Memory information determination method and device based on large model, and task processing method and device based on large model

Abstract

The invention provides a memory information determination method and device based on a large model and a task processing method and device, and relates to the technical field of artificial intelligence, in particular to the technical field of natural language processing, deep learning and large models. The method comprises the steps of performing information extraction on interaction information between a target account and an intelligent agent through a first large model to determine each piece of first candidate information; based on at least one of the confidence of the first candidate information and the importance of the first candidate information, determining first memory information associated with the target account from the first candidate information; wherein each piece of memory information associated with the target account is used for the intelligent agent to execute a target task associated with the target account.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to the fields of natural language processing, deep learning, and large model technology. Background Technology

[0002] Currently, the intelligent agent is a core concept in the field of artificial intelligence, referring to an agent capable of perceiving environmental changes, making autonomous decisions, and taking actions to achieve specific goals. It possesses autonomy, adaptability, and interactivity, and is widely used in scenarios such as automation, robotics, virtual assistants, and game characters. However, when intelligent agents interact with users, there is a problem of poor user experience. Summary of the Invention

[0003] This disclosure provides a method for determining memory information based on a large model, a method for task processing based on a large model, an apparatus, an electronic device, a storage medium, and a computer program product.

[0004] According to a first aspect of this disclosure, a method for determining memory information based on a large model is provided, comprising: extracting information from the interaction information between a target account and an agent using a first large model to determine each first candidate information; determining first memory information associated with the target account from each first candidate information based on at least one of the confidence level and the importance level of the first candidate information; wherein each memory information associated with the target account is used by the agent to perform a target task associated with the target account.

[0005] According to a second aspect of this disclosure, a task processing method based on a large model is provided, comprising: determining a target task to be performed by an agent; wherein the target task is associated with a target account; determining a processing result of the target task based on memory information associated with the target account using a second large model; wherein the memory information associated with the target account includes first memory information, which is each first candidate information extracted by a first large model from the interaction information between the target account and the agent, and determined from each first candidate information based on at least one of the confidence level and the importance level of the first candidate information; the processing result of the target task is used by a terminal device for visualization and / or voice playback.

[0006] According to a third aspect of this disclosure, a memory information determination device based on a large model is provided, comprising: an extraction module, configured to extract information from the interaction information between a target account and an agent through a first large model to determine each first candidate information; and a determination module, configured to determine first memory information associated with the target account from each of the first candidate information based on at least one of the confidence level and the importance level of the first candidate information; wherein each memory information associated with the target account is used by the agent to perform a target task associated with the target account.

[0007] According to a fourth aspect of this disclosure, a task processing apparatus based on a large model is provided, comprising: a determination module for determining a target task to be executed by an intelligent agent; wherein the target task is associated with a target account; and an execution module for determining a processing result of the target task based on memory information associated with the target account using a second large model; wherein the memory information associated with the target account includes first memory information, which is first candidate information extracted by a first large model from the interaction information between the target account and the intelligent agent, and determined from each first candidate information based on at least one of the confidence level and the importance level of the first candidate information; and the processing result of the target task is used by a terminal device for visualization and / or voice playback.

[0008] According to a fifth aspect of this disclosure, an electronic device is proposed, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the memory information determination method based on a large model proposed in the first aspect above, and / or to perform the task processing method based on a large model proposed in the second aspect above.

[0009] According to a sixth aspect of this disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause the computer to execute the memory information determination method based on a large model as proposed in the first aspect above, and / or to execute the task processing method based on a large model as proposed in the second aspect above.

[0010] According to the seventh aspect of this disclosure, a computer program product is proposed, comprising a computer program that, when executed by a processor, implements the memory information determination method based on a large model as proposed in the first aspect above, and / or implements the task processing method based on a large model as proposed in the second aspect above.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0012] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein: Figure 1 This is a flowchart illustrating a method for determining memory information based on a large model according to an embodiment of the present disclosure. Figure 2 This is a flowchart illustrating another embodiment of the memory information determination method based on a large model according to the present disclosure; Figure 3 This is a flowchart illustrating another embodiment of the memory information determination method based on a large model according to the present disclosure; Figure 4 This is a flowchart illustrating a task processing method based on a large model according to an embodiment of the present disclosure; Figure 5 This is a flowchart illustrating a task processing method based on a large model according to another embodiment of the present disclosure; Figure 6 This is a schematic diagram of the architecture of memory information according to an embodiment of the present disclosure; Figure 7 This is a schematic diagram illustrating the process of generating response information according to an embodiment of the present disclosure; Figure 8 This is a schematic diagram of the structure of a memory information determination device based on a large model according to an embodiment of the present disclosure; Figure 9 This is a schematic diagram of the structure of a task processing device based on a large model according to an embodiment of the present disclosure; Figure 10 This is a schematic block diagram of an electronic device according to an embodiment of the present disclosure. Detailed Implementation

[0013] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0014] Artificial intelligence (AI) is a technical science that studies and develops theories, methods, technologies, and application systems to simulate, extend, and expand human intelligence. Currently, AI technology has the advantages of high automation, high accuracy, and low cost, and has been widely applied.

[0015] NLU (Natural Language Processing) is a science that studies computer systems, especially their software systems, that can effectively achieve natural language communication. It is an important direction in the fields of computer science and artificial intelligence.

[0016] Deep learning (DL) is a new research direction in the field of machine learning (ML). It is a science that learns the inherent patterns and representation levels of sample data, enabling machines to have analytical and learning abilities like humans, and to recognize data such as text, images, and sound. It is widely used in speech and image recognition.

[0017] Large models refer to machine learning models with a massive number of parameters and high complexity. They require significant computing resources and storage space for training and storage, and often necessitate distributed computing and specialized hardware acceleration techniques. Large models possess stronger generalization and expressive capabilities. Large models include LLMs (Large Language Models). Large language models are deep learning models trained on large amounts of text data, capable of generating natural language text or understanding the meaning of language text. Large language models can handle various natural language tasks, such as text classification, question answering, and dialogue, and represent an important pathway to artificial intelligence.

[0018] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0019] Figure 1 This is a flowchart illustrating a method for determining memory information based on a large model according to an embodiment of this disclosure, as shown below. Figure 1 As shown, the method includes: S101 uses the first major model to extract information about the interaction between the target account and the agent in order to determine the first candidate information.

[0020] It should be noted that the execution entity of the memory information determination method based on a large model in this embodiment of the present disclosure can be a hardware device with data information processing capabilities and / or the necessary software to drive the hardware device to work. Optionally, the execution entity may include a workstation, server, computer, user terminal, and other intelligent devices. Among them, the user terminal includes, but is not limited to, mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle terminals, etc.

[0021] The first major model can be implemented using any major model from relevant technologies; no further restrictions are imposed here. For example, it could be the Transformer model, a large language model, etc. It should be noted that the Transformer model is a neural network model based on a self-attention mechanism. Based on the first major model, information is extracted from the interaction information between the target account and the agent to determine the first candidate information. This can be implemented using any information extraction method based on a major model from relevant technologies; no further restrictions are imposed here.

[0022] There are no strict limitations on the interactive information, such as dialogue information (e.g., single-turn dialogue information, multi-turn dialogue information, etc.), button clicks on the interactive interface, mouse movements, uploaded files, and interaction duration. There are also no strict limitations on the interaction methods between the target account and the intelligent agent, such as text interaction, voice interaction, and multimodal interaction (e.g., images, videos, voice, files, text, etc.).

[0023] Any candidate information includes the target account's interests, communication preferences, and content preferences. It should be noted that the acquisition, storage, and application of interest information, communication preferences, and content preferences are undertaken with the user's knowledge and consent.

[0024] Communication preferences include the tone style preferences of the AI ​​agent corresponding to the target account (e.g., formal, humorous, gentle), the granularity preferences of the AI ​​agent's response content (e.g., concise, detailed), the output format preferences of the AI ​​agent's response content (e.g., file, text, table), and the length preferences of the AI ​​agent's response content (e.g., a sentence, a paragraph, a long article).

[0025] Content preferences include content retrieval preferences, content creation preferences, and content management preferences. Content retrieval preferences include retrieval channel preferences, retrieval strategy preferences, retrieval frequency preferences, and reference content preferences. Content creation preferences include creative intent preferences, structural preferences (such as page number preferences, logical preferences, etc.), style preferences (such as visual style preferences, writing style preferences, layout preferences, color preferences, etc.), delivery preferences (such as timeliness preferences, output format preferences, completion level preferences, iteration frequency preferences, etc.), and creative habit preferences (such as template usage preferences, modification preferences, etc.).

[0026] For example, logical preferences include "users prefer a general-to-specific-to-general argument structure, especially when writing articles."

[0027] For example, the personal experience includes "The user is a third-year food safety major student who accumulated professional skills in raw material acceptance, hygiene inspection and other work during a 5-month internship at a company".

[0028] For example, personality traits include "users enjoy painting, visiting exhibitions, and discussing philosophy in their daily lives."

[0029] S102, based on at least one of the confidence level of the first candidate information and the importance level of the first candidate information, determine the first memory information associated with the target account from each first candidate information.

[0030] Among them, the memory information associated with the target account is used by the intelligent agent to execute the target task associated with the target account.

[0031] In this disclosure, confidence and / or importance can be taken into account to determine the first memory information associated with the target account from each first candidate information. For example, it can avoid using first candidate information with low confidence as the first memory information, which helps to improve the accuracy of the first memory information, thereby making the response content of the intelligent agent more accurate and optimizing the user experience.

[0032] And / or, it can avoid using low-importance first candidate information as the first memory information, and can prioritize remembering information that is more important to the user, so that the first memory information is highly focused on the user's core needs, and thus the intelligent agent's response content is more in line with the user's core needs, thereby optimizing the user experience.

[0033] In addition, considering confidence and / or importance, the first memory information associated with the target account is determined from each first candidate information, making the first memory information more concise, avoiding memory information overload, helping to speed up task processing, thereby shortening the response time of the intelligent agent and optimizing the user experience.

[0034] The memory information associated with the target account includes the primary memory information.

[0035] In this disclosure, first memory information associated with the target account is determined from each first candidate information based on at least one of the confidence level and the importance level of the first candidate information, including at least one of the following: Method 1: Based on confidence level, determine the first memory information associated with the target account from each first candidate information.

[0036] Therefore, confidence level can be taken into account to determine the first memory information associated with the target account from each first candidate information. For example, it can avoid using first candidate information with low confidence as the first memory information, which helps to improve the accuracy of the first memory information. As a result, the response content of the intelligent agent is more accurate, and the user experience is optimized.

[0037] Optionally, based on confidence level, the first memory information associated with the target account is determined from each first candidate information, including using the first candidate information with a confidence level greater than a first threshold as the first memory information associated with the target account.

[0038] Optionally, based on confidence level, the first memory information associated with the target account is determined from each first candidate information, including sorting each first candidate information in descending order of confidence level, and taking the top N first candidate information as the first memory information associated with the target account. Where N is a positive integer.

[0039] Optionally, based on confidence level, the first memory information associated with the target account is determined from each first candidate information, including sorting each first candidate information under the same dimension according to confidence level from largest to smallest, and taking the top N first candidate information as the first memory information associated with the target account.

[0040] For example, the first candidate information under the logical preference dimension includes first candidate information 1, first candidate information 2, and first candidate information 3. First candidate information 1 is "Users like the general-specific-general argumentation style, especially when writing articles", second candidate information 2 is "Users like the general-specific argumentation style", and second candidate information 3 is "Users like the timeline-based argumentation style".

[0041] If the confidence level of the first candidate information 1 is 0.5, the confidence level of the first candidate information 2 is 0.8, and the confidence level of the first candidate information 3 is 0.6, then the first candidate information 2 can be determined as the first memory information associated with the target account.

[0042] Method 2: Based on importance, determine the first memory information associated with the target account from each first candidate information.

[0043] Therefore, considering importance, the first memory information associated with the target account can be determined from each first candidate information. For example, it can avoid using low-importance first candidate information as the first memory information, and can prioritize remembering information that is more important to the user. This makes the first memory information highly focused on the user's core needs, and thus the response content of the intelligent agent is more in line with the user's core needs, thereby optimizing the user experience.

[0044] Optionally, based on importance, the first memory information associated with the target account is determined from each first candidate information, including using the first candidate information with an importance greater than a second threshold as the first memory information associated with the target account.

[0045] Optionally, based on importance, the first memory information associated with the target account is determined from each first candidate information, including sorting each first candidate information in descending order of importance, and taking the top N first candidate information as the first memory information associated with the target account. Here, N is a positive integer.

[0046] Optionally, based on importance, the first memory information associated with the target account is determined from each first candidate information, including sorting each first candidate information under the same dimension in descending order of importance, and taking the top N first candidate information as the first memory information associated with the target account.

[0047] Method 3: Based on confidence and importance, determine the first memory information associated with the target account from each first candidate information.

[0048] Therefore, by comprehensively considering confidence and importance, the first memory information associated with the target account can be determined from each first candidate information. This helps to improve the accuracy of the first memory information and can prioritize remembering information that is more important to the user. This makes the first memory information highly focused on the user's core needs, and thus the response content of the intelligent agent is more accurate and better matches the user's core needs, thereby optimizing the user experience.

[0049] Optionally, based on confidence and importance, the first memory information associated with the target account is determined from each first candidate information. This includes weighted summation of the confidence and importance of any first candidate information to determine the score of the corresponding first candidate information, and then determining the first memory information associated with the target account based on the score. Thus, a weighted summation of confidence and importance can be used to determine the score of the first candidate information, thereby determining the first memory information.

[0050] It should be noted that, based on the score, the relevant content of the first memory information associated with the target account can be determined from each first candidate information based on confidence or importance, as described in the above embodiment. This will not be repeated here.

[0051] Optionally, based on confidence and importance, the first memory information associated with the target account is determined from each first candidate information, including using the first candidate information with a confidence greater than a first threshold and an importance greater than a second threshold as the first memory information associated with the target account.

[0052] It should be noted that the target tasks are not limited in too much, and may include question answering, dialogue, retrieval, content creation (such as PPT generation, document generation, poster generation, etc.), document summary generation, text translation, etc.

[0053] Optionally, the method further includes determining the processing result of the target task based on the memory information associated with the target account using a second major model. The memory information associated with the target account includes first memory information, and the processing result of the target task is used for visualization and / or voice playback on the terminal device. Thus, by determining the processing result of the target task based on the memory information associated with the target account using a second major model, the first memory information, which has higher accuracy, helps improve the accuracy of the processing result of the target task, meaning the agent's response is more accurate, thus optimizing the user experience. And / or, the first memory information highly focuses on the user's core needs, making the processing result of the target task more aligned with the user's core needs, meaning the agent's response is more aligned with the user's core needs, thus optimizing the user experience. Furthermore, the first memory information is more concise, avoiding memory information overload, which helps accelerate the reasoning speed of the processing result of the target task, thereby shortening the agent's response time and optimizing the user experience. Optionally, the processing result of the target task is determined by the second major model based on the memory information associated with the target account. This includes determining candidate memory information from the memory information associated with the target account based on the task information of the target task, determining the target memory information from the candidate memory information based on the weight of the candidate memory information, and determining the processing result of the target task based on the target memory information by the second major model.

[0054] Optionally, based on the task information of the target task, candidate memory information is determined from the memory information associated with the target account. This includes determining the semantic matching degree between the task information of the target task and any memory information associated with the target account, determining the scene matching degree between the task scene of the target task and the task scene allowed to be used by any memory information associated with the target account, and determining candidate memory information from the memory information associated with the target account based on at least one of the semantic matching degree and the scene matching degree.

[0055] It should be noted that the second major model, based on the memory information associated with the target account, determines the relevant content of the processing result of the target task. See the following embodiments for details, which will not be repeated here. The second major model is used to determine the processing result of the task to be performed by the agent. The first and second major models may or may not be the same; no further restrictions are imposed here.

[0056] This disclosure does not impose any restrictions on the execution sequence of steps S101-S102. Figure 1 The example only demonstrates the execution of steps S101-S102 in sequence.

[0057] The proposed method for determining memory information based on a large model extracts information from the interaction information between the target account and the agent using a first large model to determine various first candidate information. Based on at least one of the confidence level and importance of the first candidate information, the method determines the first memory information associated with the target account from among these first candidate information. Each memory information associated with the target account is used by the agent to execute the target task associated with the target account. Therefore, considering confidence level and / or importance, the method determines the first memory information associated with the target account from among the first candidate information. For example, it avoids using low-confidence first candidate information as the first memory information, helping to improve the accuracy of the first memory information, resulting in more accurate responses from the agent and optimized user experience. And / or, it avoids using low-importance first candidate information as the first memory information, prioritizing information that is more important to the user, making the first memory information highly focused on the user's core needs, thus making the agent's responses more aligned with the user's core needs and optimizing user experience.

[0058] In addition, considering confidence and / or importance, the first memory information associated with the target account is determined from each first candidate information, making the first memory information more concise, avoiding memory information overload, helping to speed up task processing, thereby shortening the response time of the intelligent agent and optimizing the user experience.

[0059] In the above embodiments, the determination of the memory information associated with the target account can be combined with... Figure 2 To understand further, Figure 2 This is a flowchart illustrating another embodiment of the memory information determination method based on a large model, as shown below. Figure 2 As shown, the method includes: S201, the first model is used to extract information from the interaction between the target account and the agent to determine the first candidate information.

[0060] S202, based on at least one of the confidence level of the first candidate information and the importance level of the first candidate information, determine the first memory information associated with the target account from each first candidate information.

[0061] For a description of steps S201-S202, please refer to the relevant content in the above embodiments, which will not be repeated here.

[0062] S203, in response to obtaining first interactive information and second interactive information; wherein, the first interactive information is used to indicate that the memory information associated with the target account originates from the second interactive information.

[0063] S204, extract information from the second interaction information using the first major model to determine the second candidate information.

[0064] S205, the second candidate information is used as the second memory information associated with the target account.

[0065] In this disclosure, information can be extracted from the second interaction information through the first major model to determine the second candidate information, and the second candidate information can be directly used as the second memory information associated with the target account. That is, the user can explicitly specify the information to be remembered, so that the second memory information is highly aligned with the user's true intention, which helps to improve the accuracy of the second memory information.

[0066] It should be noted that there are no excessive restrictions on the first and second interactive information. For example, the first and second interactive information can be independent of each other, or the first interactive information can carry the second interactive information.

[0067] For example, a user can input "I like the descriptive logic of 'general-specific-general' on the interactive interface of the terminal device, please help me remember it"; accordingly, the first major model extracts information from "I like the descriptive logic of 'general-specific-general'" to determine "the user likes the argumentation style of 'general-specific-general'", and uses "the user likes the argumentation style of 'general-specific-general'" as the second memory information associated with the target account.

[0068] Additionally, the processing results of the dialogue task can be generated through the second major model. "Understood! Going forward, I will prioritize presenting the content in a structured and clear manner, avoiding fragmented expression as much as possible. If you would like me to use a certain framework (such as MECE, general-to-specific structure, timeline organization, etc.), please let me know at any time."

[0069] Correspondingly, the terminal device displays the processing results of the above dialogue task on the interactive interface.

[0070] It should be noted that in this embodiment, "I like the descriptive logic of general-specific-general, please help me remember it" is the first interactive information, "I like the descriptive logic of general-specific-general" is the second interactive information, and "Users like the argumentation style of general-specific-general" is the second candidate information and the second memory information.

[0071] For example, a user could say, "Remember I'm taking the 7 PM train, remind me at 6:30 PM." Accordingly, the first major model performs text recognition on the above voice information to determine the recognized text, and extracts information from the recognized text to determine that "the user plans to take the 7 PM train, and needs to be reminded at 6:30 PM sharp." This "the user plans to take the 7 PM train, and needs to be reminded at 6:30 PM sharp" is then used as the second memory information associated with the target account.

[0072] Additionally, the processing results of the dialogue task can be generated through the second major model: "Okay, it's all recorded for you: you plan to take the 7 pm train, and I will remind you promptly at 6:30 pm. This way you don't have to worry about missing your train. Please be careful on land."

[0073] Correspondingly, the terminal device displays the processing results of the above dialogue task on the interactive interface.

[0074] It should be noted that in this embodiment, "Remember that I will be taking the 7 pm train and remind me at 6:30 pm" is the first interactive information, "I will be taking the 7 pm train and remind me at 6:30 pm" is the second interactive information, and "The user plans to take the 7 pm train and needs to be reminded at 6:30 pm" is the second candidate information and the second memory information.

[0075] It should be noted that this disclosure does not restrict the execution sequence of steps S201-S205. For example, steps S201-S202 and steps S203-S205 can be executed in parallel.

[0076] The memory information determination method based on a large model proposed in this disclosure responds to the acquisition of first interaction information and second interaction information. The first interaction information indicates that the memory information associated with the target account originates from the second interaction information. Information is extracted from the second interaction information using a first large model to determine second candidate information, which is then used as the second memory information associated with the target account. Therefore, by extracting information from the second interaction information using the first large model to determine the second candidate information, and directly using the second candidate information as the second memory information associated with the target account, the user can explicitly specify the information to be remembered. This ensures that the second memory information is highly aligned with the user's true intent, thus improving the accuracy of the second memory information.

[0077] In the above embodiments, the update content of each memory information can be combined with... Figure 3 To understand further, Figure 3 This is a flowchart illustrating another embodiment of the memory information determination method based on a large model, as shown below. Figure 3 As shown, the method includes: S301 extracts information from the interaction between the target account and the agent using the first major model to determine the first candidate information.

[0078] S302, based on at least one of the confidence level of the first candidate information and the importance level of the first candidate information, determine the first memory information associated with the target account from each first candidate information.

[0079] S303, in response to obtaining first interactive information and second interactive information; wherein, the first interactive information is used to indicate that the memory information associated with the target account originates from the second interactive information.

[0080] S304, extract information from the second interaction information using the first major model to determine the second candidate information.

[0081] S305, the second candidate information is used as the second memory information associated with the target account.

[0082] For a description of steps S301-S305, please refer to the relevant content in the above embodiments, which will not be repeated here.

[0083] S306, determine the weight of the corresponding memory information based on at least one of the confidence level of any memory information and the importance level of the corresponding memory information; wherein, any memory information includes either the first memory information or the second memory information.

[0084] In this disclosure, the weight of a corresponding memory information is determined based on at least one of the confidence level of any memory information and the importance level of the corresponding memory information, including at least one of the following: Method 1: Determine the weight of the corresponding memory information based on the confidence level of any memory information.

[0085] Method 2: Determine the weight of the corresponding memory information based on the importance of any memory information.

[0086] Method 3: Determine the weight of the corresponding memory information based on the confidence level of any memory information and the importance level of the corresponding memory information.

[0087] Optionally, the weight of a corresponding memory information is determined based on at least one of the confidence level and the importance level of any memory information. This includes determining the usage information of any memory information for performing the task, determining the degree of semantic conflict between any memory information and newly added memory information, and determining the weight of the corresponding memory information based on the confidence level of any memory information, the importance level of the corresponding memory information, the usage information of the corresponding memory information, and the degree of semantic conflict corresponding to the memory information. Therefore, the weight of memory information can be determined by comprehensively considering confidence level, importance, usage information, and the degree of semantic conflict, thus improving the accuracy of the weighting of memory information.

[0088] Optionally, the usage information of any memory information includes at least one of the following: The last time any memory information was used to perform a task; The time interval corresponding to any memory information; wherein, the time interval corresponding to any memory information is the time interval between the last time the corresponding memory information was used to perform a task and the current time; The frequency with which any memory information is used to perform a task.

[0089] Therefore, the weight of memory information can be determined by taking into account at least one of the following: the time when any memory information was last used to perform a task, the time interval corresponding to any memory information, and the frequency of use of any memory information to perform a task, thereby improving the accuracy of the weight of memory information.

[0090] Optionally, the weight of any memory information is positively correlated with the confidence level of the corresponding memory information; or, The weight of any memory information is positively correlated with the importance of the corresponding memory information; or... The weight of any memory information is negatively correlated with the time interval corresponding to that memory information; or... The weight of any memory information is positively correlated with the frequency of its use; or... The weight of any memory information is negatively correlated with the degree of semantic conflict corresponding to that memory information.

[0091] Therefore, the weight of any memory information is positively correlated with the confidence level of the corresponding memory information. Thus, the higher the confidence level of any memory information, the greater the weight of the corresponding memory information, and the higher the confidence level of memory information can be retained first, which helps to improve the accuracy of memory information.

[0092] Alternatively, the weight of any memory information is positively correlated with the importance of the corresponding memory information. Therefore, if the importance of any memory information is greater, the weight of the corresponding memory information is greater, and memory information with high importance can be retained first. That is, information that is more important to the user can be retained first, so that memory information is highly focused on the user's core needs.

[0093] Alternatively, the weight of any memory information is negatively correlated with the time interval corresponding to that memory information. Thus, if any memory information has not been used for a long time, the weight of the corresponding memory information is smaller, and the most recently used memory information can be retained first.

[0094] Alternatively, the weight of any memory information is positively correlated with the frequency of its use. Therefore, if any memory information is frequently used, the corresponding memory information has a larger weight, and frequently used memory information can be retained first.

[0095] Alternatively, the weight of any memory information is negatively correlated with the degree of semantic conflict corresponding to that memory information. Thus, if the degree of semantic conflict between any memory information and the newly added memory information is greater, the weight of the corresponding memory information is smaller. Memory information with a smaller degree of semantic conflict with the newly added memory information can be retained first, thereby improving the semantic consistency of each memory information.

[0096] S307 updates each memory information based on weights.

[0097] In this disclosure, the weight of memory information can be determined by taking into account confidence level and / or importance, so as to update each piece of memory information. For example, memory information with high confidence level can be retained first, which helps to improve the accuracy of memory information, and / or memory information with high importance can be retained first, that is, information that is more important to the user can be retained first, so that the memory information is highly focused on the user's core needs, and the dynamic evolution of memory information can be realized.

[0098] Optionally, each memory information is updated based on its weight, including deleting the corresponding memory information when the weight of any memory information is less than or equal to a set threshold, and retaining the corresponding memory information when the weight of any memory information is greater than the set threshold.

[0099] Optionally, based on the weights, each memory information is updated, including sorting each memory information in the same dimension according to the weight from largest to smallest, and deleting the N memory information after sorting; where N is a positive integer.

[0100] It should be noted that this disclosure does not limit the execution sequence of steps S301-S307. For example, steps S301-S302 and S306-S307 can be implemented as independent embodiments, and steps S303-S307 can be implemented as independent embodiments. The memory information determination method based on a large model proposed in this disclosure determines the weight of corresponding memory information based on at least one of the confidence level and the importance level of any given memory information. Each memory information includes either a first memory information or a second memory information. Based on the weights, each memory information is updated. Thus, considering confidence level and / or importance, the weights of memory information can be determined to update each memory information. For example, high-confidence memory information can be preferentially retained, which helps improve the accuracy of memory information and thus improves the accuracy of the agent's response content, optimizing the user experience. And / or, high-importance memory information can be preferentially retained, that is, information that is more important to the user can be preferentially retained, making the memory information highly focused on the user's core needs, thereby improving the matching degree between the agent's response content and the user's core needs, and enabling the dynamic evolution of memory information.

[0101] Based on any of the above embodiments, the method further includes at least one of the following: Method 1: Determine the confidence level of the first memory information using the first major model.

[0102] Method 2: Determine the confidence level of the second memory information as a set value.

[0103] Among them, the confidence level of the second memory information is greater than or equal to the confidence level of the first memory information.

[0104] Therefore, the confidence level of the first memory information can be determined through the first major model, and / or the confidence level of the second memory information can be determined as a set value.

[0105] Understandably, the first memory information is determined based on confidence and / or importance, while the second memory information is information that the user explicitly specifies to be remembered. The confidence level of the second memory information is greater than or equal to that of the first memory information, meaning that the second memory information is more accurate.

[0106] Based on any of the above embodiments, the method further includes determining the importance of corresponding candidate information using a first major model based on at least one of the following: the correlation between any candidate information and interest information, the correlation between corresponding candidate information and candidate task, and the source information of corresponding candidate information; wherein, any candidate information includes either first candidate information or second candidate information. Thus, the importance of corresponding candidate information can be determined by considering at least one of the following: the correlation between any candidate information and interest information, the correlation between corresponding candidate information and candidate task, and the source information of corresponding candidate information, using the first major model.

[0107] Optionally, the source information for the candidate information includes at least one of the following: The emotional intensity of the interactive information from the candidate information sources; The frequency of occurrence of interactive information from candidate information sources; The task weights of candidate tasks for candidate information sources.

[0108] Therefore, the importance of candidate information can be determined by considering at least one of the following factors: the emotional intensity of the interactive information from the candidate information source, the frequency of occurrence of the interactive information from the candidate information source, and the task weight of the candidate task from the candidate information source, thereby improving the accuracy of the importance determination.

[0109] Optionally, the importance of any candidate information is positively correlated with the correlation between the corresponding candidate information and the information of interest; or, The importance of any candidate information is positively correlated with the correlation between the corresponding candidate information and the candidate task; or... The importance of any candidate piece of information is positively correlated with the emotional intensity of the interactive information from which the corresponding candidate information originates; or, The importance of any candidate information is positively correlated with the frequency of occurrence of the interaction information from the corresponding candidate information source; or, The importance of any candidate information is positively correlated with the task weight of the candidate task from which the corresponding candidate information originates.

[0110] Therefore, the importance of any candidate information is positively correlated with the degree of correlation between the corresponding candidate information and the information of interest. Thus, the greater the correlation between any candidate information and the information of interest, the greater the importance of the corresponding candidate information, and the more likely it is to be remembered or retained.

[0111] Alternatively, the importance of any candidate information is positively correlated with the degree of relevance between the corresponding candidate information and the candidate task. Therefore, the greater the degree of relevance between any candidate information and the candidate task, the greater the importance of the corresponding candidate information, and the more likely it is to be remembered or retained.

[0112] Alternatively, the importance of any candidate information is positively correlated with the emotional intensity of the interactive information from the corresponding candidate information source. Therefore, if the emotional intensity of the interactive information from any candidate information source is greater, the importance of the corresponding candidate information is greater, and information with greater emotional intensity can be remembered or retained first.

[0113] Alternatively, the importance of any candidate information is positively correlated with the frequency of the interaction information from the corresponding candidate information source. Thus, if the interaction information from any candidate information source appears frequently, such as repeatedly appearing across multiple rounds of dialogue, the greater the importance of the corresponding candidate information, the more likely it is to be remembered or retained.

[0114] Alternatively, the importance of any candidate information is positively correlated with the task weight of the candidate task from which the candidate information originates. Thus, the greater the task weight of the candidate task from which any candidate information originates, the greater the importance of the corresponding candidate information, and the more likely information with a higher task weight can be remembered or retained.

[0115] Figure 4 This is a flowchart illustrating a task processing method based on a large model according to an embodiment of the present disclosure, as shown below. Figure 4 As shown, the method includes: S401, determine the target task to be executed by the intelligent agent; wherein, the target task is associated with the target account.

[0116] S402 determines the processing result of the target task based on the memory information associated with the target account through the second major model.

[0117] It should be noted that the execution entity of the task processing method based on a large model in this embodiment can be a hardware device with data processing capabilities and / or the necessary software to drive the hardware device. Optionally, the execution entity may include a workstation, server, computer, user terminal, and other intelligent devices. The user terminal includes, but is not limited to, mobile phones, computers, intelligent voice interaction devices, smart home appliances, and in-vehicle terminals.

[0118] The memory information associated with the target account includes first memory information, which is each first candidate information extracted by the first big model from the interaction information between the target account and the agent, and determined from each first candidate information based on at least one of the confidence level and importance of the first candidate information.

[0119] The processing results of the target task are used by the terminal device for visualization and / or audio playback.

[0120] In this disclosure, the processing result of the target task is determined by the second major model based on the memory information associated with the target account. The memory information associated with the target account includes the first memory information. The accuracy of the first memory information is relatively high, which helps to improve the accuracy of the processing result of the target task. That is, the response content of the intelligent agent is more accurate, thus optimizing the user experience.

[0121] And / or, the first memory information is highly focused on the user's core needs, making the processing results of the target task more in line with the user's core needs, that is, the response content of the intelligent agent is more in line with the user's core needs, thus optimizing the user experience.

[0122] In addition, the first memory information is more concise, which can avoid memory information overload, help to speed up the reasoning speed of the processing results of the target task, thereby shortening the response time of the intelligent agent and optimizing the user experience.

[0123] Optionally, the memory information associated with the target account also includes second memory information. This second memory information is determined by extracting information from the second interaction information using the first large model when the first and second interaction information are obtained. The first interaction information indicates that the memory information associated with the target account originates from the second interaction information. Therefore, the memory information associated with the target account also includes second memory information. This second memory information is highly aligned with the user's true intent, and its accuracy is relatively high, which helps improve the accuracy of the target task's processing results. In other words, the agent's response is more accurate, thus optimizing the user experience.

[0124] It should be noted that this disclosure does not impose any restrictions on the execution sequence of steps S401-S402. Figure 4 The example only demonstrates the execution of steps S401-S402 in sequence.

[0125] The task processing method based on a large model proposed in this disclosure determines the target task to be executed by the agent. The target task is associated with a target account, and the target task is executed based on the memory information associated with the target account through a second large model. Thus, the processing result of the target task is determined by the memory information associated with the target account through the second large model. The memory information associated with the target account includes first memory information, which has higher accuracy, contributing to the accuracy of the processing result of the target task; that is, the agent's response is more accurate, optimizing the user experience. And / or, the first memory information highly focuses on the user's core needs, making the processing result of the target task more aligned with the user's core needs; that is, the agent's response is more aligned with the user's core needs, optimizing the user experience.

[0126] In addition, the first memory information is more concise, which can avoid memory information overload, help to speed up the reasoning speed of the processing results of the target task, thereby shortening the response time of the intelligent agent and optimizing the user experience.

[0127] In the above embodiments, the relevant content regarding step S402, which uses the second model to determine the processing result of the target task based on the memory information associated with the target account, can be combined with... Figure 5 To understand further, Figure 5 This is a flowchart illustrating another embodiment of the task processing method based on a large model, as shown below. Figure 5 As shown, the method includes: S501, determine the target task to be executed by the intelligent agent; wherein, the target task is associated with the target account.

[0128] For a description of step S501, please refer to the relevant content in the above embodiments, which will not be repeated here.

[0129] S502, Based on the task information of the target task, determine candidate memory information from the memory information associated with the target account.

[0130] Optionally, based on the task information of the target task, candidate memory information is determined from the memory information associated with the target account. This includes determining the semantic matching degree between the task information of the target task and any memory information associated with the target account, and determining the scenario matching degree between the task scenario of the target task and the task scenario allowed by any memory information associated with the target account. Candidate memory information is then determined from the memory information associated with the target account based on at least one of the semantic matching degree and the scenario matching degree. Therefore, considering semantic matching degree and / or scenario matching degree, candidate memory information can be determined from the memory information associated with the target account. For example, memory information with a high semantic matching degree can be used as candidate memory information, and / or memory information with a high scenario matching degree can be used as candidate memory information, which helps to improve the semantic matching degree and / or scenario matching degree between the candidate memory information and the target task.

[0131] It is understandable that different memory information may be used in different task scenarios, and any memory information may be used in at least one task scenario.

[0132] Optionally, candidate memory information is determined from the memory information associated with the target account based on at least one of semantic matching degree and scene matching degree, including at least one of the following: Method 1: Based on semantic matching degree, candidate memory information is determined from the memory information associated with the target account.

[0133] Optionally, based on semantic matching degree, candidate memory information is determined from the memory information associated with the target account, including memory information with semantic matching degree greater than a third threshold as candidate memory information.

[0134] Optionally, based on semantic matching degree, candidate memory information is determined from the memory information associated with the target account, including sorting the memory information associated with the target account from largest to smallest according to semantic matching degree, and taking the top N memory information as candidate memory information.

[0135] Method 2: Based on scene matching degree, candidate memory information is determined from the memory information associated with the target account.

[0136] Optionally, based on the scene matching degree, candidate memory information is determined from the memory information associated with the target account, including using memory information with a scene matching degree greater than a fourth threshold as candidate memory information.

[0137] Optionally, based on the scene matching degree, candidate memory information is determined from the memory information associated with the target account, including sorting the memory information associated with the target account from largest to smallest according to the scene matching degree, and taking the top N memory information as candidate memory information.

[0138] Method 3: Based on semantic matching degree and scene matching degree, candidate memory information is determined from the memory information associated with the target account.

[0139] Optionally, candidate memory information is determined from the memory information associated with the target account based on semantic matching degree and scene matching degree. This includes weighted summation of the semantic matching degree of any memory information and the scene matching degree of the corresponding memory information to determine the score of the corresponding memory information, and candidate memory information is determined from the memory information associated with the target account based on the score.

[0140] It should be noted that, based on the score, the relevant content of candidate memory information is determined from the memory information associated with the target account. This can be done by referring to the above embodiment, which uses semantic matching degree or scene matching degree to determine the relevant content of candidate memory information from the memory information associated with the target account. This will not be elaborated here.

[0141] Optionally, based on semantic matching degree and scene matching degree, candidate memory information is determined from the memory information associated with the target account, including memory information with semantic matching degree greater than a third threshold and scene matching degree greater than a fourth threshold as candidate memory information.

[0142] S503, based on the weights of the candidate memory information, determine the target memory information from each candidate memory information.

[0143] It should be noted that the relevant content regarding the weight of memory information can be found in the above embodiments, and will not be repeated here.

[0144] Optionally, the target memory information is determined from each candidate memory information based on the weight of the candidate memory information, including taking the candidate memory information with a weight greater than the fifth threshold as the target memory information.

[0145] Optionally, the target memory information is determined from each candidate memory information based on the weight of the candidate memory information, including sorting each candidate memory information in descending order of weight, and taking the top N candidate memory information as the target memory information.

[0146] S504 determines the processing result of the target task based on the target memory information through the second major model.

[0147] In this embodiment, considering the task information of the target task, candidate memory information is determined from the memory information associated with the target account. Taking into account the weight of each candidate memory information, target memory information is determined from among them. The target task is then executed based on the target memory information using a second major model. This allows for the identification of target memory information that is strongly correlated with the target task and has a high weight, thus determining the processing result of the target task, improving the quality of the agent's response content generation, and optimizing the user experience.

[0148] It should be noted that this disclosure does not impose any restrictions on the execution sequence of steps S501-S504. Figure 5 The example only demonstrates the execution of steps S501-S504 in sequence.

[0149] The task processing method based on a large model proposed in this disclosure determines candidate memory information from various memory information associated with the target account based on the task information of the target task. Then, based on the weights of each candidate memory information, the target memory information is determined from these candidate memory information. Finally, the target task is executed using a second large model based on the target memory information. This method considers the task information of the target task, determines candidate memory information from various memory information associated with the target account, and considers the weights of each candidate memory information to determine the target memory information. The target task is then executed using a second large model based on the target memory information. This allows for the identification of target memory information that is strongly correlated with the target task and has a high weight, thereby determining the processing result of the target task, improving the quality of the agent's response content generation, and optimizing the user experience.

[0150] For ease of understanding, an exemplary embodiment is provided: like Figure 6 As shown, the architecture of memory information includes the accumulation, management, processing, and application of memory information.

[0151] The accumulation of memory information includes explicit feedback and implicit feedback. Explicit feedback includes user dialogue, content editing, file upload, and self-editing, while implicit feedback includes interest information, dialogue rounds, adoption status, and dialogue context.

[0152] The management of memory information includes storage, updating, retrieval, and forgetting. Storage mainly involves encoding and persisting memory information, updating refers to correcting and integrating memory information, retrieval mainly involves similarity matching and context awareness, and forgetting mainly involves filtering and lifecycle management of memory information.

[0153] The processing of memory information includes dividing it into short-term memory, medium-term memory, and long-term memory. For example, short-term memory information may include task context, dialogue context, etc.; medium-term memory information may include user behavior data and user dialogue history over a recent period, including historical file results; and long-term memory information may include interest information, content creation preferences, task execution preferences, etc.

[0154] The application of memory information can include tasks such as content search, creation, and management, as well as daily communication scenarios such as simple question and answer and casual conversation.

[0155] Among them, the task execution scenario mainly applies personalized content preferences, such as search preferences, creation preferences (such as writing style preferences, logic preferences, etc.), management preferences, etc., while the daily communication scenario mainly applies personal communication preferences, such as communication style, personality traits, personal experience, life needs, etc.

[0156] like Figure 7 As shown, the process of generating response information (also called the processing result of the target task or the response content of the agent) includes a memory recall process and a memory injection process.

[0157] Memory evoked process: After obtaining user input information, task planning can be performed on the input information. Then, based on the input information, it is determined whether to evoked memory information. If memory information is evoked, the user's associated memory information is queried to retrieve the relevant memory information. If memory information is not evoked, the non-memory evoked scenario process is entered.

[0158] This includes querying the user's associated memory information. Figure 7 (Not shown), including: similarity matching in short-term memory information, retrieval in medium-term memory information, retrieval in long-term memory information, etc.

[0159] Memory injection process: Based on relevant memory information and task planning results, the corresponding memory information is transmitted to the corresponding tool, then the tool is invoked, the tool execution result is obtained, and the response information is obtained based on the tool execution result.

[0160] According to embodiments of this disclosure, this disclosure also provides a memory information determination apparatus based on a large model, used to implement the above-described memory information determination method based on a large model.

[0161] Figure 8 This is a block diagram of a memory information determination device based on a large model according to an embodiment of the present disclosure.

[0162] like Figure 8 As shown, the memory information determination device 800 based on a large model includes: an extraction module 801 and a determination module 802.

[0163] The extraction module 801 is used to extract information from the interaction information between the target account and the intelligent agent through the first model in order to determine each first candidate information; The determining module 802 is configured to determine, based on at least one of the confidence level and the importance level of the first candidate information, the first memory information associated with the target account from each of the first candidate information; The memory information associated with the target account is used by the intelligent agent to execute the target task associated with the target account.

[0164] In one embodiment of this disclosure, the determining module 802 is further configured to: perform a weighted summation of the confidence level and the importance level of any first candidate information to determine the score of the corresponding first candidate information; and determine the first memory information associated with the target account from each of the first candidate information based on the score.

[0165] In one embodiment of this disclosure, the determining module 802 is further configured to: respond to obtaining first interaction information and second interaction information; wherein the first interaction information is used to indicate that the memory information associated with the target account originates from the second interaction information; extract information from the second interaction information through the first big model to determine second candidate information; and use the second candidate information as the second memory information associated with the target account.

[0166] In one embodiment of this disclosure, the determining module 802 is further configured to: determine the weight of the corresponding memory information based on at least one of the confidence level of any memory information and the importance level of the corresponding memory information; wherein, any memory information includes any one of the first memory information and the second memory information; and update each of the memory information based on the weight.

[0167] In one embodiment of this disclosure, the determining module 802 is further configured to: determine the usage information of any memory information for performing a task; determine the degree of semantic conflict between any memory information and newly added memory information; and determine the weight of the corresponding memory information based on the confidence level of any memory information, the importance level of the corresponding memory information, the usage information of the corresponding memory information, and the degree of semantic conflict corresponding to the corresponding memory information.

[0168] In one embodiment of this disclosure, the usage information of any memory information includes at least one of the following: The last time any memory information was used to perform a task; The time interval corresponding to any memory information; wherein, the time interval corresponding to any memory information is the time interval between the last time the corresponding memory information was used to perform a task and the current time; The frequency with which any memory information is used to perform a task.

[0169] In one embodiment of this disclosure, the weight of any memory information is positively correlated with the confidence level of the corresponding memory information; or, The weight of any memory information is positively correlated with the importance of the corresponding memory information; or... The weight of any memory information is negatively correlated with the time interval corresponding to that memory information; or... The weight of any memory information is positively correlated with the frequency of its use; or... The weight of any memory information is negatively correlated with the degree of semantic conflict corresponding to that memory information.

[0170] In one embodiment of this disclosure, the determining module 802 is further configured to perform at least one of the following: The confidence level of the first memory information is determined using the first large model; The confidence level of the second memory information is set to a predetermined value; Wherein, the confidence level of the second memory information is greater than or equal to the confidence level of the first memory information.

[0171] In one embodiment of this disclosure, the determining module 802 is further configured to: determine the importance of the corresponding candidate information based on at least one of the following: the correlation degree between any candidate information and interest information, the correlation degree between the corresponding candidate information and the candidate task, and the source information of the corresponding candidate information; wherein, any candidate information includes any one of the first candidate information and the second candidate information.

[0172] In one embodiment of this disclosure, the source information of the candidate information includes at least one of the following: The emotional intensity of the interactive information from which the candidate information originates; The frequency of occurrence of interactive information from the candidate information source; The task weights of the candidate tasks from which the candidate information originates.

[0173] In one embodiment of this disclosure, the importance of any candidate information is positively correlated with the correlation between the corresponding candidate information and the information of interest; or, The importance of any candidate information is positively correlated with the correlation between the corresponding candidate information and the candidate task; or... The importance of any candidate piece of information is positively correlated with the emotional intensity of the interactive information from which the corresponding candidate information originates; or, The importance of any candidate information is positively correlated with the frequency of occurrence of the interaction information from the corresponding candidate information source; or, The importance of any candidate information is positively correlated with the task weight of the candidate task from which the corresponding candidate information originates.

[0174] The memory information determination device based on a large model disclosed herein extracts information from the interaction information between the target account and the agent using a first large model to determine various first candidate information. Based on at least one of the confidence level and importance of the first candidate information, first memory information associated with the target account is determined from the first candidate information. Each memory information associated with the target account is used by the agent to execute the target task associated with the target account. Therefore, considering confidence level and / or importance, the first memory information associated with the target account can be determined from the first candidate information. For example, it can avoid using first candidate information with low confidence as the first memory information, which helps improve the accuracy of the first memory information, resulting in more accurate responses from the agent and optimized user experience. And / or, it can avoid using first candidate information with low importance as the first memory information, prioritizing information that is more important to the user, making the first memory information highly focused on the user's core needs, thus making the agent's responses more aligned with the user's core needs and optimizing user experience.

[0175] In addition, considering confidence and / or importance, the first memory information associated with the target account is determined from each first candidate information, making the first memory information more concise, avoiding memory information overload, helping to speed up task processing, thereby shortening the response time of the intelligent agent and optimizing the user experience.

[0176] According to embodiments of this disclosure, this disclosure also provides a large-model-based task processing apparatus for implementing the above-described large-model-based task processing method.

[0177] Figure 9 This is a block diagram of a large-model-based task processing apparatus according to an embodiment of the present disclosure.

[0178] like Figure 9 As shown, the task processing device 900 based on a large model includes: a first determining module 901 and a second determining module 902.

[0179] The first determining module 901 is used to determine the target task to be executed by the intelligent agent; wherein the target task is associated with the target account; The second determining module 902 is used to determine the processing result of the target task based on the memory information associated with the target account through the second major model; The memory information associated with the target account includes first memory information, which is each first candidate information extracted by the first model from the interaction information between the target account and the agent, and determined from each first candidate information based on at least one of the confidence level and the importance level of the first candidate information. The processing results of the target task are used by the terminal device for visual display and / or voice playback.

[0180] In one embodiment of this disclosure, the memory information associated with the target account further includes second memory information. The second memory information is determined by extracting second candidate information from the second interaction information using the first large model when the first interaction information and the second interaction information are obtained. The first interaction information is used to indicate that the memory information associated with the target account originates from the second interaction information.

[0181] In one embodiment of this disclosure, the second determining module 902 is further configured to: determine candidate memory information from each memory information associated with the target account based on the task information of the target task; determine target memory information from each candidate memory information based on the weight of the candidate memory information; and determine the processing result of the target task based on the target memory information using the second large model.

[0182] In one embodiment of this disclosure, the second determining module 902 is further configured to: determine the semantic matching degree between the task information of the target task and any memory information associated with the target account; determine the scene matching degree between the task scene of the target task and the task scene allowed to be used by any memory information associated with the target account; and determine the candidate memory information from each memory information associated with the target account based on at least one of the semantic matching degree and the scene matching degree.

[0183] The task processing device based on a large model disclosed herein determines the target task to be executed by the intelligent agent. The target task is associated with a target account, and the target task is executed based on the memory information associated with the target account through a second large model. Thus, the processing result of the target task is determined by the memory information associated with the target account through the second large model. The memory information associated with the target account includes first memory information, which has higher accuracy, contributing to the accuracy of the processing result of the target task; that is, the intelligent agent's response is more accurate, optimizing the user experience. And / or, the first memory information highly focuses on the user's core needs, making the processing result of the target task more aligned with the user's core needs; that is, the intelligent agent's response is more aligned with the user's core needs, optimizing the user experience.

[0184] In addition, the first memory information is more concise, which can avoid memory information overload, help to speed up the reasoning speed of the processing results of the target task, thereby shortening the response time of the intelligent agent and optimizing the user experience.

[0185] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0186] Figure 10 A schematic block diagram of an example electronic device 1000 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0187] like Figure 10 As shown, device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in read-only memory (ROM) 1002 or a computer program loaded into random access memory (RAM) 1003 from storage unit 1008. The RAM 1003 may also store various programs and data required for the operation of device 1000. The computing unit 1001, ROM 1002, and RAM 1003 are interconnected via bus 1004. Input / output (I / O) interface 1005 is also connected to bus 1004.

[0188] Multiple components in device 1000 are connected to I / O interface 1005, including: input unit 1006, such as keyboard, mouse, etc.; output unit 1010, such as various types of monitors, speakers, etc.; storage unit 1008, such as disk, optical disk, etc.; and communication unit 1009, such as network card, modem, wireless transceiver, etc. Communication unit 1009 allows device 1000 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0189] The computing unit 1001 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1001 performs the various methods and processes described above, such as a large model-based memory information determination method, and / or a large model-based task processing method. For example, in some embodiments, the large model-based memory information determination method, and / or the large model-based task processing method, can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 1008. In some embodiments, part or all of the computer program can be loaded and / or installed on device 1000 via ROM 1002 and / or communication unit 1009. When the computer program is loaded into RAM 1003 and executed by computing unit 1001, one or more steps of the large-model-based memory information determination method and / or the large-model-based task processing method described above can be performed. Alternatively, in other embodiments, computing unit 1001 can be configured by any other suitable means (e.g., by means of firmware) to perform the large-model-based memory information determination method and / or the large-model-based task processing method.

[0190] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0191] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0192] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0193] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0194] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0195] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0196] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0197] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for determining memory information based on a large model, comprising: The first model is used to extract information from the interaction between the target account and the agent to determine the first candidate information. Based on at least one of the confidence level and the importance level of the first candidate information, first memory information associated with the target account is determined from each of the first candidate information; The memory information associated with the target account is used by the intelligent agent to execute the target task associated with the target account.

2. The method according to claim 1, wherein, The step of determining the first memory information associated with the target account from each of the first candidate information based on at least one of the confidence level and the importance level of the first candidate information includes: The confidence level and the importance of any first candidate information are weighted and summed to determine the score of the corresponding first candidate information. Based on the score, the first memory information associated with the target account is determined from each of the first candidate information.

3. The method according to claim 1, wherein, The method further includes: In response to obtaining first interaction information and second interaction information; wherein, the first interaction information is used to indicate that the memory information associated with the target account originates from the second interaction information; Information is extracted from the second interaction information using the first large model to determine the second candidate information; The second candidate information is used as the second memory information associated with the target account.

4. The method according to claim 3, wherein, The method further includes: The weight of a corresponding memory information is determined based on at least one of the confidence level of any memory information and the importance level of the corresponding memory information; wherein, any memory information includes either the first memory information or the second memory information. Based on the weights, each piece of memory information is updated.

5. The method according to claim 4, wherein, The determination of the weight of the corresponding memory information based on at least one of the confidence level of any memory information and the importance level of the corresponding memory information includes: Determine the usage information for any memory information to perform the task; Determine the degree of semantic conflict between any given piece of memory and newly added piece of memory; The weight of a corresponding memory information is determined based on the confidence level of any memory information, the importance of the corresponding memory information, the usage information of the corresponding memory information, and the degree of semantic conflict of the corresponding memory information.

6. The method according to claim 5, wherein, The usage information of any memory information includes at least one of the following: The last time any memory information was used to perform a task; The time interval corresponding to any memory information; wherein, the time interval corresponding to any memory information is the time interval between the last time the corresponding memory information was used to perform a task and the current time; The frequency with which any memory information is used to perform a task.

7. The method according to claim 6, wherein, The weight of any memory information is positively correlated with the confidence level of the corresponding memory information; or, The weight of any memory information is positively correlated with the importance of the corresponding memory information; or, The weight of any memory information is negatively correlated with the time interval corresponding to that memory information. or, The weight of any memory information is positively correlated with the frequency of use of the corresponding memory information. or, The weight of any memory information is negatively correlated with the degree of semantic conflict corresponding to that memory information.

8. The method according to any one of claims 3-7, wherein, The method further includes at least one of the following: The confidence level of the first memory information is determined using the first large model; The confidence level of the second memory information is determined to be a set value; wherein the confidence level of the second memory information is greater than or equal to the confidence level of the first memory information.

9. The method according to any one of claims 3-7, wherein, The method further includes: The importance of a candidate information is determined by the first large model based on at least one of the following: the correlation between any candidate information and interest information, the correlation between the corresponding candidate information and the candidate task, and the source information of the corresponding candidate information; wherein, any candidate information includes any one of the first candidate information and the second candidate information.

10. The method according to claim 9, wherein, The source information of the candidate information includes at least one of the following: The emotional intensity of the interactive information from which the candidate information originates; The frequency of occurrence of interactive information from the candidate information source; The task weights of the candidate tasks from which the candidate information originates.

11. The method according to claim 10, wherein, The importance of any candidate information is positively correlated with the degree of correlation between the corresponding candidate information and the information of interest; or, The importance of any candidate information is positively correlated with the degree of correlation between the corresponding candidate information and the candidate task; or, The importance of any candidate information is positively correlated with the emotional intensity of the interactive information from which the corresponding candidate information originates; or, The importance of any candidate information is positively correlated with the frequency of occurrence of the interaction information from the corresponding candidate information source; or, The importance of any candidate information is positively correlated with the task weight of the candidate task from which the corresponding candidate information originates.

12. A task processing method based on a large model, comprising: Determine the target task to be executed by the intelligent agent; wherein the target task is associated with the target account; The processing result of the target task is determined by the second major model based on the memory information associated with the target account; The memory information associated with the target account includes first memory information, which is each first candidate information extracted by the first model from the interaction information between the target account and the agent, and determined from each first candidate information based on at least one of the confidence level and the importance level of the first candidate information. The processing results of the target task are used by the terminal device for visual display and / or voice playback.

13. The method according to claim 12, wherein, The memory information associated with the target account also includes second memory information. The second memory information is determined by extracting information from the second interaction information using the first big model when the first interaction information and the second interaction information are obtained. The first interaction information is used to indicate that the memory information associated with the target account originates from the second interaction information.

14. The method according to claim 12, wherein, The step of determining the processing result of the target task based on the memory information associated with the target account using the second major model includes: Based on the task information of the target task, candidate memory information is determined from the memory information associated with the target account; Based on the weights of the candidate memory information, the target memory information is determined from each of the candidate memory information; The second major model determines the processing result of the target task based on the target memory information.

15. The method according to claim 14, wherein, The step of determining candidate memory information from the memory information associated with the target account based on the task information of the target task includes: Determine the semantic matching degree between the task information of the target task and any memory information associated with the target account; Determine the scene matching degree between the task scene of the target task and the task scene allowed by any memory information associated with the target account; Based on at least one of the semantic matching degree and the scene matching degree, the candidate memory information is determined from the memory information associated with the target account.

16. A device for determining memory information based on a large model, comprising: The extraction module is used to extract information from the interaction information between the target account and the agent using the first major model, so as to determine the first candidate information. The determining module is configured to determine, based on at least one of the confidence level and the importance level of the first candidate information, the first memory information associated with the target account from each of the first candidate information; The memory information associated with the target account is used by the intelligent agent to execute the target task associated with the target account.

17. A task processing device based on a large model, comprising: A determination module is used to determine the target task to be executed by the intelligent agent; wherein the target task is associated with the target account; The execution module is used to determine the processing result of the target task based on the memory information associated with the target account through the second major model; The memory information associated with the target account includes first memory information, which is each first candidate information extracted by the first model from the interaction information between the target account and the agent, and determined from each first candidate information based on at least one of the confidence level and the importance level of the first candidate information. The processing results of the target task are used by the terminal device for visual display and / or voice playback.

18. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-15.

19. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-15.

20. A computer program product comprising a computer program that, when executed by a processor, implements the steps of the method according to any one of claims 1-15.

Images