time, the historical tax review results are queried; the historical tax review results are tax review sub-results obtained from the historical review of the same tax material in the tax declaration document; The language model is called to construct the tax materials, the tax audit specification information and the historical tax audit results into guidance language.
7. The method according to claim 5, characterized in that The calling of the language model to construct the tax materials and the tax audit specification information into a guideline includes: Querying the business information of the enterprise to which the client belongs; The language model is called to construct the tax materials, the tax audit specification information and the business information into a guideline.
8. A tax audit device, characterized in that: include: A language model determination module, used to determine a language model trained based on tax data; Configuring multiple partitions for the language model in the memory; each of the partitions is configured with a task queue and tax audit specification information; A language model loading module, used for loading the language model in each of the partitions; A tax audit request receiving module is used to receive a tax audit request sent by a client; The tax audit request contains tax declaration documents; A tax material extraction module, used for extracting multiple tax materials from the tax declaration document; A tax material writing module, used for writing the tax material into the task queue of the partition adapted to the tax material; A tax material review module, used for, in each of the partitions, calling the language model to review the tax materials in the task queue according to the tax review specification information to obtain a tax review sub-result; The tax audit total result generating module is used to aggregate the tax audit sub-results into the tax audit total result of the tax declaration document.
9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can perform the tax audit method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the tax audit method according to any one of claims 1 to 7 is implemented.