A method, device, medium and equipment for matching a capital party

Abstract

Embodiments of the present application disclose a method, device, medium and equipment. The method comprises: obtaining at least one order data and obtaining a candidate capital party; extracting order parameters of the at least one order data and capital party parameters of the candidate capital party, inputting into a pre-constructed DQN matching model, and obtaining a capital party screening model; wherein the capital party screening model comprises a capital party value function; based on a current capital party value function of the capital party screening model, performing capital party matching on current order data of the at least one order data to obtain a current order data matching result, updating the capital party value function, and performing capital party matching on next order data until all order data matching is completed to obtain a capital party matching result of the at least one order data. The technical solution can maximize the overall profit of an order.

Description

TECHNICAL FIELD

[0001] The embodiment of the present application relates to the technical field of reinforcement learning, in particular to a capital party matching method, device, medium and equipment. BACKGROUND

[0002] In installment consumption, the intermediate platform usually has multiple cooperative capital parties. When a user places an order in the front end, the order is given to a certain group of capital parties for polling and auditing as an asset, until the asset is audited and paid by a certain capital party, that is, the financing matching process. The same order will have different profit amounts in different capital parties. Therefore, how to maximize the profit becomes a problem to be solved.

[0003] The current order allocation can only consider the profit of the current order and cannot consider the profit of the overall order. Although the capital party with the maximum profit is selected in the allocation of a single order, it may lead to a decrease in the overall order profit. Therefore, a capital party matching method is needed to maximize the overall order profit. SUMMARY

[0004] The embodiment of the present application provides a capital party matching method, device, medium and equipment, which can obtain a capital party screening model through a DQN matching model pre-constructed by order parameter and capital party parameter input values, and perform value matching through the current capital party value function of the capital party screening model, thereby realizing the maximization of the overall order profit.

[0005] In the first aspect, the embodiment of the present application provides a capital party matching method, which comprises:

[0006] Obtaining at least one order data and obtaining a candidate capital party;

[0007] Extracting order parameters of the at least one order data and capital party parameters of the candidate capital party, and inputting them into a pre-constructed DQN matching model to obtain a capital party screening model; wherein the capital party screening model comprises a capital party value function;

[0008] Based on the current capital party value function of the capital party screening model, performing capital party matching on the current order data of the at least one order data to obtain a current order data matching result, updating the capital party value function, and performing capital party matching on the next order data until all order data matching is completed to obtain a capital party matching result of the at least one order data.

[0009] In the embodiment of the present application, the construction process of the capital party screening model comprises:

[0010] Constructing an initial DQN matching model of the capital party matching scene;

[0011] Based on the sample order data and the sample candidate capital party, sample order parameters and sample candidate capital party parameters are extracted as training environment parameters, and the training environment parameters are input into the initial DQN matching model to obtain a trained DQN matching model.

[0012] Based on the calculation result of the capital party value function of the trained DQN matching model, a capital party ranking index corresponding to the sample order data is obtained.

[0013] The sample order data matching target capital party is determined according to the capital party ranking index, and the target parameter of the matching target capital party is taken as the maximum value, so as to train the capital party value function in the initial DQN matching model to obtain a training result of the DQN matching model.

[0014] In the embodiment of the present application, based on the current capital party value function of the capital party screening model, the current order data of the at least one order data is matched with the capital party to obtain a current order data matching result, including:

[0015] The capital party value of each capital party is calculated through the current capital party value function.

[0016] Each of the capital parties is sorted in descending order of capital party value to obtain the ranking index.

[0017] The current order data matching result is obtained according to the ranking index.

[0018] In the embodiment of the present application, the capital party value function includes:

[0019]

[0020] Wherein, The capital party value function, i is the current capital party, s is the current environment state, s' is the next environment state, A is the current action, R(s,A) is the reward obtained by taking action A in the current environment state s, π is the current policy, and Vπ(s') is the estimated value of the environment state s' when taking the policy π.

[0021] In the embodiment of the present application, the capital party value function is updated by the following formula:

[0022]

[0023] Wherein, The capital party value function updated at time t, s is the current environment state, S t The environment state at time t, i is the current capital party, a is the learning rate, R(s,i) is the reward of the current capital party i in the current environment state s, γ is the reward discount coefficient, and At+1 an action taken at a next time instant for time instant t, Q π (S t , A t+1 ) is a value of taking strategy π and action A t under environment state S t+1 .

[0024] In the embodiment of the present application, the value Q t (S t+1 , A π ) of taking strategy π and action A t under environment state S t+1 is obtained by the following formula:

[0025]

[0026] wherein P(i|S t+1 , A t+1 ) is a state transition probability of principal i in environment S t+1 at time t+1 when action A t+1 is taken, is the updated principal value function at time t+1.

[0027] In the embodiment of the present application, the current order data matching result is obtained according to the ranking index, comprising:

[0028] determining a probability of matching to each principal according to the ranking index and the audit pass rate of each principal;

[0029] determining a target principal with the highest probability, and taking the target principal as the current order data matching result.

[0030] In a second aspect, the embodiment of the present application provides a principal matching device, comprising:

[0031] an order data and candidate principal obtaining unit configured to obtain at least one order data and obtain candidate principals;

[0032] a principal screening model obtaining unit configured to extract order parameters of the at least one order data and principal parameters of the candidate principals, and input the order parameters and the principal parameters to a pre-constructed DQN matching model to obtain a principal screening model; wherein the principal screening model comprises a principal value function;

[0033] The capital party matching result acquisition unit is configured to perform capital party matching on the current order data of the at least one order data based on the current capital party value function of the capital party screening model, obtain a current order data matching result, and update the capital party value function, and perform capital party matching on the next order data until all order data matching is completed, and obtain the capital party matching result of the at least one order data.

[0034] In a third aspect, an embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the capital party matching method according to the embodiments of the present application.

[0035] In a fourth aspect, an embodiment of the present application provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable by the processor, and the processor implements the capital party matching method according to the embodiments of the present application when executing the computer program.

[0036] The technical scheme provided by the embodiments of the present application obtains order data and candidate capital parties, extracts order parameters of the order data and candidate parameters of the candidate capital parties, and inputs the order parameters and the candidate parameters into a pre-constructed DQN matching model to obtain a capital party screening model. The capital party screening model performs capital party matching on the order data through a capital party price function, obtains a current order data matching result, and updates a capital party value function each time a matching result is obtained. The capital party value function is continuously updated to match capital parties for order data, so that the overall order situation is considered instead of only considering the profit of the current order. The capital parties are matched in consideration of the overall order situation, and the overall profit is maximized. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 is a flowchart of the capital party matching method provided by the first embodiment of the present application;

[0038] Figure 2 is a flowchart of the capital party matching method provided by the first embodiment of the present application;

[0039] Figure 3 is a flowchart of the capital party screening model construction method provided by the third embodiment of the present application;

[0040] Figure 4 is a structural block diagram of the capital party matching device provided by the fourth embodiment of the present application;

[0041] Figure 5 is a structural block diagram of another capital party matching device provided by the fourth embodiment of the present application;

[0042] Figure 6Fig. 6 is a structural schematic diagram of an electronic device provided by Embodiment Six of the present application. DETAILED DESCRIPTION

[0043] The present application will be further described below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are intended to be merely illustrative of the present application and not in limitation thereof. It should also be noted that, for the purpose of description, only the parts related to the present application are shown in the accompanying drawings rather than all the parts.

[0044] Before the exemplary embodiments are described in greater detail, it is noted that some exemplary embodiments are described as processes or methods depicted as flow diagrams. Although the processes are described in a particular sequential order, many of the steps can be performed in parallel, concurrently or in any order. In addition, the order of the steps can be re-arranged. The processes can be terminated when their operations are completed, but can also have additional steps not included in the figure, which can also be performed after the operations of the processes are completed. The processes can correspond in part to method steps for implementing the methods.

[0045] Embodiment One

[0046] Figure 1 Fig. 1 is a flowchart of a method for matching investors provided by Embodiment One of the present application. As shown in Fig. 1, the method for matching investors includes: Figure 1

[0047] S110, obtaining at least one order data and obtaining candidate investors.

[0048] The order data is related information of at least one order that needs to be matched with investors, and usually includes basic information of the order and related data under a certain investor. The basic information can include order id and lending time. The at least one order data is sorted according to the lending time and / or order id. The candidate investors are investors selected from a plurality of investors for matching. Since taking all investors as matching targets will result in too many matching targets, which may affect the matching speed, several candidate investors are selected from all investors as matching objects.

[0049] S120, extracting order parameters of the at least one order data and investor parameters of the candidate investors, and inputting the order parameters and the investor parameters into a pre-constructed DQN matching model to obtain an investor screening model; wherein the investor screening model includes an investor value function.

[0050] ​The order parameters can include: period, amount, interest rate, lifecycle, and estimated profit amount, etc. The capital party parameters are related information of the candidate capital party, which can be various party limitation conditions, including whether it is a profit sharing capital party, a risk threshold, a monthly position quota, a capital party ID, etc. The DQN (Deep Q-Learning) matching model is a reinforcement learning model, which is modeled through the following parts:

[0051] Agent: an execution individual, which can operate the execution individual to make different choices.

[0052] Environment: an environment, which includes multiple states.

[0053] Action: when the agent makes an action, the environment will change, i.e., the state will change.

[0054] Reward: when the state changes, the environment will give a certain reward (the reward can be positive or negative).

[0055] In the embodiments of the present application, the environment is created according to the order parameters and the capital party parameters. The order parameters and the capital party parameters are input into the DQN matching model and the environment is constructed, i.e., the capital party screening model is obtained. In the capital party screening model, the capital party value function runs in the created environment, and the value of the capital party, i.e., the reward, is obtained.

[0056] In S130, based on the current capital party value function of the capital party screening model, the current order data of the at least one order data is matched with the capital party, a current order data matching result is obtained, the capital party value function is updated, the capital party matching of the next order data is performed, and the capital party matching of all order data is completed to obtain the capital party matching result of the at least one order data.

[0057] The current capital party value function represents the function used when matching the current order data with the capital party. Since the capital party parameters and the order parameters will change after the capital party matching, in order to maximize the overall profit, the capital party value function needs to be updated so that the capital party value function belongs to the updated state when matching the current order data with the capital party each time. The order data can be sorted by order ID and lending time, and the order data is matched with the capital party in sequence until the capital party matching result of each order data is obtained.

[0058] The technical scheme provided in the embodiments of the present application comprises the following steps: obtaining order data and candidate capital parties, extracting order parameters of the order data and candidate parameters of the candidate capital parties and inputting the order parameters and the candidate parameters into a pre-constructed DQN matching model to obtain a capital party screening model, performing capital party matching on the order data by the capital party price function of the capital party screening model to obtain a current order data matching result, updating the capital party value function each time the matching result is obtained, and matching the capital party for the order data through the continuously updated capital party value function until all order data matching is completed to obtain a capital party matching result of each order data. The scheme provided in the present application matches the capital party for the order data through continuously updating the capital party value function, so that the overall order situation is considered instead of only considering the profit of the current order, the capital party is matched in consideration of the overall order situation, and the maximization of the overall profit is realized.

[0059] On the basis of the above-mentioned embodiments, Figure 2 The flowchart of the capital party matching method provided in Embodiment One of the present application is based on the above-mentioned embodiments, and optionally, performing capital party matching on the current order data of the at least one order data based on the current capital party value function of the capital party screening model to obtain a current order data matching result, comprising: calculating the capital party value of each capital party through the current capital party value function; sorting each capital party in descending order of capital party value to obtain the sorting index; and obtaining the current order data matching result according to the sorting index.

[0060] As Figure 2 shown, the specific implementation steps of the present embodiment are as follows:

[0061] S210, calculating the capital party value of each capital party through the current capital party value function.

[0062] The capital party value function is a state-based capital party value function obtained on the basis of the value function. The value function comprises a state value function and a state-action value function, both of which are used to calculate the size of the cumulative reward. In DQN, the value function is usually a neural network, which inputs a state vector or an action state pair vector and outputs a scalar, that is, the DQN matching model is a regression prediction model. The problem of the application of the above two kinds of DQN is that the space action of the order is too large, because in asset matching, the next capital party of the order is prioritized, so the sorting of n capital parties has n! kinds, and in the case of too many capital parties corresponding to the order, all possible results cannot be facilitated. Therefore, the capital party value function is a state capital party value function obtained on the basis of the state-action value function, which is used to calculate the capital party value of each capital party.

[0063] S220, sorting each capital party in descending order of capital party value to obtain the sorting index.

[0064] The sorting index is a sorting result of the capital party according to the capital party value, and when the capital party value of each capital party is calculated, the Top-k method is used to sort the capital party value from high to low to obtain the sorting index.

[0065] S230, obtaining the current order data matching result according to the sorting index.

[0066] On the basis of the above embodiments, the current order data matching result is obtained according to the sorting index, including: determining the probability of matching to each capital party according to the sorting index and the audit pass rate of each capital party; determining the target capital party with the highest probability as the current order data matching result.

[0067] The current order data matching result is the final capital party matched to the current order data, which is obtained by combining the sorting index with other parameters. For example, the order_state.shape of the sorting index is (5, 11), that is, there are 5 candidate capital parties, each of which has 11 capital party characteristics, and the action is {3, 1, 2, 4, 0}, that is, the third capital party has the highest priority, and the capital party in the 0th row has the lowest priority. After inputting the sorting index, the probability of matching to each capital party is calculated according to the sorting and the audit pass rate, and a capital party is selected according to the order of the probability. When calculating the probability, normalization processing is performed, so that a capital party will be finally matched.

[0068] After obtaining the order data matching result, interaction with the environment is needed, including the following steps:

[0069] 1. Update the market and capital party state.

[0070] 2. Calculate the reward.

[0071] 3. Update the order state.

[0072] 4. Return the overall state, reward, matched capital party and whether the episode is over.

[0073] Embodiment two

[0074] The embodiment two of the application provides a method for constructing and updating a capital party value function, and the embodiment is optimized on the basis of the above-mentioned embodiments, and the specific optimization is that the capital party value function includes:

[0075]

[0076] Among them, For the funder value function, i is the current funder, s is the current environment state, s' is the next environment state, A is the current action taken, R(s, A) is the reward obtained by taking action A in the current environment state s, π is the current policy taken, and Vπ(s') is the estimated value of the environment state s' when taking policy π.

[0077] Wherein, the funder value function is a state funder value function, and construction of the function needs to disassemble and learn the state funder value, and needs to consider two basic assumptions:

[0078] 1. Single choice (SC), in the funder ranking given again, only one funder will be finally matched.

[0079] 2. Reward / transition dependence on selection (RTDS), the reward value order R(s, a) only depends on the finally matched funder i, and the similar state transition probability P(s'|s,A) also only depends on the finally matched funder i.

[0080] The above two assumptions are consistent in the funder matching scene, for SC, the order will finally only match one funder, but for RTDS, the current reward function does not involve the difference between funders.

[0081] According to SC, we can get:

[0082]

[0083]

[0084] According to RTDS, we can get:

[0085] P(s,A,i)=R(s,A',i)=R(s,i);

[0086] P(s',A,i)=R(s'|s,A',i)=R(s'|s,i);

[0087] The process of disassembling the state action value is as follows:

[0088]

[0089]

[0090]

[0091]

[0092] Through the above derivation process, the funder value function can be obtained.

[0093] On the basis of each of the above embodiments, optionally, the capital value function is updated by the following formula:

[0094]

[0095] wherein, is the capital value function updated at t time, s is the current environment state, S t is the environment state at t time, i is the current capital, a is the learning rate, R(s,i) is the reward of the current capital i under the current environment state s, γ is the reward discount coefficient, A t+1 is the action taken at the next time of t time, Q π (S t ,A t+1 ) is the value of taking the policy π and the action A t under the environment state S t+1 .

[0096] Specifically, the capital value function is updated by the time sequence scoring TD method. The advantage of TD is that the final output result needs to be obtained, and each step is updated once, so that the convergence is faster.

[0097] On the basis of each of the above embodiments, optionally, the value Q t (S t+1 ,A π ) of taking the policy π and the action A t under the environment state S t+1 is obtained by the following formula:

[0098]

[0099] wherein, P(i|S t+1 ,A t+1 ) is the state transition probability of the capital i in the environment S t+1 when the action A t+1 is taken at t+1 time, is the capital value function updated at t+1 time.

[0100] wherein, the selection of A t+1 may be based on Sarsa or Q-learning. In the framework of DQN, whether Sarsa or Q-learning will select A π in the manner of Q t+1 (S t+1 ,A π )=maxaQ t+1 (S t+1 ,a).This way will cause Q to be overestimated, because in the training process, the Q value of an action is abnormally overestimated once, which can easily lead to a chain reaction. Therefore, in Double DQN, the Q-function for selecting actions is not the same. Usually, the maximum value A is selected based on Q, and then A is put into another Q-function Q' of Double DQN to calculate the target value Q'(S, A).

[0101] Embodiment three

[0102] Figure 3 The flowchart of the method for constructing the investor screening model provided in Embodiment Three of the present application is optimized based on the above-mentioned embodiments. The specific optimization is that the construction process of the investor screening model comprises:

[0103] constructing an initial DQN matching model of the investor matching scenario; extracting sample order parameters and sample candidate investor parameters as training environment parameters based on sample order data and sample candidate investors, and inputting the training environment parameters into the initial DQN matching model to obtain a trained DQN matching model; obtaining an investor ranking index corresponding to the sample order data based on the calculation result of the investor value function of the trained DQN matching model; determining the matching target investor of the sample order data according to the investor ranking index, so that the target parameter of the matching target investor takes the maximum value, to train the investor value function in the initial DQN matching model and obtain the training result of the DQN matching model.

[0104] As shown in Figure 3 , the method of the present embodiment specifically comprises the following steps:

[0105] S310, constructing an initial DQN matching model of the investor matching scenario.

[0106] S320, extracting sample order parameters and sample candidate investor parameters as training environment parameters based on sample order data and sample candidate investors, and inputting the training environment parameters into the initial DQN matching model to obtain a trained DQN matching model.

[0107] Among them, the sample order data and the sample candidate investor are the data corresponding to the order and the investor as the training sample. Similarly, the sample order parameters include the sample order id and the lending time, and the sample investor parameters include various restriction conditions, including whether it is a profit-sharing investor, a risk threshold, a monthly position limit, an investor id, etc. By inputting the sample order parameters and the sample candidate investor parameters as the training environment parameters into the DQN matching model, a trained DQN matching model for training is obtained.

[0108] S330, obtaining a capital ranking index corresponding to the sample order data based on a calculation result of a capital value function of the training DQN matching model.

[0109] The capital ranking index represents the ranking result of the candidate sample capital according to the corresponding capital value. For each sample order data, the capital value of each candidate sample capital is calculated according to the capital value function, and the Top-k method is used to sort the capital value from high to low to obtain the capital ranking index.

[0110] In addition, before running through the capital value function, the environment needs to be initialized after being created, so that the capital value function runs in the initial environment state. Because when the capital is matched, some capital may have been matched to other order data in reality, resulting in changes in the capital parameters of the capital. Through environment initialization, each capital is matched with the initial capital parameter, and the matching is more accurate.

[0111] The environment initialization can be realized by performing the following operations:

[0112] 1. Set the current step number to 0.

[0113] 2. Obtain the order id according to the order parameter, and configure the features of the current order data.

[0114] 3. Initialize the market state features, which can include: total gmv, profit amount, profit rate, distribution gmv, distribution proportion, distance to target, and whether to reach the distribution proportion target.

[0115] 4. Initialize the capital parameters of each capital, which can include: loan amount, remaining position, loan proportion, etc.

[0116] 5. From all state parameters, intercept the capital parameters of the candidate capital.

[0117] 6. Summarize the order parameters, market state and candidate capital state, and return the summarized overall state.

[0118] S340, determining the target capital of the sample order data according to the capital ranking index, taking the target parameter of the matching target capital to the maximum value, training the capital value function in the initial DQN matching model, and obtaining the training result of the DQN matching model.

[0119] Since the capital ranking index represents the ranking of the candidate sample capital from high to low, the first capital in the capital ranking index is selected as the target capital matched to the sample order data, and the capital value function is trained to update the capital value function according to the matched state.

[0120] On the basis of the above-mentioned embodiments, since the order data is huge in the capital matching scenario, it will lead to an episode being too long. Multiple orders can be sorted as an action at the same time, and the iteration method needs to be modified: 100 orders each time, so that 100 orders are actually contained in a state. When using the batch order training method, the forward propagation of 100 orders needs to be run every time the fitting target is constructed, and then the average value is taken. If the batch_size is set to 128 each time, 128*100*20=256000 order forward propagations need to be predicted, which is too low in efficiency, greatly limiting the training speed and batch_size. Therefore, we use a sampling method to construct the target value, and sample 10% from 100 orders each time to construct the fitting target, which greatly improves the training efficiency.

[0121] Embodiment Four

[0122] Figure 4 A structural block diagram of a capital matching device provided in Embodiment Four of the present application is provided. The device can execute the capital matching method provided in any embodiment of the present application, and has the corresponding function modules and beneficial effects of the execution method. As shown in Figure 4 the device can include:

[0123] An order data and candidate capital obtaining unit 410 is configured to obtain at least one order data and obtain a candidate capital.

[0124] A capital screening model obtaining unit 420 is configured to extract order parameters of the at least one order data and capital parameters of the candidate capital, and input them into a pre-constructed DQN matching model to obtain a capital screening model. The capital screening model includes a capital value function.

[0125] A capital matching result obtaining unit 430 is configured to perform capital matching on a current order data of the at least one order data based on a current capital value function of the capital screening model, obtain a current order data matching result, update the capital value function, and perform capital matching on a next order data until all order data matching is completed, and obtain a capital matching result of the at least one order data.

[0126] The above-mentioned product can execute the capital matching method provided in the embodiments of the present application, and has the corresponding function modules and beneficial effects of the execution method.

[0127] In the embodiments of the present application, as shown in Figure 5 the device further includes a capital screening model construction unit 510. The capital screening model construction unit 510 is configured to perform:

[0128] An initial DQN matching model of a construction capital matching scenario is constructed;

[0129] Based on sample order data and sample candidate capital, sample order parameters and sample candidate capital parameters are extracted as training environment parameters, and the training environment parameters are input into the initial DQN matching model to obtain a trained DQN matching model;

[0130] Based on the calculation result of the capital value function of the trained DQN matching model, a capital ranking index corresponding to the sample order data is obtained;

[0131] The sample order data matching target capital is determined according to the capital ranking index, the target parameter of the matching target capital is taken as the maximum value, the capital value function in the initial DQN matching model is trained, and a training result of the DQN matching model is obtained.

[0132] In the embodiment of the present application, the capital matching result acquisition unit 430 is configured to perform:

[0133] The capital value of each capital is calculated through the current capital value function;

[0134] Each of the capitals is ranked in descending order of capital value to obtain the ranking index;

[0135] The current order data matching result is obtained according to the ranking index.

[0136] In the embodiment of the present application, the capital value function comprises:

[0137]

[0138] Wherein, The capital value function, i is the current capital, s is the current environment state, s' is the next environment state, A is the current action, R(s,A) is the reward obtained by taking action A in the current environment state s, π is the current policy, and Vπ(s') is the estimated value of the environment state s' when taking the policy π.

[0139] In the embodiment of the present application, the capital value function is updated by the following formula:

[0140]

[0141] Wherein, The capital value function updated at time t, s is the current environment state, S t The environment state at time t, i is the current capital, a is the learning rate, R(s,i) is the reward of the current capital i in the current environment state s, γ is the reward discount coefficient, and At+1 Q is an action taken at the next time of t, Q π (S t ,A t+1 ) is the value of taking strategy π and action A t under environment state S t+1 .

[0142] In the embodiment of the application, the value Q t (S t+1 ,A π ) of taking strategy π and action A t under environment state S t+1 is obtained by the following formula:

[0143]

[0144] Wherein, P(i|S t+1 ,A t+1 ) is the state transition probability of the party i in the environment S t+1 at t+1 time, using action A t+1 , is the updated party value function at t+1.

[0145] In the embodiment of the application, the party screening model construction unit 510, when executing the current order data matching result obtained according to the sorting index, specifically executes:

[0146] According to the sorting index and the audit pass rate of each party, determine the probability of matching to each party;

[0147] Determine the target party with the highest probability, and take the target party as the current order data matching result.

[0148] Embodiment five

[0149] The embodiment five of the application provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to realize a party matching method provided by all the embodiments of the application:

[0150] Obtain at least one order data, and obtain a candidate party;

[0151] Extract order parameters of the at least one order data and party parameters of the candidate party, and input the order parameters and the party parameters into a pre-constructed DQN matching model to obtain a party screening model; wherein the party screening model comprises a party value function;

[0152] Based on the current investor value function of the investor screening model, the current order data of the at least one order data is matched with the investors to obtain a current order data matching result, and the investor value function is updated, and the investor matching of the next order data is performed until all order data matching is completed, and the investor matching result of the at least one order data is obtained.

[0153] Any combination of one or more computer readable medium can be employed. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. A computer readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium include an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In this document, the computer readable storage medium can be any tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device.

[0154] A computer readable signal medium can include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated signal can take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium can be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport program for use by or in connection with an instruction execution system, apparatus, or device.

[0155] Program code embodied on a computer readable medium can be transmitted using any appropriate medium, including but not limited to wireless, wire line, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0156] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0157] Embodiment six

[0158] An electronic device is provided in embodiment six of the present application. Figure 6 is a structural schematic diagram of an electronic device provided in embodiment six of the present application. As shown in the figure, Figure 6 An electronic device 600 is provided in the embodiment, which includes one or more processors 620 and a storage device 610 configured to store one or more programs. When the one or more programs are executed by the one or more processors 620, the one or more processors 620 implement a method for matching a capital party provided in the embodiments of the present application, which includes the following steps.

[0159] obtaining at least one order data and a candidate capital party;

[0160] extracting order parameters of the at least one order data and capital party parameters of the candidate capital party, inputting the order parameters and the capital party parameters into a pre-constructed DQN matching model, and obtaining a capital party screening model; wherein the capital party screening model includes a capital party value function;

[0161] based on a current capital party value function of the capital party screening model, performing capital party matching on current order data of the at least one order data, obtaining a current order data matching result, updating the capital party value function, and performing capital party matching on next order data until all order data matching is completed, and obtaining a capital party matching result of the at least one order data.

[0162] Of course, those skilled in the art can understand that the processor 620 also implements the technical solutions of the capital party matching method provided in any embodiment of the present application.

[0163] Figure 6 The electronic device 600 shown is merely an example and should not impose any limitation on the functions and use range of the embodiments of the present application.

[0164] As Figure 6 shown, the electronic device 600 includes a processor 620, a storage device 610, an input device 630, and an output device 640; the number of processors 620 in the electronic device can be one or more, Figure 6 of which an example is one processor 620; the processor 620, the storage device 610, the input device 630, and the output device 640 in the electronic device can be connected through a bus or other means, Figure 6 of which an example is connected through the bus 650.

[0165] The storage device 610, as a kind of computer readable storage medium, can be used to store software programs, computer executable programs and module units, such as the program instructions corresponding to the method for matching the capital party in the embodiments of the present application.

[0166] The storage device 610 can mainly include a storage program area and a storage data area, wherein the storage program area can store an operating system, at least one application required by a function; the storage data area can store data created according to the use of the terminal and the like. In addition, the storage device 610 can include a high-speed random access memory, and can also include a non-volatile memory, for example, at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some examples, the storage device 610 can further include a memory disposed remotely with respect to the processor 620, and these remote memories can be connected through a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0167] The input device 630 can be used to receive input digital, character information or voice information, and generate key signal input related to user settings and function control of the electronic device. The output device 640 can include a display screen, a speaker and other electronic devices.

[0168] The capital party matching device, medium and electronic device provided in the above embodiments can execute the method for matching the capital party provided in any embodiment of the present application, and have the corresponding function modules and beneficial effects of executing the method. Technical details not described in detail in the above embodiments can be referred to the method for matching the capital party provided in any embodiment of the present application.

[0169] Note that the above merely describes preferred embodiments of the present application and the principles of the technology applied. Those skilled in the art will understand that the present application is not limited to the specific embodiments described herein, and that various obvious changes, modifications and substitutions can be made without departing from the scope of the present application. Therefore, although the present application has been described in detail through the above embodiments, the present application is not limited to the above embodiments, and can include more other equivalent embodiments without departing from the concept of the present application, and the scope of the present application is determined by the scope of the claims.

Claims

1. A method for matching investors, characterized in that, The method includes: Obtain at least one order data item, and obtain candidate investors; The order parameters of the at least one order data and the investor parameters of the candidate investor are extracted and input into a pre-built DQN matching model to obtain an investor screening model; wherein, the investor screening model includes an investor value function; Based on the current investor value function of the investor screening model, investor matching is performed on the current order data of the at least one order data to obtain the current order data matching result, and the investor value function is updated to perform investor matching on the next order data until all order data is matched to obtain the investor matching result of at least one order data. The current investor value function based on the investor screening model is used to perform investor matching on the current order data of the at least one order data to obtain the current order data matching result, including: The investor value for each investor is calculated using the current investor value function. Each investor is sorted in descending order of investor value to obtain the sorting index; The current order data matching result is obtained based on the sorting index; The capital value function includes: in, Let i be the current investor, s be the current environmental state, s' be the next environmental state, A be the current action, R(s,A) be the reward obtained by taking action A in the current environmental state s, π be the current strategy, and Vπ(s') be the estimated value of environmental state s' when strategy π is adopted. The capital value function is updated using the following formula: in, Let S be the updated value function of the investor at time t, and s be the current environmental state. t Let be the environment state at time t, i be the current investor, a be the learning rate, R(s,i) be the reward for investor i in the current environment state s, γ be the reward discount factor, and A be the learning rate. t +1 represents the action to be taken at the next time step after time t, Qπ(S) t A t+1 ) is in environmental state S t The following strategy π and action A are adopted. t+1 The value of.

2. The method according to claim 1, characterized in that, The construction process of the investor screening model includes: Construct an initial DQN matching model for investor matching scenarios; Based on sample order data and sample candidate investors, sample order parameters and sample candidate investor parameters are extracted as training environment parameters, and the training environment parameters are passed into the initial DQN matching model to obtain the trained DQN matching model; Based on the calculation results of the investor value function of the trained DQN matching model, the investor ranking index corresponding to the sample order data is obtained; Based on the investor sorting index, the sample order data is matched with the target investor, and the target parameter of the matched target investor is maximized to train the investor value function in the initial DQN matching model, thereby obtaining the training result of the DQN matching model.

3. The method according to claim 1, characterized in that, Environmental State S t The following strategy π and action A are adopted. t+1 The value of Qπ(S) t A t+1 It can be obtained through the following formula: Wherein, P(i|S t+1 A t+1 ) represents the environment S at time t+1. t+1 In the middle, action A is used. t+1 The state transition probability of investor i at time. Let be the updated capital value function at time t+1.

4. The method according to claim 1, characterized in that, The matching result of the current order data is obtained based on the sorting index, including: Based on the sorting index and the approval rate of each investor, determine the probability of matching each investor; The target investor with the highest probability is identified, and this target investor is used as the matching result for the current order data.

5. A capital matching device, characterized in that, include: The order data and candidate investor acquisition unit is used to acquire at least one order data and acquire candidate investors; The investor screening model acquisition unit is used to extract the order parameters of the at least one order data and the investor parameters of the candidate investors, and input them into a pre-built DQN matching model to obtain the investor screening model; wherein, the investor screening model includes an investor value function; The investor matching result acquisition unit is used to perform investor matching on the current order data of the at least one order data based on the current investor value function of the investor screening model, obtain the current order data matching result, update the investor value function, and perform investor matching on the next order data until all order data is matched, and obtain the investor matching result of at least one order data. The unit for obtaining the matching result of the investor is used to perform: The investor value for each investor is calculated using the current investor value function. Each investor is sorted in descending order of investor value to obtain the sorting index; The current order data matching result is obtained based on the sorting index; The capital value function includes: in, Let i be the current investor, s be the current environmental state, s' be the next environmental state, A be the current action, R(s,A) be the reward obtained by taking action A in the current environmental state s, π be the current strategy, and Vπ(s') be the estimated value of environmental state s' when strategy π is adopted. The capital value function is updated using the following formula: in, Let S be the updated value function of the investor at time t, and s be the current environmental state. t Let be the environment state at time t, i be the current investor, a be the learning rate, R(s,i) be the reward for investor i in the current environment state s, γ be the reward discount factor, and A be the learning rate. t +1 represents the action to be taken at the next time step after time t, Qπ(S) t A t+1 ) is in environmental state S t The following strategy π and action A are adopted. t+1 The value of.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the investor matching method as described in any one of claims 1-4.

7. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the investor matching method as described in any one of claims 1-4.

Images