Automatic college financial reconciliation method based on RPA and pre-training language model

Abstract

The invention relates to the technical field of RPA and pre-training language models, in particular to an automatic college financial reconciliation method based on the RPA and the pre-training language model, and the method comprises the steps: simulating a unit to download a bank deposit journal and query online banking transaction details by the RPA; formatting the data to enable the formats of the data to be compared to be consistent; judging whether the bank deposit account statement and the bank deposit journal have multiple groups of combinations with opposite loan directions, the same amount and similar dates or not; for only one group of combinations with opposite borrowing and lending directions and the same amount, the RPA directly performs comparison according to whether the amount is consistent or not; for a plurality of groups of combinations with opposite borrowing and lending directions and the same amount, the RPA splices the'account name 'and'abstract' field information of the corresponding combination of the bank deposit account statement and inputs the spliced'account name 'and'abstract' field information of the corresponding combination of the bank deposit journal into a pre-training language model together with the'abstract 'field information of the corresponding combination of the bank deposit journal for semantic comparison; according to the invention, the problems of low efficiency and one-to-many, many-to-one and many-to-many account checking amounts in traditional account checking of colleges and universities can be solved, and the account checking workload of the colleges and universities is reduced through the RPA and the pre-training language model.

Description

Technical Field

[0001] The present invention relates to the technical field of RPA and pre-trained language models, and in particular to an automatic account reconciliation method for university finances based on RPA and pre-trained language models. Background Art

[0002] Account comparison and matching is a frequent task between bank and university cashiers to ensure the accuracy and consistency of account balances and transaction records. Reconciliation will be conducted at the end of each month, or it can be conducted regularly according to the agreement between the university and the bank, such as once a week or once a quarter. The cashier will check the bank deposit journal against the bank deposit statement item by item, check the items that have been received, find out the items that have not been received, and prepare a bank deposit balance reconciliation sheet. Due to the complex relationship between the bank deposit statement and the bank deposit journal (there are one-to-many, many-to-one, and many-to-many problems in the reconciliation amount) and many matching rules. In addition, some bill details can be as high as hundreds of items, so manual reconciliation operations have the disadvantages of low efficiency and easy errors.

[0003] RPA is the abbreviation of Robotic process automation, which can assist employees in handling a large number of rule-based, repetitive workflow tasks. The use of RPA has greatly improved the efficiency of reconciliation, but RPA is still powerless when there are one-to-many, many-to-one, or many-to-many reconciliation amounts.

[0004] A pre-trained language model is an artificial intelligence model that is pre-trained using large-scale text data. Its purpose is to enable the model to understand and generate natural language. Pre-trained language models usually use unsupervised learning methods to learn language structure, grammar, semantics and other information by learning from a large amount of text data. The learned knowledge can be reused in downstream tasks because the model has already learned the language function and only needs to fine-tune its representation with a small amount of data to perform specific tasks. Summary of the invention

[0005] The purpose of the present invention is to provide an automatic reconciliation method for university finance based on RPA and pre-trained language models. The technical problem solved by the present invention is: due to the complex relationship between bank deposit reconciliation statements and bank deposit journals (there are one-to-many, many-to-one, and many-to-many problems in the reconciliation amount) and numerous matching rules. In addition, some bill details can be as high as hundreds of items. Therefore, the manual reconciliation operation process has the disadvantages of low efficiency and easy errors.

[0006] The purpose of the present invention can be achieved through the following technical solutions: The automatic reconciliation method for university finance based on RPA and pre-trained language model includes the following steps: Whenever a processing cycle is reached or the RPA automatic reconciliation button is clicked in the university financial system, RPA automatically logs into the university financial system according to the preset account data, automatically configures the query conditions, and automatically downloads the bank deposit journal from the university financial system; whenever a processing cycle is reached or the RPA automatic reconciliation button is clicked in the university financial system, RPA logs into the online banking system according to the preset account data, automatically queries and downloads the online banking transaction details; RPA formats the data to make the format of the data to be compared consistent; determines whether there are multiple groups of combinations with opposite debit and credit directions but the same amount in the bank deposit statement and bank deposit journal; For combinations with only one group of opposite debit and credit directions and the same amount, control RPA to compare directly based on the amount; For multiple combinations with opposite debit and credit directions but the same amount, RPA inputs the "counterparty account name" and "summary" field information of the corresponding combination of the bank deposit statement and the "summary" field information of the corresponding combination of the bank deposit journal into the pre-trained language model for semantic comparison; RPA performs level-by-level matching based on the similarity of the output of the pre-trained language model; After all the matching is completed, RPA will check the successfully matched items against the accounts that have been settled, and check the unsuccessfully matched items against the accounts that have not been settled. The matching results will be uploaded to the university’s financial system and the balance reconciliation table will be automatically updated.

[0007] As a further solution of the present invention: a preset method for setting the reconciliation priority by anomaly ratio includes the following steps: Step 1: Obtain daily clearing data of bank deposit journal; Step 2: Make an abnormal judgment on the project content, generate the corresponding project abnormal signal, and obtain the corresponding abnormal ratio. According to the abnormal ratio output, obtain the journal abnormal performance ratio BYR; Among them, each transaction time, summary type, and transaction amount are compared and analyzed with the project content in the historical time, and the time abnormality ratio, type abnormality ratio, and amount abnormality ratio are obtained in turn, and marked as BT, BL, and BJ respectively; By formula , and calculate the journal abnormal performance ratio BYR; where a1, a2, and a3 are all weight ratio coefficients; Step 3: If the journal abnormal performance ratio BYR is greater than or equal to the journal abnormal performance ratio threshold, the bank deposit journal is marked as an abnormal bank deposit journal.

[0008] As a further solution of the present invention: the daily clearing data includes the daily item content in the bank deposit journal; the item content includes transaction time, summary type, and transaction amount.

[0009] As a further solution of the present invention: in step 2, the time anomaly ratio BT is obtained in the following manner: Obtain the transaction time in the daily accounting, compare the transaction time with the transaction time interval, and if the transaction time is within the transaction time interval, generate a normal transaction time signal; if the transaction time is not within the transaction time interval, generate an abnormal transaction time signal; Based on the abnormal trading time signal, the difference between the trading time and the endpoint value closest to the trading time interval is calculated to obtain the trading time difference, the difference between the trading time and the midpoint value of the trading time interval is calculated to obtain the trading time control value, and the trading time difference is divided by the trading time control value to obtain the time anomaly ratio BT.

[0010] As a further solution of the present invention: the type anomaly ratio BL is obtained in the following manner: Get the number of summary types in daily accounting, mark it as the daily account type number, compare the daily account type number with the daily account type number interval, if the daily account type number is within the daily account type number interval, generate a summary type normal signal, if the daily account type number is not within the daily account type number interval, generate a summary type abnormal signal; Based on the summary type anomaly signal, the difference between the daily account type frequency and the endpoint value closest to the daily account type frequency interval is calculated to obtain the daily account type frequency difference, the difference between the daily account type frequency and the midpoint value of the daily account type frequency is calculated to obtain the daily account type frequency control value, and the daily account type frequency difference is divided by the daily account type frequency control value to obtain the type anomaly ratio BL.

[0011] As a further solution of the present invention: the method of obtaining the abnormal amount ratio BJ is as follows: Obtain the transaction amount in the daily account, compare the transaction amount with the transaction amount range, and generate a normal transaction amount signal if the transaction amount is within the transaction amount range; otherwise, generate an abnormal transaction amount signal; Based on the abnormal transaction amount signal, the transaction amount is calculated with the endpoint value closest to the transaction amount range to obtain the transaction amount difference, the transaction amount is calculated with the midpoint value of the transaction amount range to obtain the transaction amount control value, and the transaction amount difference is divided by the transaction amount control value to obtain the amount abnormality ratio BJ.

[0012] As a further solution of the present invention: it is characterized in that if the journal abnormal performance ratio BYR is less than the journal abnormal performance ratio threshold, the bank deposit journal is marked as a normal bank deposit journal.

[0013] As a further solution of the present invention: the preset method for setting the reconciliation priority by the abnormality ratio further includes the following steps: Step 4: Obtain all normal bank deposit journals within the processing period, and then obtain the abnormal signal ratio and journal abnormal performance ratio of each normal bank deposit journal, add the abnormal signal ratio and the journal abnormal performance ratio together to obtain the abnormal mark value of each normal bank deposit journal; Calculate the difference between the abnormal mark values ​​of every two normal bank deposit journals within the processing cycle to obtain the journal association value; If the journal association value is greater than or equal to the journal association threshold, it means that the contents of the two normal bank deposit journal items are highly similar, and a correlation signal is generated.

[0014] As a further solution of the present invention: if the journal association value is less than the journal association threshold, a non-association signal is generated.

[0015] As a further solution of the present invention: the process of obtaining the abnormal signal ratio is as follows: Obtain the number of abnormal signals in normal bank deposit journals. Abnormal signals include abnormal transaction time signals, summary type abnormal signals, and transaction amount abnormal signals. Divide the number of abnormal signals by the number of project content comparison and analysis to obtain the abnormal signal ratio.

[0016] Beneficial effects of the present invention: The present invention is based on the RPA and pre-trained language model of the university financial automatic reconciliation method, which applies RPA and pre-trained models, and can solve the problems of low efficiency of traditional university reconciliation and one-to-many, many-to-one, and many-to-many reconciliation amounts, thereby reducing the workload of university reconciliation; Acquire daily clearing data of bank deposit journals; make abnormal judgment on item contents (item contents are transaction time, summary type and transaction amount of bank deposit journals), generate corresponding item abnormality signals, and obtain corresponding abnormality ratios, and obtain journal abnormality performance ratio BYR according to the abnormality ratio output; make judgment according to the journal abnormality performance ratio BYR and mark the bank deposit journal; by marking the bills with abnormal records, the present invention can reduce their priority and focus on them in the matching process of bank deposit reconciliation statements and bank deposit journals, thereby improving the efficiency of automatic reconciliation in colleges and universities; Based on the normal bank deposit journal, the normal bank deposit journal within the processing cycle is associated; the present invention makes an association judgment on the normal bank deposit journal, so that in the matching process of the debit and credit amounts of the bank deposit statement and the bank deposit journal, an association match can be performed, thereby further improving the efficiency of automatic reconciliation; especially in the many-to-many matching module, the associated bank deposit journal can be matched immediately through its association. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present invention will be further described below in conjunction with the accompanying drawings.

[0018] Figure 1 A flowchart of an automatic account reconciliation method for a university based on RPA and a pre-trained language model provided by an embodiment of the present invention; Figure 2 A sub-process diagram of an automatic account reconciliation method for a university based on RPA and a pre-trained language model provided in an embodiment of the present invention; Figure 3 A schematic diagram of a process for fine-tuning a BERT pre-training model provided in an embodiment of the present invention; Figure 4 A schematic diagram of the structure of an automatic account reconciliation device for colleges and universities based on RPA and a pre-trained language model provided by an embodiment of the present invention; Figure 5 A flowchart of an implementation method of a preset method for setting reconciliation priority by abnormal ratio is provided for an embodiment of the present invention; Figure 6 A flowchart of another implementation of a preset method for setting reconciliation priority by using an abnormality ratio is provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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 are within the scope of protection of the present invention.

[0020] Example 1: Please refer to Figure 1 As shown, the present invention is an automatic reconciliation method for university finance based on RPA and pre-trained language model, comprising the following steps: Step 101: Whenever a processing cycle is reached or the RPA automatic reconciliation button is clicked in the university financial system, RPA automatically logs into the university financial system according to preset account data, automatically configures query conditions, and automatically downloads the bank deposit journal from the university financial system.

[0021] In the disclosed embodiments of the present invention, the processing cycle may be one week, one month or one quarter, which is not specifically limited in the present disclosure.

[0022] In the disclosed embodiment of the present invention, clicking the RPA automatic reconciliation button in the university financial system connects the financial system to the RPA through the Java API. The university financial system can be placed on the same server as the RPA or on different servers.

[0023] In the disclosed embodiment of the present invention, taking a processing cycle of one week as an example, RPA automatically logs into the university financial system every week.

[0024] It should be noted that before RPA periodically logs into the university financial system, a login account and login password can be pre-set for RPA so that RPA can log in to the university financial system according to the login account and password.

[0025] It should be noted that after RPA logs into the university financial system, it automatically configures query conditions according to pre-set rules. The configured query conditions include but are not limited to time and bank name.

[0026] It should be noted that before RPA automatically downloads the bank deposit journal, you can pre-set the download address and file name for RPA. The file name is the bank abbreviation plus the query time and bank journal, such as "ICBC 2023-01-08 Bank Journal"; Step 102: Whenever the processing cycle is reached or the RPA automatic reconciliation button is clicked in the university financial system, RPA logs into the online banking system according to the preset account data, automatically queries the online banking transaction details (bank deposit statement) and downloads it.

[0027] It should be noted that before the RPA periodically logs into the online banking system, a login account and a login password may be pre-set for the RPA so that the RPA can log in to the online banking system according to the login account and the password.

[0028] It should be noted that after RPA logs into the university financial system, it automatically configures query conditions according to pre-set rules. The configured query conditions include but are not limited to time.

[0029] Before RPA automatically downloads bank transaction details, you can pre-set the download address and file name for RPA. The file name is the bank abbreviation plus the query time and bank statement, such as "ICBC 2023-01-08 Bank Statement"; In step 103, RPA formats the data to make the format of the data to be compared consistent; determines the matching of the debit and credit amounts of the bank deposit statement and the bank deposit journal, that is, determines whether there are multiple groups of combinations of the bank deposit statement and the bank deposit journal with opposite debit and credit directions and the same amount.

[0030] The amount is formatted in a unified currency format with two decimal places and no currency symbol. When there is a negative number in the debit and credit amounts, the negative number is converted to a positive number and the debit and credit directions are reversed. That is, when the debit amount is a negative number, the negative number is converted to a positive number and the data is exchanged with the credit amount.

[0031] Dates are normalized to the format yyyy-mm-dd, where yyyy represents year, mm represents month, and dd represents day.

[0032] Step 104, for only one combination with opposite debit and credit directions and the same amount, control the RPA to directly compare based on the amount.

[0033] Step 105: For multiple combinations with opposite debit and credit directions and the same amount, RPA concatenates the "counterparty account name" and "summary" field information of the corresponding combination of the bank deposit statement and the "summary" field information of the corresponding combination of the bank deposit journal and inputs them into the pre-trained language model for semantic comparison. RPA performs level-by-level matching based on the similarity of the output of the pre-trained language model.

[0034] Step 106, after all the matching is completed, RPA will check the successfully matched items that have been reached, and check the unmatched items that have not been reached. The matching results will be uploaded to the university financial system and the balance reconciliation table will be automatically updated.

[0035] It should be noted that if there is ambiguity in the matching results, the cashier can directly recheck the ambiguous matching results and the uncollected items in the university financial system and regenerate the balance reconciliation statement.

[0036] In order to accurately compare multiple groups of combinations with opposite debit and credit directions and the same amount, such as Figure 2 As shown, Figure 2 It is a sub-process diagram of an automatic reconciliation method for colleges and universities based on RPA and a pre-trained language model according to an embodiment of the present invention. Figure 4 The illustrated embodiment may include the following steps: Step 401: For multiple combinations with opposite debit and credit directions but the same amount, RPA concatenates the "counterparty account name" and "summary" field information of the corresponding combination of the bank deposit statement and inputs them together with the "summary" field information of the bank deposit journal into the pre-trained language model for semantic comparison.

[0037] In step 402, the RPA performs level-by-level matching based on the similarity of the output of the pre-trained language model.

[0038] In step 403, RPA will first compare the combinations with high similarity and check the items that have been reached.

[0039] It should be noted that: if the similarity is the same, the comparison will be based on the order of date. If the date of the bank deposit statement is too different from the date of the bank deposit journal, the one with the closer date will be matched first.

[0040] The "counterparty account name" of a bank deposit statement often only shows the payment object, such as "Zhang San" and "Shenzhen Rihao Technology Modern Service Industry Co., Ltd."; the "summary" of a bank deposit statement often only shows the payment category, such as "remittance refund", "transfer withdrawal", "online banking transfer", "loan repayment"; the "summary" of a bank deposit journal shows the specific reasons for borrowing and lending, such as "expenses for purchasing smart glasses, software application, etc.", "inviting enterprise experts to give theme lectures", "Zhang San's remittance refund", "Xiamen business trip training". Among them, "Zhang San" and "remittance refund" can be completely matched with "Zhang San's remittance refund"; "Enterprise" in "inviting enterprise experts to give theme lectures" can be matched with "Company" in "Shenzhen Rihao Technology Modern Service Industry Co., Ltd.". However, more combined field information does not have such obvious vocabulary similarity and correlation. For example, "expenses for purchasing smart glasses, software application, etc.", which are different types of expenses recorded together, cannot be matched with the "counterparty account name" and "summary" corresponding to the bank deposit statement; "Xiamen business trip training" cannot be matched with the "counterparty account name" and "summary" corresponding to the bank deposit statement.

[0041] What can be used is that the annual projects or activities of colleges and universities are generally fixed, such as "competitions", "three trips to the countryside", and "paper layout fees". The objects of this type of transaction are generally fixed. In addition, the transaction objects of some activities are generally people, such as "scholarships" and expenses of different types recorded together; the transaction objects of some activities are generally enterprises and institutions, such as "paying electricity bills" and "medical reimbursement payments"; the transaction objects of some activities are generally other colleges and universities, such as "expert review fees"; the transaction type of some activities is generally "batch deductions", such as "online banking fees"; the transaction type of some activities is generally "public payment"; the transaction type of some activities is generally "loan repayment".

[0042] Preferably, the pre-trained language model of the present invention is BERT, which stands for Bidirectional Encoder Representations from Transformers. The BERT model can well learn all the information that can be used by university reconciliation.

[0043] The BERT model is a multi-layer bidirectional model composed of Transformer encoders, with about 115 million parameters. It can be deployed on a single GPU card and is fast. In comparison, other large language models have tens or hundreds of billions of parameters. They are effective, but have high deployment costs, require GPUs with large video memory, and have slow prediction speeds.

[0044] In addition, other pre-trained language models, such as GPT, are unidirectional language models, which are naturally suitable for processing generative tasks, such as text generation, dialogue generation, and language translation, but this characteristic also determines that it cannot extract contextual information for sequence encoding. The BERT model uses the masked language model task for pre-training and adopts the bidirectional Transformer as the feature extraction structure. Although it loses the ability to directly generate text, it gains the ability of bidirectional encoding, which gives the model stronger text encoding performance. The BERT model can not only learn the similarity of vocabulary, but also the semantic similarity and vocabulary categories, and can be applied to tasks different from other pre-trained language models. BERT is mainly used for natural language understanding, and its specific applications are as follows: Question Answering System: BERT can be used in question answering systems to understand questions and generate answers.

[0045] Sentence similarity comparison: BERT can be used to compare the similarity between two sentences.

[0046] Text classification: BERT can be used to classify text.

[0047] Sentiment Analysis: BERT can be used to perform sentiment analysis on text.

[0048] Named Entity Recognition: BERT can be used to identify named entities in text.

[0049] The BERT model includes the following training process: The BERT training process is divided into two parts: pre-training and fine-tuning.

[0050] Pre-training is the basic part of the BERT model, which includes using a large amount of text to train the language model. During the pre-training stage, the BERT model will learn a lot of language knowledge, such as vocabulary, grammar, sentence structure, etc. Pre-training requires a large amount of data and a large amount of computer resources. Google officially provides multiple versions of pre-trained models in different languages ​​and sizes, which greatly adapts to downstream tasks and saves training costs. The present invention directly uses the Chinese version of the BERT basic pre-trained model provided by Google and fine-tunes it on this basis.

[0051] In order to accurately fine-tune the BERT pre-trained model, such as Figure 5 As shown, Figure 5 It is a schematic diagram of the BERT pre-training model fine-tuning process according to an embodiment of the present invention. Figure 5 The illustrated embodiment may include the following steps: Step 501, optionally, when data is limited or precision is insufficient, fine-tuning of the BERT model text classification can be added. The training data will contain a sentence and a label. The label is usually represented in a tokenized form. Each tag corresponds to a category. The sentence is the "summary" of the bank deposit journal, and the label category comes from the "counterparty account name" field information of the bank deposit statement, which is abstracted as "company", "university", "individual", "institution" or "other".

[0052] It should be noted that the labels of the text classification described in the present invention are not fixed, and are mainly based on the filling of the "counterparty account name" information field of the bank deposit statement. The activities, project categories or emphasis of different universities may be different, and the filling may also be different. The filling method of different banks may also be different.

[0053] Fine-tuning means that when the model needs to perform a task in a certain professional field, it needs to collect relevant professional field data, make small adjustments to the model, and update relevant parameters. When fine-tuning BERT, the input data needs to be divided into tokens, and each token needs to be converted into a corresponding token ID. In addition, the input data also needs to include a segment ID and an attention mask to indicate the boundaries of the sentence and control the attention of the BERT model. Finally, the input data will be combined into a tensor for input into the BERT model for fine-tuning. If only the fine-tuned model is used for prediction, and the model does not need to learn and modify the weights, then the amount of calculation can be solved by the CPU.

[0054] Step 502, optionally, when the data is limited or the accuracy is insufficient, continue to fine-tune the BERT model for text classification. The sentence is the "Summary" of the bank deposit journal, and the label category comes from the "Summary" field information of the bank deposit statement, which is abstracted as "Deposit / Withdrawal", "Deposit / Withdrawal at the Counter", "Public Payment", "Loan Repayment", "Refund", "Financial Payment", "Summary" or "Other".

[0055] Depending on the bank or university, the "summary" of the corresponding bank deposit statement may be different. At this time, the label must also be different, that is, the label is different according to the different "summaries" filled in by different banks or universities. If the "summaries" of different banks or universities are too different, multiple BERT models can be fine-tuned to correspond to the bank deposit statements and bank deposit journals of different banks or universities.

[0056] Step 503, when fine-tuning the BERT model for sentence similarity comparison, explicit labels are usually not required. The training data will contain a pair of sentences and a label indicating the degree of similarity between the two sentences. The label is usually a value between 0 and 1, indicating the similarity score of the two sentences. During the fine-tuning process of the BERT model, the model will learn how to predict their similarity scores based on the input sentence pairs.

[0057] In the disclosed embodiment of the present invention, the first sentence is the "Summary" field information of the bank deposit journal; the second sentence is the concatenation of the "Counterparty Account Name" field information and the "Summary" field information of the bank deposit statement.

[0058] It should be noted that in the present invention, fine-tuning of sentence similarity comparison is essential. Finally, the BERT model compares the bank deposit journal and the bank deposit statement in the form of sentence similarity comparison.

[0059] The present invention discloses an automatic reconciliation method for colleges and universities based on RPA and pre-trained language models. Whenever a processing cycle is reached or the RPA automatic reconciliation button is clicked in the college financial system, RPA automatically logs in to the college financial system according to preset account data, automatically configures query conditions, and automatically downloads bank deposit journals from the college financial system; automatically logs in to the online banking system according to preset account data, automatically queries online banking transaction details and downloads them. The "amount", "date", "counterparty account name" and "summary" field information are compared by controlling RPA and the pre-trained language model. The staff can be released from mechanical labor, saving a lot of manpower. At the same time, the problems of many-to-one, one-to-many, and many-to-many reconciliation amounts are solved.

[0060] In order to implement the above embodiments, the present invention also proposes an automatic reconciliation device for universities based on RPA and a pre-trained language model.

[0061] Figure 4 The present invention discloses a schematic diagram of a structure of an automatic account reconciliation device for colleges and universities based on RPA and a pre-trained language model according to an embodiment of the present invention. Figure 4 As shown, the automatic reconciliation device 600 for colleges and universities based on RPA and pre-trained language models may include: an information acquisition module 601, an information judgment module 602, a one-to-one matching module 603, a many-to-many matching module 604 and an upload module 605.

[0062] Among them, the login module 601 is used for RPA to log in to the university financial system and online banking system and download the bank deposit journal and online banking transaction details (bank deposit statement) whenever the processing cycle is reached or the RPA automatic reconciliation button is clicked in the university financial system.

[0063] The information determination module 602 is used to standardize the debit and credit amounts and dates; and to determine the matching of the debit and credit amounts in the bank deposit statement and the bank deposit journal.

[0064] The one-to-one matching module 603 is used to compare based on the amount when there is only one set of combinations with opposite debit and credit directions and the same amount.

[0065] The many-to-many matching module 604 is used for multiple groups of combinations with opposite lending directions and the same amount, including one-to-many, many-to-different, and many-to-many combinations, and is used to control RPA to use the pre-trained language model to match these combinations step by step.

[0066] The upload module 605 is used to check the account items that have been successfully matched; check the account items that have not been successfully matched; upload the matching results to the university financial system and automatically update the balance adjustment table.

[0067] Example 2: Please refer to Figure 4 As shown, based on the above-mentioned embodiment 1, in step 101, the reconciliation priority is set according to the preset abnormality ratio, and the bank deposit journal can be pre-marked as abnormal. By adding the abnormal mark, the bank deposit journal can be screened, which is convenient for improving the accuracy and efficiency of judging the matching of the debit and credit amounts of the bank deposit statement and the bank deposit journal; The preset method for setting the reconciliation priority by using the abnormality ratio includes the following steps: Step 1: Obtain daily clearing data of bank deposit journal; Among them, daily clearing data includes the daily items in the bank deposit journal; In some embodiments, obtaining daily item contents in a bank deposit journal, the item contents including transaction time, summary type, and transaction amount; Step 2: Make an abnormal judgment on the project content, generate the corresponding project abnormal signal, and obtain the corresponding abnormal ratio. According to the abnormal ratio output, obtain the journal abnormal performance ratio BYR; Among them, the abnormal ratio includes time abnormal ratio, type abnormal ratio, and amount abnormal ratio; In some embodiments, the transaction time, summary type, and transaction amount corresponding to each accounting data are obtained; Compare and analyze each transaction time, summary type, and transaction amount with the project content in the historical time, and obtain the time abnormality ratio, type abnormality ratio, and amount abnormality ratio in turn, and mark them as BT, BL, and BJ respectively; By formula , calculate the journal abnormal performance ratio BYR; among them, a1, a2, and a3 are all weight ratio coefficients, a1+a2+a3=1, a1 value is 0.24, a2 value is 0.48, and a3 value is 0.28. The values ​​of a1, a2, and a3 represent the influence of the time abnormality ratio, type abnormality ratio, and amount abnormality ratio on the journal abnormal performance ratio respectively; Specifically, the time anomaly ratio BT is obtained as follows: Obtain the transaction time in the daily accounting, compare the transaction time with the transaction time interval, and if the transaction time is within the transaction time interval, generate a normal transaction time signal; if the transaction time is not within the transaction time interval, generate an abnormal transaction time signal; Based on the abnormal trading time signal, the trading time is calculated as the difference between the endpoint value closest to the trading time interval to obtain the trading time difference, the trading time is calculated as the difference between the midpoint value of the trading time interval to obtain the trading time control value, and the trading time difference is divided by the trading time control value to obtain the time abnormality ratio BT; The transaction time interval is obtained in the following way: obtain the transaction start time and transaction end time recorded daily in the historical time, mark the minimum transaction start time as the minimum endpoint value of the transaction start time interval, and mark the maximum transaction start time as the maximum endpoint value of the transaction start time interval; The type exception ratio BL is obtained as follows: Get the number of summary types in daily accounting, mark it as the daily account type number, compare the daily account type number with the daily account type number interval, if the daily account type number is within the daily account type number interval, generate a summary type normal signal, if the daily account type number is not within the daily account type number interval, generate a summary type abnormal signal; Based on the summary type abnormal signal, the daily account type frequency and the endpoint value closest to the daily account type frequency interval are calculated to obtain the daily account type frequency difference, the daily account type frequency and the midpoint value of the daily account type frequency are calculated to obtain the daily account type frequency control value, and the daily account type frequency difference is divided by the daily account type frequency control value to obtain the type abnormality ratio BL; The daily account type frequency interval is obtained by the following method: obtaining the number of different summary types of daily accounting in the historical time, marking the minimum value of the number of different summary types in the historical time as the minimum endpoint value of the daily account type frequency interval, and marking the maximum value of the number of different summary types in the historical time as the maximum endpoint value of the daily account type frequency interval; The method of obtaining the abnormal amount ratio BJ is as follows: Obtain the transaction amount in the daily account, compare the transaction amount with the transaction amount range, and generate a normal transaction amount signal if the transaction amount is within the transaction amount range; otherwise, generate an abnormal transaction amount signal; Based on the abnormal transaction amount signal, the transaction amount is calculated with the endpoint value closest to the transaction amount interval to obtain the transaction amount difference, the transaction amount is calculated with the midpoint value of the transaction amount interval to obtain the transaction amount comparison value, and the transaction amount difference is divided by the transaction amount comparison value to obtain the abnormal amount ratio BJ; The transaction amount interval is obtained in the following way: obtain the transaction amount recorded daily in the historical period, mark the minimum transaction amount as the minimum endpoint value of the transaction amount interval, and mark the maximum transaction amount as the maximum endpoint value of the transaction amount interval; The transaction amount includes the income amount and the expenditure amount; Step 3: Mark the bank deposit journal based on the journal abnormal performance ratio BYR; In some embodiments, obtaining a journal abnormal performance ratio BYR, and comparing the journal abnormal performance ratio BYR with a journal abnormal performance ratio threshold; If the journal abnormal performance ratio BYR is greater than or equal to the journal abnormal performance ratio threshold, the bank deposit journal is marked as an abnormal bank deposit journal; If the journal abnormal performance ratio BYR is less than the journal abnormal performance ratio threshold, the bank deposit journal is marked as a normal bank deposit journal; It should be explained that: the abnormal bank deposit journal means that the bank deposit journal has anomalies in the item record compared with the past. The abnormal bank deposit journal is marked, so that in the subsequent matching process between the bank deposit statement and the bank deposit journal, its priority is reduced and it is given special attention; An abnormal bank deposit journal entry means that: the bank deposit journal entry has no abnormalities compared with the past and meets the account requirements; The technical solution of the embodiment of the present invention is as follows: acquiring the daily clearing data of the bank deposit journal; making an abnormal judgment on the project content, generating a corresponding project abnormality signal, and obtaining a corresponding abnormality ratio, and obtaining the journal abnormality performance ratio BYR according to the abnormality ratio output; making a judgment according to the journal abnormality performance ratio BYR, and marking the bank deposit journal; the present invention sets the reconciliation priority according to the preset abnormality ratio, can mark the bills with abnormal records, can reduce their priority and pay special attention to them in the matching process of the bank deposit reconciliation statement and the bank deposit journal, and improve the efficiency of automatic reconciliation in colleges and universities.

[0068] Example 3: Please refer to Figure 6 As shown, based on the above-mentioned embodiment 2, the preset method for setting the reconciliation priority by the abnormality ratio further includes the following steps: Step 4: Based on the normal bank deposit journal, associate the normal bank deposit journal within the processing period; In some embodiments, all normal bank deposit journals in a processing period are obtained, and then the abnormal signal ratio and the journal abnormal performance ratio of each normal bank deposit journal are obtained, and the abnormal signal ratio and the journal abnormal performance ratio are added and summed to obtain the abnormal mark value of each normal bank deposit journal; Calculate the difference between the abnormal mark values ​​of every two normal bank deposit journals within the processing cycle to obtain the journal association value; The journal association value is then compared with the journal association threshold; If the journal association value is greater than or equal to the journal association threshold, it means that the contents of the two normal bank deposit journal items are very similar, and a correlation signal is generated; If the journal association value is less than the journal association threshold, it means that the contents of the two normal bank deposit journal items are not very similar, and a non-association signal is generated; Specifically, the process of obtaining the abnormal signal ratio is as follows: Obtain the number of abnormal signals in normal bank deposit journals. Abnormal signals include abnormal transaction time signals, abnormal summary type signals, and abnormal transaction amount signals. Divide the number of abnormal signals by the number of comparative analysis items to obtain the abnormal signal ratio. It should be noted that when a bank deposit journal statement generates a correlation signal, in the process of matching the debit and credit amounts of the bank deposit statement and the bank deposit journal, correlation matching can be performed to further improve the efficiency of automatic reconciliation; The technical solution of the embodiment of the present invention is associating the normal bank deposit journals within the processing cycle based on the normal bank deposit journals; the present invention makes an association judgment on the normal bank deposit journals, so that in the matching process of the debit and credit amounts of the bank deposit statement and the bank deposit journal, the association matching can be performed, thereby further improving the efficiency of automatic reconciliation; especially in the many-to-many matching module, the matching work can be performed on the related bank deposit journals in succession through their association.

[0069] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0070] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. An automatic reconciliation method for university finance based on RPA and pre-trained language model, characterized by Using RPA to obtain reconciliation data and using the pre-trained language model to perform semantic comparison on the reconciliation data includes the following steps: Whenever a processing cycle is reached or the RPA automatic reconciliation button is clicked in the university financial system, RPA automatically logs into the university financial system according to the preset account data, automatically configures the query conditions, and automatically downloads the bank deposit journal from the university financial system; whenever a processing cycle is reached or the RPA automatic reconciliation button is clicked in the university financial system, RPA logs into the online banking system according to the preset account data, automatically queries and downloads the online banking transaction details; RPA formats the data to make the format of the data to be compared consistent; determines whether there are multiple groups of combinations with opposite debit and credit directions but the same amount in the bank deposit statement and bank deposit journal; For combinations with only one group of opposite debit and credit directions and the same amount, control RPA to compare directly based on the amount; For multiple combinations with opposite debit and credit directions but the same amount, RPA concatenates the "Other Account Name" and "Summary" field information of the corresponding combination of the bank deposit statement, and inputs them together with the "Summary" field information of the corresponding combination of the bank deposit journal into the pre-trained language model for semantic comparison; RPA performs level-by-level matching based on the similarity of the output of the pre-trained language model; After all the matching is completed, RPA will check the successfully matched items against the accounts that have been settled, and check the unsuccessfully matched items against the accounts that have not been settled. The matching results will be uploaded to the university’s financial system and the balance reconciliation table will be automatically updated.

2. The automatic reconciliation method for university finance based on RPA and pre-trained language model according to claim 1 is characterized in that: Setting reconciliation priority by exception ratio includes the following steps: Step 1: Obtain daily clearing data of bank deposit journal; Step 2: Make an abnormal judgment on the project content, generate the corresponding project abnormal signal, and obtain the corresponding abnormal ratio. According to the abnormal ratio output, obtain the journal abnormal performance ratio BYR; Among them, each transaction time, summary type, and transaction amount are compared and analyzed with the project content in the historical time, and the time abnormality ratio, type abnormality ratio, and amount abnormality ratio are obtained in turn, and marked as BT, BL, and BJ respectively; By formula , and calculate the journal abnormal performance ratio BYR; where a1, a2, and a3 are all weight ratio coefficients; Step 3: If the journal abnormal performance ratio BYR is greater than or equal to the journal abnormal performance ratio threshold, the bank deposit journal is marked as an abnormal bank deposit journal.

3. The automatic reconciliation method for university finance based on RPA and pre-trained language model according to claim 2 is characterized in that: Daily clearing data includes the daily item contents in the bank deposit journal; the item contents include transaction time, summary type, and transaction amount.

4. The automatic reconciliation method for university finance based on RPA and pre-trained language model according to claim 1 is characterized in that: In step 2, the time anomaly ratio BT is obtained as follows: Obtain the transaction time in the daily accounting, compare the transaction time with the transaction time interval, and if the transaction time is within the transaction time interval, generate a normal transaction time signal; if the transaction time is not within the transaction time interval, generate an abnormal transaction time signal; Based on the abnormal trading time signal, the difference between the trading time and the endpoint value closest to the trading time interval is calculated to obtain the trading time difference, the difference between the trading time and the midpoint value of the trading time interval is calculated to obtain the trading time control value, and the trading time difference is divided by the trading time control value to obtain the time anomaly ratio BT.

5. The automatic reconciliation method for university finance based on RPA and pre-trained language model according to claim 4 is characterized in that: The type exception ratio BL is obtained as follows: Get the number of summary types in daily accounting, mark it as the daily account type number, compare the daily account type number with the daily account type number interval, if the daily account type number is within the daily account type number interval, generate a summary type normal signal, if the daily account type number is not within the daily account type number interval, generate a summary type abnormal signal; Based on the summary type anomaly signal, the difference between the daily account type frequency and the endpoint value closest to the daily account type frequency interval is calculated to obtain the daily account type frequency difference, the difference between the daily account type frequency and the midpoint value of the daily account type frequency is calculated to obtain the daily account type frequency control value, and the daily account type frequency difference is divided by the daily account type frequency control value to obtain the type anomaly ratio BL.

6. The automatic reconciliation method for university finance based on RPA and pre-trained language model according to claim 5 is characterized in that: The method of obtaining the abnormal amount ratio BJ is as follows: Obtain the transaction amount in the daily account, compare the transaction amount with the transaction amount range, and generate a normal transaction amount signal if the transaction amount is within the transaction amount range; otherwise, generate an abnormal transaction amount signal; Based on the abnormal transaction amount signal, the transaction amount is calculated with the endpoint value closest to the transaction amount range to obtain the transaction amount difference, the transaction amount is calculated with the midpoint value of the transaction amount range to obtain the transaction amount control value, and the transaction amount difference is divided by the transaction amount control value to obtain the amount abnormality ratio BJ.

7. The automatic reconciliation method for university finance based on RPA and pre-trained language model according to claim 1 is characterized in that: If the journal abnormal performance ratio BYR is less than the journal abnormal performance ratio threshold, the bank deposit journal is marked as a normal bank deposit journal.

8. The automatic reconciliation method for university finance based on RPA and pre-trained language model according to claim 1 is characterized in that: The preset method for setting the reconciliation priority by using the exception ratio also includes the following steps: Step 4: Obtain all normal bank deposit journals within the processing period, and then obtain the abnormal signal ratio and journal abnormal performance ratio of each normal bank deposit journal, add the abnormal signal ratio and the journal abnormal performance ratio together to obtain the abnormal mark value of each normal bank deposit journal; Calculate the difference between the abnormal mark values ​​of every two normal bank deposit journals within the processing cycle to obtain the journal association value; If the journal association value is greater than or equal to the journal association threshold, it means that the contents of the two normal bank deposit journal items are highly similar, and a correlation signal is generated.

9. The automatic reconciliation method for university finance based on RPA and pre-trained language model according to claim 8 is characterized in that: If the journal association value is less than the journal association threshold, a no association signal is generated.

10. The automatic reconciliation method for university finance based on RPA and pre-trained language model according to claim 9 is characterized in that: The process of obtaining the abnormal signal ratio is as follows: Obtain the number of abnormal signals in normal bank deposit journals. Abnormal signals include abnormal transaction time signals, summary type abnormal signals, and transaction amount abnormal signals. Divide the number of abnormal signals by the number of project content comparison and analysis to obtain the abnormal signal ratio.

Images