Question information processing method and device, electronic equipment and storage medium

Abstract

The invention provides a problem information processing method and device, electronic equipment and a storage medium, and belongs to the technical field of computers. The method comprises the steps of obtaining a pre-trained large language model, question information belonging to any business scene and an action set associated with the question information; creating a target decoding tree; for any existing target node in the target decoding tree, processing information corresponding to the target node by adopting a large language model according to at least one action in the action set to obtain at least one piece of answer information, and creating at least one child node of the target node until the target decoding tree meets a problem solving condition; and determining target answer information of the big language model for the question information from the answer information corresponding to the plurality of leaf nodes in the target decoding tree. According to the method, the problem information can be solved without training the pre-trained large language model for the business scene to which the problem information belongs, so that the labor cost and training resources required by training are saved.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a method, device, electronic device and storage medium for processing question information. Background Art

[0002] With the rapid development of artificial intelligence technology, Large Language Model (LLM) has become a new direction and new tool for solving problems in various industries and scientific research fields in the information age. Large language models can receive inputs such as natural language, audio, and images, and combine the perception of world knowledge and self-understanding to output corresponding interactive answers, proposing solutions to various human problems.

[0003] In the related art, a pre-trained large language model is usually obtained first. The pre-trained large language model is universal and can be applied to a variety of business scenarios, but it is not targeted and cannot accurately solve the problem in a certain business scenario. At this time, the data in the target business scenario can be manually labeled to obtain sample data, and the pre-trained large language model can be trained again with the sample data. The obtained target large language model can accurately solve the problem in the target business scenario.

[0004] However, the above method requires manual labeling of a large amount of data, which consumes manpower costs and consumes a large amount of training resources during the training process. Summary of the invention

[0005] The present disclosure provides a method, device, electronic device and storage medium for processing question information, which, on the basis of ensuring that the answer information has a certain degree of accuracy, does not need to further train the pre-trained large language model for the business scenario to which the question information belongs, thus saving the manpower cost and training resources required for training. The technical solution of the present disclosure is as follows:

[0006] According to one aspect of an embodiment of the present disclosure, a method for processing question information is provided, including:

[0007] Obtaining a pre-trained large language model, question information belonging to any business scenario, and an action set associated with the question information, where the actions in the action set are actions that the current user account specifies the large language model to perform when processing the question information;

[0008] Creating a target decoding tree, wherein the target decoding tree includes a root node corresponding to the question information;

[0009] For any existing target node in the target decoding tree, the large language model is used to process the information corresponding to the target node according to at least one action in the action set, to obtain at least one answer information, and at least one child node of the target node is created, wherein the at least one child node corresponds to the at least one answer information in a one-to-one manner, until the target decoding tree meets the problem solving condition;

[0010] Determine target answer information of the large language model for the question information from answer information corresponding to multiple leaf nodes in the target decoding tree.

[0011] According to another aspect of an embodiment of the present disclosure, there is provided a device for processing question information, including:

[0012] An acquisition unit is configured to acquire a pre-trained large language model, question information belonging to any business scenario, and an action set associated with the question information, wherein the actions in the action set are actions that the current user account specifies the large language model to perform when processing the question information;

[0013] A creating unit, configured to create a target decoding tree, wherein the target decoding tree includes a root node corresponding to the question information;

[0014] The creation unit is further configured to execute, for any target node already in the target decoding tree, using the large language model, processing the information corresponding to the target node according to at least one action in the action set, obtaining at least one answer information, and creating at least one child node of the target node, wherein the at least one child node corresponds to the at least one answer information in a one-to-one manner, until the target decoding tree meets the problem solving condition;

[0015] The first determination unit is configured to determine target answer information of the large language model for the question information from answer information corresponding to multiple leaf nodes in the target decoding tree.

[0016] In some embodiments, the target node is the root node;

[0017] The creation unit is configured to randomly select at least one action from the action set; for any action among the at least one action, use the large language model to process the question information corresponding to the root node according to the action, obtain the answer information corresponding to the action, create a node corresponding to the answer information, and determine the node corresponding to the answer information as a child node of the root node.

[0018] In some embodiments, the target node is any node in the target decoding tree except the root node;

[0019] The creation unit comprises:

[0020] An acquisition subunit is configured to select at least one action from the action set;

[0021] A sub-unit is created, which is configured to execute any action among the at least one action, adopt the large language model, process the answer information and the question information corresponding to the target node according to the action, obtain the answer information corresponding to the action, create a node corresponding to the answer information, and determine the node corresponding to the answer information as a child node of the target node.

[0022] In some embodiments, the acquisition subunit is configured to perform at least one of the following:

[0023] Based on the target action adopted when generating the answer information corresponding to the target node based on the parent node of the target node, selecting at least one action from the action set, wherein the at least one action is the same as or different from the target action;

[0024] Based on the action adopted in the target link of the target decoding tree, selecting the at least one action from the action set, the target link refers to the link from the root node to the target node, and the at least one action is the same as or different from the action adopted in the target link;

[0025] Selecting the at least one action from the action set based on the processing progress of the question information, wherein the processing progress is used to indicate the degree to which the answer information of the target node solves the question information;

[0026] Based on a preset width of the target decoding tree and the number of nodes in the level to which the target node belongs, determine the target number of child nodes to which the target node can be assigned in the level, and select the target number of actions from the action set, wherein the preset width represents the maximum number of nodes in each level of the target decoding tree.

[0027] In some embodiments, the acquisition subunit is configured to execute a target action adopted when acquiring answer information corresponding to the target node based on the parent node of the target node; and select at least one action related to the target action from the action set.

[0028] In some embodiments, the problem solving condition includes at least one of the following:

[0029] The depth of the target decoding tree reaches a preset depth;

[0030] The answer information corresponding to any leaf node in the target decoding tree is correct.

[0031] In some embodiments, the apparatus further includes a second determining unit configured to perform at least one of the following:

[0032] For any leaf node in the target decoding tree, in response to the current user account confirming the correctness of the answer information corresponding to the leaf node, determining that the answer information corresponding to the leaf node is correct;

[0033] For any leaf node in the target decoding tree, the answer information corresponding to the leaf node is judged to obtain a judgment result, and the judgment result is used to indicate whether the answer information of the leaf node is correct.

[0034] In some embodiments, the first determining unit is configured to determine at least one target answer information satisfying a target condition from the answer information corresponding to the plurality of leaf nodes in the target decoding tree;

[0035] The target condition is that the target answer information appears a preset number of times in the answer information corresponding to the multiple leaf nodes, or the target condition is that the target answer information is determined to be the answer to the question information.

[0036] In some embodiments, the apparatus further comprises:

[0037] The third determination unit is configured to execute, for any target answer information, input the target answer information and the question information into the large language model; determine the matching degree between the target answer information and the question information through the large language model, wherein the matching degree is used to indicate the correct degree of the target answer information in answering the question information; and select the target answer information whose matching degree satisfies the matching condition from the at least one target answer information.

[0038] In some embodiments, there are multiple large language models, and each of the multiple large language models is used to process the question information and determine the answer information of other large language models;

[0039] The device also includes:

[0040] The fourth determination unit is configured to execute the determination of multiple answer information obtained by the first large language model using at least one other large language model among the multiple large language models except the first large language model, and obtain the determination results of the multiple answer information, wherein the first large language model is any model among the multiple large language models, and the multiple answer information is the answer information corresponding to multiple leaf nodes in the target decoding tree of the first large language model, and the determination result of each answer information is used to indicate whether the answer information is correct; based on the determination results of the multiple answer information, determine the accuracy of the first large language model; determine the large language model with the highest accuracy from the multiple large language models, and determine the target answer information for the question information based on the multiple answer information of the large language model with the highest accuracy.

[0041] In some embodiments, there are multiple large language models, and the multiple large language models include multiple second large language models and one third large language model, each of the second large language models is used to process the question information, and the third large language model is used to determine the answer information of each of the second large language models;

[0042] The device also includes:

[0043] The fifth determination unit is configured to execute, for any second largest language model among the multiple large language models, using the third largest language model to determine multiple answer information obtained by the second largest language model, and obtain determination results of the multiple answer information, wherein the multiple answer information are answer information corresponding to multiple leaf nodes in the target decoding tree of the second largest language model, and the determination result of each answer information is used to indicate whether the answer information is correct; based on the determination results of the multiple answer information, filter out the correct answer information obtained by the second largest language model from the multiple answer information; summarize the correct answer information of the multiple second largest language models to obtain the target answer information for the question information.

[0044] In some embodiments, there are multiple large language models, and each of the multiple large language models is used to process the question information and determine the correctness of the answer information of the associated model;

[0045] The device also includes:

[0046] a sixth determination unit configured to execute a fourth language model to determine a plurality of answer information obtained by a fifth language model, to obtain determination results of the plurality of answer information, wherein the fourth language model is any one of the plurality of large language models, the fifth language model is an associated model of the fourth language model, and different large language models have different associated models, the plurality of answer information is answer information corresponding to a plurality of leaf nodes in a target decoding tree of the fifth language model, and the determination result of each answer information is used to indicate whether the answer information is correct;

[0047] Based on the determination results of the multiple answer information, the correct answer information output by the fifth language model is selected from the multiple answer information;

[0048] The correct answer information of the multiple large language models is aggregated to obtain target answer information for the question information.

[0049] According to another aspect of an embodiment of the present disclosure, an electronic device is provided, the electronic device comprising:

[0050] one or more processors;

[0051] A memory for storing program codes executable by the processor;

[0052] The processor is configured to execute the program code to implement the above-mentioned method for processing problem information.

[0053] According to another aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided. When a program code in the computer-readable storage medium is executed by a processor of an electronic device, the electronic device can execute the above-mentioned method for processing question information.

[0054] According to another aspect of an embodiment of the present disclosure, a computer program product is provided, including a computer program / instruction, which implements the above-mentioned method for processing question information when the computer program / instruction is executed by a processor.

[0055] The solution provided by the embodiment of the present disclosure, in the process of using a large language model to process question information, by creating a target decoding tree, the pre-trained large language model gradually processes the question information according to the actions specified by the user, that is, the decoding path of the large language model to generate answer information is interfered by the target decoding tree and the action set. Since the actions in the action set are specified by the current user account, it is guaranteed that the generated answer information meets the needs of the model user, that is, the accuracy of the answer information is guaranteed; on the basis of ensuring that the answer information has a certain accuracy, there is no need to further train the pre-trained large language model for the business scenario to which the question information belongs, which not only saves the manpower cost and training resources required for training, and enables the large language model to quickly achieve the purpose required by the user, improves the problem processing efficiency, and guarantees the versatility of the large language model, which can be applied to various business scenarios. In addition, in the process of processing question information, a variety of processing ideas can be provided for the large language model through the action set, ensuring that the answer information output by the large language model has diversity.

[0056] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description are used to explain the principles of the present disclosure, and do not constitute improper limitations on the present disclosure.

[0058] Figure 1 The present invention is a schematic diagram of an implementation environment of a method for processing problem information according to an exemplary embodiment.

[0059] Figure 2 The present invention is a flowchart of a method for processing question information according to an exemplary embodiment.

[0060] Figure 3 The figure is a flowchart of another method for processing question information according to an exemplary embodiment.

[0061] Figure 4 The diagram is a schematic diagram showing a joint processing of question information among multiple large language models according to an exemplary embodiment.

[0062] Figure 5 The figure is a schematic diagram showing another method of jointly processing question information among multiple large language models according to an exemplary embodiment.

[0063] Figure 6 The figure is a schematic diagram showing another method of jointly processing question information among multiple large language models according to an exemplary embodiment.

[0064] Figure 7 The invention is an architectural diagram showing a method of processing problem information according to an exemplary embodiment.

[0065] Figure 8 The invention is a block diagram of a device for processing question information according to an exemplary embodiment.

[0066] Fig. 9 It is a block diagram of a terminal according to an exemplary embodiment.

[0067] Fig.10 It is a block diagram of a server according to an exemplary embodiment. DETAILED DESCRIPTION

[0068] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.

[0069] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure 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 disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0070] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.) and signals involved in this disclosure are all authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions. For example, the problem information and action sets involved in this disclosure are all obtained with full authorization.

[0071] The method for processing question information provided by the embodiment of the present disclosure can be executed by an electronic device, and the electronic device can be provided as at least one of a terminal and a server. Figure 1 FIG. 1 is a schematic diagram of an implementation environment of a method for processing problem information according to an exemplary embodiment. Figure 1 , the implementation environment specifically includes: a terminal 101 and a server 102.

[0072] In an embodiment of the present disclosure, the terminal 101 is used to obtain the answer information of the user's question information and present the answer information to the user. In some embodiments, a target application is installed on the terminal 101, and the target application is used to obtain the answer information of the user's question information and present the answer information to the user. The target application can be a question-and-answer application, a query application, a game application, a browser, etc., which is not limited to this. For example, the target application takes the form of a web application, a micro program or a client, etc., and the present disclosure is not limited to this. Among them, a micro program refers to a program that runs on other applications, such as a small program. The server 102 is used to provide background services for the target application running on the terminal 101. For example, the terminal sends the user's question information to the server 102, the server 102 processes the question information, obtains the answer information of the question information, and sends it to the terminal 101, and the terminal 101 presents the answer information to the user. Of course, the server 102 can also include other functional servers to provide more comprehensive and diversified services.

[0073] Among them, the terminal 101 is at least one of a smart phone, a smart watch, a desktop computer, a laptop, an MP3 player, an MP4 player, and a laptop computer. The terminal 101 has a communication function and can access a wired network or a wireless network. The terminal 101 generally refers to one of a plurality of terminals. This embodiment is illustrated by taking the terminal 101 as an example. Those skilled in the art can know that the number of the above terminals can be more or less. For example, the above terminals can be several, or the above terminals can be dozens or hundreds, or more. The embodiment of the present disclosure does not limit the number of terminals and device types.

[0074] The server 102 may be an independent physical server, or a server cluster or distributed file system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. In some embodiments, the server 102 is directly or indirectly connected to the terminal 101 through wired or wireless communication, which is not limited in the embodiments of the present disclosure. Optionally, the number of the above-mentioned servers 102 may be more or less, which is not limited in the embodiments of the present disclosure. Of course, the server 102 may also include other functional servers to provide more comprehensive and diversified services. Among them, the server 102 undertakes the main computing work, and the terminal 101 undertakes the secondary computing work; or, the server 102 undertakes the secondary computing work, and the terminal 101 undertakes the main computing work; or, the server 102 or the terminal 101 can each undertake the computing work separately, which is not limited in the embodiments of the present disclosure.

[0075] Figure 2 is a flow chart of a method for processing problem information according to an exemplary embodiment. Figure 2 As shown, the method for processing the problem information is applied in the server, comprising the following steps:

[0076] In step 201, the server obtains a pre-trained large language model, question information belonging to any business scenario, and an action set associated with the question information, where the actions in the action set are actions that the current user account specifies the large language model to perform when processing the question information.

[0077] In the disclosed embodiment, the pre-trained large language model can be any open source large language model such as ChatGPT (Chat Generative Pre-trained Transformer), LaMDA (Learning And Mining from Data), BERT (Bidirectional Encoder Representation from Transformers), etc., which is not limited in the disclosed embodiment. The pre-trained large language model is versatile and can be applied to a variety of business scenarios.

[0078] In the disclosed embodiment, the business scenario to which the question information belongs can be a knowledge question answering scenario (e.g., asking the meaning of any term in a certain field), or an information generation scenario (e.g., requiring the generation of an article), or a mathematical calculation scenario (e.g., calculating a formula), or a game scenario (e.g., requiring the user to plan a way to pass the game), etc., which is not limited in the disclosed embodiment. For the question information, the current user account can set an action set related to the question information to indicate the action to be performed when the large language model processes the question information.

[0079] Taking the case where the question information belongs to the knowledge question and answer scenario as an example, the action set may include multiple actions such as "for explaining keyword a", "for explaining keyword b", and "generating complete answer information at one time". Taking the case where the question information belongs to the information generation scenario as an example, the action set may include multiple requirements for the information to be generated, such as "this paragraph of text contains keyword c", "the theme of this paragraph of text is artificial intelligence", "generate a paragraph of text each time", "generate the entire article at one time", and other multiple actions. Taking the case where the question information belongs to the mathematical calculation scenario as an example, the action set may include multiple algorithmic actions such as "addition", "subtraction", "multiplication", "division", and "square root". Taking the case where the question information belongs to the game scenario as an example, the action set may include multiple directional actions such as "up", "down", "left", and "right", so that the subsequent large language model can plan a route for the current user account, etc. The embodiment of the present disclosure does not limit the action set. The server can obtain the question information of the current user account and the action set related to the question information from the terminal.

[0080] In step 202, the server creates a target decoding tree, where the target decoding tree includes a root node corresponding to the question information.

[0081] In the disclosed embodiment, the server can create a target decoding tree according to the question information and the action set. The server can first create a root node of the target decoding tree according to the question information. Then, the server can create lower nodes based on the root node according to the action set to expand the depth and width of the target decoding tree. For details, please refer to the content in step 203.

[0082] In step 203, for any existing target node in the target decoding tree, the server uses a large language model to process the information corresponding to the target node according to at least one action in the action set, obtain at least one answer information, and create at least one child node of the target node, and the at least one child node corresponds to the at least one answer information one by one, until the target decoding tree meets the problem solving conditions.

[0083] In the embodiment of the present disclosure, the target node may be a root node in the target decoding tree, or may be any node in the target decoding tree except the root node, which is not limited in the embodiment of the present disclosure.

[0084] For any target node already in the target decoding tree, the server selects at least one action from the action set. For any action, the server can input the information corresponding to the action and the target node into the large language model, and process the information of the target node according to the action through the large language model to obtain the answer information corresponding to the action. Then, the server selects the node corresponding to the answer information and uses the node as the child node of the target node. That is, for any target node already in the target decoding tree, the server can create at least one child node of the target node based on at least one action in the action set.

[0085] As the depth of the target decoding tree gradually increases, the answer information obtained by the large language model becomes more and more detailed and can more and more accurately answer the question information. When the target decoding tree meets the problem-solving condition, the server stops creating new nodes in the target decoding tree. That is, when the target decoding tree meets the problem-solving condition, the server no longer further answers the question information through the large language model. The disclosed embodiment does not limit the problem intercept condition.

[0086] In step 204, the server determines target answer information of the large language model for the question information from the answer information corresponding to the plurality of leaf nodes in the target decoding tree.

[0087] In the disclosed embodiment, a leaf node is a node without child nodes in the target decoding tree. Accordingly, the answer information corresponding to the leaf node is the final and most detailed answer information obtained in the link where the leaf node is located. The server determines the target answer information of the large language model for the question information from the answer information corresponding to multiple leaf nodes in the target decoding tree.

[0088] Optionally, the server may directly regard the answer information corresponding to the multiple leaf nodes in the target decoding tree as the target answer information of the large language model for the question information. Alternatively, the server may also filter out at least one answer information from the answer information corresponding to the multiple leaf nodes in the target decoding tree, and regard the at least one answer information as the target answer information of the large language model for the question information. The embodiments of the present disclosure are not limited to this.

[0089] The disclosed embodiment provides a method for processing question information. In the process of using a large language model to process question information, by creating a target decoding tree, the pre-trained large language model gradually processes the question information according to the action specified by the user, that is, the decoding path of the large language model to generate answer information is interfered by the target decoding tree and the action set. Since the actions in the action set are specified by the current user account, it is guaranteed that the generated answer information meets the needs of the model user, that is, the accuracy of the answer information is guaranteed; on the basis of ensuring that the answer information has a certain accuracy, there is no need to further train the pre-trained large language model for the business scenario to which the question information belongs, which not only saves the manpower cost and training resources required for training, but also enables the large language model to quickly achieve the purpose required by the user, improves the problem processing efficiency, and guarantees the versatility of the large language model, which can be applied to various business scenarios. In addition, in the process of processing question information, a variety of processing ideas can be provided for the large language model through the action set, ensuring that the answer information output by the large language model has diversity.

[0090] In some embodiments, the target node is a root node;

[0091] For any target node in the target decoding tree, the large language model is used to process information corresponding to the target node according to at least one action in the action set, obtain at least one answer information, and create at least one child node of the target node, including:

[0092] Randomly select at least one action from the action set;

[0093] For any action in at least one action, a large language model is used to process the question information corresponding to the root node according to the action, to obtain the answer information corresponding to the action, and to create a node corresponding to the answer information, and to determine the node corresponding to the answer information as a child node of the root node.

[0094] The solution provided by the embodiment of the present disclosure obtains at least one action from an action set in the process of creating a child node of the root node in the target decoding tree, directly instructs the large language model to perform a first round of processing on the question information according to the action specified by the user, and obtains at least one answer information obtained in the first round of processing, thereby ensuring that the generated answer information meets the needs of the model user, that is, ensuring the accuracy of the answer information; and can record the answer information corresponding to each action by creating a child node, so that the question information can be answered in more detail later on the basis of the existing answer information, that is, by interfering with the process of generating answer information by the large language model through the target decoding tree and the action set, the answer information required by the model user can be quickly obtained without model training, thereby achieving the purpose required by the user and improving the problem processing efficiency.

[0095] In some embodiments, the target node is any node in the target decoding tree except the root node;

[0096] For any target node in the target decoding tree, the large language model is used to process information corresponding to the target node according to at least one action in the action set, obtain at least one answer information, and create at least one child node of the target node, including:

[0097] Select at least one action from the action set;

[0098] For any action in at least one action, a large language model is used to process the answer information and question information corresponding to the target node according to the action, to obtain the answer information corresponding to the action, and to create a node corresponding to the answer information, and to determine the node corresponding to the answer information as a child node of the target node.

[0099] The solution provided by the embodiment of the present disclosure obtains at least one action from an action set during the process of creating a child node of any target node except the root node in the target decoding tree, and directly instructs the large language model to perform this round of processing on the answer information and question information of the target node according to the action specified by the user, so as to obtain at least one answer information obtained in this round of processing. This not only ensures that the generated answer information meets the needs of the model user, but also can be further refined on the basis of the existing answer information, thereby ensuring the accuracy of the answer information. That is, by interfering with the process of generating answer information by the large language model through the target decoding tree and the action set, the answer information required by the model user can be quickly obtained without model training, thereby achieving the purpose required by the user and improving the problem processing efficiency.

[0100] In some embodiments, at least one action is selected from the action set, including at least one of the following:

[0101] Based on the target action used when generating answer information corresponding to the target node based on the parent node of the target node, at least one action is selected from the action set, and the at least one action is the same as or different from the target action;

[0102] Based on the actions adopted in the target link of the target decoding tree, at least one action is selected from the action set, where the target link refers to the link from the root node to the target node, and at least one action is the same as or different from the action adopted in the target link;

[0103] Based on the processing progress of the question information, at least one action is selected from the action set, and the processing progress is used to indicate the degree to which the answer information of the target node solves the question information;

[0104] Based on a preset width of the target decoding tree and the number of nodes in the level to which the target node belongs, determine the target number of child nodes to which the target node can be assigned in the level, and select the target number of actions from the action set, where the preset width represents the maximum number of nodes in each level of the target decoding tree.

[0105] In the scheme provided by the embodiment of the present disclosure, since the target action is the action adopted when generating the answer information corresponding to the target node based on the parent node of the target node, that is, the action adopted in the previous step before the current round of processing the problem information, at least one action used in the current round of processing the problem information is determined by the target action, so that each round of processing process is associated with each other, which is conducive to ensuring the accuracy of the answer information. Since the action adopted in the target link can accurately reflect the generation process of the answer information of the target node, at least one action used in the current round of processing the problem information is determined by the action in the target link, so that each round of processing process is associated with the previous processing process, which is conducive to ensuring the accuracy of the answer information. Since the processing progress of the problem information can accurately reflect the degree to which the answer information of the target node solves the problem information, at least one action used in the current round of processing the problem information is determined by the processing progress of the problem information, so that actions that can solve the remaining problems of the problem information can be obtained, ensuring that the answer information generated subsequently can accurately solve the problem information, that is, it is conducive to improving the accuracy of the answer information. Alternatively, by using the preset width of the target decoding tree and the number of nodes in the layer to which the target node belongs, a target number of actions (i.e., child nodes) are allocated to the target node, thereby ensuring that the width of the target decoding tree does not exceed the preset width. This is equivalent to limiting the amount of generated answer information while ensuring the diversity of answer information to a certain extent, thereby avoiding a large language model from generating too much answer information and causing data redundancy.

[0106] In some embodiments, based on the target action used when generating answer information corresponding to the target node based on the parent node of the target node, at least one action is selected from the action set, including:

[0107] Obtain the target action used when generating answer information corresponding to the target node based on the parent node of the target node;

[0108] From the action set, select at least one action related to the target action.

[0109] The solution provided by the embodiment of the present disclosure is that since the target action is the action adopted when generating the answer information corresponding to the target node based on the parent node of the target node, that is, it is the action adopted in the previous step before the current round of processing the question information, at least one action related to the target action is obtained as the action used for the current round of processing the question information, so that each round of processing process is related, which is beneficial to ensure the accuracy of the answer information.

[0110] In some embodiments, the problem solving condition includes at least one of the following:

[0111] The depth of the target decoding tree reaches the preset depth;

[0112] The answer information corresponding to any leaf node in the target decoding tree is correct.

[0113] The solution provided by the embodiment of the present disclosure determines that the target decoding tree meets the problem-solving conditions when the depth of the target decoding tree reaches a preset depth, and no further processing is required, that is, the number of actions (steps) for processing the problem information is limited, avoiding the large language model from performing a large number of actions without limit, and ensuring the efficiency of the large language model in generating answer information. Alternatively, when the answer information corresponding to any leaf node in the target decoding tree is correct, it is determined that the target decoding tree meets the problem-solving conditions and no further processing is required, that is, no further processing is required when the correct answer information is obtained, avoiding the large language model from performing useless actions, and improving the efficiency and accuracy of the large language model in generating answer information.

[0114] In some embodiments, the method further comprises at least one of the following:

[0115] For any leaf node in the target decoding tree, in response to a correct confirmation operation of the current user account on the answer information corresponding to the leaf node, determining that the answer information corresponding to the leaf node is correct;

[0116] For any leaf node in the target decoding tree, the answer information corresponding to the leaf node is judged to obtain a judgment result, and the judgment result is used to indicate whether the answer information of the leaf node is correct.

[0117] The solution provided by the embodiment of the present disclosure can determine whether the answer information corresponding to the leaf node in the target decoding tree is correct by the current user account, which ensures that the generated answer information meets the needs of the model user, that is, the accuracy of the answer information is guaranteed. Alternatively, the answer information corresponding to the leaf node in the target decoding tree can also be judged to determine whether the answer information is accurate, which can not only ensure the accuracy of the answer information, but also does not require user operation, so that the target answer information of the large language model for the question information can be determined quickly from the answer information corresponding to multiple leaf nodes, thereby improving the processing efficiency of the question information.

[0118] In some embodiments, determining target answer information of the large language model for question information from answer information corresponding to multiple leaf nodes in the target decoding tree includes:

[0119] Determine at least one target answer information satisfying a target condition from the answer information corresponding to the plurality of leaf nodes in the target decoding tree;

[0120] Among them, the target condition is that the number of occurrences of the target answer information in the answer information corresponding to multiple leaf nodes reaches a preset number, or the target condition is that the target answer information is determined to be the answer to the question information.

[0121] The solution provided by the embodiment of the present disclosure can more accurately reflect the processing of the question information by the large language model because the number of occurrences of the answer information reaches the preset number of times, and the number of occurrences is large. From the answer information corresponding to multiple leaf nodes in the target decoding tree, the answer information with the number of occurrences reaching the preset number of times is determined as the target answer information for the question information by the large language model, thereby ensuring the accuracy of the answer information. Alternatively, the answer information corresponding to the leaf nodes in the target decoding tree can also be determined to determine whether the answer information is the answer to the question information, which can also ensure the accuracy of the answer information.

[0122] In some embodiments, the method further comprises:

[0123] For any target answer information, the target answer information and question information are input into the large language model;

[0124] The matching degree between the target answer information and the question information is determined through a large language model. The matching degree is used to indicate the correctness of the target answer information in answering the question information.

[0125] The target answer information whose matching degree satisfies the matching condition is selected from the at least one target answer information.

[0126] The solution provided by the embodiment of the present disclosure can also determine the degree of match between the target answer information and the question information through the large language model for any target answer information obtained by the large language model for the question information, so as to further screen out qualified target answer information from at least one target answer information as the final answer to the question information, thereby ensuring the accuracy of the answer information.

[0127] In some embodiments, there are multiple large language models, and each of the multiple large language models is used to process question information and determine answer information of other large language models;

[0128] The method also includes:

[0129] Using at least one other large language model among the multiple large language models except the first large language model, to judge multiple answer information obtained by the first large language model to obtain judgment results of the multiple answer information, wherein the first large language model is any model among the multiple large language models, the multiple answer information is answer information corresponding to multiple leaf nodes in the target decoding tree of the first large language model, and the judgment result of each answer information is used to indicate whether the answer information is correct;

[0130] Based on the judgment results of the multiple answer information, determine the accuracy of the first language model;

[0131] A large language model with the highest accuracy is determined from among the multiple large language models, and target answer information for the question information is determined based on multiple answer information of the large language model with the highest accuracy.

[0132] The solution provided by the embodiment of the present disclosure may have multiple large language models, each of which can process question information to obtain corresponding answer information. The multiple large language models can integrate communication and transmit the multiple answer information obtained by each of them. The multiple answer information of the current large language model is judged by other large language models to determine whether the multiple answer information is correct, thereby determining the accuracy of the current large language model, and then determining the target answer information for the question information based on the multiple answer information of the large language model with the highest accuracy, which is beneficial to ensuring the reliability and objectivity of the answer information generated by the large language model, that is, it is beneficial to ensuring the accuracy of the answer information.

[0133] In some embodiments, there are multiple large language models, and the multiple large language models include multiple second-largest language models and one third-largest language model, each second-largest language model is used to process question information, and the third-largest language model is used to determine answer information of each second-largest language model;

[0134] The method also includes:

[0135] For any second largest language model among the multiple large language models, using the third largest language model, multiple answer information obtained by the second largest language model is judged to obtain judgment results of the multiple answer information, the multiple answer information is answer information corresponding to multiple leaf nodes in the target decoding tree of the second largest language model, and the judgment result of each answer information is used to indicate whether the answer information is correct;

[0136] Based on the determination results of the multiple answer information, the correct answer information obtained by the second largest language model is selected from the multiple answer information;

[0137] The correct answer information of multiple second largest language models is aggregated to obtain target answer information for the question information.

[0138] The solution provided by the embodiment of the present disclosure may have multiple large language models, and designate one large language model (the third largest language model) specifically for determining the correctness of the answers output by other large language models, and aggregate the correct answer information of each second largest language model as the target answer information for the question information. That is, through the model integration communication solution, the results of multiple large language models are combined and outputted to obtain the most appropriate answer, which is beneficial to ensuring the reliability and objectivity of the answer information generated by the large language model, that is, it is beneficial to ensuring the accuracy of the answer information.

[0139] In some embodiments, there are multiple large language models, each of which is used to process question information and determine the correctness of answer information associated with the model; the method further includes:

[0140] The fourth language model is used to judge the multiple answer information obtained by the fifth language model to obtain the judgment results of the multiple answer information, wherein the fourth language model is any model among the multiple large language models, the fifth language model is an associated model of the fourth language model, and the associated models of different large language models are different. The multiple answer information is the answer information corresponding to multiple leaf nodes in the target decoding tree of the fifth language model, and the judgment result of each answer information is used to indicate whether the answer information is correct;

[0141] Based on the determination results of the multiple answer information, the correct answer information output by the fifth language model is selected from the multiple answer information;

[0142] The correct answer information of multiple large language models is aggregated to obtain the target answer information for the question information.

[0143] The solution provided by the embodiment of the present disclosure may have multiple large language models, each of which can be used to determine the correctness of the answer output by the corresponding associated model, and the correct answer information of each large language model is summarized as the target answer information for the question information. That is, through the model integration communication solution, the results of multiple large language models are combined and output to obtain the most appropriate answer, which is beneficial to ensuring the reliability and objectivity of the answer information generated by the large language model, that is, it is beneficial to ensure the accuracy of the answer information.

[0144] Above Figure 2 The following is only a basic process of the present disclosure. The solution provided by the present disclosure is further described based on a specific implementation method. Figure 3 FIG. 1 is a flowchart of another method for processing problem information according to an exemplary embodiment. Figure 3 , the method comprising:

[0145] In step 301, the server obtains a pre-trained large language model, question information belonging to any business scenario, and an action set associated with the question information, where the actions in the action set are actions that the current user account specifies the large language model to perform when processing the question information.

[0146] In the embodiment of the present disclosure, the principle of the server executing step 301 is the same as the principle of executing step 201, which will not be repeated here.

[0147] In step 302, the server creates a target decoding tree, where the target decoding tree includes a root node corresponding to the question information.

[0148] In the embodiment of the present disclosure, in the process of creating the target decoding tree, the server may first set the preset depth and preset width of the target decoding tree to limit the size of the target decoding tree. The server may determine the preset width of the target decoding tree according to the number of actions in the action set. For example, the preset width of the target decoding tree is not less than the number of actions in the action set.

[0149] Then, the server can create the root node of the target decoding tree according to the question information. Based on the root node, the server can create lower nodes based on the root node according to the action set to broaden the depth and width of the target decoding tree. That is, the server can process the question information in multiple rounds according to the action set, and can create corresponding nodes for the answer information obtained in each round of processing to record each round of processing. For details, please refer to the contents in step 303 and step 304.

[0150] In step 303, for the root node in the target decoding tree, the server randomly selects at least one action from the action set; for any action in the at least one action, a large language model is used to process the question information corresponding to the root node according to the action, to obtain the answer information corresponding to the action, and to create a node corresponding to the answer information, and to determine the node corresponding to the answer information as a child node of the root node.

[0151] In an embodiment of the present disclosure, during the first round of processing of the question information, the server randomly selects at least one action from the action set. The embodiment of the present disclosure does not limit the number of selected actions. Optionally, the number of at least one action selected is equal to the preset width. Then, for any action, the server uses a large language model to process the question information corresponding to the root node according to the action to obtain the answer information corresponding to the action. Then, the server creates a node corresponding to the answer information and uses the node as a child node of the root node. Among them, the server can splice the question information and the action, input it into the large language model for processing, and output the answer information corresponding to the action. Then, the server can directly create the corresponding node based on the answer information. Alternatively, the server can also splice the answer information and the question information and then create the corresponding node. That is, except for the root node, each node in the target decoding tree can only store the corresponding answer information, or can also store the corresponding answer information and question information, etc., which is not limited by the embodiment of the present disclosure.

[0152] The solution provided by the embodiment of the present disclosure obtains at least one action from an action set in the process of creating a child node of the root node in the target decoding tree, directly instructs the large language model to perform a first round of processing on the question information according to the action specified by the user, and obtains at least one answer information obtained in the first round of processing, thereby ensuring that the generated answer information meets the needs of the model user, that is, ensuring the accuracy of the answer information; and can record the answer information corresponding to each action by creating a child node, so that the question information can be answered in more detail later on the basis of the existing answer information, that is, by interfering with the process of generating answer information by the large language model through the target decoding tree and the action set, the answer information required by the model user can be quickly obtained without model training, thereby achieving the purpose required by the user and improving the problem processing efficiency.

[0153] After creating the child nodes of the root node, the server can detect whether the target decoding tree meets the problem solving conditions. That is, after the first round of processing of the problem information, the server can detect whether the obtained answer information meets the requirements.

[0154] The problem solving conditions include at least one of the following:

[0155] First, the depth of the target decoding tree reaches a preset depth.

[0156] The second item is that the answer information corresponding to any leaf node in the target decoding tree is correct.

[0157] The solution provided by the embodiment of the present disclosure determines that the target decoding tree meets the problem-solving conditions when the depth of the target decoding tree reaches a preset depth, and no further processing is required, that is, the number of actions (steps) for processing the problem information is limited, avoiding the large language model from performing a large number of actions without limit, and ensuring the efficiency of the large language model in generating answer information. Alternatively, when the answer information corresponding to any leaf node in the target decoding tree is correct, it is determined that the target decoding tree meets the problem-solving conditions and no further processing is required, that is, no further processing is required when the correct answer information is obtained, avoiding the large language model from performing useless actions, and improving the efficiency and accuracy of the large language model in generating answer information.

[0158] The disclosed embodiment does not limit the method of determining whether the answer information is correct. It can be determined manually by the user of the large language model or automatically by the device.

[0159] In some embodiments, for any leaf node in the target decoding tree, in response to the correct confirmation operation of the current user account on the answer information corresponding to the leaf node, the server determines that the answer information corresponding to the leaf node is correct. That is, the server can transmit the answer information corresponding to multiple leaf nodes to the terminal of the current user account. The terminal displays the answer information corresponding to multiple leaf nodes. For the answer information of any leaf node, the terminal can display the confirmation control and the rejection control corresponding to the answer information. In response to the triggering operation of the confirmation control of the answer information of any leaf node, the terminal determines that the answer information is correct and sends a notification to the server so that the large language model can further process the question information. Alternatively, in response to the triggering operation of the rejection control of the answer information of any leaf node, the terminal determines that the answer information is not correct and sends a notification to the server so that the large language model can further process the question information.

[0160] In other embodiments, for any leaf node in the target decoding tree, the server determines the answer information corresponding to the leaf node and obtains a determination result, and the determination result is used to indicate whether the answer information of the leaf node is correct. The server may use any heuristic algorithm to determine the answer information. The heuristic algorithm may be an ant algorithm, a simulated annealing method, or a genetic algorithm, etc., which is not limited in the embodiments of the present disclosure.

[0161] The solution provided by the embodiment of the present disclosure can determine whether the answer information corresponding to the leaf node in the target decoding tree is correct by the current user account, which ensures that the generated answer information meets the needs of the model user, that is, the accuracy of the answer information is guaranteed. Alternatively, the answer information corresponding to the leaf node in the target decoding tree can also be judged to determine whether the answer information is accurate, which can not only ensure the accuracy of the answer information, but also does not require user operation, so that the target answer information of the large language model for the question information can be determined quickly from the answer information corresponding to multiple leaf nodes, thereby improving the processing efficiency of the question information.

[0162] In the disclosed embodiment, after creating the child nodes of the root node, if the target decoding tree does not meet the problem-solving conditions, the server continues to use the large language model to perform a new round of processing on the question information and create lower-level nodes. Please continue to refer to the specific content in step 304. The processing rounds of the question information correspond one-to-one to the levels of the nodes in the target decoding tree. If the target decoding tree meets the problem-solving conditions, the server controls the large language model to stop the processing of the question information and no longer creates lower-level nodes. After that, step 305 can be directly executed to determine the target answer information of the large language model for the question information. It can be seen that step 304 is an optional step.

[0163] In step 304, for any target node except the root node in the target decoding tree, the server selects at least one action from the action set; for any action in the at least one action, the large language model is used to process the answer information and question information corresponding to the target node according to the action, to obtain the answer information corresponding to the action, and to create a node corresponding to the answer information, and to determine the node corresponding to the answer information as a child node of the target node, until the target decoding tree meets the problem solving condition.

[0164] In the disclosed embodiment, during the non-first-round processing of the question information, the server still selects at least one action from the action set. The disclosed embodiment does not limit the number of selected actions. Then, for any action, the server uses a large language model to process the answer information and question information corresponding to the target node according to the action to obtain the answer information corresponding to the action. Then, the server creates a node corresponding to the answer information and uses the node as a child node of the target node.

[0165] Among them, the server can splice the answer information, question information and action corresponding to the target node, input them into the large language model for processing, and output the answer information corresponding to the action. Then, the server can splice the answer information corresponding to the action with the answer information corresponding to the target node, and create a corresponding node based on the spliced ​​answer information. That is, except for the root node, the answer information in each node in the target decoding tree is determined by a series of actions from the root node to the node. Correspondingly, the answer information corresponding to each leaf node in the target decoding tree is the answer information obtained as a whole from a series of actions in the link where the leaf node is located.

[0166] The solution provided by the embodiment of the present disclosure obtains at least one action from an action set during the process of creating a child node of any target node except the root node in the target decoding tree, and directly instructs the large language model to perform this round of processing on the answer information and question information of the target node according to the action specified by the user, so as to obtain at least one answer information obtained in this round of processing. This not only ensures that the generated answer information meets the needs of the model user, but also can be further refined on the basis of the existing answer information, thereby ensuring the accuracy of the answer information. That is, by interfering with the process of generating answer information by the large language model through the target decoding tree and the action set, the answer information required by the model user can be quickly obtained without model training, thereby achieving the purpose required by the user and improving the problem processing efficiency.

[0167] In the embodiment of the present disclosure, the server may randomly select at least one action from the action set to process the problem information, or the server may also selectively select at least one action based on a certain rule to process the problem information. The embodiment of the present disclosure does not limit the method of selecting the action. Four methods are described below by way of example, but are by no means limited thereto.

[0168] In the first way, the server selects at least one action from the action set based on the target action used when generating the answer information corresponding to the target node based on the parent node of the target node. At least one action is the same as or different from the target action. In other words, the server can select at least one action from the action set based on a target action used in the process of creating the target node. In other words, the server can determine the target action required in the current processing process (the current round of processing step) based on the action used in the previous round of processing (that is, the previous step).

[0169] The solution provided by the embodiment of the present disclosure is that the target action is the action adopted when generating the answer information corresponding to the target node based on the parent node of the target node, that is, the action adopted in the previous step before the current round of processing the question information. At least one action used for the current round of processing the question information is determined through the target action, so that each round of processing process is related, which is beneficial to ensure the accuracy of the answer information.

[0170] The server may obtain the target action used when generating the answer information corresponding to the target node based on the parent node of the target node; and select at least one action related to the target action from the action set. The action relatedness may be that the answer information corresponding to the target action is the information required when a certain action is executed, for example, if the target action is "explain keyword a", the action related to the target action may be "how to achieve the target process based on keyword a".

[0171] The solution provided by the embodiment of the present disclosure is that since the target action is the action adopted when generating the answer information corresponding to the target node based on the parent node of the target node, that is, it is the action adopted in the previous step before the current round of processing the question information, at least one action related to the target action is obtained as the action used for the current round of processing the question information, so that each round of processing process is related, which is beneficial to ensure the accuracy of the answer information.

[0172] In the second mode, the server selects at least one action from the action set based on the action used in the target link of the target decoding tree. The target link refers to the link from the root node to the target node. At least one action is the same as or different from the action used in the target link.

[0173] The solution provided by the embodiment of the present disclosure, since the actions adopted in the target link can accurately reflect the generation process of the answer information of the target node, at least one action used for this round of processing question information is determined through the actions in the target link, so that each round of processing process is associated with the previous processing process, which is beneficial to ensure the accuracy of the answer information.

[0174] In a third manner, the server selects at least one action from the action set based on the processing progress of the question information. The processing progress is used to indicate the degree to which the answer information of the target node solves the question information. The processing progress of the question information can be determined by the matching situation between the question information and the answer information, the information granularity of the answer information, and other information, which is not limited in the embodiments of the present disclosure.

[0175] The solution provided by the embodiment of the present disclosure can accurately reflect the degree to which the answer information of the target node solves the problem information, and determines at least one action used in this round of processing the problem information through the processing progress of the problem information, so that actions that can solve the remaining problems of the problem information can be obtained, thereby ensuring that the answer information generated subsequently can accurately solve the problem information, which is beneficial to improving the accuracy of the answer information.

[0176] In the fourth mode, the server determines the target number of child nodes that the target node can be assigned to in the hierarchy based on the preset width of the target decoding tree and the number of nodes in the hierarchy to which the target node belongs, and selects the target number of actions from the action set, where the preset width represents the maximum number of nodes in each hierarchy of the target decoding tree. For example, if the preset width of the target decoding tree is 8 and the number of nodes in the hierarchy to which the target node belongs is 4, then the target number of child nodes that the target node can be assigned to is 8 / 4=2.

[0177] The solution provided by the disclosed embodiment allocates a target number of actions (i.e., child nodes) to the target node through the preset width of the target decoding tree and the number of nodes in the layer to which the target node belongs, thereby ensuring that the width of the target decoding tree does not exceed the preset width. This is equivalent to limiting the amount of generated answer information on the basis of ensuring the diversity of answer information to a certain extent, thereby avoiding a large language model from generating too much answer information and causing data redundancy.

[0178] The above four methods can be freely combined to obtain a new method, which is not limited in the embodiments of the present disclosure.

[0179] After creating any layer of nodes in the target decoding, the server can detect whether the target decoding tree meets the problem-solving conditions. If the target decoding tree does not meet the problem-solving conditions, the server continues to use the large language model to perform a new round of processing on the question information and create lower-level nodes. That is, the server continues to execute step 304. If the target decoding tree meets the problem-solving conditions, the server controls the large language model to stop processing the question information and no longer creates lower-level nodes. Then, step 305 can be directly executed to determine the target answer information of the large language model for the question information.

[0180] In step 305, the server determines target answer information of the large language model for the question information from the answer information corresponding to the multiple leaf nodes in the target decoding tree.

[0181] In the disclosed embodiment, the server may directly regard the answer information corresponding to the multiple leaf nodes in the target decoding tree as the target answer information of the large language model for the question information. Alternatively, the server may also filter out at least one answer information from the answer information corresponding to the multiple leaf nodes in the target decoding tree, and regard the at least one answer information as the target answer information of the large language model for the question information.

[0182] The embodiment of the present disclosure does not limit the manner in which the server screens the target answer information. Optionally, the server determines at least one target answer information that meets the target condition from the answer information corresponding to multiple leaf nodes in the target decoding tree.

[0183] Among them, the target condition is that the number of occurrences of the target answer information reaches a preset number of times in the answer information corresponding to multiple leaf nodes. That is, the server regards the answer information that appears the preset number of times as the answer to the question information of the large language model. For example, the question information is "the calculation result of solving formula A". The answer information corresponding to multiple leaf nodes are "4, 4, 5, 5, 5, 5, 5, 4, 5, 3" respectively. Since the answer information "5" appears the most times, "5 is" the answer information of formula A. Alternatively, the target condition is that the target answer information is determined to be the answer to the question information. That is, the server can judge the multiple answer information separately according to the question information to determine whether each answer information is the answer to the question information. Among them, the server can regard the answer information with keywords (such as 'the answer is') in the answer information as the target answer information. Alternatively, the server can also input the question information and each answer information into the large language model, and identify whether each answer information is the answer to the question information through the large language model's ability to understand natural language. This is not limited in the embodiments of the present disclosure.

[0184] The solution provided by the embodiment of the present disclosure can more accurately reflect the processing of the question information by the large language model because the number of occurrences of the answer information reaches the preset number of times, and the number of occurrences is large. From the answer information corresponding to multiple leaf nodes in the target decoding tree, the answer information with the number of occurrences reaching the preset number of times is determined as the target answer information for the question information by the large language model, thereby ensuring the accuracy of the answer information. Alternatively, the answer information corresponding to the leaf nodes in the target decoding tree can also be determined to determine whether the answer information is the answer to the question information, which can also ensure the accuracy of the answer information.

[0185] In some embodiments, after the server filters out the target answer information, the server can further filter out answer information that is more consistent with the question information based on the match between the target answer information and the question information. Accordingly, for any target answer information, the server inputs the target answer information and the question information into the large language model. The server then determines the match between the target answer information and the question information through the large language model. The match is used to indicate the degree of accuracy of the target answer information in answering the question information. The server then selects the target answer information whose match meets the matching conditions from at least one target answer information. That is, the server can transmit the answer information output by the large language model back to the large language model for analysis, so that the large language model can analyze and feedback the quality of its own output results, so as to provide accurate answer information to the user.

[0186] The solution provided by the embodiment of the present disclosure can also determine the degree of match between the target answer information and the question information through the large language model for any target answer information obtained by the large language model for the question information, so as to further screen out qualified target answer information from at least one target answer information as the final answer to the question information, thereby ensuring the accuracy of the answer information.

[0187] In some embodiments, there may be multiple large language models. Multiple large language models can work together to jointly process question information in any business scenario. Among them, multiple large language models can communicate with each other and transmit the answer information obtained by the model itself for the question information to interact, so as to make full use of the advantages of different large language models. The answer information of multiple large language models draws on the strengths of others and is integrated and output, thereby improving the accuracy and robustness of problem processing. The following exemplifies three ways of jointly processing question information between multiple large language models, but is by no means limited to this.

[0188] In the first way, each of the multiple large language models is used to process the question information and judge the answer information of other large language models. Accordingly, the server uses at least one other large language model except the first large language model among the multiple large language models to judge the multiple answer information obtained by the first large language model to obtain the judgment results of the multiple answer information. The first large language model is any model among the multiple large language models. The multiple answer information is the answer information corresponding to the multiple leaf nodes in the target decoding tree of the first large language model. The judgment result of each answer information is used to indicate whether the answer information is correct. Then, the server determines the accuracy of the first large language model based on the judgment results of the multiple answer information. Then, the server determines the large language model with the highest accuracy from the multiple large language models, and determines the target answer information for the question information based on the multiple answer information of the large language model with the highest accuracy.

[0189] In the process of determining the target answer information for the question information based on the multiple answer information of the large language model with the highest accuracy, the server may further select the answer information that meets the condition from the multiple answer information of the large language model with the highest accuracy based on at least one of the target condition and the degree of matching between the answer information and the question information, as the target answer information for the question information. The disclosed embodiments are not limited to this.

[0190] The solution provided by the embodiment of the present disclosure may have multiple large language models, each of which can process question information to obtain corresponding answer information. The multiple large language models can integrate communication and transmit the multiple answer information obtained by each of them. The multiple answer information of the current large language model is judged by other large language models to determine whether the multiple answer information is correct, thereby determining the accuracy of the current large language model, and then determining the target answer information for the question information based on the multiple answer information of the large language model with the highest accuracy, which is beneficial to ensuring the reliability and objectivity of the answer information generated by the large language model, that is, it is beneficial to ensuring the accuracy of the answer information.

[0191] For example, Figure 4 FIG. 1 is a schematic diagram showing a method of jointly processing question information between multiple large language models according to an exemplary embodiment. Figure 4 , taking the large language model 1 as an example, the large language model 1 transmits the multiple answer information obtained by itself to the large language model 2 and the large language model 3. The large language model 2 judges the multiple answer information of the large language model 1 respectively, and obtains the judgment results of the multiple answer information. Then, based on the judgment results of the multiple answer information of the large language model 1, the large language model 2 determines that the accuracy of the large language model 1 is 0.9. Similarly, the large language model 3 determines that the accuracy of the large language model 1 is 0.8. Then, the server can average the accuracy of 0.9 obtained by the large language model 2 and the accuracy of 0.8 obtained by the large language model 3, and obtain the accuracy of the large language model 1 as 0.85. Similarly, the accuracy of the large language model 2 is 0.6, and the accuracy of the large language model 3 is 0.675. It can be seen that the accuracy of the large language model 1 is the highest. Then, the server uses the multiple answer information of the large language model 1 as the target answer information for the question information.

[0192] In some embodiments, the plurality of large language models include a plurality of second largest language models and a third largest language model. Each second largest language model is used to process question information. The third largest language model is used to determine the answer information of each second largest language model. Accordingly, for any second largest language model in the plurality of large language models, the server uses the third largest language model to determine the plurality of answer information obtained by the second largest language model to obtain determination results of the plurality of answer information. The plurality of answer information is the answer information corresponding to the plurality of leaf nodes in the target decoding tree of the second largest language model. The determination result of each answer information is used to indicate whether the answer information is correct. Then, based on the determination results of the plurality of answer information, the server filters out the correct answer information obtained by the second largest language model from the plurality of answer information. The server summarizes the correct answer information of the plurality of second largest language models to obtain the target answer information for the question information.

[0193] The solution provided by the embodiment of the present disclosure may have multiple large language models, and designate one large language model (the third largest language model) specifically for determining the correctness of the answers output by other large language models, and aggregate the correct answer information of each second largest language model as the target answer information for the question information. That is, through the model integration communication solution, the answer information of multiple large language models is combined and outputted to obtain the most appropriate answer, which is beneficial to ensuring the reliability and objectivity of the answer information generated by the large language model, that is, it is beneficial to ensuring the accuracy of the answer information.

[0194] For example, Figure 5 FIG. 1 is a schematic diagram showing another method of jointly processing question information between multiple large language models according to an exemplary embodiment. Figure 5 , large language model 1, large language model 2 and large language model 3 are the second large language model described above. Large language model A is the third large language model described above. Taking large language model 1 as an example, large language model 1 transmits the multiple answer information obtained by itself to large language model A. Large language model A judges the multiple answer information of large language model 1 respectively to obtain the judgment results of the multiple answer information. Then, based on the judgment results of the multiple answer information of large language model 1, large language model A filters out the correct answer information of large language model 1 from the multiple answer information. Similarly, large language model A filters out the correct answer information of large language model 2 and the correct answer information of large language model 3. Then, the server aggregates the correct answer information of each model of large language model 1, large language model 2 and large language model 3 to obtain the target answer information for the question information.

[0195] In some embodiments, each of the multiple large language models is used to process question information and determine the correctness of the answer information of the associated model. Accordingly, the server uses the fourth largest language model to determine the multiple answer information obtained by the fifth largest language model to obtain the determination results of the multiple answer information. The fourth largest language model is any model among the multiple large language models. The fifth largest language model is an associated model of the fourth largest language model. The multiple answer information is the answer information corresponding to multiple leaf nodes in the target decoding tree of the fifth largest language model, and the determination result of each answer information is used to indicate whether the answer information is correct. Then, based on the determination results of the multiple answer information, the server filters out the correct answer information output by the fifth largest language model from the multiple answer information. Then, the server summarizes the correct answer information of the multiple large language models to obtain the target answer information for the question information.

[0196] Among them, the association models of different large language models are different. The disclosed embodiment does not limit the method of determining the association models of each large language model. Optionally, the association models of each large language model can be manually set in advance, so that in the process of processing question information, the answer information of each large language model can be directly transmitted to the corresponding association model according to the pre-set association relationship to determine the correctness of the answer information. Alternatively, the server can also determine the association model of each large language model based on at least one of the performance, architecture and algorithm principle of each large language model.

[0197] The solution provided by the embodiment of the present disclosure may have multiple large language models, each of which can be used to determine the correctness of the answer output by the corresponding associated model, and the correct answer information of each large language model is summarized as the target answer information for the question information. That is, through the model integration communication solution, the results of multiple large language models are combined and output to obtain the most appropriate answer, which is beneficial to ensuring the reliability and objectivity of the answer information generated by the large language model, that is, it is beneficial to ensure the accuracy of the answer information.

[0198] For example, Figure 6 FIG. 1 is a schematic diagram showing another method of jointly processing question information between multiple large language models according to an exemplary embodiment. Figure 6 , large language model 1, large language model 2 and large language model 3 are the fourth large language model described above. Large language model 1 is an associated model of large language model 2, and large language model 2 can determine whether multiple answer information of large language model 1 is correct. Large language model 2 is an associated model of large language model 3, and large language model 3 can determine whether multiple answer information of large language model 2 is correct. Large language model 3 is an associated model of large language model 1, and large language model 1 can determine whether multiple answer information of large language model 3 is correct. Then, the correct answer information of each large language model is screened out. The server aggregates the correct answer information of each model of large language model 1, large language model 2 and large language model 3 to obtain the target answer information for the question information.

[0199] In order to more clearly describe the method for processing problem information provided by the embodiment of the present disclosure, the method is further described below in conjunction with the accompanying drawings. Figure 7 is a diagram showing an architecture for processing problem information according to an exemplary embodiment. Figure 7, the architecture for processing question information includes multiple large language models such as large language model 1, large language model 2, large language model 3, an integrated communication module, and a voting module. Taking large language model 1 as an example, the server can input the question information of the current user account into large language model 1 to initialize the target decoding tree. That is, set the hyperparameters (such as preset width, preset depth, etc.) of the target decoding tree, and create the root node of the target decoding tree based on the question information. Then, for any target node already in the target decoding tree, the server uses the large language model to process the information corresponding to the target node according to at least one action in the action set, obtain at least one answer information, and create at least one child node of the target node, which is equivalent to expanding the target decoding tree or can be regarded as updating the target decoding tree. Then, the server determines whether the target decoding tree meets the problem solving conditions, such as whether it reaches the preset depth, or whether the answer information corresponding to any leaf node in the target decoding tree is correct. When the target decoding tree does not meet the problem-solving conditions, the server continues to obtain at least one action from the action set, uses the large language model to perform a new round of processing on the question information, and creates lower-level nodes until the target decoding tree meets the problem-solving conditions. Then, the server transmits the multiple answer information obtained by each large language model to other large language models through the integrated communication module. Then, the server instructs each large language model to judge the answer information of other models through the voting module to obtain the judgment results of each answer information. Then, the server determines the final answer information for the question information based on the judgment results of each answer information.

[0200] The disclosed embodiment provides a method for processing question information. In the process of using a large language model to process question information, by creating a target decoding tree, the pre-trained large language model gradually processes the question information according to the action specified by the user, that is, the decoding path of the large language model to generate answer information is interfered by the target decoding tree and the action set. Since the actions in the action set are specified by the current user account, it is guaranteed that the generated answer information meets the needs of the model user, that is, the accuracy of the answer information is guaranteed; on the basis of ensuring that the answer information has a certain accuracy, there is no need to further train the pre-trained large language model for the business scenario to which the question information belongs, which not only saves the manpower cost and training resources required for training, but also enables the large language model to quickly achieve the purpose required by the user, improves the problem processing efficiency, and guarantees the versatility of the large language model, which can be applied to various business scenarios. In addition, in the process of processing question information, a variety of processing ideas can be provided for the large language model through the action set, ensuring that the answer information output by the large language model has diversity.

[0201] In addition, through the model integration communication solution, the answer information of multiple large language models is learned and integrated to obtain the most appropriate answer, which is conducive to ensuring the reliability and objectivity of the answer information generated by the large language model, that is, it is also conducive to ensuring the accuracy of the answer information.

[0202] All the above optional technical solutions can be arbitrarily combined to form optional embodiments of the present disclosure, and will not be described in detail here.

[0203] Figure 8 is a block diagram of a device for processing question information according to an exemplary embodiment. Figure 8 , the device comprises:

[0204] An acquisition unit 801 is configured to acquire a pre-trained large language model, question information belonging to any business scenario, and an action set associated with the question information, wherein the action in the action set is an action to be performed by the large language model specified by the current user account when processing the question information;

[0205] A creating unit 802 is configured to create a target decoding tree, where the target decoding tree includes a root node corresponding to the question information;

[0206] The creation unit 802 is further configured to execute, for any target node already in the target decoding tree, using the large language model, processing the information corresponding to the target node according to at least one action in the action set, obtaining at least one answer information, and creating at least one child node of the target node, wherein the at least one child node corresponds to the at least one answer information one by one, until the target decoding tree meets the problem solving condition;

[0207] The first determining unit 803 is configured to determine target answer information of the large language model for the question information from answer information corresponding to multiple leaf nodes in the target decoding tree.

[0208] In some embodiments, the target node is a root node;

[0209] The creation unit 802 is configured to execute randomly selecting at least one action from the action set; for any action of the at least one action, using a large language model, processing the question information corresponding to the root node according to the action, obtaining the answer information corresponding to the action, and creating a node corresponding to the answer information, and determining the node corresponding to the answer information as a child node of the root node.

[0210] In some embodiments, the target node is any node in the target decoding tree except the root node;

[0211] The creation unit 802 includes:

[0212] The acquisition subunit is configured to select at least one action from the action set;

[0213] A sub-unit is created, which is configured to execute any action of at least one action, use a large language model, process the answer information and question information corresponding to the target node according to the action, obtain the answer information corresponding to the action, create a node corresponding to the answer information, and determine the node corresponding to the answer information as a child node of the target node.

[0214] In some embodiments, the acquisition subunit is configured to perform at least one of the following:

[0215] Based on the target action used when generating answer information corresponding to the target node based on the parent node of the target node, at least one action is selected from the action set, and the at least one action is the same as or different from the target action;

[0216] Based on the actions adopted in the target link of the target decoding tree, at least one action is selected from the action set, where the target link refers to the link from the root node to the target node, and at least one action is the same as or different from the action adopted in the target link;

[0217] Based on the processing progress of the question information, at least one action is selected from the action set, and the processing progress is used to indicate the degree to which the answer information of the target node solves the question information;

[0218] Based on a preset width of the target decoding tree and the number of nodes in the level to which the target node belongs, determine the target number of child nodes to which the target node can be assigned in the level, and select the target number of actions from the action set, where the preset width represents the maximum number of nodes in each level of the target decoding tree.

[0219] In some embodiments, the acquisition subunit is configured to execute a target action adopted when generating answer information corresponding to the target node based on the parent node of the target node; and select at least one action related to the target action from the action set.

[0220] In some embodiments, the problem solving condition includes at least one of the following:

[0221] The depth of the target decoding tree reaches the preset depth;

[0222] The answer information corresponding to any leaf node in the target decoding tree is correct.

[0223] In some embodiments, the apparatus further includes a second determining unit configured to perform at least one of the following:

[0224] For any leaf node in the target decoding tree, in response to a correct confirmation operation of the current user account on the answer information corresponding to the leaf node, determining that the answer information corresponding to the leaf node is correct;

[0225] For any leaf node in the target decoding tree, the answer information corresponding to the leaf node is judged to obtain a judgment result, and the judgment result is used to indicate whether the answer information of the leaf node is correct.

[0226] In some embodiments, the first determining unit is configured to determine at least one target answer information satisfying a target condition from answer information corresponding to a plurality of leaf nodes in the target decoding tree;

[0227] Among them, the target condition is that the number of occurrences of the target answer information in the answer information corresponding to multiple leaf nodes reaches a preset number, or the target condition is that the target answer information is determined to be the answer to the question information.

[0228] In some embodiments, the apparatus further comprises:

[0229] The third determination unit is configured to execute, for any target answer information, input the target answer information and the question information into the large language model; determine the matching degree between the target answer information and the question information through the large language model, the matching degree being used to indicate the correct degree of the target answer information in answering the question information; and select the target answer information whose matching degree meets the matching condition from at least one target answer information.

[0230] In some embodiments, there are multiple large language models, and each of the multiple large language models is used to process question information and determine answer information of other large language models;

[0231] The device also includes:

[0232] The fourth determination unit is configured to execute the determination of multiple answer information obtained by the first large language model using at least one other large language model among the multiple large language models except the first large language model, and obtain determination results of the multiple answer information, wherein the first large language model is any model among the multiple large language models, and the multiple answer information is answer information corresponding to multiple leaf nodes in the target decoding tree of the first large language model, and the determination result of each answer information is used to indicate whether the answer information is correct; based on the determination results of the multiple answer information, determine the accuracy of the first large language model; determine the large language model with the highest accuracy from the multiple large language models, and determine the target answer information for the question information based on the multiple answer information of the large language model with the highest accuracy.

[0233] In some embodiments, there are multiple large language models, and the multiple large language models include multiple second-largest language models and one third-largest language model, each second-largest language model is used to process question information, and the third-largest language model is used to determine answer information of each second-largest language model;

[0234] The device also includes:

[0235] The fifth determination unit is configured to execute, for any second largest language model among the multiple large language models, using the third largest language model to determine multiple answer information obtained by the second largest language model, and obtain determination results of the multiple answer information, where the multiple answer information are answer information corresponding to multiple leaf nodes in the target decoding tree of the second largest language model, and the determination result of each answer information is used to indicate whether the answer information is correct; based on the determination results of the multiple answer information, filter out the correct answer information obtained by the second largest language model from the multiple answer information; summarize the correct answer information of the multiple second largest language models to obtain the target answer information for the question information.

[0236] In some embodiments, there are multiple large language models, and each of the multiple large language models is used to process question information and determine the correctness of answer information of the associated model;

[0237] The device also includes:

[0238] a sixth determination unit configured to execute a fourth language model to determine multiple answer information obtained by the fifth language model to obtain determination results of the multiple answer information, wherein the fourth language model is any model among the multiple language models, the fifth language model is an associated model of the fourth language model, and different large language models have different associated models, the multiple answer information is answer information corresponding to multiple leaf nodes in the target decoding tree of the fifth language model, and the determination result of each answer information is used to indicate whether the answer information is correct;

[0239] Based on the determination results of the multiple answer information, the correct answer information output by the fifth language model is selected from the multiple answer information;

[0240] The correct answer information of multiple large language models is aggregated to obtain the target answer information for the question information.

[0241] The disclosed embodiment provides a device for processing question information. In the process of using a large language model to process question information, by creating a target decoding tree, the pre-trained large language model gradually processes the question information according to the action specified by the user, that is, the decoding path of the large language model to generate answer information is interfered by the target decoding tree and the action set. Since the actions in the action set are specified by the current user account, it is guaranteed that the generated answer information meets the needs of the model user, that is, the accuracy of the answer information is guaranteed; on the basis of ensuring that the answer information has a certain accuracy, there is no need to further train the pre-trained large language model for the business scenario to which the question information belongs, which not only saves the manpower cost and training resources required for training, but also enables the large language model to quickly achieve the purpose required by the user, improves the problem processing efficiency, and guarantees the versatility of the large language model, which can be applied to various business scenarios. In addition, in the process of processing question information, a variety of processing ideas can be provided for the large language model through the action set, ensuring that the answer information output by the large language model has diversity.

[0242] It should be noted that the apparatus for processing problem information provided in the above embodiment only uses the division of the above functional units as an example when processing problem information in any business scenario. In actual applications, the above functions can be assigned to different functional units as needed, that is, the internal structure of the electronic device is divided into different functional units to complete all or part of the functions described above. In addition, the apparatus for processing problem information provided in the above embodiment and the embodiment of the method for processing problem information belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0243] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0244] When an electronic device is provided as a terminal, Fig. 9 FIG. 1 is a block diagram of a terminal 900 according to an exemplary embodiment. Fig. 9 The structure block diagram of a terminal 900 provided by an exemplary embodiment of the present disclosure is shown. The terminal 900 may be: a smart phone, a tablet computer, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV), a laptop computer or a desktop computer. The terminal 900 may also be called a user device, a portable terminal, a laptop terminal, a desktop terminal or other names.

[0245] Typically, the terminal 900 includes a processor 901 and a memory 902 .

[0246] The processor 901 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 901 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 901 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 901 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 901 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0247] The memory 902 may include one or more computer-readable storage media, which may be non-transitory. The memory 902 may also include a high-speed random access memory and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 902 is used to store at least one computer program, which is used to be executed by the processor 901 to implement the method for processing problem information provided in the method embodiment of the present application.

[0248] In some embodiments, the terminal 900 may further optionally include: a peripheral device interface 903 and at least one peripheral device. The processor 901, the memory 902 and the peripheral device interface 903 may be connected via a bus or a signal line. Each peripheral device may be connected to the peripheral device interface 903 via a bus, a signal line or a circuit board. Specifically, the peripheral device includes: at least one of a radio frequency circuit 904, a display screen 905, a camera assembly 906, an audio circuit 907 and a power supply 908.

[0249] The peripheral device interface 903 may be used to connect at least one peripheral device related to I / O (Input / Output) to the processor 901 and the memory 902. In some embodiments, the processor 901, the memory 902, and the peripheral device interface 903 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 901, the memory 902, and the peripheral device interface 903 may be implemented on a separate chip or circuit board, which is not limited in this embodiment.

[0250] The radio frequency circuit 904 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The radio frequency circuit 904 communicates with the communication network and other communication devices through electromagnetic signals. The radio frequency circuit 904 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals into electrical signals. In some embodiments, the radio frequency circuit 904 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, and the like. The radio frequency circuit 904 can communicate with other terminals through at least one wireless communication protocol. The wireless communication protocol includes, but is not limited to: the World Wide Web, a metropolitan area network, an intranet, various generations of mobile communication networks (2G, 3G, 4G and 5G), a wireless local area network and / or a WiFi (Wireless Fidelity) network. In some embodiments, the radio frequency circuit 904 may also include circuits related to NFC (Near Field Communication), which is not limited in this application.

[0251] The display screen 905 is used to display the UI (User Interface). The UI may include graphics, text, icons, videos, and any combination thereof. When the display screen 905 is a touch display screen, the display screen 905 also has the ability to collect touch signals on the surface or above the surface of the display screen 905. The touch signal can be input to the processor 901 as a control signal for processing. At this time, the display screen 905 can also be used to provide virtual buttons and / or virtual keyboards, also known as soft buttons and / or soft keyboards. In some embodiments, the display screen 905 can be one, set on the front panel of the terminal 900; in other embodiments, the display screen 905 can be at least two, respectively set on different surfaces of the terminal 900 or in a folding design; in other embodiments, the display screen 905 can be a flexible display screen, set on the curved surface or folding surface of the terminal 900. Even, the display screen 905 can also be set to a non-rectangular irregular shape, that is, a special-shaped screen. The display screen 905 can be made of materials such as LCD (Liquid Crystal Display), OLED (Organic Light-Emitting Diode, organic light-emitting diode).

[0252] The camera assembly 906 is used to capture images or videos. In some embodiments, the camera assembly 906 includes a front camera and a rear camera. Typically, the front camera is disposed on the front panel of the terminal, and the rear camera is disposed on the back of the terminal. In some embodiments, there are at least two rear cameras, which are any one of a main camera, a depth of field camera, a wide-angle camera, and a telephoto camera, so as to realize the fusion of the main camera and the depth of field camera to realize the background blur function, the fusion of the main camera and the wide-angle camera to realize panoramic shooting and VR (Virtual Reality) shooting function or other fusion shooting functions. In some embodiments, the camera assembly 906 may also include a flash. The flash may be a monochrome temperature flash or a dual-color temperature flash. A dual-color temperature flash refers to a combination of a warm light flash and a cold light flash, which can be used for light compensation at different color temperatures.

[0253] The audio circuit 907 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, and convert the sound waves into electrical signals and input them into the processor 901 for processing, or input them into the radio frequency circuit 904 to achieve voice communication. For the purpose of stereo acquisition or noise reduction, there may be multiple microphones, which are respectively arranged at different parts of the terminal 900. The microphone may also be an array microphone or an omnidirectional acquisition microphone. The speaker is used to convert the electrical signal from the processor 901 or the radio frequency circuit 904 into sound waves. The speaker may be a traditional film speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can not only convert the electrical signal into sound waves audible to humans, but also convert the electrical signal into sound waves inaudible to humans for purposes such as ranging. In some embodiments, the audio circuit 907 may also include a headphone jack.

[0254] The power supply 908 is used to power various components in the terminal 900. The power supply 908 can be an alternating current, a direct current, a disposable battery, or a rechargeable battery. When the power supply 908 includes a rechargeable battery, the rechargeable battery can be a wired rechargeable battery or a wireless rechargeable battery. A wired rechargeable battery is a battery charged through a wired line, and a wireless rechargeable battery is a battery charged through a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0255] Those skilled in the art will understand that Fig. 9 The structure shown in the figure does not constitute a limitation on the terminal 900, and the terminal 900 may include more or less components than those shown in the figure, or combine some components, or adopt a different component arrangement.

[0256] When the electronic device is provided as a server, Fig.10 This is a block diagram of a server 1000 according to an exemplary embodiment. The server 1000 may have relatively large differences due to different configurations or performances, and may include one or more processors (Central Processing Units, CPU) 1001 and one or more memories 1002, wherein the memory 1002 stores at least one program code, and the at least one program code is loaded and executed by the processor 1001 to implement the problem information processing method provided by the above-mentioned various method embodiments. Of course, the server may also have components such as a wired or wireless network interface, a keyboard, and an input and output interface for input and output. The server 1000 may also include other components for implementing device functions, which will not be described in detail here.

[0257] In an exemplary embodiment, a computer-readable storage medium including instructions is also provided, such as a memory 902 or a memory 1002 including instructions, and the above instructions can be executed by the processor 901 of the terminal 900 or the processor 1001 of the server 1000 to complete the above-mentioned problem information processing method. Optionally, the computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0258] A computer program product includes a computer program / instruction, which implements the above-mentioned method for processing problem information when executed by a processor.

[0259] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0260] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A method for processing problem information, characterized in that: The method comprises: Obtaining a pre-trained large language model, question information belonging to any business scenario, and an action set associated with the question information, where the actions in the action set are actions that the current user account specifies the large language model to perform when processing the question information; Creating a target decoding tree, wherein the target decoding tree includes a root node corresponding to the question information; For any existing target node in the target decoding tree, the large language model is used to process the information corresponding to the target node according to at least one action in the action set, to obtain at least one answer information, and at least one child node of the target node is created, wherein the at least one child node corresponds to the at least one answer information in a one-to-one manner, until the target decoding tree meets the problem solving condition; Determine target answer information of the large language model for the question information from answer information corresponding to multiple leaf nodes in the target decoding tree.

2. The method for processing problem information according to claim 1, characterized in that: The target node is the root node; For any target node already in the target decoding tree, using the large language model, respectively processing information corresponding to the target node according to at least one action in the action set, obtaining at least one answer information, and creating at least one child node of the target node, including: Randomly selecting at least one action from the action set; For any of the at least one action, the large language model is used to process the question information corresponding to the root node according to the action to obtain answer information corresponding to the action, and a node corresponding to the answer information is created, and the node corresponding to the answer information is determined as a child node of the root node.

3. The method for processing problem information according to claim 1, characterized in that: The target node is any node in the target decoding tree except the root node; For any target node already in the target decoding tree, using the large language model, respectively processing information corresponding to the target node according to at least one action in the action set, obtaining at least one answer information, and creating at least one child node of the target node, including: Selecting at least one action from the action set; For any of the at least one action, the large language model is used to process the answer information and the question information corresponding to the target node according to the action to obtain the answer information corresponding to the action, and a node corresponding to the answer information is created, and the node corresponding to the answer information is determined as a child node of the target node.

4. The method for processing problem information according to claim 3, characterized in that: The selecting at least one action from the action set includes at least one of the following: Based on the target action adopted when generating the answer information corresponding to the target node based on the parent node of the target node, selecting at least one action from the action set, wherein the at least one action is the same as or different from the target action; Based on the action adopted in the target link of the target decoding tree, selecting the at least one action from the action set, the target link refers to the link from the root node to the target node, and the at least one action is the same as or different from the action adopted in the target link; Selecting the at least one action from the action set based on the processing progress of the question information, wherein the processing progress is used to indicate the degree to which the answer information of the target node solves the question information; Based on a preset width of the target decoding tree and the number of nodes in the level to which the target node belongs, determine the target number of child nodes to which the target node can be assigned in the level, and select the target number of actions from the action set, wherein the preset width represents the maximum number of nodes in each level of the target decoding tree.

5. The method for processing problem information according to claim 4, characterized in that: The target action adopted when generating the answer information corresponding to the target node based on the parent node of the target node, selecting the at least one action from the action set, comprises: Obtaining a target action used when generating answer information corresponding to the target node based on the parent node of the target node; At least one action related to the target action is selected from the action set.

6. The method for processing problem information according to claim 1, characterized in that: The problem-solving condition includes at least one of the following: The depth of the target decoding tree reaches a preset depth; The answer information corresponding to any leaf node in the target decoding tree is correct.

7. The method for processing problem information according to claim 6, characterized in that: The method further comprises at least one of the following: For any leaf node in the target decoding tree, in response to the current user account confirming the correctness of the answer information corresponding to the leaf node, determining that the answer information corresponding to the leaf node is correct; For any leaf node in the target decoding tree, the answer information corresponding to the leaf node is judged to obtain a judgment result, and the judgment result is used to indicate whether the answer information of the leaf node is correct.

8. The method for processing problem information according to claim 1, characterized in that: The step of determining target answer information of the large language model for the question information from answer information corresponding to a plurality of leaf nodes in the target decoding tree comprises: Determine at least one target answer information satisfying a target condition from the answer information corresponding to the plurality of leaf nodes in the target decoding tree; The target condition is that the target answer information appears a preset number of times in the answer information corresponding to the multiple leaf nodes, or the target condition is that the target answer information is determined to be the answer to the question information.

9. The method for processing problem information according to claim 8, characterized in that: The method further comprises: For any target answer information, input the target answer information and the question information into the large language model; Determining the matching degree between the target answer information and the question information by using the large language model, wherein the matching degree is used to indicate the correctness of the target answer information in answering the question information; The target answer information whose matching degree satisfies the matching condition is selected from the at least one target answer information.

10. The method for processing problem information according to claim 1, characterized in that: There are multiple large language models, and each of the multiple large language models is used to process the question information and determine the answer information of other large language models; The method further comprises: Using at least one other large language model among the multiple large language models except the first large language model, to determine multiple answer information obtained by the first large language model to obtain determination results of the multiple answer information, wherein the first large language model is any one of the multiple large language models, the multiple answer information is answer information corresponding to multiple leaf nodes in a target decoding tree of the first large language model, and the determination result of each answer information is used to indicate whether the answer information is correct; Determining the accuracy of the first language model based on the determination results of the multiple answer information; A large language model with the highest accuracy is determined from the plurality of large language models, and target answer information for the question information is determined based on a plurality of answer information of the large language model with the highest accuracy.

11. The method for processing problem information according to claim 1, characterized in that: There are multiple large language models, the multiple large language models include multiple second large language models and one third large language model, each of the second large language models is used to process the question information, and the third large language model is used to determine the answer information of each of the second large language models; The method further comprises: For any second largest language model among the multiple large language models, using the third largest language model, determining multiple answer information obtained by the second largest language model to obtain determination results of the multiple answer information, wherein the multiple answer information is answer information corresponding to multiple leaf nodes in the target decoding tree of the second largest language model, and the determination result of each answer information is used to indicate whether the answer information is correct; Based on the determination results of the multiple answer information, filter out the correct answer information obtained by the second largest language model from the multiple answer information; The correct answer information of the plurality of second largest language models is aggregated to obtain target answer information for the question information.

12. The method for processing problem information according to claim 1, characterized in that: There are multiple large language models, and each of the multiple large language models is used to process the question information and determine the correctness of the answer information of the associated model; The method further comprises: Using a fourth language model, judging multiple answer information obtained by a fifth language model to obtain judgment results of the multiple answer information, wherein the fourth language model is any model among the multiple large language models, the fifth language model is an associated model of the fourth language model, and different large language models have different associated models, the multiple answer information is answer information corresponding to multiple leaf nodes in a target decoding tree of the fifth language model, and the judgment result of each answer information is used to indicate whether the answer information is correct; Based on the determination results of the multiple answer information, the correct answer information output by the fifth language model is screened out from the multiple answer information; The correct answer information of the multiple large language models is aggregated to obtain target answer information for the question information.

13. A device for processing question information, characterized in that: The device comprises: An acquisition unit is configured to acquire a pre-trained large language model, question information belonging to any business scenario, and an action set associated with the question information, wherein the actions in the action set are actions that the current user account specifies the large language model to perform when processing the question information; A creating unit, configured to create a target decoding tree, wherein the target decoding tree includes a root node corresponding to the question information; The creation unit is further configured to execute, for any target node already in the target decoding tree, using the large language model, processing the information corresponding to the target node according to at least one action in the action set, obtaining at least one answer information, and creating at least one child node of the target node, wherein the at least one child node corresponds to the at least one answer information in a one-to-one manner, until the target decoding tree meets the problem solving condition; The first determination unit is configured to determine target answer information of the large language model for the question information from answer information corresponding to multiple leaf nodes in the target decoding tree.

14. An electronic device, characterized in that: The electronic device comprises: one or more processors; a memory for storing program code executable by the processor; The processor is configured to execute the program code to implement the method for processing question information as claimed in any one of claims 1 to 12.

15. A computer-readable storage medium, characterized in that: When the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the method for processing question information as described in any one of claims 1 to 12.

16. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method for processing question information as claimed in any one of claims 1 to 12 is implemented.

Images