OLLAMA-based data center clothing management method and system

Abstract

The present application relates to the technical field of data processing, and particularly relates to a data center clothing management method and system based on OLLAMA, the method comprises the following steps: collecting computing resource state data; performing predictive semantic preprocessing, including: when the computing resource state data indicates that a graphic processor is in a preset low load state, according to query mode information obtained by analyzing a user query log, selecting unstructured clothing data to be pre-calculated, and driving an OLLAMA model to perform semantic analysis on the unstructured clothing data to be pre-calculated to obtain a pre-calculated semantic analysis result, and storing the pre-calculated semantic analysis result in a semantic cache; processing a user query request; through predictive semantic preprocessing, the application can make full use of the idle time of computing resources, and complete time-consuming semantic analysis tasks in advance, thereby significantly reducing the delay of real-time queries and improving the response speed of the system.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a data center clothing management method and system based on OLLAMA. Background Technology

[0002] In the rapidly changing fashion industry, enterprise data centers need to process massive amounts of diverse apparel data, including structured information such as sales and inventory, as well as unstructured text such as design sketches, market reports, and social media comments. To extract business insights, enterprises have introduced solutions based on the OLLAMA model. However, this solution faces several challenges: rapidly updating fashion vocabulary, lagging data cleaning, frequent manual intervention, unstable model inference performance, and inconsistent outputs. These issues severely impact the real-time performance and accuracy of data analysis, hindering the enterprise's ability to respond quickly to the market.

[0003] Specifically, the rapid evolution of fashion terminology and the differences in expression among consumers in different regions pose significant challenges to traditional data cleaning methods based on pre-set dictionaries and grammatical patterns, making it difficult to accurately identify new terms and multilingual mixed text. The lag in data cleaning leads to unstable data quality input into the OLLAMA model, which in turn affects the depth of semantic understanding and the accuracy of the analysis results.

[0004] To compensate for deficiencies in the data cleaning process, data analysts have had to adopt a compensatory strategy: more frequently and meticulously adjusting the wording of OLLAMA model queries in an attempt to extract valuable information from incomplete or noisy raw data. This manual, iterative trial-and-error approach not only consumes significant human resources but also, because each interaction requires complex reasoning and contextual understanding from the OLLAMA model, significantly lengthens the single-response time of the OLLAMA model interface calls and result processing, failing to meet real-time analysis needs.

[0005] Furthermore, as market analysis teams increasingly rely on the OLLAMA model, and with the implementation of parallel query strategies to address the aforementioned response latency, the server graphics processing units (GPUs) responsible for OLLAMA model inference within the data center operate under consistently high loads. This sustained high-intensity computational demand causes the temperature of some server GPU cores to rise continuously, frequently reaching their thermal limits. When the server GPU temperature becomes too high, the hardware automatically triggers a thermal throttling mechanism, reducing its operating frequency to prevent overheating and damage. This mechanism results in periodic declines and recoveries in computational performance, manifesting as unstable fluctuations rather than a simple linear performance degradation. These performance fluctuations further exacerbate data processing latency and increase the uncertainty of the OLLAMA model's output results.

[0006] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of existing technologies by proposing a data center clothing management method and system based on OLLAMA.

[0008] In a first aspect, the present invention provides a data center clothing management method based on OLLAMA, the method comprising the following steps:

[0009] Collect computing resource status data;

[0010] Perform predictive semantic preprocessing, including:

[0011] When the computing resource status data indicates that the graphics processor is in a preset low load state, the unstructured clothing data to be pre-computed is selected based on the query pattern information obtained from the user query log analysis, and the OLLAMA model is driven to perform semantic analysis on the unstructured clothing data to be pre-computed to obtain the pre-computed semantic analysis result, and the pre-computed semantic analysis result is stored in the semantic cache.

[0012] Processing user query requests includes:

[0013] Receive user query requests and perform a search in the semantic cache according to the user query requests;

[0014] Generate a query response based on the search results.

[0015] Secondly, an OLLAMA-based data center apparel management system is provided, which includes:

[0016] The data acquisition module is used to collect computing resource status data;

[0017] The preprocessing module performs predictive semantic preprocessing, including:

[0018] When the computing resource status data indicates that the graphics processor is in a preset low load state, the unstructured clothing data to be pre-computed is selected based on the query pattern information obtained from the user query log analysis, and the OLLAMA model is driven to perform semantic analysis on the unstructured clothing data to be pre-computed to obtain the pre-computed semantic analysis result, and the pre-computed semantic analysis result is stored in the semantic cache.

[0019] The query processing module is used to process user query requests, including:

[0020] Receive user query requests and perform a search in the semantic cache according to the user query requests;

[0021] Generate a query response based on the search results.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] By employing predictive semantic preprocessing, this application fully utilizes idle computing resources to complete time-consuming semantic analysis tasks ahead of time, thereby significantly reducing real-time query latency and improving system response speed. Simultaneously, the introduction of semantic caching reduces repeated calls to the OLLAMA model, alleviating the real-time computational load on the graphics processor, avoiding hot throttling and performance fluctuations caused by high load operation, and improving model operational stability. Furthermore, through automated preprocessing and caching mechanisms, the reliance on data analysts manually adjusting model query wording is reduced, lowering labor costs and improving the real-time nature and accuracy of data analysis, enabling enterprises to respond more quickly to market changes. Attached Figure Description

[0024] Figure 1 This is a flowchart of the method of the present invention.

[0025] Figure 2 This is a schematic diagram of the system structure of the present invention.

[0026] In the diagram: 201, data acquisition module; 202, preprocessing module; 203, query processing module. Detailed Implementation

[0027] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0028] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0029] To better understand the technical solution proposed in this application, the key terms involved are explained first. The OLLAMA model refers to a large-scale language model deployed and run in a data center environment, primarily used for semantic understanding, analysis, and generation of unstructured text data. The Graphics Processing Unit (GPU) is the core computing hardware upon which the OLLAMA model relies, and its workload directly affects the model's inference performance. The semantic cache is an area storing pre-computed semantic analysis results, designed to accelerate the response to subsequent user queries by trading space for time. Unstructured clothing data refers to clothing-related information that is not stored in a fixed format, such as design sketches, market reports, and social media comments. Query pattern information is extracted from historical user query logs, revealing patterns such as user query habits, high-frequency query terms, and query themes.

[0030] like Figure 1 The illustrated method for managing clothing in a data center based on OLLAMA includes the following steps:

[0031] S101. Collect computing resource status data;

[0032] It's important to note that the purpose of this step is to obtain real-time information about the OLLAMA model's operating environment within the data center. For example, a monitoring agent deployed on the server can periodically collect data such as the graphics processor's workload, temperature, and the response time of the OLLAMA model during inference tasks. This data can be stored in log files or transmitted in real-time to a central monitoring system via an API interface. Alternatively, a resource monitoring module can be integrated into the OLLAMA model scheduler to automatically record and summarize relevant resource usage before and after each model inference task.

[0033] S102. Perform predictive semantic preprocessing, including:

[0034] When the computing resource status data indicates that the graphics processor is in a preset low load state, the unstructured clothing data to be pre-computed is selected based on the query pattern information obtained from the user query log analysis, and the OLLAMA model is driven to perform semantic analysis on the unstructured clothing data to be pre-computed, so as to obtain the pre-computed semantic analysis results and store the pre-computed semantic analysis results in the semantic cache.

[0035] It's important to note that when the collected computing resource status data indicates that the graphics processor is in a preset low-load state, the system intelligently selects unstructured clothing data to be pre-processed based on query pattern information obtained from user query log analysis. For example, if historical query logs show that users frequently query information about "new summer dresses," and the current graphics processor load is low, the system will prioritize preprocessing unstructured data related to "new summer dresses," such as design sketches and market reports. Subsequently, the system drives the OLLAMA model to perform semantic analysis on these selected unstructured clothing data to obtain pre-computed semantic analysis results. These results are then stored in a semantic cache for rapid retrieval later. As an alternative, query pattern information can also be used to predict user behavior through machine learning models, thereby more accurately identifying potential high-frequency query content.

[0036] Unstructured clothing data includes, for example, 3D fabric effect data, 3D garment effect data, and pattern / pattern data from the digital sample room; live-action try-on data and dynamic try-on video data generated from clothing modeling; and basic fabric information text and inventory quantity description text from the fabric management module. An example addressing the "unstructured clothing data type limitation": Assume a clothing data center is connected to multiple data sources. In the digital sample room module, the designer uses 3D modeling software to generate 3D effect data for a summer dress, including the garment's three-dimensional pattern and dynamic pleating effects of the material; simultaneously, various pattern data are created for the dress's design. Regarding fabric management, the system inputs basic fabric information for the dress, such as fabric composition (60% cotton, 40% polyester), weight, width, etc., as well as real-time updated fabric inventory data, recorded in text form in the inventory table. Furthermore, during the clothing modeling process, AI generates try-on data for models of different body types and dynamic runway video data for the dress. When the data center performs predictive semantic preprocessing, the 3D effect data and pattern data between these digital samples, the basic information text of fabric management, inventory data, and AI-generated try-on images and video data are all included in the scope of pre-computation as unstructured clothing data, driving the OLLAMA model to perform semantic analysis.

[0037] S103. Processing user query requests, including:

[0038] Receive user query requests and search the semantic cache according to the user query requests;

[0039] Generate a query response based on the search results.

[0040] It's important to note that this step begins by receiving a query request from the user. For example, a user might enter queries like "latest fashion trends" or "a brand's autumn / winter collection" through the front-end interface. Upon receiving these requests, the system first searches the semantic cache. If a pre-computed semantic analysis result highly matches the user's query, it directly extracts that result as the query response, thus avoiding real-time invocation of the OLLAMA model for inference and significantly shortening the response time. If no pre-computed result matching the conditions is found in the semantic cache, or if the confidence level of the real-time inference result is lower than a preset threshold, the system initiates a real-time inference process and generates a query response based on the inference result. For example, if a user queries "eco-friendly fabric clothing," the system first searches the cache; if no match is found, it then drives the OLLAMA model in real-time to analyze the relevant data and returns the analysis results.

[0041] The overall working principle of this application is to construct an efficient semantic cache by proactively utilizing idle periods of data center computing resources to perform predictive semantic preprocessing on potential user queries. When a user submits a query request, the system first attempts to quickly retrieve pre-computed results from the semantic cache to meet queries with high real-time requirements. This mechanism effectively alleviates the computational pressure faced by the OLLAMA model during peak periods and reduces the response time of model inference. Specifically, by continuously collecting computing resource status data, the system can intelligently identify low-load periods of the graphics processor. During these periods, the system combines in-depth analysis of user query patterns to accurately select unstructured clothing data related to high-frequency or potentially high-frequency queries. Subsequently, the OLLAMA model is driven to perform semantic analysis on this data, and the obtained pre-computed semantic analysis results are stored in the semantic cache. When a user query request arrives, the system prioritizes searching in the semantic cache. If a match is found, the pre-computed result is returned directly, greatly accelerating the response process. If a match is not found or the confidence of the real-time inference result is insufficient, the system will initiate real-time inference and may combine a multi-model arbitration mechanism to ensure the accuracy and consistency of the query response. This collaborative approach enables various technical features to work closely together, jointly solving problems such as data processing lag, long response time, unstable model performance, and inconsistent results in existing technologies, and significantly improving the efficiency and accuracy of data center clothing management methods.

[0042] Compared to existing technologies, the core innovation of this application lies in the introduction of a predictive semantic preprocessing mechanism based on computing resource status. Traditional methods often only initiate the inference process of the OLLAMA model after receiving a user query request. This results in long response times when processing massive amounts of unstructured data, especially when the graphics processor is under heavy load, where the performance bottleneck becomes more pronounced. This application effectively avoids this problem by performing semantic analysis on unstructured clothing data that users may query in advance and caching the results when the graphics processor is in a preset low-load state. For example, at night or during off-peak business hours, the system can utilize idle computing resources to preprocess the latest fashion trend reports, social media comments, etc., and pre-store the semantic analysis results of high-frequency queries such as "2024 Spring / Summer trending colors" and "sustainable fashion brands." When users query this content during peak daytime hours, the system can directly retrieve the results from the semantic cache without needing to call the OLLAMA model in real time for time-consuming inference, thereby reducing the query response time from several seconds or even tens of seconds to the millisecond level. This forward-looking approach not only significantly improves the user experience but also optimizes the utilization efficiency of data center computing resources, reducing energy consumption and hardware wear caused by real-time model inference. Furthermore, by analyzing user query logs, this application can more accurately predict user needs, making the preprocessing process more targeted and avoiding resource waste caused by blind pre-computation. Therefore, this application demonstrates significant advancements in real-time performance, efficiency, and resource utilization.

[0043] As one embodiment of the present invention, the computing resource status data includes: workload data and temperature data of the graphics processor on which the OLLAMA model depends for running, as well as the response time of the OLLAMA model in completing the inference task.

[0044] The graphics processing unit (GPU) workload data refers to the workload of the GPU during a specific time period, usually expressed as a percentage, reflecting the utilization rate of its computing resources. Temperature data refers to the real-time temperature of the GPU during operation, a key indicator for assessing hardware health and overheating risk. The response time of the OLLAMA model inference task measures the time required for the OLLAMA model to complete and return a result after receiving an inference request, directly reflecting the model's processing efficiency and the system's real-time performance. These data collectively constitute a comprehensive monitoring of the OLLAMA model's operating environment and performance.

[0045] This application's solution, by collecting workload and temperature data from the graphics processing unit (GPU), can monitor the usage and health status of the hardware resources upon which the OLLAMA model depends in real time. When workload data indicates that the GPU is at a low utilization level, and temperature data is within the normal range, it can more accurately determine that the GPU is in a preset low-load state. Simultaneously, by monitoring the response time of the OLLAMA model in completing inference tasks, the actual performance of the model under the current load can be evaluated, thus providing a more reliable basis for triggering predictive semantic preprocessing. This multi-dimensional data acquisition method enables the system to more intelligently identify resource idle periods and optimize the scheduling of preprocessing tasks.

[0046] As one embodiment of the present invention, the step of generating a query response based on the search results includes:

[0047] If the search result is a pre-computed semantic analysis result that satisfies the user's query request and exists in the semantic cache, then the pre-computed semantic analysis result is extracted as the query response.

[0048] If the search result is that there is no pre-computed semantic analysis result in the semantic cache that satisfies the user's query request, or the confidence of the result in real-time inference by the OLLAMA model is lower than the preset threshold, then the user's query request is submitted to multiple OLLAMA model instances in parallel, and the output results of multiple OLLAMA model instances are received.

[0049] It should be noted that multiple OLLAMA model instances can be OLLAMA models configured with different parameters, different versions, or fine-tuned on different datasets, or they can simply be multiple copies of the same OLLAMA model deployed on different computing resources to achieve redundancy and diversity. Parallel submissions allow for obtaining independent inference results from multiple models simultaneously.

[0050] The outputs of multiple OLLAMA model instances are compared, and the semantic analysis results with high consistency and high confidence are selected as the final query response according to the preset arbitration rules.

[0051] It should be noted that the outputs of multiple OLLAMA model instances are compared. This comparison process aims to evaluate the consistency and differences between the outputs of different models. Based on the comparison results, the system will select the semantic analysis result with high consistency and high confidence as the final query response according to preset arbitration rules. Preset arbitration rules may include, but are not limited to, majority voting mechanisms, weighted averaging, weight allocation based on historical evaluations of model performance, or decision-making in conjunction with expert systems. The purpose is to effectively avoid the biases, errors, or "illusions" that may exist in a single model by combining the wisdom of multiple models, thereby improving the quality and reliability of the final query response.

[0052] This application's solution effectively addresses the low confidence issue that may arise during real-time inference with a single OLLAMA model, as well as the response quality problem when the semantic cache misses, by introducing a parallel inference and arbitration mechanism involving multiple OLLAMA model instances. Specifically, when the semantic cache cannot provide a valid response or the inference result of a single model is unreliable, the system no longer passively waits or directly outputs uncertain results. Instead, it actively utilizes the parallel processing capabilities of multiple OLLAMA model instances. By simultaneously acquiring independent insights from multiple models and comparing and arbitrating these insights, the accuracy of the semantic analysis results can be verified from multiple dimensions. High consistency results generally imply higher reliability, while high confidence results directly reflect the model's confidence in its own judgment. This mechanism is essentially an application of ensemble learning or consensus mechanisms. Through a "brainstorming" approach, it significantly reduces the risk of error propagation from a single model, thereby ensuring the generation of high-quality, highly reliable query responses even in complex or uncertain scenarios.

[0053] Through the above technical solution, this application can significantly improve the robustness and accuracy of the data center clothing management method in processing user query requests. Especially in cases of semantic cache misses or low confidence levels in real-time inference results from the OLLAMA model, by utilizing multiple OLLAMA model instances in parallel and performing intelligent arbitration, erroneous or inaccurate responses caused by the limitations of a single model can be effectively avoided. This not only improves the reliability and consistency of query responses and reduces the probability of "illusion" phenomena, but also provides users with a more stable and reliable query service experience, thereby enhancing the intelligence level and practical value of the entire data center clothing management system.

[0054] In some preferred embodiments, a specific example is given below. Suppose a user queries "Please recommend a breathable, quick-drying T-shirt suitable for summer outdoor sports." The system first searches in the semantic cache but fails to find a perfectly matching pre-computed semantic analysis result. At this point, if only an OLLAMA model is relied upon for real-time inference, the model may provide a recommendation with low confidence due to training data bias or query complexity, such as recommending a "cotton T-shirt," which contradicts the requirement of "quick-drying."

[0055] According to the scheme of this application, the system will submit the user query request to three OLLAMA model instances in parallel.

[0056] The first OLLAMA model instance might recommend "CoolMax quick-drying T-shirt from a certain brand" with a high confidence level.

[0057] The second OLLAMA model instance might recommend "a certain brand of sweat-wicking and breathable sports T-shirt" and also give a high confidence level.

[0058] The third OLLAMA model instance might recommend "a certain brand of pure cotton casual T-shirt", but with low confidence.

[0059] After receiving these three outputs, the system compares them. The comparison reveals that the recommendations from the first two model instances focus on core semantic features such as "quick-drying," "breathable," and "sports," and exhibit high confidence and consistency. However, the recommendations from the third model instance deviate significantly from user needs and have low confidence. Based on pre-defined arbitration rules (e.g., selecting the majority of high-confidence and semantically consistent results), the system chooses the "quick-drying, breathable sports T-shirt" recommended by the first two model instances as the final query response, potentially integrating the advantages of both or providing more detailed recommendation information. Therefore, even in cases of cache misses or potential errors in a single model, the system can provide accurate, reliable, and user-expected query responses.

[0060] As one embodiment of the present invention, the steps of comparing the output results of multiple OLLAMA model instances and selecting the semantic analysis result with high consistency and high confidence as the final query response according to a preset arbitration rule include:

[0061] Capture the semantic analysis output of multiple OLLAMA model instances and convert them into a unified format;

[0062] It's important to clarify that this specifically refers to collecting semantic analysis results generated by various parallel-running OLLAMA model instances in response to the same user query. Since different OLLAMA model instances may employ different internal representations or output structures, these raw outputs need to be standardized into a unified, parsable data format for effective subsequent comparison and analysis. For example, all outputs can be converted to JSON, XML, or a specific semantic triple format to ensure data consistency and operability. The aim is to provide a standardized input foundation for subsequent consensus and disagreement quantification.

[0063] Consensus and divergence quantification is performed on the semantic analysis output in a unified format to obtain entity consensus degree, sentiment polarity divergence degree, and summary content similarity.

[0064] It's important to note that this can be understood as a fine-grained semantic comparison of the standardized model output. Entity consensus measures the consistency among different models in identifying and mentioning the same entities (e.g., specific clothing brands, styles, materials, etc.). Sentiment polarity divergence assesses the differences among different models in their judgments of sentimental inclination (e.g., positive, negative, neutral) towards a specific entity or concept. Summary content similarity quantifies the similarity in core content expression by comparing the text summaries or key information output by the models. The purpose of these quantitative metrics is to gain a deeper understanding of the similarities and differences in model outputs from multiple dimensions.

[0065] Based on the quantification results of consensus and disagreement, a semantic graph is constructed. The semantic graph presents semantic consensus regions and semantic disagreement regions, and associates underlying entities with abstract concepts.

[0066] It's important to note that this means using the aforementioned quantitative metrics to visualize and structure the entities, concepts, and their relationships in the model output as a graph. Nodes in the semantic graph can represent entities or abstract concepts, while edges represent the relationships between them. Consensus regions refer to highly consistent semantic information across multiple model outputs, while divergence regions highlight areas where differences or conflicts exist between models. By associating underlying entities (such as "dress" or "linen") with abstract concepts (such as "summer wear" or "casual style"), a more comprehensive understanding of the semantic content can be achieved. The aim is to provide an intuitive and structured view to facilitate the identification and analysis of commonalities and individual characteristics in the model output.

[0067] Quantitative indicators of consensus and disagreement are appended as metadata to semantic graph elements;

[0068] It should be noted that this refers to attaching calculated quantitative values ​​such as entity consensus, sentiment polarity divergence, and summary content similarity as additional information tags to the corresponding nodes or edges in the semantic graph. For example, an entity node representing "dress" could be appended with metadata of high consensus, while an associated edge representing "suitable occasion" might be appended with metadata of medium divergence. The purpose is to provide richer contextual information for each element in the semantic graph, assisting in subsequent arbitration decisions.

[0069] Based on the preset arbitration rules and semantic graph elements, semantic analysis results with high consistency and high confidence are extracted from the semantic graph as the final query response.

[0070] It's important to note that this means during the arbitration phase, the system no longer relies solely on simple text matching or voting mechanisms. Instead, it comprehensively considers consensus regions, divergence regions, the relationships between underlying entities and abstract concepts presented in the semantic graph, as well as additional consensus and divergence quantification metadata. Pre-defined arbitration rules can leverage this rich information—such as prioritizing entities with high consensus, conducting in-depth analysis of divergence regions to identify their nature, and combining this with the model's own confidence level—to make a more informed and accurate final decision.

[0071] This application's solution effectively addresses the potential robustness and accuracy issues when comparing the outputs of multiple OLLAMA model instances by introducing multi-dimensional consensus and disagreement quantification and semantic graph construction. First, by converting the semantic analysis outputs of different model instances into a unified format, standardization and comparability of subsequent analyses are ensured. Second, by quantifying entity consensus, sentiment polarity disagreement, and summary content similarity, the system can gain a deeper understanding of the similarities and differences in model outputs from multiple fine-grained dimensions, going beyond mere surface text matching. It is this meticulous quantitative analysis that enables the system to identify the commonalities and differences between models in specific entities, sentiment tendencies, and core content. Based on this, a semantic graph is constructed, and quantitative indicators are appended as metadata to the graph elements, making the semantic information of the model output structured, visualized, and rich in contextual information. The semantic graph clearly presents semantic consensus and disagreement regions and connects underlying entities with abstract concepts, thus providing a more comprehensive and intuitive basis for decision-making in the arbitration process. Ultimately, by combining preset arbitration rules with these rich semantic graph elements, the system can more accurately identify and extract truly consistent and confident semantic analysis results, thereby avoiding misjudgments or information omissions that may result from simple comparisons.

[0072] Through the aforementioned technical solutions, this application significantly improves the accuracy, robustness, and interpretability of semantic analysis results in data center clothing management methods when handling complex user queries. Compared to simple output comparisons, this application, through meticulous consensus and disagreement quantification, can gain deeper insights into subtle differences and potential conflicts between the outputs of different OLLAMA model instances, thereby avoiding arbitration errors caused by incomplete information or misunderstandings. Furthermore, the construction of the semantic graph not only provides an intuitive semantic representation but also, by associating underlying entities with abstract concepts, enables the system to better understand the deeper intent of user queries and the inherent logic of model outputs. Additional metadata further enriches the graph information, allowing arbitration rules to make decisions based on a more comprehensive context. Therefore, the final generated query response not only has higher consistency and confidence but also greater transparency and traceability in its generation process, greatly enhancing the system's intelligence and user experience.

[0073] As one embodiment of the present invention, based on the consensus and divergence quantification results, a semantic graph is constructed. The semantic graph presents semantic consensus regions and semantic divergence regions, and the steps of associating underlying entities with abstract concepts include:

[0074] Based on the consensus and disagreement quantification results, we identify newly emerging abstract concepts or entities in the semantic analysis output, as well as semantic consensus regions and semantic disagreement regions.

[0075] Based on the newly identified abstract concepts or entities, as well as semantic consensus regions and semantic divergence regions, the nodes and associated edges of the semantic graph are expanded, and the underlying entities are associated with the abstract concepts.

[0076] Specifically, in constructing a semantic graph, the first step is to conduct an in-depth analysis of the semantic analysis output of OLLAMA model instances. "Identifying newly emerging abstract concepts or entities in the semantic analysis output" refers to the system using Natural Language Processing (NLP) techniques, such as Named Entity Recognition (NER), keyword extraction, and topic modeling, to discover and extract words or phrases with independent semantic meaning that do not yet exist in the existing semantic graph from the text after consensus and disagreement quantification. These newly emerging words or phrases can be specific entities such as clothing brands, styles, and materials, or abstract concepts such as fashion trends, styles, and occasions. Simultaneously, "identifying semantic consensus regions and semantic disagreement regions" refers to determining, based on previously quantified indicators such as entity consensus, sentiment polarity disagreement, and summary content similarity, which parts of the semantic analysis output represent consensus regions (where multiple OLLAMA model instances agree) and which parts exhibit significant differences or contradictions (where disagreement regions).

[0077] Furthermore, "expanding the nodes and edges of the semantic graph" refers to adding newly identified abstract concepts or entities as new nodes to the semantic graph. Simultaneously, based on the semantic relationships between these new nodes and existing nodes—such as hierarchical relationships, synonymous relationships, causal relationships, and attribute relationships—related edges in the graph are created or updated. For example, if the abstract concept of "retro style" is identified and found to be strongly associated with entities such as "flared pants" and "polka dot skirt," corresponding nodes and edges will be added to the graph. In addition, "associating underlying entities with abstract concepts" means ensuring that the semantic graph not only contains high-level abstract concepts but also concrete, perceptible underlying entities, connecting them through explicit related edges to form a multi-layered, structured knowledge network. For example, the abstract concept of "dress" is associated with entities of specific materials such as "silk dress" and "cotton and linen dress," or the abstract concept of "summer clothing" is associated with specific clothing entities such as "short-sleeved T-shirt" and "shorts."

[0078] This application's solution, through in-depth utilization of consensus and disagreement quantification results, makes the semantic graph construction process more dynamic and precise. By identifying newly emerging abstract concepts or entities in the semantic analysis output, it ensures that the semantic graph can continuously absorb and integrate the latest fashion information and user query patterns, avoiding content aging and information loss. Simultaneously, by identifying semantic consensus regions and semantic disagreement regions, the graph can be expanded in a targeted manner. Knowledge connections are strengthened in consensus regions, while potential ambiguities or polysemy are marked in disagreement regions, providing a basis for subsequent arbitration and interpretation. Therefore, by using these newly identified elements as nodes and expanding associated edges based on their semantic relationships, a more comprehensive, detailed semantic graph that reflects the differences in multi-model outputs can be constructed, thus providing a richer and more accurate semantic context for subsequent query response generation.

[0079] As one embodiment of the present invention, the step of constructing a semantic graph based on the consensus and disagreement quantification results, wherein the semantic graph presents semantic consensus regions and semantic disagreement regions, and associating underlying entities with abstract concepts, further includes:

[0080] Based on the contextual information contained in the user query, and combined with the confidence in the association between specific entities and abstract concepts in the current context, the strength of the association edges in the semantic graph is adjusted.

[0081] It's important to note that contextual information can be understood as additional information such as the specific background, time, location, user preferences, historical interaction records, or current fashion trends in which the user queries. For example, when a user queries "summer dress," contextual information might include the user's geographical location (e.g., tropical regions), the current season (summer), and the user's past preferences for lightweight, breathable fabrics. The confidence level of the association between a specific entity and an abstract concept in the current context refers to the reliability or probability of the association between a certain entity (e.g., "cotton and linen fabric") and an abstract concept (e.g., "breathability") in a specific context. This confidence level can be evaluated through historical data analysis, expert knowledge, or machine learning models. Adjusting the strength of the association edges in the semantic graph refers to dynamically increasing or decreasing the weights of the connections between nodes in the semantic graph based on the aforementioned contextual information and confidence level, to reflect the strength of their relevance in the current context. For example, in a summer context, the strength of the association edge between "dress" and "breathability" would be increased.

[0082] Based on the user's query intent, by identifying the core abstract concepts of the query and performing multi-step semantic association, we explore and make explicit the deep-level association paths connected by intermediate entities or secondary abstract concepts, and construct multi-level abstract concept association paths in the semantic graph.

[0083] It's important to clarify that user query intent refers to the true purpose or need implied when a user initiates a query. For example, when a user queries "business casual wear," their intent might be to find clothing suitable for the office but not too formal. Identifying the core abstract concepts of a query means extracting the high-level concepts that best represent the user's intent from the query text. For instance, identifying core concepts like "spring," "outing," "lightweight," and "jacket" from "lightweight jacket suitable for a spring outing." Multi-step semantic association refers to exploring and discovering other indirectly related concepts or entities in a semantic graph, starting from a single concept or entity and progressing through multiple intermediate nodes and connecting edges. For example, associating "dress" with "fabric," then with "breathability," and finally with "cotton and linen." Exploring and making explicit the deep-level association paths connected by intermediate entities or secondary abstract concepts means clearly representing these indirect association paths discovered through multi-step association, making them visible and usable in the semantic graph. Constructing multi-level abstract concept association paths in a semantic graph refers to establishing and maintaining connections between concepts at different levels of abstraction in the semantic graph, such as from "clothing" to "top" and then to "shirt," forming a hierarchical association structure.

[0084] This application's solution addresses the problem of fixed association strength and lack of flexibility in traditional semantic graphs when handling dynamic contexts by dynamically adjusting the strength of association edges through the introduction of contextual information. Specifically, when a user's query request contains contextual information, the system can combine this information to assess the association confidence between a specific entity and an abstract concept in the current context, and dynamically adjust the strength of the corresponding association edges in the semantic graph accordingly. For example, in the context of "summer," if a user queries "dress," the association strength between "dress" and "breathable fabric" will be significantly enhanced, while the association strength with "wool fabric" will be weakened, thus enabling the semantic graph to more accurately reflect the semantic associations in the current context.

[0085] Furthermore, this application, by identifying the core abstract concepts in a user's query and performing multi-step semantic association, can explore and explicit deep-level association paths, constructing multi-level abstract concept association paths. This enables the system not only to identify direct associations but also to discover indirect and hidden semantic relationships, thus solving the problem of insufficient depth that traditional solutions may have when understanding complex user intent. For example, when a user queries "clothing suitable for outdoor sports," the system can not only directly associate it with "sportswear," but also, through multi-step association, associate "outdoor sports" with "sunscreen" and "quick-drying," then with specific clothing items such as "sunscreen clothing" and "quick-drying pants," and further with attributes such as "lightweight" and "durable," thereby constructing a more comprehensive semantic association path that better meets the user's deeper needs.

[0086] As one embodiment of the present invention, after performing consensus and divergence quantification on the semantic analysis output in a unified format to obtain entity consensus degree, sentiment polarity divergence degree, and summary content similarity, the method includes:

[0087] Based on entity consensus, sentiment polarity divergence, and summary content similarity, identify text fragments with divergence in the semantic analysis output;

[0088] Extract the semantic features that lead to the disagreement from the disagreeing text fragments;

[0089] Based on semantic features, the disagreements are categorized into predefined disagreement types;

[0090] Generate an explanatory description of the divergence type and attach it as metadata to the corresponding divergence region in the semantic graph.

[0091] Specifically, after quantifying consensus and divergence in the semantic analysis output of a unified format to obtain entity consensus, sentiment polarity divergence, and summary content similarity, the system uses these quantitative indicators to identify text fragments that show significant differences or inconsistencies in the outputs of different OLLAMA model instances. For example, by setting thresholds, a text fragment is marked as divergent text fragment when its entity consensus in different model outputs is below a certain threshold or its sentiment polarity divergence is above a certain threshold. Further, once divergent text fragments are identified, the specific semantic features causing the divergence are extracted from these fragments. These semantic features can be words, phrases, sentence structures, or even implicit contextual information. For example, Natural Language Processing (NLP) techniques, such as keyword extraction, dependency parsing, and named entity recognition, can be used to locate and extract these key semantic elements. The goal is to accurately identify the root causes of differences in understanding between different models. Based on this, the identified divergences are categorized into preset divergence types according to the extracted semantic features. These types of disagreements can be defined based on practical application scenarios, such as polysemous ambiguity, context-dependent disagreement, regional cultural differences disagreement, factual error disagreement, and opinion differences disagreement. Classification allows for structured management of disagreements, facilitating subsequent analysis and processing. Finally, for each categorized disagreement type, a corresponding explanatory description is generated. This description aims to clearly elucidate the nature of the disagreement, its possible causes, and its impact on overall semantic understanding. For example, for the "polysemous ambiguity" type, the explanatory description can indicate which specific word is ambiguous and how different models interpret it differently. These explanatory descriptions are then appended as metadata to the corresponding disagreement regions in the semantic graph, enabling the semantic graph to not only display consensus and disagreement but also provide the underlying causes and explanations of the disagreements, greatly enhancing the graph's diagnostic and interpretability.

[0092] This application's solution, after quantifying consensus and disagreement, introduces the identification of disagreeing text fragments, semantic feature extraction, disagreement type classification, and explanatory description generation, effectively addressing the limitations of the aforementioned technical foundations that only quantify disagreement without deeply understanding its causes. Specifically, when multiple OLLAMA model instances produce different outputs for the same user query, the system first uses quantified entity consensus, sentiment polarity disagreement, and summary content similarity to accurately locate inconsistent text fragments in the semantic analysis output. Subsequently, through detailed semantic analysis of these disagreeing text fragments, the system extracts key semantic features that lead to differences in understanding between models, such as the polysemy of a word, different interpretations in specific contexts, or regional differences in expression. Based on these extracted semantic features, the system can categorize complex disagreements into predefined, easily understood types, thus concretizing abstract disagreements. Finally, a clear explanatory description is generated for each disagreement type and integrated as supplementary information into the semantic graph, enabling the semantic graph to not only present semantic consensus and disagreement areas but also provide a "diagnostic report" of the disagreements. This series of steps enables the system to move from "knowing there is a disagreement" to "knowing what the disagreement is and why there is a disagreement," providing richer and more detailed evidence for subsequent arbitration decisions.

[0093] As one embodiment of the present invention, semantic features include polysemous words, contextual modifiers, or regional expressions; the types of divergence include polysemous ambiguity, context-dependent divergence, or regional cultural difference divergence.

[0094] Specifically, polysemous words are words that have multiple meanings in different contexts. For example, the word "fashion" may refer to different styles or trends in different cultural backgrounds. Contextual modifiers are words whose meanings or effects change depending on the specific context. For example, "leisure" has different specific requirements in loungewear and business casual wear. Regional expressions refer to vocabulary, phrases, or expressions unique to a particular region or cultural sphere. For example, the understanding of "formal wear" in some regions may differ from internationally accepted standards.

[0095] Among these, polysemous ambiguity refers to the discrepancy caused by different OLLAMA model instances interpreting the same word differently due to the existence of polysemous words. Context-dependent ambiguity refers to the differences in semantic processing among different OLLAMA model instances in specific contexts due to the presence of contextual modifiers. Regional cultural difference ambiguity refers to the inconsistency among different OLLAMA model instances in understanding and interpreting clothing concepts within a specific cultural context due to the existence of regional expressions.

[0096] This application's solution identifies specific semantic features that lead to semantic discrepancies, such as polysemous words, contextual modifiers, or regional expressions. Based on this, the discrepancies are categorized into polysemous ambiguity, context-dependent discrepancies, or regional cultural differences. This allows the system to perform detailed analysis of discrepancies in the output of OLLAMA model instances. It is precisely this accurate identification of the causes of discrepancies that enables subsequent explanatory descriptions of the discrepancies to more accurately reflect the essence of the problem, providing users with more insightful information.

[0097] As one embodiment of the present invention, the step of adjusting the strength of the association edges in the semantic graph based on the contextual information contained in the user query and the confidence level of the association between a specific entity and an abstract concept in the current context includes:

[0098] Monitoring the co-occurrence of entities and abstract concepts in context within fashion data streams;

[0099] Specifically, "monitoring the co-occurrence of entities and abstract concepts in context within fashion data streams" refers to the system continuously collecting fashion-related information from various real-time or near-real-time data sources, such as social media posts, fashion blog articles, product descriptions and user reviews on online retail platforms, and fashion magazine content. These data streams contain a large amount of text, image, and video information. Through natural language processing and image recognition technologies, specific "entities" (such as "dress," "jeans," and "sneakers") and "abstract concepts" (such as "retro style," "minimalism," and "street fashion") can be extracted. "Context" refers to the specific background in which these entities and concepts appear, such as "summer outfits," "business casual," and "party occasions." Co-occurrence indicates that two or more entities or concepts appear simultaneously or are closely related within the same context.

[0100] Within a time window, the co-occurrence frequency and recency of entities and abstract concepts in a context are quantified. This refers to counting the number of times a specific entity and abstract concept co-occur in a specific context within a defined time period (e.g., the past 24 hours, week, or month), i.e., the "co-occurrence frequency." Meanwhile, "recency" measures how close these co-occurring events are to the current moment; more recently occurring co-occurring events are typically given higher weight to reflect the latest trends. For example, an exponential decay function can be used to calculate the recency score, making more recent events contribute more to the total score.

[0101] Comparing co-occurrence frequency and recency with a benchmark helps identify changes in the trend of association between entities and abstract concepts within a given context. This involves comparing the co-occurrence frequency and recency scores calculated within the current time window with pre-defined historical averages, long-term trends, or standard values ​​for a specific context (i.e., the "benchmark"). This comparison allows for the determination of whether the current association strength between entities and abstract concepts is significantly increasing, decreasing, or remaining stable, thus identifying changes in the association trend. For example, if the co-occurrence frequency and recency scores of "oversized suit" and "business casual" over the past week are significantly higher than their historical benchmarks, it indicates that the association trend between these two is strengthening.

[0102] Based on the contextual information contained in the user query, combined with the confidence level of the association between a specific entity and an abstract concept in the current context, and according to the identified trend changes in association, the system adjusts the strength of the association edges in the semantic graph. Specifically, when receiving a user query, the system first parses the implicit or explicit contextual information in the query. Then, it combines the existing confidence levels of the association between a specific entity and an abstract concept in the current context (e.g., based on historical data or expert knowledge) and comprehensively considers the trend changes in association identified through the above steps. If the trend shows that a certain association is strengthening, the strength of the association edge connecting these two entities or concepts in the semantic graph is increased accordingly; conversely, if the trend weakens, the edge strength is decreased. This dynamic adjustment ensures that the semantic graph can reflect the latest developments in the fashion field in real time.

[0103] The solution proposed in this application effectively addresses the lag issue in adjusting the strength of semantic graph association edges in existing technologies because it introduces a real-time monitoring and trend analysis mechanism for fashion data streams. By continuously monitoring the co-occurrence of entities and abstract concepts within context in the fashion data stream, the system can capture the nascent and evolving relationships between various elements in the fashion field. Subsequently, by quantifying the co-occurrence frequency and proximate occurrence of entities and abstract concepts within a time window, the system not only focuses on the number of co-occurrences but also emphasizes recent co-occurrence events, thus ensuring sensitivity to the latest trends. It is precisely by comparing the co-occurrence frequency and proximate occurrence with a benchmark to identify changes in the trend of association between entities and abstract concepts within context that the system can objectively determine which associations are becoming stronger or weaker, rather than relying solely on static experience. Ultimately, by "adjusting the strength of the association edges in the semantic graph based on the contextual information contained in the user query, the confidence level of the association between specific entities and abstract concepts in the current context, and the changes in the identified association trends," the system ensures that the associations in the semantic graph can dynamically and accurately reflect the real connections in the current fashion context, thereby providing users with more timely and relevant query results.

[0104] like Figure 2The illustrated data center apparel management system based on OLLAMA includes:

[0105] Acquisition module 201 is used to collect computing resource status data;

[0106] Preprocessing module 202, used to perform predictive semantic preprocessing, includes:

[0107] When the computing resource status data indicates that the graphics processor is in a preset low load state, the unstructured clothing data to be pre-computed is selected based on the query pattern information obtained from the user query log analysis, and the OLLAMA model is driven to perform semantic analysis on the unstructured clothing data to be pre-computed, so as to obtain the pre-computed semantic analysis results and store the pre-computed semantic analysis results in the semantic cache.

[0108] Query processing module 203 is used to process user query requests, including:

[0109] Receive user query requests and search the semantic cache according to the user query requests;

[0110] Generate a query response based on the search results.

[0111] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A data center clothing management method based on OLLAMA, characterized in that, The method includes the following steps: S101. Collect computing resource status data; S102. Perform predictive semantic preprocessing, including: When the computing resource status data indicates that the graphics processor is in a preset low load state, the unstructured clothing data to be pre-computed is selected based on the query pattern information obtained from the user query log analysis, and the OLLAMA model is driven to perform semantic analysis on the unstructured clothing data to be pre-computed to obtain the pre-computed semantic analysis result, and the pre-computed semantic analysis result is stored in the semantic cache. S103. Processing user query requests, including: Receive user query requests and perform a search in the semantic cache according to the user query requests; Generate a query response based on the search results; The step of generating a query response based on the search results includes: If the search result shows that there is a pre-computed semantic analysis result in the semantic cache that satisfies the user's query request, then the pre-computed semantic analysis result is extracted as the query response; If the search result is that there is no pre-computed semantic analysis result in the semantic cache that satisfies the user's query request, or the confidence level of the result inferred in real time by the OLLAMA model is lower than a preset threshold, then the user's query request is submitted to multiple OLLAMA model instances in parallel, and the output results of the multiple OLLAMA model instances are received. The output results of the multiple OLLAMA model instances are compared, and the semantic analysis result with high consistency and high confidence is selected as the final query response according to the preset arbitration rules. The step of comparing the output results of the multiple OLLAMA model instances and selecting the semantic analysis result with high consistency and high confidence as the final query response according to the preset arbitration rules includes: Capture the semantic analysis output of the multiple OLLAMA model instances and convert it into a unified format; Consensus and divergence quantifications are performed on the semantic analysis output of the unified format to obtain entity consensus degree, sentiment polarity divergence degree, and summary content similarity. Based on the consensus and disagreement quantification results, a semantic graph is constructed, which presents semantic consensus regions and semantic disagreement regions, and associates underlying entities with abstract concepts; The consensus and disagreement quantification indicators are appended as metadata to semantic graph elements; Based on the preset arbitration rules and semantic graph elements, semantic analysis results with high consistency and high confidence are extracted from the semantic graph as the final query response.

2. The data center clothing management method based on OLLAMA according to claim 1, characterized in that, The computing resource status data includes: workload and temperature data of the graphics processor on which the OLLAMA model depends for operation, as well as the response time of the OLLAMA model in completing inference tasks.

3. The data center clothing management method based on OLLAMA according to claim 1, characterized in that, The step of constructing a semantic graph based on the consensus and disagreement quantification results, wherein the semantic graph presents semantic consensus regions and semantic disagreement regions, and associates underlying entities with abstract concepts, includes: Based on the consensus and disagreement quantification results, newly emerging abstract concepts or entities in the semantic analysis output are identified, as well as the semantic consensus region and semantic disagreement region. Based on the newly identified abstract concepts or entities, as well as the semantic consensus region and semantic divergence region, the nodes and associated edges of the semantic graph are expanded, and the underlying entities are associated with the abstract concepts.

4. The data center clothing management method based on OLLAMA according to claim 3, characterized in that, The step of constructing a semantic graph based on the consensus and disagreement quantification results, wherein the semantic graph presents semantic consensus regions and semantic disagreement regions, and associates underlying entities with abstract concepts, further includes: Based on the contextual information contained in the user query, and combined with the confidence level of the association between a specific entity and an abstract concept in the current context, the strength of the association edges in the semantic graph is adjusted. Based on the user's query intent, by identifying the core abstract concept of the query and performing multi-step semantic association, the system explores and makes explicit the deep-level association paths connected by intermediate entities or secondary abstract concepts, thus constructing multi-level abstract concept association paths in the semantic graph.

5. The data center clothing management method based on OLLAMA according to claim 1, characterized in that, After the steps of quantifying consensus and divergence in the semantic analysis output of the unified format to obtain entity consensus degree, sentiment polarity divergence degree, and summary content similarity, the following steps are included: Based on the entity consensus, sentiment polarity divergence, and summary content similarity, identify the text fragments with divergence in the semantic analysis output; Extract the semantic features that lead to the disagreement from the disagreeing text fragments, including polysemous words, contextual modifiers, or regional expressions; Based on the semantic features, the disagreements are classified into preset disagreement types, including polysemous ambiguity, context-dependent disagreement, or regional cultural difference disagreement. An explanatory description of the divergence type is generated and appended as metadata to the corresponding divergence region in the semantic graph.

6. The data center clothing management method based on OLLAMA according to claim 4, characterized in that, The step of adjusting the strength of the association edges in the semantic graph based on the contextual information contained in the user query and the confidence level of the association between a specific entity and an abstract concept in the current context includes: Monitoring the co-occurrence of entities and abstract concepts in context within fashion data streams; Within a time window, quantify the co-occurrence frequency and proximity of the entity and the abstract concept in the context; By comparing the co-occurrence frequency and proximity with a benchmark, the changing trend of the association between the entity and the abstract concept in the context can be identified. Based on the contextual information contained in the user query, combined with the confidence level of the association between specific entities and abstract concepts in the current context, and according to the identified trend changes in association, the strength of the association edges in the semantic graph is adjusted.

7. A data center clothing management system based on OLLAMA, used to execute a data center clothing management method based on OLLAMA as described in any one of claims 1-6, characterized in that, The system includes: The data acquisition module is used to collect computing resource status data; The preprocessing module performs predictive semantic preprocessing, including: When the computing resource status data indicates that the graphics processor is in a preset low load state, the unstructured clothing data to be pre-computed is selected based on the query pattern information obtained from the user query log analysis, and the OLLAMA model is driven to perform semantic analysis on the unstructured clothing data to be pre-computed to obtain the pre-computed semantic analysis result, and the pre-computed semantic analysis result is stored in the semantic cache. The query processing module is used to process user query requests, including: Receive user query requests and perform a search in the semantic cache according to the user query requests; Generate a query response based on the search results.

Images