Financial big data analysis platform based on artificial intelligence

Abstract

The invention relates to the technical field of financial big data analysis, and discloses a financial big data analysis platform based on artificial intelligence, and the platform comprises the steps: obtaining the historical behavior characteristics of buyers, calculating value scores, and dividing the buyers into two classes according to the value scores according to a set score threshold value; according to a buying request initiated by a buyer, predicting a buying success probability of the buyer, and obtaining a transaction probability; the cooling interval is calculated according to the value score and the transaction probability of the buyer, and when the buyer initiates a buying request, the buyer delays the cooling interval and then sends the buying request to the modification sorting matchmaking module; receiving a buying request and a selling request, calculating the average number of error sequence pairs of the buying request, and comparing the average number with an error sequence number threshold value; the buying requests are modified according to the maximum modification times, the average number of the error sequence pairs of the buying requests is controlled to be smaller than the error sequence number threshold value, intelligent optimization and sorting of the buying requests before transaction matching are achieved, and the platform stability is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of financial big data analysis, and specifically to a financial big data analysis platform based on artificial intelligence. Background Art

[0002] Prioritizing and optimizing buy requests before trades are matched helps improve the overall efficiency of the matching system and reduce resource usage by low-quality requests. By pre-processing buy requests, the system can filter or adjust difficult-to-fill requests in advance without interfering with matching rules, thereby improving system response speed and transaction success rate.

[0003] Existing trade matching systems typically employ fixed sorting rules and lack the ability to dynamically modify buy requests. In high-concurrency environments, identifying and controlling misordering is difficult, making it impossible to determine which low-priority requests should not appear at the front of the queue. This leads to wasted matching resources and reduced transaction efficiency. Furthermore, existing technologies lack a "minimum modification" approach. Attempts to optimize request order often rely on wholesale reordering or forced discarding of original requests, which can disrupt the original submission order and impact transaction fairness and auditability. Furthermore, current platforms lack flexible mechanisms for adjusting request content (such as IP address, price, and quantity), and lack a method for evaluating misordering after request swaps. Consequently, quantitative control and refined optimization of the degree of misordering are impossible. During the request optimization process, a dynamic balance cannot be achieved between maximizing request order preservation and minimizing modification costs, resulting in limited sorting optimization effectiveness, constrained system throughput, and difficulty in stable implementation in real-world trading scenarios.

[0004] This solution proposes an artificial intelligence-based financial big data analysis platform that integrates user ratings, transaction probability prediction, dynamic flow control, and out-of-order control mechanisms to achieve intelligent optimization and sorting of buy requests before transaction matching. Summary of the Invention

[0005] The present invention provides a financial big data analysis platform based on artificial intelligence to promote the solution of the problems mentioned in the above background technology.

[0006] The present invention provides the following technical solution: a financial big data analysis platform based on artificial intelligence, comprising:

[0007] The user rating module is used to obtain the buyer's historical behavior characteristics to calculate the value score. Based on the set rating threshold, the buyer is divided into two categories according to the value score;

[0008] The users are divided into buyers and sellers, where buyers are users who initiate buy requests on the buy side of the platform, and sellers are users who initiate sell requests on the sell side of the platform;

[0009] The transaction probability prediction module is used to predict the probability of the buyer's purchase success based on the purchase request initiated by the buyer and obtain the transaction probability;

[0010] The real-time dynamic current limiting module is used to calculate the cooling-off interval based on the buyer's value score and transaction probability. When the buyer initiates a buy request, the buyer delays the cooling-off interval and then sends the buy request to the modified sorting matching module.

[0011] Modify the sorting and matching module to receive buy requests and sell requests, modify the buy requests to sort the buy requests, and match the buy requests and sell requests, specifically including:

[0012] Set the threshold for the number of wrong sequences and the maximum number of modifications;

[0013] Calculate the average number of misordered buy request pairs and compare the average number with the misordered number threshold.

[0014] If the average quantity is less than the out-of-sequence quantity threshold, the buy and sell requests are matched;

[0015] If the average quantity is greater than or equal to the out-of-sequence quantity threshold, the buy request is modified according to the maximum number of modifications to control the average quantity of the out-of-sequence pairs in which the buy request is located to be less than the out-of-sequence quantity threshold.

[0016] Optionally, the user scoring module is configured to obtain historical behavioral characteristics of the buyer to calculate a value score, and to classify the buyer into two categories according to the value score based on a set scoring threshold, including:

[0017] For the financial products analyzed in this article, they are sold by a single seller and purchased by multiple buyers concurrently;

[0018] The purchase request includes the purchase quantity and purchase price of the financial product;

[0019] The selling request includes the quantity and selling price of the financial product to be sold;

[0020] The historical behavior characteristics include historical transaction success rate, cumulative transaction volume and cumulative transaction amount;

[0021] Set weight coefficients for historical transaction success rate, cumulative transaction volume, and cumulative transaction amount respectively;

[0022] For any buyer, calculate the weighted average of the buyer's historical transaction success rate, cumulative transaction volume, cumulative transaction amount, and their respective weight coefficients;

[0023] The weighted average result is recorded as the buyer's value score;

[0024] Setting scoring thresholds;

[0025] If the value score is greater than or equal to the score threshold, the buyer is a high-value buyer;

[0026] If the value score is less than the score threshold, the buyer is a low-value buyer.

[0027] Optionally, the transaction probability prediction module is used to predict the probability of the buyer's purchase being successful based on the purchase request initiated by the buyer, including:

[0028] Get the buying prices in all buyers' buy requests, sort the buyers from high to low based on their buying prices, and then assign a ranking number to each buyer.

[0029] Calculate the price ranking factor of the buyer with order number k, specifically:

[0030] Get the number N of all buyers;

[0031] calculate , the result is used as the buyer's price ranking factor, where when k=1, the price ranking factor is equal to 1, and when k=N, the price ranking factor is equal to .

[0032] Optionally, the transaction probability prediction module is used to predict the probability of the buyer's purchase being successful based on the purchase request initiated by the buyer, and further includes:

[0033] Calculate the quantity factor of the buyer with order number k, specifically:

[0034] Get the remaining quantity of the financial product at the moment the buyer initiates the buy request ;

[0035] Get the quantity of financial products requested by the buyer ;

[0036] calculate , among which, when , the quantity factor is equal to 1, when , the quantity factor is less than 1;

[0037] Calculate the price ranking factor of the buyer with the ranking number k multiplied by the quantity factor, and use the result as the transaction probability of the buyer.

[0038] Optionally, the real-time dynamic flow limiting module is used to calculate the cooling-off interval based on the buyer's value score and transaction probability, including:

[0039] Set minimum cooling interval and maximum cooling interval ;

[0040] For any buyer:

[0041] Setting a high value ratio and a low value ratio, wherein the high value ratio is greater than the low value ratio, and the sum of the high value ratio and the low value ratio is equal to 1;

[0042] If the buyer is a high-value buyer, calculate the product of the high-value ratio and the buyer's transaction probability, and use the result as the buyer's priority. ;

[0043] If the buyer is a low-value buyer, calculate the product of the low-value ratio and the buyer's transaction probability, and use the result as the buyer's priority. ;

[0044] Calculate the buyer's final cooling-off interval. The specific calculation steps are:

[0045] ;

[0046] The moment when the buyer sends the buy request is recorded as the first moment;

[0047] The time of the cooling interval after the first time is calculated and recorded as the second time. The control sending end sends the buy request to the modified sort matching module at the second time.

[0048] Optionally, the modification and sorting matching module is configured to receive buy requests and sell requests, modify the buy requests to sort the buy requests, and match the buy requests with the sell requests, including:

[0049] Collect all buy requests sent by the buyer, arrange them in the modification sorting and matching module in the order of receipt, and number the buy requests in the order of arrangement;

[0050] Record any buy request as a mark request;

[0051] The buy request with a number lower than the marked request number is recorded as a preceding request;

[0052] Obtain the priorities of the marking request and the preceding request, and compare the priority of the marking request with the priorities of all preceding requests;

[0053] If the priority of the mark request is greater than the priority of the predecessor request, the mark request and the predecessor request are recorded as a misordered pair;

[0054] Obtain all buy requests with misordered pairs, obtain the number of misordered pairs in the buy requests, calculate the average, and record the result as the average number.

[0055] Optionally, the modification sorting and matching module is configured to receive buy requests and sell requests, modify the buy requests to sort the buy requests, and match the buy requests with the sell requests, further comprising:

[0056] When the average number is greater than or equal to the out-of-sequence number threshold:

[0057] Obtaining a buy request with the largest number of out-of-order pairs, and modifying the buy request;

[0058] Modifying the buy request includes modifying the buy-end IP address, the buy price, and the quantity of the financial product corresponding to the buy request;

[0059] Among them, one buying IP corresponds to a unique buyer;

[0060] S1. Obtain all preceding requests of the out-of-order pair containing the buy request, traverse each preceding request in turn, and record the preceding request being traversed as a candidate request;

[0061] S2. Swap the buying IP addresses of the buy request and the candidate request;

[0062] Swap the buy price and quantity of the financial instrument between the buy request and the candidate request;

[0063] S3. Reacquire all buy requests with misordered pairs, obtain the number of misordered pairs in the buy requests, calculate the average, and record it as a single replacement value;

[0064] S4. Obtain the single replacement value corresponding to each pre-request and buy request after they are swapped, and obtain the pre-request corresponding to the minimum single replacement value, which is recorded as the target request;

[0065] Use the target request as a pre-request to replace the final and buy requests.

[0066] Optionally, the modification sorting and matching module is configured to receive buy requests and sell requests, modify the buy requests to sort the buy requests, and match the buy requests with the sell requests, further comprising:

[0067] S5. Obtain a single replacement value corresponding to the target request, and compare the single replacement value with a misordered number threshold;

[0068] S6. If the single replacement value is less than the misordered quantity threshold, the buy request and the sell request are matched;

[0069] If the single replacement value is greater than or equal to the misorder threshold, the number of existing target requests is counted and recorded as the number of swaps;

[0070] S7. Compare the number of swaps and the maximum number of modifications;

[0071] If the number of swaps is greater than or equal to the maximum number of modifications, the buy request and the sell request are matched;

[0072] If the number of swaps is less than the maximum number of modifications, the buy request with the largest number of misordered pairs is obtained and S1-S7 is repeated.

[0073] The present invention has the following beneficial effects:

[0074] 1. This AI-based financial big data analysis platform can handle transactions between multiple sellers and multiple buyers. This solution only targets transactions between one seller and multiple buyers, analyzes core issues, simplifies the analysis process, and calculates a value score based on each buyer's historical behavioral characteristics. It then divides high-value buyers into low-value buyers based on the value score, thereby increasing the participation of high-value buyers and helping maintain transaction stability.

[0075] 2. This AI-based financial big data analysis platform calculates the probability of each buyer's successful purchase and sets a price ranking factor before actually matching the transaction. When the buyer enters the price and quantity of the financial product in the buy request, the seller sets the selling price and quantity in the sell request. The platform also sets a quantity factor, assuming that the higher the buy price, the higher the buyer's purchase success rate. Since the total number of sellers is limited and multiple buyers participate in the competition, the larger the purchase quantity requested, the more likely it is to exceed the allocable resources, and the higher the probability of being judged as untradeable by the platform. At the same price, small orders are more likely to be partially executed or matched first. Therefore, the smaller the purchase quantity, the higher the quantity factor. The transaction probability obtained by multiplying the price ranking factor by the quantity factor is matched with the matching mechanism of the trading system based on price priority and quantity consideration, thereby improving the accuracy of the prediction.

[0076] 3. This AI-based financial big data analysis platform sets high-value and low-value ratios for high-value buyers and low-value buyers, quantifies high-value buyers and low-value buyers, and then calculates the cooling-off interval based on the buyer's transaction probability. The lower the transaction probability of low-value users, the lower the priority, and the longer the cooling-off interval. The higher the transaction probability of high-value users, the higher the priority, and the shorter the cooling-off interval. The cooling-off interval controls transactions in the time dimension, increases the transaction probability of high-value buyers, improves the stability of platform transactions, and maintains a good trading environment.

[0077] 4. Although this AI-based financial big data analysis platform controls the sending time of buy requests on both the buying and selling sides to increase the probability of successful high-quality transactions, the time when the buy requests actually arrive at the platform's modified sorting and matching module is uncertain. Due to network instability or different distances between different buying sides and the modified sorting and matching module, the buy requests that ultimately arrive at the modified sorting and matching module may have low-value, low-transaction-probability buy requests before high-value, high-transaction-probability buy requests. In other words, low-priority buy requests may arrive before high-priority buy requests. Since actual matching is performed in the order of the buy requests, multiple high-quality buy requests cannot be successfully traded. The large number of low-priority buy requests causes platform instability. Therefore, by modifying the transaction requests, high-priority buy requests are brought forward to enhance the stability of platform transactions and reduce the time cost of unsuccessful transactions of low-priority buy requests that are ranked higher.

[0078] 5. This AI-based financial big data analysis platform limits the maximum number of modifications, as modifying a buy request requires the platform's time, memory, and processor resources. This is to prevent the total duration of the actual modification from being longer than the time wasted judging the transaction as unsuccessful before the modification, which would defeat the purpose of saving time and resources. The purpose of setting a misorder threshold is to allow for some misorders. If the restriction is that there are no misorders, the time it takes for the platform to sort all buy requests from high to low priority may be much longer than the duration of unsorted transactions. Sorting can also increase the possibility of successful transactions for high-priority buy requests. Therefore, setting a misorder threshold is used to balance sorting time and improve transaction stability, which can not only improve the stability of platform transactions but also improve the utilization of time resources.

[0079] 6. When sorting buy requests, the AI-based financial big data analysis platform first selects the buy request with the largest number of misordered pairs, and swaps the buy-end IP, buy price, and quantity of the buy request with the candidate request in the misordered pair where the buy request is located. Each time the swap is made, the overall number of misordered pairs is updated, and a single replacement value is obtained. The candidate request with the smallest single replacement value is finally swapped with the buy request, which completes the modification of a buy request. Based on the single replacement value and the maximum number of modifications, it is determined whether the buy request and the sell request can be matched. If matching is possible, it is directly matched. Otherwise, the buy request with the largest number of misordered pairs is selected for modification. The entire process is limited by the misordered number threshold and the maximum number of modifications, which not only improves the stability of platform transactions but also saves time resources.

[0080] 7. This AI-based financial big data analysis platform proactively optimizes the sorting structure before requests enter the matching stage by dynamically modifying buy requests (such as adjusting the buy price, quantity, and IP identifier). This platform quickly responds to system state changes, achieves near-real-time sorting reconstruction, and improves processing flexibility and sensitivity. It emphasizes a "minimum modification strategy," selecting only local modifications and swaps among the most severely misordered requests. By calculating the mean change in misordering after each replacement (a single replacement value), it achieves global improvements at minimal cost. While ensuring sorting optimization, it also maximizes the preservation of the original submission order and business continuity. By explicitly modeling the number of misordered pairs and the effect of replacements, each operation can be tracked and explained, ensuring the platform is stable, secure, auditable, and compliant with financial regulatory requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 Schematic diagram of the module of the present invention. DETAILED DESCRIPTION

[0082] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0083] Example 1, refer to Figure 1 , an AI-based financial big data analysis platform, including:

[0084] The user rating module is used to obtain the buyer's historical behavior characteristics to calculate the value score. Based on the set scoring threshold, the buyer is divided into two categories according to the value score, including:

[0085] For the financial products analyzed in this article, they are sold by a single seller and purchased by multiple buyers concurrently;

[0086] The purchase request includes the purchase quantity and purchase price of the financial product;

[0087] The selling request includes the quantity and selling price of the financial product to be sold;

[0088] The historical behavior characteristics include historical transaction success rate, cumulative transaction volume and cumulative transaction amount;

[0089] In this example, the transaction success rate is the ratio of the number of actual transactions among the buy requests submitted by buyers on the platform to the total number of submissions. For example, if user A submitted 50 buy requests in the past month, 40 of which were successfully completed (i.e., successfully matched with sellers), user A's historical transaction success rate is: 40 / 50 = 0.8. A high transaction success rate indicates that the user's bids are reasonable, their intentions are clear, their behavior is stable, and their credibility is high. A low success rate may indicate malicious attempts, false quotes, or risk control.

[0090] Cumulative trading volume refers to the total number of financial products that a buyer has successfully purchased within a certain period of time. It is the sum of the successful purchase quantities of each transaction. The larger the cumulative trading volume, the stronger the user's activity and capital operation ability, and the potential "platform big player".

[0091] The cumulative transaction amount represents the total amount actually paid by the buyer in historical successful transactions. It is equal to the sum of "purchase price × quantity" in all successful orders. It is used to comprehensively measure the user's financial strength, trading willingness and actual contribution.

[0092] Set weight coefficients for historical transaction success rate, cumulative transaction volume, and cumulative transaction amount respectively;

[0093] For any buyer, calculate the weighted average of the buyer's historical transaction success rate, cumulative transaction volume, cumulative transaction amount, and their respective weight coefficients;

[0094] The weighted average result is recorded as the buyer's value score;

[0095] Setting scoring thresholds;

[0096] If the value score is greater than or equal to the score threshold, the buyer is a high-value buyer;

[0097] If the value score is less than the score threshold, the buyer is a low-value buyer.

[0098] The transaction probability prediction module is used to predict the probability of a successful purchase based on the purchase request initiated by the buyer, including:

[0099] Get the buying prices in all buyers' buy requests, sort the buyers from high to low based on their buying prices, and then assign a ranking number to each buyer.

[0100] Calculate the price ranking factor of the buyer with order number k, specifically:

[0101] Get the number N of all buyers;

[0102] calculate , the result is used as the buyer's price ranking factor, where when k=1, the price ranking factor is equal to 1, and when k=N, the price ranking factor is equal to .

[0103] Calculate the quantity factor of the buyer with order number k, specifically:

[0104] Get the remaining quantity of the financial product at the moment the buyer initiates the buy request ;

[0105] Get the quantity of financial products requested by the buyer ;

[0106] calculate , among which, when , the quantity factor is equal to 1, when , the quantity factor is less than 1;

[0107] Calculate the price ranking factor of the buyer with the ranking number k multiplied by the quantity factor, and use the result as the transaction probability of the buyer.

[0108] A real-time dynamic flow control module is used to calculate the cooling-off interval based on the buyer's value score and transaction probability, including:

[0109] Set minimum cooling interval Seconds and maximum cooldown interval Second;

[0110] For any buyer:

[0111] Setting a high value ratio and a low value ratio, wherein the high value ratio is greater than the low value ratio, and the sum of the high value ratio and the low value ratio is equal to 1;

[0112] If the buyer is a high-value buyer, calculate the product of the high-value ratio and the buyer's transaction probability, and use the result as the buyer's priority. ;

[0113] If the buyer is a low-value buyer, calculate the product of the low-value ratio and the buyer's transaction probability, and use the result as the buyer's priority. ;

[0114] Calculate the buyer's final cooling-off interval. The specific calculation steps are:

[0115] ;

[0116] The moment when the buyer sends the buy request is recorded as the first moment;

[0117] The time of the cooling interval after the first time is calculated and recorded as the second time. The control sending end sends the buy request to the modified sort matching module at the second time.

[0118] In this embodiment, there is a seller who sells a total of 100 units of financial products at a price of RMB 10 per unit.

[0119] There are 5 buy requests on the buy side, which are numbered in the order they arrive at the modification sorting matching module, as follows:

[0120] The scoring threshold is 0.7;

[0121] The high value ratio is 0.7, and the low value ratio is 0.3. The purpose of setting the high value ratio and the low value ratio here is to prevent the cooling interval from being less than 0;

[0122] No. 1: Buyer A, purchase price 9.8, purchase quantity 10, transaction probability 0.5, value score 0.9, high-value buyer, priority = 0.5 × 0.7 = 0.35, cooldown interval = 10 - 0.35 × 9.5 = 6.675 seconds;

[0123] No. 2: Buyer B, purchase price 9.5, purchase quantity 20, transaction probability 0.4, value score 0.6, low-value buyer, priority = 0.4 × 0.3 = 0.12, cooldown interval = 10 - 0.12 × 9.5 = 8.86 seconds;

[0124] No. 3: Buyer C, purchase price 10.0, purchase quantity 10, transaction probability 0.9, value score 0.7, high-value buyer, priority = 0.9 × 0.7 = 0.63, cooldown interval = 10 - 0.63 × 9.5 = 4.015 seconds;

[0125] No. 4: Buyer D, purchase price 9.7, purchase quantity 15, transaction probability 0.6, value score 0.8, high-value buyer, priority = 0.6 × 0.7 = 0.42, cooldown interval = 10 - 0.42 × 9.5 = 6.01 seconds;

[0126] No. 5: Buyer E, purchase price 9.4, purchase quantity 30, transaction probability 0.3, value score 0.4, low-value buyer, priority = 0.3 × 0.3 = 0.09, cooldown interval = 10 - 0.09 × 9.5 = 9.145;

[0127] It has intelligent control capabilities to avoid resource occupation caused by a large influx of requests with low transaction probability.

[0128] Modify the sorting and matching module to receive buy requests and sell requests, modify the buy requests to sort the buy requests, and match the buy requests and sell requests, including:

[0129] Collect all buy requests sent by the buyer, arrange them in the modification sorting and matching module in the order of receipt, and number the buy requests in the order of arrangement;

[0130] Record any buy request as a mark request;

[0131] The buy request with a number lower than the marked request number is recorded as a preceding request;

[0132] Obtain the priorities of the marking request and the preceding request, and compare the priority of the marking request with the priorities of all preceding requests;

[0133] If the priority of the mark request is greater than the priority of the predecessor request, the mark request and the predecessor request are recorded as a misordered pair;

[0134] In this example, buyer C, numbered 3, has a higher priority than buyer A and buyer B, and the number of misordered pairs of buyer C is 2;

[0135] Buyer D, numbered 4, has a higher priority than Buyer A and Buyer B, and the number of misordered pairs that Buyer D has is 2;

[0136] The average number of misordered pairs is calculated to be 2, and the misordered number threshold is set to 1. If the average number is greater than the misordered number threshold, the buy request needs to be modified.

[0137] Obtaining a buy request with the largest number of out-of-order pairs, and modifying the buy request;

[0138] Modifying the buy request includes modifying the buy-end IP address, the buy price, and the quantity of the financial product corresponding to the buy request;

[0139] Among them, one buying IP corresponds to a unique buyer;

[0140] S1. Obtain all preceding requests of the out-of-order pair containing the buy request, traverse each preceding request in turn, and record the preceding request being traversed as a candidate request;

[0141] S2. Swap the buying IP addresses of the buy request and the candidate request;

[0142] Swap the buy price and quantity of the financial instrument between the buy request and the candidate request;

[0143] S3. Re-obtain the number of mis-ordered pairs of each buy request, calculate the average, and record it as the single replacement value;

[0144] S4. Obtain the single replacement value corresponding to each pre-request and buy request after they are swapped, and obtain the pre-request corresponding to the minimum single replacement value, which is recorded as the target request;

[0145] Use the target request as a pre-request to replace the final and buy requests.

[0146] S5. Obtain a single replacement value corresponding to the target request, and compare the single replacement value with a misordered number threshold;

[0147] S6. If the single replacement value is less than the misordered quantity threshold, the buy request and the sell request are matched;

[0148] If the single replacement value is greater than or equal to the misorder threshold, the number of existing target requests is counted and recorded as the number of swaps;

[0149] S7. Compare the number of swaps and the maximum number of modifications;

[0150] If the number of swaps is greater than or equal to the maximum number of modifications, the buy request and the sell request are matched;

[0151] If the number of swaps is less than the maximum number of modifications, the buy request with the largest number of misordered pairs is obtained and S1-S7 is repeated.

[0152] In this embodiment, the maximum number of modifications is set to 3;

[0153] First, modify Buyer D:

[0154] The first step is to swap the buying IP, buying price, and buying quantity of buyer D and buyer A. The new buyer order is numbered 4, 2, 3, 1, and 5, with corresponding priorities of 0.42, 0.12, 0.63, 0.35, and 0.09, respectively. The resulting single replacement value is 3 / 2=1.5. Among them, buyer C has two misordered pairs and buyer A has one misordered pair, and 1.5>1;

[0155] By swapping the buying IP, buying price and buying quantity of buyer D and buyer A, it can be considered that buyer D's buying request is moved to position numbered 1, and buyer A's buying request is moved to position numbered 4.

[0156] The second step is to swap the buying IP addresses, buying prices, and buying quantities of buyers D and B. The new buyers are numbered 1, 4, 3, 2, and 5, with corresponding priorities of 0.35, 0.42, 0.63, 0.12, and 0.09, respectively. The resulting single replacement value is 3 / 2=1.5. Buyer C has two misordered pairs, and buyer D has one misordered pair, so 1.5>1.

[0157] Among them, the buy requests of buyer D and buyer A are swapped. The order of the new buyers is 4, 2, 3, 1, and 5, and the corresponding priorities are 0.42, 0.12, 0.63, 0.35, and 0.09 respectively;

[0158] At this time, buyer C has the largest number of misordered pairs, 2. Modify buyer C:

[0159] The first step is to swap the buying IP addresses, buying prices, and buying quantities of buyers D and C. The new buyers are numbered 3, 2, 4, 1, and 5, with corresponding priorities of 0.63, 0.12, 0.42, 0.35, and 0.09, respectively. The resulting single replacement value is 1 / 2 = 0.5. Buyer D has one misordered pair, and buyer A has one misordered pair. 0.5 < 1, so the transaction between the five buyers and the single seller can be directly matched.

[0160] As a logical data structure before entering the matching phase, a buy request is essentially a pre-processing object controlled by the platform, rather than an immutable trade instruction committed to by the user. The platform dynamically modifies fields such as the price, quantity, and identification information of the request before matching, which is a scheduling behavior to optimize the request ranking. This modification occurs before the request is formally matched and does not affect the user's account assets or change their core trading intent. By treating buy requests as optimizable data inputs, the platform can adjust the overall ranking structure while ensuring fairness. This improves matching efficiency, reduces out-of-sequence interference, and maximizes the preservation of the original request order, enhancing the stability of the trading system and its resource scheduling capabilities. This achieves a technical balance between request optimization, user intent, and platform controllability, with a clear operational basis.

[0161] Under the premise of ensuring matching fairness, the original request order is retained to the greatest extent possible, avoiding meaningless large-scale reordering and reducing the system's computational complexity.

[0162] By introducing dual comparison logic between single replacement values ​​and the threshold for the number of misorders, and using the maximum number of modifications as the final tolerance limit, the replacement process can be effectively prevented from over-consuming resources or falling into an infinite loop. This prevents resource waste caused by ineffective fine-tuning while ensuring that matching tasks can still be completed in a timely manner even if replacements are ineffective, thus achieving an optimal balance between system efficiency, stability, and user experience.

[0163] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0164] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A financial big data analysis platform based on artificial intelligence, characterized by: include: The user rating module is used to obtain the buyer's historical behavior characteristics to calculate the value score. Based on the set rating threshold, the buyer is divided into two categories according to the value score; The users are divided into buyers and sellers, where buyers are users who initiate buy requests on the buy side of the platform, and sellers are users who initiate sell requests on the sell side of the platform; The transaction probability prediction module is used to predict the probability of the buyer's purchase success based on the purchase request initiated by the buyer and obtain the transaction probability; The real-time dynamic current limiting module is used to calculate the cooling-off interval based on the buyer's value score and transaction probability. When the buyer initiates a buy request, the buyer delays the cooling-off interval and then sends the buy request to the modified sorting matching module. Modify the sorting and matching module to receive buy requests and sell requests, modify the buy requests to sort the buy requests, and match the buy requests and sell requests, specifically including: Set the threshold for the number of wrong sequences and the maximum number of modifications; Calculate the average number of misordered buy request pairs and compare the average number with the misordered number threshold. If the average quantity is less than the out-of-sequence quantity threshold, the buy and sell requests are matched; If the average quantity is greater than or equal to the out-of-sequence quantity threshold, the buy request is modified according to the maximum number of modifications to control the average quantity of the out-of-sequence pairs in which the buy request is located to be less than the out-of-sequence quantity threshold.

2. The artificial intelligence-based financial big data analysis platform according to claim 1, characterized in that: The user scoring module is used to obtain the buyer's historical behavior characteristics to calculate the value score, and divide the buyers into two categories according to the value score based on the set scoring threshold, including: For the financial products analyzed in this article, they are sold by a single seller and purchased by multiple buyers concurrently; The purchase request includes the purchase quantity and purchase price of the financial product; The selling request includes the quantity and selling price of the financial product to be sold; The historical behavior characteristics include historical transaction success rate, cumulative transaction volume and cumulative transaction amount; Set weight coefficients for historical transaction success rate, cumulative transaction volume, and cumulative transaction amount respectively; For any buyer, calculate the weighted average of the buyer's historical transaction success rate, cumulative transaction volume, cumulative transaction amount, and their respective weight coefficients; The weighted average result is recorded as the buyer's value score; Setting scoring thresholds; If the value score is greater than or equal to the score threshold, the buyer is a high-value buyer; If the value score is less than the score threshold, the buyer is a low-value buyer.

3. The artificial intelligence-based financial big data analysis platform according to claim 1, characterized in that: The transaction probability prediction module is used to predict the probability of a successful purchase by the buyer based on the purchase request initiated by the buyer, including: Get the buying prices in all buyers' buy requests, sort the buyers from high to low based on their buying prices, and then assign a ranking number to each buyer. Calculate the price ranking factor of the buyer with order number k, specifically: Get the number N of all buyers; calculate , the result is used as the buyer's price ranking factor, where when k=1, the price ranking factor is equal to 1, and when k=N, the price ranking factor is equal to .

4. The artificial intelligence-based financial big data analysis platform according to claim 3 is characterized in that: The transaction probability prediction module is used to predict the probability of the buyer's purchase success based on the purchase request initiated by the buyer, and further includes: Calculate the quantity factor of the buyer with order number k, specifically: Get the remaining quantity of the financial product at the moment the buyer initiates the buy request ; Get the quantity of financial products requested by the buyer ; calculate , among which, when , the quantity factor is equal to 1, when , the quantity factor is less than 1; Calculate the price ranking factor of the buyer with the ranking number k multiplied by the quantity factor, and use the result as the transaction probability of the buyer.

5. The artificial intelligence-based financial big data analysis platform according to claim 1, characterized in that: The real-time dynamic flow limiting module is used to calculate the cooling-off interval based on the buyer's value score and transaction probability, including: Set minimum cooling interval and maximum cooling interval ; For any buyer: Setting a high value ratio and a low value ratio, wherein the high value ratio is greater than the low value ratio, and the sum of the high value ratio and the low value ratio is equal to 1; If the buyer is a high-value buyer, calculate the product of the high-value ratio and the buyer's transaction probability, and use the result as the buyer's priority. ; If the buyer is a low-value buyer, calculate the product of the low-value ratio and the buyer's transaction probability, and use the result as the buyer's priority. ; Calculate the buyer's final cooling-off interval. The specific calculation steps are: ； The moment when the buyer sends the buy request is recorded as the first moment; The time of the cooling interval after the first time is calculated and recorded as the second time. The control sending end sends the buy request to the modified sort matching module at the second time.

6. The artificial intelligence-based financial big data analysis platform according to claim 5, characterized in that: The modification sorting and matching module is used to receive buy requests and sell requests, modify the buy requests to sort the buy requests, and match the buy requests and sell requests, including: Collect all buy requests sent by the buyer, arrange them in the modification sorting and matching module in the order of receipt, and number the buy requests in the order of arrangement; Record any buy request as a mark request; The buy request with a number lower than the marked request number is recorded as a preceding request; Obtain the priorities of the marking request and the preceding request, and compare the priority of the marking request with the priorities of all preceding requests; If the priority of the mark request is greater than the priority of the predecessor request, the mark request and the predecessor request are recorded as a misordered pair; Obtain all buy requests with misordered pairs, obtain the number of misordered pairs in the buy requests, calculate the average, and record the result as the average number.

7. The artificial intelligence-based financial big data analysis platform according to claim 6, characterized in that: If the average quantity is greater than or equal to the out-of-sequence quantity threshold, modifying the buy request according to the maximum number of modifications to control the average quantity of the out-of-sequence pairs in which the buy request is located to be less than the out-of-sequence quantity threshold includes: When the average number is greater than or equal to the out-of-sequence number threshold: Obtaining a buy request with the largest number of out-of-order pairs, and modifying the buy request; Modifying the buy request includes modifying the buy-end IP address, the buy price, and the quantity of the financial product corresponding to the buy request; Among them, one buying IP corresponds to a unique buyer; S1. Obtain all preceding requests of the out-of-order pair containing the buy request, traverse each preceding request in turn, and record the preceding request being traversed as a candidate request; S2. Swap the buying IP addresses of the buy request and the candidate request; Swap the buy price and quantity of the financial instrument between the buy request and the candidate request; S3. Reacquire all buy requests with misordered pairs, obtain the number of misordered pairs in the buy requests, calculate the average, and record it as a single replacement value; S4. Obtain the single replacement value corresponding to each pre-request and buy request after they are swapped, and obtain the pre-request corresponding to the minimum single replacement value, which is recorded as the target request; Use the target request as a pre-request to replace the final and buy requests.

8. The artificial intelligence-based financial big data analysis platform according to claim 7, characterized in that: If the average number is greater than or equal to the out-of-sequence number threshold, modifying the buy request according to the maximum number of modifications to control the average number of out-of-sequence pairs in which the buy request is located to be less than the out-of-sequence number threshold, further comprising: S5. Obtain a single replacement value corresponding to the target request, and compare the single replacement value with a misordered number threshold; S6. If the single replacement value is less than the misordered quantity threshold, the buy request and the sell request are matched; If the single replacement value is greater than or equal to the misorder threshold, the number of existing target requests is counted and recorded as the number of swaps; S7. Compare the number of swaps and the maximum number of modifications; If the number of swaps is greater than or equal to the maximum number of modifications, the buy request and the sell request are matched; If the number of swaps is less than the maximum number of modifications, the buy request with the largest number of misordered pairs is obtained and S1-S7 is repeated.

Images