Hotspot information tracking method of marine transportation supply chain and related equipment

Abstract

The invention discloses a hot spot information tracking method and related equipment for a sea transportation supply chain, and aims to accurately focus professional contents in the sea transportation field and accurately understand complex causal relationships by constructing a multi-dimensional feature knowledge base, establishing dynamic mapping between the multi-dimensional feature knowledge base and diversified data sources and driving a first language model in combination with a field specific abstract prompt template. The problems of inaccurate information screening and hot spot identification errors or omission are effectively solved. The second large language model is driven by means of a multi-dimensional classification prompt template, deep analysis is conducted on the information abstract from multiple dimensions such as event types and influence degrees, the limitation that in the prior art, the analysis dimension is single is broken through, and deep decision support is provided. The powerful understanding ability of the large language model is utilized to realize effective integration of multi-source heterogeneous data, and the data fusion dilemma is broken. And inputting a multi-dimensional analysis result, a freight rate index and other external quantitative indexes into a hotspot scoring model to realize timely and accurate identification of complex hotspots and risks.

Description

Technical Field

[0001] The present application relates to the technical field of maritime supply chain, and more specifically, to a method and related equipment for tracking hot spot information in a maritime supply chain. Background Art

[0002] In the field of maritime supply chain, accurate, in-depth and timely tracking of industry hotspots is crucial for supply chain management. Timely tracking of industry hotspots can not only help companies plan transportation routes in advance, allocate resources reasonably, and reduce operating costs, but also help companies respond to unexpected risks in a timely manner, such as port strikes and bad weather, effectively ensuring the stable operation of the supply chain and occupying an advantageous position in the fierce market competition.

[0003] However, existing technologies in this area suffer from numerous shortcomings. First, they often rely on general models or simple keyword systems for information processing, lacking industry-specific depth and precision. Difficulties in understanding the specialized terminology and complex causal relationships within supply chain shipping make it impossible to predict, for example, the delayed impact of a strike in one location on freight rates on a specific route. This ultimately leads to inaccurate information screening and misidentification or omission of hotspots. Second, existing technologies employ a single analytical dimension, focusing solely on surface-level information aggregation or simple categorization. They fail to assess events from multiple dimensions, such as event type, logistics links, and geographic scope, providing in-depth insights and meeting the analytical needs of businesses for complex market conditions. Third, faced with heterogeneous data from multiple sources, existing technologies struggle to effectively integrate and comprehensively assess it, failing to fully leverage the value of diverse data types. Fourth, existing technologies rely on a single information source or simple rules. This poor identification of complex hotspots or risks, which require the integration of multiple factors such as supply and demand changes, related events, and quantitative indicators, results in delayed and inaccurate early warnings, making it difficult for businesses to take timely countermeasures.

[0004] Therefore, a new method for tracking hot spot information in the maritime supply chain is urgently needed to solve the above problems and meet industry needs. Summary of the Invention

[0005] The present application provides a method and related equipment for tracking hot spot information in the maritime supply chain. By constructing a multi-dimensional feature knowledge base of the maritime supply chain and establishing a dynamic mapping with diversified data sources, the application combines domain-specific prompt templates to drive a large language model to extract and analyze information. The large language model is used to integrate multi-source heterogeneous data, and the multi-dimensional analysis results are combined with quantitative indicators to perform hot spot judgment. This achieves efficient fusion of multi-source data and accurate tracking and identification of hot spot information in the maritime supply chain.

[0006] A method for tracking hot spot information in a maritime supply chain, comprising:

[0007] Build a multi-dimensional feature knowledge base for the maritime supply chain and establish dynamic mapping relationships between the multi-dimensional feature knowledge base and diverse data sources, including text news, structured freight rate data, Automatic Vessel Identification System data, and meteorological data.

[0008] Based on the domain-specific summary prompt template, the first language model is driven to extract key logistics elements from the diverse data sources and generate information summaries;

[0009] The second language model is driven by a multi-dimensional classification prompt template to classify the information summary by event type, logistics link, geographical scope, impact level and novelty, and output a multi-dimensional analysis result;

[0010] The multi-dimensional analysis results and external quantitative indicators are input into the hotspot scoring model to obtain a normalized hotspot index and sort and filter it to generate a push hotspot set. The external quantitative indicators include freight rate index, ship waiting time and container availability index.

[0011] Optionally, the adjustment of the dynamic mapping association between the multi-dimensional feature knowledge base and the diversified data sources includes at least one of rule matching, model recommendation, and update frequency adjustment;

[0012] The rule matching method is based on a preset industry mapping rule library, matching the fields in the diversified data sources with the multi-dimensional features in the multi-dimensional feature knowledge library to generate a mapping relationship;

[0013] The model recommendation method is to calculate the semantic similarity between the data source and the knowledge base features through the model, generate a dynamic association weight matrix, and iteratively update the weight coefficient based on user feedback;

[0014] The update frequency adjustment method is to adjust the priority weight of the mapping association based on the update frequency of the diversified data source and the confidence score output by the data quality assessment model.

[0015] Optionally, the domain-specific summary prompt template includes a first type of template embedded in an industry causal trigger vocabulary, a second type of template embedded in freight rate data association, and a third type of template embedded in a spatiotemporal rule library of ship trajectories and a meteorological event keyword table;

[0016] The domain-specific summary prompt template drives the first language model to extract key logistics elements from the diverse data sources and generate information summaries, including:

[0017] For text data sources, the first language model is driven by the first template to extract event causal chains from the text and generate a text information summary containing objects, events, time, location, values, and causes.

[0018] For structured data sources, the first language model is driven based on the second type of template to semantically align freight rate index data and vessel position data with the multi-dimensional features in the multi-dimensional feature knowledge base, thereby generating a structured information summary containing the relationship between numerical indicators and industry characteristics;

[0019] For semi-structured data sources, the first language model is driven by the third type of template to analyze the probability of port congestion triggered by ship trajectory deviation thresholds or meteorological event keywords, and generate a semi-structured information summary containing spatiotemporal abnormal events and impact predictions;

[0020] The text information summary, the structured information summary, and the semi-structured information summary are integrated to generate an information summary.

[0021] Optionally, the multi-dimensional classification prompt template includes a causal chain reasoning template, a novelty evaluation template, and an impact degree classification template;

[0022] The multi-dimensional classification prompt template drives the second language model to classify the information summary by event type, logistics link, geographical scope, impact level and novelty, and output multi-dimensional analysis results, including:

[0023] The second language model is driven by the causal chain reasoning template to extract event timestamps and geographic tags from the information summary, and a model for the lagged impact of events on shipping rates or vessel waiting times is constructed by combining historical freight rate data and port operation records. The causal relationship weights are quantified based on a probabilistic causal graph to generate causal chain analysis results.

[0024] Driving the second language model to compare the semantic similarity between the current event and the historical events in the multi-dimensional feature knowledge base based on the novelty evaluation template, and determining a novelty tag according to the semantic similarity;

[0025] Based on the impact classification template, the second language model is driven to classify the impact level according to the fluctuation range of the freight rate index, the growth rate of the ship waiting time, and the decline rate of the container availability index;

[0026] The causal chain analysis results, the novelty mark and the impact level are integrated to output a multi-dimensional analysis result including event type, logistics link, geographical scope, impact degree and novelty.

[0027] Optionally, the training process of the hotspot scoring model includes:

[0028] Collect historical hot event data and corresponding sample external quantitative indicators;

[0029] Extract the event type, logistics link, geographical scope, impact degree and novelty classification from the sample multi-dimensional classification results as feature vectors, and combine them with the sample external quantitative indicators as input features;

[0030] A supervised learning algorithm is used to output the training model parameters with the manually labeled normalized hotspot index as the target, and the hyperparameters are adjusted through cross-validation to optimize the model generalization ability.

[0031] Optionally, also include:

[0032] Based on the user's click feedback, marking operation or manual correction data on the normalized hot spot index or the pushed hot spot set, the parameters of the dynamic mapping association, the field-specific summary prompt template, the multi-dimensional classification prompt template and the hot spot scoring model are iteratively optimized to form a closed-loop optimization mechanism.

[0033] A hot spot information tracking device for a maritime supply chain, comprising:

[0034] A mapping association unit is used to construct a multi-dimensional feature knowledge base for the maritime supply chain and establish dynamic mapping associations between the multi-dimensional feature knowledge base and diverse data sources, including text news, structured freight rate data, automatic vessel identification system data, and meteorological data.

[0035] An information summary unit is used to drive the first language model based on a domain-specific summary prompt template, extract key logistics elements from the diversified data sources, and generate an information summary;

[0036] a multi-dimensional analysis unit, configured to drive a second language model based on a multi-dimensional classification prompt template, classify the information summary by event type, logistics link, geographical scope, impact degree, and novelty, and output a multi-dimensional analysis result;

[0037] The hotspot screening unit is used to input the multi-dimensional analysis results and external quantitative indicators into the hotspot scoring model to obtain a normalized hotspot index and perform sorting and screening to generate a push hotspot set. The external quantitative indicators include freight rate index, ship waiting time and container availability index.

[0038] A hot spot information tracking device for a maritime supply chain, comprising a memory and a processor;

[0039] The memory is used to store programs;

[0040] The processor is used to execute the program to implement each step of the hot spot information tracking method for the shipping supply chain as described in any one of the above items.

[0041] A readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of any of the above-mentioned methods for tracking hot spot information in a shipping supply chain.

[0042] A computer program product includes a computer program, which, when executed by a processor, executes the steps of any of the above-mentioned methods for tracking hot spot information in a shipping supply chain.

[0043] It can be seen from the above technical solutions that the embodiment of the present application provides a method and related equipment for tracking hot spot information in the shipping supply chain. By constructing a multi-dimensional feature knowledge base of the shipping supply chain including port codes, routes, shipping companies, key event words, etc., and establishing dynamic mapping associations between it and diversified data sources such as text news, structured freight rate data, ship automatic identification system data and meteorological data, it achieves precise focusing and integration of professional information in the field of shipping supply chain, solving the problem that the existing technology lacks industry depth and accuracy, resulting in inaccurate information screening and hot spot identification.

[0044] By driving the first language model based on a domain-specific summary prompt template, key logistics elements are extracted from diverse data sources to generate information summaries, and by using a multi-dimensional classification prompt template to drive the second language model, the information summaries are classified according to event type, logistics link, geographical scope, degree of impact and novelty, and multi-dimensional analysis results are output. This enables in-depth mining and multi-dimensional analysis of industry information, making up for the defects of existing technologies with single analysis dimensions and superficial insights, and can reveal the nature, severity and potential impact of events.

[0045] By leveraging the powerful understanding capabilities of large language models and combining them with dynamic mapping associations, we can uniformly process multi-source heterogeneous data such as text and structured data, effectively integrate multi-source heterogeneous data, solve the difficulties of existing technologies in data fusion, and give full play to the value of different types of data.

[0046] The multi-dimensional analysis results and external quantitative indicators such as freight rate index, ship waiting time and container availability index are input into the hotspot scoring model to obtain a normalized hotspot index and sort and filter it to generate a push hotspot set. This allows for comprehensive judgment of hotspot information based on multiple factors, enabling more timely and accurate identification of complex hotspots and potential risks, solving the problem of delayed and inaccurate hotspot judgment in existing technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0048] Figure 1 This is a flow chart of a method for tracking hotspot information in a shipping supply chain disclosed in an embodiment of the present application;

[0049] Figure 2 This is a schematic diagram of a hotspot information tracking device for a shipping supply chain disclosed in an embodiment of the present application;

[0050] Figure 3 This is a hardware structure block diagram of a hot spot information tracking device for a shipping supply chain disclosed in an embodiment of the present application. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0052] The present application can be used in a variety of general or special computing device environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multi-processor devices, and distributed computing environments including any of the above devices or devices.

[0053] Next, we will introduce the application scheme. This application proposes the following technical scheme, please see below for details.

[0054] Figure 1 This is a flowchart of a method for tracking hot spot information in a shipping supply chain disclosed in an embodiment of the present application.

[0055] like Figure 1 As shown, the method may include:

[0056] Step S1: Build a multi-dimensional feature knowledge base for the shipping supply chain and establish a dynamic mapping association between the multi-dimensional feature knowledge base and diversified data sources, where the multi-dimensional features include port codes, routes, shipping companies, and key event words, and the diversified data sources include text news, structured freight rate data, ship automatic identification system data, and meteorological data.

[0057] Specifically, the multi-dimensional feature knowledge base is a specialized knowledge system built for the supply chain shipping sector. Its core dimensions encompass internationally recognized port codes, major global shipping routes, shipping company names and codes, and key industry event terms such as "port congestion," "freight rate fluctuations," and "strikes," forming a structured set of domain features. The diverse data sources include unstructured text news (such as real-time reports from shipping information websites), structured freight rate data (such as the SCFI and CCFI indices), Automatic Identification System (AIS) data (including vessel position, speed, arrival time, etc.), and meteorological data (such as typhoon paths and maritime weather warnings). Through rule matching, model recommendation, and frequency adjustment, a dynamic mapping is established between each dimensional feature in the knowledge base and its corresponding data source. For example, "Port Code A" is associated with text news related to Port A, its corresponding AIS data, and freight rate data. This mapping relationship can be adjusted in real time based on industry dynamics (such as new route openings and port policy changes) or user needs, enabling precise data capture and association based on domain features, providing a domain-focused, dynamically adjustable data foundation for subsequent information processing.

[0058] The adjustment of the dynamic mapping association between the multi-dimensional feature knowledge base and the diversified data sources includes at least one of rule matching, model recommendation, and update frequency adjustment;

[0059] First, the rule matching method is based on a preset industry mapping rule library, matching the fields in the diversified data sources with the multi-dimensional features in the multi-dimensional feature knowledge base to generate a mapping relationship.

[0060] Based on a pre-set industry mapping rule base, fields in diverse data sources (such as port names in news text and route identifiers in freight rate data) are semantically and structurally matched with features such as port codes, routes, and shipping companies in a multi-dimensional feature knowledge base. Initial mapping relationships are generated using rules such as regular expressions and keyword matching. For example, the rule base pre-sets that "Port A" corresponds to the port code "157618." When the text "Port A is congested" appears in the data source, it is automatically associated with the "157618" feature node in the knowledge base, achieving rapid mapping based on industry rules.

[0061] Secondly, the model recommendation method is to calculate the semantic similarity between the data source and the knowledge base features through the model, generate a dynamic association weight matrix, and iteratively update the weight coefficient based on user feedback.

[0062] Using semantic similarity calculation models (such as BERT and Word2Vec), data source content and knowledge base features are vectorized, generating a dynamic association weight matrix to quantify the strength of association between different data sources and feature dimensions. For example, for news about "XY route freight rate fluctuations," the model calculates the semantic similarity score between it and the "XY route" feature in the knowledge base and generates a corresponding weight coefficient. Simultaneously, based on user feedback on the mapping results (such as flagging cases of inaccurate associations), the weight matrix is ​​iteratively updated using a machine learning algorithm, achieving intelligent optimization of the mapping relationship.

[0063] Third, the update frequency adjustment method is to adjust the priority weight of the mapping association based on the update frequency of the diversified data source and the confidence score output by the data quality assessment model.

[0064] By analyzing the update frequency of diverse data sources (such as real-time weather data, daily freight rate indices, and periodically released industry reports), combined with the confidence scores output by the data quality assessment model (such as data completeness, timeliness, and accuracy indicators), the priority weights of mapping associations are dynamically adjusted. For example, real-time Automatic Identification System data is given a higher mapping priority due to its high timeliness and accuracy. For less frequently updated industry report data, the association weights are dynamically adjusted based on its historical confidence score. This ensures that the system prioritizes the mapping of high-value, time-sensitive data, thereby improving overall information processing efficiency and accuracy.

[0065] Step S2: driving the first language model based on the domain-specific summary prompt template to extract key logistics elements from the diversified data sources and generate an information summary.

[0066] Specifically, key logistics elements include involved objects (such as ports, routes, shipping companies, and ship IDs), event types (such as strikes, congestion, freight rate fluctuations, abnormal speeds, and typhoon disasters), locations of occurrence (such as Port A, Route XY, Sea Area B, Port C, and Sea Area D), time (such as an estimated arrival time delayed by 12 hours and a duration of 48 hours), values ​​(such as a 5% increase in the freight rate index), scope of impact (such as affecting the berthing of ships on Route XY), causes (such as being affected by congestion at Port A), and potential impacts (such as causing ships on Route XY passing through the sea area to be diverted).

[0067] For different types of diverse data sources such as text news, structured freight rate data, ship automatic identification system data and meteorological data, we design and call domain-specific summary prompt templates that adapt to their data characteristics. For example, for unstructured data such as text news, a prompt template containing instructions such as "extracting the event subject, event type, location, impact range, and related routes" is used to drive the first language model to perform semantic analysis on the news content, extract key logistics elements such as "a workers' strike occurred in Port A, affecting the berthing of ships on Route XY", and generate a structured information summary; for structured freight rate data, a preset prompt template of "analyzing the fluctuation range of the freight rate index, corresponding routes, and influencing factors" is used to trigger the model to extract "the freight rate index of Route XY rose by 5% this week, affected by the congestion at Port A" from the numerical data. " and other core information; for ship automatic identification system data, through prompt templates containing instructions such as "identify ship ID, current position, abnormal speed, expected arrival time delay", the model is driven to extract key elements such as "ship M's speed suddenly dropped in sea area B, and the estimated arrival time at port C is delayed by 12 hours" from the real-time data stream; for meteorological data, the prompt template of "extracting disaster type, affected sea area, duration, and potential impact on shipping" is used to enable the model to generate information summaries such as "Typhoon N is expected to affect sea area D within 48 hours, which may cause ships on route XY passing through this sea area to be diverted". Through the targeted drive of the above-mentioned field-specific templates, the accurate extraction of key logistics elements in multi-source data and the generation of structured summaries are achieved, providing standardized input for subsequent multi-dimensional analysis.

[0068] Step S3: Based on the multi-dimensional classification prompt template, the second language model is driven to classify the information summary by event type, logistics link, geographical scope, impact degree and novelty, and output a multi-dimensional analysis result.

[0069] Specifically, the multi-dimensional classification prompt template is designed for the business needs of the supply chain shipping field, and includes five core classification dimensions: event type, logistics link, geographical scope, impact level, and novelty:

[0070] Event type classification: This drives the second language model to identify core event attributes in information summaries. For example, it classifies "workers' strike at Port A" as a "labor incident," "5% increase in freight rates on route XY" as "freight rate fluctuation," "sudden drop in speed of vessel M" as "transportation anomaly," and "Typhoon N affecting sea area D" as a "natural disaster."

[0071] Logistics Link Classification: Based on the supply chain nodes involved in the event, the information summary is matched to specific logistics links such as "port operations," "shipping," "vessel scheduling," and "cargo warehousing." For example, port congestion events are associated with the "port operations" link, while shipping rate fluctuations are associated with the "shipping" link.

[0072] Geographic scope classification: Extract geographic identifiers from information summaries and match them to geographic dimensions in the knowledge base. For example, "Port A" and "Port C" are mapped to specific port codes, and "Route XY", "Sea Area B", and "Sea Area D" are associated with corresponding routes or sea area scopes to achieve precise positioning of the area affected by the event.

[0073] Impact Classification: The model analyzes the event's quantitative indicators (such as freight rate increases and vessel delays) and potential consequences (such as rerouting and suspension of sailings), and combines them with pre-set scoring rules to categorize the impact into three levels: "Minor," "Moderate," and "Severe." For example, a 12-hour delay in estimated arrival time affecting more than 10 vessels is classified as "Moderate," while a complete suspension of sailings due to a typhoon is classified as "Severe."

[0074] Novelty classification: Based on historical event data in the knowledge base, semantic similarity is calculated to determine whether the current event represents a new industry trend or a rare occurrence. For example, a new maritime policy interpretation appearing for the first time in Sea Area E is classified as "high novelty," while regular seasonal freight rate fluctuations are classified as "low novelty."

[0075] Driven by the aforementioned multi-dimensional classification prompt template, the second language model performs structured analysis of the information summary and outputs an analysis report containing the classification results of each dimension. For example, "Event Type: Natural Disaster - Typhoon, Logistics Link: Route Transportation, Geographical Scope: Sea Area D and Part of Route XY, Impact Level: Severe (Expected 3-day Suspension), Novelty: Low (Common in the same period in history)" provides multi-dimensional data support for subsequent hotspot index calculation and risk assessment.

[0076] Step S4: Input the multi-dimensional analysis results and external quantitative indicators into the hotspot scoring model to obtain a normalized hotspot index and perform sorting and screening to generate a push hotspot set. The external quantitative indicators include freight rate index, ship waiting time and container availability index.

[0077] Specifically, the hotspot scoring model is a comprehensive evaluation model that integrates multi-source data, and its input includes two parts:

[0078] Multi-dimensional analysis results: structured categorized data from step S3, including event type, logistics link, geographical scope, impact, and novelty. The impact and novelty dimensions are converted into numerical indicators using preset scoring rules.

[0079] External quantitative indicators: Real-time collected freight rate indices (such as the month-on-month increase in SCFI and CCFI), vessel waiting times (such as the average waiting time at port anchorages), container availability indexes (such as the empty container turnover rate at the target port), and other numerical data need to be normalized to unify the dimensions.

[0080] The model calculates the comprehensive hotspot index through a weighted summation algorithm, and the formula is:

[0081] Hotspot index = α × multi-dimensional score + β × quantitative indicator score + γ × geographical scope weight

[0082] Among them, α, β, and γ are adjustable weight coefficients (the default values ​​are 0.5, 0.4, and 0.1, respectively). The geographical scope weight is dynamically adjusted according to the economic importance of the event-affected area (such as hub ports and trunk shipping routes) (for example, if Port A is a hub port, the geographical scope weight of its related events will be increased by 10%).

[0083] The calculated hotspot index is normalized and sorted from high to low by score, filtering out the top-N hotspot events (e.g., the top 20). This generates a push notification collection containing an event summary, scores for each dimension, and the hotspot index. For example, an event scored 92 points (85 points for "severe" impact, 88 points for the quantitative indicator corresponding to a 7% increase in the freight rate index, and a 10% bonus for the geographic scope of hub port A), ranking third, will be pushed to the user terminal and trigger the risk warning process. This mechanism enables comprehensive multi-factor assessment and prioritization of complex hotspots, ensuring timely access to high-value information and decision support.

[0084] The training process of the hotspot scoring model includes:

[0085] ① Collect historical hot event data and corresponding sample external quantitative indicators;

[0086] ② Extract the event type, logistics link, geographical scope, impact degree and novelty classification from the sample multi-dimensional classification results as feature vectors, and combine them with the sample external quantitative indicators as input features;

[0087] ③ A supervised learning algorithm is used to output the training model parameters with the manually labeled normalized hotspot index as the target, and the hyperparameters are adjusted through cross-validation to optimize the model generalization ability.

[0088] First, historical hot event data and corresponding external quantitative indicators are collected. The historical hot event data includes information on manually identified hot events in the supply chain and shipping industries. The external quantitative indicators correspond to numerical data such as freight rate index, vessel waiting time, and container availability index at the time of each event. Next, the event type, logistics link, geographic scope, impact level, and novelty classifications from the multi-dimensional classification results are extracted as feature vectors and combined with the external quantitative indicators to form input features. For example, information such as "Event Type: Port Congestion," "Impact Level: Severe," and "Freight Rate Index Increased by 8%" for a historical event is converted into structured numerical feature vectors (e.g., event type coded as 03, impact level assigned as 90, and freight rate index increase normalized as 85), forming the input dataset for the model. Finally, a supervised learning algorithm (such as a random forest, gradient boosting tree, or neural network) is used to train the model parameters using the manually labeled normalized hot index as the target output. Through cross-validation (such as K-fold cross-validation), hyperparameters (such as weight coefficients, learning rate, tree depth, etc.) are adjusted to optimize the model's generalization ability for unseen data, ensuring that the model can accurately calculate the hot index and sort and filter hot events in actual applications.

[0089] In addition, this application can also add a feedback closed-loop optimization process, specifically:

[0090] Based on the user's click feedback, marking operation or manual correction data on the normalized hot spot index or the pushed hot spot set, the parameters of the dynamic mapping association, the field-specific summary prompt template, the multi-dimensional classification prompt template and the hot spot scoring model are iteratively optimized to form a closed-loop optimization mechanism.

[0091] Specifically, based on the user's click feedback, marking operation or manual correction data (such as the user marking an event classification error, adjusting the hot index weight), feedback features (such as the type of incorrect classification, the weight adjustment range) on the normalized hot index or the pushed hot spot set are extracted and input into the dynamic mapping association module, the domain-specific summary prompt template module, the multi-dimensional classification prompt template module and the hot spot scoring model, respectively, to trigger the iterative optimization of the parameters of each module. For example, if the user marks that the extraction of key logistics elements of a text news omits the "scope of impact", the feedback data will be used to adjust the instruction weight of the domain-specific summary prompt template to enhance the accuracy of subsequent element extraction of similar data; if the user corrects the impact degree classification of an event, the correction result can be used as a training sample to input the optimization model of the multi-dimensional classification prompt template to update the classification rules; for the hot spot scoring model, the user's manual adjustment of the hot spot index can be converted into the target output of the model training, and the model parameters can be retrained through the supervised learning algorithm to optimize the weighting coefficient and geographical range weight and other configurations. Through the continuous collection of the above-mentioned feedback data and the coordinated optimization of each module, a closed-loop optimization mechanism covering the entire process of data association, feature extraction, classification analysis and hotspot determination is formed, enabling the system to adapt to dynamic changes in the industry and personalized needs of users, and continuously improve the accuracy and intelligence level of hotspot tracking.

[0092] It can be seen from the above technical solutions that the embodiment of the present application provides a method and related equipment for tracking hot spot information in the shipping supply chain. By constructing a multi-dimensional feature knowledge base of the shipping supply chain including port codes, routes, shipping companies, key event words, etc., and establishing dynamic mapping associations between it and diversified data sources such as text news, structured freight rate data, ship automatic identification system data and meteorological data, it achieves precise focusing and integration of professional information in the field of shipping supply chain, solving the problem that the existing technology lacks industry depth and accuracy, resulting in inaccurate information screening and hot spot identification.

[0093] By driving the first language model based on a domain-specific summary prompt template, key logistics elements are extracted from diverse data sources to generate information summaries, and by using a multi-dimensional classification prompt template to drive the second language model, the information summaries are classified according to event type, logistics link, geographical scope, degree of impact and novelty, and multi-dimensional analysis results are output. This enables in-depth mining and multi-dimensional analysis of industry information, making up for the defects of existing technologies with single analysis dimensions and superficial insights, and can reveal the nature, severity and potential impact of events.

[0094] By leveraging the powerful understanding capabilities of large language models and combining them with dynamic mapping associations, we can uniformly process multi-source heterogeneous data such as text and structured data, effectively integrate multi-source heterogeneous data, solve the difficulties of existing technologies in data fusion, and give full play to the value of different types of data.

[0095] The multi-dimensional analysis results and external quantitative indicators such as freight rate index, ship waiting time and container availability index are input into the hotspot scoring model to obtain a normalized hotspot index and sort and filter it to generate a push hotspot set. This allows for comprehensive judgment of hotspot information based on multiple factors, enabling more timely and accurate identification of complex hotspots and potential risks, solving the problem of delayed and inaccurate hotspot judgment in existing technologies.

[0096] In some embodiments of the present application, the field-specific summary prompt template includes a first type of template embedded in an industry causal trigger vocabulary, a second type of template embedded in freight rate data association, and a third type of template embedded in a spatiotemporal rule library of ship trajectories and a meteorological event keyword table.

[0097] On this basis, step S2, driving the first language model based on the domain-specific summary prompt template to extract key logistics elements from the diverse data sources and generate information summaries, may specifically include:

[0098] Step S21: For a text data source, the first language model is driven based on the first template to extract an event causal chain from the text, and generate a text information summary including objects, events, time, location, values, and causes.

[0099] Step S22: For the structured data source, the first language model is driven based on the second type of template to semantically align the freight index data and the vessel position data with the multi-dimensional features in the multi-dimensional feature knowledge base, thereby generating a structured information summary including the correlation between the numerical index and the industry feature;

[0100] Step S23: For the semi-structured data source, the first language model is driven based on the third type of template to analyze the port congestion probability triggered by the ship trajectory deviation threshold or the meteorological event keyword, and generate a semi-structured information summary including spatiotemporal abnormal events and impact predictions;

[0101] Step S24: Integrate the text information summary, the structured information summary, and the semi-structured information summary to generate an information summary.

[0102] Specifically, the industry causal trigger vocabulary is a set of causal vocabulary extracted from shipping industry texts, including a set of trigger words that express causal relationships, which is used to assist the large language model in parsing the event causal logic in the text; freight data association is a mechanism for establishing semantic matching between freight index data (such as freight index, shipping company quotations) and industry features (such as routes, ports, shipping companies) in the multi-dimensional feature knowledge base; the spatiotemporal rule library of ship trajectories contains spatiotemporal laws and abnormal thresholds such as the normal speed range of ships, route navigation time thresholds, port docking time standards, etc., which are used to identify trajectory anomalies; the meteorological event keyword table includes keywords related to shipping risks such as "typhoon", "heavy fog", and "severe convective weather", which are used to trigger meteorological impact forecasts.

[0103] The diverse data sources include:

[0104] Text data sources: Unstructured natural language text, such as shipping news and policy announcements, requires semantic parsing to extract key elements;

[0105] Structured data sources: Numerical data in predefined formats, such as freight index and automatic identification system data, can be directly associated with knowledge base features;

[0106] Semi-structured data sources: data with a certain format but not strictly structured, such as ship trajectory reports and weather warning documents, which require the use of rules or keywords to parse implicit information.

[0107] For text data sources, the first language model, driven by the first type of template, extracts the causal chain of events by scanning the text for industry causal trigger words, generating a summary of the text information including the object, event, time, location, value, and cause. For example, for the news article "Port A has suspended loading and unloading operations starting today due to equipment failure, which is expected to affect the berthing of 8 ships on route XY," the model identifies the causal relationship through trigger words, extracts "Cause: Equipment failure, Event: Suspension of operations at Port A, Impact: Delays of 8 ships," and outputs a structured summary.

[0108] For structured data sources, the first language model is driven by the second type of templates. It semantically aligns freight rate index data and vessel position data with industry characteristics in the knowledge base, generating a structured summary that links numerical indicators and industry characteristics. For example, for data such as "The CCFI index for route XY reached 2000 points this week, up 3%." the model associates the "XY route" feature with it and outputs the summary "The XY route freight rate index (CCFI) rose 3% month-over-month (2000 points), associated with XY route."

[0109] For semi-structured data sources, the first language model is driven by the third type of template. Using the spatiotemporal rule base of ship trajectories, it identifies speed anomalies or identifies risks using a meteorological event keyword list. This generates a semi-structured information summary containing spatiotemporal anomalies and impact predictions. For example, for a report such as "Ship M's speed dropped sharply from 10 knots to 4 knots in Sea Area B, resulting in an orange weather warning for heavy rain," the model combines the rule base (speed anomaly) with the keyword (heavy rain) to produce the summary "Ship M's speed in Sea Area B was abnormal (4 knots), and its arrival at Port C is expected to be delayed by 10 hours. Heavy rain may disrupt operations at Port C, increasing the probability of congestion to 70%."

[0110] The data fusion module integrates text, structured, and semi-structured information summaries, removes duplicate elements, and standardizes the format to generate a unified information summary. For example, the above example would produce the following: "Event: Equipment failure at Port A causes operations to be suspended, freight rates on Route XY increase by 3%, vessel M's speed is abnormal, and congestion may occur at Port C; Affected entities: Port A, Route XY, vessel M, Port C; Time: Starting today; Location: Route XY, Sea Area B, Port C; Value: Affects 8 vessels, freight index 2000 points, delay 10 hours; Cause: Equipment failure, heavy rain; Forecast: 70% probability of congestion at Port C."

[0111] In some embodiments of the present application, the multi-dimensional classification prompt template includes a causal chain reasoning template, a novelty evaluation template, and an impact degree classification template.

[0112] On this basis, step S3, driving the second language model based on the multi-dimensional classification prompt template to classify the information summary by event type, logistics link, geographical scope, impact degree, and novelty, and outputting the multi-dimensional analysis results, may specifically include:

[0113] Step S31: Based on the causal chain reasoning template, the second language model is driven to extract the event timestamp and geographic tag in the information summary, and a model of the lagged impact of events on shipping rates or vessel waiting times is constructed by combining historical freight rate data and port operation records. The causal relationship weights are quantified based on a probabilistic causal graph to generate a causal chain analysis result.

[0114] Step S32: driving the second language model to compare the semantic similarity between the current event and the historical events in the multi-dimensional feature knowledge base based on the novelty evaluation template, and determining a novelty flag according to the semantic similarity;

[0115] Step S33: Based on the impact classification template, the second language model is driven to classify the impact level according to the fluctuation range of the freight index, the growth rate of the ship waiting time, and the decline rate of the container availability index;

[0116] Step S34: Integrate the causal chain analysis results, the novelty mark, and the impact level, and output a multi-dimensional analysis result including event type, logistics link, geographical scope, impact degree, and novelty.

[0117] Specifically, the causal chain reasoning template is used to drive the second language model to analyze the causal relationship between events. By extracting the event timestamps and geographic tags in the information summary, combined with historical freight data, port operation records, etc., it constructs a lagged impact model of events on shipping freight rates or ship waiting times, and quantifies the causal relationship weights based on the probabilistic causal graph to reveal the deep connection between events; the novelty evaluation template determines the novelty tag (such as "high", "medium", "low") according to the similarity threshold by comparing the semantic similarity between the current event and the historical events in the multi-dimensional feature knowledge base (such as using BERT and other models to calculate the vector distance); the impact degree classification template divides the impact level (such as "serious", "medium", "minor") according to quantitative indicators such as the fluctuation range of the freight rate index, the growth rate of ship waiting time, and the decline rate of the container availability index, combined with preset level standards.

[0118] Based on the causal chain reasoning template, the second language model extracts the timestamp and geotag of the event from the information summary. It then retrieves historical data on freight rate fluctuations and port vessel waiting times within the same geographic area to construct a time series lag model, analyzing the correlation between the event and subsequent changes in freight rates and waiting times. For example, regarding the "Port A Workers' Strike" incident, the model found through historical data that such events typically lead to a 3%-8% increase in freight rates on route XY after 72 hours. Combined with the probabilistic causal graph, the model calculated a causal relationship weight of 0.82, generating a causal chain analysis result: "Strike Event → Lagged Increase in Freight Rates on Route XY."

[0119] Based on the novelty assessment template, the model calculates semantic similarity between key elements of the current event (such as event type and impact) and feature vectors of historical events in the knowledge base. For example, if the current event is "Port Z implements its first drone cargo pilot," and its similarity with "Other ports' traditional cargo modes" in the knowledge base is 32%, below the 50% threshold, the model will be labeled "high" for novelty. If the event is "Conventional typhoon affects air routes," and its similarity with historical typhoon events is 85%, the model will be labeled "low."

[0120] Based on the impact classification template, the model extracts quantitative indicators from the information summary and calculates the rate of change. For example, if an event causes the freight rate index on route XY to increase by 6% (exceeding the severe threshold of 5%), vessel waiting times to increase by 40% (exceeding the moderate threshold of 30%), and the container availability index to decrease by 25% (below the severe threshold of 30%), the overall impact level is determined to be "severe."

[0121] This approach integrates causal chain analysis results, novelty markers, and impact levels, combining event type, logistics link, geographic scope, and other dimensions to produce structured, multi-dimensional analysis results. For example, "Event Type: Labor Incident - Strike, Logistics Link: Port Operations, Geographic Scope: Port A, Route XY, Causal Chain: Strike → Delayed Freight Rate Increase (Weight 0.82), Impact: Severe (Freight Rate Increase 6%), Novelty: Low (Represents ≥ 60% of Historical Similar Events)."

[0122] The following describes a hotspot information tracking device for a shipping supply chain provided by an embodiment of the present application. The hotspot information tracking device for a shipping supply chain described below and the hotspot information tracking method for a shipping supply chain described above can correspond to each other.

[0123] See also Figure 2 , Figure 2 This is a schematic diagram of a hot spot information tracking device for a shipping supply chain disclosed in an embodiment of the present application.

[0124] like Figure 2 As shown, the hot spot information tracking device of the shipping supply chain may include:

[0125] A mapping association unit 110 is configured to construct a multi-dimensional feature knowledge base for the maritime supply chain and establish dynamic mapping associations between the multi-dimensional feature knowledge base and diverse data sources, wherein the multi-dimensional features include port codes, routes, shipping companies, and key event words. The diverse data sources include text news, structured freight rate data, automatic vessel identification system data, and meteorological data.

[0126] An information summary unit 120 is configured to drive a first language model based on a domain-specific summary prompt template, extract key logistics elements from the diverse data sources, and generate an information summary;

[0127] A multi-dimensional analysis unit 130 is configured to drive a second language model based on a multi-dimensional classification prompt template to classify the information summary by event type, logistics link, geographical scope, impact, and novelty, and output a multi-dimensional analysis result;

[0128] The hotspot screening unit 140 is used to input the multi-dimensional analysis results and external quantitative indicators into the hotspot scoring model to obtain a normalized hotspot index and perform sorting and screening to generate a push hotspot set. The external quantitative indicators include freight rate index, ship waiting time and container availability index.

[0129] It can be seen from the above technical solutions that the embodiment of the present application provides a method and related equipment for tracking hot spot information in the shipping supply chain. By constructing a multi-dimensional feature knowledge base of the shipping supply chain including port codes, routes, shipping companies, key event words, etc., and establishing dynamic mapping associations between it and diversified data sources such as text news, structured freight rate data, ship automatic identification system data and meteorological data, it achieves precise focusing and integration of professional information in the field of shipping supply chain, solving the problem that the existing technology lacks industry depth and accuracy, resulting in inaccurate information screening and hot spot identification.

[0130] By driving the first language model based on a domain-specific summary prompt template, key logistics elements are extracted from diverse data sources to generate information summaries, and by using a multi-dimensional classification prompt template to drive the second language model, the information summaries are classified according to event type, logistics link, geographical scope, degree of impact and novelty, and multi-dimensional analysis results are output. This enables in-depth mining and multi-dimensional analysis of industry information, making up for the defects of existing technologies with single analysis dimensions and superficial insights, and can reveal the nature, severity and potential impact of events.

[0131] By leveraging the powerful understanding capabilities of large language models and combining them with dynamic mapping associations, we can uniformly process multi-source heterogeneous data such as text and structured data, effectively integrate multi-source heterogeneous data, solve the difficulties of existing technologies in data fusion, and give full play to the value of different types of data.

[0132] The multi-dimensional analysis results and external quantitative indicators such as freight rate index, ship waiting time and container availability index are input into the hotspot scoring model to obtain a normalized hotspot index and sort and filter it to generate a push hotspot set. This allows for comprehensive judgment of hotspot information based on multiple factors, enabling more timely and accurate identification of complex hotspots and potential risks, solving the problem of delayed and inaccurate hotspot judgment in existing technologies.

[0133] Optionally, the adjustment of the dynamic mapping association between the multi-dimensional feature knowledge base and the diversified data sources includes at least one of rule matching, model recommendation, and update frequency adjustment;

[0134] The rule matching method is based on a preset industry mapping rule library, matching the fields in the diversified data sources with the multi-dimensional features in the multi-dimensional feature knowledge library to generate a mapping relationship;

[0135] The model recommendation method is to calculate the semantic similarity between the data source and the knowledge base features through the model, generate a dynamic association weight matrix, and iteratively update the weight coefficient based on user feedback;

[0136] The update frequency adjustment method is to adjust the priority weight of the mapping association based on the update frequency of the diversified data source and the confidence score output by the data quality assessment model.

[0137] Optionally, the domain-specific summary prompt template includes a first type of template embedded in an industry causal trigger vocabulary, a second type of template embedded in freight rate data association, and a third type of template embedded in a spatiotemporal rule library of ship trajectories and a meteorological event keyword table;

[0138] The domain-specific summary prompt template drives the first language model to extract key logistics elements from the diverse data sources and generate information summaries, including:

[0139] For text data sources, the first language model is driven by the first template to extract event causal chains from the text and generate a text information summary containing objects, events, time, location, values, and causes.

[0140] For structured data sources, the first language model is driven based on the second type of template to semantically align freight rate index data and vessel position data with the multi-dimensional features in the multi-dimensional feature knowledge base, thereby generating a structured information summary containing the relationship between numerical indicators and industry characteristics;

[0141] For semi-structured data sources, the first language model is driven by the third type of template to analyze the probability of port congestion triggered by ship trajectory deviation thresholds or meteorological event keywords, and generate a semi-structured information summary containing spatiotemporal abnormal events and impact predictions;

[0142] The text information summary, the structured information summary, and the semi-structured information summary are integrated to generate an information summary.

[0143] Optionally, the multi-dimensional classification prompt template includes a causal chain reasoning template, a novelty evaluation template, and an impact degree classification template;

[0144] The multi-dimensional classification prompt template drives the second language model to classify the information summary by event type, logistics link, geographical scope, impact level and novelty, and output multi-dimensional analysis results, including:

[0145] The second language model is driven by the causal chain reasoning template to extract event timestamps and geographic tags from the information summary, and a model for the lagged impact of events on shipping rates or vessel waiting times is constructed by combining historical freight rate data and port operation records. The causal relationship weights are quantified based on a probabilistic causal graph to generate causal chain analysis results.

[0146] Driving the second language model to compare the semantic similarity between the current event and the historical events in the multi-dimensional feature knowledge base based on the novelty evaluation template, and determining a novelty tag according to the semantic similarity;

[0147] Based on the impact classification template, the second language model is driven to classify the impact level according to the fluctuation range of the freight rate index, the growth rate of the ship waiting time, and the decline rate of the container availability index;

[0148] The causal chain analysis results, the novelty mark and the impact level are integrated to output a multi-dimensional analysis result including event type, logistics link, geographical scope, impact degree and novelty.

[0149] Optionally, the training process of the hotspot scoring model includes:

[0150] Collect historical hot event data and corresponding sample external quantitative indicators;

[0151] Extract the event type, logistics link, geographical scope, impact degree and novelty classification from the sample multi-dimensional classification results as feature vectors, and combine them with the sample external quantitative indicators as input features;

[0152] A supervised learning algorithm is used to output the training model parameters with the manually labeled normalized hotspot index as the target, and the hyperparameters are adjusted through cross-validation to optimize the model generalization ability.

[0153] Optionally, also include:

[0154] Based on the user's click feedback, marking operation or manual correction data on the normalized hot spot index or the pushed hot spot set, the parameters of the dynamic mapping association, the field-specific summary prompt template, the multi-dimensional classification prompt template and the hot spot scoring model are iteratively optimized to form a closed-loop optimization mechanism.

[0155] The hot spot information tracking device for a shipping supply chain provided in the embodiment of the present application can be applied to hot spot information tracking equipment for a shipping supply chain. Figure 3 The hardware structure diagram of the hot spot information tracking device of the maritime supply chain is shown. Figure 3 ,The hardware structure of the hot spot information tracking device for the maritime supply chain may include: at least one processor 1, at least one communication interface 2, at least one memory 3 and at least one communication bus 4;

[0156] In the embodiment of the present application, the number of the processor 1, the communication interface 2, the memory 3, and the communication bus 4 is at least one, and the processor 1, the communication interface 2, and the memory 3 communicate with each other through the communication bus 4;

[0157] The processor 1 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention;

[0158] The memory 3 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory;

[0159] The memory stores a program, and the processor can call the program stored in the memory, wherein the program is used to:

[0160] Build a multi-dimensional feature knowledge base for the maritime supply chain and establish dynamic mapping relationships between the multi-dimensional feature knowledge base and diverse data sources, including text news, structured freight rate data, Automatic Vessel Identification System data, and meteorological data.

[0161] Based on the domain-specific summary prompt template, the first language model is driven to extract key logistics elements from the diverse data sources and generate information summaries;

[0162] The second language model is driven by a multi-dimensional classification prompt template to classify the information summary by event type, logistics link, geographical scope, impact level and novelty, and output a multi-dimensional analysis result;

[0163] The multi-dimensional analysis results and external quantitative indicators are input into the hotspot scoring model to obtain a normalized hotspot index and sort and filter it to generate a push hotspot set. The external quantitative indicators include freight rate index, ship waiting time and container availability index.

[0164] Optionally, the refined functions and extended functions of the program may refer to the above description.

[0165] The present application also provides a readable storage medium, which may store a program suitable for execution by a processor, wherein the program is used to:

[0166] Build a multi-dimensional feature knowledge base for the maritime supply chain and establish dynamic mapping relationships between the multi-dimensional feature knowledge base and diverse data sources, including text news, structured freight rate data, Automatic Vessel Identification System data, and meteorological data.

[0167] Based on the domain-specific summary prompt template, the first language model is driven to extract key logistics elements from the diverse data sources and generate information summaries;

[0168] The second language model is driven by a multi-dimensional classification prompt template to classify the information summary by event type, logistics link, geographical scope, impact level and novelty, and output a multi-dimensional analysis result;

[0169] The multi-dimensional analysis results and external quantitative indicators are input into the hotspot scoring model to obtain a normalized hotspot index and sort and filter it to generate a push hotspot set. The external quantitative indicators include freight rate index, ship waiting time and container availability index.

[0170] Optionally, the refined functions and extended functions of the program may refer to the above description.

[0171] The present application also provides a computer program product, including a computer program. When the computer program is executed by a processor, the execution method is as follows:

[0172] Build a multi-dimensional feature knowledge base for the maritime supply chain and establish dynamic mapping relationships between the multi-dimensional feature knowledge base and diverse data sources, including text news, structured freight rate data, Automatic Vessel Identification System data, and meteorological data.

[0173] Based on the domain-specific summary prompt template, the first language model is driven to extract key logistics elements from the diverse data sources and generate information summaries;

[0174] The second language model is driven by a multi-dimensional classification prompt template to classify the information summary by event type, logistics link, geographical scope, impact level and novelty, and output a multi-dimensional analysis result;

[0175] The multi-dimensional analysis results and external quantitative indicators are input into the hotspot scoring model to obtain a normalized hotspot index and sort and filter it to generate a push hotspot set. The external quantitative indicators include freight rate index, ship waiting time and container availability index.

[0176] Optionally, the refined functions and extended functions of the program may refer to the above description.

[0177] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0178] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0179] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for tracking hot spot information in a shipping supply chain, characterized in that: include: Build a multi-dimensional feature knowledge base for the maritime supply chain and establish dynamic mapping relationships between the multi-dimensional feature knowledge base and diverse data sources, including text news, structured freight rate data, Automatic Vessel Identification System data, and meteorological data. Based on the domain-specific summary prompt template, the first language model is driven to extract key logistics elements from the diverse data sources and generate information summaries; The second language model is driven by a multi-dimensional classification prompt template to classify the information summary by event type, logistics link, geographical scope, impact level and novelty, and output a multi-dimensional analysis result; The multi-dimensional analysis results and external quantitative indicators are input into the hotspot scoring model to obtain a normalized hotspot index and sort and filter it to generate a push hotspot set. The external quantitative indicators include freight rate index, ship waiting time and container availability index.

2. The method according to claim 1, characterized in that The adjustment of the dynamic mapping association between the multi-dimensional feature knowledge base and the diversified data sources includes at least one of rule matching, model recommendation, and update frequency adjustment; The rule matching method is based on a preset industry mapping rule library, matching the fields in the diversified data sources with the multi-dimensional features in the multi-dimensional feature knowledge library to generate a mapping relationship; The model recommendation method is to calculate the semantic similarity between the data source and the knowledge base features through the model, generate a dynamic association weight matrix, and iteratively update the weight coefficient based on user feedback; The update frequency adjustment method is to adjust the priority weight of the mapping association based on the update frequency of the diversified data source and the confidence score output by the data quality assessment model.

3. The method according to claim 1, characterized in that The domain-specific summary prompt templates include a first type of template embedded in an industry causal trigger vocabulary, a second type of template embedded in freight rate data association, and a third type of template embedded in a spatiotemporal rule library of ship trajectories and a meteorological event keyword table; The domain-specific summary prompt template drives the first language model to extract key logistics elements from the diverse data sources and generate information summaries, including: For text data sources, the first language model is driven by the first template to extract event causal chains from the text and generate a text information summary containing objects, events, time, location, values, and causes. For structured data sources, the first language model is driven based on the second type of template to semantically align freight rate index data and vessel position data with the multi-dimensional features in the multi-dimensional feature knowledge base, thereby generating a structured information summary containing the relationship between numerical indicators and industry characteristics; For semi-structured data sources, the first language model is driven by the third type of template to analyze the probability of port congestion triggered by ship trajectory deviation thresholds or meteorological event keywords, and generate a semi-structured information summary containing spatiotemporal abnormal events and impact predictions; The text information summary, the structured information summary, and the semi-structured information summary are integrated to generate an information summary.

4. The method according to claim 1, wherein The multi-dimensional classification prompt template includes a causal chain reasoning template, a novelty evaluation template, and an impact degree classification template; The multi-dimensional classification prompt template drives the second language model to classify the information summary by event type, logistics link, geographical scope, impact level and novelty, and output multi-dimensional analysis results, including: The second language model is driven by the causal chain reasoning template to extract event timestamps and geographic tags from the information summary, and a model for the lagged impact of events on shipping rates or vessel waiting times is constructed by combining historical freight rate data and port operation records. The causal relationship weights are quantified based on a probabilistic causal graph to generate causal chain analysis results. Driving the second language model to compare the semantic similarity between the current event and the historical events in the multi-dimensional feature knowledge base based on the novelty evaluation template, and determining a novelty tag according to the semantic similarity; Based on the impact classification template, the second language model is driven to classify the impact level according to the fluctuation range of the freight rate index, the growth rate of the ship waiting time, and the decline rate of the container availability index; The causal chain analysis results, the novelty mark and the impact level are integrated to output a multi-dimensional analysis result including event type, logistics link, geographical scope, impact degree and novelty.

5. The method according to claim 1, characterized in that The training process of the hotspot scoring model includes: Collect historical hot event data and corresponding sample external quantitative indicators; Extract the event type, logistics link, geographical scope, impact degree and novelty classification from the sample multi-dimensional classification results as feature vectors, and combine them with the sample external quantitative indicators as input features; A supervised learning algorithm is used to output the training model parameters with the manually labeled normalized hotspot index as the target, and the hyperparameters are adjusted through cross-validation to optimize the model generalization ability.

6. The method according to claim 1, characterized in that Also includes: Based on the user's click feedback, marking operation or manual correction data on the normalized hot spot index or the pushed hot spot set, the parameters of the dynamic mapping association, the field-specific summary prompt template, the multi-dimensional classification prompt template and the hot spot scoring model are iteratively optimized to form a closed-loop optimization mechanism.

7. A hot spot information tracking device for a maritime supply chain, characterized in that: include: A mapping association unit is used to construct a multi-dimensional feature knowledge base for the maritime supply chain and establish dynamic mapping associations between the multi-dimensional feature knowledge base and diverse data sources, including text news, structured freight rate data, automatic vessel identification system data, and meteorological data. An information summary unit is used to drive the first language model based on a domain-specific summary prompt template, extract key logistics elements from the diversified data sources, and generate an information summary; a multi-dimensional analysis unit, configured to drive a second language model based on a multi-dimensional classification prompt template, classify the information summary by event type, logistics link, geographical scope, impact degree, and novelty, and output a multi-dimensional analysis result; The hotspot screening unit is used to input the multi-dimensional analysis results and external quantitative indicators into the hotspot scoring model to obtain a normalized hotspot index and perform sorting and screening to generate a push hotspot set. The external quantitative indicators include freight rate index, ship waiting time and container availability index.

8. A hot spot information tracking device for a maritime supply chain, characterized in that: including memory and processor; The memory is used to store programs; The processor is configured to execute the program to implement the steps of the hot spot information tracking method for a shipping supply chain according to any one of claims 1 to 6.

9. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step of the method for tracking hot spot information in a shipping supply chain according to any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, each step of the method for tracking hot spot information in a shipping supply chain as claimed in any one of claims 1 to 6 is executed.

Images