Interaction method and device applied to large language model

Abstract

The embodiment of the invention discloses an interaction method and device applied to a large language model. The method comprises the steps of obtaining a to-be-processed language instruction input by a user through a page based on a large language model, determining and displaying at least one target personalized demand description according to a target category to which the to-be-processed language instruction belongs, and responding to the fact that the at least one target personalized demand description is selected. And adjusting the to-be-processed language instruction according to the selected target personalized demand description to generate a personalized language instruction, inputting the personalized language instruction into the large language model for processing, and outputting and displaying corresponding personalized feedback information. According to the method, a plurality of target personalized demand descriptions are provided for the user according to the category to which the to-be-processed language instruction belongs, so that the user can selectively add the target personalized demand descriptions to the to-be-processed language instruction according to the own demand, and the original instruction can be individually adjusted through man-machine interaction, so that the personalized feedback information of the large language model better conforms to the user expectation.

Description

Technical Field

[0001] The present invention relates to the technical field of large language models, and in particular, to an interaction method and device applied to large language models. Background Art

[0002] With the continuous progress of artificial intelligence technology, especially the breakthroughs in the field of deep learning, large language models, as a powerful natural language processing tool, are receiving extensive attention and applications globally. When users use large language models, they usually first briefly describe their needs. The large language model generates corresponding personalized feedback information according to the needs. After generating halfway or completely, the user finds that the generated personalized feedback information does not meet their needs, so they will terminate the conversation and readjust the need description to let the large language model generate again. This correction process will repeat many times. When users input their needs, it is indeed difficult to accurately describe them clearly at once and can only be continuously corrected during the trial-and-error process, resulting in a poor user experience. Summary of the Invention

[0003] In view of this, embodiments of the present invention provide an interaction method and device applied to large language models to provide users with multiple target personalized need descriptions for users to selectively add to the to-be-processed language instructions according to their own needs. Through the above human-computer interaction, personalized adjustment of the original instructions can be realized, so that the personalized feedback information of the large language model better meets the user's expectations and improves the user experience.

[0004] In the first aspect, an interaction method applied to a large language model is provided. The method includes:

[0005] Obtaining a to-be-processed language instruction input by a user based on the large language model usage page;

[0006] Determining and displaying at least one target personalized need description according to the target category to which the to-be-processed language instruction belongs;

[0007] In response to at least one of the target personalized need descriptions being selected, adjusting the to-be-processed language instruction according to the selected target personalized need description to generate a corresponding personalized language instruction;

[0008] Inputting the personalized language instruction into the large language model for processing, and outputting and displaying the corresponding personalized feedback information.

[0009] In the second aspect, an interaction method applied to a large language model is provided. The method includes:

[0010] Displaying a language instruction input control on the large language model usage page, where the language instruction input control includes a language input box and an input confirmation control;

[0011] Receive a to-be-processed language instruction input by a user through the language input box;

[0012] In response to the input confirmation control being triggered, display the to-be-processed language instruction in a first area on the large language model usage page, and display a list of personalized requirement descriptions in a second area on the large language model usage page, where the list of personalized requirement descriptions includes at least one target personalized requirement description, and the target personalized requirement description is determined according to the target category to which the to-be-processed language instruction belongs;

[0013] In response to at least one of the target personalized requirement descriptions being selected, display personalized feedback information in a third area on the large language model usage page, where the personalized feedback information is generated by the large language model according to a personalized language instruction, and the personalized language instruction is determined according to the to-be-processed language instruction and the selected target personalized requirement description.

[0014] In a third aspect, there is provided an interaction device applied to a large language model, the device including:

[0015] An acquisition module, configured to acquire a to-be-processed language instruction input by a user based on a large language model usage page;

[0016] A determination module, configured to determine and display at least one target personalized requirement description according to the target category to which the to-be-processed language instruction belongs;

[0017] A generation module, configured to, in response to at least one of the target personalized requirement descriptions being selected, adjust the to-be-processed language instruction according to the selected target personalized requirement description, and generate a corresponding personalized language instruction;

[0018] An output module, configured to input the personalized language instruction into the large language model for processing, and output and display the corresponding personalized feedback information.

[0019] In a fourth aspect, there is provided an interaction device applied to a large language model, the device including:

[0020] A first display module, configured to display a language instruction input control on a large language model usage page, where the language instruction input control includes a language input box and an input confirmation control;

[0021] A receiving module, configured to receive a to-be-processed language instruction input by a user through the language input box;

[0022] A second display module, configured to, in response to the input confirmation control being triggered, display the to-be-processed language instruction in a first area on the large language model usage page, and display a list of personalized requirement descriptions in a second area on the large language model usage page, where the list of personalized requirement descriptions includes at least one target personalized requirement description, and the target personalized requirement description is determined according to the target category to which the to-be-processed language instruction belongs;

[0023] A third display module, configured to, in response to at least one of the target personalized requirement descriptions being selected, display personalized feedback information in a third area on the large language model usage page, where the personalized feedback information is generated by the large language model according to a personalized language instruction, and the personalized language instruction is determined according to the to-be-processed language instruction and the selected target personalized requirement description.

[0024] In a fifth aspect, there is provided an electronic device, including a memory and a processor, where the memory is configured to store one or more computer program instructions, and wherein the one or more computer program instructions are executed by the processor to implement the method according to the first aspect or the second aspect as described above.

[0025] In a sixth aspect, there is provided a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the method according to the first aspect or the second aspect as described above is implemented.

[0026] In a seventh aspect, there is provided a computer program product, including a computer program / instructions, and when the computer program / instructions are executed by a processor, the method according to the first aspect or the second aspect as described above is implemented.

[0027] The technical solution of the embodiments of the present invention is to obtain the to-be-processed language instruction input by the user based on the large language model usage page, determine and display at least one target personalized requirement description according to the target category to which the to-be-processed language instruction belongs, in response to at least one target personalized requirement description being selected, adjust the to-be-processed language instruction according to the selected target personalized requirement description, generate a corresponding personalized language instruction, input the personalized language instruction into the large language model for processing, and output and display the corresponding personalized feedback information. The above technical solution provides multiple target personalized requirement descriptions for the user according to the category to which the to-be-processed language instruction input by the user belongs, so that the user can selectively add them to the to-be-processed language instruction according to their own needs. Through the above human-computer interaction, the original instruction can be adjusted in a personalized manner, so that the personalized feedback information of the large language model better meets the user's expectations. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Through the following description of the embodiments of the present invention with reference to the drawings, the above and other objects, features, and advantages of the present invention will become clearer. In the drawings:

[0029] Figure 1 Schematic diagram of the interaction system applied to the large language model according to the embodiment of the present invention;

[0030] Figure 2 Interaction flowchart between the user side and the server applied to the large language model according to the embodiment of the present invention;

[0031] Figure 3 Flowchart of the interaction method applied to the large language model according to the embodiment of the present invention;

[0032] Figure 4 Flowchart of the training sample set construction method according to the embodiment of the present invention;

[0033] Figure 5 Flowchart of the category determination method according to the embodiment of the present invention;

[0034] Figure 6 Flowchart of the method for determining the target personalized demand description according to the embodiment of the present invention;

[0035] Figure 7 Flowchart of the method for determining the target personalized demand description according to the embodiment of the present invention;

[0036] Figure 8 Interaction schematic diagram of the large model usage page according to the embodiment of the present invention;

[0037] Figure 9 Flowchart of the interaction method applied to the large language model according to the embodiment of the present invention;

[0038] Figure 10 Schematic diagram of the large language model usage page according to the embodiment of the present invention;

[0039] Figure 11 Schematic diagram of the interaction device applied to the large language model according to the embodiment of the present invention;

[0040] Figure 12 Schematic diagram of the interaction device applied to the large language model according to the embodiment of the present invention;

[0041] Figure 13 Schematic diagram of the electronic device according to the embodiment of the present invention. Detailed implementation manners

[0042] The following describes the present application based on embodiments, but the present application is not limited to these embodiments. In the following detailed description of the present application, some specific details are described in detail. Those skilled in the art can fully understand the present application without the description of these details. In order to avoid obscuring the essence of the present application, well-known methods, processes, flows, elements, and circuits are not described in detail.

[0043] In addition, persons of ordinary skill in the art will appreciate that the drawings provided herein are for illustration purposes and are not necessarily drawn to scale.

[0044] Unless the context clearly requires otherwise, the words "include", "comprising" and similar words throughout the application should be interpreted as including rather than exclusive or exhaustive; that is, the meaning is "including but not limited to".

[0045] In the description of this application, it should be understood that the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. In addition, in the description of this application, unless otherwise specified, the meaning of "plurality" is two or more.

[0046] Figure 1 FIG. 1 is a schematic diagram of an interactive system applied to a large language model according to an embodiment of the present invention. Figure 1 As shown, the interactive system applied to the large language model includes a user terminal 11 and a server 12.

[0047] Among them, the user terminal 11 is a general terminal that can run applications, small programs or web pages of the large language model usage platform. The terminal can be a smart phone, desktop computer, tablet computer, car computer, smart wearable device, notebook computer or other types of communication terminals. In some possible implementations, the user terminal 11 can also be a dedicated terminal with a corresponding application solidified in a dedicated integrated circuit. Among them, the open platform enables third-party developers to develop their own small programs based on the open platform by providing an application programming interface (Application Programming Interface, API) to third-party developers. The small program can specifically be a program developed on the basis of the application in the platform and used to perform corresponding operations.

[0048] The server 12 refers to a general data processing device that can provide computing or application services for a large language model usage platform. It can be a single computer, a cluster of multiple computers, or a cloud server that can flexibly adjust computing resources through cloud technology.

[0049] The user terminal 11 uses the page for human-computer interaction based on the large language model, and interacts with the server 12 through the network, thereby assisting the user to use the large language model efficiently.

[0050] Figure 2 FIG. 1 is a flowchart of the interaction between a client and a server applied to a large language model according to an embodiment of the present invention. Figure 2As shown in the figure, the interaction process between the client and the server applied to the large language model includes:

[0051] Step S201, the client 11 displays the large language model usage page.

[0052] Step S202, the client 11 obtains the language instruction to be processed input by the user.

[0053] Step S203, the client 11 sends the language instruction to be processed to the server 12.

[0054] Step S204, the server 12 determines and displays at least one target personalized requirement description according to the target category to which the language instruction to be processed belongs.

[0055] Step S205, the server 12 sends the determined target personalized requirement description to the client 11.

[0056] Step S206, the client 11 displays the received target personalized requirement description.

[0057] Step S207, the client 11 determines the selected target personalized requirement description.

[0058] Specifically, after the client 11 displays at least one target personalized requirement description, the user can select and determine one or more target personalized requirement descriptions on the page, and the client 11 can determine the target personalized requirement description selected by the user based on human-computer interaction.

[0059] Step S208, the client 11 sends the selected target personalized requirement description to the server 12.

[0060] Step S209, the server 12 adjusts the language instruction to be processed according to the selected target personalized requirement description to generate a corresponding personalized language instruction.

[0061] In a possible implementation manner, after the server 12 generates the personalized language instruction, it can directly input the personalized language instruction into the large language model for processing, so that the large language model outputs and displays the corresponding personalized feedback information.

[0062] In a possible implementation manner, after the server 12 generates the personalized language instruction, it can also send the personalized language instruction to the client 11, and the client 11 displays the personalized language instruction. After the user views the personalized language instruction, the user can determine or modify the personalized language instruction according to their own needs.

[0063] If the client 11 receives a confirmation instruction from the user based on human-computer interaction, it directly sends the original personalized language instruction to the server 12. The server 12 inputs the original personalized language instruction into the large language model for processing, so that the large language model outputs and displays the corresponding personalized feedback information.

[0064] If the client 11 receives a modification instruction from the user based on human-computer interaction, it sends the modified personalized language instruction to the server 12. The server 12 inputs the modified personalized language instruction into the large language model for processing, so that the large language model outputs and displays the corresponding personalized feedback information.

[0065] The following steps S210 - S214 will take the interaction path where the client 11 receives a confirmation instruction from the user based on human-computer interaction and directly sends the original personalized language instruction to the server 12 as an example for subsequent description. Step S210, the client 11 sends a personalized language instruction to the server 12.

[0066] Step S211, the client 11 displays the personalized language instruction.

[0067] Step S212, the client 11 obtains a confirmation instruction.

[0068] Among them, the confirmation instruction refers to the instruction generated after the user confirms the personalized language instruction.

[0069] Step S213, the client 11 sends the confirmation instruction to the server 12.

[0070] Step S214, after receiving the confirmation instruction, the server 12 inputs the personalized language instruction into the large language model for processing, so that the large language model outputs and displays the corresponding personalized feedback information.

[0071] Step S215, the server 12 sends the personalized feedback information to the client 11.

[0072] Step S211, the client 11 displays the personalized feedback information.

[0073] The above method provides multiple personalized demand descriptions for the user to choose from based on the information interaction between the client 11 and the server 12, as well as the human-computer interaction between the client 11 and the user, thereby assisting the user in constructing a personalized language instruction.

[0074] In a possible implementation manner, based on the fact that the large language model can be used offline, the process of constructing the above personalized language instruction and the large language model outputting personalized feedback information based on the personalized language instruction can also be implemented separately on the client 11. That is to say, the interaction process between the client 11 and the server 12 can be omitted, and only the human-computer interaction between the client 11 and the user is used to assist the user in efficiently using the large language model.

[0075] Optionally, during the process of the client 11 independently implementing the generation of personalized feedback information, the information interaction process between the front end (i.e., the display device) and the back end (i.e., the processor) of the client 11 can participate in the above-mentioned interaction protocol between the client 11 and the server 12, which will not be elaborated here.

[0076] For the specific implementation manners of the steps in the above method, please refer to the following embodiments, which will not be elaborated here.

[0077] The interaction system according to the embodiment of the present invention is applied to a large language model. Through interaction, it can obtain a to-be-processed language instruction input by a user based on a large language model usage page, determine and display at least one target personalized requirement description according to the target category to which the to-be-processed language instruction belongs, in response to at least one target personalized requirement description being selected, adjust the to-be-processed language instruction according to the selected target personalized requirement description, generate a corresponding personalized language instruction, input the personalized language instruction into the large language model for processing, and output and display the corresponding personalized feedback information. The interaction system of this embodiment provides multiple target personalized requirement descriptions for the user according to the category to which the to-be-processed language instruction input by the user belongs, so that the user can selectively add them to the to-be-processed language instruction according to their own needs. Through the above human-computer interaction, the original instruction can be personalized adjusted, so that the personalized feedback information of the large language model can better meet the user's expectations.

[0078] Figure 3 It is a flowchart of an interaction method applied to a large language model according to an embodiment of the present invention. As Figure 3 shown, the interaction method applied to a large language model includes the following steps:

[0079] Step S301, obtain a to-be-processed language instruction input by a user based on a large language model usage page.

[0080] Among them, the to-be-processed language instruction is an instruction proposed by the user in natural language.

[0081] Step S302, determine and display at least one target personalized requirement description according to the target category to which the to-be-processed language instruction belongs.

[0082] Among them, the categories corresponding to different language instructions may be different. For example, the categories may include data calculation, article polishing, children's education, advertising creativity, investment advice, knowledge answering, text translation, code writing, fault troubleshooting, art creation, medical advice, cooking process, event planning, and so on. The target category is the category to which the to-be-processed language instruction being processed belongs.

[0083] The personalized requirement description refers to the restrictive conditions for the language instruction to be processed, and the corresponding restrictive conditions are different for different categories. For example, when the category is code writing, the corresponding personalized requirement description can be "implemented in Python language", "implemented in C language", "lowest control complexity", "add comments to key code", or "give an input-output example", etc.

[0084] In a possible implementation manner, the target category to which the language instruction to be processed belongs can be determined based on keyword recognition technology. Specifically, keywords are extracted from the language instruction to be processed, and its category is determined according to the extracted keywords.

[0085] In a possible implementation manner, Prompt-based Learning can be used to determine the target category to which the language instruction to be processed belongs. Specifically, a reasonable category determination prompt template is designed in advance. After obtaining the language instruction to be processed, a corresponding category determination prompt is constructed based on the category determination prompt template and the language instruction to be processed, and then the category determination prompt is input into the large language model for processing to obtain the target category to which the language instruction to be processed belongs. Exemplarily, the category determination prompt template can be designed as "Please determine which of the following categories [input place for the language instruction to be processed] belongs to, and the categories include data calculation, article polishing, knowledge Q&A, text translation, code writing". If the language instruction to be processed is "Please help me write a binary search code", then the constructed category determination prompt is "Please determine which of the following categories [Please help me write a binary search code] belongs to, and the categories include data calculation, article polishing, knowledge Q&A, text translation, code writing", and this category determination prompt guides the large language model to output the type to which the language instruction to be processed belongs.

[0086] In a possible implementation manner, the target category to which the language instruction to be processed belongs can also be determined based on a pre-trained language classification model. Specifically, the language instruction to be processed is input into the pre-trained language classification model to determine the corresponding target category. Among them, the language classification model is trained based on a training sample set.

[0087] Figure 4 It is a flowchart of the method for constructing a training sample set according to an embodiment of the present invention. As Figure 4 shown, the method for constructing the training sample set includes the following steps:

[0088] Step S401, obtain historical language instructions.

[0089] Among them, the historical language instruction is the language instruction input to the large language model in the past, and the form of the language instruction is a natural language sentence. The historical language instruction can be all the language instructions previously input by all users, or the language instructions used by some users in some past time periods. The restricted time period and / or restricted user range of the historical language instruction can be adjusted according to actual needs and are not limited here.

[0090] Step S402: Cluster the historical language instructions to obtain multiple clusters.

[0091] In a possible implementation manner, the historical language instruction can be converted from a natural language sentence into a vector, and the historical language instruction is clustered based on the vector clustering technology. The vector conversion can be implemented by a bag of words model (BOW), word embeddings, term frequency-inverse document frequency (TF-IDF), or a BERT (Bidirectional Encoder Representations from Transformers) model, etc. The clustering algorithm can be K-means, hierarchical clustering, density-based spatial clustering of applications with noise (DBSCAN), or spectral clustering. Inputting the vector corresponding to the historical language instruction into the above clustering algorithm can achieve vector clustering. The above methods are only examples, and this embodiment does not limit the vector conversion method and vector clustering.

[0092] Specifically, convert the historical language instruction into a corresponding instruction vector, and cluster the instruction vector to obtain multiple clusters.

[0093] Step S403: Determine the category of the cluster according to the common attributes of each cluster.

[0094] Among them, the category refers to the category name of the cluster.

[0095] Figure 5 This is the flowchart of the category determination method according to the embodiment of the present invention. As Figure 5 shown, the category determination method includes the following steps:

[0096] Step S501: Extract a third predetermined number of historical language instructions from the historical language instruction groups corresponding to each cluster.

[0097] Among them, the third predetermined quantity can be set and adjusted according to the usage effect. The third predetermined quantity can be a predetermined value, or a quantity calculated according to a preset ratio. For example, the third predetermined quantity can be 10, 20, or 50, etc., or 10%, 20%, or 50% of the total clustering quantity, etc.

[0098] Step S502, construct a corresponding category output prompt based on the historical language instructions.

[0099] Among them, the category output prompt is used to prompt the large language model to extract the common attributes of multiple historical language instructions to determine the corresponding category.

[0100] Specifically, the category output prompt is constructed based on a preset category output prompt template and the historical language instructions extracted from the clustering clusters above.

[0101] For example, the category output prompt template is:

[0102] "Please judge which topic the following [third predetermined quantity] language instructions can be classified under?

[0103] (1), [to be input]

[0104] (2), [to be input]

[0105] (3), [to be input] ......

[0107] (third predetermined quantity), [to be input]"

[0108] The historical language instructions extracted from the clustering clusters are:

[0109] "(1), I have a dictionary containing 1 billion key-value pairs. How can I quickly count the number of key-value pairs with values greater than 1000 while minimizing memory usage?

[0110] (2), When using asyncio.gather, if a task fails, how can I capture the error without affecting the continued execution of other tasks?

[0111] (3), How can I calculate the standard deviation of each column value in a DataFrame using Pandas and extract the columns with a standard deviation higher than 2?

[0112] (4), When using multithreading to accelerate a large computing task, due to the GIL, the performance has not improved. How can I avoid it through multiprocessing or other methods?

[0113] (5), In scenarios of frequent insertion and deletion, which one has better performance, std::vector or std::deque? Please give examples.

[0114] (6) What is the type of decltype(x + y) in the following code? How to ensure that the template function supports different combinations of parameter types? ......

[0116] (n) What are the memory sizes of the following structure on 32-bit and 64-bit systems respectively? How to manually optimize its memory alignment?

[0117] The category output prompt constructed based on the pre-set category output prompt template and the historical language instructions extracted from the clustering cluster is as follows:

[0118] "Please judge which theme the following multiple (i.e., the third predetermined quantity) language instructions can be classified into?

[0119] (1) I have a dictionary containing 1 billion key-value pairs. How to quickly count the number of key-value pairs with values greater than 1000 while minimizing memory usage?

[0120] (2) When using asyncio.gather, if a task fails, how to capture the error without affecting the continued execution of other tasks?

[0121] (3) How to calculate the standard deviation of each column value in a DataFrame using Pandas and extract the columns with a standard deviation higher than 2?

[0122] (4) When using multithreading to accelerate a large computing task, due to the GIL, the performance is not improved. How to avoid it through multiprocessing or other methods?

[0123] (5) In the scenario of frequent insertion and deletion, which one has better performance, std::vector or std::deque? Please give examples.

[0124] (6) What is the type of decltype(x + y) in the following code? How to ensure that the template function supports different combinations of parameter types? ......

[0126] (n) What are the memory sizes of the following structure on 32-bit and 64-bit systems respectively? How to manually optimize its memory alignment?

[0127] Step S503, input the category output prompt into the large language model for processing, and output the category corresponding to the clustering cluster.

[0128] Specifically, the category output prompt is input into the large language model. The large language model can extract the attributes of each historical language instruction, determine the common attributes of multiple historical language instructions, and name the corresponding category of the clustering cluster with the attribute name of the common attribute. That is, the corresponding category (category name) of the clustering cluster is determined as the attribute name of the common attribute.

[0129] Among them, the common attribute extraction method can be a statistical feature analysis method, a feature selection algorithm, and / or an attribute induction method, etc.

[0130] Continuing with the above example, the output result of the large language model can be "code writing".

[0131] Step S404, determine the category of the clustering cluster as the category corresponding to each historical language instruction corresponding to the clustering cluster.

[0132] Step S405, construct a training sample set based on each historical language instruction and the corresponding category.

[0133] In a possible implementation manner, the large language model can be used to perform secondary verification on the training sample set to ensure the quality of the training sample set. When all historical language instructions are associated with the corresponding categories, the large language model is used to verify all or part of the historical language instructions. Specifically, the verification prompt can be constructed based on the pre-set verification prompt template and each historical language instruction, and then the verification prompt is input into the large language model for processing to obtain the corresponding category, and this category is compared with the category associated with the historical language instruction to determine the verification result. If the categories are the same, the verification result is passed; if the categories are different, the verification result is not passed. Finally, the training sample set is scored and screened according to the verification result. Exemplarily, the screening process is to retain the historical language instructions with the verification result passed as samples under this category, and determine the historical language instructions with the verification result not passed as the "other" category, so as to remove noise. Thus, a training sample set with better quality can be obtained.

[0134] For example, the verification prompt template can be:

[0135] "Which category does

historical language instruction to be verified

[0136] The currently to-be-verified historical language instruction is: "What are the memory sizes of the following structure on 32-bit and 64-bit systems respectively? How to manually optimize its memory alignment?"

[0137] Then the verification prompt constructed based on the pre-set verification prompt template and each historical language instruction is:

[0138] "Which category does the following question belong to: 'What are the memory sizes of the following structures on 32-bit and 64-bit systems respectively? How to manually optimize their memory alignment?' among data calculation, article polishing, knowledge Q&A, text translation, and code writing?"

[0139] Through the above method, the training sample set for training the language classification model can be obtained. Using the training sample set to train the language classification model, a classification model that can directly determine the category to which the language instruction to be processed belongs can be obtained.

[0140] Figure 6 It is a flowchart of the method for determining the target personalized demand description in the embodiments of the present invention. As Figure 6 shown, the method for determining the target personalized demand description includes the following steps:

[0141] Step S601, determine the target category to which the language instruction to be processed belongs.

[0142] Specifically, the target category is determined by the above method, which will not be elaborated here.

[0143] Step S602, obtain the historical language instructions corresponding to the target category.

[0144] Among them, the historical language instructions include the historical language instructions corresponding to the user who inputs the language instruction to be processed and the historical language instructions corresponding to other users. Similarly, the restricted time period and / or restricted user range of the historical language instructions can be adjusted according to actual needs, which will not be limited here.

[0145] Step S603, respectively determine the personalized demand descriptions corresponding to each of the historical language instructions.

[0146] Among them, each historical language instruction may correspond to one or more personalized demand descriptions.

[0147] In a possible implementation manner, a large language model can be used to determine the personalized demand descriptions corresponding to each historical language instruction in real time. Specifically, description output prompts corresponding to each of the historical language instructions are respectively constructed, and the description output prompts are used to prompt the large language model to determine the personalized demand descriptions in each of the historical language instructions, and then the description output prompts are input into the large language model for processing to output at least one personalized demand description corresponding to the historical language instruction.

[0148] Among them, the description output prompt can be constructed according to a preset description output template and the historical language instruction.

[0149] For example, the description output template can be "Analyze

Historical Language Instruction

Write a binary search code in Python

[0150] The output result of the large language model can be:

[0151] "Let's analyze this sentence and extract "what needs to be done" and "the requirements for this thing" separately

[0152] Original sentence: Write a binary search code in Python.

[0153] What needs to be done: Write a binary search code (implement a binary search algorithm).

[0154] Requirements for doing this thing: Write in Python (the code must be implemented in the Python language for summary)"

[0155] Thus, the personalized requirement descriptions corresponding to each historical language instruction can be determined in real time based on the large language model.

[0156] Optionally, before determining the personalized requirement descriptions corresponding to each historical language instruction, the historical language instructions can be refined to obtain historical language instructions with higher accuracy.

[0157] Specifically, input each of the historical language instructions into a trained language classification model to re - check the category corresponding to each of the historical language instructions.

[0158] After the above - mentioned historical language instructions are determined by the large language model and re - determined by the language classification model, the accuracy is relatively high. The higher the accuracy of each historical language instruction category, the higher the quality of all personalized requirement descriptions at the user group level, and the more closely the target personalized requirement descriptions provided for users match the actual needs of users.

[0159] In a possible implementation manner, to improve the processing speed, the above - mentioned method can be used to pre - determine the personalized requirement descriptions corresponding to each historical language instruction, and construct a corresponding mapping relationship table according to each of the historical language instructions and the corresponding personalized requirement descriptions. When the personalized requirements need to be called, determine the personalized requirement descriptions corresponding to each of the historical language instructions according to the mapping relationship table.

[0160] That is to say, a mapping relationship table is pre-constructed. When personalized requirements need to be invoked, according to the mapping relationship between historical language instructions and personalized requirements, the corresponding personalized requirement description can be found.

[0161] Optionally, the mapping relationship table can be stored according to users. The mapping relationship can be that a user corresponds to multiple historical language instructions, and each historical language instruction corresponds to at least one personalized requirement description.

[0162] Step S604, determine at least one target personalized requirement description from the personalized requirement descriptions corresponding to the historical language instructions.

[0163] In a possible implementation manner, the target personalized requirement description can be determined based on the occurrence frequency of the personalized requirements of all historical language instructions called.

[0164] In a possible implementation manner, the target personalized requirement description can be determined according to the similarity with the to-be-processed language instruction.

[0165] Figure 7 It is a flowchart of the method for determining the target personalized requirement description in the embodiment of the present invention. As Figure 7 shown, the method for determining the target personalized requirement description includes the following steps:

[0166] Step S701, determine the similarity between each of the historical language instructions and the to-be-processed language instruction respectively.

[0167] Among them, the determination of similarity can be implemented based on any similarity algorithm. This embodiment does not limit the similarity algorithm. Exemplarily, the similarity can be determined based on methods such as vector space and word embedding.

[0168] Step S702, determine the historical language instructions with the similarity higher than a predetermined similarity threshold as target historical language instructions.

[0169] Optionally, the target historical language instructions can be selected respectively from the user group level and the user level, and then personalized requirement descriptions can be provided for the user according to the historical language levels at different levels. Determining the personalized requirement description from the user level can ensure the personalization of the personalized requirement description, and at the same time determining the personalized requirement description from the user group level can help the user break away from the information cocoon and avoid amplifying and continuing the stereotypes and biases in the user data.

[0170] Specifically, the method for determining the personalized requirement description from the user group level is: determine the historical language instructions with the similarity higher than the first predetermined similarity threshold as the first target historical language instructions, and select the first predetermined number of personalized requirement descriptions from the personalized requirement descriptions corresponding to the first target historical language instructions as the personalized requirement descriptions at the group level.

[0171] The method for determining personalized requirement descriptions from the user level is as follows: Determine the second target historical language instructions corresponding to the user, which have a similarity higher than the second predetermined similarity threshold, and select a second predetermined number of personalized requirement descriptions from the personalized requirement descriptions corresponding to the second target historical language instructions as the personalized requirement descriptions at the user level.

[0172] Since the same personalized requirement descriptions may be obtained from the user group level and the user level, deduplication is also required. Specifically, deduplicate the personalized requirement descriptions at the group level and the personalized requirement descriptions at the user level to obtain the target personalized requirement descriptions.

[0173] Among them, the first predetermined similarity threshold and the second predetermined similarity threshold may be the same or different. The first predetermined number and the second predetermined number may also be the same or different.

[0174] Step S703: Select a predetermined number of personalized requirement descriptions from the personalized requirement descriptions corresponding to the target historical language instructions and determine them as the target personalized requirement descriptions.

[0175] Optionally, the predetermined number should be less than the sum of the first predetermined similarity threshold and the second predetermined similarity threshold.

[0176] Step S303: In response to at least one of the target personalized requirement descriptions being selected, adjust the to-be-processed language instruction according to the selected target personalized requirement description to generate a corresponding personalized language instruction.

[0177] Specifically, display multiple target personalized requirement descriptions on the large language model usage page, and each target personalized requirement description can be selected. If there is a to-be-processed language instruction that is selected, adjust it according to the to-be-processed language instruction to generate a personalized language instruction.

[0178] Step S304: Input the personalized language instruction into the large language model for processing, and output and display the corresponding personalized feedback information.

[0179] Among them, the large language model (Large Language Model, abbreviated as LLM) is an artificial intelligence model trained with a large amount of data, aiming to understand and generate natural language text. These models are usually based on deep learning technology and can capture the complexity and diversity of language.

[0180] In a possible implementation manner, after determining the personalized language instruction, the personalized language instruction can be directly input into the large language model. This eliminates the interactive process of user confirmation or modification, and the processing speed is relatively fast.

[0181] In a possible implementation, after generating the personalized language instruction, the personalized language instruction can be displayed to enable the user to confirm or modify the personalized language instruction. Adding the interaction process of confirming or modifying the personalized language instruction can ensure that the personalized language instruction is closer to the actual needs of the user and provide more personalized services for the user.

[0182] After displaying the generated personalized language instruction to the user, the user determines whether the current personalized requirement instruction needs to be modified according to their own needs. If no modification is required, a confirmation instruction is directly generated based on human-computer interaction. The user terminal or the server, in response to receiving the user's confirmation instruction, inputs the original personalized language instruction into the large language model for processing.

[0183] If the user believes that modification is needed, modification information is input based on human-computer interaction and a corresponding modification instruction is generated. The user terminal or the server, in response to receiving the user's modification instruction, updates the personalized language instruction according to the modification information carried in the modification instruction, and then inputs the updated personalized language instruction into the large language model for processing.

[0184] Optionally, the updated personalized language instruction can also be displayed to the user for secondary confirmation.

[0185] The following combines Figure 8 the above method is explained with reference to the

[0186] Figure 8 This is an interaction schematic diagram of the large model usage page according to an embodiment of the present invention. Figure 8 All the subgraphs shown, including subgraphs 810 - 850, are large language model usage pages.

[0187] Subgraph 810 is the initial page of the large language model usage page, and the page includes a language instruction input control 811 and a language instruction determination control 812. The language instruction input control 811 is used to receive the language instruction to be processed input by the user.

[0188] In response to detecting that the language instruction to be processed is obtained in the language instruction input control 811, the target personalized requirement description corresponding to the language instruction to be processed is determined, and subgraph 820 is displayed.

[0189] Sub - figure 820 includes multiple target personalized requirement description controls 821 and a language instruction determination control 812. The target personalized requirement description controls 821 are used to display corresponding target personalized requirement descriptions, and the target personalized requirement description controls 821 can be selected. The language instruction determination control 812 is used to determine the selected target personalized requirement description according to the selection status of each target personalized requirement description control 821 after being triggered, and then adjust the language instruction to be processed according to the selected target personalized requirement description to generate a corresponding personalized language instruction.

[0190] In a possible implementation manner, after generating the personalized language instruction, directly input the personalized language instruction into the large - language model to generate and display personalized feedback information, where the personalized feedback information can be displayed based on sub - figure 850.

[0191] In a possible implementation manner, after generating the personalized language instruction, the personalized language instruction can also be displayed to the user for the user to confirm or modify. The personalized language instruction is displayed based on sub - figure 830.

[0192] Sub - figure 830 includes a personalized language instruction display control 831 and an instruction determination control 832. The personalized language instruction display control 831 is used to display the personalized language instruction, and the personalized language instruction display control 831 is an input - enabled control, that is, the personalized language instruction display control 831 can receive the modification information input by the user.

[0193] If the personalized language instruction display control 831 does not receive the user's modification information before the instruction determination control 832 is triggered, directly input the personalized language instruction into the large - language model to generate the corresponding personalized feedback information, and the personalized feedback information is displayed based on sub - figure 850.

[0194] If the personalized language instruction display control 831 receives the user's modification information before the instruction determination control 832 is triggered, that is, the user modifies, adds, or deletes the personalized language instruction in the personalized language instruction display control 831, then after the instruction determination control 832 is triggered, input the adjusted personalized language instruction into the large - language model to generate the corresponding personalized feedback information, and the personalized feedback information is displayed based on sub - figure 850.

[0195] The method of the embodiment of the present invention is to use a page based on a large language model to obtain a to-be-processed language instruction input by a user, determine and display at least one target personalized requirement description according to the target category to which the to-be-processed language instruction belongs, and in response to the selection of at least one target personalized requirement description, adjust the to-be-processed language instruction according to the selected target personalized requirement description to generate a corresponding personalized language instruction, input the personalized language instruction into the large language model for processing, and output and display the corresponding personalized feedback information. The above method provides multiple target personalized requirement descriptions for the user according to the category to which the to-be-processed language instruction input by the user belongs, so that the user can selectively add them to the to-be-processed language instruction according to their own needs. Through the above human-computer interaction, the original instruction can be personalized adjusted to make the personalized feedback information of the large language model more in line with the user's expectations.

[0196] Figure 9 It is a flowchart of an interaction method applied to a large language model according to an embodiment of the present invention. Figure 10 It is a schematic diagram of a page for using a large language model according to an embodiment of the present invention. The following is combined with Figure 10 The page for using the large language model shown Figure 9 The method shown is explained. Figure 10 The subgraphs 1010 - 1030 shown are the page update processes when the page for using the large language model interacts with the user.

[0197] As Figure 9 shown, the interaction method applied to the large language model includes the following steps:

[0198] Step S901, display a language instruction input control on the page for using the large language model, and the language instruction input control includes a language input box and an input confirmation control.

[0199] Step S902, receive the to-be-processed language instruction input by the user through the language input box.

[0200] Step S903, in response to the trigger of the input confirmation control, display the to-be-processed language instruction in the first area on the page for using the large language model, and display a list of personalized requirement descriptions in the second area on the page for using the large language model. The list of personalized requirement descriptions includes at least one target personalized requirement description, and the target personalized requirement description is determined according to the target category to which the to-be-processed language instruction belongs.

[0201] Step S904, in response to at least one of the selected target personalized requirement descriptions, display personalized feedback information in the third area of the large language model usage page. The personalized feedback information is generated by the large language model according to a personalized language instruction, and the personalized language instruction is determined according to the to-be-processed language instruction and the selected target personalized requirement description.

[0202] For the specific implementation manners of the above steps, please refer to the above embodiments and will not be elaborated here.

[0203] For the initial state of the large language model usage page, please refer to Figure 10 sub - figure 1010 in it. The large language model usage page includes a language instruction input control 1011. The language instruction input control 1011 includes a language input box 1012 and an input confirmation control 1013. The user inputs the to - be - processed language instruction through the language input box 1012. After the input is completed, the input confirmation control 1013 is triggered. After the client 11 detects that the input confirmation control 1013 is triggered, the large language model usage page is updated to be as shown in sub - figure 1020. The to - be - processed language instruction is displayed in the first area 1023, and a list of personalized requirement descriptions is displayed in the second area 1024. The list of personalized requirement descriptions includes selectable controls corresponding to multiple target personalized requirement descriptions. The user can change the selection state of the selectable control by triggering it. If at least one of the target personalized requirement descriptions is selected, the large language model usage page is updated to be as shown in sub - figure 1030, and the personalized feedback information is displayed in the fourth area 1031.

[0204] In a possible implementation manner, the first area further includes a to - be - processed language instruction modification control, which is used to re - modify the to - be - processed language instruction after being triggered, and then jump to step S903 to redisplay the target personalized requirement descriptions and personalized feedback information updated according to the modified to - be - processed language instruction.

[0205] In a possible implementation manner, before displaying the personalized feedback information in the third area of the large language model usage page, the method further includes: displaying the personalized language instruction in the fourth area of the large language model usage page. The personalized language instruction has a corresponding instruction confirmation control and an instruction modification control. In response to the triggering of the instruction modification control, receive and display the modified information based on the personalized language instruction by the user in the third area. In response to the triggering of the instruction confirmation control, display the personalized feedback information.

[0206] Referring to the sub - figure 1030, before displaying the personalized feedback information, personalized language instructions can also be displayed in the fourth area 1032 for the user to confirm or modify. Specifically, the fourth area 1032 includes a personalized language instruction modification control 1033 and a personalized language instruction confirmation control 1034. The personalized language instruction modification control 1033 can be a specific control displayed on the page, such as a button, or an invisible control embedded in the fourth area 1032. When the user triggers the fourth area 1032 by clicking or other triggering methods, the personalized language instructions displayed in the fourth area 1032 can be modified. If the user has no other modification opinions on the original personalized language instructions or the modified personalized language instructions displayed, the personalized language instruction confirmation control 1034 can be triggered. After the personalized language instruction confirmation control 1034 is triggered, the large - language model uses the page to display the third area.

[0207] In a possible implementation manner, in addition to being able to display personalized feedback information, the fourth area can also display comparison feedback information generated according to the language instruction to be processed, so as to compare and display the personalized feedback information and the comparison feedback information.

[0208] In a possible implementation manner, the large - language model usage page shown in the sub - figure 1010 also includes a personalized service switch control. If the switch state corresponding to the personalized service switch control is on, the above steps S903 - step S904 are executed. If the switch state corresponding to the personalized service switch control is off, the language instruction to be processed is displayed in the first area of the large - language model usage page, and the comparison feedback information is displayed in the third area. The comparison feedback information is the feedback information directly generated by the large - language model according to the language instruction to be processed.

[0209] Among them, the personalized service switch control is a visible switch control, and the user can switch the switch state by clicking or dragging the "switch button". Or, the personalized service switch control is an invisible embedded control, and the user can display the personalized service switch control by clicking the mouse or setting it through the shortcut menu.

[0210] Optionally, when there is no personalized service switch control, the personalized service can also be requested in the form of a command stream. The personalized service is to generate and display the target personalized requirement description, generate personalized language instructions according to the selected target personalized requirement description, and generate personalized feedback information according to the personalized language instructions, etc.

[0211] The method of the embodiment of the present invention is to use a page based on a large language model to obtain a to-be-processed language instruction input by a user. According to the target category to which the to-be-processed language instruction belongs, at least one target personalized requirement description is determined and displayed. In response to at least one target personalized requirement description being selected, the to-be-processed language instruction is adjusted according to the selected target personalized requirement description to generate a corresponding personalized language instruction. The personalized language instruction is input into the large language model for processing, and the corresponding personalized feedback information is output and displayed. The above method provides multiple target personalized requirement descriptions for the user according to the category to which the to-be-processed language instruction input by the user belongs, so that the user can selectively add them to the to-be-processed language instruction according to their own needs. Through the above human-computer interaction, the original instruction can be adjusted personalized to make the personalized feedback information of the large language model more in line with the user's expectations.

[0212] Figure 11 It is a schematic diagram of an interaction device applied to a large language model according to an embodiment of the present invention. As Figure 11 shown, the interaction device applied to the large language model includes:

[0213] An acquisition module 1101, configured to obtain a to-be-processed language instruction input by a user using a page based on a large language model.

[0214] A determination module 1102, configured to determine and display at least one target personalized requirement description according to the target category to which the to-be-processed language instruction belongs.

[0215] A generation module 1103, configured to, in response to at least one of the target personalized requirement descriptions being selected, adjust the to-be-processed language instruction according to the selected target personalized requirement description to generate a corresponding personalized language instruction.

[0216] An output module 1104, configured to input the personalized language instruction into the large language model for processing, and output and display the corresponding personalized feedback information.

[0217] The device according to an embodiment of the present invention is used to obtain a to-be-processed language instruction input by a user based on a large language model using page, determine and display at least one target personalized requirement description according to the target category to which the to-be-processed language instruction belongs, and in response to at least one target personalized requirement description being selected, adjust the to-be-processed language instruction according to the selected target personalized requirement description, generate a corresponding personalized language instruction, input the personalized language instruction into the large language model for processing, and output and display the corresponding personalized feedback information. The above device provides a plurality of target personalized requirement descriptions for the user according to the category to which the to-be-processed language instruction input by the user belongs, so that the user can selectively add them to the to-be-processed language instruction according to their own needs. Through the above human-computer interaction, the original instruction can be personalized adjusted to make the personalized feedback information of the large language model more in line with the user's expectations.

[0218] Figure 12 It is a schematic diagram of an interaction device applied to a large language model according to an embodiment of the present invention. As Figure 12 shown, the interaction device applied to the large language model includes:

[0219] A first display module 1201, configured to display a language instruction input control on the large language model using page, where the language instruction input control includes a language input box and an input confirmation control.

[0220] A receiving module 1202, configured to receive a to-be-processed language instruction input by a user through the language input box.

[0221] A second display module 1203, configured to, in response to the input confirmation control being triggered, display the to-be-processed language instruction in a first area on the large language model using page, and display a personalized requirement description list in a second area on the large language model using page, where the personalized requirement description list includes at least one target personalized requirement description, and the target personalized requirement description is determined according to the target category to which the to-be-processed language instruction belongs.

[0222] A third display module 1204, configured to, in response to at least one of the target personalized requirement descriptions being selected, display personalized feedback information in a third area on the large language model using page, where the personalized feedback information is generated by the large language model according to the personalized language instruction, and the personalized language instruction is determined according to the to-be-processed language instruction and the selected target personalized requirement description.

[0223] The device according to an embodiment of the present invention is used to obtain a to-be-processed language instruction input by a user based on a large language model, determine and display at least one target personalized requirement description according to the target category to which the to-be-processed language instruction belongs, in response to the selection of at least one target personalized requirement description, adjust the to-be-processed language instruction according to the selected target personalized requirement description, generate a corresponding personalized language instruction, input the personalized language instruction into the large language model for processing, and output and display the corresponding personalized feedback information. The above device provides multiple target personalized requirement descriptions for the user according to the category to which the to-be-processed language instruction input by the user belongs, so that the user can selectively add them to the to-be-processed language instruction according to their own needs. Through the above human-computer interaction, the original instruction can be adjusted personalized to make the personalized feedback information of the large language model more in line with the user's expectations.

[0224] Figure 13 It is a schematic diagram of an electronic device according to an embodiment of the present invention. In this embodiment, the electronic device 1300 includes a server, a terminal, etc. As Figure 13 shown, the electronic device 1300: includes at least one processor 1301; and, a memory 1302 communicatively connected to at least one processor 1301; and, a communication component 1303 communicatively connected to a scanning device, and the communication component 1303 receives and sends data under the control of the processor 1301; wherein, the memory 1302 stores instructions executable by at least one processor 1301, and the instructions are executed by at least one processor 1301 to implement the above interaction method of the large language model.

[0225] Specifically, the electronic device includes: one or more processors 1301 and a memory 1302, Figure 13 taking one processor 1301 as an example. The processor 1301 and the memory 1302 can be connected by a bus or other means, Figure 13 taking the connection by a bus as an example. The memory 1302, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. The processor 1301 executes various functional applications and data processing of the device by running the non-volatile software programs, instructions, and modules stored in the memory 1302, that is, to implement the above interaction method of the large language model.

[0226] The memory 1302 may include a program storage area and a data storage area. The program storage area may store an operating system and application programs required for at least one function. The data storage area may store an option list and the like. In addition, the memory 1302 may include high-speed random access memory and may also include non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory 1302 may optionally include a memory remotely provided with respect to the processor 1301, and these remote memories may be connected to an external device through a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0227] One or more modules are stored in the memory 1302 and, when executed by one or more processors 1301, execute the interaction method of the large language model in any of the above method embodiments.

[0228] The above product can execute the method provided in the embodiments of the present application, and has corresponding functional modules and beneficial effects for executing the method. For technical details not described in detail in this embodiment, reference may be made to the method provided in the embodiments of the present application.

[0229] The technical solution of the embodiment of the present invention is to use a page based on a large language model to obtain a to-be-processed language instruction input by a user, determine and display at least one target personalized requirement description according to the target category to which the to-be-processed language instruction belongs, and in response to the selection of at least one target personalized requirement description, adjust the to-be-processed language instruction according to the selected target personalized requirement description to generate a corresponding personalized language instruction, input the personalized language instruction into the large language model for processing, and output and display the corresponding personalized feedback information. The above technical solution provides multiple target personalized requirement descriptions for the user according to the category to which the to-be-processed language instruction input by the user belongs, so that the user can selectively add them to the to-be-processed language instruction according to his own needs. Through the above human-computer interaction, the original instruction can be personalized adjusted so that the personalized feedback information of the large language model better meets the user's expectations.

[0230] Another embodiment of the present invention relates to a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, it implements the above-mentioned partial or all method embodiments.

[0231] Another embodiment of the present invention relates to a computer program product, including a computer program / instructions, and when the computer program / instructions are executed by a processor, it implements the above-mentioned partial or all method embodiments.

[0232] That is, those skilled in the art can understand that all or part of the steps in the methods of the above embodiments can be completed by instructing relevant hardware through a program. The program is stored in a storage medium, including several instructions to enable a device (which can be a single-chip microcomputer, a chip, etc.) or a processor to execute all or part of the steps of the methods described in the embodiments of the present application. The foregoing storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs.

[0233] The foregoing are only the preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An interactive method applied to a large language model, characterized in that: The method comprises: Using the page to obtain the language instructions to be processed input by the user based on the large language model; Determine and display at least one target personalized demand description according to the target category to which the language instruction to be processed belongs; In response to at least one of the target personalized demand descriptions being selected, adjusting the to-be-processed language instruction according to the selected target personalized demand description to generate a corresponding personalized language instruction; The personalized language instruction is input into the large language model for processing, and the corresponding personalized feedback information is output and displayed.

2. The method according to claim 1, characterized in that: The step of inputting the personalized language instruction into the large language model for processing, and outputting and displaying corresponding personalized feedback information, comprises: displaying the personalized language instruction; In response to receiving a modification instruction from a user, updating the personalized language instruction according to modification information carried in the modification instruction; In response to receiving a confirmation instruction from the user, the personalized language instruction is input into the large language model for processing.

3. The method according to claim 1, characterized in that The step of determining and displaying at least one target personalized requirement description according to the target category to which the language instruction to be processed belongs includes: Inputting the language instruction to be processed into a pre-trained language classification model to determine the corresponding target category; Obtaining historical language instructions corresponding to the target category; Determine the personalized demand description corresponding to each of the historical language instructions; respectively determining the similarity between each of the historical language instructions and the language instructions to be processed; Determine the historical language instruction whose similarity is higher than a predetermined similarity threshold as a target historical language instruction; A predetermined number of personalized demand descriptions are selected from the personalized demand descriptions corresponding to the target historical language instructions to be determined as target personalized demand descriptions.

4. The method according to claim 3, characterized in that The selecting a predetermined number of personalized demand descriptions from the personalized demand descriptions corresponding to the target historical language instructions to determine as target personalized demand descriptions includes: Determine the historical language instruction whose similarity is higher than a first predetermined similarity threshold as a first target historical language instruction; Selecting a first predetermined number of personalized demand descriptions from the personalized demand descriptions corresponding to the first target historical language instruction as group-level personalized demand descriptions; Determine the historical language instruction corresponding to the user and having a similarity higher than a second predetermined similarity threshold as a second target historical language instruction; Selecting a second predetermined number of personalized demand descriptions from the personalized demand descriptions corresponding to the second target historical language instruction as user-level personalized demand descriptions; The group-level personalized demand description and the user-level personalized demand description are deduplicated to obtain the target personalized demand description.

5. The method according to claim 3, characterized in that: The language classification model is trained based on a training sample set, and the training sample set is constructed by the following steps: Acquire a historical language instruction, where the historical language instruction is a natural language sentence; Converting the historical language instruction into a corresponding instruction vector; Clustering the instruction vectors to obtain a plurality of clusters; Extracting a third predetermined number of historical language instructions from the historical language instruction groups corresponding to each of the clusters; Building a corresponding category output prompt based on the historical language instruction, the category output prompt is used to prompt the large language model to extract common attributes of multiple historical language instructions to determine the corresponding category; Inputting the category output prompt into a large language model for processing, and outputting the category corresponding to the cluster; Determining the category of the cluster as the category corresponding to each historical language instruction corresponding to the cluster; A training sample set is constructed based on each of the historical language instructions and the corresponding category.

6. The method according to claim 5, characterized in that The method further comprises: Each of the historical language instructions is input into the language classification model to perform a secondary check on the category corresponding to each of the historical language instructions.

7. The method according to claim 3, characterized in that The step of respectively determining the personalized requirement descriptions corresponding to the historical language instructions includes: Constructing corresponding description output prompts according to each of the historical language instructions respectively, wherein the description output prompts are used to prompt the large language model to determine the personalized demand descriptions in each of the historical language instructions; The description output prompt is input into a large language model for processing, and at least one personalized demand description corresponding to the historical language instruction is output.

8. The method according to claim 3, characterized in that The method further comprises: Constructing corresponding description output prompts according to each of the historical language instructions respectively, wherein the description output prompts are used to prompt the large language model to determine the personalized demand descriptions in each of the historical language instructions; Inputting the description output prompt into a large language model for processing, and outputting at least one personalized demand description corresponding to the historical language instruction; A corresponding mapping relationship table is constructed according to each of the historical language instructions and the corresponding personalized demand description.

9. The method according to claim 8, characterized in that The step of respectively determining the personalized requirement descriptions corresponding to the historical language instructions includes: According to the mapping relationship table, the personalized requirement description corresponding to each of the historical language instructions is determined.

10. An interactive method applied to a large language model, characterized in that: The method comprises: Displaying a language instruction input control on the large language model usage page, wherein the language instruction input control includes a language input box and an input confirmation control; Receiving a language instruction to be processed input by a user through the language input box; In response to the input confirmation control being triggered, the language instruction to be processed is displayed in a first area of ​​the large language model use page, and a personalized demand description list is displayed in a second area of ​​the large language model use page, wherein the personalized demand description list includes at least one target personalized demand description, and the target personalized demand description is determined according to the target category to which the language instruction to be processed belongs; In response to at least one of the target personalized demand descriptions being selected, personalized feedback information is displayed in a third area in the large language model usage page, wherein the personalized feedback information is generated by the large language model according to personalized language instructions, and the personalized language instructions are determined according to the language instructions to be processed and the selected target personalized demand description.

11. The method according to claim 10, characterized in that Before displaying personalized feedback information in the third area of ​​the large language model usage page, the method further includes: Displaying the personalized language instruction in a fourth area of ​​the large language model use page, wherein the personalized language instruction has a corresponding instruction confirmation control and an instruction modification control; In response to the instruction modification control being triggered, receiving and displaying in the third area modification information modified by the user based on the personalized language instruction; In response to the instruction confirmation control being triggered, the personalized feedback information is displayed.

12. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method according to any one of claims 1-11.

13. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 11 is implemented.

14. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the method according to any one of claims 1 to 11 is implemented.

Images