Digital human question and answer method and device, electronic equipment and storage medium

Abstract

The invention discloses a digital human question answering method and device, electronic equipment and a storage medium, which are used for solving the technical problem that an existing digital human question answering method is limited to a single model and cannot meet different question answering requirements, so that the answering effect is relatively poor. The method comprises the following steps: receiving question data of a user; identifying the question classification and the corresponding difficulty level of the question data; determining a processing model according to the question classification and the difficulty level; and inputting the question data into the processing model, and outputting an answer to the question data. According to the method, classification and difficulty grade division are performed on the question data through intention recognition, and the corresponding processing model is selected according to question classification and difficulty grades, so that flexible adaptation of the processing model is realized, and the question and answer effect is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent question-answering technology, and in particular to a digital human question-answering method, device, electronic equipment and storage medium. Background Art

[0002] In modern dialogue systems, large language models have powerful generation and understanding capabilities. However, in real-time dialogue scenarios, especially when applied to digital human all-in-one machines, speed is a key factor. Although traditional large language models have powerful understanding and generation capabilities, they have high computing resource requirements and slow response speeds. Especially in real-time dialogue scenarios, they cannot provide the instant feedback that users expect, which may lead to a poor dialogue experience. In addition, the continuous calling of large models requires a lot of computing resources and storage space. Especially in embedded devices (such as digital human all-in-one machines) or low-resource environments, the frequent calling of large models will significantly affect the performance of the device and user experience. Lightweight small models are efficient in handling simple tasks, but their ability to handle complex problems is limited, and they are prone to misunderstandings or inaccurate responses. Summary of the invention

[0003] The present invention provides a digital human question-and-answer method, device, electronic device and storage medium, which are used to solve the technical problem that the existing digital human question-and-answer method is limited to a single model and cannot meet different question-and-answer requirements, resulting in poor answering effect.

[0004] The present invention provides a digital human question-answering method, comprising:

[0005] Receive user's question data;

[0006] Identify a question classification and a corresponding difficulty level of the question data;

[0007] Determining a processing model according to the problem classification and the difficulty level;

[0008] The question data is input into the processing model, and the answer to the question data is output.

[0009] Optionally, the step of identifying the question classification and the corresponding difficulty level of the question data includes:

[0010] Performing intent recognition on the problem data to obtain problem classification of the problem data;

[0011] The user's historical conversation data is obtained, and the difficulty level of the question data is determined according to the user's historical conversation data and / or the question classification.

[0012] Optionally, before the step of determining a processing model according to the problem classification and the difficulty level, the step further includes:

[0013] Retrieving the problem data in a preset cache;

[0014] When there is a question in the cache whose similarity to the question data reaches a preset threshold, the answer corresponding to the question is output as the answer to the question data.

[0015] Optionally, the step of inputting the question data into the processing model and outputting the answer to the question data comprises:

[0016] When the difficulty level is lower than a preset level, determining the processing model to be a small model;

[0017] The question data is input into the small model, and the answer to the question data is output.

[0018] Optionally, the step of inputting the question data into the processing model and outputting the answer to the question data comprises:

[0019] When the difficulty level is not less than a preset level, determining the processing model to be a small model and a large language model;

[0020] Inputting the question data into the small model and the large language model;

[0021] When it is detected that the small model outputs an answer guide in response to the question data and the large language model does not generate an output result, output the answer guide;

[0022] When the large language model generates an answer to the question data in response to the question data, the answer to the question data is output.

[0023] Optionally, the question categories include: business consultation, daily chat, instruction execution, and common sense questions.

[0024] The present invention also provides a digital human question-answering device, comprising:

[0025] A question data receiving module, used for receiving question data from users;

[0026] An identification module, used to identify the problem classification and corresponding difficulty level of the problem data;

[0027] A processing model determination module, used to determine a processing model according to the problem classification and the difficulty level;

[0028] The first answer output module is used to input the question data into the processing model and output the answer to the question data.

[0029] Optionally, the identification module includes:

[0030] An intention recognition submodule, used to perform intention recognition on the question data to obtain a question classification of the question data;

[0031] The difficulty level determination submodule is used to obtain user historical conversation data and determine the difficulty level of the question data according to the user historical conversation data and / or the question classification.

[0032] The present invention also provides an electronic device, the device comprising a processor and a memory:

[0033] The memory is used to store program code and transmit the program code to the processor;

[0034] The processor is used to execute the digital human question-answering method as described in any one of the above items according to the instructions in the program code.

[0035] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store program code, and the program code is used to execute the digital human question-answering method as described in any one of the above items.

[0036] It can be seen from the above technical solutions that the present invention has the following advantages: the present invention discloses a digital human question-and-answer method, and specifically discloses: receiving user question data; identifying the question classification and corresponding difficulty level of the question data; determining a processing model according to the question classification and the difficulty level; inputting the question data into the processing model, and outputting the answer to the question data. The present invention classifies question data and divides the difficulty level through intention recognition, so as to select the corresponding processing model according to the question classification and difficulty level, thereby realizing flexible adaptation of the processing model and improving the question-and-answer effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0038] Figure 1 A step flow diagram of a digital human question-answering method provided by an embodiment of the present invention;

[0039] Figure 2 A flowchart of a digital human question-answering method provided by another embodiment of the present invention;

[0040] Figure 3 A flowchart of a digital human question-answering method provided by an embodiment of the present invention;

[0041] Figure 4 A structural block diagram of a digital human question-and-answer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The embodiments of the present invention provide a digital human question-and-answer method, device, electronic device, and storage medium, which are used to solve the technical problem that the existing digital human question-and-answer method is limited to a single model and cannot meet different question-and-answer requirements, resulting in poor answering effect.

[0043] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] See also Figure 1 , Figure 1 A step flow chart of a digital human question-answering method provided by an embodiment of the present invention.

[0045] The digital human question-answering method provided by the present invention may specifically include the following steps:

[0046] Step 101, receiving user's question data;

[0047] In the embodiment of the present invention, the user can input question data via text or voice.

[0048] Step 102, identifying the question classification and corresponding difficulty level of the question data;

[0049] In an embodiment of the present invention, the problem classification and difficulty level of the user's problem data can be identified to select an appropriate response strategy. Among them, the problem classification can be set according to the user's common user type, such as business consultation, daily chat, instruction execution, common sense questions, etc. The difficulty level corresponding to the problem data can also be freely set according to its processing complexity. In an example, the difficulty level can be divided into 0 and 1, corresponding to two different processing logics.

[0050] Step 103, determining a processing model according to the problem classification and the difficulty level;

[0051] Step 104, input the question data into the processing model, and output the answer to the question data.

[0052] In the embodiment of the present invention, different problem categories and difficulty levels may correspond to different processing models. For example, a processing model with a fast processing flow may be selected for simple problems, and a model with a higher accuracy rate may be selected for complex problems.

[0053] After the corresponding processing model is determined, the processing model can be used to analyze the problem data and obtain the corresponding answer for output.

[0054] The present invention classifies question data and divides the difficulty levels through intent recognition, so as to select the corresponding processing model according to the question classification and difficulty level, thereby achieving flexible adaptation of the processing model and improving the question-answering efficiency.

[0055] See also Figure 2 , Figure 2 This is a flowchart of a digital human question-answering method provided by another embodiment of the present invention. Specifically, it may include the following steps:

[0056] Step 201, receiving user's question data;

[0057] In the embodiment of the present invention, the user can input question data via text or voice.

[0058] Step 202, performing intent recognition on the question data to obtain question classification of the question data;

[0059] Intent recognition refers to a technical task that automatically identifies the user's intention or purpose through natural language text. The accuracy of intent recognition can greatly affect the accuracy of search and the intelligence of the dialogue system. In the fields of human-computer interaction, speech recognition, natural language processing, etc., intent recognition plays a vital role.

[0060] In the embodiment of the present invention, the question data may be subjected to intent recognition to obtain question classification of the question data.

[0061] In the specific implementation, the question classification of the user's historical conversation data can be manually annotated and fine-tuned using a pre-trained model, such as Bert or a smaller-scale large language model. Then the user's historical conversation data and current question data are input into the pre-trained model to determine the question classification of the current question data.

[0062] Step 203, obtaining user historical conversation data, and determining the difficulty level of the question data according to the user historical conversation data and / or the question classification;

[0063] The user's historical conversation data is the user's previous question and answer data, including the user's historical question data and the corresponding question classification, difficulty level and answer.

[0064] In the embodiment of the present invention, the difficulty level of the question data may be determined according to the question classification of the question data and / or the user's historical conversation data.

[0065] In one example, the difficulty level can be divided directly according to the intent classification. For example, problems related to command execution and rule-based dialogue (weather query, daily chat, etc.) can be classified as simple difficulty problems, and complex problems that require querying the knowledge base, such as business problems or problems that require multi-step reasoning, can be classified as complex difficulty problems.

[0066] In another example, sometimes the same question classification may have different difficulty levels. For example, for business consulting questions, simple questions can be answered directly by querying the knowledge base. For complex questions, such as those containing a large number of professional terms or rare words, more rounds of dialogue are needed to clarify the questions and collect user information before an accurate answer can be given. Therefore, after completing the problem classification of the question data, the difficulty level of the question data can also be determined based on the user's historical conversation data. For example, an intent recognition model whose output result is the difficulty level is trained, and the user's historical conversation data is input into the trained intent recognition model to predict the difficulty level of the question data. Alternatively, a difficulty judgment model is trained, and its input data is the user's historical conversation data and the current question data.

[0067] In another example, the difficulty level of the question data may also be determined directly through the user's historical conversation data and the current question data.

[0068] It should be noted that, in the embodiment of the present invention, different numbers of difficulty levels can be set according to the number of processing logics set, so as to respond to the question data more specifically. Among them, those skilled in the art can subdivide the difficulty levels according to actual conditions, and the embodiment of the present invention does not specifically limit this.

[0069] Step 204, retrieving the problem data in a preset cache;

[0070] Step 205, when there is a question in the cache whose similarity with the question data reaches a preset threshold, outputting the answer corresponding to the question as the answer to the question data;

[0071] In the embodiment of the present invention, for repetitive questions or dialogue scenarios, the generated responses can be saved in the cache. When a question similar to the question data is retrieved in the cache, the answer corresponding to the question in the cache can be directly returned without re-calling the model for generation.

[0072] In a specific implementation, the data similarity judgment of question data can be performed by semantically vectorizing the question data and quickly finding the most similar questions from the cache through vector retrieval.

[0073] Furthermore, in order to improve the matching accuracy, a rerank model can be trained to more accurately calculate the similarity between the question and the question in the cache.

[0074] The preset threshold value may be set according to actual needs, and the embodiment of the present invention does not specifically limit this.

[0075] Step 206, determining a processing model according to the problem classification and the difficulty level;

[0076] Step 207: input the question data into the processing model and output the answer to the question data.

[0077] In the embodiment of the present invention, different problem categories and difficulty levels may correspond to different processing models. For example, a processing model with a fast processing flow may be selected for simple problems, and a model with a higher accuracy rate may be selected for complex problems.

[0078] After the corresponding processing model is determined, the processing model can be used to analyze the problem data and obtain the corresponding answer for output.

[0079] In one example, the step of inputting the question data into the processing model and outputting the answer to the question data may specifically include the following sub-steps:

[0080] S11, when the difficulty level is lower than a preset level, determining the processing model to be a small model;

[0081] S12, inputting the question data into the small model, and outputting the answer to the question data.

[0082] Small models refer to AI models with fewer parameters and data. These AI models are usually suitable for edge devices and low-resource environments, and can achieve efficient computing and reasoning capabilities.

[0083] In an embodiment of the present invention, when the difficulty level of the question data is less than a preset level, the problem is characterized by a low processing complexity, and a lightweight small model can be used to respond, such as instruction execution, regularized dialogue, and other issues.

[0084] In one example, question-answer pairs can be collected and fine-tuned based on a large open source language model with a smaller parameter scale to obtain the small model required by the present invention. For example, the model qwen2.5-7b is selected and fine-tuned using SFT (supervised fine-tuning), DPO (direct preference alignment), etc.

[0085] By using small models to handle simple problems, we can respond to problems quickly and enhance user experience.

[0086] In another example, the step of inputting the question data into the processing model and outputting the answer to the question data may specifically include the following sub-steps:

[0087] S21, when the difficulty level is not less than a preset level, determining that the processing model is a small model and a large language model;

[0088] S22, inputting the question data into the small model and the large language model;

[0089] S23, when it is detected that the small model outputs an answer guide in response to the question data and the large language model does not generate an output result, output the answer guide;

[0090] S24, when the large language model generates an answer to the question data in response to the question data, outputs the answer to the question data.

[0091] Large Language Model (LLM) refers to a deep learning model trained with a large amount of text data that can generate natural language text or understand the meaning of language text. Large language models usually have more parameters and complex structures and can handle higher-level tasks.

[0092] In the specific implementation, when the difficulty level is not less than the preset level, the problem data is characterized as a complex problem. At this time, the user experience can be optimized through a parallel mechanism. The small model and the large language model are called synchronously, and the small model is allowed to generate the answer guide first. For example, the user asks: "I want to know the process of applying for a passport"; the answer guide generated by the small model can be "Please wait, we are helping you find the passport application process". Before the large language model finds the output answer, the answer guide is output to the user first to ensure that the user gets a preliminary and quick response while waiting for the large language model to be generated.

[0093] Furthermore, after completing the response to the question data, the question data and the corresponding answer may be cached in a preset cache to facilitate subsequent rapid response to similar questions.

[0094] The present invention classifies question data and divides the difficulty levels through intention recognition, so as to select the corresponding processing model according to the question classification and difficulty level, thereby achieving flexible adaptation of the processing model and improving the question-answering efficiency.

[0095] For ease of understanding, the embodiments of the present invention are described below by using specific examples:

[0096] Figure 3 A flowchart of a digital human question-answering method provided by an embodiment of the present invention; Figure 3 As shown in the figure, the user first inputs the question through text or voice. Then, the user's current intent is identified through intent recognition, the intent is classified, and the difficulty level of the question is identified. Then similar questions are searched in the cache. When similar questions are found, the answers corresponding to the similar questions are output to achieve a quick response to the questions; when different similar questions are searched in the cache, tasks are allocated between models according to the question classification and the difficulty level of the questions: when the question is not a complex question, the small model is used to generate the corresponding answer; when the question is a complex question, the small model and the large language model are called synchronously, and the small model is first used to generate the answer guide to ensure that the user gets a preliminary quick response while waiting for the large language model to generate the answer, and when the large language model generates the answer, the corresponding answer is replied to the user.

[0097] See also Figure 4 , Figure 4 A structural block diagram of a digital human question-and-answer device provided in an embodiment of the present invention.

[0098] The embodiment of the present invention provides a digital human question-answering device, comprising:

[0099] The question data receiving module 401 is used to receive the question data of the user;

[0100] An identification module 402, for identifying the problem classification and the corresponding difficulty level of the problem data;

[0101] A processing model determination module 403, for determining a processing model according to the problem classification and the difficulty level;

[0102] The first answer output module 404 is used to input the question data into the processing model and output the answer to the question data.

[0103] In the embodiment of the present invention, the identification module 402 includes:

[0104] An intention recognition submodule, used to perform intention recognition on the question data to obtain a question classification of the question data;

[0105] The difficulty level determination submodule is used to obtain user historical conversation data and determine the difficulty level of the question data according to the user historical conversation data and / or the question classification.

[0106] In an embodiment of the present invention, it also includes:

[0107] A cache retrieval module, used to retrieve the problem data in a preset cache;

[0108] The second answer output module is used to output the answer corresponding to the question as the answer to the question data when there is a question in the cache whose similarity with the question data reaches a preset threshold.

[0109] The first answer output module 404 includes:

[0110] A small model determination submodule, used for determining that the processing model is a small model when the difficulty level is less than a preset level;

[0111] The first answer output submodule is used to input the question data into the small model and output the answer to the question data.

[0112] The first answer output module 404 includes:

[0113] A small model and a large language model determination submodule, used to determine the processing model as a small model and a large language model when the difficulty level is not less than a preset level;

[0114] A question data input submodule, used for inputting the question data into the small model and the large language model;

[0115] An answer guide output submodule, configured to output the answer guide when detecting that the small model outputs the answer guide in response to the question data and the large language model does not generate an output result;

[0116] The second answer output submodule is used to output the answer to the question data when the large language model generates the answer to the question data in response to the question data.

[0117] In the embodiment of the present invention, the problem classification includes: business consultation, daily chat, instruction execution, and common sense questions.

[0118] An embodiment of the present invention further provides an electronic device, the device comprising a processor and a memory:

[0119] The memory is used to store program code and transmit the program code to the processor;

[0120] The processor is used to execute the digital human question-answering method described in the embodiment of the present invention according to the instructions in the program code.

[0121] The embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store program codes, and the program codes are used to execute the digital human question-answering method described in the embodiment of the present invention.

[0122] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0123] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0124] It will be appreciated by those skilled in the art that the embodiments of the present invention may be provided as methods, devices, or computer program products. Therefore, the embodiments of the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0125] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0126] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0127] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1The steps for the functions specified in one or more boxes.

[0128] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0129] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0130] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A digital human question-answering method, characterized in that: include: Receive user's question data; Identify a question classification and a corresponding difficulty level of the question data; Determining a processing model according to the problem classification and the difficulty level; The question data is input into the processing model, and the answer to the question data is output.

2. The method according to claim 1, characterized in that The step of identifying the question classification and the corresponding difficulty level of the question data comprises: Performing intent recognition on the problem data to obtain problem classification of the problem data; The user's historical conversation data is obtained, and the difficulty level of the question data is determined according to the user's historical conversation data and / or the question classification.

3. The method according to claim 1, characterized in that Before the step of determining the processing model according to the problem classification and the difficulty level, the method further includes: Retrieving the problem data in a preset cache; When there is a question in the cache whose similarity to the question data reaches a preset threshold, the answer corresponding to the question is output as the answer to the question data.

4. The method according to claim 1, characterized in that: The step of inputting the question data into the processing model and outputting the answer to the question data comprises: When the difficulty level is lower than a preset level, determining the processing model to be a small model; The question data is input into the small model, and the answer to the question data is output.

5. The method according to claim 1, characterized in that: The step of inputting the question data into the processing model and outputting the answer to the question data comprises: When the difficulty level is not less than a preset level, determining the processing model to be a small model and a large language model; Inputting the question data into the small model and the large language model; When it is detected that the small model outputs an answer guide in response to the question data and the large language model does not generate an output result, output the answer guide; When the large language model generates an answer to the question data in response to the question data, the answer to the question data is output.

6. The method according to any one of claims 1 to 5, characterized in that: The problem categories include: business consultation, daily chat, instruction execution, and common sense questions.

7. A digital human question-answering device, characterized in that: include: A question data receiving module, used for receiving question data from users; An identification module, used to identify the problem classification and corresponding difficulty level of the problem data; A processing model determination module, used to determine a processing model according to the problem classification and the difficulty level; The first answer output module is used to input the question data into the processing model and output the answer to the question data.

8. The device according to claim 7, characterized in that The identification module comprises: An intention recognition submodule, used to perform intention recognition on the question data to obtain a question classification of the question data; The difficulty level determination submodule is used to obtain user historical conversation data and determine the difficulty level of the question data according to the user historical conversation data and / or the question classification.

9. An electronic device, characterized in that: The device comprises a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the digital human question-answering method according to any one of claims 1 to 6 according to the instructions in the program code.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store program codes, and the program codes are used to execute the digital human question-answering method according to any one of claims 1 to 6.

Images