Financial reimbursement method and device based on artificial intelligence and electronic equipment

Abstract

The invention discloses a financial reimbursement method and device based on artificial intelligence, and electronic equipment. The method comprises the following steps: obtaining target reimbursement data input by a target user; inputting the target reimbursement data into a semantic analysis model to obtain filling intention data output by the semantic analysis model; the semantic analysis model is obtained by training sample reimbursement data and a corresponding order filling intention label result; matching the form filling intention data with a form template library to obtain a target reimbursement template matched with the form filling intention data; performing data filling operation on the target reimbursement template based on the filling intention data to obtain a target reimbursement form; and auditing the target reimbursement form, and carrying out financial reporting based on an auditing confirmation result. According to the method, the problem that mistakes are easy to make when bills are input manually is avoided, and the trouble caused by the fact that new and old employees are not familiar with reimbursement processes and form differences can be avoided.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based financial accounting method, device and electronic equipment. Background Art

[0002] Financial accounting is an important part of the financial management of an enterprise or organization. It mainly refers to the process in which employees or departments organize and record relevant expense information in accordance with the company's established financial regulations and procedures in order to reimburse expenses incurred in carrying out business activities, and submit it to the financial department for review and processing, ultimately realizing expense reimbursement and financial accounting.

[0003] As the scale of enterprises expands and the complexity of business increases, financial management becomes more and more important. As the types of financial reimbursement attachments increase, the traditional financial reimbursement method of manually entering business documents leads to errors in the process of entering and filling out forms, and the filling and review process is time-consuming and inefficient. At the same time, due to differences in reimbursement processes and form formats among different companies, employees often face great challenges in adapting to new reimbursement requirements, which further increases the complexity of reimbursement and the possibility of errors. Summary of the invention

[0004] The purpose of the present invention is to solve the problems in the prior art and to propose a financial accounting method, device and electronic equipment based on artificial intelligence.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: A financial accounting method based on artificial intelligence comprises the following steps: Obtain target reimbursement data input by target users; Input the target reimbursement data into a semantic parsing model to obtain the form filling intention data output by the semantic parsing model; the semantic parsing model is trained by the sample reimbursement data and its corresponding form filling intention label results; Matching the form filling intention data with a form template library to obtain a target reimbursement template that matches the form filling intention data; Performing data filling operation on the target reimbursement template based on the form filling intention data to obtain a target reimbursement form; The target reimbursement form is reviewed and financial reporting is performed based on the review and confirmation results.

[0006] A financial accounting device based on artificial intelligence, comprising: An acquisition unit, used for acquiring target reimbursement data input by a target user; A model output unit, used to input the target reimbursement data into a semantic parsing model to obtain the form filling intention data output by the semantic parsing model; the semantic parsing model is trained by the sample reimbursement data and its corresponding form filling intention label results; A matching unit, configured to match the form filling intention data with a form template library based on the form filling intention data to obtain a target reimbursement template that matches the form filling intention data; A writing unit, configured to perform a data filling operation on the target reimbursement template based on the form filling intention data to obtain a target reimbursement form; Uploading unit: used to review the target reimbursement form and make financial reports based on the review and confirmation results.

[0007] The present invention also provides an electronic device, comprising a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned artificial intelligence-based financial accounting method when executing the program.

[0008] The present invention also provides a non-transitory readable storage medium on which a program is stored, characterized in that when the program is executed by a processor, the steps of the above-mentioned financial accounting method based on artificial intelligence are implemented.

[0009] Compared with the prior art, the present invention has the following advantages: The present invention provides an artificial intelligence-based financial reimbursement method, device and electronic equipment, which first inputs the target reimbursement data into a semantic parsing model for semantic parsing to obtain the filling intention, and then matches the filling intention with the form database to obtain the corresponding target reimbursement template. Compared with the traditional method, there is no need for manual entry one by one, which not only avoids the problems that are easy to occur in the process of manual document entry, but also avoids the problems of unfamiliarity with the reimbursement process and differences in reimbursement forms for new and old employees, and can flexibly match the corresponding reimbursement template according to different reimbursement data to adapt to changes in corporate reimbursement policies and diversified business scenarios; then, by automatically filling in the target reimbursement form with data, and reviewing and reporting the filled target reimbursement form, the automation of the form data is realized, the reimbursement cycle is greatly shortened, the probability of manual errors is further reduced, and the difficulty of the overall financial reimbursement process is reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0011] Figure 1 A flowchart of a financial accounting method based on artificial intelligence provided by an embodiment of the present invention; Figure 2 A schematic diagram of the structure of a financial accounting device based on artificial intelligence provided by an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of the electronic device proposed by the present invention. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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 are within the scope of protection of the present invention.

[0013] Combine the following Figure 1 - Figure 3 The present invention describes an artificial intelligence-based financial accounting method, device and electronic equipment.

[0014] Figure 1 The figure is a flowchart of a financial accounting method based on artificial intelligence provided by the present invention. Figure 1 As shown, the method includes: Step 101, obtaining target reimbursement data input by a target user.

[0015] Specifically, in the process of obtaining the target reimbursement data, the target user enters the reimbursement data through a special financial reimbursement software platform, the company's internal information system, etc., so as to obtain the target reimbursement data, and the target reimbursement data includes the voice information, text information or picture information entered by the target user within the preset time. The preset time is a timed entry time set according to demand. The combination of multiple input methods and shortcut functions reduces the time and energy of users to manually fill in tedious information and improves the user experience. Among them, the preset time refers to the timed entry time set according to demand, so that when the target user enters the reimbursement data, he only needs to enter part of the data to carry out the subsequent reimbursement process, without having to enter all of it, so as to reduce the difficulty of entry recognition and improve the overall entry efficiency.

[0016] Furthermore, voice information refers to audio data input during the input process. When the system inputs, it is converted into text through the built-in voice conversion technology and then subsequently entered. Text information includes text description information (such as the purpose of expenses, itinerary details, etc.), numerical information (such as amount, quantity, etc.), date data (such as the date of expense incurred, the date of reimbursement, etc.) and related file information (such as contracts, etc.). Image information includes related voucher maps and invoice photos. After the image information is entered into the system, the text content in the image is extracted and converted into an editable text format through optical character recognition technology.

[0017] Furthermore, during the data acquisition process, the system will also perform preliminary format verification and integrity checks on the input data, such as checking whether the date format conforms to the "YYYY-MM-DD" specification, whether the amount is a number and is retained to an appropriate number of decimal places, whether the required fields have content filled in, etc. If the format is found to be wrong or incomplete, the system will promptly pop up a prompt box to inform the user to correct it, so as to ensure that the data format of the subsequent process is basically correct, and improve data quality and processing efficiency.

[0018] Step 102, input the target reimbursement data into the semantic parsing model to obtain the form filling intention data output by the semantic parsing model; the semantic parsing model is trained by the sample reimbursement data and its corresponding form filling intention label results.

[0019] Specifically, when training the semantic parsing model, a large amount of historical reimbursement data covering various types of reimbursements in the enterprise (such as travel expenses, office supplies procurement expenses, business entertainment expenses, etc. in different departments and business scenarios) is first collected, and then the corresponding form filling intention label results are annotated for each sample reimbursement data according to the company's established reimbursement rules. For example, a travel expense reimbursement sample may have a form filling intention label that includes business trips, transportation methods and cost details, accommodation arrangements and expenses, and catering subsidy standards.

[0020] Furthermore, the sample reimbursement data needs to be preprocessed before being used for model training, including text segmentation, removal of stop words, extraction of key feature vectors (such as expense type, amount range, payment method, etc.), with feature vectors as input and form filling intention labels as output, and training using deep learning algorithms (such as RNN, LSTM or models based on attention mechanisms, etc.), and parameters are continuously adjusted during training to minimize the loss function between the predicted form filling intention and the actual label. For details, see the description of steps 1021 to 1023.

[0021] Furthermore, in one embodiment, when the target reimbursement data is input into the trained semantic parsing model, the model will perform feature extraction and semantic understanding on it. For example, for the reimbursement data of the above-mentioned salesperson, the model will identify the expense type of travel expenses, extract the amount information of various expenses such as transportation, accommodation, and catering, as well as key intention information such as the purpose of the business trip such as business negotiation, and output a similar form filling intention data structure such as {Expense type: travel expenses, transportation expenses: [round-trip airfare amount, city transportation amount], accommodation expenses: [hotel name, number of days of accommodation, amount], business purpose: customer negotiation, catering expenses: [amount]}. It should be noted that not all form filling data structures adopt the above form, and can be adjusted according to actual needs and company requirements.

[0022] Step 103, matching the form filling intention data with the form template library to obtain a target reimbursement template that matches the form filling intention data.

[0023] Specifically, during the construction process, the form template library is designed and established by the enterprise according to its own financial reimbursement system, tax regulations and business needs. Each template is customized for different expense types and business scenarios, covering everything from reimbursement header information (such as company name, reimbursement form number, reimbursement date, etc.) to detailed expense details (setting corresponding required and optional fields according to different expense types, such as transportation methods, itinerary start and end locations, accommodation days and standards, catering subsidy standards, etc. in the travel expense template; item name, specifications, quantity, unit price, supplier information, etc. in the office supplies procurement template), as well as approval process related information (clarifying the positions and authority of approval personnel at all levels, such as department managers approving expenses within their authority, financial managers approving key information related to financial compliance, and senior company leaders approving large or special reimbursements, etc.). Reimbursement templates are stored in a structured format in the database for quick retrieval and call-up.

[0024] Furthermore, the form filling intention data is matched with the form template library to obtain a target reimbursement template that matches the form filling intention data. The specific process is as described in steps 1031 to 1035.

[0025] Step 104, based on the form filling intention data, the target reimbursement template is filled with data to obtain the target reimbursement form.

[0026] Specifically, according to the matched target reimbursement template, the information in the form filling intention data is accurately filled in according to the structure and field requirements of the template. For text-type fields, such as the description of the purpose of the expense, the name of the business project, etc., the corresponding text content in the form filling intention data is directly filled in the corresponding position of the template to ensure that the information is complete and accurate. For numeric fields, such as the amount and quantity of each expense, strict data format verification and conversion (such as converting the amount string in the form filling intention data to a numeric type and retaining the appropriate number of decimal places according to the template requirements) are performed before filling in the template. At the same time, for some fields that need to be calculated according to the company's financial rules, such as the conversion of the amount including tax and the amount excluding tax, and the calculation of the actual reimbursement amount according to the reimbursement ratio, the system will automatically call the pre-set calculation formula for calculation and filling. For example, if the company's reimbursement policy stipulates that certain expenses are reimbursed at a rate of 80%, the system will automatically calculate the actual reimbursement amount based on the original expense amount in the form filling intention data and fill it in the corresponding field.

[0027] In addition, for some fields involving associated information, such as department information in the form filling intention data, the budget code or person in charge information of the department is automatically associated and filled into the template to improve the information integrity and accuracy of the reimbursement form, so that it complies with the company's financial management standards and approval process requirements, and generates a complete, standardized and accurate target reimbursement form, including all necessary header information, expense details, approval signature column and other content, which is convenient for subsequent review and financial processing. Through the automated data filling process, the problems of typos, omissions or inconsistent formats that may occur in manual filling are avoided, the accuracy and consistency of the data in the reimbursement form are guaranteed, the quality of financial data is improved, and reliable basic data is provided for financial review and statistical analysis. At the same time, the form filling intention data is quickly and automatically filled into the target reimbursement template, which greatly shortens the time to generate the reimbursement form. Compared with the traditional method of manual filling or manual copying and pasting of data, it significantly improves work efficiency and speeds up the entire financial reimbursement process.

[0028] Step 105, review the target reimbursement form and make a financial report based on the review and confirmation result.

[0029] Specifically, during the audit process, the target reimbursement forms are first audited comprehensively and automatically based on the company's pre-set financial rules and audit standards. For example, the reasonableness of expenses is audited by comparing with the company's internal expense standard database to determine whether each expense is within a reasonable range. Taking travel expenses as an example, according to the travel subsidy standards for employees of different levels, check whether transportation expenses (such as the price cap corresponding to the class of air tickets, regulations on train seat types, etc.), accommodation expenses (according to the accommodation standard cap in different cities) and catering subsidies comply with company policies.

[0030] Furthermore, for data integrity review, ensure that the required fields in the reimbursement form (such as the name of the reimbursement person, department, expense details, invoice number, etc.) are accurately filled in to avoid failure of the review or subsequent financial processing difficulties due to missing information. At the same time, verify the invoice information, including the authenticity query of the invoice (through the connection with the invoice query interface of the tax department, enter the invoice number, code, invoice date and other key information, and obtain the authenticity verification result of the invoice), the consistency check of the invoice header and the company name, and the matching check of the invoice amount and the reimbursement amount.

[0031] For logical consistency review, check whether the logical relationship between the data in the reimbursement form is correct. For example, the total reimbursement amount should be equal to the sum of the detailed amounts of each expense; if there is an expense sharing situation, check whether the sum of the sharing ratio is 100%, and whether the amount after sharing is calculated accurately. For expense reimbursements involving multiple projects or departments, verify the rationality and accuracy of expense allocation to ensure that the expenses borne by each project or department are consistent with the actual business situation.

[0032] At the same time, for reimbursement forms marked as abnormal or high-risk during the automated review process, as well as some complex business scenarios (such as expense reimbursement involving major projects, expense processing for special business matters, etc.), the system automatically transfers them to the manual review stage. Financial personnel can view detailed information on the reimbursement form in the system, including the original target reimbursement data, the form filling intention data after semantic analysis, the matching target reimbursement template, and the review doubts and risk warning information given by the system, and make further review judgments based on their own professional knowledge and experience.

[0033] Furthermore, there is also an audit record and feedback function. The system records the results of automated and manual audits in detail, including different audit statuses and opinions such as audit pass, audit failure (specific reasons for failure must be clearly listed, such as excessive expenses, missing invoices, data logic errors, etc.), and need for additional materials. The audit result information is fed back to the target user in a timely manner, so that the user can clearly understand the processing of the reimbursement application and make corresponding modifications and improvements based on the feedback. At the same time, the audit result data will also be stored in the company's financial audit database for subsequent statistical analysis and audit tracing. After the audit is passed, the target reimbursement form and approval record will be uploaded to facilitate subsequent process operations. In this way, a large number of reimbursement forms can be quickly processed through the full implementation of automated audit rules, which greatly shortens the audit cycle and improves the audit efficiency. Compared with the traditional method of manually reviewing reimbursement forms one by one, the data can be accurately compared and verified in a short time, avoiding audit errors and omissions caused by human factors, and significantly improving the accuracy of the audit.

[0034] The present invention relates to the field of artificial intelligence, and proposes a financial accounting method, device and electronic device based on artificial intelligence. In the present invention, the proposed financial accounting method based on artificial intelligence, compared with the traditional method, does not require manual entry one by one, which not only avoids the problem of easy errors when manually entering documents, but also allows new and old employees to avoid the trouble caused by unfamiliarity with the reimbursement process and form differences, and can flexibly match the corresponding reimbursement template according to different reimbursement data to adapt to changes in corporate reimbursement policies and diversified business scenarios; then, by automatically filling in the target reimbursement form with data, and reviewing and reporting the filled target reimbursement form, the automation of the form data is realized, which greatly shortens the accounting cycle, further reduces the probability of manual errors, and reduces the difficulty of the overall financial accounting process.

[0035] In one embodiment, in step 102, the semantic parsing model is obtained based on the fusion of the first target model, the second target model and the third target model.

[0036] Wherein, step 1021, the training step of the first target model, includes: Acquire first sample data; the first sample data includes structured data in historical financial reimbursement data; input the first sample data into the first pre-training model to obtain a first prediction result output by the first pre-training model; obtain a first loss value based on the first prediction result based on the first loss function of the first pre-training model; adjust the first optimization function in the first pre-training model based on the first loss value; predict the prediction result of the first sample data based on the adjusted first pre-training model until the loss value output by the first loss function is less than or equal to the first preset value, the square of the difference between adjacent loss values ​​is less than 1, and the difference between at least three consecutive loss values ​​is greater than zero, thereby obtaining a first target model. The first preset value can be set to 0.01.

[0037] Specifically, for the first sample data, historical financial reimbursement records are extracted from the company's financial database, and these records are preprocessed to filter out the structured data part. Structured data has a clear format and fixed semantics, such as date (following a specific date format, such as "YYYY-MM-DD"), amount (precise numeric type, which may contain decimal places and currency units), employee number (usually a number or a specific encoding format), expense category code (predefined classification code, such as 1 for travel expenses, 2 for office supplies expenses, etc.). These data can be efficiently extracted and sorted through database query statements and data cleaning rules to form the first sample data set.

[0038] Furthermore, the first pre-trained model can be a multi-layer perceptron MLP, a deep neural network, etc. Taking the multi-layer perceptron MLP as an example, in a model based on MLP, the input layer receives the feature values ​​of structured data, which are weighted summed and nonlinearly transformed (using activation functions such as ReLU functions) through neurons in the hidden layer, and finally outputs the prediction results (such as 0 or 1 for a binary classification problem, indicating non-compliance or compliance) at the output layer.

[0039] Furthermore, the first loss function is: ; Where N represents the number of samples of the first pre-training data, and M represents the dimension of the semantic representation of professional vocabulary; Represents the value of the true label of the i-th sample in the first pre-training data (the correct semantic representation of professional vocabulary after standardization) on the j-th dimension, It represents the value of the first model's predicted output for the i-th sample in the j-th dimension. Represents the weight coefficient. The first half of the first loss function uses the cross entropy function to reflect the inaccuracy of the model prediction. When the probability distribution of the model prediction is closer to the probability distribution of the true label, the cross entropy value is smaller. The second half introduces constraints on the overall difference between the prediction result and the true label. To adjust the importance of this part in the loss function, there may be some cases in the financial reimbursement data, although there is a certain deviation in the classification probability (the cross entropy may not have changed much), but the overall prediction result is numerically different from the true label. For example, in a certain dimension of the predicted reimbursement amount, the sum of the model prediction value and the true value is quite different. This difference is reflected by the absolute value part (the second half), and its influence in the loss function is controlled by the weight coefficient to avoid the model only focusing on the classification probability and ignoring the accuracy of the overall value.

[0040] By substituting the prediction results into the formula, the first loss value is calculated, which reflects the degree of difference between the model prediction results and the true label. The larger the loss value, the less accurate the model prediction.

[0041] Furthermore, during the model training process, the adjustment of the first optimization function is repeated continuously until the three stop conditions are met to meet the training requirements. Among them, the loss value output by the first loss function is less than or equal to the first preset value, which indicates that the model has reached a certain accuracy requirement. Continuing training may not bring significant performance improvement, but may lead to overfitting. The square of the difference between adjacent loss values ​​is less than 1. This is to avoid excessive fluctuations or instability in the model during training. Let the loss value of the tth iteration be , the loss value of the t-1th iteration is , then the square of the difference between adjacent loss values ​​is . If this condition is not met, it means that the model may have abnormal changes under the current parameter update, and it is necessary to further adjust the optimization strategy or check whether there are problems with the data. For at least three consecutive loss values, the difference between the adjacent values ​​is greater than zero, which means that the model is still improving in several consecutive iterations, and the loss value is steadily decreasing, rather than falling into a local minimum or oscillating. In summary, setting multiple stop conditions can comprehensively consider the accuracy, stability and improvement trend of the model, avoid premature termination of training (underfitting) or overtraining (overfitting) of the model, and ensure that the first target model obtained has good generalization ability and performance. Through fine control of the training process, the most suitable model for processing structured financial reimbursement data can be obtained within limited training time and resources, thereby improving the applicability and reliability of the model.

[0042] Step 1022, a step of training a second target model, includes: Obtain second sample data; the second sample data includes semi-structured data in historical financial reimbursement data; input the second sample data into the second pre-training model to obtain a second prediction result output by the second pre-training model; obtain a second loss value according to the second prediction result based on the second loss function of the second pre-training model; adjust the second optimization function in the second pre-training model based on the second loss value; predict the prediction result of the second sample data based on the adjusted second pre-training model until the loss value output by the second loss function is less than or equal to the second preset value and the product of adjacent loss values ​​is greater than 0 and less than 1, thereby obtaining a second target model. The second preset value can be set to 0.05.

[0043] Specifically, semi-structured data is also obtained from historical financial reimbursement data, such as expense details lists (listing the name, amount and brief description of each expense in a specific format, but the format is not as strictly fixed as structured data), approval process records (including approvers, approval opinions and approval time, but there may be some non-standard expressions and format differences), some text data with simple tags or classifications, etc. By using text parsing technology and data extraction tools, these semi-structured data are converted into a format suitable for model input, such as converting expense details lists into tabular form, extracting approval process records into a list of key information pairs (approver-approval opinion), etc., to form the second sample data set.

[0044] Furthermore, in the stopping condition of the second target model, the loss value output by the second loss function is less than or equal to the second preset value, which indicates that the prediction error of the model on semi-structured data has been reduced to an acceptable range, and the model has achieved a certain accuracy requirement. The product of adjacent loss values ​​is greater than 0 and less than 1. This is to ensure the stability and continuous improvement of the model during the training process. It also prevents the model from oscillating or converging to the local minimum too early during the training process, ensuring that the model has good generalization ability and adaptability to new data while achieving a certain accuracy.

[0045] Furthermore, the second pre-trained model may use a recurrent neural network (RNN), a long short-term memory network LSTM, and a gated recurrent unit GRU. Taking the recurrent neural network RNN ​​as an example, for the expense list data, each expense item is input into the RNN model in sequence. The model gradually updates its internal state based on the information of the previous expense item and the characteristics of the current item, and finally outputs the prediction results of the entire expense details, such as whether the predicted expense is reasonable, whether there are any abnormalities, etc.

[0046] Furthermore, the second loss function is: ; in, represents the number of samples of the second pre-training data, represents the dimension of semantic representation of professional vocabulary after conversion, It is represented as the value of the true label of the professional vocabulary converted from the original spoken vocabulary of the i-th sample in the second pre-training data on the j-th dimension. represents the value of the predicted output of the i-th sample in the second pre-training data in the j-th dimension, is an indicator function, indicating whether the second model successfully integrates the spoken vocabulary into the existing semantic system for the i-th sample (if the integration is successful, ;otherwise ), represents the penalty coefficient for fusion failure. The second loss function uses different methods to measure the loss according to the success or failure of the model in fusion of spoken words, thereby guiding the model to better process semi-structured financial reimbursement data containing spoken words. hour, This part of the calculation loss helps the model to make the predicted professional vocabulary semantic representation when the fusion is successful. Close to the true semantic representation of professional vocabulary in probability distribution , which can prompt the model to select more accurate professional vocabulary representations when predicting and improve the accuracy of semantic understanding. When , a fixed penalty term is introduced , so that the model will receive an explicit penalty when fusion fails, prompting the model to work harder to correctly integrate spoken vocabulary, so as to avoid problems such as data misunderstanding and inaccurate classification in financial reimbursement data if spoken vocabulary cannot be correctly integrated into the semantic system.

[0047] Step 1023, a training step of the third target model, includes: Obtain third sample data; the third sample data includes unstructured data in historical financial reimbursement data; input the third sample data into the third pre-trained model to obtain a third prediction result output by the third pre-trained model; obtain a third loss value based on the third prediction result based on the third loss function of the third pre-trained model; adjust the third optimization function in the third pre-trained model based on the third loss value; predict the prediction result of the third sample data based on the adjusted third pre-trained model until the loss value output by the third loss function is less than or equal to the third preset value and the inverse of the product of the inter-loss values ​​is greater than the fourth preset value, thereby obtaining a third target model. Among them, the third preset value can be set to 0.1. The third pre-trained model can be a language model based on the Transformer architecture (such as GPT, BERT and its variants). The pre-processed third sample data is input into the model, and the model models the relationship between each word in the text and other words through its multi-head attention mechanism, thereby capturing the semantic information and contextual dependencies of the text.

[0048] Specifically, unstructured data is collected from historical financial reimbursement records, such as descriptions of expense purposes (for example, due to business negotiations with customers, banquets were held at XX restaurant to discuss cooperation details, and catering and transportation expenses were incurred during the period, in order to promote project cooperation), special circumstances (due to the temporary cancellation of flights, the cost of repurchasing tickets increased, and the airline has been communicated with and relevant certificates have been retained), and text content in attachments (such as descriptions of terms in contracts, discussion items in meeting minutes, etc.). Then, natural language processing technology is used, such as text cleaning tools to remove noise (such as redundant punctuation, special characters, garbled characters, etc.), lexical analysis tools for word segmentation (splitting text into individual words) and part-of-speech tagging (marking the part of speech of each word, such as nouns, verbs, adjectives, etc.), and then key entities (such as people, places, organizations, amounts, time, etc.) are extracted through entity recognition technology to form a structured text representation as the third sample data.

[0049] Furthermore, in the stopping condition of the third target model, the loss value output by the third loss function is less than or equal to the third preset value, which indicates that the prediction error of the model on unstructured data has been reduced to an acceptable range, the model has reached a certain accuracy requirement, and the inverse of the product of the phase loss values ​​is greater than the fourth preset value, which is to ensure the stability and continuous improvement of the model during the training process. The fourth preset value can be adjusted according to the actual training requirements, and can prevent the model from oscillating or converging to the local minimum too early during the training process, ensuring that the model has good generalization ability and adaptability to new data while achieving a certain accuracy.

[0050] The third preset loss function is: ; ; ; in, Mainly used to deal with the correction of fuzzy words, Mainly used to process the semantic analysis of special words, through the weight coefficient To balance the importance of these two parts in the overall loss function, it is possible to perform effective loss calculation and model optimization for different characteristics (fuzzy words and special words) in unstructured financial reimbursement data. If it is close to 1, grinding pays more attention to fuzzy word correction. Close to 0, more emphasis is placed on semantic analysis of special words. K represents the number of fuzzy words in the third pre-training data. is an indicator function, indicating whether the third model's prediction correction of the fuzzy words in the i-th sample is correct. (If correct, ;otherwise ), It represents the penalty coefficient when the fuzzy vocabulary is corrected. is the weight coefficient, It indicates that the prediction correction result of the third model for the fuzzy words in the i-th sample is converted into the correct probability value; L indicates the number of samples of the fourth pre-training data, P indicates the dimension of the special word semantic label, represents the value of the special vocabulary true semantic label of the i-th sample in the fourth pre-training data on the j-th dimension, It represents the value of the third model's prediction and parsing result of the special vocabulary of the i-th sample on the j-th dimension. Indicates the preset minimum value, which is used to prevent the denominator in logarithmic operations from being zero.

[0051] Furthermore, by combining the two parts of fuzzy word correction and special word semantic analysis, the loss function can comprehensively evaluate the performance of the model when processing unstructured financial reimbursement data. During the training process, the model should strive to accurately correct fuzzy words (through Partial optimization), and correctly parse the semantics of special words (through This can improve the model's ability to handle complex vocabulary in unstructured data, allowing the model to better adapt to the diversity and complexity of financial reimbursement data in actual business.

[0052] It should be noted that the first target model, the second target model, and the third target model are trained for structured, semi-structured, and unstructured data, respectively, and their respective stop conditions ensure that they achieve good performance when processing the corresponding types of data. When fused, this accuracy can be retained and integrated. For example, the high-precision judgment of the first target model on structured data (such as accurate identification of amounts, dates, etc.), the precise grasp of the logical relationship of semi-structured data by the second target model (such as expense details and approval processes), and the accurate understanding of the semantics of unstructured data by the third target model (such as description of expense usage) complement each other. When processing a complete set of financial reimbursement data, the fusion model can accurately parse from multiple angles, avoiding errors caused by the inadequacy of a single model in processing a certain type of data, thereby improving the overall accuracy. In addition, different stop conditions enable each model to maintain its own stability during the training process. When fused, this stability can prevent the occurrence of unstable training after the model fusion. The last three models are trained for different data types, and the stop conditions take into account the generalization ability of the model. After fusion, the fusion model can inherit this generalization advantage and better adapt to the financial reimbursement data of different enterprises.

[0053] In one embodiment, steps 1031-1035 are as follows: Step 1031, extract key features from the form filling intention data and the form template library to obtain a first form filling feature vector and a form feature vector respectively.

[0054] Specifically, for the form filling intention data, natural language processing technology and data mining algorithms are used to extract key features such as expense type (travel expenses, office supplies expenses, business entertainment expenses, etc.), amount range (through numerical recognition and range division, such as 0-1000 yuan, 1001-5000 yuan, etc.), business association (project name, department, customer involved, etc.), time information (time of expense occurrence, reimbursement time, etc.), and convert them into vector form, namely the first form filling feature vector.

[0055] Furthermore, for each template in the form template library, key features are extracted, such as the expense type applicable to the template (determined by the template name and the preset expense type label), the amount range preset in the template (the amount filling range set in the template according to the different expense standards stipulated by the enterprise's financial regulations), the business department or project type associated with the template (obtained from the template's instructions for use or preset associated information), the time format requirements of the template (such as a specific date filling format and time limit), etc. These features are converted into vector form to obtain a form feature vector.

[0056] By extracting key features and converting them into vector form, complex text data and template information can be quantified, facilitating efficient calculation and comparative analysis.

[0057] Step 1032: Obtain a historical reimbursement feature vector based on the first form filling feature vector and the form feature vector, and calculate the template historical usage frequency of each template in the form template library in the historical reimbursement records.

[0058] Specifically, first, historical reimbursement records related to the current form filling intention data and form template are collected from the company's historical reimbursement database. For each historical reimbursement record, the corresponding form filling feature vector and the used form template information (i.e., the first form filling feature vector and the form feature vector) are extracted, and the form filling feature vectors are aggregated and analyzed. For example, the K-Means clustering algorithm is used to cluster similar form filling feature vectors into different categories, each category representing a historical reimbursement feature. Then, the number of occurrences of each form template in each historical reimbursement feature category is counted, and divided by the total number of reimbursement records in the category to obtain the usage frequency of each template under different historical reimbursement features.

[0059] Further, in one embodiment, the goal of the K-Means clustering algorithm is to cluster the data set , ( It can be the filling feature vector of historical reimbursement records) divided into Cluster , so that the sum of squares within the cluster is minimized, and the iterative update process is as follows: First, randomly initialize Cluster Centers .

[0060] Repeat the following steps until convergence: For each data point , calculate its relationship with each cluster center Distance (such as Euclidean distance): ,in and Respectively expressed as and No. eigenvalues, n represents the dimension of the eigenvector, Assign to the cluster with the closest cluster center For each cluster , update its cluster center: ;in Represented as a cluster The data point data.

[0061] When the clustering converges, for each cluster , count where each form template is used The number of , then the template In cluster Frequency of use in for: ; LL represents the total number of templates in the form template library, and the usage frequency of each template in each cluster is combined to form the historical reimbursement feature vector , used for subsequent similarity weighted calculation.

[0062] Step 1033: weighting the original similarity based on the historical usage frequency of the template to obtain a weighted similarity.

[0063] Specifically, the cosine similarity is used to calculate the original similarity between the first form feature vector and the form feature vector. The calculation formula is: ,in, express and The dot product operation, Representation vector The module length, Representation vector The module length; For the first single feature vector = ( ), Characteristic vector of the form = ), and then the template historical usage frequency in the historical reimbursement feature vector calculated in the previous step is used as the weight to weight the original similarity. (R is the number of categories of historical reimbursement characteristics), then the weighted similarity ,in, It represents the original similarity calculated under the jth historical reimbursement characteristic (the original feature vector needs to be appropriately adjusted or recalculated according to different historical reimbursement characteristics). In this way, the template that better matches the historical reimbursement situation gets a larger weight in the similarity calculation, which increases its probability of being selected.

[0064] Step 1034: determine the clusters to be selected from the multiple cluster sets based on the weighted similarity; the clusters to be selected are the first preset clusters in the multiple cluster sets.

[0065] Specifically, all templates in the form template library are clustered according to clustering methods (such as hierarchical clustering, DBSCAN, etc.) to form multiple cluster sets. The basis for clustering can be the similarity of the key feature vectors of the templates (such as clustering templates with similar features such as expense type, amount range, and business association into one category). Afterwards, the templates in each cluster are sorted according to the weighted similarity calculated in the previous step, and the first preset clusters with higher weighted similarity are selected as the clusters to be selected. The first preset clusters can be set as needed. For example, the form template library is divided into 5 cluster sets using the hierarchical clustering method, and each cluster set contains several templates. For each cluster set, the weighted similarity between its internal templates and the form filling intention data is calculated, and they are sorted from high to low according to the similarity. If the first two clusters are preset as the clusters to be selected, the clusters containing the top two templates with the highest similarity are selected from each cluster set, and finally two clusters to be selected are obtained. The templates in the two clusters to be selected will be used as candidate sets for further screening of target reimbursement templates, which narrows the scope of subsequent searches and improves the efficiency of template selection.

[0066] Step 1035, obtaining the overall similarity between the second form filling feature vector in the cluster to be selected and each template, and determining the target reimbursement template based on the overall similarity; the second form filling feature vector is the vector after the dimension of the first form filling feature vector is reduced.

[0067] Specifically, the first feature vector is reduced in dimension using dimensionality reduction techniques (such as principal component analysis (PCA), singular value decomposition (SVD), etc.) to obtain the second feature vector. The purpose of dimensionality reduction is to reduce the dimension of the feature vector and reduce the computational complexity while retaining the main information, while removing some possible noise features. Projected into a low-dimensional subspace, the second single-character eigenvector is obtained , the specific process is as follows: First, calculate the covariance matrix of the first fill-in eigenvector , (where Q is the sample size. That is, the number of historical form filling data); Then perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues and the corresponding eigenvector , sort the eigenvectors in descending order of eigenvalue; Select the first w eigenvectors (w is the dimension after dimensionality reduction) to form the projection matrix ; The second fill-in feature vector Calculated by the following formula: .

[0068] After that, for each template in the cluster to be selected, calculate its overall similarity with the second form filling feature vector. The overall similarity here can take into account a variety of factors. In addition to the previous feature vector similarity (such as cosine similarity), some business rule similarity (such as the matching degree between the approval process preset in the template and the approval requirements implicit in the form filling intention), text semantic similarity (semantic analysis of the text description in the template and the expense use description in the form filling intention, and calculation of similarity), etc. For example, the weighted average method is used to calculate the overall similarity. ;in, represents the feature vector of the template, Represents the cosine similarity calculation function, represents the business rule similarity calculation function, Represents the text semantic similarity calculation function, , , Indicates the corresponding weight coefficient, which is set according to the actual business needs and data characteristics of the enterprise. Finally, the template with the highest overall similarity is selected as the target reimbursement template to complete the template matching and selection process, providing an accurate template basis for subsequent data filling operations.

[0069] Figure 2 Schematic diagram of a financial accounting device based on artificial intelligence provided by the present invention. Figure 2 As shown, the device includes: an acquisition unit 10, used to acquire target reimbursement data input by a target user; a model output unit 20, used to input the target reimbursement data into a semantic parsing model to obtain form filling intention data output by the semantic parsing model; the semantic parsing model is trained by sample reimbursement data and its corresponding form filling intention label results; a matching unit 30, used to match the form filling intention data with a form template library to obtain a target reimbursement template that matches the form filling intention data; a writing unit 40, used to perform data filling operations on the target reimbursement template based on the form filling intention data to obtain a target reimbursement form; an uploading unit 50: used to review the target reimbursement form and make financial reports based on the review confirmation results.

[0070] Figure 3 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 3As shown, the electronic device may include: a processor 310, a communication interface 320, a memory 330 and a communication bus 340, wherein the processor 310, the communication interface 320 and the memory 330 communicate with each other through the communication bus 340. The processor 310 may call the logic instructions in the memory 330 to execute a financial accounting method, device and electronic device based on artificial intelligence, the method comprising: obtaining target reimbursement data input by a target user; inputting the target reimbursement data into a semantic parsing model to obtain the form filling intention data output by the semantic parsing model; the semantic parsing model is trained by the sample reimbursement data and its corresponding form filling intention label results; matching the form filling intention data with the form template library to obtain the target reimbursement template matching the form filling intention data; performing data filling operation on the target reimbursement template based on the form filling intention data to obtain the target reimbursement form; reviewing the target reimbursement form and performing financial reporting based on the review confirmation result.

[0071] In addition, the logic instructions in the above-mentioned memory 330 can be implemented in the form of a software functional unit and can be stored in a readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a device (which can be a person, a server, or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0072] On the other hand, the present invention also provides a program product, which includes a program stored on a non-transitory readable storage medium, and the program includes program instructions. When the program instructions are executed, it can execute an artificial intelligence-based financial accounting method provided by the above-mentioned methods, and the method includes: obtaining target reimbursement data input by a target user; inputting the target reimbursement data into a semantic parsing model to obtain form filling intention data output by the semantic parsing model; the semantic parsing model is trained by sample reimbursement data and its corresponding form filling intention label results; matching the form filling intention data with a form template library to obtain a target reimbursement template matching the form filling intention data; performing data filling operations on the target reimbursement template based on the form filling intention data to obtain a target reimbursement form; reviewing the target reimbursement form, and making financial reports based on the review confirmation results.

[0073] On the other hand, the present invention also provides a non-transitory readable storage medium having a program stored thereon, which, when executed by a processor, is implemented to execute the above-mentioned artificial intelligence-based financial accounting method, the method comprising: obtaining target reimbursement data input by a target user; inputting the target reimbursement data into a semantic parsing model to obtain form filling intention data output by the semantic parsing model; the semantic parsing model is trained by sample reimbursement data and its corresponding form filling intention label results; matching the form filling intention data with a form template library to obtain a target reimbursement template matching the form filling intention data; performing data filling operations on the target reimbursement template based on the form filling intention data to obtain a target reimbursement form; reviewing the target reimbursement form, and making financial reports based on the review confirmation results.

[0074] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.

[0075] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, or of course by hardware. Based on this understanding, the above technical solution, in essence or in other words, the part that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a device (which can be a person, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0076] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A financial accounting method based on artificial intelligence, characterized in that: The following steps are involved: Obtain target reimbursement data input by target users; Inputting the target reimbursement data into a semantic parsing model to obtain form filling intention data output by the semantic parsing model; The semantic parsing model is trained by sample reimbursement data and its corresponding form filling intention label results; Matching the form filling intention data with a form template library to obtain a target reimbursement template that matches the form filling intention data; Performing data filling operation on the target reimbursement template based on the form filling intention data to obtain a target reimbursement form; The target reimbursement form is reviewed and financial reporting is performed based on the review and confirmation results.

2. The financial accounting method based on artificial intelligence according to claim 1, characterized in that: The semantic parsing model is obtained based on the fusion of the first target model, the second target model and the third target model.

3. The financial accounting method based on artificial intelligence according to claim 2 is characterized in that: The training step of the first target model comprises: Acquire first sample data; the first sample data includes structured data in historical financial reimbursement data; Input the first sample data into the first pre-trained model to obtain a first prediction result output by the first pre-trained model; obtain a first loss value according to the first prediction result based on the first loss function of the first pre-trained model; adjust the first optimization function in the first pre-trained model based on the first loss value; predict the prediction result of the first sample data based on the adjusted first pre-trained model until the loss value output by the first loss function is less than or equal to a first preset value, the square of the difference between adjacent loss values ​​is less than 1, and the difference between at least three consecutive loss values ​​is greater than zero, so as to obtain the first target model.

4. The financial accounting method based on artificial intelligence according to claim 2, characterized in that: The training step of the second target model comprises: Acquire second sample data; the second sample data includes semi-structured data in historical financial reimbursement data; Input the second sample data into the second pre-trained model to obtain a second prediction result output by the second pre-trained model; obtain a second loss value according to the second prediction result based on the second loss function of the second pre-trained model; adjust the second optimization function in the second pre-trained model based on the second loss value; predict the prediction result of the second sample data based on the adjusted second pre-trained model until the loss value output by the second loss function is less than or equal to the second preset value and the product of adjacent loss values ​​is greater than 0 and less than 1, so as to obtain the second target model.

5. The financial accounting method based on artificial intelligence according to claim 2, characterized in that: The training step of the third target model comprises: Acquire third sample data; the third sample data includes unstructured data in historical financial reimbursement data; Input the third sample data into the third pre-trained model to obtain a third prediction result output by the third pre-trained model; obtain a third loss value according to the third prediction result based on the third loss function of the third pre-trained model; adjust the third optimization function in the third pre-trained model based on the third loss value; predict the prediction result of the third sample data based on the adjusted third pre-trained model, until the loss value output by the third loss function is less than or equal to the third preset value and the inverse of the product of the phase loss values ​​is greater than the fourth preset value, so as to obtain the third target model.

6. The financial accounting method based on artificial intelligence according to claim 1, characterized in that: The step of performing a data filling operation on the target reimbursement template based on the form filling intention data to obtain a target reimbursement form includes: Extract key features from the form filling intention data and the form template library to obtain a first form filling feature vector and a form feature vector respectively; Acquire a historical reimbursement feature vector based on the first form filling feature vector and the form feature vector, and calculate the template historical usage frequency of each template in the form template library in the historical reimbursement record; Based on the historical usage frequency of the template, the original similarity is weighted to obtain a weighted similarity; Based on the weighted similarity, determining a cluster to be selected from a plurality of cluster sets; the cluster to be selected is a pre-set number of clusters in the plurality of cluster sets; The overall similarity between the second form filling feature vector and each template in the cluster to be selected is obtained, and based on the overall similarity, the target reimbursement template is determined; the second form filling feature vector is a vector after the dimension of the first form filling feature vector is reduced.

7. The financial accounting method based on artificial intelligence according to claim 6 is characterized in that: The calculation formula of the weighted similarity is: ； ； in, represents weighted similarity; express and The original similarity between represents the first fill-in feature vector; represents the form feature vector; express and The dot product operation of Representation vector The module length, Representation vector The module length; represents the original similarity calculated under the jth historical reimbursement feature, R represents the number of categories of historical reimbursement features, Represents the historical usage frequency of the template under the j-th historical reimbursement feature.

8. A financial accounting device based on artificial intelligence, characterized in that: Applicable to the artificial intelligence-based financial accounting method as claimed in any one of claims 1 to 7; The financial accounting device based on artificial intelligence includes: An acquisition unit, used for acquiring target reimbursement data input by a target user; A model output unit, used to input the target reimbursement data into a semantic parsing model to obtain the form filling intention data output by the semantic parsing model; the semantic parsing model is trained by the sample reimbursement data and its corresponding form filling intention label results; A matching unit, configured to match the form filling intention data with a form template library based on the form filling intention data to obtain a target reimbursement template that matches the form filling intention data; A writing unit, configured to perform a data filling operation on the target reimbursement template based on the form filling intention data to obtain a target reimbursement form; Uploading unit: used to review the target reimbursement form and make financial reports based on the review and confirmation results.

9. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the artificial intelligence-based financial accounting method as described in any one of claims 1 to 7 are implemented.

10. A non-transitory readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps of the artificial intelligence-based financial accounting method as described in any one of claims 1 to 7 are implemented.

Images