Cloud mobile phone market demand analysis method and related equipment

Abstract

The invention discloses a cloud mobile phone market demand analysis method and related equipment, and relates to the technical field of cloud mobile phones, the method comprises the following steps: obtaining multi-source heterogeneous data, the multi-source heterogeneous data comprising structured sales data and unstructured user behavior data; based on a preset semantic alignment rule, performing cross-modal fusion on the multi-source heterogeneous data to generate space-time correlation features; based on a dynamic federated learning framework, training a market demand prediction model by using the space-time correlation features, and determining a prediction result of the cloud mobile phone market; and based on the association network atlas, performing association analysis on the prediction result, and determining market decision information. According to the application, the comprehensiveness, prediction precision and decision support efficiency of market analysis are improved, and an accurate and prospective strategic decision basis is provided for each link of the cloud mobile phone industry chain.

Description

Technical Field

[0001] This application relates to the technical field of cloud mobile phones, and particularly to a method for analyzing the market demand of cloud mobile phones and related devices. Background Art

[0002] With the rapid development of the cloud mobile phone market, the existing technologies face significant challenges in processing multi-source heterogeneous data. Traditional analysis methods mainly rely on a single structured data source, making it difficult to comprehensively reflect market dynamics. In addition, static prediction models cannot dynamically adapt to sudden market fluctuations, and the problem of cross-institutional data silos leads to insufficient coverage of model training samples and limited generalization performance, making it difficult to guarantee the accuracy and reliability of prediction results. Therefore, there is an urgent need for a method for analyzing the market demand of cloud mobile phones to solve the above-mentioned technical problems. Summary of the Invention

[0003] A series of simplified concepts are introduced in the Summary of the Invention section, which will be further elaborated in the Detailed Description section. The Summary of the Invention section of this application does not mean to attempt to define the key features and essential technical features of the claimed technical solution, nor does it mean to attempt to determine the protection scope of the claimed technical solution.

[0004] In a first aspect, this application provides a method for analyzing the market demand of cloud mobile phones, the method comprising:

[0005] Obtain multi-source heterogeneous data, wherein the multi-source heterogeneous data includes structured sales data and unstructured user behavior data;

[0006] Based on a preset semantic alignment rule, perform cross-modal fusion on the multi-source heterogeneous data to generate spatio-temporal correlation features;

[0007] Based on a dynamic federated learning framework, use the spatio-temporal correlation features to train a market demand prediction model to determine the prediction result of the cloud mobile phone market;

[0008] Based on an association network graph, perform association analysis on the prediction result to determine market decision-making information.

[0009] In some embodiments, obtaining multi-source heterogeneous data includes:

[0010] Obtain structured sales data from a database through an API interface, wherein the structured sales data includes timestamp, sales volume, and price fields;

[0011] Obtain unstructured user behavior data from a social media platform through a web crawler, wherein the unstructured user behavior data includes user comment texts and online behavior logs;

[0012] Perform noise injection processing on structured sales data and unstructured user behavior data based on the differential privacy mechanism to determine the intermediate data after adding noise;

[0013] Perform masking processing on sensitive fields in the intermediate data based on preset desensitization rules to determine the standardized data stream as multi-source heterogeneous data.

[0014] In some embodiments, perform cross-modal fusion on the multi-source heterogeneous data based on preset semantic alignment rules to generate spatio-temporal correlation features, including:

[0015] Perform sliding window slicing on the structured sales data based on a convolutional neural network, extract spatio-temporal features, and generate a spatio-temporal feature matrix;

[0016] Perform semantic feature extraction on the unstructured user behavior data based on a dynamic layer selection mechanism to determine semantic feature vectors;

[0017] Calculate the correlation weights between the spatio-temporal feature matrix and the semantic feature vectors based on a multi-head attention mechanism to determine the cross-modal correlation matrix;

[0018] Perform weighted fusion on the spatio-temporal feature matrix and the cross-modal correlation matrix based on an adaptive gating function to generate spatio-temporal correlation features.

[0019] In some embodiments, based on a dynamic federated learning framework, use the spatio-temporal correlation features to train a market demand prediction model to determine the prediction results for the cloud mobile phone market, including:

[0020] In the federated client, train the local initial model based on the spatio-temporal correlation features to generate local gradients;

[0021] Perform norm clipping on the local gradients based on a preset gradient clipping threshold to determine the constrained gradients;

[0022] Perform noise processing on the constrained gradients based on a phased privacy strategy to generate encrypted gradients;

[0023] Perform encrypted aggregation on the encrypted gradients of multiple federated clients based on a homomorphic encryption protocol to determine the global gradients;

[0024] Update the global prediction model parameters based on the global gradients and determine the target market demand prediction model;

[0025] Input the real-time market data into the target market demand prediction model and output the prediction results for the cloud mobile phone market.

[0026] In some embodiments, perform correlation analysis on the prediction results based on an association network graph to determine market decision-making information, including:

[0027] Extract technical entities and product entities from patent texts to determine the entity set;

[0028] Based on the co-occurrence frequency and semantic similarity of the entity set, determine the association relationship between technical entities and product entities;

[0029] According to the association relationship, construct a three-dimensional association network graph including patents, products and markets;

[0030] Based on the preset strategy deduction rules, perform path analysis on the three-dimensional association network graph to generate a competitive strategy deduction path;

[0031] Map and match the prediction results with the competitive strategy deduction path to determine market decision-making information.

[0032] In some embodiments, it further includes:

[0033] Based on a sliding window, monitor the error value between the prediction result and the actual market data;

[0034] When the error value is greater than the preset error threshold, determine the dynamic calibration parameter;

[0035] Based on the dynamic calibration parameter, online fine-tune the parameters of the market demand prediction model through an incremental learning algorithm to generate an updated prediction model;

[0036] Based on the updated prediction model, recalculate the prediction result of the cloud phone market.

[0037] In some embodiments, it further includes:

[0038] Based on Monte Carlo simulation, generate the probability distribution of market fluctuations to determine the risk level threshold;

[0039] When the risk probability in the prediction result is greater than the risk level threshold, trigger a risk warning signal and generate decision-making supplementary information including coping strategies.

[0040] In a second aspect, the present application proposes an analysis device for the market demand of cloud phones, including:

[0041] A multi-source heterogeneous data acquisition unit for acquiring multi-source heterogeneous data, where the multi-source heterogeneous data includes structured sales data and unstructured user behavior data;

[0042] A spatio-temporal association feature generation unit, based on preset semantic alignment rules, performs cross-modal fusion on the multi-source heterogeneous data to generate spatio-temporal association features;

[0043] A cloud phone demand prediction unit, based on a dynamic federated learning framework, uses spatio-temporal association features to train a market demand prediction model to determine the prediction result of the cloud phone market;

[0044] A market decision information determination unit performs correlation analysis on the prediction results based on the associated network graph to determine market decision information.

[0045] In a third aspect, an electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program stored in the memory, it implements the steps of the analysis method for the market demand of cloud phones according to any one of the first aspects.

[0046] In a fourth aspect, the present application provides a computer-readable storage medium with a computer program stored thereon. When the computer program is executed by a processor, it implements the analysis method for the market demand of cloud phones according to any one of the first aspects.

[0047] In summary, through multi-source heterogeneous data fusion, dynamic federated learning, and the construction of an associated network graph, the present application significantly improves the comprehensiveness and prediction accuracy of market analysis. First, the cross-modal data fusion mechanism performs semantic alignment and spatio-temporal correlation modeling on structured sales data and unstructured user behavior data, solving the problem of data dimension fragmentation in traditional methods and providing a more comprehensive data basis for analysis. Second, the dynamic federated learning framework realizes cross-institutional model collaborative training through phased privacy policies and error chain calibration, enhancing the adaptability of the model to market dynamic changes while ensuring data privacy. Finally, the correlation analysis based on the associated network graph generates multi-dimensional strategy deduction paths, which, combined with the user-defined visualization engine, can output refined and customizable decision support information, improving the accuracy and foresight of market decisions.

[0048] For the analysis method of the market demand of cloud phones proposed in the present application, other advantages, objectives, and features of the present application will be partially reflected by the following description and partially understood by those skilled in the art through the research and practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] By reading the detailed description of the preferred embodiments below, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to limit this specification. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0050] Figure 1 It is a schematic flowchart of an analysis method for the market demand of cloud phones provided by an embodiment of the present application;

[0051] Figure 2 It is a schematic structural diagram of an analysis device for the market demand of cloud phones provided by an embodiment of the present application;

[0052] Figure 3Schematic diagram of the structure of an electronic device for analyzing the market demand of cloud mobile phones provided by an embodiment of the present application. Specific embodiments

[0053] The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims and above-mentioned drawings of the present application are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily limit to those clearly listed steps or units, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices. Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments.

[0054] Please refer to Figure 1 , which is a schematic diagram of the process of a method for analyzing the market demand of cloud mobile phones provided by an embodiment of the present application, and specifically may include:

[0055] S110. Obtain multi-source heterogeneous data, where the multi-source heterogeneous data includes structured sales data and unstructured user behavior data;

[0056] Exemplarily, the acquisition of multi-source heterogeneous data is a basic link in the analysis of the market demand of cloud mobile phones. The core lies in integrating data sources in different forms to construct a comprehensive market portrait. The structured sales data is directly extracted from the enterprise system through the API interface or database direct connection method, reflecting the objective trading rules of the market; the unstructured user behavior data is captured from the open platform through technologies such as web crawlers to capture users' subjective preferences and potential needs. The heterogeneity of the two types of data stems from the differences in their generation scenarios and representation forms, but together they constitute the multi-dimensional input for market demand analysis, providing basic data support for subsequent cross-modal fusion.

[0057] During the data acquisition process, privacy and compliance are core issues that cannot be ignored. Both structured and unstructured data may contain sensitive information. It is necessary to inject noise into the original data through differential privacy mechanisms to reduce the risk of data leakage; at the same time, combined with de-identification rules, keyword fields are masked to ensure that the data meets the requirements of regulations such as GDPR in subsequent analysis. This process constructs a secure and reliable data circulation link on the premise of ensuring data availability, laying a compliance foundation for cross-institutional collaboration and model training, and avoiding analysis limitations caused by data silos or privacy issues.

[0058] S120. Based on preset semantic alignment rules, perform cross-modal fusion on multi-source heterogeneous data to generate spatio-temporal correlation features;

[0059] Exemplarily, the core of cross-modal fusion lies in solving the semantic gap between structured and unstructured data, and realizing the deep association of multi-source information through preset rules. Structured data extracts regular features in the time and space dimensions through sliding window segmentation and spatio-temporal convolutional networks; unstructured data extracts semantic information at different granularities through a dynamic layer selection mechanism to form a feature expression reflecting market sentiment and demand preferences. Under the semantic alignment framework, the two types of features establish a cross-modal mapping relationship based on the attention mechanism to eliminate the impact of data form differences on analysis and provide a unified representation basis for subsequent modeling.

[0060] The generation of spatio-temporal correlation features focuses on mining the multi-dimensional coupling relationship of market dynamics. Through the spatio-temporal attention mechanism, the time continuity of sales trends and the spatial distribution characteristics of user behavior are jointly modeled to capture the correlation laws such as the fluctuations of regional market demand and the popularity of social media public opinion. This feature not only retains the spatio-temporal attributes of the original data, but also dynamically adjusts the contribution weights of different modalities through adaptive gating fusion to form a comprehensive feature that can reflect historical laws and adapt to real-time changes, providing an input with both interpretability and predictive power for the prediction model.

[0061] S130. Based on the dynamic federated learning framework, use spatio-temporal correlation features to train a market demand prediction model to determine the prediction results of the cloud mobile phone market;

[0062] Exemplarily, the dynamic federated learning framework realizes the collaborative modeling of multi-source spatio-temporal correlation features through a distributed cooperation mechanism on the premise of ensuring data privacy. Each participant (such as cloud mobile phone manufacturers, operators) independently trains an initial prediction model and calculates gradients based on local data and spatio-temporal correlation features; through gradient clipping and staged noise injection strategies, the risk of gradient explosion is suppressed while avoiding privacy leakage. The encrypted gradients are aggregated to the coordination node through a secure aggregation protocol to generate global model parameter updates, forming a prediction model that takes into account both individual data characteristics and global market laws, providing technical support for cross-institutional joint analysis.

[0063] The market demand prediction model takes spatio-temporal correlation features as the core input. By capturing the spatio-temporal patterns of structured sales data and the semantic associations of unstructured data, it constructs a multi-dimensional dynamic market representation. Under the federated learning framework, the global model integrates the distributed features of multi-source data, combines the sliding window error monitoring and incremental learning algorithms, and dynamically optimizes the prediction weights. The finally output prediction results not only reflect historical trends but also integrate real-time market feedback, forming an accurate prediction of the market demand changes in the cloud mobile phone market, providing a reliable basis for subsequent competitive strategy deduction and risk warning.

[0064] S140. Based on the association network graph, conduct association analysis on the prediction results to determine market decision-making information.

[0065] Exemplarily, the construction of the association network graph is based on multi-dimensional entity relationship mining. By extracting technical entities and product entities from patent texts and combining their co-occurrence frequencies and semantic similarities, an association network between entities is established. This network covers the three-dimensional association relationships of patent technologies, product characteristics, and market dynamics, forming a structured knowledge base that reflects the market technology ecosystem and competitive landscape. By deeply analyzing the association paths in the graph through preset strategy deduction rules, potential technology evolution trends and market opportunities can be identified, providing data-driven logical support for decision-making.

[0066] The generation of market decision-making information depends on the intelligent mapping of the prediction results and the association network graph. Match the cloud mobile phone market demand prediction results with the technology-product association paths in the graph to identify key influencing factors and potential risk nodes. For example, predicting the market demand growth in a certain technology field can be associated with relevant patent layouts through the graph, and then generate targeted product optimization or market expansion strategies. This mapping mechanism enables decision-making information to be based not only on historical data but also on the global associations of the technology ecosystem, significantly improving the forward-looking and operability of decision-making.

[0067] In summary, in the embodiments of the present application, through multi-source heterogeneous data fusion, dynamic federated learning, and associated network graph construction, the comprehensiveness of cloud mobile phone market demand analysis and the accuracy of decision-making are improved. First, by integrating structured sales data and unstructured user behavior data, the limitations of traditional single data sources are broken through. Through cross-modal semantic alignment and spatio-temporal correlation modeling, a multi-dimensional market dynamic representation is established, providing a more complete input basis for analysis. Second, the dynamic federated learning framework supports cross-institutional collaborative training of models while protecting data privacy. Combining phased privacy strategies and error chain calibration mechanisms enhances the model's adaptability to sudden market fluctuations and improves the reliability and real-time nature of prediction results. Finally, based on the path deduction of the associated network graph, the market demand prediction results are intelligently mapped to the technology ecosystem and product layout to identify potential risks and opportunities, generating interpretable competitive strategies and risk warning schemes. In addition, while improving prediction accuracy, market uncertainty is reduced through risk warning and dynamic calibration mechanisms, helping enterprises quickly respond to market changes, optimize resource allocation, and strategic layout.

[0068] In some examples, obtaining multi-source heterogeneous data includes:

[0069] Obtaining structured sales data from a database through an API interface, where the structured sales data includes timestamp, sales volume, and price fields;

[0070] Obtaining unstructured user behavior data from a social media platform through a web crawler, where the unstructured user behavior data includes user comment text and online behavior logs;

[0071] Performing noise injection processing on the structured sales data and unstructured user behavior data based on the differential privacy mechanism to determine the intermediate data after adding noise;

[0072] Performing masking processing on sensitive fields in the intermediate data based on preset desensitization rules to determine a standardized data stream as multi-source heterogeneous data.

[0073] Exemplarily, in the data acquisition stage, structured sales data is extracted from an enterprise database or cloud platform through an API interface, and its core fields include key transaction metrics such as timestamp, sales volume, and price. The API interface is designed according to the RESTful specification, supporting efficient data query and batch transmission to ensure the timeliness and integrity of sales data. Structured data is stored in a standardized table form and directly connected to an enterprise-level database through the JDBC or ODBC protocol to achieve real-time or scheduled data synchronization. Such data accurately records the time series characteristics and spatial distribution laws of market transaction behaviors, providing a basic input for subsequent spatio-temporal feature extraction.

[0074] Unstructured user behavior data is dynamically crawled from social media platforms (such as Weibo and forums) through distributed web crawler technology, covering diverse information such as user comment texts and online behavior logs. The crawler component is built based on the Scrapy framework, supporting multi-threaded concurrent crawling and anti-crawling strategy avoidance to ensure the efficiency and compliance of data collection. The original text data undergoes preprocessing processes such as sentence splitting, denoising, and stemming, and is converted into a standardized text stream. Such data deeply reflects users' sentiment tendencies, product preferences, and potential needs, providing semantically rich unstructured inputs for semantic feature extraction and cross-modal fusion.

[0075] To ensure data privacy and security, structured and unstructured data need to undergo differential privacy processing before fusion. The Laplace mechanism is used to inject noise into the original data, and the noise intensity is dynamically controlled by the phased privacy budget parameter (ε value): a higher privacy budget (such as ε = 3.0) is set at the initial stage of training to retain the statistical characteristics of the data, gradually reduced in the middle stage (such as ε = 2.2), and further tightened in the later stage (such as ε = 1.8) to balance privacy protection and data availability. The noise injection process is monitored in real time through a privacy budget manager, and the noise scale is dynamically adjusted in combination with training stage detection signals (such as training round counts) to ensure that the data privacy intensity in different training stages adapts to the model convergence requirements.

[0076] The noisy intermediate data needs to further perform sensitive field desensitization processing to comply with data compliance requirements such as GDPR. A preset desensitization rule engine masks or generalizes fields containing sensitive information such as user identities and geographical locations, for example, using hash encryption, interval generalization, or mask replacement techniques. The desensitized data is converted into a unified data stream according to the JSON-LD standard format. The JSON-LD template defines the semantic associations and spatio-temporal attributes of data fields, ensuring semantic consistency and resolvability of multi-source heterogeneous data in subsequent cross-modal fusion. The standardized data stream is transmitted to the analysis engine through an AES-256 encrypted channel to build a secure and reliable data processing link, providing compliant inputs for federated learning and market prediction.

[0077] In some instances, based on preset semantic alignment rules, multi-source heterogeneous data is cross-modally fused to generate spatio-temporal correlation features, including:

[0078] The structured sales data is sliced by a sliding window based on a convolutional neural network to extract spatio-temporal features and generate a spatio-temporal feature matrix;

[0079] Semantic feature extraction is performed on unstructured user behavior data based on a dynamic layer selection mechanism to determine semantic feature vectors;

[0080] The association weights between the spatio-temporal feature matrix and the semantic feature vectors are calculated based on a multi-head attention mechanism to determine the cross-modal association matrix;

[0081] Based on an adaptive gating function, the spatio-temporal feature matrix and the cross-modal correlation matrix are weighted and fused to generate spatio-temporal correlation features.

[0082] Exemplarily, structured sales data is modeled for spatio-temporal features through a convolutional neural network. The sliding window slicing technique is adopted to divide the time-series sales data into continuous windows with a fixed time period (such as 7 days) to capture the periodic fluctuations of market demand. The data within each window is input into a hybrid network architecture, where the Temporal CNN extracts local temporal dependencies through multi-scale convolutional kernels (such as 3×1, 5×1), and the dilation convolutional layer (dilation rate = 2) enhances the ability to capture long-term trends. The output forms a spatio-temporal feature matrix, whose dimensions include time steps, regional distributions, and feature channels, representing the spatio-temporal evolution law of sales data.

[0083] Unstructured user behavior data realizes semantic feature extraction through a dynamic layer selection mechanism. After the original text is preprocessed in chunks, it is input into an improved hierarchical BERT model, and the intermediate hidden layer combination (such as layers 6-9) is dynamically selected according to the text complexity score (based on the TF-IDF entropy value). Low-complexity texts focus on shallow syntactic features (layers 6-7), and high-complexity texts extract deep semantic features (layers 8-9). By non-linearly weighting and fusing the outputs of each layer, a 768-dimensional semantic feature vector is generated, comprehensively reflecting the sentiment polarity of user comments, the attention to product attributes, and the semantic information of behavior patterns.

[0084] The spatio-temporal feature matrix and the semantic feature vector establish cross-modal correlations through the multi-head attention mechanism. The spatio-temporal features are used as query vectors, and the semantic features are used as key-value pairs to calculate the attention weight matrix between the two. Eight parallel attention branches are adopted to capture different-dimensional feature interaction patterns respectively, such as the correlation between regional sales trends and sentiment tendencies, and the mapping between price fluctuations and user behavior patterns. The attention weights are scaled by a temperature parameter (scaling factor adjustment parameter d_k = 64) and normalized by Softmax to generate a cross-modal correlation matrix, quantifying the semantic matching strength between structured and unstructured data.

[0085] The cross-modal correlation matrix and the original spatio-temporal feature matrix are dynamically fused through an adaptive gating function. The gating coefficient is generated by the Sigmoid function, and the contribution ratio of unstructured semantic features is dynamically adjusted according to the correlation weights. Specifically, in regions with high correlation weights, the fusion intensity of semantic features is enhanced, and in regions with low weights, the original statistical characteristics of spatio-temporal features are retained. A residual connection mechanism is introduced during the fusion process, adding the original spatio-temporal features and the weighted cross-modal features to form spatio-temporal correlation features. This feature simultaneously retains the precise spatio-temporal laws of sales data and the semantic context information of unstructured data, providing a high-information-density input representation for subsequent prediction models.

[0086] In some instances, based on the dynamic federated learning framework, a market demand prediction model is trained using spatio-temporal correlation features to determine the prediction results for the cloud mobile phone market, including:

[0087] In the federated client, the local initial model is trained based on spatio-temporal correlation features to generate local gradients;

[0088] Based on a preset gradient clipping threshold, the norm of the local gradient is clipped to determine the constrained gradient;

[0089] Based on the staged privacy strategy, noise processing is performed on the constrained gradient to generate the encrypted gradient;

[0090] Based on the homomorphic encryption protocol, the encrypted gradients of multiple federated clients are encrypted and aggregated to determine the global gradient;

[0091] Based on the global gradient, the global prediction model parameters are updated and the target market demand prediction model is determined;

[0092] The real-time market data is input into the target market demand prediction model, and the prediction results for the cloud mobile phone market are output.

[0093] Exemplarily, the process of training the market demand prediction model based on the dynamic federated learning framework follows a multi-stage collaborative optimization mechanism, and its core principle is as follows: First, each federated client performs local training on the initial prediction model based on the spatio-temporal correlation features stored locally. The spatio-temporal correlation features are generated by cross-modal fusion and contain the spatio-temporal patterns of structured sales data and the semantic associations of unstructured user behavior data. The client calculates the local gradients of the model parameters through the backpropagation algorithm, and these gradients reflect the association patterns between the local data distribution characteristics and the dynamics of market demand.

[0094] To ensure the security of gradient transmission and suppress the interference of outliers, a norm clipping operation needs to be performed on the local gradient. The preset gradient clipping threshold is 1.5, and the L2 norm constraint method is used to limit the magnitude of the gradient vector within the threshold range to generate the constrained gradient. This process effectively avoids the problem of gradient explosion and at the same time reduces the noise interference intensity of subsequent privacy protection processing, ensuring the convergence stability of the model.

[0095] The constrained gradient needs to be processed by the staged privacy strategy to meet the requirements of differential privacy. A higher privacy budget ε = 3.0 is set in the initial stage of training, adjusted to 2.2 in the middle stage, and reduced to 1.8 in the later stage. According to the current training stage identifier, the Laplace mechanism is used to inject dynamic noise into the gradient, and the noise scale is jointly determined by the privacy budget parameter and the gradient sensitivity. The gradient after noise processing is converted into an encrypted gradient, and homomorphic encryption protocol is used to achieve encrypted transmission, preventing the gradient information from being stolen or tampered with during the transmission process.

[0096] The encrypted gradients are globally integrated at the coordination node through the secure aggregation protocol. Specifically, the Paillier homomorphic encryption algorithm is used to perform ciphertext addition operations on the encrypted gradients of multiple clients to generate the global gradient. This process ensures that participants cannot obtain the gradient information of other clients, achieving a balance between data privacy and model performance. The global gradient represents the common laws of multi-party data and provides a direction for global model updates.

[0097] The parameters of the global prediction model are iteratively updated based on the aggregated global gradients. An adaptive learning rate strategy is adopted, with an initial learning rate of 0.01, which decays exponentially with the number of training rounds (the decay factor is 0.1). The parameter update process synchronously receives the error chain calibration signal. When the prediction error exceeds the threshold monitored by the sliding window, the incremental learning algorithm is triggered to fine-tune the model parameters online, enhancing the adaptability of the model to market dynamic changes.

[0098] Finally, the real-time collected market data is input into the trained target prediction model, and the prediction results of the cloud mobile phone market demand are output. The prediction model integrates the multi-source spatio-temporal correlation features and the distributed modeling advantages of the federated learning framework, and can accurately capture the market demand trends, regional distribution differences, and sudden fluctuation laws. The prediction results are synchronously input into the associated network graph for strategy deduction, providing a quantitative basis with both timeliness and interpretability for market decisions.

[0099] In some instances, based on the associated network graph, correlation analysis is performed on the prediction results to determine market decision-making information, including:

[0100] Extract technical entities and product entities from the patent text to determine the entity set;

[0101] Based on the co-occurrence frequency and semantic similarity of the entity set, determine the association relationship between technical entities and product entities;

[0102] According to the association relationship, construct a three-dimensional association network graph including patents, products, and the market;

[0103] Based on the preset strategy deduction rules, perform path analysis on the three-dimensional association network graph to generate a competitive strategy deduction path;

[0104] Map and match the prediction results with the competitive strategy deduction path to determine market decision-making information.

[0105] Exemplarily, first, extract technical entities and product entities from patent texts to form an entity set. Use the BERT-BiLSTM-CRF combined model to preprocess the patent text and extract entities. Among them, the BERT model extracts context semantic features, the BiLSTM captures sequence dependencies, and the CRF decoding layer outputs entity recognition results that conform to the IOB annotation specification. Technical entities include key technical terms such as chip architectures and communication protocols, and product entities cover product identifiers such as cloud mobile phone models and functional modules. Eliminate naming ambiguities through semantic normalization processing to form a standardized entity set.

[0106] Construct association relationships based on entity co-occurrence frequencies and semantic similarities. Use the sliding window co-occurrence analysis method to count the adjacent co-occurrence times of entities in patent texts, calculate the association strength between entities in combination with pointwise mutual information (PMI), and set a threshold θ = 0.35 to filter out low-correlation relationships. At the same time, generate entity semantic vectors through the BERT model, calculate the cosine similarity to quantify the semantic association degree, and fuse the co-occurrence statistics and semantic similarity scores to generate a multi-dimensional association relationship matrix of technical entities and product entities, providing a relational data basis for graph construction.

[0107] Construct a three-dimensional association network graph of patents, products, and markets. Use the graph convolutional network to perform embedding representations on entity nodes and association edges, map patent technologies, product features, and market dynamics to independent dimensions in the graph, and achieve three-dimensional association modeling through cross-dimensional edge connections. For example, a 5G patent node is linked to a cloud mobile phone product node that supports this technology through a "technology implementation" edge, and then associated with a target regional market node through a "market application" edge, forming a structured knowledge network that reflects the technology ecosystem and market layout. The graph supports an incremental update mechanism. When a new product is released or a policy changes are detected, it triggers a dynamic graph version iteration.

[0108] Generate a competitive strategy deduction path based on preset strategy deduction rules. Use the random walk algorithm to traverse the three-dimensional association network and identify high-frequency path patterns and potential association links. For example, starting from the "edge computing patent" node, deduce along the paths of technology implementation, cloud mobile phone product, and regional market, and evaluate the commercialization potential of the technology in combination with the market demand prediction results, generating a strategic recommendation of "strengthening the research and development of edge computing technology to expand the smart city market". The deduction process integrates historical market performance and real-time prediction data, and optimizes the path weights through a feedback mechanism to improve the feasibility and foresight of the strategic recommendations.

[0109] Intelligently map and match the prediction results with the competitive strategy deduction path, and output market decision-making information. Align the cloud mobile phone market demand prediction indicators (such as regional sales growth rate, technology demand heat) with the entity association paths in the graph through a semantic matching algorithm to identify key influencing factors. For example, when predicting a sharp increase in market demand in a certain region, automatically associate the technology patent layout of the mainstream products in that region, and generate decision-making suggestions such as "prioritize deploying cloud mobile phone models with relevant technology patents". The decision-making information is presented through a visualization interface, supporting multi-dimensional drill-down analysis, and providing a data-driven decision-making basis for enterprises to formulate product optimization, market expansion, and technology investment strategies.

[0110] In some instances, it further includes:

[0111] Monitor the error value between the prediction result and the actual market data based on a sliding window;

[0112] When the error value is greater than the preset error threshold, determine the dynamic calibration parameter;

[0113] Based on the dynamic calibration parameter, online fine-tune the parameters of the market demand prediction model through an incremental learning algorithm to generate an updated prediction model;

[0114] Recalculate the prediction result of the cloud mobile phone market based on the updated prediction model.

[0115] Exemplarily, use a sliding window monitoring mechanism to track the deviation between the prediction result and the actual market data in real time. Set the window length to 5 time units (which can be days or weeks, etc.), and slide it sequentially in chronological order to cover the latest prediction data and actual data. Within the window, calculate the mean absolute error (MAE) using a moving average algorithm by accumulating the absolute differences between the predicted values and the actual values at each time point and taking the average, so as to form a dynamic error sequence, thereby quantifying the model prediction deviation in real time. When it is monitored that the MAE of consecutive windows exceeds the preset threshold (such as 0.15), trigger an error warning signal, indicating that there is a risk of cumulative prediction deviation in the current model, and it is necessary to start the dynamic calibration process.

[0116] Determine dynamic calibration parameters based on the error statistical characteristics within the sliding window. The error feedback control module analyzes the error distribution pattern and fluctuation amplitude, and calculates the calibration coefficient. The calibration coefficient is positively correlated with the MAE value within the window. The MAE is mapped to the interval [0, 1] through normalization processing, and weighted smoothing is performed in combination with historical calibration records. For example, if the current MAE is 0.2 and the preset threshold is 0.15, the calibration coefficient can be calculated as (0.2 - 0.15) / 0.15 ≈ 0.33, which means that the model parameters need to be adjusted by 33%. At the same time, the calibration coefficient also considers the error change trend. When the error shows an increasing trend, a negative feedback coefficient is generated to suppress overfitting; when the error fluctuates randomly, an adaptive adjustment factor is generated by weighting the variance of the sliding window. After normalization processing, the calibration coefficient is mapped to the model parameter adjustment range to ensure the stability of incremental learning.

[0117] Perform online fine-tuning on the market demand prediction model by means of an incremental learning algorithm. Retain the main structure of the global model, and only perform local optimization on the parameters of some sensitive layers such as the fully connected layer and attention weights. Use the mini-batch gradient descent method to extract the latest samples (such as the data of the recent 7 days) from the real-time data stream, and adjust the learning rate with the calibration coefficient as the learning rate scaling factor. For example, multiply the base learning rate (such as 0.01) by the calibration coefficient to generate an adaptive learning rate. During the update process, introduce the momentum term and weight decay mechanism to avoid parameter oscillation and control the model complexity, ensuring that the fine-tuned model not only adapts to the latest market dynamics but also retains the learning results of historical laws and maintains the generalization ability of the model.

[0118] Recalculate the cloud mobile phone market demand prediction result based on the updated prediction model. Input the real-time market data into the fine-tuned model, generate feature vectors through the spatio-temporal correlation feature extraction module, and calculate the prediction result based on the fine-tuned parameters. The recalculation process uses an incremental data set (only including the latest data and historical key node data), reducing the calculation overhead and improving the timeliness. The updated prediction result is synchronously fed back to the error monitoring window to form a closed-loop control link of prediction, error evaluation, calibration, and re-prediction. If the prediction error recalculated falls below the threshold, the current model parameters are maintained; if it still exceeds the standard, a secondary calibration process is triggered. Through this dynamic iterative optimization mechanism, continuously improve the adaptability of the prediction model to sudden market fluctuations, ensure the accuracy and timeliness of the output results, and ensure its high consistency with market dynamics.

[0119] In some instances, it also includes:

[0120] Generate the probability distribution of market fluctuations based on Monte Carlo simulation, and determine the risk level threshold;

[0121] When the risk probability in the prediction result is greater than the risk level threshold, trigger a risk warning signal and generate decision supplementary information including coping strategies.

[0122] Exemplarily, first, the random sampling and probability modeling of the cloud mobile phone market demand prediction results are carried out through Monte Carlo simulation. Based on historical market data, a multi-dimensional parameter model including price fluctuations, user demand changes and competition situation is constructed, and the probability distribution curve of market fluctuations is generated through tens of thousands of random samplings. This distribution curve quantifies the occurrence probabilities of different risk events (such as sudden demand drop, technology substitution shock), and based on the quantiles, the risk level thresholds are divided (such as the high-risk threshold is the 90% quantile, and the medium-risk is the 75% quantile), providing a quantitative basis for risk early warning.

[0123] The risk probability indicators in the prediction results are monitored in real time and compared with the preset thresholds. When a certain risk dimension in the prediction results (such as the probability of regional market demand shrinkage) exceeds the corresponding level threshold, a risk early warning signal is triggered. For example, if the probability of a regional market demand decline reaches the high-risk threshold (such as ≥15%), an early warning event including the regional identifier, risk type and probability value is generated, and the decision-making supplementary information generation process is started synchronously.

[0124] Combined with the risk type and the associated network graph to generate coping strategies. The strategy deduction engine extracts the technical entities, product nodes and market paths related to the early warning event from the three-dimensional associated network graph, and identifies potential coping measures through path analysis. For example, for the technology substitution risk, it automatically associates the relevant patent layout and competitor dynamics, and generates strategy suggestions such as "accelerating technology iteration" or "patent cross-licensing"; for the market demand fluctuation risk, it associates the historical sales data and user behavior characteristics, and generates "dynamic price adjustment" or "inventory optimization" solutions.

[0125] Integrate the risk early warning signal and the coping strategy into decision-making supplementary information, and push it to the user terminal through the visualization interface. The early warning information is presented in the form of a risk heat map, marking the high-risk areas and triggering factors; the coping strategies are sorted by priority, and the strategy implementation path and expected effect evaluation are attached. For example, for the high-risk area, a decision-making suggestion of "reducing the investment scale and strengthening competitor monitoring" is generated, and at the same time, the associated patent technology upgrade roadmap is provided. This mechanism significantly improves the enterprise's response efficiency and decision-making scientificity to market uncertainties through the closed-loop of probabilistic risk quantification and strategy deduction.

[0126] Please refer to Figure 2 , which is a schematic structural diagram of an analysis device for the cloud mobile phone market demand provided by an embodiment of the present application, including:

[0127] A multi-source heterogeneous data acquisition unit 21, configured to acquire multi-source heterogeneous data, where the multi-source heterogeneous data includes structured sales data and unstructured user behavior data;

[0128] The spatio-temporal correlation feature generation unit 22 performs cross-modal fusion on multi-source heterogeneous data based on a preset semantic alignment rule to generate spatio-temporal correlation features;

[0129] The cloud mobile phone demand prediction unit 23 trains a market demand prediction model using spatio-temporal correlation features based on a dynamic federated learning framework to determine the prediction result of the cloud mobile phone market;

[0130] The market decision information determination unit 24 performs correlation analysis on the prediction result based on an association network graph to determine market decision information.

[0131] Please refer to Figure 3 , The embodiment of the present application also provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored on the memory 310 and executable on the processor. When the processor 320 executes the computer program 311, it implements the steps of any method for analyzing the demand of the cloud mobile phone market.

[0132] Since the electronic device introduced in this embodiment is the device adopted by an analysis device for the demand of the cloud mobile phone market in the embodiment of the present application, based on the method introduced in the embodiment of the present application, those skilled in the art can understand the specific implementation manners and various variations of the electronic device in this embodiment. Therefore, the specific implementation of how this electronic device implements the method in the embodiment of the present application will not be described in detail here. As long as the device adopted by those skilled in the art to implement the method in the embodiment of the present application belongs to the scope protected by the present application.

[0133] In the specific implementation process, when the computer program 311 is executed by the processor, it can implement any implementation manner in the corresponding embodiment of the first aspect.

[0134] It should be noted that in the above embodiments, the descriptions of each embodiment have their own focuses. For the parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0135] Those skilled in the art should understand that the embodiments of the present application can provide methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-readable storage media containing computer-readable program codes.

[0136] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can also be implemented. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks.

[0137] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implement the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks.

[0138] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks.

[0139] Embodiments of the present application also provide a computer program product, which includes computer software instructions. When the computer software instructions run on a processing device, the processing device is caused to execute Figure 1 the process of an analysis method for the market demand of a cloud mobile phone in the corresponding embodiment.

[0140] A computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, the processes or functions according to the embodiments of the present application are all or partially generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired or wireless manner. The computer-readable storage medium may be any available medium that can be stored by a computer or a data storage device such as a server or a data center that includes one or more integrated available media. The available medium may be a magnetic medium, an optical medium, or a semiconductor medium, etc.

[0141] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0142] In several embodiments provided in the present application, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the displayed or discussed couplings, direct couplings, or communication connections to each other may be indirect couplings or communication connections through some interfaces, devices, or units, and may be in electrical, mechanical, or other forms.

[0143] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0144] In addition, each functional unit in various embodiments of the present application may be integrated in a processing unit, or each unit may exist physically alone, or two or more units may be integrated in one unit. The above-mentioned integrated units may be implemented in the form of hardware and / or software functional units.

[0145] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device to execute all or part of the steps of the methods of the various embodiments of this application.

[0146] The above embodiments are only used to illustrate the technical solutions of this application, rather than to limit them; although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of this application.

[0147] Although the preferred embodiments of this specification have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of this specification.

[0148] Obviously, those skilled in the art can make various changes and deformations to this specification without departing from the spirit and scope of this specification. In this way, if these modifications and deformations of this specification fall within the scope of the claims of this specification and their equivalent technologies, this specification also intends to include these modifications and deformations.

Claims

1. A method for analyzing the market demand of cloud mobile phones, characterized in that, Including: Obtain multi-source heterogeneous data, where the multi-source heterogeneous data includes structured sales data and unstructured user behavior data; Based on preset semantic alignment rules, perform cross-modal fusion on the multi-source heterogeneous data to generate spatio-temporal correlation features; Based on a dynamic federated learning framework, use the spatio-temporal correlation features to train a market demand prediction model to determine the prediction result of the cloud mobile phone market; Based on the associated network graph, perform association analysis on the prediction result to determine market decision-making information.

2. The method according to claim 1, characterized in that, The obtaining of the multi-source heterogeneous data includes: Obtain structured sales data from a database through an API interface, where the structured sales data includes timestamp, sales volume, and price fields; Obtain unstructured user behavior data from a social media platform through a web crawler, where the unstructured user behavior data includes user comment texts and online behavior logs; Based on the differential privacy mechanism, perform noise injection processing on the structured sales data and the unstructured user behavior data to determine the intermediate data with noise; Based on preset desensitization rules, perform masking processing on the sensitive fields in the intermediate data to determine the standardized data stream as the multi-source heterogeneous data.

3. The method according to claim 1, wherein The performing of cross-modal fusion on the multi-source heterogeneous data based on preset semantic alignment rules to generate spatio-temporal correlation features includes: Based on a convolutional neural network, perform sliding window segmentation on the structured sales data, extract spatio-temporal features, and generate a spatio-temporal feature matrix; Based on a dynamic layer selection mechanism, perform semantic feature extraction on the unstructured user behavior data to determine semantic feature vectors; Based on a multi-head attention mechanism, calculate the association weights between the spatio-temporal feature matrix and the semantic feature vectors to determine a cross-modal association matrix; Based on an adaptive gating function, perform weighted fusion on the spatio-temporal feature matrix and the cross-modal association matrix to generate spatio-temporal correlation features.

4. The method according to claim 1, characterized in that The training of a market demand prediction model using the spatio-temporal correlation features based on a dynamic federated learning framework to determine the prediction result of the cloud mobile phone market includes: In the federated client, train the local initial model based on the spatio-temporal correlation features to generate local gradients; Based on a preset gradient clipping threshold, perform norm clipping on the local gradients to determine constrained gradients; Based on a phased privacy strategy, perform noise processing on the constrained gradients to generate encrypted gradients; Based on a homomorphic encryption protocol, perform encrypted aggregation on the encrypted gradients of multiple federated clients to determine global gradients; Based on the global gradients, update the global prediction model parameters and determine the target market demand prediction model; Input real-time market data into the target market demand prediction model and output the prediction result of the cloud mobile phone market.

5. The method according to claim 1, characterized in that The performing of association analysis on the prediction result based on the associated network graph to determine market decision-making information includes: Extract technical entities and product entities from patent texts to determine an entity set; Based on the co-occurrence frequency and semantic similarity of the entity set, determine the association relationship between the technical entities and the product entities; According to the association relationship, construct a three-dimensional association network graph including patents, products, and the market; Perform path analysis on the three-dimensional association network graph based on preset strategy deduction rules to generate a competitive strategy deduction path; Map and match the prediction result with the competitive strategy deduction path to determine market decision information.

6. The method according to claim 1, wherein It further includes: Monitor the error value between the prediction result and the actual market data based on a sliding window; When the error value is greater than a preset error threshold, determine the dynamic calibration parameter; Based on the dynamic calibration parameter, online fine-tune the parameters of the market demand prediction model through an incremental learning algorithm to generate an updated prediction model; Recalculate the prediction result of the cloud mobile phone market based on the updated prediction model.

7. The method according to claim 1, wherein It further includes: Generate a probability distribution of market fluctuations based on Monte Carlo simulation to determine the risk level threshold; When the risk probability in the prediction result is greater than the risk level threshold, trigger a risk warning signal and generate decision supplementary information including countermeasures.

8. An analysis device for the market demand of cloud mobile phones, characterized in that, It includes: A multi-source heterogeneous data acquisition unit for acquiring multi-source heterogeneous data, where the multi-source heterogeneous data includes structured sales data and unstructured user behavior data; A spatio-temporal association feature generation unit that performs cross-modal fusion on the multi-source heterogeneous data based on preset semantic alignment rules to generate spatio-temporal association features; A cloud mobile phone demand prediction unit that trains a market demand prediction model using the spatio-temporal association features based on a dynamic federated learning framework to determine the prediction result of the cloud mobile phone market; A market decision information determination unit that performs association analysis on the prediction result based on an association network graph to determine market decision information.

9. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor is configured to implement the steps of the method for analyzing the market demand of a cloud mobile phone according to any one of claims 1 to 7 when executing the computer program stored in the memory.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The computer program, when executed by the processor, implements the method for analyzing the market demand of a cloud mobile phone according to any one of claims 1 to 7.

Images