Supply chain optimization result visualization and feedback method and device based on artificial intelligence interaction

Abstract

The embodiment of the invention provides a supply chain optimization result visualization and feedback method and device based on artificial intelligence interaction, and relates to the technical field of supply chain optimization result visualization technologies. The method comprises the following steps: receiving language information for at least one supply chain optimization result; generating a structured query request according to the language information; based on the structured query request, obtaining optimization scheme data corresponding to the decision object entity, and generating and displaying at least one visual chart according to a visual demand; receiving multi-modal feedback information input by a user on the visual chart; performing fusion analysis on the multi-modal feedback information to generate a structured algorithm optimization instruction; and performing recalculation based on the algorithm optimization instruction to generate an updated supply chain optimization result. According to the method and the device, the problem of poor interpretation of the algorithm result in the existing supply chain optimization system is solved, and the effect of improving the interpretation of the algorithm result in the supply chain optimization system is further achieved.

Description

Technical Field

[0001] This invention relates to the field of supply chain optimization result visualization, and more specifically, to a method and apparatus for visualization and feedback of supply chain optimization results based on artificial intelligence interaction. Background Technology

[0002] Supply chain management is a core component of business operations, and its complexity continues to increase with changes in the market environment. To improve operational efficiency, reduce inventory costs, and optimize logistics routes, companies widely adopt supply chain optimization algorithms, such as inventory optimization algorithms, route planning algorithms, and demand forecasting algorithms, as core decision support tools.

[0003] Traditional supply chain optimization systems typically output algorithmic calculations in static, predefined reports or charts, such as suggested inventory levels for each warehouse, optimal transportation routes for vehicles, or future demand forecasts. When enterprise users, such as supply chain planners or operations managers, are faced with these results, they often find it difficult to intuitively understand the underlying reasons why the algorithm made a particular decision. For example, for questions like "Why is it recommended to increase the inventory of a certain SKU by 500 units in the Shanghai warehouse?", existing systems cannot provide dynamic, interactive explanations. Users must manually consult large amounts of raw data, business documents, or directly consult algorithm engineers, resulting in low decision-making efficiency and requiring a high level of professional knowledge from the users.

[0004] Therefore, how to solve the technical problems of poor interpretability of algorithm results and poor human-computer interaction experience in existing supply chain optimization systems has become an urgent technical challenge in this field. Summary of the Invention

[0005] This invention provides a method and apparatus for visualizing and providing feedback on supply chain optimization results based on artificial intelligence interaction, in order to at least solve the problem of poor interpretability of algorithm results in existing supply chain optimization systems in related technologies.

[0006] According to an embodiment of the present invention, a method for visualizing and providing feedback on supply chain optimization results based on artificial intelligence interaction is provided, comprising:

[0007] Receive language information for at least one supply chain optimization result;

[0008] A structured query request is generated based on the language information, wherein the structured query request includes user intent identifiers, decision object entities, and visualization requirements;

[0009] Based on the structured query request, obtain the optimization scheme data corresponding to the decision object entity, and generate and display at least one visualization chart according to the visualization requirements;

[0010] Receive multimodal feedback information input by the user on the visualization chart, wherein the multimodal feedback information includes visual annotation information associated with chart elements in the visualization chart, and supplementary text information bound to the visual annotation information;

[0011] The multimodal feedback information is fused and analyzed to generate structured algorithm optimization instructions, wherein the algorithm optimization instructions include executable constraints or parameter adjustment items.

[0012] The algorithm is used to optimize the instructions and then recalculate to generate updated supply chain optimization results.

[0013] In one exemplary embodiment, generating a structured query request based on the language information includes:

[0014] The language information is preprocessed, including text cleaning, word segmentation, and stop word filtering;

[0015] The preprocessed language information is input into a pre-trained natural language processing model for joint parsing to output the user intent identifier and the decision object entity: wherein the user intent identifier is selected from a preset intent set, the intent set including at least one of cause explanation, scheme comparison, cost analysis, effect prediction, indicator query and constraint supplementation; the decision object entity includes the core business entity in the language information.

[0016] In one exemplary embodiment, receiving multimodal feedback information input by the user on the visualization chart includes:

[0017] Listen for user annotation operations on the visual chart to capture the visual annotation information, wherein the visual annotation information includes the identifier of the annotated chart element and the identifier of the business entity associated with the chart element;

[0018] After the annotation operation is completed, activate the text input interface near the chart element to receive the supplementary text information;

[0019] The visual annotation information and the text supplementary information are encapsulated into a unified multimodal feedback data structure, wherein the business entity identifier is used to eliminate semantic ambiguity in the text supplementary information.

[0020] In an exemplary embodiment, the step of fusing and parsing the multimodal feedback information to generate structured algorithm optimization instructions includes:

[0021] The business entity identifier in the multimodal feedback information is used as a forced context and input together with the supplementary text information into a preset language model;

[0022] Using the language model and based on the preset algorithm knowledge base metadata, key field values ​​are extracted from the supplementary text information to generate structured optimization suggestions.

[0023] In one exemplary embodiment, the generation of structured algorithm optimization instructions further includes:

[0024] Based on the algorithm type in the structured optimization suggestions, the corresponding algorithm function template is queried from the algorithm knowledge base;

[0025] The key field values ​​are mapped and populated into the algorithm function template to generate the structured algorithm optimization instructions;

[0026] Before mapping and filling, the data type and value range of the key field value are verified according to the verification rules of the key field value defined in the algorithm knowledge base.

[0027] In one exemplary embodiment, after recalculating based on the algorithm optimization instructions to generate updated supply chain optimization results, the method further includes:

[0028] Receive the updated supply chain optimization results, which include updated core indicator values ​​and parameter change records;

[0029] The updated supply chain optimization results are pushed to the user terminal through a real-time communication protocol;

[0030] On the user terminal, a comparison view before and after optimization is generated, showing the changes in the core indicator values.

[0031] According to another embodiment of the present invention, a supply chain optimization result visualization and feedback device based on artificial intelligence interaction is provided, comprising:

[0032] The language information acquisition module is used to receive language information related to at least one supply chain optimization result;

[0033] The request generation module is used to generate a structured query request based on the language information, wherein the structured query request includes user intent identifiers, decision object entities, and visualization requirements;

[0034] The chart generation module is used to obtain optimization scheme data corresponding to the decision object entity based on the structured query request, and generate and display at least one visualization chart according to the visualization requirements.

[0035] The feedback collection module is used to receive multimodal feedback information input by the user on the visualization chart, wherein the multimodal feedback information includes visual annotation information associated with chart elements in the visualization chart, and supplementary text information bound to the visual annotation information;

[0036] The parsing module is used to perform fusion parsing on the multimodal feedback information to generate structured algorithm optimization instructions, wherein the algorithm optimization instructions include executable constraints or parameter adjustment items;

[0037] The optimization module is used to recalculate based on the algorithm optimization instructions to generate updated supply chain optimization results.

[0038] In one exemplary embodiment, generating a structured query request based on the language information includes:

[0039] The language information is preprocessed, including text cleaning, word segmentation, and stop word filtering;

[0040] The preprocessed language information is input into a pre-trained natural language processing model for joint parsing to output the user intent identifier and the decision object entity; wherein, the user intent identifier is selected from a preset intent set, the intent set including at least one of reason explanation, scheme comparison, cost analysis, effect prediction, indicator query and constraint supplementation; the decision object entity includes the core business entity in the language information.

[0041] According to yet another embodiment of the present invention, a computer-readable storage medium is also provided, wherein a computer program is stored therein, wherein the computer program is configured to perform the steps in any of the above method embodiments when executed.

[0042] According to yet another embodiment of the present invention, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0043] This invention uses a natural language model to parse natural language questions, dynamically generate structured queries and visualizations, and achieves intelligent interpretation of algorithm results. Therefore, it can solve the problem of poor interpretability of algorithm results in existing supply chain optimization systems, thereby improving the interpretability of algorithm results and enhancing the user interaction experience. Attached Figure Description

[0044] Figure 1 This is a flowchart of a method for visualizing and providing feedback on supply chain optimization results based on artificial intelligence interaction according to an embodiment of the present invention;

[0045] Figure 2 This is a structural block diagram of a supply chain optimization result visualization and feedback device based on artificial intelligence interaction according to an embodiment of the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of this application will be described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0047] In the following description, the terms "first," "second," etc., 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 with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0048] Furthermore, in this application, directional terms such as "upper," "lower," "left," and "right" may be defined relative to the orientation of the components shown in the accompanying drawings. It should be understood that these directional terms can be relative concepts, used for relative description and clarification, and may change accordingly depending on the orientation of the components in the accompanying drawings.

[0049] In this application, unless otherwise expressly specified and limited, the term "connection" should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral part; it can be a direct connection or an indirect connection through an intermediate medium. Furthermore, the term "coupled" can refer to a method of electrical connection for signal transmission.

[0050] As used herein, “about,” “approximately,” or “approximately” includes the stated value and the average value within an acceptable range of deviation from the given value, wherein the acceptable range of deviation is determined by a person skilled in the art taking into account the measurement under discussion and the error associated with the measurement of the given quantity (i.e., the limitations of the measurement system).

[0051] This embodiment solves the technical problems of poor interpretability of supply chain optimization results, rigid human-computer interaction, and disconnect between feedback and optimization in the prior art by constructing a closed-loop system that integrates natural language understanding, dynamic visualization, multimodal feedback, intelligent parsing, and automatic algorithm iteration. It achieves the beneficial effects of improving decision-making efficiency, enhancing system adaptability, and realizing continuous algorithm optimization.

[0052] Specifically, the AI-based supply chain optimization result visualization and feedback method provided in this embodiment is executed in a layered architecture consisting of a front-end display layer, an AI service layer, an algorithm optimization layer, and a data storage layer. The layers interact with each other through standardized application programming interfaces (APIs) and real-time communication protocols.

[0053] like Figure 1 As shown, the method includes the following steps:

[0054] The technical solution of this application will be described in detail below.

[0055] S100: Receive natural language questions (i.e., language information) for at least one supply chain optimization result.

[0056] In this embodiment, the system provides a user interface through its front-end presentation layer, which includes a natural language input interface, such as a text input box. When reviewing the initial optimization results (e.g., inventory levels of each warehouse, transportation plans, etc.) generated by the underlying supply chain optimization algorithm, users (such as supply chain planners) can input relevant supply chain questions into this interface in a natural, conversational manner. The corresponding supply chain optimization results are pre-calculated by one or more algorithm modules (such as inventory optimization algorithms and route planning algorithms) and stored in the data storage layer.

[0057] The core of this step is to capture users' exploratory and interpretive needs for complex data results. These needs are often difficult to meet through traditional interfaces based on fixed filtering and sorting functions. Natural language questions take many forms.

[0058] For example, users might enter the following types of natural language questions:

[0059] Explanation of reasons: "Why is it recommended that the Shanghai warehouse add 500 units of safety stock for SKU-123?"

[0060] Comparison of plans: "Comparing the current optimized plan with the plan from last month, what are the differences in transportation costs and delivery time in East China?"

[0061] Cost analysis question: "What specific components make up the total inventory cost of Option C?"

[0062] Effect prediction question: "If the safety stock level in the Guangzhou warehouse is increased by 20%, how will the stockout rate change in the next 3 months?"

[0063] Constraint Supplement Intent Class: "Please add a constraint: the daily outbound volume of the Suzhou warehouse cannot exceed 2000 pieces."

[0064] The front-end presentation layer's issue receiving component is configured as a JavaScript module running in the browser environment. This module monitors the status of the natural language input interface in real time. When user input is detected as complete (e.g., the user presses the Enter key or clicks the "Submit" button), the component captures the complete, unprocessed raw text string. After capture, the component encapsulates this raw text string, along with necessary contextual information (e.g., the current user's identifier `user_id`, session identifier `session_id`, and the currently viewed optimization scheme identifier `scheme_id`), into a data packet. This data packet is then sent to the back-end AI service layer via a secure Hypertext Transfer Protocol (HTTPS) request for further parsing and processing.

[0065] To ensure the fluency and accuracy of user input, the natural language input interface can also integrate auxiliary functions. For example, it can provide input history lookup, allowing users to quickly re-initiate previous questions; or it can integrate input suggestions or auto-completion functions. This function can dynamically recommend relevant words during user input based on a pre-built dictionary (which contains high-frequency supply chain terms such as "safety stock," "in-transit inventory," "out-of-stock rate," and "inventory turnover days"), thereby reducing user input errors and guiding the way users ask questions, making them easier for the backend model to understand. For example, when a user inputs "inventory," the system can automatically suggest options such as "inventory turnover days," "safety stock level," and "inventory cost," thereby lowering the threshold for human-computer interaction and enabling even users who are not fully familiar with supply chain terminology to interact with the system efficiently.

[0066] S200: Perform intent recognition and entity extraction on language information to generate a structured query request, wherein the structured query request includes user intent identifiers, decision object entities, and visualization requirements.

[0067] In this embodiment, after the AI ​​service layer receives raw language information from the front end, it initiates a Natural Language Processing (NLP) process to convert unstructured text input into structured instructions that the machine can accurately understand and execute. This process is a crucial bridge connecting human intent and system functionality, and its accuracy directly determines the effectiveness of all subsequent steps. This process is executed by an AI intelligent parsing module, which includes a preprocessing unit and an NLP model fine-tuned with supply chain data. The processing includes:

[0068] S210: Preprocess the language information.

[0069] Preprocessing is to purify and standardize the input text to provide better input data for the core model. This process usually includes multiple sub-steps: First is text cleaning to remove characters irrelevant to semantics in the original text, such as extra spaces, line breaks, special symbols, or HTML tags (if the input source is rich text); then tokenization is performed, that is, splitting continuous sentences into independent words or morphemes. For Chinese, specialized tokenization algorithms are needed, such as tokenizers based on Hidden Markov Model (HMM) or Conditional Random Field (CRF), or using mature tokenization libraries like Jieba; among them, a custom dictionary in the supply chain domain can be loaded during tokenization to ensure that domain-specific terms (such as "safety stock", "in-transit inventory", "regional distribution center") are correctly recognized as a whole instead of being wrongly split. Finally is stop word filtering, that is, removing words that frequently appear in the text but contribute little to semantics, such as "of", "is", "a", "how", etc. These words will increase the computational burden and may interfere with the model's judgment.

[0070] Exemplarily, for the original question "Why is it recommended to increase 500 pieces of safety stock for SKU-123 in the Shanghai warehouse?", after preprocessing, a word sequence may be obtained: ['Why', 'is recommended', 'Shanghai warehouse', 'SKU-123', 'increase', '500 pieces','safety stock']. Among them, "Shanghai warehouse", "SKU-123", and "safety stock" are correctly recognized as domain terms.

[0071] S220: Use the NLP model for joint parsing and output the user intention identifier and decision object entity.

[0072] The preprocessed word sequence is input into the NLP model, which adopts a pre-trained language model based on the Transformer architecture (such as BERT (Bidirectional Encoder Representations from Transformers) or its variants). After being pre-trained on a large amount of general corpus, it is then fine-tuned using a specially constructed and manually annotated supply chain domain dataset to enable the model to learn two joint tasks: intent classification and slot filling, that is, entity extraction.

[0073] When classifying intents, the model outputs the label with the highest probability from a preset intent set as the user intent identifier. This preset intent set is predefined based on the supply chain analysis scenario. In this embodiment, the set includes: reason_explain, scenario_compare, cost_analysis, effect_predict, indicator_detail, and constraint_add.

[0074] The task of entity extraction is to identify and extract words containing specific information from sentences and classify them into predefined entity types. These entities constitute the decision object entities. In this embodiment, the predefined entity types are divided into two main categories:

[0075] Decision object: Directly refers to the operation or analysis object in the supply chain. Examples include: inventory (such as "safety stock" or "in-transit stock"), transport route (such as "route from Beijing warehouse to Hangzhou RDC"), cost item (such as "storage cost" or "transportation cost"), and order (such as "e-commerce retail order").

[0076] Dimension entities are used to define the scope of the decision-making object. Examples include: region (geographical area entity, such as "East China" or "Yangtze River Delta"), and time cycle (time period entity, such as "monthly" or "Q3 2024"). In addition, there are more specific metric entities and constraint entities, such as stockout rate, warehouse (warehouse, such as "Shanghai warehouse"), and SKU (material unit, such as "SKU-123").

[0077] The model predicts which entity type a word belongs to by attaching a classifier to each word (using the BIO labeling system, i.e., B-Beginning, I-Inside, O-Outside).

[0078] For example, for the input sequence ['Why', 'Recommendation', 'Shanghai Warehouse', 'SKU-123', 'Increase', '500 Items', 'Safety Stock'], the joint parsing output of the NLP model might be:

[0079] User intent identifier: reason_explain

[0080] Decision-making entity:

[0081] type: inventory

[0082] details: {warehouse: 'Shanghai warehouse', sku: 'SKU-123', inventory_type: 'safety stock', adjustment: '+500 pieces'}

[0083] S230: Generate a structured query request.

[0084] The AI-powered intelligent parsing module integrates the output of the NLP model into a standardized, structured query request. This request typically uses JSON (JavaScript Object Notation) format due to its good readability and machine parsing capabilities. This JSON object must contain several core fields: the `intent` field stores the user's intent identifier, and the `decision_object` field is an object storing the decision object entity and its detailed attributes. Furthermore, to drive subsequent visualization, the system intelligently infers the user's visualization needs based on the parsing results. For example, for the `reason_explain` intent, the system might default to recommending a "bar chart - scheme indicator comparison" to compare key indicators before and after adjustments, and a "line chart - out-of-stock rate time trend" to predict future changes. For the `scenario_compare` intent, it might recommend a "dual-axis line chart" or a "radar chart."

[0085] The resulting structured query request JSON object not only contains all the user's explicit information but also includes best practice suggestions inferred by the system based on experience. This structured data packet will serve as the unified input for all subsequent modules, ensuring the accuracy and consistency of information transmission.

[0086] For example, in response to the aforementioned user questions, the final generated complete structured query request might look like this:

[0087]

[0088]

[0089] The confidence field represents the AI ​​model's confidence in the parsing results, which can be used to determine whether human intervention is needed; the JSON object will then be passed to the dynamic visualization generation module.

[0090] S300: In response to a structured query request, retrieve the optimization scheme data corresponding to the decision object entity, and generate and display at least one visualization chart according to visualization requirements.

[0091] In this embodiment, the abstract structured query request is transformed into visual information that users can intuitively perceive. Specifically, the structured query request is received as input, and after three sub-processes—data acquisition, data processing, and chart rendering—a dynamic and interactive visual chart is finally presented on the front-end interface. The specific steps are as follows:

[0092] S310: Obtain optimization scheme data.

[0093] First, the input structured query request is parsed, and fields such as decision_object, comparison_baseline, metrics, and dimensions are extracted to form the core parameters for data query. Then, based on the extracted fields, one or more data query requests are sent to the backend business service layer. After receiving the request, the business service layer will access the data storage layer (e.g., a system consisting of a PostgreSQL database and a Redis cache) to retrieve the required data.

[0094] For example, the data query logic for the aforementioned request to "explain the reason for the increase of 500 units in the inventory of SKU-123 in the Shanghai warehouse" is as follows:

[0095] Query current scheme data: Based on the scheme_id of the current session, and the warehouse: 'Shanghai warehouse' and sku: 'SKU-123' in decision_object, query all core metrics of this SKU in the current scheme, such as stockout_rate and inventory_turnover_days.

[0096] Query and compare baseline data: Based on comparison_baseline: 'original plan', query the corresponding core indicator values ​​of this SKU in the "original plan" before this inventory adjustment.

[0097] Query trend forecast data: The system may also call a lightweight forecast model to simulate the monthly trend of stockout rate changes over a future time period (such as 2024-Q3) before and after inventory adjustments; all of the data retrieved are usually raw and tabular, and will be aggregated and returned by the business service layer.

[0098] S320: Data processing and conversion.

[0099] After receiving the raw data, it needs to be cleaned, transformed, and aggregated according to the requirements of the `visualization_needs` field to adapt it to the rendering format of a specific chart. For example, for a "bar chart - comparison of plan indicators," the indicator values ​​(such as stockout rate and turnover days) of the current and original plans need to be integrated into a dataset format suitable for the bar series in ECharts. For a "line chart - stockout rate time trend," the forecast data before and after adjustment need to be organized into two independent series data suitable for the line series by month. This process ensures that any structured data returned by the backend can be correctly parsed and drawn by the frontend rendering engine.

[0100] S330: Generate and display visualization charts.

[0101] Once the data is ready, the dynamic visualization generation module calls the front-end chart rendering engine (such as Apache ECharts or D3.js) to dynamically generate the visualization chart. At this time, according to the chart type in visualization_needs (such as "bar chart" or "line chart"), the corresponding chart configuration item (option) is generated. This configuration item is a complex JavaScript object that defines in detail all visual and functional attributes of the chart, such as title, axes, legend, data series, colors, and interactive behaviors.

[0102] After the configuration options are generated, the module calls the rendering engine's `setOption` method to apply the configuration options to a reserved chart container (DOM element) in the front-end interface. The rendering engine then instantly draws the required visualization charts in the browser. These charts are not static images but are highly interactive. Users can hover their mouse over a chart element (such as a bar or data point) to view detailed values, click on the legend to show or hide specific data series, and perform operations such as data zooming or drill-down to further explore the data from different dimensions and granularities.

[0103] The above steps achieve a seamless transition from a natural language problem to a dynamic interactive chart. Users no longer need to face cold, impersonal tables of numbers, but can quickly understand the logic and potential impact behind algorithmic decisions through intuitive visual comparisons and trend changes, greatly reducing cognitive load and improving decision-making efficiency.

[0104] S400: Receives multimodal feedback information input by the user on a visualization chart, wherein the multimodal feedback information includes visual annotation information associated with chart elements in the visualization chart, and supplementary text information bound to the visual annotation information.

[0105] In this embodiment, after users understand the algorithm results through the generated visualizations, they may find that the results deviate from the actual business situation, or that new business constraints need to be added. At this time, users can provide feedback directly at the "site" where they have questions (i.e., on the visualization). Specifically:

[0106] S410: Capture visual annotation information.

[0107] By integrating an event listener system with the front-end chart rendering engine, users can directly perform visual annotation operations on the chart.

[0108] For map-based charts (such as route planning maps and delivery area maps): listen for events such as mapselectchanged or draw:created from the chart library; users can use tools provided in the toolbar to select points on the map (e.g., select a delivery center), circle an area (e.g., mark a traffic congestion area), or draw a line (e.g., indicate an unusable road segment); when the user completes the operation, the event listener will be triggered; the handling function will capture two core pieces of information from the event object: first, visual features, such as the set of geographic coordinate vertices of the circled area: coordinates: [{lat: 31.23, lng: 121.47},...]; second, the business entity identifier related_entity_id associated with this visual feature. This ID is obtained by querying the underlying business data bound during chart rendering; for example, the ID corresponding to the delivery center selected by the user in the layer data is dc-005. This association can bind a pure geometric shape (visual label) to a clear and unique business entity (delivery center dc-005), solving the semantic ambiguity that may exist in simple coordinates.

[0109] For indicator-based charts (such as bar charts and line charts): listen for the `click` or `brushselected` event. Users can click on a specific element in the chart, such as a bar representing transportation costs in June 2024. The event handler will parse all the dimensional information corresponding to that element from the callback parameters (params), such as `{time: "2024-06", metric: "transport_cost", scenario: "Scenario A"}`, and its unique identifier in the chart: `chart_element_id: "bar-202406"`.

[0110] S420: Receive and bind supplementary text information.

[0111] Once the visual annotation is completed, the system will immediately trigger an interactive change: a text input box will dynamically pop up next to the user's annotation location to achieve a "strong binding" between visual and text information.

[0112] Interaction Lock: The text input field is locked to the previously completed annotation. The system prevents the user from performing new annotations until they finish entering text or actively close the input field. This ensures that each text feedback clearly corresponds to a visual annotation.

[0113] Data binding: During the user's text input process, the captured visual annotation information (especially the key related_entity_id: "dc-005") will be bound as a hidden field in real time to the same memory data object as the user's text input content.

[0114] Real-time preview: To provide a better user experience, the system can display marked areas / elements on the map or next to charts in real time, along with a draft of the text the user is typing, allowing the user to confirm and modify it at any time, ensuring the accuracy of the feedback.

[0115] S430: Encapsulated as a multimodal feedback data structure.

[0116] After the user completes the text input and submits it, the user feedback collection module encapsulates the visual information, text information, and other contextual information (such as user ID, timestamp, and scheme ID) into a unified, structured multimodal feedback JSON object. This object is designed to be both machine-readable and human-readable.

[0117] For example, if a user selects "Shanghai Pudong Distribution Center" on a map and enters the text "This distribution center is closed and cannot receive any inbound orders before July 1st. Inventory allocation in surrounding warehouses needs to be adjusted," the system will generate the following multimodal feedback JSON object:

[0118]

[0119] This structured data object, through the `related_business_entity` field, establishes an unambiguous and strong association between the visual annotation "where" (distribution center dc-005) and the textual supplement "what problem" (closed, unable to be entered into the database). It contains both the core ID required for machine processing and the name and coordinates needed for front-end display, making it ideal input for subsequent intelligent parsing. This data object will be instantly sent to the backend AI service layer via real-time communication protocols such as WebSocket.

[0120] S500: Fusion and analysis of multimodal feedback information to generate structured algorithm optimization instructions, which include executable constraints or parameter adjustment terms for the algorithm.

[0121] In this embodiment, the aforementioned collected multimodal feedback containing human business knowledge is converted into precise, structured instructions that the underlying optimization algorithm can directly understand and execute. This process is completed collaboratively by the feedback analysis module in the backend AI service layer and an algorithm instruction conversion service (middle layer). The specific process is as follows:

[0122] S510: AI Fusion Analysis.

[0123] After receiving the multimodal feedback JSON object, the AI ​​service layer's feedback analysis module first performs data validation and preprocessing. It first checks whether visual_info.related_business_entity.entity_id (e.g., dc-005) exists in the business database to exclude invalid annotations. After the validation passes, it extracts the context of this validated business entity (i.e., "distribution center dc-005") as a mandatory constraint for subsequent NLP parsing.

[0124] Next, this mandatory context and the text information in text_info.content are input into a large language model (LLM) (such as GPT-4). The model is guided by a pre-defined prompt engineering to associate pronouns (such as "here") in the text with the mandatory context when understanding the text.

[0125] For example, a Prompt might be designed as follows:

[0126] Task: Analyze user feedback on supply chain optimization solutions to identify core issues and optimization suggestions.

[0127] Given the visual context: The user marked "Shanghai Pudong Distribution Center (ID: dc-005)" on the map.

[0128] User text feedback: "The distribution center here is closed and cannot receive any inbound orders before July 1st. Inventory allocation in surrounding warehouses needs to be adjusted."

[0129] Please output the analysis results in JSON format, including:

[0130] 1. core_issue: Description of the core issue.

[0131] 2. constraint_to_add: The business constraints that need to be added. The format must follow the definition of the algorithm knowledge base.

[0132] 3. optimization_suggestion: Algorithm optimization suggestions.

[0133] Based on this prompt, the large language model can accurately parse "here" as "distribution center dc-005" and output structured analysis results, namely "structured optimization suggestions".

[0134] The model outputs (structured optimization suggestions) at this point:

[0135]

[0136] This output is very close to the executable instructions of the algorithm, but its constraint_content is still a natural language description and needs to be further converted into precise algorithm parameters.

[0137] S520: Instruction conversion based on algorithm knowledge base.

[0138] To address the challenge of translating AI-suggested "natural language" constraints into algorithmic "code" instructions, the system utilizes an algorithm instruction conversion service. This service relies on a pre-built algorithm knowledge base, stored in database tables. This knowledge base defines all acceptable constraints and parameters for underlying optimization algorithms (such as inventory optimization and path planning). Each record defines a constraint or parameter, specifically including the following core fields:

[0139] constraint_id / param_id: Unique identifiers, such as route_time_restriction.

[0140] algorithm_type: The type of algorithm to apply, such as path_planning.

[0141] required_fields: A list of fields required to perform the constraint or parameter adjustment. Each field contains a name, data type, and validation rules (such as regular expressions).

[0142] allowed_values: Allowed values ​​for the field, such as enumerations or numeric ranges.

[0143] field_desc: The business meaning of the field.

[0144] The conversion process includes:

[0145] Knowledge base matching: The instruction conversion service receives optimization suggestions and queries the algorithm knowledge base for matching records based on constraint_to_add.constraint_type (such as distribution_center_status).

[0146] Field Extraction and Mapping: The service reads the required_fields of the record. Then, using lightweight NLP techniques (such as regular expressions and keyword matching) and a rule engine, it extracts the field values ​​required by required_fields from constraint_to_add.constraint_content (such as "dc-005 is prohibited from being added to the database from June 10, 2024 to July 1, 2024"). For example, it extracts start_date: '2024-06-10', end_date: '2024-07-01', and status: 'no_inbound'.

[0147] Validation and formatting: The service performs format validation and conversion on the extracted values ​​according to the validation rules defined in the knowledge base (for example, converting dates to ISO 8601 format).

[0148] Instruction generation: Based on the "algorithm function template" associated with the knowledge base, the service fills all the extracted and validated field values ​​into a predefined JSON or Protobuf structure to generate a fully structured algorithm optimization instruction.

[0149] For example, based on the aforementioned AI optimization suggestions, the final generated algorithm optimization instructions are as follows:

[0150]

[0151] This JSON object is an unambiguous machine instruction that contains all the precise information needed for the algorithm to execute, and it is associated with the initial user feedback via feedback_id, ensuring traceability throughout the entire process.

[0152] S600: Transmits the algorithm optimization instruction to one or more supply chain optimization modules to trigger one or more supply chain optimization modules to recalculate based on the algorithm optimization instruction, thereby generating an updated supply chain optimization result.

[0153] In this embodiment, after generating structured algorithm optimization instructions, the algorithm instruction conversion service transmits the instructions to the corresponding algorithm optimization layer by calling a standard API interface. Specifically:

[0154] S610: Command reception and verification.

[0155] The algorithm optimization layer exposes a series of API endpoints, such as ` / api / algorithm / execute`. Upon receiving an algorithm optimization command, the layer first performs a compliance check. At this point, it calls the algorithm knowledge base interface again to verify the validity of the parameter combination in the command and whether the values ​​are within a safe range. For example, if the command attempts to adjust the safety stock of a SKU to a value exceeding the maximum warehouse capacity, the verification will fail, and an error message will be returned to the caller, thus preventing invalid or dangerous parameters from being executed.

[0156] S620: Algorithm parameter / constraint update.

[0157] After successful verification, the corresponding operation is executed according to the command_type of the instruction.

[0158] For parameter adjustment (adjust_param): the algorithm will directly update the parameter value of the corresponding entity in its internal "parameter configuration pool"; for example, update the safety_stock parameter of SKU-123 in warehouse WH-001 from 600 to 800. This parameter pool is usually composed of a memory cache (such as Redis) and a persistent database (such as PostgreSQL) to ensure high performance and data consistency.

[0159] For constraint updates (add_constraint): the algorithm updates its internal "constraint matrix"; for example, for a path planning algorithm, this might be a multidimensional array representing road network connectivity, and the instruction would set the weight of the A->B segment for truck traffic to be infinite during nighttime. For an inventory optimization algorithm, this might be marking dc-005's inbound limit as 0 within a specified time period in the "warehouse-time-inbound / outbound quantity" constraint matrix.

[0160] S630: Trigger recalculation.

[0161] After the parameters or constraints are updated, the system will automatically trigger the "recalculation" interface of the corresponding algorithm. At this time, the system will use its original core calculation logic (such as the Economic Order Quantity (EOQ) calculation function and the Dijkstra path search function), but the updated input conditions will be used. Since only the input conditions have changed locally, the system can use a variety of strategies to control the calculation efficiency, such as caching historical basic data and parallel computing of non-core dimensions, to ensure that the time taken for a single recalculation is controlled within an acceptable range (e.g., within 3 seconds).

[0162] S660: Generates and returns the updated optimization results.

[0163] After the algorithm completes the recalculation, it generates a completely new set of optimization data. This data not only includes new decision results (such as new inventory levels and new routes), but must also contain metadata for comparative analysis. Its output format is typically a JSON object, which contains:

[0164] scheme_id: A unique ID for the new scheme, such as scheme-125.

[0165] optimized_params / applied_constraints: Parameters that were changed or constraints that were applied in this calculation.

[0166] indicator_change: Changes in core indicators, such as {"stockout_rate": "5% -> 2.5%", "inventory_cost": "120,000 -> 135,000"}.

[0167] feedback_id: The original feedback ID that triggered this calculation, fb-20240610-001.

[0168] After generating this result, the algorithm optimization layer will push it back to the business service layer via an internal message queue or API callback. Upon receiving the result, the business service layer will, on the one hand, persist the full data of the new solution to the data storage layer, and on the other hand, proactively and in real time push the simplified result, including the core change (indicator_change), back to the user's front-end interface that initiated this feedback via a WebSocket long connection.

[0169] Simultaneously, the system updates the processing status of the feedback. For example, it updates the status of feedback item with ID fb-20240610-001 from "Optimizing" to "Completed" and generates a clickable link pointing to a detailed comparison view of the old and new solutions. At this point, a complete closed-loop process from unstructured user feedback to automated algorithm updates is completed. Users can see the quantifiable impact of their business feedback within seconds, achieving true "what you say is what you get" continuous optimization.

[0170] In summary, the method and system provided in this application, by deeply integrating artificial intelligence technology with supply chain optimization processes, construct a complete intelligent interaction and feedback closed loop. It not only solves the shortcomings of traditional systems in terms of result interpretability, but more importantly, it establishes a channel that allows business knowledge to be automatically injected into and optimize algorithm models at low cost and high efficiency. This enables the supply chain optimization system to evolve from a static computing tool into an intelligent decision-making tool that can co-evolve with business experts and continuously adapt to business changes.

[0171] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0172] This embodiment also provides a visualization and feedback device for supply chain optimization results based on artificial intelligence interaction. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0173] Figure 2 This is a structural block diagram of a supply chain optimization result visualization and feedback device based on artificial intelligence interaction according to an embodiment of the present invention, such as... Figure 2 As shown, the device includes:

[0174] Language information acquisition module 21 is used to receive language information for at least one supply chain optimization result;

[0175] The request generation module 22 is used to generate a structured query request based on the language information, wherein the structured query request includes user intent identifiers, decision object entities, and visualization requirements;

[0176] The chart generation module 23 is used to obtain optimization scheme data corresponding to the decision object entity based on the structured query request, and generate and display at least one visualization chart according to the visualization requirements.

[0177] The feedback collection module 24 is used to receive multimodal feedback information input by the user on the visualization chart, wherein the multimodal feedback information includes visual annotation information associated with chart elements in the visualization chart, and text supplementary information bound to the visual annotation information;

[0178] The parsing module 25 is used to perform fusion parsing on the multimodal feedback information to generate structured algorithm optimization instructions, wherein the algorithm optimization instructions include executable constraints or parameter adjustment items of the algorithm;

[0179] The optimization module 26 is used to recalculate based on the algorithm optimization instructions to generate updated supply chain optimization results.

[0180] In an optional embodiment, generating a structured query request based on the language information includes:

[0181] The language information is preprocessed, including text cleaning, word segmentation, and stop word filtering;

[0182] The preprocessed language information is input into a pre-trained natural language processing model for joint parsing to output the user intent identifier and the decision object entity; wherein, the user intent identifier is selected from a preset intent set, the intent set including at least one of reason explanation, scheme comparison, cost analysis, effect prediction, indicator query and constraint supplementation; the decision object entity includes the core business entity in the language information.

[0183] It should be noted that the above modules can be implemented by software or hardware. For the latter, they can be implemented in the following ways, but are not limited to: all the above modules are located in the same processor; or, the above modules are located in different processors in any combination.

[0184] Embodiments of the present invention also provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to perform the steps in any of the above method embodiments when executed.

[0185] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0186] Embodiments of the present invention also provide an electronic device including a memory and a processor, the memory storing a computer program and the processor being configured to run the computer program to perform the steps in any of the above method embodiments.

[0187] In one exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.

[0188] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0189] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another apparatus, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0190] The units described as separate components may or may not be physically separate. A component shown as a unit can be one or more physical units; that is, it can be located in one place or distributed in multiple different locations. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0191] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0192] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0193] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for visualizing and providing feedback on supply chain optimization results based on artificial intelligence interaction, characterized in that, include: Receive language information for at least one supply chain optimization result; A structured query request is generated based on the language information, wherein the structured query request includes user intent identifiers, decision object entities, and visualization requirements; Based on the structured query request, obtain the optimization scheme data corresponding to the decision object entity, and generate and display at least one visualization chart according to the visualization requirements; Receive multimodal feedback information input by the user on the visualization chart, wherein the multimodal feedback information includes visual annotation information associated with chart elements in the visualization chart, and supplementary text information bound to the visual annotation information; The multimodal feedback information is fused and analyzed to generate structured algorithm optimization instructions, wherein the algorithm optimization instructions include executable constraints or parameter adjustment items. The algorithm is used to optimize the instructions and then recalculate to generate updated supply chain optimization results.

2. The method according to claim 1, characterized in that, The step of generating a structured query request based on the language information includes: The language information is preprocessed, including text cleaning, word segmentation, and stop word filtering; The preprocessed language information is input into a pre-trained natural language processing model for joint parsing to output the user intent identifier and the decision object entity; wherein, the user intent identifier is selected from a preset intent set, the intent set including at least one of reason explanation, scheme comparison, cost analysis, effect prediction, indicator query and constraint supplementation; the decision object entity includes the core business entity in the language information.

3. The method according to claim 1, characterized in that, The receiving of multimodal feedback information input by the user on the visualized chart includes: Listen for user annotation operations on the visual chart to capture the visual annotation information, wherein the visual annotation information includes the identifier of the annotated chart element and the identifier of the business entity associated with the chart element; After the annotation operation is completed, activate the text input interface near the chart element to receive the supplementary text information; The visual annotation information and the text supplementary information are encapsulated into a unified multimodal feedback data structure, wherein the business entity identifier is used to eliminate semantic ambiguity in the text supplementary information.

4. The method according to any one of claims 1-3, characterized in that, The step of fusing and parsing the multimodal feedback information to generate structured algorithm optimization instructions includes: The business entity identifier in the multimodal feedback information is used as a forced context and input together with the supplementary text information into a preset language model; Using the language model and based on the preset algorithm knowledge base metadata, key field values ​​are extracted from the supplementary text information to generate structured optimization suggestions.

5. The method according to claim 4, characterized in that, The algorithm optimization instructions for generating structured data also include: Based on the algorithm type in the structured optimization suggestions, the corresponding algorithm function template is queried from the algorithm knowledge base; The key field values ​​are mapped and populated into the algorithm function template to generate the structured algorithm optimization instructions; Before mapping and filling, the data type and value range of the key field value are verified according to the verification rules of the key field value defined in the algorithm knowledge base.

6. The method according to claim 1, characterized in that, After recalculating based on the algorithm optimization instructions to generate updated supply chain optimization results, the method further includes: Receive the updated supply chain optimization results, which include updated core indicator values ​​and parameter change records; The updated supply chain optimization results are pushed to the user terminal through a real-time communication protocol; On the user terminal, a comparison view before and after optimization is generated, showing the changes in the core indicator values.

7. A visualization and feedback device for supply chain optimization results based on artificial intelligence interaction, characterized in that, include: The language information acquisition module is used to receive language information related to at least one supply chain optimization result; The request generation module is used to generate a structured query request based on the language information, wherein the structured query request includes user intent identifiers, decision object entities, and visualization requirements; The chart generation module is used to obtain optimization scheme data corresponding to the decision object entity based on the structured query request, and generate and display at least one visualization chart according to the visualization requirements. The feedback collection module is used to receive multimodal feedback information input by the user on the visualization chart, wherein the multimodal feedback information includes visual annotation information associated with chart elements in the visualization chart, and supplementary text information bound to the visual annotation information; The parsing module is used to perform fusion parsing on the multimodal feedback information to generate structured algorithm optimization instructions, wherein the algorithm optimization instructions include executable constraints or parameter adjustment items; The optimization module is used to recalculate based on the algorithm optimization instructions to generate updated supply chain optimization results.

8. The apparatus according to claim 7, characterized in that, The step of generating a structured query request based on the language information includes: The language information is preprocessed, including text cleaning, word segmentation, and stop word filtering; The preprocessed language information is input into a pre-trained natural language processing model for joint parsing to output the user intent identifier and the decision object entity; wherein, the user intent identifier is selected from a preset intent set, the intent set including at least one of reason explanation, scheme comparison, cost analysis, effect prediction, indicator query and constraint supplementation; the decision object entity includes the core business entity in the language information.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program is configured to perform the method described in any one of claims 1 to 6 when executed.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the method as described in any one of claims 1 to 6.

Images