LLM-based multi-cloud storage heterogeneous user intention recognition method and device

Abstract

The invention provides an LLM-based multi-cloud storage heterogeneous user intention recognition method and device, and belongs to the technical field of cloud computing. The method comprises the following steps: collecting a user demand identification data set through a multi-cloud platform, and setting a unified data management format; constructing a professional keyword dictionary, and judging whether the user is a professional user or not according to user query content; dividing the data set into four sub-data sets according to an interaction information field and a user background field in the data management format; performing data enhancement on the non-professional user sub-data set until the coverage of the data set reaches a set threshold value; and training the LLM model by using the sub-data sets D1 and D2 and the data-enhanced sub-data sets D3 and D4 to obtain a fine-tuned model F for multi-cloud storage heterogeneous user intention recognition. According to the method, a multi-source heterogeneous multi-cloud environment and diversified user intentions are considered, and the intention recognition capability of heterogeneous users in the multi-cloud storage environment is remarkably improved.

Description

Technical Field

[0001] The present invention relates to a multi-cloud storage heterogeneous user intention recognition method and device based on LLM, belonging to the technical field of cloud computing. Background Art

[0002] In the field of cloud computing, multi-cloud platforms have emerged to meet the needs of cloud-network integration for future service customization networks. Cloud resources usually include three categories: computing, storage, and network. Among them, storage resources are an important component. At present, the management and control of cloud resources such as storage is mainly manual, which requires high professional knowledge from users and is inefficient. Therefore, the problem of intelligent management and control of multi-cloud storage needs to be solved urgently.

[0003] The prerequisite for intelligent management and control of multi-cloud storage is to correctly identify the true intentions implied by users. Multi-cloud platform users are not only practitioners with professional knowledge, but also non-professionals. These two types of users need more or less assistance from multi-cloud platform developers when deploying multi-cloud platforms. Developers need to help users convert their intentions into specific multi-cloud parameters before using the multi-cloud platform. Due to the differences in user industries, knowledge backgrounds, and application scenarios, their intention expressions are diverse.

[0004] At present, the related research on intent recognition usually adopts reinforcement learning and non-generative deep learning methods. Although this method has certain learning capabilities, it is difficult to apply to multi-source heterogeneous multi-cloud environments. The large language model LLM is a powerful artificial intelligence algorithm that learns the grammar, semantics and contextual information of the language by training a large amount of text data, so that it can model natural language text. This model has a wide range of applications in the field of natural language processing (NLP), including text generation, text classification, machine translation, sentiment analysis, etc. At the same time, some studies have adopted LLM-based methods, but most of the related methods rely on advanced general LLMs such as GPT 4.0 and use prompt methods. Although this method can complete the function of intent recognition, its cost and controllability are limited. Summary of the invention

[0005] The present invention aims to provide a method and device for multi-cloud storage heterogeneous user intention recognition based on LLM, thereby improving the intention recognition capability of heterogeneous users in a multi-cloud storage environment.

[0006] In order to achieve the above object, the present invention is implemented through the following technical solutions: A multi-cloud storage heterogeneous user intention recognition method based on LLM includes the following steps: Collect user demand identification data sets through multi-cloud platforms and set a unified data management format, including general fields, interactive information fields, and user background fields; Build a professional keyword dictionary, compare the user query content with the professional keyword dictionary, calculate the proportion of professional vocabulary, and if it exceeds the set threshold, mark the user background field as a professional user, otherwise mark it as a non-professional user; According to the interactive information field and user background field in the data management format, the dataset is divided into four sub-datasets: the single intention of professional users is sub-dataset D1, the compound intention of professional users is sub-dataset D2, the single intention of non-professional users is sub-dataset D3, and the compound intention of non-professional users is sub-dataset D4; the data in each sub-dataset includes the user's natural language intention expression, storage mode and basic tasks; Data enhancement is performed on the single intent sub-dataset D3 of non-professional users and the compound intent sub-dataset D4 of non-professional users until the coverage of the data set reaches the set threshold; The LLM model is trained using sub-datasets D1, D2, and data-enhanced sub-datasets D3 and D4 to obtain a fine-tuned model F for heterogeneous user intent recognition in multi-cloud storage.

[0007] Preferably, the general fields include timestamp, platform, user ID, session ID and request ID; the interaction information field includes interaction type, query content, number of intents, intent recognition; and the intent recognition classification format is storage mode-basic task.

[0008] Preferably, the storage formats include block storage, object storage and file storage, and the basic tasks include single basic tasks and multiple basic tasks.

[0009] Preferably, the data enhancement includes: grammatical structure transformation, vocabulary transformation, introduction of text errors, and mixing of uppercase and lowercase letters; The grammatical structure changes include adjusting the order of subject, predicate and object; The vocabulary transformation includes synonym replacement and word redundancy to generate non-professional expressions; The introduced text errors include errors that violate Chinese or English grammatical rules in written expressions. Preferably, the coverage calculation formula is as follows: , in, The number of types of expression enhancement transformations, For the dataset The number of unique patterns in , describing the actual diversity of the dataset, The number of data corresponding to different enhanced transformation sets, is the weight, indicating the The importance of class transformation, Indicates the number of transforms for each augmented transform type.

[0010] Preferably, the LLM model training loss function is as follows: , in, is the standard cross entropy loss for the professional user dataset, is the standard cross entropy loss for the non-professional user dataset, is the alignment semantic loss for the professional user dataset, is the alignment semantic loss for the non-professional user dataset, is the confidence-regulated loss, , is the dynamic weight, , and represents the weight balancing parameter, and They are used to adjust the influence of semantic alignment loss for professional and non-professional users, Represents the weight of confidence control of non-professional users.

[0011] Preferably, the weight balancing parameter The calculation is as follows: , , Weight balancing parameters The calculation is as follows: , , in, , , , are the proportions of samples in data sets D1, D2, D3, and D4 respectively; The value range of weight balance parameter is: .

[0012] Preferably, the standard cross entropy loss of the professional user dataset is as follows: The standard cross entropy loss for the non-expert user dataset is as follows: in, represents the number of samples in the mini-batch data for calculating the classification loss of the professional user dataset, represents the true probability value of the jth sample in category k, represents the probability that the jth sample in the professional user dataset belongs to the kth class, represents the number of samples in the mini-batch data for computing the classification loss of the non-professional user dataset, represents the probability that the jth sample in the non-professional user dataset belongs to the kth class, Indicates the number of elements in the basic task set; The alignment semantic loss of the professional user dataset is as follows: , in, is the number of professional user samples in the mini-batch data, is the semantic embedding of professional user samples, Target semantic embedding corresponding to professional user samples; The alignment semantic loss of the non-professional user dataset is as follows: , in is the number of professional user samples in the mini-batch data, is the semantic embedding of professional user samples, Target semantic embedding corresponding to professional user samples; The confidence control loss is as follows: , in, is the confidence threshold, is the highest class confidence in the prediction of the jth sample of the model pair.

[0013] Preferably, the dynamic weight is calculated as follows: , in, is the current training round, is the total number of training rounds, Initial weight for professional users, is the initial weight of non-professional users, and . A data set construction device for a multi-cloud storage heterogeneous user intention recognition method based on LLM comprises a processor and a memory storing program instructions, wherein the processor is configured to execute the multi-cloud storage heterogeneous user intention recognition method based on LLM when running the program instructions.

[0014] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the LLM-based multi-cloud storage heterogeneous user intention recognition method.

[0015] The advantages of the present invention are: (1) Multi-level user intent recognition By modeling the multi-objective intent recognition problem, the model can effectively respond to diverse user requests in a multi-cloud storage environment. This method not only improves the accuracy of the model, but also enhances the model's ability to identify different types of user needs, thereby achieving more efficient resource management and service optimization in complex environments. In particular, for non-professional users, it can better understand and respond to their diverse needs.

[0016] (2) Special optimization for non-professional users In view of the spelling errors, colloquial expressions and non-standard language that non-professional users may make, this method improves the ability to understand the intentions of non-professional users through data augmentation technology. By introducing a dynamic loss adjustment strategy for non-professional users, it can flexibly adapt to the user's expression and ensure accurate intention recognition even in the case of non-standard language.

[0017] (3) Flexible coverage evaluation mechanism In order to ensure the diversity and coverage of the dataset, this method proposes an innovative coverage evaluation mechanism. By evaluating the diversity of the dataset in terms of grammar, vocabulary, and expression, it ensures that the model can cover as wide a range of user expression scenarios as possible. This mechanism enables the model to cope with various potential language deviations or non-standard expressions when processing practical applications, improving the stability and reliability of the model.

[0018] (4) Adaptive fine-tuning strategy This method uses an adaptive fine-tuning loss function to dynamically adjust the loss weight at different stages according to the user's background and intent type. This fine-tuning strategy enables the model to flexibly adjust the optimization direction according to the characteristics of the training data, thereby avoiding the overfitting or underfitting problems that may be caused by fixed loss weights in traditional methods. This adaptive fine-tuning method not only improves accuracy, but also enhances the generalization ability of the model, and can cope with various complex practical scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0020] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] Example 1 like Figure 1 As shown, a multi-cloud storage heterogeneous user intention recognition method based on LLM includes the following steps: S1: Collect user demand identification data sets through the multi-cloud platform and set a unified data management format, namely the log management structure, including general fields, interaction information fields, and user background fields.

[0023] As a refinement of the above embodiment, the specific goal of data collection is determined to be user demand identification, the multi-cloud platform includes Alibaba Cloud, Tencent Cloud, Huawei Cloud, AWS Cloud, Microsoft Cloud and Google Cloud, and the specific types of interaction data to be collected are conversation content and search queries, etc. Use common log aggregation tools (such as Elastic Stack, Splunk) and the log management services provided by the cloud platform (such as AWS CloudTrail, Google Cloud Logging, Azure Monitor, etc.) to set the same log management structure for conversation and query records. The structure contains general fields and interaction information fields.

[0024] The general fields are divided into the following parts: timestamp, platform, user ID, session ID, and request ID. The platform indicates the cloud platform where the log comes from, such as AWS, Azure, Huawei, etc., in order to distinguish between multiple cloud sources.

[0025] The interactive information field is divided into the following parts: Interaction Type: Distinguish the type of interaction, such as "conversation query", "search query", "API request", etc.

[0026] Query content: the specific content entered by the user, such as conversation messages, search keywords, etc.

[0027] Number of intents: Classify user intents into compound intents and single intents.

[0028] Intent recognition: Intent recognition corresponding to user input content. The classification format is: storage mode (M)-basic task (t). For example: object storage-add cloud disk.

[0029] Specifically, the basic task set of multi-cloud platform storage resources is defined as According to the storage type of the platform, it can be divided into three storage modes: block storage, object storage, and file storage. A collection of basic tasks related to block storage It has functions of adding cloud disks, mounting cloud disks, unmounting cloud disks, expanding cloud disks, deleting cloud disks, etc. A collection of basic tasks related to object storage It can be broken down into adding an object storage bucket, setting bucket access control permissions, deleting an object storage bucket, etc. Basic file storage task set It can be decomposed into adding access groups, modifying access groups, deleting access groups, etc. According to the storage type, the result after the user's natural language intention expression x is recognized It could be: ,in, , The fine-tuned model.

[0030] because It may contain a single basic task or two or more basic block storage tasks. Possibly a single intention , or possibly a composite intent .but It can be further expressed as: .

[0031] in, Corresponding to the single basic task of large-scale multi-cloud exchange platform block storage resources, Corresponding to two or more basic tasks of block storage resources on large-scale multi-cloud exchange platforms.

[0032] Similarly, and They correspond to the single basic tasks of platform object storage and file storage respectively. and They correspond to the multiple basic tasks of platform object storage and file storage resources. and It can be further expressed as: ; .

[0033] S2: Build a professional keyword dictionary, compare the user query content with the professional keyword dictionary, calculate the proportion of professional vocabulary, and if it exceeds the set threshold, mark the user background field as a professional user, otherwise mark it as a non-professional user. The threshold in this embodiment is 10%. During the application process, the threshold can be modified as needed.

[0034] As a refinement of the above embodiment, the following steps are specifically included: Build a professional keyword dictionary. Collect and organize professional terms in the field of cloud computing, including technical documents, professional papers, industry reports, and API documents. Classify the vocabulary by storage mode to facilitate subsequent management and expansion.

[0035] Define a professional threshold. The professional threshold is defined as the ratio of professional keywords to the total number of query words. If the user's query content is considered a professional query.

[0036] Traverse the word segmentation results, compare each word with the professional keyword dictionary, and record the matched professional vocabulary.

[0037] Count the number of professional keywords matched in the query content and calculate their frequency.

[0038] If the proportion P of professional vocabulary in the query content exceeds the set threshold, it is marked as a "professional user", otherwise it is marked as a "non-professional user".

[0039] Mark the user background field as professional user, otherwise mark it as non-professional user.

[0040] S3: According to the interaction information field and user background field in the data management format, the dataset is divided into four sub-datasets: the single intention of professional users is sub-dataset D1, the compound intention of professional users is sub-dataset D2, the single intention of non-professional users is sub-dataset D3, and the compound intention of non-professional users is sub-dataset D4; the data in each sub-dataset contains the user's natural language intention expression, storage mode and basic tasks.

[0041] As a refinement of the above embodiment, the log management structure is formatted after the division is completed, wherein the formatted data set refers to the format of the annotated data set used by the benchmark model.

[0042] S4: Data enhancement is performed on the sub-dataset D3 for the single intention of non-professional users and the sub-dataset D4 for the compound intention of non-professional users until the data set coverage reaches a set threshold. The threshold in this embodiment is 10%. During the application process, the threshold can be modified as needed.

[0043] The aim is to improve the coverage and diversity of the sub-dataset for non-expert users.

[0044] As a refinement of the above embodiment, the enhanced transformation set is: .

[0045] (1) Grammatical Structure Transformation Grammatical structure changes refer to disrupting the order of the subject, predicate, and object of a sentence, such as help me add cloud disk → help me add cloud disk.

[0046] (2) Vocabulary transformation Refers to the generation of non-professional expressions through synonym replacement and word redundancy. The difference between professional expressions and non-professional expressions is mainly reflected in the accuracy of terms, technicality and professionalism of context. Professional expressions usually use industry-recognized terms and concepts, which have clear definitions and uses and can accurately describe cloud computing-related technologies, services and concepts. Non-professional expressions may use more popular and colloquial words, which may not be precise enough but are easier for non-professionals to understand. For example: the professional expression is cloud storage service, and the corresponding non-professional expression is stored online.

[0047] Specifically, the basis for dividing professional expressions and non-professional expressions is whether the expression is a term in the cloud computing keyword dictionary. Currently, the public dictionaries include Microsoft's Azure Cloud Computing Dictionary (https: / / www.clouddefense.ai / glossary / azure) and IBM Cloud Computing Glossary (https: / / cloud.ibm.com / docs / overview?topic=overview-glossary&locale=zh-CN).

[0048] Word redundancy achieves the transformation of professional expressions into non-professional expressions, which means explaining professional terms in more understandable language without sacrificing information transmission. This method includes synonym replacement, as well as definitions, explanations, examples or analogies to help understand complex or professional terms. Example: The professional expression API is transformed into "Application Programming Interface" through word redundancy, which is like a bridge between different software, allowing them to "talk" to each other and exchange information.

[0049] (3) Introducing text errors Refers to but is not limited to inappropriate use of words, inappropriate collocation, incomplete components, and redundant components.

[0050] For example: Storage Bucket → Storage Pass, NFS → NFC, Help me add a cloud disk → Help add a disk, I plan to store data in the cloud → I plan to store data in the network cloud.

[0051] The enhanced non-professional user datasets D3 and D4 are marked as .

[0052] The coverage index R is used to measure the enhanced data set Whether the requirements of coverage diversity and complexity are met, the coverage calculation formula is as follows: , in, The number of types of expression-enhancing transformations, such as grammar, vocabulary, spelling, colloquialism, etc., For the dataset The number of unique patterns in , describing the actual diversity of the dataset, The number of data corresponding to different enhanced transformation sets, is the weight, indicating the The importance of class transformation, represents the number of transformations of each enhanced transformation type, and the denominator represents the total number of all possible non-specialized expression patterns. Indicates the number of transforms for each augmented transform type.

[0053] The closer R is to 1, the better the generated dataset can simulate the extensiveness of non-professional expressions; the closer R is to 0, the more the dataset is too single and cannot represent the diverse expressions of non-professional users.

[0054] Setting the coverage threshold ,like Then the data enhancement ends. Otherwise, continue the data enhancement process.

[0055] S5: Use sub-datasets D1, D2, and data-enhanced sub-datasets D3 and D4 to train the LLM model and obtain the fine-tuned model F for multi-cloud storage heterogeneous user intent recognition.

[0056] As a refinement of the above embodiment, the LLM model training loss function as follows: , in, is the standard cross entropy loss for the professional user datasets (D1 and D2), Standard cross entropy loss for non-expert user datasets (D3 and D4) to handle irregular expressions; is the alignment semantic loss of the professional user datasets (D1 and D2), is the alignment semantic loss of the non-professional user datasets (D3 and D4). The alignment semantic loss is used to constrain the consistency of the output semantic embedding and the target semantic embedding; To adjust the confidence loss, the prediction confidence of non-professional users is adjusted to avoid the model from incorrectly amplifying low-confidence samples; , is the dynamic weight, , and represents the weight balancing parameter, and They are used to adjust the influence of semantic alignment loss for professional and non-professional users, Represents the weight of confidence control of non-professional users.

[0057] , Both are greater than or equal to 0. There is no rigid normalization requirement, but it can be set according to the ratio of the number of professional user data sets (D1 and D2) and non-professional user data sets (D3 and D4).

[0058] , , and are the proportions of samples in data sets D1, D2, D3 and D4 respectively Greater than or equal to 0, no rigid normalization requirement, but usually the value range is .

[0059] The standard cross entropy loss for the professional user dataset is as follows: The standard cross entropy loss for the non-expert user dataset is as follows: Where M represents the number of samples in the mini-batch data for calculating the classification loss, Represents the number of elements in the basic task set, It refers to the probability that the jth sample belongs to the kth class.

[0060] The alignment semantic loss of the professional user dataset is as follows: , in, is the number of professional user samples in the mini-batch data, is the semantic embedding of professional user samples, Target semantic embedding corresponding to professional user samples; The alignment semantic loss of the non-professional user dataset is as follows: , in is the number of professional user samples in the mini-batch data, is the semantic embedding of professional user samples, Target semantic embedding corresponding to professional user samples; The confidence control loss is as follows: , in, is the confidence threshold, usually 0.9, is the highest category confidence in the jth sample corresponding to the model. This mechanism is used to ensure that for datasets of non-professional users (D3 and D4), the model optimization direction will focus on improving the reliability of predictions rather than blindly increasing the recognition accuracy.

[0061] Specifically, the dynamic weight calculation method is as follows: , in, is the current training round, is the total number of training rounds, Initial weight for professional users, is the initial weight of non-professional users, and , to ensure the training optimization of non-professional user datasets.

[0062] In the early stages of training ( ), Smaller, Larger, the fine-tuning of the model focuses more on non-professional expressions Understanding and optimization of ), and Gradually approach and reach a balance, realize the balanced optimization of professional users and non-professional users, and enhance the generalization ability of the model.

[0063] Finally, we get the fine-tuned model F, which is: ,in, The embodiment of the present disclosure also provides a multi-cloud storage heterogeneous user intention identification method device based on LLM, including a processor and a memory. Optionally, the device may also include a communication interface and a bus. Among them, the processor, the communication interface, and the memory can communicate with each other through the bus. The communication interface can be used for information transmission. The processor can call the logic instructions in the memory to execute the multi-cloud storage heterogeneous user intention identification method based on LLM of the above embodiment.

[0064] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product.

[0065] The memory, as a computer-readable storage medium, can be used to store software programs and computer executable programs, such as program instructions / modules corresponding to the method in the embodiment of the present disclosure. The processor executes the functional application and data processing by running the program instructions / modules stored in the memory, that is, the multi-cloud storage heterogeneous user intention recognition method based on LLM in the above embodiment is implemented.

[0066] The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required for at least one function; the data storage area may store data created according to the use of the terminal device, etc. In addition, the memory may include a high-speed random access memory and may also include a non-volatile memory.

[0067] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-cloud storage heterogeneous user intention recognition method based on LLM, characterized in that: The following steps are involved: Collect user demand identification data sets through multi-cloud platforms and set a unified data management format, including general fields, interactive information fields, and user background fields; Build a professional keyword dictionary, compare the user query content with the professional keyword dictionary, calculate the proportion of professional vocabulary, and if it exceeds the set threshold, mark the user background field as a professional user, otherwise mark it as a non-professional user; According to the interactive information field and user background field in the data management format, the dataset is divided into four sub-datasets: the single intention of professional users is sub-dataset D1, the compound intention of professional users is sub-dataset D2, the single intention of non-professional users is sub-dataset D3, and the compound intention of non-professional users is sub-dataset D4; the data in each sub-dataset includes the user's natural language intention expression, storage mode and basic tasks; Data enhancement is performed on the single intent sub-dataset D3 of non-professional users and the compound intent sub-dataset D4 of non-professional users until the coverage of the data set reaches the set threshold; The LLM model is trained using sub-datasets D1, D2, and data-enhanced sub-datasets D3 and D4 to obtain a fine-tuned model F for heterogeneous user intent recognition in multi-cloud storage.

2. The LLM-based multi-cloud storage heterogeneous user intention recognition method according to claim 1 is characterized in that: The general fields include timestamp, platform, user ID, session ID and request ID; the interaction information field includes interaction type, query content, number of intents, and intent recognition; the intent recognition classification format is storage mode-basic task.

3. The LLM-based multi-cloud storage heterogeneous user intention identification method according to claim 2 is characterized in that: The storage formats include block storage, object storage and file storage, and the basic tasks include single basic tasks and multiple basic tasks.

4. The LLM-based multi-cloud storage heterogeneous user intention identification method according to claim 1 is characterized in that: The data enhancement includes: grammatical structure transformation, vocabulary transformation, introduction of text errors, and mixing of uppercase and lowercase letters; The grammatical structure changes include adjusting the order of subject, predicate and object; The vocabulary transformation includes synonym replacement and word redundancy to generate non-professional expressions; The introduced text errors include errors that violate Chinese or English grammatical rules in written expressions.

5. The LLM-based multi-cloud storage heterogeneous user intention identification method according to claim 4 is characterized in that: The coverage calculation formula is as follows: ， in, The number of types of expression enhancement transformations, For the dataset The number of unique patterns in , describing the actual diversity of the dataset, The number of data corresponding to different enhanced transformation sets, is the weight, indicating the The importance of class transformation, Indicates the number of transforms for each augmented transform type.

6. The LLM-based multi-cloud storage heterogeneous user intention identification method according to claim 1 is characterized in that: LLM model training loss function as follows: ， in, is the standard cross entropy loss for the professional user dataset, is the standard cross entropy loss for the non-professional user dataset, is the alignment semantic loss for the professional user dataset, is the alignment semantic loss for the non-professional user dataset, is the confidence-regulated loss, , is the dynamic weight, , and represents the weight balancing parameter, and They are used to adjust the influence of semantic alignment loss for professional and non-professional users, Represents the weight of confidence control of non-professional users.

7. The LLM-based multi-cloud storage heterogeneous user intention identification method according to claim 6 is characterized in that: Weight balancing parameters The calculation is as follows: ， ， Weight balancing parameters The calculation is as follows: ， ， in, , , , are the proportions of samples in data sets D1, D2, D3 and D4 respectively.

8. The LLM-based multi-cloud storage heterogeneous user intention identification method according to claim 6 is characterized in that: The standard cross entropy loss for the professional user dataset is as follows: The standard cross entropy loss for the non-expert user dataset is as follows: in, represents the number of samples in the mini-batch data for calculating the classification loss of the professional user dataset, represents the true probability value of the jth sample in category k, represents the probability that the jth sample in the professional user dataset belongs to the kth class, represents the number of samples in the mini-batch data for computing the classification loss of the non-professional user dataset, represents the probability that the jth sample in the non-professional user dataset belongs to the kth class, Indicates the number of elements in the basic task set; The alignment semantic loss of the professional user dataset is as follows: ， in, is the number of professional user samples in the mini-batch data, is the semantic embedding of professional user samples, Target semantic embedding corresponding to professional user samples; The alignment semantic loss of the non-professional user dataset is as follows: ， in is the number of professional user samples in the mini-batch data, is the semantic embedding of professional user samples, Target semantic embedding corresponding to professional user samples; The confidence control loss is as follows: ， in, is the confidence threshold, is the highest class confidence in the prediction of the jth sample of the model pair.

9. The LLM-based multi-cloud storage heterogeneous user intention identification method according to claim 8 is characterized in that: The dynamic weight calculation method is as follows: ， in, is the current training round, is the total number of training rounds, Initial weight for professional users, is the initial weight of non-professional users, and .

10. A data set construction device for a multi-cloud storage heterogeneous user intention recognition method based on LLM, comprising a processor and a memory storing program instructions, characterized in that: The processor is configured to execute the LLM-based multi-cloud storage heterogeneous user intention recognition method as described in any one of claims 1-9 when running the program instructions.

Images