Model parameter adjustment method, device and equipment based on recall knowledge

Abstract

The invention discloses a model parameter adjustment method, device and equipment based on recall knowledge, and relates to the field of artificial intelligence. The model parameter adjustment method based on recall knowledge comprises the following steps: dynamically extracting candidate recall knowledge points based on current question and answer sample data and a target question knowledge base; determining multiple features of the current question and answer sample according to the candidate recall knowledge points; and adjusting model hyper-parameters of the question and answer model based on the multiple features of the current question and answer sample and the confidence scores of the candidate recalled knowledge points. According to the technical scheme, the question and answer model can present different answer effects in different recall scenes, the applicability of the question and answer model is improved, and the question and answer effect of the question and answer model is greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a model parameter adjustment method, device and equipment based on recall knowledge. Background Art

[0002] The question-answering system is based on the RAG (Retrieval-Augmented Generation) architecture, which breaks down the knowledge source data into knowledge points according to a certain split granularity, finds the background knowledge points related to the question more quickly in the vertical field, and generates answers to the questions based on the language understanding ability of the large model (i.e., large language model). Among them, the RAG architecture will be combined with external knowledge retrieval, specifically connecting real-time data to the large model, so as to understand the questions raised based on the large model, and generate answers based on the knowledge points related to the questions, that is, using the large model as a question-answering model.

[0003] In actual application scenarios, the question-answering model will present different styles depending on the hyperparameter settings: when the randomness is low, the question-answering model will be rigorous and meticulous; when the randomness is high, the answers of the question-answering model are relatively random, and the thinking of the question-answering model is relatively divergent, with more model illusions. If the hyperparameters are set to fixed values, the performance style of the question-answering model will be relatively rigid, and the applicability to the scenario will be insufficient. How to determine the appropriate hyperparameters has become an urgent problem to be solved in the question-answering model. Summary of the invention

[0004] The present invention provides a model parameter adjustment method, device and equipment based on recall knowledge to solve the problem that the current question-answering model is not applicable to the scene and the answering effect is not good.

[0005] According to one aspect of the present invention, a method for adjusting model parameters based on recall knowledge is provided, comprising:

[0006] Dynamically extract candidate recall knowledge points based on the current question and answer sample data and the target question knowledge base;

[0007] Determine multiple features of the current question and answer sample based on candidate recall knowledge points;

[0008] Adjust the model hyperparameters of the question-answering model based on the multiple features of the current question-answering sample and the confidence scores of the candidate recalled knowledge points.

[0009] According to another aspect of the present invention, a device for adjusting model parameters of recall knowledge is provided, comprising:

[0010] The recall knowledge point dynamic extraction module is used to dynamically extract candidate recall knowledge points based on the current question and answer sample data and the target question knowledge base;

[0011] A multiple feature determination module is used to determine multiple features of the current question and answer sample based on candidate recall knowledge points;

[0012] The model parameter adjustment module is used to adjust the model hyperparameters of the question-answering model based on the multiple features of the current question-answering sample and the confidence scores of the candidate recalled knowledge points.

[0013] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0014] at least one processor; and

[0015] a memory communicatively connected to the at least one processor; wherein,

[0016] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the model parameter adjustment method based on recall knowledge described in any embodiment of the present invention.

[0017] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the model parameter adjustment method based on recall knowledge described in any embodiment of the present invention when executed.

[0018] According to another aspect of the present invention, a computer program product is provided, characterized in that the computer program product includes a computer program, and when the computer program is executed by a processor, the model parameter adjustment method based on recall knowledge described in any embodiment of the present invention is implemented.

[0019] The technical solution of the embodiment of the present invention dynamically extracts candidate recall knowledge points based on the current question and answer sample data and the target question knowledge base, thereby determining multiple features of the current question and answer sample based on the candidate recall knowledge points, and then adjusting the model hyperparameters of the question and answer model based on the multiple features of the current question and answer sample and the confidence scores of the candidate recall knowledge points. In this solution, candidate recall knowledge points can be dynamically extracted for the current question and answer sample data in different scenarios to ensure that the candidate recall knowledge points are highly adapted to the recall scenarios, and the features of the candidate recall knowledge points are extracted from multiple dimensions, combined with the confidence scores of the candidate recall knowledge points, and the model hyperparameters adapted to the recall scenarios are obtained by fitting the question and answer model, which solves the problem that the current question and answer model is not strong in applicability to the scenarios and the answering effect is not good, and enables the question and answer model to present different answering effects in different recall scenarios, thereby improving the applicability of the question and answer model and greatly improving the question and answer effect of the question and answer model.

[0020] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 A flowchart of a model parameter adjustment method based on recall knowledge provided in Example 1 of the present invention;

[0023] Figure 2 A flowchart of a method for adjusting model parameters of recall knowledge provided in the second embodiment of the present invention;

[0024] Figure 3 A schematic diagram of the structure of a model parameter adjustment device for recalling knowledge provided in Embodiment 3 of the present invention;

[0025] Figure 4 A schematic diagram of the structure of an electronic device that can be used to implement an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0027] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0028] Embodiment 1

[0029] Figure 1 This is a flowchart of a model parameter adjustment method based on recall knowledge provided in the first embodiment of the present invention. This embodiment can be applied to the situation where the question-answering system accurately answers the user's question. The method can be executed by a model parameter adjustment device based on recall knowledge. The model parameter adjustment device based on recall knowledge can be implemented in the form of hardware and / or software. The model parameter adjustment device based on recall knowledge can be configured in an electronic device. Figure 1 As shown, the method includes:

[0030] Step 110: Dynamically extract candidate recall knowledge points based on the current question and answer sample data and the target question knowledge base.

[0031] The current question and answer sample data may be a question and answer sample set used to adjust the model hyperparameters of the question and answer model. The target question knowledge base may be a database storing background knowledge points related to the current question and answer sample data. The candidate recall knowledge points may be knowledge points dynamically extracted from the target question and answer database and related to the current question and answer sample data.

[0032] In an embodiment of the present invention, knowledge points related to the current question and answer sample data can be dynamically extracted from the target question knowledge base based on the association attributes between the knowledge points and the current question and answer sample data to obtain candidate recall knowledge points, that is, the number of candidate recall knowledge points extracted from the target question knowledge base for different question and answer sample data can be different.

[0033] Optionally, the current question and answer sample data can be preprocessed in combination with the RAG architecture, and various processes in the search field, such as recall, rough sorting, fine sorting, and re-ranking, can be used to recall the background knowledge related to the question of the current question and answer sample data from the target question knowledge base, and dynamically extract the top-ranked knowledge points as candidate recall knowledge points. Among them, the higher the ranking of the candidate recall knowledge point, the higher the confidence.

[0034] Step 120: Determine multiple features of the current question and answer sample based on the candidate recall knowledge points.

[0035] Among them, the multiple features of the current question and answer sample can be the features extracted from the candidate recalled knowledge points based on multiple preset dimensions.

[0036] In an embodiment of the present invention, multi-dimensional feature extraction can be performed on candidate recall knowledge points based on a preset feature extraction type to obtain multiple features of the current question and answer sample.

[0037] Step 130: Adjust the model hyperparameters of the question-answering model based on the multiple features of the current question-answering sample and the confidence scores of the candidate recalled knowledge points.

[0038] Among them, the confidence score of the candidate recalled knowledge point can be the confidence score of the candidate recalled knowledge point, which can be understood as the probability of the knowledge point being a positive sample (i.e., the correct answer). The higher the probability, the more credible the knowledge point is as the correct answer. Optionally, the value range of the confidence score of the candidate recalled knowledge point is (0,1). Exemplarily, the confidence score of the candidate recalled knowledge point can be determined by a deep learning model or a dual tower model. The question-answering model can be a large language model that can generate answers based on questions. The large language model is a deep learning model with a huge number of parameters, such as BERT (Bidirectional Encoder Representations from Transformers, a pre-trained language representation model). Large language models usually have hundreds of millions of parameters and are trained on large-scale data, so that they can better fit complex data distributions and better capture long-distance dependencies in language, thereby improving the ability of language understanding and generation.

[0039] In an embodiment of the present invention, multiple features of the current question and answer sample and confidence scores of candidate recalled knowledge points can be used as model input parameters to train the question and answer model and fit model hyperparameters that affect the expression of answers generated by the question and answer model.

[0040] The technical solution of the embodiment of the present invention dynamically extracts candidate recall knowledge points based on the current question and answer sample data and the target question knowledge base, thereby determining multiple features of the current question and answer sample based on the candidate recall knowledge points, and then adjusting the model hyperparameters of the question and answer model based on the multiple features of the current question and answer sample and the confidence scores of the candidate recall knowledge points. In this solution, candidate recall knowledge points can be dynamically extracted for the current question and answer sample data in different scenarios to ensure that the candidate recall knowledge points are highly adapted to the recall scenarios, and the features of the candidate recall knowledge points are extracted from multiple dimensions, combined with the confidence scores of the candidate recall knowledge points, and the model hyperparameters adapted to the recall scenarios are obtained by fitting the question and answer model, which solves the problem that the current question and answer model is not strong in applicability to the scenarios and the answering effect is not good, and enables the question and answer model to present different answering effects in different recall scenarios, thereby improving the applicability of the question and answer model and greatly improving the question and answer effect of the question and answer model.

[0041] Embodiment 2

[0042] Figure 2 This is a flowchart of a method for adjusting model parameters of recall knowledge provided in the second embodiment of the present invention. This embodiment is specific based on the above embodiment and provides a specific optional implementation method for dynamically extracting candidate recall knowledge points based on the current question and answer sample data and the target question knowledge base. Figure 2 As shown, the method includes:

[0043] Step 210: Obtain a knowledge point dynamic extraction model.

[0044] In an optional embodiment of the present invention, obtaining a dynamic knowledge point extraction model may include: determining supervised learning question and answer samples and a recall return candidate set; training a supervised learning model based on the supervised learning question and answer samples and the recall return candidate set to obtain a dynamic knowledge point extraction model.

[0045] The supervised learning question and answer samples may be pre-collected question and answer samples used for supervised learning training. The recall return candidate set may be a set of recall knowledge points corresponding to the supervised learning question and answer samples. The knowledge point dynamic extraction model may be a model for dynamically extracting candidate recall knowledge points.

[0046] In an embodiment of the present invention, supervised learning question and answer samples and recall return candidate sets can be first obtained, and then the supervised learning question and answer samples and the recall return candidate sets can be input into the supervised learning model for supervised model training, and the trained supervised learning model can be used as a knowledge point dynamic extraction model to generate a knowledge point dynamic extraction model through supervised learning, thereby effectively ensuring the extraction reliability of candidate recall knowledge points of the knowledge point dynamic extraction model.

[0047] Step 220: input the current question and answer sample data and the target question knowledge base into the knowledge point dynamic extraction model to obtain the candidate recall knowledge points of the dynamic extraction quantity of knowledge points.

[0048] Among them, the number of dynamically extracted knowledge points can be the number of candidate recall knowledge points output by the knowledge point dynamic extraction model, specifically a dynamic number adapted to the recall scenario.

[0049] In an embodiment of the present invention, the current question and answer sample data and the target question knowledge base are used as input data and input into a knowledge point dynamic extraction model. The knowledge point dynamic extraction model is used to query the target question knowledge base for a number of candidate recall knowledge points that are related to the background knowledge of the current question and answer sample data and are dynamically extracted.

[0050] Step 230: Determine multiple features of the current question and answer sample based on the candidate recall knowledge points.

[0051] In an optional embodiment of the present invention, determining multiple features of the current question and answer sample based on candidate recall knowledge points may include: screening candidate recall knowledge points based on confidence scores of the candidate recall knowledge points to obtain feature knowledge points to be extracted; performing multi-dimensional feature extraction on the feature knowledge points to be extracted to obtain multiple features of the current question and answer sample.

[0052] The feature knowledge points to be extracted may be knowledge points that meet the confidence requirements and are selected from the candidate recalled knowledge points based on the confidence scores of the candidate recalled knowledge points.

[0053] In an embodiment of the present invention, the confidence score of the candidate recall knowledge point can be determined, so that based on the confidence score of the candidate recall knowledge point, the candidate recall knowledge point is screened, and the knowledge point that meets the confidence requirement is used as the feature knowledge point to be extracted, and then based on the preset feature extraction type, multi-dimensional feature extraction is performed on the feature knowledge point to be extracted to obtain multiple features of the current question and answer sample. By using the confidence score of the candidate recall knowledge point to screen the candidate recall knowledge point, it can be ensured that the knowledge point that is subsequently subjected to multi-dimensional feature extraction has a higher probability of being a correct sample, thereby reliably ensuring the accuracy of the answer of the final question and answer model.

[0054] When the probability of including the correct candidate answer in the candidate recall knowledge points is higher, the question-answering model performs more reasonably and well-founded. At this time, the randomness of the generated answers should be reduced. On the contrary, when the probabilities in the candidate recall knowledge points are not high, the question-answering model performs more cautiously, and the randomness of the answers can be improved accordingly, thereby improving the adaptability of the question-answering system.

[0055] Step 240: Adjust the model hyperparameters of the question-answering model based on the multiple features of the current question-answering sample and the confidence scores of the candidate recalled knowledge points.

[0056] In an optional embodiment of the present invention, adjusting the model hyperparameters of the question and answer model based on multiple features of the current question and answer sample and the confidence scores of the candidate recalled knowledge points may include: using the multiple features of the current question and answer sample and the confidence scores of the candidate recalled knowledge points as model input data, and using the first hyperparameter and the second hyperparameter as model output data; and fitting the hyperparameters through the question and answer model according to the model input data and the model output data.

[0057] Among them, the first hyperparameter can be temperature. The second hyperparameter can be top_p. Temperature is used to adjust the degree of random sampling from the model. After triggering the "generate" instruction, even the same prompt word may produce different outputs. When temperature is 0, the question-answering model will always produce the same output. The higher the value of temperature, the greater the randomness. top_p is used to dynamically set the size of the tokens candidate list, and the toptokens whose sum of probabilities does not exceed a certain value are included in the candidate list. top_p is usually set to a higher value (such as 0.75) in order to limit the length of low-probability tokens that may be sampled.

[0058] In an embodiment of the present invention, the first hyperparameter and the second hyperparameter that affect the answer effect in different recall scenarios in the question-answering model can be determined, or the first hyperparameter and the second hyperparameter in the question-answering model can be selected by the user, so that the multiple features of the current question-answering sample and the confidence score of the candidate recall knowledge point are used as model input data, and the first hyperparameter and the second hyperparameter are used as model output data, and then the model input data and the model output data are used to perform hyperparameter fitting through the first hyperparameter and the second hyperparameter in the question-answering model. That is, this scheme supports multiple ways to determine the first hyperparameter and the second hyperparameter, and does not restrict the selection rules of the first hyperparameter and the second hyperparameter. The confidence scores of the relevant candidate knowledge points recalled are also used as variable inputs to construct a fitting question-answering model, and the model hyperparameters used to control randomness in the question-answering model are used as model outputs. The model hyperparameters of the question-answering model are dynamically adjusted, so that the question-answering model presents different answer effects in different recall scenarios, effectively avoiding the shortcomings of the question-answering model that has a single performance and poor adaptability.

[0059] For example, temper = func(a0, a1, a2, ..., a k ); top_p=func(a0, a1, a2,..., a k ), where a0 represents the multiple features extracted from the first candidate recall knowledge point, which may include a series of model feature engineering output data such as the candidate recall knowledge point confidence score, related word frequency, semantic similarity score, etc. a1 represents the multiple features extracted from the second candidate recall knowledge point, and so on, which will not be repeated here.

[0060] In an optional embodiment of the present invention, performing hyperparameter fitting through a question-and-answer model based on model input data and model output data may include: obtaining target question-and-answer interaction behavior characteristics, and supplementing the model input data based on the target question-and-answer interaction behavior characteristics; and adjusting the model hyperparameters of the question-and-answer model based on the supplemented model input data and model output data.

[0061] The target question-answering interaction behavior feature may be the question-asking behavior feature of the user asking the question corresponding to the current question-answering sample data. For example, the target question-answering interaction behavior feature may include, but is not limited to, the user's preference for the brevity of the question reply, and other features of the interaction between the user and the question-answering model, such as the user's question like rate within a statistical period.

[0062] In an embodiment of the present invention, the target question-answering interaction behavior characteristics of the user who asks the question corresponding to the current question-answering sample data can be obtained, so that the target question-answering interaction behavior characteristics are added to the model input data, and the model hyperparameters of the question-answering model are adjusted using the supplemented model input data and model output data. By enriching the model input data with the target question-answering interaction behavior characteristics, the final question-answering model can provide personalized answers to users and improve user satisfaction.

[0063] The technical solution of the embodiment of the present invention obtains a knowledge point dynamic extraction model, thereby inputting the current question and answer sample data and the target question knowledge base into the knowledge point dynamic extraction model, obtaining the candidate recall knowledge points of the number of knowledge point dynamic extraction, and then determining the multiple features of the current question and answer sample according to the candidate recall knowledge points, and further adjusting the model hyperparameters of the question and answer model based on the multiple features of the current question and answer sample and the confidence score of the candidate recall knowledge points. In this solution, the dynamic extraction model of the knowledge point automatically extracts the recall knowledge points from the target question knowledge base for the recall scenario corresponding to the input question and answer data, ensuring that the candidate recall knowledge points are highly adapted to the recall scenario, and extracting the features of the candidate recall knowledge points from multiple dimensions, combined with the confidence score of the candidate recall knowledge points, and fitting the model hyperparameters adapted to the recall scenario with the help of the question and answer model, solving the problem that the current question and answer model is not strong in the applicability of the scenario and the answering effect is not good, enabling the question and answer model to present different answering effects in different recall scenarios, improving the applicability of the question and answer model, and greatly improving the question and answer effect of the question and answer model.

[0064] Embodiment 3

[0065] Figure 3 This is a schematic diagram of the structure of a model parameter adjustment device for recalling knowledge provided in the third embodiment of the present invention. Figure 3 As shown, the device comprises:

[0066] A recall knowledge point dynamic extraction module 310 is used to dynamically extract candidate recall knowledge points based on current question and answer sample data and a target question knowledge base;

[0067] A multiple feature determination module 320, used to determine multiple features of the current question and answer sample based on the candidate recall knowledge points;

[0068] The model parameter adjustment module 330 is used to adjust the model hyperparameters of the question-answering model based on the multiple features of the current question-answering sample and the confidence scores of the candidate recalled knowledge points.

[0069] The technical solution of the embodiment of the present invention dynamically extracts candidate recall knowledge points based on the current question and answer sample data and the target question knowledge base, thereby determining multiple features of the current question and answer sample based on the candidate recall knowledge points, and then adjusting the model hyperparameters of the question and answer model based on the multiple features of the current question and answer sample and the confidence scores of the candidate recall knowledge points. In this solution, candidate recall knowledge points can be dynamically extracted for the current question and answer sample data in different scenarios to ensure that the candidate recall knowledge points are highly adapted to the recall scenarios, and the features of the candidate recall knowledge points are extracted from multiple dimensions, combined with the confidence scores of the candidate recall knowledge points, and the model hyperparameters adapted to the recall scenarios are obtained by fitting the question and answer model, which solves the problem that the current question and answer model is not strong in applicability to the scenarios and the answering effect is not good, and enables the question and answer model to present different answering effects in different recall scenarios, thereby improving the applicability of the question and answer model and greatly improving the question and answer effect of the question and answer model.

[0070] Optionally, the recall knowledge point dynamic extraction module 310 includes a dynamic extraction model acquisition unit and a recall knowledge point dynamic extraction unit. The dynamic extraction model acquisition unit is used to obtain a knowledge point dynamic extraction model. The recall knowledge point dynamic extraction unit is used to input the current question and answer sample data and the target question knowledge base into the knowledge point dynamic extraction model to obtain the candidate recall knowledge points of the knowledge point dynamic extraction quantity.

[0071] Optionally, a dynamic extraction model acquisition unit is specifically used to determine supervised learning question and answer samples and recall return candidate sets; based on the supervised learning question and answer samples and the recall return candidate sets, a supervised learning model is trained to obtain the knowledge point dynamic extraction model.

[0072] Optionally, the multiple feature determination module 320 is specifically used to screen the candidate recalled knowledge points according to the confidence scores of the candidate recalled knowledge points to obtain feature knowledge points to be extracted; perform multi-dimensional feature extraction on the feature knowledge points to be extracted to obtain multiple features of the current question and answer sample.

[0073] Optionally, the model parameter adjustment module 330 is specifically used to use the multiple features of the current question and answer sample and the confidence score of the candidate recalled knowledge point as model input data, and use the first hyperparameter and the second hyperparameter as model output data; and perform hyperparameter fitting through the question and answer model based on the model input data and the model output data.

[0074] Optionally, the model parameter adjustment module 330 is specifically used to obtain target question-answering interaction behavior characteristics, and supplement the model input data according to the target question-answering interaction behavior characteristics; and adjust the model hyperparameters of the question-answering model based on the supplemented model input data and the model output data.

[0075] The model parameter adjustment device for recall knowledge provided in the embodiment of the present invention can execute the model parameter adjustment method for recall knowledge provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0076] Embodiment 4

[0077] Figure 4 A schematic diagram of an electronic device that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0078] like Figure 4 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0079] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0080] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as a model parameter adjustment method for recalling knowledge.

[0081] In some embodiments, the model parameter adjustment method for recall knowledge may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the model parameter adjustment method for recall knowledge described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to execute the model parameter adjustment method for recall knowledge in any other appropriate manner (e.g., by means of firmware).

[0082] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0083] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0084] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may 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.

[0085] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0086] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0087] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of traditional physical hosts and VPS servers, which are difficult to manage and have weak business scalability.

[0088] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0089] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A model parameter adjustment method based on recall knowledge, characterized in that: include: Dynamically extract candidate recall knowledge points based on the current question and answer sample data and the target question knowledge base; Determine multiple features of the current question and answer sample based on the candidate recall knowledge points; Based on the multiple features of the current question and answer sample and the confidence scores of the candidate recalled knowledge points, the model hyperparameters of the question and answer model are adjusted.

2. The method according to claim 1, characterized in that The method of dynamically extracting candidate recall knowledge points based on the current question and answer sample data and the target question knowledge base includes: Obtain a dynamic knowledge point extraction model; The current question and answer sample data and the target question knowledge base are input into the knowledge point dynamic extraction model to obtain the candidate recall knowledge points of the dynamic extraction quantity of knowledge points.

3. The method according to claim 2, characterized in that The method of obtaining a knowledge point dynamic extraction model includes: Determine supervised learning question answering samples and recall return candidate sets; Based on the supervised learning question and answer samples and the recalled return candidate set, a supervised learning model is trained to obtain the knowledge point dynamic extraction model.

4. The method according to claim 1, characterized in that: Determining multiple features of the current question and answer sample based on the candidate recall knowledge points includes: According to the confidence scores of the candidate recalled knowledge points, the candidate recalled knowledge points are screened to obtain feature-to-be-extracted knowledge points; Multi-dimensional feature extraction is performed on the feature-to-be-extracted knowledge points to obtain multiple features of the current question and answer sample.

5. The method according to claim 1, characterized in that The adjusting the model hyperparameters of the question-answering model based on the multiple features of the current question-answering sample and the confidence scores of the candidate recalled knowledge points includes: Using the multiple features of the current question and answer sample and the confidence score of the candidate recalled knowledge point as model input data, and using the first hyperparameter and the second hyperparameter as model output data; Hyperparameter fitting is performed through the question-answering model according to the model input data and the model output data.

6. The method according to claim 5, characterized in that The step of performing hyperparameter fitting through the question-answering model according to the model input data and the model output data includes: Acquire target question-answer interaction behavior characteristics, and supplement the model input data according to the target question-answer interaction behavior characteristics; Based on the supplemented model input data and the model output data, adjust the model hyperparameters of the question-answering model.

7. A model parameter adjustment device based on recall knowledge, characterized in that: include: The recall knowledge point dynamic extraction module is used to dynamically extract candidate recall knowledge points based on the current question and answer sample data and the target question knowledge base; A multiple feature determination module, used to determine multiple features of the current question and answer sample based on the candidate recall knowledge points; The model parameter adjustment module is used to adjust the model hyperparameters of the question-answering model based on the multiple features of the current question-answering sample and the confidence scores of the candidate recalled knowledge points.

8. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the model parameter adjustment method based on recall knowledge described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the model parameter adjustment method based on recall knowledge described in any one of claims 1 to 6 when executed.

10. A computer program product, characterized in that The computer program product comprises a computer program, which, when executed by a processor, implements the model parameter adjustment method based on recall knowledge according to any one of claims 1 to 6.

Images