Large model application optimization method and device based on user feedback, equipment and medium

Abstract

The application discloses a large model application optimization method and device based on user feedback, equipment and medium, and relates to the field of large model application. The method comprises the following steps: constructing a user feedback database based on user historical input information, output results of the large model to the user historical input information and user feedback information; determining feedback data instances according to user new input information and the user feedback database; generating new prompt words according to the user new input information and the feedback data instances; inputting the new prompt words into the large model to obtain output results of the large model to the user new input information. The application does not need to train and fine-tune the large model, reduces the optimization cost of the large model, can optimize the output results of the large model in real time according to historical user feedback information during the use of the large model, improves the adaptability and flexibility of the large model, and makes the output of the large model more in line with user expectations.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of large model application, in particular to a large model application optimization method and device based on user feedback, equipment and medium. BACKGROUND

[0002] With the increasing application of large models and agents, the optimization process is facing significant bottlenecks. Traditional optimization modes highly depend on model training and fine-tuning, which not only require a large amount of labeled data support, but also consume huge computing resources, resulting in high optimization costs. At the same time, the training and fine-tuning process is long, which is difficult to realize real-time response and cannot quickly adapt to the dynamic changes of user needs in actual use.

[0003] In addition, the existing optimization method does not fully utilize user feedback, and cannot timely and effectively integrate user's positive and negative feedback information into the output adjustment of large model application, resulting in that the output result of large model application sometimes cannot meet the actual expectations of users, and the user experience needs to be further improved.

[0004] In summary, the current large model has the following shortcomings: (1) poor optimization real-time performance: training and fine-tuning take a long time and cannot be optimized in real time with user use; (2) high data dependency: usually requires a large amount of manually labeled data, which is difficult and costly to obtain; (3) low feedback utilization rate: the effective information in user's positive and negative feedback has not been fully mined and applied; (4) weak universality: generally for specific model architecture, poor cross-model adaptability. SUMMARY

[0005] The purpose of the present application is to provide a large model application optimization method and device based on user feedback, equipment and medium, which can reduce the optimization cost of large model, improve the adaptability and flexibility of large model, and make the output of large model more in line with user expectations.

[0006] To achieve the above purpose, the present application provides the following solutions:

[0007] In a first aspect, the present application provides a large model application optimization method based on user feedback, comprising:

[0008] Based on user historical input information, output results of large models on user historical input information and user feedback information, a user feedback database is constructed;

[0009] Obtain new user input information, and determine feedback data instances according to the new user input information and the user feedback database;

[0010] Generate new prompt words according to the new user input information and the feedback data instances;

[0011] input the new prompt word into the large model to obtain an output result of the large model on the new input information of the user.

[0012] In a second aspect, the present application provides a large model application optimization device based on user feedback, comprising:

[0013] A database construction module is configured to construct a user feedback database based on historical input information of a user, an output result of a large model on the historical input information of the user, and user feedback information.

[0014] An instance retrieval module is configured to obtain new input information of a user, and determine feedback data instances according to the new input information of the user and the user feedback database.

[0015] A prompt word generation module is configured to generate a new prompt word according to the new input information of the user and the feedback data instances.

[0016] A large model application module is configured to input the new prompt word into the large model to obtain an output result of the large model on the new input information of the user.

[0017] In a third aspect, the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the large model application optimization method based on user feedback.

[0018] In a fourth aspect, the present application provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executable by a processor to implement the large model application optimization method based on user feedback.

[0019] According to the embodiments provided in the present application, the present application has the following technical effects:

[0020] The present application provides a large model application optimization method, device, equipment and medium based on user feedback, which adopts a whole context learning method, so that the large model adjusts the output according to the provided context information when processing the current task (i.e. the new input information of the user). Therefore, the large model does not need to be trained and fine-tuned, which reduces the optimization cost of the large model, and the output result of the large model can be optimized in real time according to the historical user feedback information during the use of the large model, which improves the adaptability and flexibility of the large model, and makes the output of the large model more consistent with the user's expectation. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0022] Figure 1 An application environment diagram of a large model application optimization method based on user feedback according to an embodiment of the present application.

[0023] Figure 2 A flowchart of a large model application optimization method based on user feedback according to an embodiment of the present application.

[0024] Figure 3 A functional module diagram of a large model application optimization device based on user feedback according to an embodiment of the present application.

[0025] Figure 4 A structural diagram of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only constitute some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0027] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0028] The large model application optimization method based on user feedback provided by the embodiments of the present application can be applied to, for example, Figure 1The application environment shown. Among them, the terminal 101 communicates with the server 102 through the network. The data storage system can store the data required by the server 102 to process. The data storage system can be set up separately, or integrated on the server 102, or placed on the cloud or other servers. The terminal 101 can send the user new input information to the server 102. After the server 102 receives the user new input information, it determines the feedback data instance according to the user new input information and the user feedback database; generate new prompt words according to the user new input information and the feedback data instance; input the new prompt words into the large model to get the output result of the large model to the user new input information. The server 102 can feed back the output result of the large model to the user new input information to the terminal 101. In addition, in some embodiments, the large model application optimization method based on user feedback can also be implemented by the server 102 or the terminal 101 alone.

[0029] Among them, the terminal 101 can be but not limited to various desktop computers, notebook computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things device can be a smart speaker, a smart TV, a smart air conditioner, a smart vehicle device, etc. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc. The server 102 can be implemented by an independent server or a server cluster composed of multiple servers, and can also be a cloud server.

[0030] In an exemplary embodiment, as Figure 2 shown, a large model application optimization method based on user feedback is provided, which is executed by a computer device, specifically by a terminal or a server computer device alone, or by a terminal and a server together. In the embodiment of the present application, the method is applied to the server 102 in Figure 1 The following steps 201 to 204 are described.

[0031] Step 201, based on the user historical input information, the output result of the large model to the user historical input information and the user feedback information, a user feedback database is constructed. Among them, the user historical input information is usually a question for C-end users, and it can also be a task instruction for B-end. For example, writing a document according to a content outline or classifying a work order into a specified category, etc.

[0032] In the present application, the user feedback database includes a good review relationship database, a bad review relationship database, a good review vector database and a bad review vector database. Among them, the good review relationship database and the bad review relationship database are both relational databases.

[0033] Each piece of data in the good review relational database comprises an association number, user historical input information, user historical input prompt words, output result of the large model on the user historical input information, and a good review label.

[0034] Each piece of data in the bad review relational database comprises an association number, user historical input information, user historical input prompt words, output result of the large model on the user historical input information, and a bad review label.

[0035] Each piece of data in the good review vector database comprises an association number, user historical input vector, and a good review label. The association number in the good review vector database is correspondingly bound to the association number in the good review relational database.

[0036] Each piece of data in the bad review vector database comprises an association number, user historical input vector, and a bad review label. The association number in the bad review vector database is correspondingly bound to the association number in the bad review relational database.

[0037] In one specific application example, step 201 comprises steps 11 to 14.

[0038] Step 11: For any user historical input information, user historical input prompt words are generated according to the user historical input information, and output result of the large model on the user historical input information and user feedback information are obtained. The user feedback information is a good review or a bad review.

[0039] Step 12: The user historical input information is converted into a vector form to obtain a user historical input vector.

[0040] Step 13: If the user feedback information is a good review, the user historical input information, the user historical input prompt words, and the output result of the large model on the user historical input information are stored as one piece of data in the good review relational database, and an association number and a good review label are generated. The user historical input vector, the association number, and the good review label are stored as one piece of data in the good review vector database.

[0041] Specifically, a structured data table is first created, and the fields thereof comprise user historical input information, user historical input prompt words, output result of the large model on the user historical input information, feedback time, and the like. Then, data storage is realized through a database management system (such as MySQL, PostgreSQL, and the like), and it is ensured that each piece of data is associated with a unique good review label.

[0042] Step 14, if the user feedback information is a bad review, the user historical input information, the user historical input prompt word, and the output result of the user historical input information by the large model are stored as a piece of data in the bad review relationship database, an association number and a bad review mark are generated, and the user historical input vector, the association number and the bad review mark are stored as a piece of data in the bad review vector database.

[0043] Specifically, first, a structured data table is created, and the fields include user historical input information, user historical input prompt word, output result of user historical input information by large model, feedback time, etc. Then data storage is realized through a database management system (such as MySQL, PostgreSQL, etc.), ensuring that each piece of data is associated with a unique bad review mark.

[0044] The present application uses an embedded large model to convert the user historical input information in the good review relationship database and the bad review relationship database into vector form and store them in the good review vector database and the bad review vector database respectively, so as to facilitate subsequent semantic retrieval.

[0045] Specifically, a text embedding large model (such as a pre-trained embedding large model such as BGE, Qwen3 Embedding, etc.) is selected, the model calling interface and vector generation related parameters (such as vector dimension, etc.) are configured, the contents of the “user historical input information” field in the good review relationship database are batch extracted, and the embedded large model is used to convert them into fixed dimension vectors, which are stored in the good review vector database (such as Milvus, etc.) in the format of “user historical input information-user historical input vector-association number”. The same process is used to process the bad review relationship database, generate a bad review vector dataset and store it in the bad review vector database, and complete the construction of vector index to support efficient semantic retrieval.

[0046] Step 202, obtaining user new input information, and determining feedback data instances according to the user new input information and the user feedback database. The feedback data instances include good review instances and bad review instances.

[0047] In one specific application example, step 202 includes steps 21 to 24.

[0048] Step 21, converting the user new input information into vector form to obtain a user new input vector. Specifically, an embedded large model is called to convert the user new input information into vector form, ensuring that the vector space is consistent with the vectors in the database.

[0049] Step 22, retrieve the top N data with the highest semantic matching degree with the user new input vector in the good review vector database and the bad review vector database respectively, to obtain N good review vector instances and N bad review vector instances. N ≥ 1, which can be set according to the actual application scenario, for example, N = 3, 5, etc.

[0050] Specifically, the retrieval enhancement generation technology (such as cosine similarity or other similarity calculation methods) is used to retrieve the user new input vector in the good review vector database and the bad review vector database respectively, to obtain the top N data with the highest semantic matching degree with the user new input vector.

[0051] In addition, a reordering model can also be introduced to reorder the semantic retrieval results, and output the top N data in descending order of scores as good review vector instances or bad review vector instances. According to the actual scene demand, the number of good review vector instances and bad review vector instances can be optimized by adjusting the value of N or setting a similarity threshold.

[0052] Step 23, retrieve the corresponding user historical input information, user historical input prompt words, and output results of the large model on the user historical input information from the good review relationship database according to the association number of each good review vector instance, to obtain good review instances.

[0053] Each good review instance is encapsulated in the format of “

good review instance n

[0054] Step 24, retrieve the corresponding user historical input information, user historical input prompt words, and output results of the large model on the user historical input information from the bad review relationship database according to the association number of each bad review vector instance, to obtain bad review instances.

[0055] Each bad review instance is encapsulated in the format of “

bad review instance n

[0056] Step 203, generate new prompt words according to the user new input information and the feedback data instance.

[0057] In one specific application instance, step 203 includes steps 31 and 32.

[0058] Step 31, generating a preliminary prompt word based on the new user input information. If the new user input information is not standardized or lacks information that needs to be supplemented, it is converted into a preliminary prompt word, otherwise the new user input information is directly taken as the preliminary prompt word. Among them, the program or large model can be used to convert the new user input information into a preliminary prompt word.

[0059] Step 32, concatenating the feedback data instance with the preliminary prompt word to obtain a new prompt word. The small sample learning method is used to concatenate the feedback data instance with the preliminary prompt word to obtain a new prompt word.

[0060] Specifically, using the small sample learning method, the preliminary prompt word is concatenated based on the preliminary prompt word, and the first N positive instance and the first N negative instance obtained from the positive vector database and the negative vector database are concatenated to form a new prompt word.

[0061] The application takes the preliminary prompt word as the basic framework, adds the instruction in the header: “Please refer to the advantages of the following positive instances and the disadvantages of the negative instances, and optimize the answer to the problem:”. Then concatenate N positive instances and N negative instances in the order of “positive instances first, negative instances second” to form a new prompt word containing context learning information.

[0062] Step 204, inputting the new prompt word into the large model to obtain the output result of the large model to the new user input information.

[0063] Specifically, the new prompt word is input into the large model (such as ChatGPT, DeepSeek, Qwen, etc.). Through the learning of positive instances and negative instances, the large model analyzes the difference characteristics (such as answer structure, detail depth, accuracy, etc.) of positive instances and negative instances, and generates a targeted optimized output result.

[0064] In another exemplary embodiment, the large model application optimization method based on user feedback further comprises the following step 205.

[0065] Step 205, obtaining feedback information (positive / negative) of the user to the output result, and storing the new user input information and the output result of the large model to the new user input information into the user feedback database based on the feedback information.

[0066] Specifically, based on the feedback information, the new user input information, the new prompt word and the corresponding output result are updated to the positive relationship database or the negative relationship database. Then, the positive vector database and the negative vector database are synchronized, the new user input information is converted into a vector through an embedded large model, and is updated to the corresponding vector database, realizing the dynamic update iteration of the retrieval data.

[0067] The application further provides good review examples, bad review examples, user new input information, and output results.

[0068] (1) Good review example.

[0069] User historical input information: "I plan to travel to Chengdu next week and want to find some suitable scenic spots for family travel, preferably with both interest and educational significance. I also want to recommend some food that the whole family can share, and hope to find accommodation near the scenic spots with convenient transportation."

[0070] User historical input prompt words: "Please recommend scenic spots in Chengdu suitable for family travel, with both interest and educational significance; recommend food suitable for the whole family to share; recommend accommodation near the scenic spots with convenient transportation."

[0071] Large model output results: "Chengdu is very suitable for family travel. The following are scenic spots, food and accommodation recommendations, with both interest, educational significance and convenient transportation.

[0072] Scenic spots: Chengdu Giant Panda Breeding Research Base is a must-visit place, where children can observe the living habits of giant pandas and learn about their growth process. The base has a beautiful environment and is very interesting. Chengdu Natural Museum is also very good, with rich natural specimens and interactive exhibitions, allowing children to learn about nature while having fun.

[0073] Food: Shu Jiuxiang Hotpot (Yulin Store) has a suitable children's yunyang pot, with fresh ingredients and thoughtful service. The whole family can enjoy delicious hotpot. There is also Long Chao Hand (Chunxi Road Store), with delicious signature hand taste and various snacks to meet different taste needs.

[0074] Accommodation: Chengdu Long Dream Hotel is close to Chengdu Giant Panda Breeding Research Base and has convenient transportation. The hotel has family rooms with complete facilities to provide a comfortable stay experience for the whole family. In addition, Chunxi Road Yada Hotel is located in the city center, close to several scenic spots and food streets, with convenient transportation and good accommodation environment."

[0075] (2) Bad review example.

[0076] User historical input information: "I plan to travel to Chengdu next week and want to find some suitable scenic spots for family travel, preferably with both interest and educational significance. I also want to recommend some food that the whole family can share, and hope to find accommodation near the scenic spots with convenient transportation."

[0077] User historical input prompt words: "Please recommend scenic spots in Chengdu suitable for family travel, with both interest and educational significance; recommend food suitable for the whole family to share; recommend accommodation near the scenic spots with convenient transportation."

[0078] The large model output result: "Chengdu has many places to go, the scenic spots are good, and the food can try the husband and wife lung piece, and the wife and husband lung piece. The accommodation is random to find a hotel."

[0079] Shortcomings: "The recommended scenic spots, Wuhou Temple and Du Fu's Thatched Cottage, are more inclined to cultural relics, and the interest and educational significance are not enough for family tours; the recommended food, husband and wife lung piece and maopo tofu, is more spicy and not suitable for children; the recommended accommodation is too general and does not reflect the characteristics of being near the scenic spot and convenient transportation."

[0080] (3) User new input information: "I plan to take my child to Xi'an for tourism next month, and want to find some suitable scenic spots for family tours, which are interesting and can learn something, and also recommend some food that the whole family can eat together, and the accommodation is better near the scenic spot and convenient transportation."

[0081] (4) Preliminary prompt words: "Please recommend scenic spots in Xi'an suitable for family tours, which require both interest and educational significance; recommend food suitable for the whole family to share; recommend accommodation near the scenic spot with convenient transportation."

[0082] (5) New prompt words: "Please refer to the advantages of the following good examples and the shortcomings of the poor examples to optimize the answer to the question.

[0083]

Good example 1

[0084] User historical input information: I plan to go to Chengdu for tourism next week, and want to find some suitable scenic spots for family tours, which are better to combine interest and educational significance, and also recommend some food that the whole family can eat together, and the accommodation is better near the scenic spot and convenient transportation.

[0085] User historical input prompt words: Please recommend scenic spots in Chengdu suitable for family tours, which require both interest and educational significance; recommend food suitable for the whole family to share; recommend accommodation near the scenic spot with convenient transportation.

[0086] High-quality output: Chengdu is very suitable for family travel. The following are recommended for you: attractions, food and accommodation, which take into account interest, educational significance and convenient transportation. Attractions: Chengdu Giant Panda Breeding Research Base is a must-see place, where children can observe the living habits of giant pandas at close range and understand their growth process. It is very educational and the environment inside the base is beautiful and interesting. Chengdu Natural Museum is also very good, with rich natural specimens and interactive exhibitions, allowing children to learn about nature in a fun way. Food: Shu Jiuxiang Hotpot (Yulin Store) has a suitable children's yunyang pot, fresh ingredients and thoughtful service, so the whole family can enjoy delicious hotpot. There is also Long Chao Hand (Chunxi Road Store), which has delicious signature hand and various snacks to meet different taste needs. Accommodation: Chengdu Dragon Dream Hotel is close to Chengdu Giant Panda Breeding Research Base and has convenient transportation. The hotel has a family room with complete facilities to provide a comfortable stay experience for the whole family. In addition, Chunxi Road Yada Hotel is located in the city center, close to many attractions and food streets, with convenient transportation and a good living environment.

[0087] [Good example 2]…

[0088] [Good example N]…

[0089] [Bad example 1]

[0090] User historical input information: I plan to travel to Chengdu next week and want to find some attractions suitable for family travel, preferably with both interest and educational significance. I also want to recommend some food that the whole family can share, and I hope the accommodation is near the attractions and has convenient transportation.

[0091] User historical input prompt words: Please recommend Chengdu attractions suitable for family travel, with both interest and educational significance. Recommend food suitable for the whole family to share. Recommend accommodation near attractions with convenient transportation.

[0092] To be improved output: Chengdu has many places to go. The attractions of Wuhou Temple and Du Fu's Thatched Cottage are good. You can try husband and wife lung slices and mapo tofu. You can just find a hotel.

[0093] Shortcomings: The recommended attractions of Wuhou Temple and Du Fu's Thatched Cottage are more inclined to cultural relics, which are not enough for family travel in terms of interest and educational significance. The recommended food of husband and wife lung slices and mapo tofu is too spicy and not suitable for children. The accommodation recommendation is too general and does not reflect the characteristics of being near the attractions and having convenient transportation.

[0094] [Bad example 2]…

[0095] [Bad example N]…

[0096] Please recommend suitable scenic spots for a family trip in Xi'an, which should have both interest and educational significance. Recommend family-friendly food. Recommend accommodations near the scenic spots with convenient transportation.

[0097] (6) Large model output result: "Xi'an is a city with rich historical heritage and suitable for family travel. Based on your needs, we recommend the following scenic spots, food, and accommodations.

[0098] Scenic spots: Shaanxi History Museum has a special children's interactive area, where children can learn about the long history of Shaanxi through lively exhibits and explanations. Xi'an Qujiang Ocean Polar Park allows children to get close to marine life and watch exciting marine animal performances.

[0099] Food: Hulusu Chicken and Squirrel Fish at Xi'an Restaurant (Bell Tower Branch) are moderately flavored and suitable for the whole family to enjoy. The restaurant has a unique environment. There is also Zhang Ji Roujiaomo (Cuihua Road Branch) in Zijin Road, which has authentic meat jiao and light congee that children will also enjoy.

[0100] Accommodation: Xi'an Westin Hotel is close to Shaanxi History Museum and has convenient transportation. The hotel has family-friendly facilities to ensure a comfortable stay for the whole family. Xi'an Qujiang International Hotel is located in Qujiang New Area, close to Xi'an Qujiang Ocean Polar Park, with convenient transportation and good accommodation conditions.

[0101] In summary, the beneficial effects of this application include at least the following points.

[0102] (1) No need for training and fine-tuning: Using the large model context learning method, there is no need for training and fine-tuning of the large model, saving a lot of labeled data and computing resources, and reducing optimization costs.

[0103] (2) Real-time optimization: Real-time optimization can be performed during the use of the large model, allowing for quick response to user needs and improving the adaptability and flexibility of the large model.

[0104] (3) Make full use of feedback data: By building a good review relationship database and a bad review relationship database, user feedback is effectively integrated into the optimization process of the large model, making the output of the large model more in line with user expectations and improving user experience.

[0105] (4) Wide range of applications: This method does not depend on a specific large model architecture and can be applied to various types of large models, making it highly versatile and scalable.

[0106] Based on the same inventive concept, the embodiments of the present application also provide a user feedback based large model application optimization device for implementing the above-mentioned method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more user feedback based large model application optimization device embodiments provided below can refer to the limitations of the method in the foregoing, which will not be described here again.

[0107] In one exemplary embodiment, as shown in Figure 3 A user feedback based large model application optimization device is provided, which includes a database construction module 301, an instance retrieval module 302, a prompt word generation module 303, and a large model application module 304.

[0108] The database construction module 301 is configured to construct a user feedback database based on user historical input information, output results of a large model on the user historical input information, and user feedback information.

[0109] The instance retrieval module 302 is configured to obtain user new input information, and determine feedback data instances according to the user new input information and the user feedback database.

[0110] The prompt word generation module 303 is configured to generate new prompt words according to the user new input information and the feedback data instances.

[0111] The large model application module 304 is configured to input the new prompt words into the large model to obtain output results of the large model on the user new input information.

[0112] In another exemplary embodiment, the user feedback based large model application optimization device further includes a database update module 305. The database update module 305 is configured to obtain feedback information (good / dissatisfied) of a user on the output results, and store the user new input information and the output results of the large model on the user new input information into the user feedback database based on the feedback information.

[0113] In an exemplary embodiment, a computer device is provided, which can be a server or a terminal, and an internal structure diagram thereof can be as shown in Figure 4As shown in the figure. The computer device includes a processor, a memory, an input / output interface (Input / Output, I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through the system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capability. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is a user feedback database. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement a large model application optimization method based on user feedback.

[0114] Those skilled in the art can understand that, Figure 4 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0115] In one exemplary embodiment, a computer readable storage medium is provided, which stores a computer program. The computer program is executed by a processor to implement the steps in each of the method embodiments described above.

[0116] In one exemplary embodiment, a computer program product is provided, which includes a computer program. The computer program is executed by a processor to implement the steps in each of the method embodiments described above.

[0117] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0118] In the present application, all actions of obtaining signals, information or data are carried out in compliance with the corresponding data protection regulations and policies of the country where the device is located, and with the authorization given by the owner of the corresponding device.

[0119] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to a memory, a database or other medium used in the embodiments provided in the present application can include at least one of a non-volatile and a volatile memory. The non-volatile memory can include a read-only memory (ROM), a magnetic tape, a floppy disk, a flash memory, an optical storage, a high-density embedded non-volatile memory, a resistive random access memory (ReRAM), a magnetoresistive random access memory (MRAM), a ferroelectric random access memory (FRAM), a phase change memory (PCM), a graphene memory, etc. The volatile memory can include a random access memory (RAM) or an external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as a static random access memory (SRAM) or a dynamic random access memory (DRAM), etc.

[0120] The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a blockchain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0121] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present application.

[0122] The principles and implementation modes of the present application are described by using specific examples in the present application. The above embodiments are only used to help understand the method and its core idea of the present application; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range can be changed. In conclusion, the content of the present application should not be understood as a limitation.

Claims

1. A large model application optimization method based on user feedback, characterized by, The method comprises: constructing a user feedback database based on user historical input information, output results of the large model on the user historical input information, and user feedback information; The user feedback database comprises a good review relationship database, a bad review relationship database, a good review vector database, and a bad review vector database; Each piece of data in the good review relationship database comprises an association number, user historical input information, a user historical input prompt word, an output result of the large model on the user historical input information, and a good review label; Each piece of data in the bad review relationship database comprises an association number, user historical input information, a user historical input prompt word, an output result of the large model on the user historical input information, and a bad review label; Each piece of data in the good review vector database comprises an association number, a user historical input vector, and a good review label; the association number in the good review vector database is correspondingly bound to the association number in the good review relationship database; Each piece of data in the bad review vector database comprises an association number, a user historical input vector, and a bad review label; the association number in the bad review vector database is correspondingly bound to the association number in the bad review relationship database; obtaining user new input information, and determining feedback data instances according to the user new input information and the user feedback database; the feedback data instances comprise good review instances and bad review instances; determining feedback data instances according to the user new input information and the user feedback database specifically comprises: converting the user new input information into a vector form to obtain a user new input vector; respectively searching the good review vector database and the bad review vector database for the first N pieces of data with the highest semantic matching degree with the user new input vector to obtain N good review vector instances and N bad review vector instances; N≥1; specifically, a retrieval enhancement generation technology is used to search the user new input vector in the good review vector database and the bad review vector database to obtain the first N pieces of data with the highest semantic matching degree with the user new input vector, or a reordering model is introduced to reorder the semantic search results, and the first N pieces of data are output in descending order of scores as good review vector instances or bad review vector instances; retrieving corresponding user historical input information, a user historical input prompt word, and an output result of the large model on the user historical input information from the good review relationship database according to the association numbers of the good review vector instances to obtain good review instances; retrieving corresponding user historical input information, a user historical input prompt word, and an output result of the large model on the user historical input information from the bad review relationship database according to the association numbers of the bad review vector instances to obtain bad review instances; generating a new prompt word according to the user new input information and the feedback data instances, specifically comprising: generating a preliminary prompt word according to the user new input information; using a small sample learning method to splice the feedback data instances and the preliminary prompt word to obtain a new prompt word; inputting the new prompt word into the large model to obtain an output result of the large model on the user new input information; Obtaining feedback information of the user on the output result, and storing the new input information of the user and the output result of the large model on the new input information of the user into the user feedback database based on the feedback information.

2. The large model application optimization method based on user feedback according to claim 1, characterized in that, Based on the historical input information of the user, the output result of the large model on the historical input information of the user and the user feedback information, a user feedback database is constructed, specifically including: For any historical input information of the user, a historical input prompt word of the user is generated according to the historical input information of the user, and the output result of the large model on the historical input information of the user and the user feedback information are obtained; the user feedback information is a good comment or a bad comment; The historical input information of the user is converted into a vector form to obtain a historical input vector of the user; If the user feedback information is a good comment, the historical input information of the user, the historical input prompt word of the user and the output result of the large model on the historical input information of the user are stored as a piece of data into a good comment relationship database, an association number and a good comment mark are generated at the same time, and the historical input vector of the user, the association number and the good comment mark are stored as a piece of data into a good comment vector database; If the user feedback information is a bad comment, the historical input information of the user, the historical input prompt word of the user and the output result of the large model on the historical input information of the user are stored as a piece of data into a bad comment relationship database, an association number and a bad comment mark are generated at the same time, and the historical input vector of the user, the association number and the bad comment mark are stored as a piece of data into a bad comment vector database.

3. A large model application optimization apparatus based on user feedback, characterized by, The device is applied to the large model application optimization method based on user feedback in any one of claims 1-2, and the device comprises: A database construction module is configured to construct a user feedback database based on historical input information of the user, output result of the large model on the historical input information of the user and user feedback information; An instance retrieval module is configured to obtain new input information of the user, and determine feedback data instances according to the new input information of the user and the user feedback database; A prompt word generation module is configured to generate a new prompt word according to the new input information of the user and the feedback data instances; A large model application module is configured to input the new prompt word into the large model to obtain an output result of the large model on the new input information of the user.

4. A computer device comprising: A memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor executes the computer program to implement the large model application optimization method based on user feedback in any one of claims 1-2.

5. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the large model application optimization method based on user feedback in any one of claims 1-2.

Images