Abnormal baggage processing method and system based on vector matching

Abstract

The present invention relates to the technical field of abnormal baggage processing, and discloses a method and system for abnormal baggage processing based on vector matching. The method includes: constructing a hybrid retrieval knowledge base, integrating multi-source heterogeneous data to generate relational and vector knowledge bases and indexing associations; establishing a hybrid retrieval rule engine, performing keyword analysis and semantic vectorization on case descriptions, generating query instructions and performing parallel retrieval to obtain a set of candidate solutions; developing a domain-enhanced decision-making system, generating structured processing suggestions through a multi-dimensional evidence chain and RAG framework, and visually outputting process specifications and compensation basis; constructing a disposal system to integrate various modules to achieve full-process automation. The system includes hybrid retrieval knowledge base construction, a rule engine, domain-enhanced decision-making, and an integration module. The present invention improves the efficiency, standardization, and scientific nature of abnormal baggage processing by combining vector matching with rule retrieval and enhancing decision-making with a large model, and is suitable for intelligent disposal of abnormal baggage in air transportation.

Description

Technical Field

[0001] The present invention relates to the technical field of abnormal baggage processing, and in particular to a method and system for abnormal baggage processing based on vector matching. Background Art

[0002] In the air transport sector, handling abnormal baggage is a crucial component in protecting passenger rights and ensuring the quality of air service. With the growing demand for air travel, abnormal baggage cases are becoming more diverse and complex, presenting numerous challenges to traditional handling methods.

[0003] From a data processing perspective, abnormal baggage cases involve heterogeneous data from multiple sources, including policy documents from different airlines, historical case records, and images of damaged baggage. Traditional methods lack the ability to effectively integrate and utilize this data. For one thing, paper-based or fragmented policy documents are difficult to quickly retrieve and structure, resulting in inconsistent and inadequate processing rules. Furthermore, historical case data lacks sufficient semantic connections to provide reference for new cases through case similarity analysis, leading to high duplication of effort and low processing efficiency.

[0004] In the search and decision-making stages, traditional processing relies on manual review of rules and empirical judgment, which presents significant limitations. Manual keyword searches struggle to capture deep semantic connections. For example, different airlines may have different definitions and compensation standards for "damaged luggage," and simple keyword matching cannot accurately adapt to specific scenarios. Furthermore, manual decision-making lacks multi-dimensional evidence support, making liability determinations and compensation calculations susceptible to subjective factors. This results in insufficient fairness and transparency in processing results, leading to a high rate of passenger complaints.

[0005] In terms of visualization and interaction, traditional systems lack intuitive process displays and human-machine collaboration mechanisms. Processing procedures are often presented in text reports, with unclear key decision nodes and regulatory basis, making it difficult for frontline staff to quickly grasp the processing logic. Furthermore, the lack of visual case comparison and risk assessment tools means that optimization and adjustment of processing solutions lack data support, making it difficult to adapt to the dynamic changes in complex cases.

[0006] With the development of artificial intelligence (AI), techniques such as vector matching and large language models have provided new approaches to addressing these issues. However, existing technologies have yet to form a complete system for handling abnormal baggage. For one thing, the fusion mechanism between vector retrieval and rule-based retrieval is not mature enough, making it impossible to achieve efficient collaborative queries across cross-modal data. Furthermore, the depth of domain knowledge enhancement and decision-making reasoning is insufficient, making it difficult to organically integrate policy rules, historical cases, and real-time case data, limiting the accuracy and reliability of automated processing. Therefore, an abnormal baggage handling method and system that can integrate multi-source data, implement intelligent retrieval and decision-making, and improve processing efficiency, standardize the processing process, and ensure the scientific and fair nature of the results is urgently needed. Summary of the Invention

[0007] The object of the present invention is to provide a method and system for handling abnormal baggage based on vector matching, so as to solve the problems raised in the above background technology.

[0008] To achieve the above-mentioned object, the present invention provides the following technical solution: a method for handling abnormal baggage based on vector matching, the method comprising:

[0009] Step S1: Construct a hybrid retrieval knowledge base; integrate multi-source heterogeneous data of abnormal baggage case data to generate a relational knowledge base and a vector knowledge base; index and associate the relational knowledge base and the vector knowledge base to establish a hybrid retrieval knowledge base;

[0010] Step S2: Establish a hybrid search rule engine; perform keyword parsing and semantic vectorization on the input abnormal baggage case description to generate a hybrid search query instruction; perform parallel search based on the hybrid search knowledge base to generate a set of candidate processing solutions;

[0011] Step S3: Develop a domain-enhanced decision-making system based on a large model; enhance the candidate treatment solution set with domain knowledge and generate structured treatment suggestions; output treatment process specifications and compensation basis through a visual interactive interface;

[0012] Step S4: Build an abnormal baggage handling system; integrate a hybrid retrieval rule engine, a domain-enhanced decision-making system, and a visual interactive interface to achieve full-process automation of abnormal baggage cases from retrieval to decision-making.

[0013] Preferably, step S1 includes the following steps:

[0014] Step S11: Build a relational knowledge base; collect abnormal baggage handling policy documents from different airlines, perform document structure processing using a localized parsing tool, and generate a triplet dataset of airline-abnormality type-handling rule;

[0015] Step S12: Construct a vector knowledge base; collect historical abnormal baggage case data, perform missing value cleaning and field standardization, generate text vectors and multimodal joint vectors through a pre-trained language model, and store them in a vector index library;

[0016] Step S13: Establish an index association mechanism; set the airline code primary key for the relational knowledge base, configure a dynamic sharding index for the vector knowledge base, and establish a cross-modal query channel for the hybrid retrieval knowledge base through primary key mapping.

[0017] Preferably, step S2 includes the following steps:

[0018] Step S21: Parse the input case description, extract the airline code, exception type, and item feature fields, and generate a keyword search expression;

[0019] Step S22: semantically encode the unstructured text to generate a query vector; calculate the cosine similarity between the query vector and the vectors in the vector knowledge base, and select the top N similar cases by ranking by score;

[0020] Step S23: Merge the keyword search results and the vector search results, perform conflict resolution processing, and generate a candidate processing solution set including processing procedures, compensation standards, and historical cases.

[0021] Preferably, step S22 includes the following steps:

[0022] Step S221: Perform word segmentation and entity recognition on the case description text to extract the airline identifier, abnormal item category, and damage level features;

[0023] Step S222: Generate a text semantic vector using a pre-trained language model, perform feature dimensionality reduction using a multi-layer perceptron, and generate a fixed-dimensional query vector;

[0024] Step S223: Perform shard parallel search in the vector index library, use weighted cosine similarity algorithm to calculate similarity score, merge the shard results and take the top N similar cases.

[0025] Preferably, step S3 includes the following steps:

[0026] Step S31: Perform domain knowledge enhancement; extract airline-specific processing rules from the relational knowledge base, retrieve similar case processing records from the vector knowledge base, and form a multi-dimensional evidence chain;

[0027] Step S32: Constructing a RAG reasoning framework; converting search results into structured prompt words, inputting them into a large language model to generate processing suggestions, including responsibility determination logic and compensation amount calculation basis;

[0028] Step S33: Design a visual output template; convert the processing suggestions into a structured flowchart, mark the key decision nodes and regulatory reference sources, and generate an editable disposal plan document.

[0029] Preferably, step S31 includes the following steps:

[0030] Step S311: Establish a rule weight matrix; assign weight coefficients to evidence from different sources based on airline policy priorities and case confidence levels;

[0031] Step S312: performing evidence fusion calculations; using a Bayesian network model to integrate structured rules and unstructured case data to generate a comprehensive treatment recommendation confidence score;

[0032] Step S313: Construct a decision tree model; convert the fusion result into a decision tree structure that includes abnormality type judgment, applicable compensation standards, and processing flow path.

[0033] Preferably, step S4 includes the following steps:

[0034] Step S41: deploying a hybrid search engine module; configuring a multi-threaded concurrent controller to implement synchronous query scheduling of the relational database and the vector index library;

[0035] Step S42: Develop a decision-making reasoning engine; integrate the rule engine and probabilistic reasoning model to support compliance verification and risk assessment of the treatment plan;

[0036] Step S43: Build a human-computer interaction layer; design a multi-view visualization panel to support comparative analysis of treatment plans, regulatory clause tracing, and electronic signature confirmation functions.

[0037] Preferably, an adversarial training mechanism is introduced in the construction of the vector knowledge base in step S12, and adversarial baggage damage image data is generated through a GAN network.

[0038] Preferably, the domain enhanced decision-making system in step S3 includes a dynamic knowledge distillation module: a three-layer architecture including a basic rule layer, a case reasoning layer and a meta-learning layer is established, and the training weight of complex cases is gradually increased through a curriculum learning strategy.

[0039] Preferably, the present invention further includes an abnormal baggage handling system based on vector matching, the system comprising:

[0040] A hybrid retrieval knowledge base construction module is used to integrate multi-source heterogeneous data on abnormal baggage cases, generate a relational knowledge base and a vector knowledge base, and associate the two through indexing to form a hybrid retrieval knowledge base;

[0041] A hybrid search rule engine module is used to perform keyword parsing and semantic vectorization on the input abnormal baggage case description to generate a hybrid search query instruction, perform parallel search based on the hybrid search knowledge base, and output a set of candidate processing solutions;

[0042] A domain enhancement decision module integrates a large language model to enhance the candidate treatment solution set with domain knowledge to generate structured treatment suggestions, and outputs treatment process specifications and compensation basis through a visual interactive interface;

[0043] The integration module is used to logically integrate the hybrid retrieval knowledge base construction module, the hybrid retrieval rule engine module and the domain enhancement decision module to realize the full-process automated processing of abnormal baggage cases from retrieval to decision-making.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] In terms of data integration and management, a relational knowledge base and a vector knowledge base are generated through the integration of multi-source heterogeneous data, and an index association mechanism is established. The relational knowledge base structures the policy documents of different airlines into "airline-anomaly type-handling rule" triples, addressing the fragmentation and difficulty of retrieval of traditional policy documents and ensuring the consistency and traceability of handling rules. The vector knowledge base generates text vectors and multimodal joint vectors through a pre-trained language model. At the same time, an adversarial training mechanism is introduced to generate adversarial baggage damage image data. This not only achieves semantic-level representation of historical case data but also enhances the model's generalization capabilities for complex damage scenarios. The index association between the two establishes a cross-modal query channel, providing a data foundation for subsequent hybrid retrieval and improving the depth and breadth of data utilization.

[0046] In the intelligent search phase, the hybrid search rule engine generates hybrid search queries through keyword parsing and semantic vectorization, enabling the parallel execution of rule-based and vector-based search. Keyword search accurately extracts structured information such as airline codes and exception types, quickly locating matching policy rules. Semantic vectorization generates query vectors through word segmentation, entity recognition, and pre-trained models, calculates cosine similarity to screen for similar cases, and captures the deep semantic connections within unstructured text. The combined results are then used to perform conflict resolution, generating a set of candidate solutions encompassing processing procedures, compensation standards, and historical cases. Compared to traditional single-search methods, this significantly improves the comprehensiveness and accuracy of searches and shortens case handling response times.

[0047] At the decision support level, the domain-enhanced decision-making system achieves intelligent generation of processing suggestions through the construction of a multi-dimensional evidence chain and the RAG reasoning framework. It extracts airline-specific rules from the relational knowledge base and retrieves similar case handling records from the vector knowledge base to form an evidence chain combining rules and cases. Evidence is then fused using a rule weight matrix and a Bayesian network model to generate a comprehensive confidence score, ensuring the scientific nature of the decision-making basis. The RAG framework converts search results into structured prompt words and inputs them into a large language model to generate processing suggestions that include responsibility determination logic and compensation calculation basis. It also structures the fusion results through a decision tree model to clarify key nodes such as abnormality type judgment and applicable compensation standards. The three-layer architecture of the dynamic knowledge distillation module (basic rule layer, case reasoning layer, meta-learning layer) gradually increases the weight of complex cases through a curriculum learning strategy, continuously improving the model's ability to handle complex cases, making decision results more in line with actual business needs, reducing the cost of manual intervention, and improving the fairness and transparency of processing results.

[0048] In terms of system integration and interaction, the abnormal baggage handling system achieves full automation from search to decision-making. The hybrid search engine module's multi-threaded concurrent controller ensures synchronized scheduling between the relational database and vector index library, improving search efficiency. The decision-making inference engine integrates a rule engine and a probabilistic reasoning model to support compliance verification and risk assessment, ensuring the legality and security of handling solutions. The multi-view visualization panel at the human-computer interaction layer supports solution comparison, regulatory traceability, and electronic signature confirmation, making the handling process intuitive and transparent, facilitating operations for frontline personnel and communication with passengers, improving the user experience while reducing disputes caused by unclear procedures.

[0049] Overall, this invention, through technological integration and system innovation, has built an intelligent and standardized abnormal baggage handling system, effectively solving problems such as insufficient data utilization, low retrieval decision-making efficiency, and highly subjective processing results in traditional processing methods. It significantly improves the efficiency, standardization, and scientific nature of abnormal baggage handling, and provides strong technical support for service optimization and passenger rights protection in the aviation transportation industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a working principle diagram of the abnormal baggage handling method based on vector matching according to the present invention;

[0051] Figure 2 Design diagram for building a hybrid retrieval knowledge base;

[0052] Figure 3 A design diagram for building a hybrid search rule engine;

[0053] Figure 4 A design diagram for developing domain-enhanced decision-making systems based on large models;

[0054] Figure 5 A design diagram for building an abnormal baggage handling system. DETAILED DESCRIPTION

[0055] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0056] See also Figure 1-Figure 5 The present invention relates to a method and system for handling abnormal baggage based on vector matching, and the specific implementation steps are as follows:

[0057] Step S1: Construct a hybrid retrieval knowledge base; integrate multi-source heterogeneous data of abnormal baggage case data to generate a relational knowledge base and a vector knowledge base, and index and associate the relational knowledge base and the vector knowledge base to establish a hybrid retrieval knowledge base.

[0058] Step S2: Establish a hybrid search rule engine; perform keyword parsing and semantic vectorization on the input abnormal baggage case description to generate a hybrid search query instruction; perform parallel search based on the hybrid search knowledge base to generate a set of candidate processing solutions.

[0059] Step S3: Develop a domain-enhanced decision-making system based on a large model; enhance the candidate processing solution set with domain knowledge and generate structured processing suggestions; output processing process specifications and compensation basis through a visual interactive interface.

[0060] Step S4: Build an abnormal baggage handling system; integrate a hybrid retrieval rule engine, a domain-enhanced decision-making system, and a visual interactive interface to achieve full-process automation of abnormal baggage cases from retrieval to decision-making.

[0061] The present invention will be further described below in conjunction with Examples 1 to 4:

[0062] Example 1: In the process of constructing a hybrid search knowledge base in step S1, the specific implementation method is as follows:

[0063] Step S11 constructs a relational knowledge base. Abnormal baggage handling policy documents from different airlines are collected. These documents come from sources including, but are not limited to, official baggage transportation rule manuals, baggage handling-related contractual agreements, and industry regulations. Because different airlines' policy documents may exist in different formats, such as PDF documents, Word documents, and image-based announcements, these documents must be structured using localized parsing tools. Localized parsing tools can employ parsing algorithms tailored to different document formats. For example, for PDF documents, optical character recognition (OCR) technology is used to extract text content; for structured table data, field information is directly read from the table. During the parsing process, key information from the policy documents must be identified, including the airline name or code, abnormal baggage type (e.g., lost, damaged, delayed, damaged, or oversized), and corresponding handling rules (e.g., compensation standards, processing timelines, and liability determination criteria). By extracting and structuring this key information, a triplet dataset of airline, abnormality type, and handling rule is generated. For example, if an airline's code is CA and the exception type is damaged baggage, the handling rules might include specific clauses such as "Verify the damage within 72 hours. If the damage is caused by the carrier, compensation will be no more than RMB 100 per kilogram, with a maximum of RMB 2,000." These triples are stored according to the table structure of a relational database, forming a relational knowledge base. Each triple is a record, with the airline code and exception type as the primary key fields, and the handling rules as the detailed description field. This facilitates subsequent keyword searches to quickly locate a specific airline's abnormal baggage handling policy.

[0064] Step S12 constructs a vector knowledge base. Historical data on abnormal baggage cases is collected from sources such as airlines' internal baggage handling work orders, case descriptions in customer service systems, and information filed at airport baggage inquiry counters. This data may have inconsistent formats, missing fields, and ambiguous descriptions, so data preprocessing is required. First, missing values are cleaned. Records with missing key fields, such as airline code, exception type, case occurrence time, and item characteristics, are completed through manual verification or comparison with data from other related systems. Records that cannot be completed are removed. Field standardization is then performed, unifying fields with the same meaning but different descriptions into a standard format. For example, "luggage lost" and "lost luggage" are unified into "lost," and "laptop" and "computer" in item characteristics are unified into "laptop." The preprocessed case data includes structured data (such as airline code, exception type, and handling results) and unstructured data (such as detailed case descriptions, photos of baggage damage, and surveillance video footage). For text data, pre-trained language models (such as BERT and RoBERTa) are used for vectorization, converting the text into dense vectors in a high-dimensional vector space to capture the semantic information contained in the text. For multimodal data (such as damaged luggage images), multimodal pre-trained models (such as CLIP and ViT-BERT) are used to generate a joint representation of text and image vectors, forming a multimodal joint vector. All generated vectors are stored in a vector index library, which uses efficient indexing structures (such as FAISS and Annoy) to facilitate subsequent fast retrieval. It is important to note that an adversarial training mechanism is introduced during the construction of the vector knowledge base, using a GAN (Generative Adversarial Network) network to generate adversarial examples of damaged luggage images. Specifically, the generator network learns the distribution characteristics of real damaged luggage images and generates adversarial examples that are highly similar to real images but have subtle differences. The discriminator network is responsible for distinguishing between real and generated images. Through adversarial training, the model's feature extraction capabilities for complex and variable damaged luggage images are improved, enhancing the robustness and generalization capabilities of the vector knowledge base. The generated adversarial image data is manually reviewed and then mixed with real data and stored in the vector index library to enrich the sample diversity of the vector knowledge base.

[0065] Step S13 establishes an index association mechanism. Set the airline code primary key for the relational knowledge base. The airline code adopts the internationally accepted two-letter code (such as CA for Air China, MU for China Eastern Airlines, etc.) to ensure the uniqueness and standardization of the primary key. For the vector knowledge base, configure a dynamic shard index, divide the vector data into multiple shards according to the characteristic distribution of the data or dimensions such as timestamp, and independently establish an index for each shard to support parallel retrieval and improve retrieval efficiency. A cross-modal query channel for the hybrid retrieval knowledge base is established through primary key mapping. The specific implementation method is as follows: the airline code field is retained in both the relational knowledge base and the vector knowledge base. When a cross-modal query is required, the airline code primary key of the relational knowledge base is first used to locate the processing policy of a specific airline, and then the corresponding historical case vector data is retrieved in the vector knowledge base based on the airline code, to achieve associated retrieval of structured rules and unstructured cases. For example, when querying a specific airline (code CZ) about its lost baggage handling procedures, the system first retrieves the airline's rules for handling lost baggage from a relational knowledge base. Then, using the airline code CZ, it searches the vector knowledge base for similar historical case vectors to obtain handling records for similar cases, thereby generating a hybrid search result that includes both rules and cases. This index association mechanism breaks down the barriers between structured and unstructured data, enabling the system to leverage both the accuracy of rules and the empirical nature of cases, providing a solid foundation for subsequent hybrid searches.

[0066] Throughout the construction of the hybrid retrieval knowledge base, data accuracy and consistency must be ensured. The processing rules in the relational knowledge base are regularly compared with the latest policy documents released by each airline, and outdated or invalid rules are promptly updated. For historical case data in the vector knowledge base, regular data cleaning and deduplication are performed to remove duplicate records or invalid data to ensure the quality of the vector index library. At the same time, a data security protection mechanism is established to encrypt and store sensitive data (such as passenger personal information and undisclosed details of airline processing policies) and control access to prevent data leakage or misuse. Through these steps, a hybrid retrieval knowledge base with a clear structure, efficient indexing, and rich data is constructed, providing strong data support for the retrieval and processing of abnormal baggage cases.

[0067] Example 2: In the process of establishing a hybrid search rule engine in step S2, the specific implementation method is as follows:

[0068] Step S21 parses the input case description and generates keyword search expressions. The system receives a user-entered description of an abnormal baggage case. This description can be in text form (such as manually entered case details or customer service chat logs) or text content converted from speech. The input text is first segmented using a dictionary-based segmentation algorithm (such as the forward maximum matching algorithm or the reverse maximum matching algorithm) combined with a deep learning segmentation model (such as LSTM-CRF) to break the continuous text sequence into independent words or phrases. During the segmentation process, a custom domain dictionary (containing specialized terms such as airline codes, anomaly types, and item names) is used to improve segmentation accuracy. For example, "Air China" can be identified as "CA" (airline code) and "cracked luggage" as "damaged" (anomaly type). After segmentation, entity recognition is performed using a named entity recognition (NER) model (such as BERT-NER) to extract key entities from the segmentation results, including airline codes, anomaly types, and item feature fields. For example, if the input case description is "A passenger on China Eastern Airlines flight MU5102 reported lost checked luggage containing a laptop and some clothing," entity recognition can extract the airline code "MU," the exception type "lost," and the item characteristics "laptop, clothing." Based on the extracted entity information, a keyword search expression is generated in the format "airline code = MU AND exception type = lost AND item characteristics = laptop." This expression is used to retrieve corresponding processing rules in the relational knowledge base, such as China Eastern Airlines' procedures and compensation standards for lost luggage containing electronic items.

[0069] Execute step S22 to perform semantic vectorization on the unstructured text and search for similar cases. First, semantically encode the unstructured text portion of the case description (such as detailed events, passenger statements, etc.). The specific steps are as follows:

[0070] Step S221: Word segmentation and entity recognition are performed on the case description text. This process is similar to the word segmentation and entity recognition process in step S21, but focuses on extracting key semantic features, such as airline identifiers (e.g., "MU" for China Eastern Airlines), abnormal item categories (e.g., "electronics," "clothing"), and damage level characteristics (e.g., "minor damage," "severe deformation," "complete loss," etc.). Entity recognition converts the features of the natural language description into standardized labels that the system can recognize. For example, "broken luggage wheel" is converted to "damaged - damaged components" and "luggage not delivered for more than 72 hours" is converted to "delayed - timeout."

[0071] Step S222: Generate a text semantic vector using a pre-trained language model. This pre-trained language model, such as RoBERTa-wwm-ext-finetuned-airline, is fine-tuned on a large amount of aviation text data. This model is capable of better capturing the semantic associations between aviation terms. The processed text is input into the model, which outputs a hidden layer vector of fixed dimension as the text semantic vector. To reduce the vector dimension and improve computational efficiency, feature dimensionality reduction is performed using a multi-layer perceptron (MLP), mapping the high-dimensional vector to a query vector of a preset dimension (e.g., 768). For example, for a text describing "a luggage tie rod broke due to improper loading and unloading during transportation," the generated query vector must include vector representations of semantic features such as "improper loading and unloading" and "tie rod broken."

[0072] Step S223: Perform sharded parallel search in the vector index library. The vector index library utilizes a dynamic sharding mechanism, dividing vector data into multiple logical shards (e.g., by airline code, by anomaly type, etc.), with each shard assigned an independent search thread. Upon receiving a query vector, the system sends a search request to all shards. Each shard then concurrently calculates the similarity score between the query vector and the vectors within its own shard. This similarity calculation utilizes a weighted cosine similarity algorithm, assigning weights to different feature dimensions. For example, the anomaly type dimension is weighted more highly than the item color dimension to emphasize the impact of key features on similarity. Weights are determined based on domain expert experience or historical data statistics, such as a weight of 0.4 for the airline identifier, 0.3 for the anomaly item category, and 0.3 for the damage grade. After the calculations are complete for each shard, the top K cases with the highest similarity score are returned. The system then merges the results from all shards, sorting them from highest to lowest score, and selects the top N similar cases (N being a preset threshold, such as 50) as the vector search results. For example, if the weighted cosine similarity between the query vector and a historical case vector is 0.85 and ranks third, then the case is included in the candidate result set.

[0073] Then, step S23 is executed to merge the search results and generate a set of candidate processing solutions. The keyword search results of step S21 (processing rules from the relational knowledge base) are merged with the vector search results of step S22 (historical cases from the vector knowledge base). First, a data mapping relationship is established to associate the processing rules in the keyword search results with the historical cases in the vector search results according to the airline code and exception type. For example, the baggage loss rules of the same airline are associated with the loss cases. Then, conflict resolution is performed. When there is an inconsistency between the rules and the cases (such as the rules stipulate that the upper limit of compensation is 2,000 yuan, and the compensation amount of a certain case is 2,500 yuan), the following strategy is used to resolve it: give priority to the latest effective rules. If the case is a historical record before the rule update, the rules shall prevail; if the case is a special processing case after the rule update, the special circumstances in the case (such as the passenger has purchased luggage insurance, the value of the items in the luggage has been notarized, etc.) are extracted as supplementary explanations to the rules. After the conflict is resolved, the processing process, compensation standards, historical cases and other information are integrated to generate a set of candidate processing solutions. Each solution contains the following contents:

[0074] Airline information: such as airline name and code;

[0075] Abnormal type: such as loss, damage;

[0076] Processing rules: reference specific clauses in the relational knowledge base, such as Article X of the "XX Airlines Baggage Transportation Rules";

[0077] Similar cases: List the first M cases (M≤N) retrieved by the vector, including case number, processing results, and key feature comparison;

[0078] Compensation standards: Based on rules and cases, clarify the calculation method of compensation amount, time limit requirements, etc.;

[0079] Processing flow: The operating specifications of each link are displayed in the form of a timeline, such as "filing a case within 24 hours → internal verification within 48 hours → completion of compensation approval within 7 working days".

[0080] Throughout the process of establishing a hybrid search rule engine, retrieval efficiency and accuracy need to be optimized. For keyword retrieval, queries are accelerated by establishing database indexes (such as airline code indexes and exception type indexes). For vector retrieval, the vector index library is regularly updated, invalid case vectors are deleted, new case vectors are added, and the shard index is recalculated. At the same time, a manual intervention interface is provided. When the candidate solutions generated by the system do not meet actual needs, operators can manually adjust the search keywords, similarity thresholds, or conflict resolution strategies to ensure that the search results meet the business scenario. In addition, a logging mechanism is used to record the input parameters, search results, processing time, etc. of each search, facilitating subsequent system performance analysis and optimization. Through the above steps, the hybrid search rule engine can efficiently and accurately generate a set of candidate processing solutions containing multi-dimensional information, providing rich input material for subsequent domain knowledge enhancement and decision-making.

[0081] Example 3: In the process of developing a domain-enhanced decision-making system based on a large model in step S3, the specific implementation method is as follows:

[0082] First, step S31 is executed to enhance domain knowledge. Airlines' specific processing rules are extracted from the relational knowledge base. For example, one airline's compensation policy for international baggage delays exceeding 48 hours is "US$50 per day, with a maximum of US$300." Simultaneously, similar case handling records are retrieved from the vector knowledge base. For example, a historical case involving a 72-hour delay with urgently needed medication in the luggage resulted in "priority delivery + an additional 200 yuan in compensation." By combining rules with case studies, a multi-dimensional chain of evidence is formed, encompassing the basis for determining liability, compensation standards, and the handling process.

[0083] To integrate evidence from different sources, a rule weight matrix is established (step S311). A weight coefficient is assigned to each piece of evidence based on the airline policy priority (e.g., IATA standards > airline general rules > local branch rules) and the confidence level of the case (based on the historical accuracy of the case processing results). Suppose the evidence set is , the corresponding weight coefficient is ,satisfy .in, Indicates the pieces of evidence (such as an airline's rules or a historical case), It represents the credibility weight of the evidence, which is comprehensively assessed by domain experts based on the authority and timeliness of the evidence.

[0084] Then, the evidence fusion calculation is performed (step S312), using a Bayesian network model to integrate structured rules and unstructured case data. The nodes of the Bayesian network include anomaly types (such as loss, damage), responsible parties (such as carriers, passengers, third parties), compensation amount ranges, etc., and the edges represent the probabilistic dependencies between nodes. The conditional probability table of the Bayesian network is trained using historical data to calculate the confidence score of the comprehensive treatment recommendation. To process the confidence of the suggestion, the formula is: .

[0085] in, Indicated in evidence Support processing results The conditional probability of is derived through Bayesian network reasoning. This formula integrates the support for the treatment outcome from different pieces of evidence through a weighted summation method to generate a comprehensive confidence score, which is used to measure the reliability of the treatment recommendation.

[0086] A decision tree model is then constructed (step S313), transforming the fusion results into a tree structure. The root node of the decision tree represents the anomaly type (e.g., "Is the luggage damaged?"), intermediate nodes represent detailed features (e.g., "Is the damage greater than 50%?", "Are there obvious signs of external force?"), and leaf nodes represent the processing conclusions (e.g., "Compensate at full price" or "Compensate after repair"). Each path corresponds to a logical chain of decisions, for example: "Luggage damaged → Damage greater than 50% → Carrier's liability → Compensate at current market value."

[0087] Step S32 constructs a RAG reasoning framework (retrieval-augmented generation framework). The evidence chain and decision tree structure extracted in step S31 are converted into structured prompts in the format of "[case characteristics] + [evidence set] + [decision logic] + content to be generated." For example, the prompts might be "Case type: International flight luggage lost, airline code: CA, loss time: over 72 hours, evidence: CA airline rules Article X (compensation cap of 2,000 yuan), similar case Y (compensation of 1,800 yuan), please generate liability determination logic and compensation amount calculation basis." The prompts are input into a large language model (such as a fine-tuned LLaMA-2). Based on pre-trained knowledge and input prompts, the model generates processing suggestions, including liability determination logic (e.g., "According to the Warsaw Convention and CA airline rules, the carrier failed to prove it exercised reasonable care and must bear full liability") and compensation amount calculation basis (e.g., "Taking into account the 2,000 yuan limit in the rules and the 1,800 yuan in similar cases, and considering item depreciation, a compensation of 1,500 yuan is recommended").

[0088] Step S33 designs a visual output template. The proposed action is converted into a structured flowchart, using a swimlane diagram to differentiate the operational processes for different responsible parties (e.g., airline, passenger, airport). Key decision nodes are annotated (e.g., "Is the responsible party confirmed?", "Does the compensation amount exceed 5,000 yuan and require approval from a higher authority") and regulatory citations (e.g., Article 125 of the Civil Aviation Law). An editable action plan document is also generated, including a tabular comparison of compensation standards and case characteristics, to facilitate quick understanding and adjustment of the plan by the operator.

[0089] The domain-enhanced decision-making system includes a dynamic knowledge distillation module and a three-layer architecture: a basic rule layer stores standardized processing rules; a case-based reasoning layer stores feature vectors and processing logic for historical cases; and a meta-learning layer automatically adjusts training weights using a curriculum learning strategy. This strategy gradually increases training weights based on case complexity (e.g., simple loss, complex damage, multi-party disputes). This allows the system to first master basic rules before learning the processing logic for complex cases, improving the model's generalization and decision-making accuracy.

[0090] Through the above steps, the domain-enhanced decision-making system can generate structured processing suggestions based on multi-source evidence, and clearly present the processing process and compensation basis through a visual interface, providing intelligent support for automated decision-making in abnormal baggage cases.

[0091] Example 4: In the process of constructing the abnormal baggage handling system in step S4, the specific implementation method is as follows:

[0092] Execute step S41 to deploy the hybrid search engine module. This module must implement simultaneous query scheduling between a relational database (such as MySQL) and a vector index library (such as FAISS). Search efficiency is improved by configuring a multi-threaded concurrent controller. For example, consider an airline's baggage incident. A passenger reported damaged checked luggage on flight CA1203 through the customer service system, which contained photographic equipment. After receiving the case description, the multi-threaded concurrent controller launches two independent threads: one thread accesses the relational database and retrieves the corresponding processing rules based on the extracted airline code "CA" and the "damage" exception type. For example, Air China stipulates that "damage to checked luggage requires a damage assessment within 48 hours. If the cause is a mechanical failure, compensation will be based on actual losses, but not exceeding 3,000 yuan." The other thread accesses the vector index library, generates a query vector for the case description, and searches for similar cases. The resulting results return the top 50 most similar cases from 150 baggage damage cases handled by Air China in the past six months. Thirty of these cases involved photographic equipment, and the case details indicate that most compensation is based on the equipment repair invoice amount. Two threads execute in parallel, completing data retrieval in approximately 200 milliseconds, avoiding the potentially lengthy delays associated with traditional serial retrieval. The multi-threaded controller also coordinates inter-thread retrieval. For example, if a thread times out or returns abnormal data, it automatically triggers a retry mechanism or switches to a backup index node, ensuring the stability of the retrieval process.

[0093] Execute step S42 to develop a decision-making inference engine. This engine integrates a rules engine and a probabilistic reasoning model to verify the compliance of handling plans and conduct risk assessments. Taking the damaged baggage case on flight CA1203 as an example, the set of candidate handling plans obtained from the hybrid search engine includes three types of solutions: Plan 1, based on Air China's basic rules, provides compensation of 100 yuan per kilogram (assuming the baggage weighs 5 kilograms, compensation is 500 yuan); Plan 2, based on similar cases, compensates based on the cost of photographic equipment repairs (estimated to be 2,000 yuan); Plan 3, based on the latest industry standards, allows compensation based on the actual value of the equipment if the passenger can provide proof of value (the passenger provided a purchase invoice, valued at 5,000 yuan). The rules engine first verifies each solution for compliance with current regulations and airline policies. It finds that while Plan 1 complies with Air China's basic rules, it fails to consider the supplementary provisions regarding the declaration of valuables in the new industry regulations. Plan 3 requires confirmation that the passenger declared the photographic equipment at check-in. If not, it does not comply with the airline's policy that "undeclared valuables are compensated as ordinary baggage." The probabilistic reasoning model analyzes the execution risks of each plan. For example, Plan 2 relies on the authenticity of the repair invoice, which carries the risk of passengers falsely reporting expenses. Historical data indicates a 15% probability of this risk. Plan 3, on the other hand, could lead to subsequent legal disputes if passengers fail to declare their belongings, with a 25% probability of risk. The decision-making reasoning engine, combining compliance and risk assessment results, prioritizes Plan 2 and prompts operators to verify the authenticity of the repair invoice and whether passengers have declared their belongings in advance.

[0094] Execute step S43 to build the human-computer interaction layer. This layer features a multi-view visualization panel that supports comparative analysis of handling options, regulatory tracing, and electronic signature verification. In the case of flight CA1203, after logging into the system, the operator views basic case information (passenger name, flight number, and exception type) on the left side of the visualization panel. The center area displays a comparison table of candidate handling options, with the horizontal layout showing the option number, compensation amount, handling time limit, and reference rule / case, and the vertical layout showing key dimensions (such as liability determination, compensation standard, and risk level). For example, Option 1 provides 500 yuan in compensation with a time limit of 3 working days, based on Article 8 of the Air China Baggage Carriage Rules; Option 2 provides 2,000 yuan in compensation with a time limit of 5 working days, based on Case CA20230405 (damaged photographic equipment is compensated based on repair costs); Option 3 provides 5,000 yuan in compensation with a time limit of 7 working days, based on Article 37 of the Civil Aviation Transport Regulations. By clicking the "Reference Rule / Case" link for any option, the operator can jump to the regulatory detail page or case archive page to view the specific content and effective date. The right side of the panel provides a risk assessment chart, which displays the scores of each plan in terms of compliance, implementation cost, passenger satisfaction and other dimensions in the form of a radar chart.

[0095] During the solution confirmation phase, the operator pushes a detailed description of Solution 2 to the passenger via a visual panel, and the passenger confirms the solution online using the electronic signature function. The system automatically generates an electronic confirmation letter, which includes the solution content, the rights and obligations of both parties, and the method of compensation payment. The electronic signature uses blockchain technology for evidence storage to ensure that it cannot be tampered with. If the passenger disagrees with the solution, they can initiate a negotiation process through the panel, and the operator will readjust the search parameters or initiate a manual review mechanism. For example, if a passenger claims that the photographic equipment is a limited edition and the repair cost exceeds 2,000 yuan, the operator can add search keywords such as "limited edition item" and "repair cost exceeds expectations" to the vector index library, re-search similar cases, and obtain solutions that include the handling of special items.

[0096] During system integration, the hybrid search engine module, the decision-making and reasoning engine module, and the human-computer interaction layer interact with each other through APIs. For example, after receiving a case description input by the user, the human-computer interaction layer calls the hybrid search engine's search interface through an API to obtain a set of candidate solutions. It then calls the decision-making and reasoning engine's evaluation interface through an API to obtain the solution's compliance and risk scores. Finally, the results are returned to the human-computer interaction layer via an API for display. Each module is deployed on a cloud server using a microservices architecture that supports dynamic scalability and load balancing, ensuring system stability in high-concurrency scenarios.

[0097] The system also features log management, recording the entire processing process for each case, including search parameters, candidate solution generation time, decision-making and reasoning process, operator approval records, and passenger confirmation time. Log data is stored on a separate log server and can be queried by case number, processing time, operator, and other dimensions, facilitating subsequent audits and process optimization. For example, through log analysis, managers may discover that the average processing time for a certain type of abnormal baggage case is 48 hours, exceeding the industry standard of 24 hours. This allows them to identify bottlenecks in search parameter settings or decision-making processes and adjust system configurations to improve efficiency.

[0098] Through the above steps, the abnormal baggage handling system achieves full automation, from case information entry, hybrid search, decision-making reasoning, to solution confirmation. Each module has a clear division of labor and close collaboration, ensuring efficient, accurate, and traceable handling. Operators no longer need to manually review extensive policy documents and historical cases. The system allows them to quickly obtain comprehensive, rule-based and case-based handling recommendations. Visualization tools also provide a clear understanding of the key nodes and basis for case handling, significantly improving the standardization and intelligence of abnormal baggage handling.

[0099] 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," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0100] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for handling abnormal baggage based on vector matching, characterized in that: The following steps are involved: Step S1: Constructing a hybrid retrieval knowledge base: Integrate multi-source heterogeneous data on abnormal baggage case data to generate a relational knowledge base and a vector knowledge base; index and associate the relational knowledge base and the vector knowledge base to establish a hybrid retrieval knowledge base; Step S2: Establishing a hybrid search rule engine: Perform keyword parsing and semantic vectorization on the input abnormal baggage case description to generate a hybrid search query instruction; perform parallel search based on the hybrid search knowledge base to generate a set of candidate processing solutions; Step S3: Develop a domain-enhanced decision-making system based on a large model: enhance the candidate treatment solution set with domain knowledge and generate structured treatment recommendations; Output processing flow specifications and compensation basis through a visual interactive interface; Step S4: Build an abnormal baggage handling system: Integrate a hybrid search rule engine, a domain-enhanced decision-making system, and a visual interactive interface to automate the entire abnormal baggage case process from search to decision-making; Step S2 includes the following steps: Step S21: Parse the input case description, extract the airline code, exception type, and item feature fields, and generate a keyword search expression; Step S22: semantically encode the unstructured text to generate a query vector; calculate the cosine similarity between the query vector and the vectors in the vector knowledge base, and select the top N similar cases by ranking by score; Step S23: Merge the keyword search results and the vector search results, perform conflict resolution, and generate a candidate solution set including processing procedures, compensation standards, and historical cases; Step S22 includes the following steps: Step S221: Perform word segmentation and entity recognition on the case description text to extract the airline identifier, abnormal item category, and damage level features; Step S222: Generate a text semantic vector using a pre-trained language model, perform feature dimensionality reduction using a multi-layer perceptron, and generate a fixed-dimensional query vector; Step S223: Performing shard parallel search in the vector index library, using the weighted cosine similarity algorithm to calculate the similarity score, merging the shard results and taking the top N similar cases; Step S3 includes the following steps: Step S31: Perform domain knowledge enhancement: extract airline-specific processing rules from the relational knowledge base, retrieve similar case processing records from the vector knowledge base, and form a multi-dimensional evidence chain; Step S32: Constructing a RAG reasoning framework: Converting search results into structured prompt words, inputting them into a large language model to generate processing suggestions, including responsibility determination logic and compensation amount calculation basis; Step S33: Design a visual output template: convert the treatment suggestions into a structured flowchart, mark the key decision nodes and regulatory reference sources, and generate an editable disposal plan document.

2. The abnormal baggage handling method based on vector matching according to claim 1 is characterized in that: Step S1 includes the following steps: Step S11: Build a relational knowledge base: collect abnormal baggage handling policy documents from different airlines, perform document structure processing using a localized parsing tool, and generate a triplet dataset of airline-abnormality type-handling rule; Step S12: Constructing a vector knowledge base: Collect historical abnormal baggage case data, perform missing value cleaning and field standardization, generate text vectors and multimodal joint vectors using a pre-trained language model, and store them in a vector index library; Step S13: Establish an index association mechanism: set the airline code primary key for the relational knowledge base, configure a dynamic sharding index for the vector knowledge base, and establish a cross-modal query channel for the hybrid retrieval knowledge base through primary key mapping.

3. The abnormal baggage handling method based on vector matching according to claim 1 is characterized in that: Step S31 includes the following steps: Step S311: Establish a rule weight matrix: assign weight coefficients to evidence from different sources based on the airline's policy priority and the confidence level of the case; Step S312: Execute evidence fusion calculation: Use the Bayesian network model to integrate structured rules and unstructured case data to generate a comprehensive treatment recommendation confidence score; Step S313: Construct a decision tree model: convert the fusion result into a decision tree structure that includes abnormality type judgment, applicable compensation standards, and processing flow path.

4. The abnormal baggage handling method based on vector matching according to claim 1 is characterized in that: Step S4 includes the following steps: Step S41: deploying a hybrid search engine module: configuring a multi-threaded concurrent controller to implement synchronous query scheduling of the relational database and the vector index library; Step S42: Develop a decision-making reasoning engine: Integrate a rule engine and a probabilistic reasoning model to support compliance verification and risk assessment of treatment solutions; Step S43: Build the human-computer interaction layer: Design a multi-view visualization panel to support comparative analysis of treatment plans, regulatory clause tracing, and electronic signature confirmation functions.

5. The abnormal baggage handling method based on vector matching according to claim 2 is characterized in that: In step S12, an adversarial training mechanism is introduced into the construction of the vector knowledge base, and adversarial baggage damage image data is generated through the GAN network.

6. The abnormal baggage handling method based on vector matching according to claim 1 is characterized in that: The domain-enhanced decision-making system in step S3 includes a dynamic knowledge distillation module: a three-layer architecture consisting of a basic rule layer, a case-based reasoning layer, and a meta-learning layer is established, and the training weight of complex cases is gradually increased through a curriculum learning strategy.

7. An abnormal baggage handling system based on vector matching, characterized in that: include: A hybrid retrieval knowledge base construction module is used to integrate multi-source heterogeneous data on abnormal baggage cases, generate a relational knowledge base and a vector knowledge base, and associate the two through indexing to form a hybrid retrieval knowledge base; The hybrid search rule engine module is used to perform keyword parsing and semantic vectorization on the input abnormal baggage case description to generate a hybrid search query instruction, perform parallel search based on the hybrid search knowledge base, and output a set of candidate processing solutions. Specifically, it includes: Parse the input case description, extract the airline code, exception type, and item feature fields, and generate keyword search expressions; Semantically encode unstructured text to generate a query vector; calculate the cosine similarity between the query vector and the vectors in the vector knowledge base, and select the top N similar cases by ranking by score; Merge keyword search results and vector search results, perform conflict resolution, and generate a set of candidate solutions that include processing procedures, compensation standards, and historical cases; A domain enhancement decision module integrates a large language model to enhance the candidate treatment solution set with domain knowledge to generate structured treatment suggestions, and outputs treatment process specifications and compensation basis through a visual interactive interface; An integration module, configured to logically integrate the hybrid retrieval knowledge base construction module, the hybrid retrieval rule engine module, and the domain enhancement decision module, to achieve full automation of the entire process from retrieval to decision-making for abnormal baggage cases; Step S22 includes the following steps: Step S221: Perform word segmentation and entity recognition on the case description text to extract the airline identifier, abnormal item category, and damage level features; Step S222: Generate a text semantic vector using a pre-trained language model, perform feature dimensionality reduction using a multi-layer perceptron, and generate a fixed-dimensional query vector; Step S223: Performing shard parallel search in the vector index library, using the weighted cosine similarity algorithm to calculate the similarity score, merging the shard results and taking the top N similar cases; Step S3 includes the following steps: Step S31: Perform domain knowledge enhancement: extract airline-specific processing rules from the relational knowledge base, retrieve similar case processing records from the vector knowledge base, and form a multi-dimensional evidence chain; Step S32: Constructing a RAG reasoning framework: Converting search results into structured prompt words, inputting them into a large language model to generate processing suggestions, including responsibility determination logic and compensation amount calculation basis; Step S33: Design a visual output template: convert the treatment suggestions into a structured flowchart, mark the key decision nodes and regulatory reference sources, and generate an editable disposal plan document.

Images