Mobile cloud disk service processing method, device, equipment, medium and product

Abstract

The invention discloses a mobile cloud disk service processing method and device, equipment, a medium and a product, and the method comprises the steps: inputting an obtained input sequence into a preset multi-modal model, and analyzing a processing demand of a user; determining to-be-processed data in a mobile cloud disk according to the processing demand; and inputting the to-be-processed data into a preset multi-mode machine learning model for processing, and outputting a business processing result. According to the scheme, user input can be accurately understood, and the service processing result is accurately output.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cloud computing, in particular to a mobile cloud disk service processing method, device, equipment, medium and product. BACKGROUND

[0002] The mobile cloud disk is a cloud disk product, which aims to provide users with safe, intelligent and unlimited-speed cloud storage services. The product not only provides users with free cloud storage services, but also supports multi-end synchronization, online management and group sharing functions, and is committed to becoming a digital asset management center for users and families in the 5G era. The mobile cloud disk not only has data storage capabilities such as photo albums, audio and video, but also can have AI conversations and intelligent generation based on user input requirements, in order to improve the accuracy of mobile cloud disk processing 5G services and improve the user's cloud disk experience. Based on cloud computing, the processing of 5G services in the mobile cloud disk has gradually become a key research direction.

[0003] The prior art mainly trains multiple single-modal models for different tasks using machine learning algorithms according to the functions of the mobile cloud disk, and processes the services received by the cloud disk based on the trained models. However, due to the richness of human language, the prior art cannot accurately understand user input, and the AI tasks in the mobile cloud disk often need to combine multiple data to generate user answers, which leads to inaccurate output. SUMMARY

[0004] In order to solve the above problems, the present application provides a mobile cloud disk service processing method, device, equipment, medium and product, which can accurately understand user input and accurately output service processing results.

[0005] The embodiment of the present application provides a mobile cloud disk service processing method, which comprises the following steps: inputting the obtained input sequence into a preset multi-modal model to analyze the processing demand of the user; determining the data to be processed in the mobile cloud disk according to the processing demand; inputting the data to be processed into a preset multi-modal machine learning model for processing, and outputting a service processing result.

[0006] Preferably, the step of inputting the obtained input sequence into a preset multi-modal model to analyze the processing demand of the user comprises the following steps: obtaining an input sequence input by the user; vectorizing each word in the text of the input sequence to obtain an input vector; adding a time stamp to the input vector of each word according to the order of each word in the input sequence; inputting the input vector with the added timestamp into a multi-head attention mechanism, adopting a preset multi-head attention mechanism algorithm, calculating a weighted sum of each word based on the correlation of each word in the input sequence with all words, and obtaining an output vector of multi-head attention; adopting a preset processing layer to analyze the processing demand of the user according to the input vector and the output vector.

[0007] Preferably, the processing layer includes an encoder that encodes the input sequence to obtain the semantics and relationships in the input sequence.

[0008] Preferably, the adopting a preset processing layer to analyze the processing demand of the user according to the input vector and the output vector includes: residual connecting and normalizing the input vector and the output vector to obtain a normalized result; adding a timestamp to the normalized result based on the output time of each word; inputting the normalized result with the added timestamp into two preset neural network layers in sequence to perform input purpose recognition and input requirement recognition in sequence, and obtaining the processing demand.

[0009] Preferably, the calculating a weighted sum of each word based on the correlation of each word in the input sequence with all words to obtain an output vector of multi-head attention includes: converting the input vector corresponding to each word in the input sequence into a query vector, a key vector, and a value vector, respectively; splitting the query vector, the key vector, and the value vector corresponding to each word into a plurality of attention heads according to a preset weight matrix to obtain a query vector matrix, a key vector matrix, and a value vector matrix of each word under different attention heads; calculating the similarity between the query vector matrix and the key vector matrix of each word under different attention heads according to the query vector matrix and the key vector matrix of each word under different attention heads; calculating the weight of the value vector matrix of each word under different attention heads according to the similarity of each word under different attention heads; calculating a weighted sum of each word based on the similarity and the weight of the value vector matrix of each word under different attention heads to obtain an output vector of multi-head attention.

[0010] Preferably, the calculating the similarity between the query vector matrix and the key vector matrix of each word under different attention heads includes: calculating the number of elements in the query vector matrix and the key vector matrix of each word under different attention heads that are within a preset range as a similarity number. The similarity between the query vector matrix and the key vector matrix under different attention heads is calculated according to the number of similar words under different attention heads.

[0011] Preferably, the weighted sum of each word is calculated according to the similarity of each word under different attention heads and the weight of the value vector matrix, and the output vector of the multi-head attention is obtained, comprising: The attention heads with weights of the value vector matrix less than a preset weight threshold are filtered out, and the product of the value vector matrix and the corresponding weight of the remaining attention heads after filtering is calculated to obtain the output of each attention head; The outputs of each attention head are spliced and linearly transformed to obtain the output vector of the multi-head attention.

[0012] Preferably, the two neural network layer training processes specifically include: The pre-collected input sequence training set is sequentially input into the two neural network layers for weight and bias training of the neural network layers; Based on the negative feedback of the output result of the second neural network layer, the two weights and biases of the neural network layers are adjusted.

[0013] Preferably, the determining of the to-be-processed data in the mobile cloud disk according to the processing requirement comprises: According to the storage mode of the mobile cloud disk, the data type corresponding to the processing requirement, and the data condition, the to-be-processed data is determined from the aligned multi-modal data in the mobile cloud disk.

[0014] Preferably, when the storage mode is distributed storage, the determining of the to-be-processed data from the aligned multi-modal data in the mobile cloud disk according to the storage mode of the mobile cloud disk, the data type corresponding to the processing requirement, and the data condition comprises: According to the data type corresponding to the processing requirement and the data condition, the target data is determined; According to the reverse transmission of the storage path formed by the switch sequence record experienced by the target data in the data storage process, the to-be-processed data is obtained.

[0015] Preferably, after determining the to-be-processed data in the mobile cloud disk according to the processing requirement, the method further comprises: According to a preset data alignment manner, the to-be-processed data is aligned in time and space.

[0016] Preferably, the inputting of the to-be-processed data into the preset multi-modal machine learning model for processing and the outputting of the business processing result comprise: input the to-be-processed data into a preset multi-modal machine learning model, process the aligned data based on a user demand according to a decoder in the multi-modal machine learning model, and output a service processing result.

[0017] Preferably, the method further comprises: obtaining a satisfaction score table of the user based on the service processing result; determining each scoring item according to the satisfaction score table; based on each scoring item, adjusting weights and biases of the multi-modal model and the multi-modal machine learning model by using a negative feedback neural network for model tuning.

[0018] Preferably, the adjusting of the weights and biases of the multi-modal model and the multi-modal machine learning model by using the negative feedback neural network for model tuning based on each scoring item comprises: when there is a scoring item that does not reach a preset qualified score, adjusting the weights and biases of the multi-modal model and the multi-modal machine learning model by using the negative feedback neural network for model tuning, and re-processing demand analysis and service processing result calculation according to the adjusted multi-modal model and multi-modal machine learning model; and updating each scoring item according to a satisfaction score table obtained based on the service processing result; when there is no scoring item that does not reach the qualified score, the tuning is completed.

[0019] Preferably, the satisfaction score table comprises to-be-scored items associated with weights and biases of an encoder of the multi-modal model and weights and biases of a decoder of the multi-modal machine learning model.

[0020] The embodiment of the application further provides a mobile cloud disk service processing device, and the device comprises: a demand analysis module configured to input an input sequence obtained to a preset multi-modal model, and analyze a processing demand of a user; a data determination module configured to determine to-be-processed data in a mobile cloud disk according to the processing demand; a processing module configured to input the to-be-processed data into a preset multi-modal machine learning model, process the to-be-processed data, and output a service processing result.

[0021] The embodiment of the application further provides a communication device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor implements the mobile cloud disk service processing method according to any one of the above embodiments when executing the computer program.

[0022] The embodiment of the present application also provides a computer readable storage medium, which comprises a stored computer program, wherein the computer readable storage medium controls a device where the computer readable storage medium is located to execute the mobile cloud disk service processing method according to any one of the above embodiments when the computer program runs.

[0023] The embodiment of the present application also provides a computer program product, comprising computer programs / instructions, which are executed by a processor to implement the steps of any one of the above methods.

[0024] The present application provides a mobile cloud disk service processing method, device, equipment, medium and product, by inputting the obtained input sequence into the preset multi-modal model, analyzing the processing demand of the user; determining the data to be processed in the mobile cloud disk according to the processing demand; inputting the data to be processed into the preset multi-modal machine learning model for processing, and outputting the service processing result. The application scheme can accurately understand the user input and accurately output the service processing result. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is a flowchart of the mobile cloud disk service processing method provided by the embodiment of the present application; Figure 2 is a structural diagram of a multi-modal model provided by the embodiment of the present application; Figure 3 is a structural diagram of a mobile cloud disk service processing device provided by the embodiment of the present application; Figure 4 is a structural diagram of a communication device provided by the embodiment of the present application. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0027] To solve the above technical problems, the present application provides a mobile cloud disk service processing method, referring to Figure 1 is a flowchart of the mobile cloud disk service processing method provided by the embodiment of the present application, and the method comprises: Step S1, inputting the obtained input sequence into the preset multi-modal model to analyze the processing demand of the user; Step S2, determining the data to be processed in the mobile cloud disk according to the processing demand; Step S3, inputting the to-be-processed data into a preset multi-modal machine learning model for processing, and outputting a business processing result.

[0028] In the implementation of the embodiment, an input sequence of a user is obtained, and the input sequence is input into the improved multi-modal large model to analyze and output a processing demand of the user based on an encoder layer of the model.

[0029] The multi-modal model, as a core technology support, has the ability to integrate text, voice, image and other multi-modal input modal information. Through the cross-modal correlation knowledge learned in the pre-training process, it can uniformly encode and semantically analyze the sequence input by the user in different forms (such as text description, voice instruction, picture containing demand, etc.), break the information barrier of a single mode, and accurately extract the core processing demand of the user, As a data carrier and management hub, the mobile cloud disk provides a basis for positioning of to-be-processed data through its structured data storage system (such as classification according to file type, creation time, and label). The scheme converts the processing demand analyzed by the multi-modal model into specific data retrieval conditions, realizes the rapid screening and locking of target data in the cloud disk.

[0030] For the positioned to-be-processed data, the preset multi-modal machine learning model is called again for special processing. The model adapts to the corresponding processing module according to the specific demand, such as an OCR recognition module for image data, a voice-to-text module for voice data, and an association analysis module for multi-modal data. Through the processes of feature extraction, semantic understanding, and task reasoning inside the model, the business processing result meeting the user's demand is output, and the full-link closed loop from demand to result is completed.

[0031] The diversified input sequence of the user is converted into an explicit and executable processing demand. The input sequence of the user is received, which can cover multiple modalities, such as a text message “extract the name and number of the customer's ID photo in the cloud disk” sent by the user, an attached ID picture sample, or a voice containing a demand description. Then, the multi-modal input sequence is input into the preset multi-modal model. The model first pre-processes each modal data, such as denoising and feature extraction for voice data, size standardization for image data, and word segmentation and encoding for text data. Then, through the cross-modal attention mechanism, the modal features are fused, the semantic association in the input information is mined, and finally the structured processing demand description is output, which explicitly describes the key information such as processing task type, target data feature, and output result requirement.

[0032] The system identifies the data to be processed in the mobile cloud drive. Based on the user's processing needs, it determines the data stored in the cloud drive that the user requires. Based on the parsed structured processing requirements, the system constructs data retrieval rules. For example, if the processing requirement is "transcribe meeting recordings since October 2025 into text," the retrieval rule can be set as "file type = audio, file tag contains 'meeting recording,' creation time ≥ 2025-10-01." Subsequently, the system establishes communication with the mobile cloud drive's data management interface, converting the constructed retrieval rules into query commands recognizable by the cloud drive, and sends them to the mobile cloud drive. Based on the query commands, the system traverses the internal file index, filters out all data that meets the conditions, and forms a list of data to be processed.

[0033] Based on the modality type of the data to be processed and the processing requirements, the corresponding specialized processing module in the multimodal model is invoked.

[0034] For example, for the "ID card information extraction" requirement where the data to be processed is an image, the model's image recognition and OCR modules are invoked; for the "meeting recording transcription" requirement where the data to be processed is audio, the speech recognition and semantic processing modules are invoked. Subsequently, the system obtains the data to be processed in batches through the cloud disk interface and inputs it into the corresponding processing modules in batches. The model performs in-depth processing on the input data. For example, the OCR module first identifies the text regions in the image, then extracts the specific content through text recognition algorithms, and filters out key information such as name and ID number by combining the format features of the ID card; the speech transcription module first converts the audio signal into a text sequence, then removes redundant information through semantic analysis, and organizes it into a structured meeting minutes according to the speaking logic. After processing, the model outputs preliminary business processing results, completing the entire business processing flow.

[0035] Compared to traditional single-modal requirement recognition methods (such as those that only support text commands), multimodal models can integrate multiple input information such as text, voice, and images, effectively avoiding requirement misunderstandings caused by incomplete single-modal information, breaking modal limitations, making requirement recognition more accurate, and generating user answers based on multimodal data stored in the cloud disk, thereby accurately outputting business processing results.

[0036] In yet another embodiment provided by the present invention, see Figure 2 This is a schematic diagram of the structure of a multimodal model provided in an embodiment of the present invention.

[0037] Employing a layered processing and residual connection design, text feature processing is completed progressively across three core levels: The underlying layer receives the user's text input sequence through input vectorization and transforms each word into a high-dimensional input vector.

[0038] Add timestamps: By concatenating / merging ⊕ operations, the word order features, i.e. timestamps, are injected into the input vector, preserving the semantic order of the text.

[0039] The system uses a multi-head attention mechanism to receive input vectors with timestamps, calculates the global correlation between words, and outputs a feature vector that integrates the context.

[0040] Based on residual connection normalization, the output of the multi-head attention mechanism is residually connected to the input vector, and then the fused features are passed to the next layer through normalization.

[0041] Neural network layers 1 and 2 are multi-layer fully connected neural networks (or convolutional layers) that perform further non-linear transformations, dimensionality compression / expansion on the features output by the attention mechanism to extract more abstract semantic features.

[0042] Based on residual connections and normalization, the features of the previous layer are preserved again through residual connections, and the final semantic feature vector is output after normalization, which is used by the subsequent processing layer to analyze user needs.

[0043] The input sequence (i.e., the text entered by the user) is converted into input vectors by input vectorization. Timestamps are added to each vector based on the order of each word in the received input sequence, so that the subsequent analysis of the input sequence can be performed according to the word order entered by the user.

[0044] The input vector with added timestamps is fed into the multi-head attention mechanism. The improved multi-head attention mechanism algorithm calculates the weighted sum of the current word based on the relevance of all words in the input sequence to the current word.

[0045] The multi-head attention mechanism uses a pre-defined algorithm to calculate the relevance score between the input vector of each word and all words in the sequence, with words having higher relevance receiving greater weight. Then, the vectors of all words are weighted and summed to obtain the output vector for that word. The "multi-head" design allows the model to capture the relationships between words from multiple dimensions (such as syntactic and semantic connections).

[0046] After obtaining the output of the multi-head attention mechanism, the user's processing requirements are analyzed using a preset processing layer based on the input vector and the output vector.

[0047] The processing layer typically consists of a fully connected layer, activation functions (such as ReLU and Softmax), and a classification module. Its core function is to integrate the features of the input vector (basic semantics of words) and the output vector (related semantics of words). Through nonlinear transformation and feature selection, it maps high-dimensional vector features into a structured description of requirements.

[0048] By acquiring input sequences, vectorizing words, adding timestamps, processing multi-head attention, and analyzing the processing layer, the system achieves accurate identification of user needs. Through structured parsing and deep understanding of text semantics, and through multi-stage technical collaboration, unstructured text input is transformed into clear processing requirements, providing precise guidance for subsequent cloud disk data processing.

[0049] In another embodiment of the present invention, the processing layer includes an encoder that encodes the input sequence to obtain semantics and relationships in the input sequence.

[0050] Specifically, when the processing layer is acquired, the structure of the encoder layer in the existing transformer model is improved to generate a processing layer that can encode the input sequence and capture the semantics and relationships in the input sequence.

[0051] An improved multi-head attention mechanism and a structurally enhanced multimodal model encoder layer were used to improve the analysis method of user input sequences, thereby increasing the accuracy of user input sequence analysis.

[0052] In another embodiment provided by the present invention, the step of analyzing the user's processing requirements using a preset processing layer based on the input vector and the output vector includes: After obtaining the output of the multi-head attention mechanism, the input and output vectors are residually concatenated and normalized to alleviate the problems of vanishing and exploding gradients during model training. The normalized results are then individually timestamped based on their output timestamps to identify the location of newly generated data according to their generation time.

[0053] The data after the second timestamp is added is sequentially input into neural network layer 1 and neural network layer 2 to identify the user's input purpose and input requirements, respectively. Neural network layer 1 and neural network layer 2 each have an activation function at the output to increase the non-linearity of the input. The weights and biases in neural network layer 1 and neural network layer 2 can be trained separately based on a pre-collected input sequence training set, and the weights and biases can be adjusted based on negative feedback. Because neural network layer 1 and neural network layer 2 have different functions, the weights and biases in neural network layer 1 and neural network layer 2 are different. The user's processing requirements in the input sequence are analyzed through neural network layer 1 and neural network layer 2.

[0054] In yet another embodiment of the present invention, when calculating the multi-head attention output vector, the specific improved multi-head attention algorithm includes: The input vector corresponding to each word in the input sequence will be transformed into Q (query), K (key), and V (value) according to the following formula: , ; ; Here, W is predicted in advance based on a trained neural network model, and each element is split into three sets of data: Q, K, and V; X represents a certain element in the input sequence.

[0055] Taking Q as an example, we can further break it down into multiple attention heads: ; ;.....;. .

[0056] Where n is the number of attention heads; W is the weight matrix, which is also predicted by the pre-trained model. Different attention heads focus on different directions, and the corresponding data of Q under these different attention directions are calculated by multiplying W with Q.

[0057] Based on the query vector matrix and key vector matrix of each word under different attention heads, calculate the similarity between the query vector matrix and key vector matrix of each word under different attention heads; Based on the similarity of each word under different attention heads, calculate the weights of the value vector matrix of each word under different attention heads; Based on the similarity of each word under different attention heads and the weights of the value vector matrix, the weighted sum of each word is calculated to obtain the output vector of multi-head attention.

[0058] In another embodiment provided by the present invention, when calculating similarity, specifically: The similarity between Q and K is calculated for each attention head using the following formula: ; Where matrices A and B represent the Q and K matrices for the same word under the same attention head, respectively. This represents the number of identical or highly similar elements in matrix A and matrix B. This is the sum of the number of elements in matrices A and B. The method determines whether an element is identical or highly similar to an element in matrices A and B by calculating the absolute value of the difference between corresponding elements in matrices A and B and checking if this difference falls within a preset range. of Elements and Elements are identified as identical or highly similar. y is a pre-set empirical value.

[0059] Improving the weighted sum calculation method of the multi-head attention mechanism by using weighted filtering can directly filter out data that is too low in relevance to the current word, reduce the amount of computation when using the multi-head attention mechanism for input analysis, and improve computational efficiency.

[0060] In another embodiment of the present invention, the output vector of multi-head attention is calculated, specifically: Further calculations are performed on the similarity results between Q and K to obtain the weights of V corresponding to the same word under the same attention head: ; in, For words in the input sequence The weight corresponding to V under the i-th attention head, where j is the number of attention heads.

[0061] If the weights are less than the preset weights, the corresponding V is filtered out. The filtered weights are multiplied by the corresponding V, and the outputs of each attention head are calculated. The outputs of each attention head are then concatenated and linearly transformed to obtain the output result of the multi-head attention.

[0062] In another embodiment provided by the present invention, the training process of the two neural network layers specifically includes: The weights and biases in neural network layer 1 and neural network layer 2 can be trained separately based on a pre-collected input sequence training set, and the weights and biases can be adjusted based on negative feedback. Since neural network layer 1 and neural network layer 2 have different functions, the weights and biases in neural network layer 1 and neural network layer 2 are also different. The user's processing requirements in the input sequence can be analyzed using neural network layer 1 and neural network layer 2.

[0063] In another embodiment provided by the present invention, step S2 constructs a mapping relationship between cloud disk storage rules and user processing needs, and quickly filters out target data that meets the requirements from massive cloud disk data by aligning a structured index of multimodal data. Specifically: Mobile cloud storage features multimodal data alignment and structured storage. It doesn't simply store raw multimodal data (such as images, audio, and text files). Instead, it first constructs a structured index for data alignment, providing a foundation for subsequent retrieval. For multimodal related data in the cloud drive, association mappings are established using key information such as file IDs, timestamps, and tags, ensuring that data from the same source but different modalities can be retrieved uniformly. The cloud drive creates standardized index entries for each piece of data, and the index contains two core types of information: Basic storage attributes, such as storage path, file format, creation / modification time, file size, storage partition, etc. Data characteristics and attributes, such as data type tags, content keywords, and associated data IDs.

[0064] These indexes constitute the data retrieval library of the cloud disk, supporting fast filtering based on multi-dimensional conditions.

[0065] User processing requirements are goal-oriented, structured descriptions that need to be transformed into retrieval rules recognizable by the cloud drive. The transformation process is based on the adaptation of core elements: Adapting to cloud disk storage methods (such as partitioned storage, categorized folder storage, distributed storage, etc.) determines the scope of retrieval, narrowing the search range and improving efficiency.

[0066] The data type is matched to the requirements, and data modality filtering is performed. The processing requirements clarify the modality / type of the target data. The search rules will directly match the data type tags in the cloud disk index to filter out data that meets the modality / type requirements.

[0067] The data conditions in the implementation requirements, namely the precise selection of dimensions, and the processing requirements include specific data constraints.

[0068] The system accurately matches the data to be processed from the aligned data. Based on the transformed retrieval rules, the system interacts with the cloud disk's retrieval library to determine the data to be processed.

[0069] The data required by the user is determined based on the user's processing needs. For example, if the user's processing need is to generate a video with sound based on the image in a selfie stored on September 3, 2023 and the audio on October 5, 2024, then the data required by the user is the selfie from September 3, 2023 and the audio from October 5, 2024 stored in the cloud storage.

[0070] Based on the structured storage and multimodal data alignment of cloud disks, user processing requirements are broken down into executable retrieval rules. Through filtering, the data to be processed is accurately and efficiently located, ensuring that subsequent data processing steps are only carried out on the target data, avoiding invalid calculations and improving the efficiency of the overall solution.

[0071] In another embodiment of the present invention, the mobile cloud disk stores the received data in a distributed storage manner to the cloud, and the process of determining the data to be processed includes: The target data is determined based on the data type and data conditions corresponding to the processing requirements. The data to be processed is obtained by reverse transmission of the storage path formed by the sequential records of the switches that the target data traverses during data storage.

[0072] Specifically, the mobile cloud drive transmits and exchanges data through multiple switches. Data uploaded by users via PC or mobile app is distributed and stored on the corresponding cloud server, or data already stored on the cloud server is sent to the user's terminal, all requiring transmission through layers of switches. As shown in the diagram, gateways, load balancing hardware, and servers all interact with the switches. Therefore, by recording the sequence of switches involved in the data storage process, the storage path of the data after it is sent to the mobile cloud drive can be determined. The reverse transmission of this storage path serves as the path to retrieve the data needed by the user.

[0073] In this solution, the distributed storage of the mobile cloud disk is mainly based on the elastic load balancing principle shown in the diagram for data scheduling and storage. Specifically, xxl-job-admin, associated with the MySQL data, is the scheduling center of a distributed task scheduling platform, responsible for managing and scheduling scheduled tasks. Skywalking in the KCS cluster is an application performance monitoring tool for distributed systems, serving as an observational analysis platform and application performance management system, providing an integrated solution for distributed tracing, service mesh telemetry analysis, metric aggregation, and visualization.

[0074] In another embodiment of the present invention, after determining the data to be processed in the mobile cloud disk according to the processing requirements, the method further includes a data alignment process, specifically: In multimodal processing, it is necessary to align data from multiple modalities both temporally and spatially so that data from different modalities can be understood and used by the model. This allows the model to more comprehensively understand the input information and perform more complex and intelligent tasks. Data alignment is performed on the multimodal data required by the user based on existing data alignment methods.

[0075] In another embodiment provided by the present invention, step S3 is specifically implemented as follows: The aligned data is then fed into an improved multimodal machine learning model, which processes the data to produce the output.

[0076] The aligned data is then input into the decoder layer of the improved multimodal machine learning model described above. The decoder layer processes the aligned data based on user requirements to obtain the output result. The specific structure and principle of the decoder layer can be found in existing technologies and will not be elaborated here.

[0077] When mobile cloud storage uploads data to the cloud, it requires user nodes, public service nodes, cloud space nodes, and PDS nodes to store and manage the uploaded data, device interfaces, and user information. A cloud node is a computing node in a cloud computing environment; it is part of the cloud computing service and is used to process computing tasks and data. Cloud nodes can be physical servers or virtual servers, and they constitute a computing resource pool in a cloud computing environment, dynamically allocating computing resources according to demand.

[0078] In another embodiment provided by the present invention, the method further includes: Obtain the user-uploaded output result satisfaction rating table; on the user's display interface, after generating and outputting the processing results based on the input sequence, display the output result satisfaction rating table to the user below the results.

[0079] Based on the aforementioned satisfaction rating table, determine each rating item and the user's answer preferences; Based on the answer preference, a negative feedback neural network is used to optimize the model and adjust the model weights.

[0080] By adjusting the model parameters based on user satisfaction feedback data, the weights and biases of the encoder and decoder layers of a multimodal large model can be adjusted. This allows for customized mobile cloud storage for different users, thereby improving user satisfaction.

[0081] In another embodiment provided by the present invention, the process of performing weight and bias optimization is specifically as follows: Based on user ratings, the target adjustment direction for the weights and biases in the levels associated with the rating item can be determined. Based on the target adjustment direction, the weights and biases of the corresponding levels are adjusted, and the next rating result is used as negative feedback to continue adjusting the weights and biases of the corresponding levels until all rating items in the output satisfaction rating table reach the qualified score.

[0082] When there are scoring items that do not reach the preset passing score, the negative feedback neural network is used to optimize the model, adjust the weights and biases of the multimodal model and the multimodal machine learning model, and re-analyze the processing requirements and calculate the business processing results based on the adjusted multimodal model and multimodal machine learning model. The optimization is complete when there are no scoring items that have not reached the required score.

[0083] By adjusting the model parameters based on user satisfaction feedback data, the weights and biases of the encoder and decoder layers of a multimodal large model can be adjusted. This allows for customized mobile cloud storage for different users, thereby improving user satisfaction.

[0084] In another embodiment of the present invention, the items to be rated in the satisfaction rating table are associated with the weights and biases in the encoder and decoder layers of the model, and the ratings corresponding to each rating item have been pre-classified into levels. Based on the user rating, the target adjustment direction of the weights and biases in the level associated with the rating item can be determined to achieve precise optimization.

[0085] Specifically, as shown in Table 1 below, the output results satisfaction rating table includes pre-set rating items, user-input rating scores, and rating levels automatically matched by the system based on the rating scores.

[0086] Table 1 Satisfaction Rating Table

[0087] User dissatisfaction with the output may stem from a discrepancy between the user's input habits and the model's encoder layer's understanding. For example, if a user habitually inputs their requirements using a mix of Chinese and English, even if our improved encoder layer is more accurate and comprehensive than a conventional encoder layer, this input habit represents a high degree of user personalization. In this case, the user might rate "Accurate Understanding" lower than "Comprehensive Material Finding" and "Accurate Material Finding" higher. In such cases, we can correlate these ratings with the encoder layer and adjust parameters based on the user's input style.

[0088] See Figure 3 This is a schematic diagram of a mobile cloud storage service processing device provided in an embodiment of the present invention. The device includes: The requirement parsing module is used to input the acquired input sequence into a preset multimodal model and parse the user's processing requirements. The data determination module is used to determine the data to be processed in the mobile cloud drive based on the processing requirements. The processing module is used to input the data to be processed into a preset multimodal machine learning model for processing and output the business processing results.

[0089] The mobile cloud storage service processing device provided in this embodiment can execute all the steps and functions of the mobile cloud storage service processing method provided in any of the above embodiments. The specific functions of the device will not be described in detail here.

[0090] See Figure 4This is a schematic diagram of a communication device provided in an embodiment of the present invention. The communication device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a mobile cloud storage service processing program. When the processor executes the computer program, it implements the steps described in the various embodiments of the mobile cloud storage service processing method above, for example... Figure 1 The steps shown. Alternatively, when the processor executes the computer program, it implements the functions of each module in the above-described device embodiments.

[0091] For example, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the mobile cloud storage service processing device. For example, the computer program can be divided into several modules, the specific functions of which have been described in detail in the mobile cloud storage service processing method provided in any of the above embodiments; therefore, the specific functions of the device will not be repeated here.

[0092] The communication device described can be a desktop computer, laptop, handheld computer, or cloud server, etc. The communication device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the schematic diagram is merely an example of a communication device and does not constitute a limitation on a mobile cloud storage service processing device. It may include more or fewer components than illustrated, or combine certain components, or use different components. For example, the communication device may also include input / output devices, network access devices, buses, etc.

[0093] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the mobile cloud storage service processing device, connecting all parts of the device via various interfaces and lines.

[0094] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the mobile cloud storage service processing device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital card (SD card), flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0095] If the modules integrated into the mobile cloud storage service processing device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0096] This invention also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the functional network element implementing the method described in the above embodiments.

[0097] The computer program product provided in this embodiment can execute all the steps and functions of the mobile cloud disk service processing method provided in any of the above embodiments. The specific functions of the product will not be described in detail here.

[0098] It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications are also considered to be within the scope of protection of this invention.

Claims

1. A method for processing mobile cloud storage services, characterized in that, The method includes: The acquired input sequence is input into a preset multimodal model to parse the user's processing requirements; The data to be processed in the mobile cloud drive is determined based on the processing requirements. The data to be processed is input into a preset multimodal machine learning model for processing, and the business processing results are output.

2. The mobile cloud storage service processing method according to claim 1, characterized in that, The step of inputting the acquired input sequence into a preset multimodal model to parse the user's processing requirements includes: Obtain the input sequence from the user; Each word in the text of the input sequence is vectorized to obtain the input vector; Add timestamps to the input vectors of each word according to the order of each word in the input sequence; The input vector with added timestamps is input into the multi-head attention mechanism. Using a preset multi-head attention mechanism algorithm, the weighted sum of each word is calculated based on the correlation between each word in the input sequence and all words, and the output vector of the multi-head attention is obtained. Based on the input vector and the output vector, a preset processing layer is used to analyze the user's processing requirements.

3. The mobile cloud storage service processing method according to claim 2, characterized in that, The processing layer includes an encoder that encodes the input sequence to obtain the semantics and relationships in the input sequence.

4. The mobile cloud storage service processing method according to claim 2, characterized in that, The step of analyzing the user's processing requirements using a preset processing layer based on the input vector and the output vector includes: The input vector and the output vector are subjected to residual concatenation and normalized cost processing to obtain the normalized result. Add timestamps to the normalization results based on the output times of each word; The normalized result with added timestamps is sequentially input into two preset neural network layers to sequentially identify the input purpose and input requirements, thereby obtaining the processing requirements.

5. The mobile cloud storage service processing method according to claim 2, characterized in that, The step of calculating a weighted sum of each word based on its relevance to all words in the input sequence to obtain the multi-head attention output vector includes: The input vectors corresponding to each word in the input sequence are converted into query vectors, key vectors, and value vectors, respectively. Based on the preset weight matrix, the query vector, key vector, and value vector corresponding to each word are split into several attention heads, so as to obtain the query vector matrix, key vector matrix, and value vector matrix of each word under different attention heads; Based on the query vector matrix and key vector matrix of each word under different attention heads, calculate the similarity between the query vector matrix and key vector matrix of each word under different attention heads; Based on the similarity of each word under different attention heads, calculate the weights of the value vector matrix of each word under different attention heads; Based on the similarity of each word under different attention heads and the weights of the value vector matrix, the weighted sum of each word is calculated to obtain the output vector of multi-head attention.

6. The mobile cloud storage service processing method according to claim 5, characterized in that, The step of calculating the similarity between the query vector matrix and the key vector matrix under different attention heads for each word includes: The number of elements whose difference between the query vector matrix and the key vector matrix for each word is within a preset range under different attention heads is used as the number of similar words; Calculate the similarity between the query vector matrix and the key vector matrix under different attention heads based on the number of similar words for each word.

7. The mobile cloud storage service processing method according to claim 5, characterized in that, The process involves calculating a weighted sum of the similarities of each word under different attention heads and the weights of the value vector matrix to obtain the output vector of the multi-head attention, including: Attention heads whose weights in the value vector matrix are less than a preset weight threshold are filtered out. The value vector matrix of the remaining attention heads after filtering is multiplied by the corresponding weights to obtain the output of each attention head. The outputs of each attention head are concatenated and linearly transformed to obtain the output vector of the multi-head attention.

8. The mobile cloud storage service processing method according to claim 4, characterized in that, The training process for the two neural network layers specifically includes: The pre-collected input sequence training set is sequentially input into two neural network layers to train the weights and biases of the neural network layers. Based on the negative feedback from the output of the second neural network layer, the two weights and biases of the neural network layer are adjusted.

9. The mobile cloud storage service processing method according to claim 1, characterized in that, The step of determining the data to be processed in the mobile cloud drive according to the processing requirements includes: Based on the storage method of the mobile cloud disk, the data type corresponding to the processing requirements, and the data conditions, the data to be processed is determined from the aligned multimodal data in the mobile cloud disk.

10. The mobile cloud storage service processing method according to claim 9, characterized in that, When the storage method is distributed storage, determining the data to be processed from the aligned multimodal data in the mobile cloud disk according to the storage method of the mobile cloud disk, the data type corresponding to the processing requirements, and the data conditions includes: The target data is determined based on the data type and data conditions corresponding to the processing requirements. The data to be processed is obtained by reverse transmission of the storage path formed by the sequential records of the switches that the target data traverses during data storage.

11. The mobile cloud storage service processing method according to claim 1, characterized in that, After determining the data to be processed in the mobile cloud drive based on the processing requirements, the method further includes: The data to be processed is aligned in time and space according to a preset data alignment method.

12. The mobile cloud storage service processing method according to claim 1, characterized in that, The step of inputting the data to be processed into a preset multimodal machine learning model for processing and outputting business processing results includes: The data to be processed is input into a preset multimodal machine learning model. The decoder in the multimodal machine learning model processes the aligned data based on user requirements and outputs the business processing results.

13. The mobile cloud storage service processing method according to claim 1, characterized in that, The method further includes: Obtain a user satisfaction rating table based on the results of the business processing; Each rating item is determined based on the aforementioned satisfaction rating form; Based on each scoring item, a negative feedback neural network is used to optimize the model, adjusting the weights and biases of the multimodal model and the multimodal machine learning model.

14. The mobile cloud storage service processing method according to claim 13, characterized in that, The step of optimizing the model using a negative feedback neural network based on each scoring item, adjusting the weights and biases of the multimodal model and the multimodal machine learning model, includes: When there are scoring items that do not reach the preset passing score, the negative feedback neural network is used to optimize the model, adjust the weights and biases of the multimodal model and the multimodal machine learning model, and re-analyze the processing requirements and calculate the business processing results based on the adjusted multimodal model and multimodal machine learning model; and update each scoring item based on the satisfaction rating table obtained from the business processing results. The optimization is complete when there are no scoring items that have not reached the required score.

15. The mobile cloud storage service processing method according to claim 14, characterized in that, The satisfaction rating scale includes items to be rated that are associated with the weights and biases of the encoder of the multimodal model and the weights and biases of the decoder of the multimodal machine learning model.

16. A mobile cloud storage service processing device, characterized in that, The device includes: The requirement parsing module is used to input the acquired input sequence into a preset multimodal model and parse the user's processing requirements. The data determination module is used to determine the data to be processed in the mobile cloud drive based on the processing requirements. The processing module is used to input the data to be processed into a preset multimodal machine learning model for processing and output the business processing results.

17. A communication device, characterized in that, The device includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the mobile cloud storage service processing method as described in any one of claims 1 to 15.

18. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the mobile cloud disk service processing method as described in any one of claims 1 to 15.

19. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1 to 15.

Images