Model training method and device, electronic equipment and storage medium

Abstract

The invention relates to a model training method and apparatus, an electronic device and a storage medium. The method comprises the steps of obtaining a to-be-trained recommendation model and sample guide information of the round; inputting the sample guide information into the recommendation model to be trained in the current round to obtain a first recommendation result; evaluating the first recommendation result by adopting multiple evaluation strategies to obtain a comprehensive evaluation score of the first recommendation result; and with the purpose of improving the comprehensive evaluation score of the first recommendation result, updating parameters in the recommendation model to be trained in the round. The method is essentially characterized in that a first recommendation result of a to-be-trained recommendation model is evaluated through multiple evaluation means instead of evaluating the first recommendation result only through a single evaluation means, a comprehensive evaluation score of the first recommendation result is introduced into a loss function, and parameters in the recommendation model are updated, so that the first recommendation result is recommended. The purpose that the trained recommendation model can output a high-quality recommendation result can be achieved.

Description

Technical Field

[0001] The present disclosure relates to the field of model training technology, and in particular to a model training method, device, electronic device, and storage medium. Background Art

[0002] In conversational applications, users can ask questions to an agent via text. The agent generates answers based on the user's questions and displays a set of recommended questions after the answers. Recommended questions are follow-up questions that the user can choose from. If a recommended question meets the user's follow-up question needs, the user can directly click on the recommended question provided by the agent to ask the follow-up question without having to type it. In related technologies, recommended questions are typically generated by recommendation models. Improving the quality of the recommendation results generated by recommendation models is one of the key technologies to reduce the time cost and input complexity of users when asking follow-up questions, and it is also a problem that needs to be solved urgently. Summary of the Invention

[0003] In order to solve the above technical problems or at least partially solve the above technical problems, the present disclosure provides a model training method, device, electronic device and storage medium.

[0004] In a first aspect, the present disclosure provides a model training method, comprising:

[0005] Obtain the recommendation model and sample guidance information to be trained in this round;

[0006] Inputting the sample guidance information into the recommendation model to be trained in this round to obtain a first recommendation result;

[0007] Using multiple evaluation strategies to evaluate the first recommendation result to obtain a comprehensive evaluation score of the first recommendation result;

[0008] With the goal of improving the comprehensive evaluation score of the first recommendation result, the parameters of the recommendation model to be trained in this round are updated.

[0009] In a second aspect, the present disclosure further provides a model training device, comprising:

[0010] The acquisition module is used to obtain the recommendation model to be trained in this round and the sample guidance information;

[0011] an inference module, configured to input the sample guidance information into the recommendation model to be trained in this round to obtain a first recommendation result;

[0012] a scoring module, configured to evaluate the first recommendation result using a plurality of evaluation strategies to obtain a comprehensive evaluation score of the first recommendation result;

[0013] An updating module is used to update the parameters of the recommendation model to be trained in this round with the goal of improving the comprehensive evaluation score of the first recommendation result.

[0014] In a third aspect, the present disclosure further provides an electronic device, comprising:

[0015] one or more processors;

[0016] a storage device for storing one or more programs;

[0017] When the one or more programs are executed by the one or more processors, the one or more processors implement the model training method as described above.

[0018] In a fourth aspect, the present disclosure also provides a computer-readable storage medium on which a computer program is stored, which implements the model training method described above when executed by a processor.

[0019] The technical solution provided by the embodiments of the present disclosure has the following advantages over the prior art:

[0020] The technical solution provided by the embodiment of the present disclosure is to obtain the recommendation model to be trained in this round and the sample guidance information; input the sample guidance information into the recommendation model to be trained in this round to obtain the first recommendation result; adopt a variety of evaluation strategies to evaluate the first recommendation result to obtain the comprehensive evaluation score of the first recommendation result; with the goal of improving the comprehensive evaluation score of the first recommendation result, the parameters in the recommendation model to be trained in this round are updated. Its essence is to evaluate the first recommendation result of the recommendation model to be trained by multiple evaluation means, rather than by only a single evaluation means, and introduce the comprehensive evaluation score of the first recommendation result into the loss function. By updating the parameters in the recommendation model, the purpose of the trained recommendation model being able to output high-quality recommendation results can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0022] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0023] Figure 1A flowchart of a model training method provided in an embodiment of the present disclosure;

[0024] Figure 2 Schematic diagram of the structure of a model training device in an embodiment of the present disclosure;

[0025] Figure 3 Schematic diagram of the structure of an electronic device in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0026] In order to more clearly understand the above-mentioned objectives, features and advantages of the present disclosure, the scheme of the present disclosure will be further described below. It should be noted that the embodiments of the present disclosure and the features therein can be combined with each other in the absence of conflict.

[0027] In the following description, many specific details are set forth to facilitate a full understanding of the present disclosure, but the present disclosure may also be implemented in other ways different from those described herein; it is obvious that the embodiments in the specification are only part of the embodiments of the present disclosure, rather than all of the embodiments.

[0028] Figure 1 A flowchart of a model training method provided in an embodiment of the present disclosure. This embodiment is applicable to the case where model training is performed in a client. The method can be executed by a model training device, which can be implemented in software and / or hardware. The device can be configured in an electronic device, such as a terminal, specifically including but not limited to a smartphone, a PDA, a tablet computer, a wearable device with a display screen, a desktop computer, a laptop computer, an all-in-one computer, a smart home device, etc. Alternatively, this embodiment is applicable to the case where model training is performed in a server. The method can be executed by a model training device, which can be implemented in software and / or hardware. The device can be configured in an electronic device, such as a server.

[0029] It should be noted that in practice, when training a recommendation model, multiple rounds of training are required to ensure that the trained recommendation model outputs high-quality recommendation results. The model training method provided in this application can be applied to any one or more training rounds in the recommendation model training process.

[0030] like Figure 1 As shown, the method may specifically include:

[0031] S110: Obtain the recommendation model to be trained in this round and sample guidance information.

[0032] A recommendation model, for example, can provide recommendation results based on guidance information. In some scenarios, recommendation models are referred to as search engines, recommendation engines, query recommendation systems, and query auto-completion engines. Guidance information, for example, can be information that needs to be input into the recommendation model to control its reasoning process and produce recommendation results. The recommendation results correspond to the guidance information. In other words, different guidance information results in different recommendation results.

[0033] The specific use cases of recommendation models are diverse and are not limited in this application. For example, in a conversational application, the recommendation model is used to provide recommended questions for the user to choose based on historical communication records between the user and the agent. In a query scenario, the recommendation model is used to provide query keywords related to the partial information entered by the user.

[0034] The recommendation model to be trained may, for example, already be capable of providing recommendations based on guidance information, but the quality of its recommendations is poor and requires optimization. The goal of training the recommendation model to be trained is to optimize its parameters so that it outputs high-quality recommendations.

[0035] If this round is the first training round, the recommendation model to be trained in this round is the recommendation model in its initial state. The recommendation model in its initial state is the recommendation model at training round 0. It should be noted that in actual applications, there may be multiple versions of a recommendation model, with later versions being obtained by optimizing earlier versions. The process of optimizing an earlier version to a later version requires multiple rounds of training. In this case, the recommendation model in its initial state is the recommendation model of the earlier version.

[0036] If this round is the second or subsequent round of training, the recommendation model to be trained in this round is the recommendation model obtained after the previous round of training.

[0037] The sample guidance information may be, for example, guidance information used during the training of the recommendation model to be trained in this round.

[0038] S120: Input the sample guidance information into the recommendation model to be trained in this round to obtain a first recommendation result.

[0039] It should be noted that in some scenarios, the recommendation model to be trained in this round may output multiple first recommendation results. This corresponds to the specific usage scenario of the recommendation model. For example, in a scenario where the recommendation model generates a recommendation question based on historical communication records between the user and the agent, the first recommendation result specifically corresponds to the recommendation question. Therefore, based on the sample guidance information, the recommendation model to be trained may generate multiple first recommendation results.

[0040] S130: Evaluate the first recommendation result using multiple evaluation strategies to obtain a comprehensive evaluation score of the first recommendation result.

[0041] The evaluation strategy may be, for example, a method for evaluating the quality of recommendation results. The differences between different evaluation strategies may be reflected in at least one of the following aspects: different evaluation criteria, different evaluation rules, and different evaluation tools.

[0042] Exemplarily, the multiple evaluation strategies include at least two of the following: a preset first evaluation rule; an evaluation model set for the usage scenario of the recommendation model; a language processing model with the ability to evaluate the recommendation results based on a preset second evaluation rule; and a language processing model with the ability to evaluate the recommendation results based on a preset third evaluation rule and retrieval results.

[0043] The preset first evaluation rule may be, for example, a pre-set clear evaluation rule. For example, it specifies what the score of the recommendation result will be when certain conditions are met. It may be manually set or generated by a language processing model. The present application does not limit the specific dimension or dimensions from which the preset evaluation rules evaluate the recommendation results. When setting it specifically, it may be set in combination with the purpose of the recommendation results. For example, if the recommendation model is used to give recommendation questions based on the historical communication records between the user and the intelligent agent, the first evaluation rule may be, for example, setting the evaluation rules from dimensions such as the diversity, language consistency, format, and text length of the recommendation results.

[0044] An evaluation model tailored to a recommendation model's usage scenario can, for example, be a model customized to improve the model's performance in that specific scenario. For example, if the recommendation model is used in a scenario where it generates recommendations based on historical communication records between users and agents, the user's adoption of the recommended questions is particularly important. Based on this, the evaluation model can be trained using data reflecting the user's adoption of the recommended questions, resulting in an evaluation model tailored to the recommendation model's usage scenario.

[0045] In practice, multiple evaluation models can be trained specifically for the use cases of the recommendation model. Different evaluation models have different evaluation functions. For example, if the recommendation model is used in a scenario where recommendations are given based on historical communication records between users and agents, two evaluation models can be trained for this scenario: one evaluation model is used to select high-quality recommendations from multiple results, while the other evaluation model evaluates each recommendation result and provides reasons for its evaluation.

[0046] The first language processing model capable of evaluating the recommendation results based on the preset second evaluation rule may be, for example, a general, conventional language processing model that is not customized for a usage scenario of the recommendation model.

[0047] The second language processing model, capable of evaluating recommendation results based on pre-set third-level evaluation rules and search results, can be, for example, a general-purpose, conventional language processing model that is not customized for the recommendation model's usage scenario. It should be noted that the primary difference between the first and second language processing models is that the second language processing model first searches for the inference results to be evaluated and then evaluates them based on the search results and the third-level evaluation rules.

[0048] The evaluation dimensions involved in any two of the first, second, and third evaluation rules can be completely identical, partially identical, or completely different. In some scenarios, the evaluation granularity of the second and third evaluation rules may be coarser than that of the first evaluation rule. For example, the second and third evaluation rules may only indicate the dimensions along which the evaluation is conducted; or the second and third evaluation rules may specify the criteria for determining each evaluation level without quantifying the evaluation level into a score.

[0049] If the evaluation result obtained based on certain evaluation strategies is not a score, the evaluation result can be quantified to obtain an evaluation score.

[0050] In practice, there are multiple methods for implementing this step, which are not limited in this application. For example, the implementation method of this step may include: evaluating the first recommendation result using multiple evaluation strategies to obtain an evaluation score for the first recommendation result under each evaluation strategy; determining a weight value corresponding to each evaluation strategy; and obtaining a comprehensive evaluation score for the first recommendation result based on the evaluation score of the first recommendation result under each evaluation strategy and the weight value corresponding to each evaluation strategy.

[0051] For example, if there are four evaluation strategies, namely evaluation strategy F1, evaluation strategy F2, evaluation strategy F3, and evaluation strategy F4, where the weight value corresponding to evaluation strategy F1 is a, the weight value corresponding to evaluation strategy F2 is b, the weight value corresponding to evaluation strategy F3 is c, and the weight value corresponding to evaluation strategy F4 is d. Using F1(X) to represent the evaluation score of the first recommendation result X under evaluation strategy F1, using F2(X) to represent the evaluation score of the first recommendation result X under evaluation strategy F2, using F3(X) to represent the evaluation score of the first recommendation result X under evaluation strategy F3, and using F4(X) to represent the evaluation score of the first recommendation result X under evaluation strategy F4, the comprehensive evaluation score Y(X) of the first recommendation result X can be expressed as:

[0052] Y(X)=F1(X)·a+F2(X)·b+F3(X)·c+F4(X)·d.

[0053] S140 : With the goal of improving the comprehensive evaluation score of the first recommendation result, update the parameters of the recommendation model to be trained in this round.

[0054] The essence of this step is to introduce the comprehensive evaluation score of the first recommendation result into the loss function. By updating the parameters in the recommendation model, the comprehensive evaluation score of the recommendation result of the recommendation model gradually develops in an upward direction, thereby achieving the goal of enabling the optimized recommendation model to output high-quality recommendation results.

[0055] It should be noted that updating the parameters of the recommendation model to be trained in this round may mean updating some or all of the parameters of the recommendation model to be trained in this round.

[0056] The above technical solution is to obtain the recommendation model to be trained in this round and the sample guidance information; input the sample guidance information into the recommendation model to be trained in this round to obtain the first recommendation result; adopt multiple evaluation strategies to evaluate the first recommendation result and obtain the comprehensive evaluation score of the first recommendation result; with the goal of improving the comprehensive evaluation score of the first recommendation result, the parameters in the recommendation model to be trained in this round are updated. Its essence is to evaluate the first recommendation result of the recommendation model to be trained by multiple evaluation methods, rather than by only a single evaluation method, and introduce the comprehensive evaluation score of the first recommendation result into the loss function. By updating the parameters in the recommendation model, the purpose of the trained recommendation model being able to output high-quality recommendation results can be achieved.

[0057] On the basis of the above technical solution, optionally, S130 may also include: with the goal of improving the comprehensive evaluation score of the first recommendation result, performing a first update on the parameters of the recommendation model to be trained in this round to obtain the recommendation model after the first update; determining the recognition degree of the recommendation model in the initial state for the recommendation model after the first update; the recommendation model in the initial state is the recommendation model with a training round of 0; with the goal of improving the recognition degree of the recommendation model in the initial state for the recommendation model after the first update, performing a second update on the parameters of the recommendation model after the first update.

[0058] The degree of recognition of the recommendation model in the initial state for the recommendation model after the first update can be, for example, the degree of recognition of the recommendation model in the initial state for the recommendation results of the recommendation model after the first update. If the recommendation model in the initial state has a high degree of recognition of the recommendation results of the recommendation model after the first update, it means that, based on the same conditions (such as the same guidance information), the recommendation model in the initial state has a high probability of outputting the recommendation results of the recommendation model after the first update. It also means that the first update makes the recommendation model evolve in a direction that can generate correct, reasonable, and improved recommendation results with a comprehensive evaluation score. If the recommendation model in the initial state has a low degree of recognition of the recommendation results of the recommendation model after the first update, it means that, based on the same conditions (such as the same guidance information), the recommendation model in the initial state has a low probability of outputting the recommendation results of the recommendation model after the first update. It also means that the first update makes the recommendation model evolve in a direction that improves the comprehensive evaluation score but ignores the correctness and rationality of the generated recommendation results.

[0059] Furthermore, the degree of recognition of the recommendation model in the initial state for the recommendation model after the first update is determined, including: inputting the sample guidance information into the recommendation model after the first update to obtain a second recommendation result; determining the probability that the recommendation model in the initial state will infer the second recommendation result when the sample guidance information is input into the recommendation model in the initial state; and using the probability that the recommendation model in the initial state will infer the second recommendation result as the degree of recognition of the recommendation model in the initial state for the recommendation model after the first update.

[0060] With the goal of improving the recognition of the recommendation model in the initial state for the recommendation model after the first update, the parameters of the recommendation model after the first update are updated for the second time. Its essence is to introduce the recognition of the recommendation model in the initial state for the recommendation model after the first update into the loss function. By updating the parameters in the recommendation model for the second time, the recommendation model is prompted to evolve in the direction of generating correct, reasonable and comprehensively evaluated recommendation results.

[0061] Based on the above technical solutions, the method may optionally further include: inputting the sample guidance information into the recommendation model after the second update to obtain a third recommendation result; using multiple evaluation strategies to evaluate the third recommendation result to obtain the evaluation score of the third recommendation result under each evaluation strategy; updating the weight value corresponding to the evaluation strategy based on the evaluation score of the first recommendation result under each evaluation strategy and the evaluation score of the third recommendation result under each evaluation strategy.

[0062] Furthermore, based on the evaluation scores of the first recommendation result under each evaluation strategy and the evaluation scores of the third recommendation result under each evaluation strategy, the weight values ​​corresponding to the evaluation strategies are updated, including: determining a target evaluation strategy among multiple evaluation strategies, the evaluation score of the third recommendation result under the target evaluation strategy is lower than the evaluation score of the first recommendation result under the target evaluation strategy; and increasing the weight value corresponding to the target evaluation strategy.

[0063] For example, assume that there are four evaluation strategies, namely evaluation strategy F1, evaluation strategy F2, evaluation strategy F3 and evaluation strategy F4. The evaluation score of the first recommendation result X1 under evaluation strategy F1 is F1(X1), and the evaluation score of the third recommendation result X3 under evaluation strategy F1 is F1(X3). Compare F1(X1) and F1(X3). If F1(X1)>F1(X3), increase the weight value corresponding to evaluation strategy F1. If F1(X1)≤F1(X3), keep the weight value corresponding to evaluation strategy F1 unchanged. The method for updating the weight values ​​corresponding to evaluation strategy F2, evaluation strategy F3 and evaluation strategy F4 is similar to the method for updating the weight value corresponding to evaluation strategy F1, and will not be repeated here.

[0064] By setting the evaluation scores for the first and third recommended results under each evaluation strategy, and updating the corresponding weights for each evaluation strategy, the goal is to improve the evaluation scores of the recommended results output by the recommendation model under one evaluation strategy without sacrificing their scores under other evaluation strategies during the training of the recommendation model. In other words, this setting can prevent the problem of optimal combination coefficients shifting as the recommendation model trains. Here, optimal combination coefficient shift refers to the gradual change in the quality characteristics of the recommended results output by the recommendation model as the recommendation model trains, causing the originally optimal weight combination for each evaluation strategy to become ineffective in evaluating the recommended results. In other words, the originally optimal weight combination for each evaluation strategy no longer reflects the changed quality characteristics of the recommended results. Continuing to use the originally optimal weight combination for each evaluation strategy will lead to a deviation in the overall evaluation score of the recommended results, failing to accurately reflect the actual quality of the recommended results.

[0065] It is understandable that before using the technical solutions disclosed in the various embodiments of this disclosure, the type, scope of use, usage scenarios, etc. of the personal information involved in this disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0066] For example, in response to a user's active request, a prompt message is sent to the user to clearly inform the user that the operation requested will require the acquisition and use of the user's personal information. This allows the user to independently choose whether to provide personal information to the electronic device, application, server, storage medium, or other software or hardware that performs the operations of the disclosed technical solution based on the prompt message.

[0067] As an optional but non-limiting implementation, in response to receiving a user's active request, the prompt information may be sent to the user in the form of a pop-up window, in which the prompt information may be presented in text form. Furthermore, the pop-up window may also contain a selection control for the user to select "agree" or "disagree" to provide personal information to the electronic device.

[0068] It is understandable that the above notification and user authorization process are merely illustrative and do not limit the implementation of the present disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of the present disclosure.

[0069] It should be noted that for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by this application.

[0070] Figure 2 This is a schematic diagram of the structure of a model training device in an embodiment of the present disclosure. The model training device provided in the embodiment of the present disclosure can be configured in the client or in the server. Figure 2 , the model training device specifically includes:

[0071] Acquisition module 210, used to obtain the recommendation model to be trained in this round and sample guidance information;

[0072] An inference module 220 is configured to input the sample guidance information into the recommendation model to be trained in this round to obtain a first recommendation result;

[0073] A scoring module 230 is configured to evaluate the first recommendation result using multiple evaluation strategies to obtain a comprehensive evaluation score for the first recommendation result;

[0074] The updating module 240 is configured to update the parameters of the recommendation model to be trained in this round with the goal of improving the comprehensive evaluation score of the first recommendation result.

[0075] Furthermore, the scoring module 230 is used to:

[0076] Using multiple evaluation strategies to evaluate the first recommendation result, and obtaining evaluation scores of the first recommendation result under each of the evaluation strategies;

[0077] Determining the weight value corresponding to each of the evaluation strategies;

[0078] Based on the evaluation scores of the first recommendation result under each of the evaluation strategies and the weight values ​​corresponding to each of the evaluation strategies, a comprehensive evaluation score of the first recommendation result is obtained.

[0079] Furthermore, the update module 240 is configured to:

[0080] With the goal of improving the comprehensive evaluation score of the first recommendation result, the parameters of the recommendation model to be trained in this round are updated for the first time to obtain the recommendation model after the first update;

[0081] Determining the recognition degree of the recommendation model in the initial state to the recommendation model after the first update; the recommendation model in the initial state is the recommendation model with a training round of 0;

[0082] With the goal of improving the recognition of the recommendation model in the initial state to the recommendation model after the first update, the parameters of the recommendation model after the first update are updated for the second time.

[0083] Furthermore, the update module 240 is configured to:

[0084] Inputting the sample guidance information into the recommendation model after the first update to obtain a second recommendation result;

[0085] determining a probability that the recommendation model in the initial state infers the second recommendation result when the sample guidance information is input into the recommendation model in the initial state;

[0086] The probability that the recommendation model in the initial state infers the second recommendation result is used as the recognition degree of the recommendation model in the initial state to the recommendation model after the first update.

[0087] Furthermore, the device also includes a weight updating module, which is used to:

[0088] Inputting the sample guidance information into the recommendation model after the second update to obtain a third recommendation result;

[0089] Using the multiple evaluation strategies, the third recommendation result is evaluated to obtain an evaluation score of the third recommendation result under each of the evaluation strategies;

[0090] Based on the evaluation scores of the first recommendation result under each of the evaluation strategies and the evaluation scores of the third recommendation result under each of the evaluation strategies, the weight values ​​corresponding to the evaluation strategies are updated.

[0091] Furthermore, the weight update module is used to:

[0092] Determining a target evaluation strategy among the multiple evaluation strategies, wherein the evaluation score of the third recommendation result under the target evaluation strategy is lower than the evaluation score of the first recommendation result under the target evaluation strategy;

[0093] Increase the weight value corresponding to the target evaluation strategy.

[0094] Furthermore, the multiple evaluation strategies include at least two of the following:

[0095] Preset first evaluation rule;

[0096] An evaluation model set for the usage scenario of the recommendation model;

[0097] A first language processing model capable of evaluating recommendation results based on a preset second evaluation rule;

[0098] A second language processing model capable of evaluating recommendation results based on preset third evaluation rules and retrieval results.

[0099] The model training device provided in the embodiment of the present disclosure can execute the steps executed by the client or server in the model training method provided in the embodiment of the method of the present disclosure, and has the execution steps and beneficial effects, which will not be repeated here.

[0100] The following specific reference Figure 3 , which shows a schematic structural diagram of an electronic device 1000 suitable for implementing the embodiments of the present disclosure. The electronic device 1000 in the embodiments of the present disclosure may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), wearable electronic devices, and fixed terminals such as digital TVs, desktop computers, smart home devices, and the like. Figure 3 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0101] like Figure 3As shown, the electronic device 1000 may include a processing device (such as a central processing unit, a graphics processing unit, etc.) 1001, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1008 into a random access memory (RAM) 1003 to implement the model training method of the embodiment as described in the present disclosure. In the RAM 1003, various programs and information required for the operation of the electronic device 1000 are also stored. The processing device 1001, the ROM 1002, and the RAM 1003 are connected to each other via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0102] Typically, the following devices may be connected to the I / O interface 1005: an input device 1006 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 1007 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 1008 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 may allow the electronic device 1000 to communicate with other devices wirelessly or by wire to exchange information. Although Figure 3 The electronic device 1000 is shown with various devices, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead.

[0103] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart, thereby implementing the model training method as described above. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device 1009, or installed from the storage device 1008, or installed from the ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment of the present disclosure are performed.

[0104] It should be noted that the computer-readable medium mentioned above in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, device, or component. In the present disclosure, a computer-readable signal medium may include an information signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated information signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to wires, optical cables, RF (radio frequency), etc., or any suitable combination thereof.

[0105] In some embodiments, the client and server can communicate using any known or future developed network protocol, such as HTTP (HyperText Transfer Protocol), and can be interconnected with any form or medium of digital information communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any known or future developed network.

[0106] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0107] The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device:

[0108] Obtain the recommendation model and sample guidance information to be trained in this round;

[0109] Inputting the sample guidance information into the recommendation model to be trained in this round to obtain a first recommendation result;

[0110] Using multiple evaluation strategies to evaluate the first recommendation result to obtain a comprehensive evaluation score of the first recommendation result;

[0111] With the goal of improving the comprehensive evaluation score of the first recommendation result, the parameters of the recommendation model to be trained in this round are updated.

[0112] Optionally, when the above one or more programs are executed by the electronic device, the electronic device may also execute other steps described in the above embodiments.

[0113] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages, or a combination thereof, including, but not limited to, object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server. In cases involving a remote computer, the remote computer may 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 may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0114] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0115] The units involved in the embodiments described in this disclosure may be implemented in software or hardware, wherein the name of a unit does not necessarily limit the unit itself.

[0116] The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, and without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0117] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0118] According to one or more embodiments of the present disclosure, the present disclosure provides an electronic device, including:

[0119] one or more processors;

[0120] a memory for storing one or more programs;

[0121] When the one or more programs are executed by the one or more processors, the one or more processors implement any model training method provided in the present disclosure.

[0122] According to one or more embodiments of the present disclosure, the present disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any model training method as provided in the present disclosure.

[0123] An embodiment of the present disclosure also provides a computer program product, which includes a computer program or instructions, and when the computer program or instructions are executed by a processor, implements the model training method described above.

[0124] It should be noted that, in this document, relational terms such as "first" and "second" 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, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0125] The foregoing description is intended only to provide specific embodiments of the present disclosure, intended to enable those skilled in the art to understand and implement the present disclosure. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure is not intended to be limited to the embodiments described herein, but rather to be construed in the broadest manner consistent with the principles and novel features disclosed herein.

Claims

1. A model training method, characterized in that: include: Obtain the recommendation model and sample guidance information to be trained in this round; Inputting the sample guidance information into the recommendation model to be trained in this round to obtain a first recommendation result; Using multiple evaluation strategies to evaluate the first recommendation result to obtain a comprehensive evaluation score of the first recommendation result; With the goal of improving the comprehensive evaluation score of the first recommendation result, the parameters of the recommendation model to be trained in this round are updated.

2. The method according to claim 1, characterized in that The adopting multiple evaluation strategies to evaluate the first recommendation result to obtain a comprehensive evaluation score of the first recommendation result includes: Using multiple evaluation strategies to evaluate the first recommendation result, and obtaining evaluation scores of the first recommendation result under each of the evaluation strategies; Determining the weight value corresponding to each of the evaluation strategies; Based on the evaluation scores of the first recommendation result under each of the evaluation strategies and the weight values ​​corresponding to each of the evaluation strategies, a comprehensive evaluation score of the first recommendation result is obtained.

3. The method according to claim 2, characterized in that The updating of the parameters of the recommendation model to be trained in this round with the goal of improving the comprehensive evaluation score of the first recommendation result further includes: With the goal of improving the comprehensive evaluation score of the first recommendation result, performing a first update on the parameters of the recommendation model to be trained in this round to obtain the recommendation model after the first update; Determining the recognition degree of the recommendation model in the initial state to the recommendation model after the first update; the recommendation model in the initial state is the recommendation model with a training round of 0; With the goal of improving the recognition of the recommendation model in the initial state to the recommendation model after the first update, the parameters of the recommendation model after the first update are updated for the second time.

4. The method according to claim 3, characterized in that The determining of the recognition degree of the recommendation model in the initial state to the recommendation model after the first update includes: Inputting the sample guidance information into the recommendation model after the first update to obtain a second recommendation result; determining a probability that the recommendation model in the initial state infers the second recommendation result when the sample guidance information is input into the recommendation model in the initial state; The probability that the recommendation model in the initial state infers the second recommendation result is used as the recognition degree of the recommendation model in the initial state to the recommendation model after the first update.

5. The method according to claim 3, characterized in that The method further comprises: Inputting the sample guidance information into the recommendation model after the second update to obtain a third recommendation result; Using the multiple evaluation strategies, the third recommendation result is evaluated to obtain an evaluation score of the third recommendation result under each evaluation strategy; Based on the evaluation scores of the first recommendation result under each of the evaluation strategies and the evaluation scores of the third recommendation result under each of the evaluation strategies, the weight values ​​corresponding to the evaluation strategies are updated.

6. The method according to claim 5, characterized in that The updating of the weight values ​​corresponding to the evaluation strategies based on the evaluation scores of the first recommendation results under the evaluation strategies and the evaluation scores of the third recommendation results under the evaluation strategies includes: Determining a target evaluation strategy among the multiple evaluation strategies, wherein the evaluation score of the third recommendation result under the target evaluation strategy is lower than the evaluation score of the first recommendation result under the target evaluation strategy; Increase the weight value corresponding to the target evaluation strategy.

7. The method according to claim 1, characterized in that The multiple evaluation strategies include at least two of the following: Preset first evaluation rule; An evaluation model set for the usage scenario of the recommendation model; A first language processing model capable of evaluating recommendation results based on a preset second evaluation rule; A second language processing model capable of evaluating recommendation results based on preset third evaluation rules and retrieval results.

8. A model training device, characterized in that: include: The acquisition module is used to obtain the recommendation model to be trained in this round and the sample guidance information; an inference module, configured to input the sample guidance information into the recommendation model to be trained in this round to obtain a first recommendation result; a scoring module, configured to evaluate the first recommendation result using a plurality of evaluation strategies to obtain a comprehensive evaluation score of the first recommendation result; An updating module is used to update the parameters of the recommendation model to be trained in this round with the goal of improving the comprehensive evaluation score of the first recommendation result.

9. An electronic device, characterized in that: The electronic device comprises: one or more processors; a storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images