Scenic spot ticket inventory and intelligent question and answer linkage management method and system

Abstract

The invention discloses a scenic spot ticket inventory and intelligent question and answer linkage management method and system, and relates to the technical field of data processing. Receiving a user query statement, dividing the statement into a fragment set with a semantic identifier, matching a preset training semantic fragment to obtain a candidate fragment, generating a semantic decoding sequence after hierarchical analysis, and constructing thinking chain guide information; combining the information with the target fragment, extracting key elements through a large language model, generating a verification path set and a ticket scheme set through a business rule network, and outputting the verification path set and the ticket scheme set; and in response to user selection, executing inventory linkage operation. According to the method, semantic analysis is simplified by means of thinking chain guide information, interpretability is enhanced, and ticket buying elements are accurately extracted; coherent response from inquiry to ticket business scheme and inventory operation is realized through a whole-process closed loop, consultation and inventory information islands are broken, a ticket buying link is optimized, real-time and accurate scheduling is guaranteed, and service accuracy is improved.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically relating to a management method and system for linking scenic spot ticket inventory with intelligent question answering. Background Technology

[0002] With the advancement of digital transformation in the cultural tourism industry, scenic spot ticketing management has been upgraded from the traditional manual ticketing and verification model to an intelligent model dominated by smart ticketing management systems. By integrating resources from multiple channels such as OTA platforms, proprietary mini-programs, and offline windows, it has achieved real-time order synchronization, dynamic inventory management, and rapid verification through multiple methods. Some systems also incorporate big data analysis functions, which can provide real-time statistics on order sales, cancellations, and tourist profiles. At the same time, it supports the integration of ticketing with various business scenarios such as accommodation and catering. Combined with automated equipment such as self-service ticketing machines and smart gates, it has significantly improved the operational efficiency of scenic spots, reduced labor costs, and promoted the transformation of scenic spots from dependence on traffic to value creation.

[0003] However, existing scenic spot ticketing systems have significant shortcomings in parsing user statements. Their intelligent question-and-answer modules can only perform shallow parsing of general statements such as scenic spot overview and opening hours. They cannot break down user queries into precise segments with semantic tags, nor do they have the ability to analyze the hierarchical structure of statements. This makes it difficult to understand the internal logic of semantic structure, resulting in high semantic parsing complexity and poor interpretability. Consequently, existing systems cannot accurately extract key elements related to ticket purchase from user statements, nor can they link the statement parsing results with ticket inventory management. This means that users' statements involving core ticketing needs such as remaining tickets and reservation slots do not receive accurate responses, creating information silos between consultation and inventory management, thus leading to low accuracy in ticketing service responses. Summary of the Invention

[0004] The purpose of this invention is to solve the problem that existing scenic spot ticketing systems lack the ability to parse user statements and cannot effectively link the parsing results with ticket inventory management, resulting in low accuracy of ticketing service response. Therefore, this invention proposes a management method and system that links scenic spot ticket inventory with intelligent question answering.

[0005] In a first aspect of this invention, a management method for linking scenic spot ticket inventory with intelligent question answering is first proposed, the method comprising: Receive query statements input by the user; The query statement is divided into multiple semantic segments to obtain a semantic segment set; the semantic segments in the semantic segment set include semantic type identifiers and their corresponding text segments; Based on the target semantic segment, a matching search is performed on a preset training semantic segment set, and the semantic segment with the highest similarity to the target semantic segment is selected to obtain candidate semantic segments; the target semantic segment is any one of the semantic segments in the set of semantic segments; The candidate semantic segments are parsed hierarchically to obtain the semantic decoding sequence; The thought chain guidance information is constructed based on the semantic decoding sequence. The thought chain guidance information is a semantic parsing guidance method that decomposes the query statement into local semantic fragments and inputs the local semantic fragments into a preset large language model for processing according to priority. This guidance information is used to reduce the complexity of semantic parsing and enable the preset large language model to focus on the recognition and combination of local semantic fragments. The target semantic fragment and the thought chain guidance information are input into a preset large language model to decode the semantic fragment and obtain the key elements; A key element set is obtained by statistically analyzing the key elements of all semantic segments. Based on the key element set, multiple initial verification paths are generated in a preset business rule network to obtain an initial verification path set. Generate a ticketing scheme set based on the initial verification path set; The ticketing scheme set is output to the user interface, responding to the user's selection instruction for a specific alternative ticketing scheme, and performing inventory linkage operation based on the verification path corresponding to the scheme.

[0006] This solution breaks down user queries into semantically tagged fragments, matches candidate fragments, and hierarchically parses them to construct a thought chain guiding information. This reduces the complexity of semantic parsing, enhances the interpretability of semantic structure, and can accurately extract key elements with the help of a large language model. Combined with a business rule network, it generates an appropriate ticketing solution and links it with inventory, thus efficiently achieving intelligent and accurate response to users' ticketing queries.

[0007] Optionally, based on the target semantic segment, a matching search is performed on a preset training semantic segment set, and candidate semantic segments are obtained by filtering the semantic segments with the highest similarity to the target semantic segment, including: Extract semantic type identifiers from the target semantic segment to obtain a semantic type identifier set; Based on the semantic type identifier set, all training semantic segments containing the same semantic type identifier are selected from the preset training semantic segment set to obtain the initial training semantic segment set; Extract the semantic type identifiers of the target initial training semantic fragments to obtain an initial semantic type identifier set; the target initial training semantic fragment is any one of the initial training semantic fragments in the initial training semantic fragment set; Calculate the Jaccard similarity between the semantic type identifier set and the initial semantic type identifier set to obtain a first similarity value; A second similarity value is obtained by calculating the semantic similarity between the text segment corresponding to the target semantic segment and the training text segment corresponding to the target initial training semantic segment; A weighted average of the first similarity value and the second similarity value is obtained to obtain a comprehensive similarity value. The overall similarity value between all initial training semantic segments and the target semantic segment is calculated, and the training semantic segment with the highest overall similarity value is selected as the candidate semantic segment.

[0008] This approach first filters training segments based on semantic type, then calculates similarity across dimensions, and finally selects the optimal weighted average. The steps progress from coarse screening to fine evaluation, leveraging semantic type identifiers to quickly narrow down the matching range and improve retrieval efficiency, while using dual similarity weighted evaluation to ensure candidate segments more accurately match user intent. This avoids redundancy from invalid matches and ensures the accuracy of subsequent semantic parsing, laying a solid foundation for accurately extracting key elements for ticket purchase.

[0009] Optionally, constructing thought chain guidance information based on the semantic decoding sequence includes: A semantic tree is constructed based on the semantic decoding sequence. Non-terminal nodes of the semantic tree are extracted as initial semantic segments. After filtering out redundant segments in the candidate semantic segments, a hierarchical semantic sequence is obtained by performing hierarchical traversal parsing on the filtered initial semantic segments through a preset target traversal algorithm. Based on the hierarchical relationship of the semantic tree, the semantic type identifier, the association relationship of the text segments, and the traversal order of each initially selected semantic segment in the hierarchical semantic sequence are identified to obtain the sequence structure feature information; Based on structural feature information, the semantic decoding sequence is decomposed into ordered steps to obtain an ordered step set; The ordered set of steps is combined according to the traversal order in the sequence structure feature information to obtain the thought chain guidance information.

[0010] This solution connects semantic tree construction, fragment processing, and hierarchical parsing to transform semantic information from scattered fragments into a structured sequence. Then, through feature recognition and step decomposition, the sequence logic is transformed into ordered guiding information. This achieves both a progressive and logically coherent information processing approach and ensures that the semantic analysis results of the preceding steps accurately support the construction of the guiding information in the following steps, avoiding information gaps or redundancy. This makes the entire semantic processing flow more coherent and efficient, while also providing a precisely adapted semantic foundation for the generation of subsequent business solutions.

[0011] Optionally, the target semantic fragment and the thought chain guidance information are input into a preset large language model for semantic fragment decoding to obtain a key element set, including: Based on the thought chain guidance information, the hierarchical association rules and priorities in the semantic decoding sequence corresponding to the target semantic segment are parsed to obtain semantic parsing information; The parsing order of the target semantic segment is determined based on the semantic parsing information, and the parsing task of the target semantic segment is decomposed into multiple sequentially executed semantic parsing subtasks to obtain a subtask set based on the parsing order; The target semantic segment is tokenized to obtain an encoded sequence; Obtain the contextual semantic information corresponding to the target semantic fragment; In the target subtask, a candidate token set is obtained by filtering valid tokens that conform to the semantic rules of the corresponding level of the subtask from the encoded sequence based on the context semantic information; the target subtask is any one of the subtasks in the set. For the candidate token set, hierarchical semantic information is obtained by semantic combination and relation construction in the order that the core semantics take precedence over the subordinate semantics; Key elements are obtained by extracting key information and removing redundant information from the hierarchical semantic information of all subtasks.

[0012] This solution guides the hierarchical rule parsing of semantic decoding sequences through a thought chain, then uses the parsed information to connect task decomposition, and task decomposition to connect token encoding and context information acquisition. Subsequently, sub-tasks serve as carriers to connect effective token filtering and semantic combination. Each link is interconnected, ensuring that the preceding semantic parsing logic can accurately guide the subsequent task decomposition and token processing. Furthermore, through semantic combination within sub-tasks and information integration across sub-tasks, it achieves a coherent implementation of semantic parsing from rules to execution, and from encoding to information extraction. This avoids information gaps between steps and allows semantic processing to more efficiently extract key elements through orderly decomposition and precise filtering.

[0013] Optionally, generating a ticketing scheme set based on the initial verification path set includes: A verification state matrix is ​​constructed based on the set of verification paths; the rows of the verification state matrix correspond to the initial verification paths, and the columns correspond to the verification nodes. For each verification node in the verification status matrix, the corresponding inventory status data is obtained in real time, and the inventory status data is filled into the corresponding position in the matrix to obtain the inventory status matrix. The feasibility score set is obtained by calculating the feasibility score of each initial verification path in the initial verification path set based on the inventory status matrix. The initial verification paths with a feasibility score greater than a preset threshold in the feasibility score set are selected as valid verification paths to obtain a set of valid verification paths. For each valid verification path in the set of valid verification paths, a ticketing scheme is generated based on the constraints of each verification node and real-time inventory data, ultimately resulting in a set of ticketing schemes.

[0014] This solution first integrates path and node information in a matrix manner, making the correlation of inventory data more intuitive and orderly. Then, it filters out paths that meet the constraints through feasibility scoring, which not only ensures the rationality of the solution but also avoids redundant calculations of invalid paths. Finally, based on the effective paths and real-time inventory, a solution is generated, which not only makes the ticketing solution fit the actual inventory status but also improves the efficiency and accuracy of solution generation, helping users quickly obtain suitable ticketing options.

[0015] In a second aspect of this invention, a management system linking scenic spot ticket inventory with intelligent question answering is proposed, comprising: The query receiving module is used to receive query statements input by the user; The segmentation module is used to divide the query statement into multiple semantic segments to obtain a semantic segment set; the semantic segments in the semantic segment set include semantic type identifiers and their corresponding text segments; The filtering module is used to perform matching and retrieval on a preset training semantic segment set based on the target semantic segment, and filter the semantic segment with the highest similarity to the target semantic segment to obtain candidate semantic segments; the target semantic segment is any semantic segment in the semantic segment set; The hierarchical parsing module is used to perform hierarchical traversal parsing on the candidate semantic segments to obtain the semantic decoding sequence; The thought chain information construction module is used to construct thought chain guidance information based on the semantic decoding sequence. The thought chain guidance information is a semantic parsing guidance method that decomposes the query statement into local semantic fragments and inputs the local semantic fragments into a preset large language model for processing according to priority. This guidance information is used to reduce the complexity of semantic parsing and enable the preset large language model to focus on the recognition and combination of local semantic fragments. The decoding module is used to input the target semantic fragment and the thought chain guidance information into a preset large language model to decode the semantic fragment and obtain key elements. The statistics module is used to count the key elements of all semantic segments to obtain a key element set, and to generate multiple initial verification paths in the preset business rule network based on the key element set to obtain an initial verification path set. The scheme generation module is used to generate a ticketing scheme set based on the initial verification path set; The inventory linkage module is used to output the ticketing scheme set to the user interface, respond to the user's selection instruction for a specific alternative ticketing scheme, and perform inventory linkage operation based on the verification path corresponding to the scheme.

[0016] Optionally, the filtering module includes: The first extraction module is used to extract semantic type identifiers from the target semantic segment to obtain a semantic type identifier set; The first filtering module is used to filter all training semantic segments containing the same semantic type identifier from the preset training semantic segment set according to the semantic type identifier set to obtain an initial training semantic segment set; The second extraction module is used to extract the semantic type identifier of the target initial training semantic segment to obtain an initial semantic type identifier set; the target initial training semantic segment is any one of the initial training semantic segments in the initial training semantic segment set; The first calculation module is used to calculate the Jaccard similarity between the semantic type identifier set and the initial semantic type identifier set to obtain a first similarity value; The second calculation module is used to calculate the semantic similarity between the text segment corresponding to the target semantic segment and the training text segment corresponding to the target initial training semantic segment to obtain a second similarity value; The weighting module is used to weight the first similarity value and the second similarity value to obtain a comprehensive similarity value; The candidate semantic segment generation module is used to calculate the comprehensive similarity value between all initial training semantic segments and the target semantic segment, and select the training semantic segment with the highest comprehensive similarity value as the candidate semantic segment.

[0017] Optionally, the thought chain information construction module includes: The hierarchical semantic generation module is used to construct a semantic tree based on the semantic decoding sequence, extract non-terminal nodes of the semantic tree as initial semantic fragments, filter redundant fragments in the candidate semantic fragments, and then perform hierarchical traversal parsing on the filtered initial semantic fragments through a preset target traversal algorithm to obtain a hierarchical semantic sequence. The identification module is used to identify the semantic type identifier, text segment association relationship and traversal order of each initially selected semantic segment in the hierarchical semantic sequence according to the hierarchical relationship of the semantic tree to obtain sequence structure feature information; The step decomposition module is used to decompose the semantic decoding sequence into ordered steps based on structural feature information to obtain an ordered step set; The combination module is used to combine the ordered set of steps according to the traversal order in the sequence structure feature information to obtain the thought chain guidance information.

[0018] Optionally, the decoding module includes: The semantic information generation module is used to parse the hierarchical association rules and priorities in the semantic decoding sequence corresponding to the target semantic segment according to the thought chain guidance information to obtain semantic parsing information; The subtask generation module is used to determine the parsing order of the target semantic segment based on the semantic parsing information, and to decompose the parsing task of the target semantic segment into multiple sequentially executed semantic parsing subtasks to obtain a subtask set based on the parsing order. The encoding module is used to perform token encoding on the target semantic segment to obtain an encoded sequence; The information acquisition module is used to acquire the contextual semantic information corresponding to the target semantic fragment; The second filtering module is used to filter valid tokens that conform to the corresponding hierarchical semantic rules of the corresponding subtask from the encoded sequence based on the context semantic information to obtain a candidate token set; the target subtask is any one of the subtasks in the set. The semantic information generation module is used to perform semantic combination and relationship construction on the candidate token set in the order that the core semantics are superior to the subordinate semantics to obtain hierarchical semantic information; The key element generation module is used to extract key information and remove redundant information from the hierarchical semantic information of all subtasks to obtain key elements.

[0019] Optionally, the scheme generation module includes: A matrix construction module is used to construct a verification state matrix based on the verification path set; the rows of the verification state matrix correspond to the initial verification path, and the columns correspond to the verification nodes. The inventory status matrix generation module is used to obtain the corresponding inventory status data in real time for each verification node in the verification status matrix, and fill the inventory status data into the corresponding position of the matrix to obtain the inventory status matrix. The scoring module is used to calculate the feasibility score of each initial verification path in the initial verification path set based on the inventory status matrix to obtain a feasibility score set; the feasibility score is calculated based on the proportion of verification nodes that satisfy constraints and the key constraint satisfaction status in the path. The effective path filtering module is used to select the initial verification paths with a feasibility score greater than a preset threshold from the feasibility score set as effective verification paths to obtain an effective verification path set. The ticketing scheme generation module is used to generate a ticketing scheme for each valid verification path in the valid verification path set, based on the constraints of each verification node and real-time inventory data, and finally obtain a ticketing scheme set.

[0020] The beneficial effects of this invention are as follows: This invention proposes a management method that links scenic spot ticket inventory with intelligent question answering. First, it breaks down user queries into semantically labeled fragments. After matching candidate fragments, it performs hierarchical parsing to obtain a semantic decoding sequence, and then constructs a thought chain guiding information based on this. Next, it combines this information with the target fragment, extracts key elements through a large language model, generates a ticketing solution set through a business rule network, and finally responds to the user's selection and links inventory operations. This method reduces the complexity of semantic parsing and enhances the interpretability of semantic structures through thought chain guiding information, helping to accurately extract key elements for ticket purchase. It also achieves a precise and coherent response from user query to ticketing solutions and inventory operations through a closed-loop collaborative process, breaking down information silos between consultation and inventory, optimizing the user's ticket purchase process, ensuring real-time accuracy of ticket scheduling, and significantly improving service precision. Attached Figure Description

[0021] The present invention will now be further described with reference to the accompanying drawings.

[0022] Figure 1 A flowchart illustrating a management method for linking scenic spot ticket inventory with intelligent question answering, provided as an embodiment of the present invention; Figure 2 A flowchart for screening candidate semantic segments provided in an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0024] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] This invention provides a management method that links scenic area ticket inventory with intelligent question-and-answer functionality. See also... Figure 1 , Figure 1 A flowchart illustrating a management method for scenic area ticket inventory linked with intelligent question answering, provided as an embodiment of the present invention. The method includes the following steps: S101, Receive the query statement input by the user; S102, the query statement is divided into multiple semantic segments to obtain a semantic segment set; the semantic segments in the semantic segment set include semantic type identifiers and their corresponding text segments; S103, perform matching and retrieval on the preset training semantic segment set according to the target semantic segment, and select the semantic segment with the highest similarity to the target semantic segment to obtain candidate semantic segments; the target semantic segment is any semantic segment in the semantic segment set; S104, perform hierarchical traversal parsing on candidate semantic segments to obtain semantic decoding sequences; S105, construct thought chain guidance information based on semantic decoding sequence; S106, Input the target semantic fragment and thought chain guidance information into the preset large language model to decode the semantic fragment and obtain the key elements; S107, count the key elements of all semantic segments to obtain a key element set, and generate multiple initial verification paths in the preset business rule network based on the key element set to obtain an initial verification path set. S108, Generate a ticketing scheme set based on the initial verification path set; S109, output the ticketing scheme set to the user interface, respond to the user's selection instruction for a specific alternative ticketing scheme, and perform inventory linkage operation based on the verification path corresponding to the scheme; The semantic fragments in the semantic fragment set include semantic type identifiers and their corresponding text fragments; The target semantic fragment is any one semantic fragment in the set of semantic fragments; The thought chain guidance information is a semantic parsing guidance method that breaks down the query statement into local semantic fragments and inputs the local semantic fragments into a preset large language model for processing according to priority. This guidance information is used to reduce the complexity of semantic parsing and enable the preset large language model to focus on the identification and combination of local semantic fragments.

[0026] This invention provides a method for managing scenic area ticket inventory in conjunction with intelligent question-and-answer interaction. The method first breaks down user queries into semantically identifiable segments, matches candidate segments, and performs hierarchical parsing to obtain a semantic decoding sequence. Then, it constructs a thought chain guidance information based on this sequence. Next, it combines this guidance information with the target semantic segments, extracts key elements using a large language model, generates a set of ticketing solutions using a business rule network, and finally responds to the user's solution selection while simultaneously executing inventory linkage operations. This method simplifies the semantic parsing process through thought chain guidance information, enhances the understandability of the semantic structure, and helps accurately capture core elements related to ticket purchase. With a closed-loop collaborative model throughout the entire process, it achieves a precise and coherent response from user-initiated queries to ticketing solution output and dynamic inventory adjustment. This breaks down information barriers between consultation and inventory management, optimizes the user's ticket purchase process, ensures the real-time performance and accuracy of ticketing scheduling, and effectively improves service precision.

[0027] In one implementation, the preset business rule network is a relational network that transforms ticketing rules into something a computer can understand and execute. This network consists of many rule points and the connecting lines between them. Each rule point is responsible for checking a specific ticketing condition, and the connecting lines indicate the sequential relationship and logical dependency between the rule points. For example, when a user requests a weekend park visit for two adults and one child, the system will find several different checking routes in this network. The first route will pass through the weekend ticket price rule point, the family discount rule point, and the child free ticket rule point in sequence. The second route will pass through the individual ticket price rule point and the adult and child separate pricing rule point. The third route may pass through the group discount rule point but will stop at the number of people verification point. Each complete checking route is an initial verification path, which represents a possible ticketing solution that is formed after a series of rule checks starting from the user's needs.

[0028] In one implementation, the query statement segmentation process involves inputting the query statement into a pre-trained semantic encoding model for token-level encoding to obtain a statement encoding vector containing contextual semantic information. Based on a span parsing mechanism, all possible consecutive token fragments in the query statement are traversed. Each token fragment is semantically matched and scored using a pre-defined semantic type system, which includes intent-type semantic types and slot-type semantic types. The score is calculated by a feedforward network using the span embedding features and semantic type features of the token fragment. Valid token fragments with scores higher than a pre-defined threshold and no partial overlap are selected. Each valid token fragment corresponds to a unique semantic type, forming a semantic fragment set composed of semantic type-text fragment tuples. All semantic fragments in the semantic fragment set together completely cover the core semantic information of the query statement and meet the non-redundant and non-omission structured requirements of hierarchical semantic parsing.

[0029] In one implementation, hierarchical traversal parsing involves first extracting the semantic type and corresponding text fragment of each candidate semantic segment, determining the master-slave relationship between each candidate semantic segment according to preset semantic hierarchy rules, constructing a semantic hierarchy tree with the core semantic type as the root node and subordinate semantic types as child nodes, and using a depth-first search algorithm to traverse the semantic hierarchy tree layer by layer starting from the root node, parsing the semantic connotation of the candidate semantic segment corresponding to each node and its association logic with other nodes in turn, and recording the parsing priority, semantic type affiliation, and text fragment correspondence of each candidate semantic segment in the traversal order to form an ordered and complete semantic decoding sequence that reflects the semantic hierarchy structure.

[0030] In one implementation, the inventory linkage operation process is as follows: when the system receives a user's selection instruction for a specific ticketing scheme, it does not directly call the predefined inventory interface. Instead, it first parses the complete verification path corresponding to the scheme. This path records each constraint verification node from the initial requirement to the final scheme and the real-time data snapshot used during verification. The system then traces back along this verification path, extracting the precise inventory data identifiers that each verified node in the path depends on, such as the specific ticket number, the inventory pool identifier for a specific time period, and the discount rule identifier used to calculate the price. These discrete data identifiers are then compiled into an atomic, transactional inventory operation chain according to the business logic sequence. This operation chain not only includes the final inventory deduction instruction but also embeds pre-check instructions for related locked inventory, rollback contingency plans for parallel operation conflicts, and state synchronization instructions after successful operation. This ensures the accuracy and data consistency of inventory linkage in multi-user concurrent scenarios, realizing an intelligent transformation from a flexible verification path to a rigid execution instruction.

[0031] In one embodiment, see Figure 2 , Figure 2 This is a flowchart of the candidate semantic segment selection process provided in this embodiment of the invention. The process involves matching and retrieving a preset training semantic segment set based on the target semantic segment, and selecting the semantic segment with the highest similarity to the target semantic segment to obtain candidate semantic segments, including: S201, Extract semantic type identifiers from the target semantic segment to obtain a semantic type identifier set; S202, based on the semantic type identifier set, select all training semantic segments containing the same semantic type identifier from the preset training semantic segment set to obtain the initial training semantic segment set; S203, extract the semantic type identifiers of the initial training semantic segments of the target to obtain the initial semantic type identifier set; S204, calculate the Jaccard similarity between the semantic type identifier set and the initial semantic type identifier set to obtain the first similarity value; S205, calculate the semantic similarity between the text segment corresponding to the target semantic segment and the training text segment corresponding to the target initial training semantic segment to obtain the second similarity value; S206, The first similarity value and the second similarity value are weighted to obtain the comprehensive similarity value; S207, Calculate the comprehensive similarity value between all initial training semantic segments and target semantic segments, and select the training semantic segment with the highest comprehensive similarity value as the candidate semantic segment; The target initial training semantic segment is any one of the initial training semantic segments in the initial training semantic segment set.

[0032] In one implementation, the process of obtaining the preset training semantic fragment set involves first collecting multi-domain training corpora corresponding to the hierarchical semantic parsing task, mapping each training statement in the corpus to an original hierarchical semantic representation containing nested structures, and then, based on the semantic framework rules corresponding to predefined logical tokens, decomposing the original hierarchical semantic representation into multiple non-terminal nodes. Each non-terminal node is a training semantic fragment, and each training semantic fragment explicitly contains a semantic type identifier and a corresponding text fragment. After deduplication and standardization, all training semantic fragments are classified and stored according to the semantic type identifier, ultimately forming a pre-set training semantic fragment set with a well-organized structure that can be directly used for retrieval and matching.

[0033] In one implementation, the process of selecting from a preset training semantic fragment set involves extracting the core semantic category and hierarchical attributes corresponding to each semantic type identifier based on a set of semantic type identifiers. Unlike existing technologies that rely solely on text surface similarity or single keyword matching, this implementation uses precise matching of semantic type identifiers as the core selection criterion. It iterates through the training semantic fragment set to retrieve all training semantic fragments that are completely consistent with the semantic type identifier. Simultaneously, a semantic association verification mechanism is used. This mechanism is based on the historical association data between the text fragments of the training semantic fragments and the corresponding semantic type identifiers. It excludes redundant fragments that only have the same semantic type identifier but whose text fragment semantic connotations are unrelated. This ensures that the selected training semantic fragments not only accurately correspond to the target semantic type identifier but also completely cover the core semantic scenario under that semantic type identifier. Finally, the selected fragments are aggregated to form an initial training semantic fragment set that is non-redundant, highly relevant, and can support subsequent hierarchical parsing.

[0034] In one implementation, the semantic similarity calculation process involves extracting the text fragments corresponding to the target semantic fragment and the training text fragments corresponding to the target initial training semantic fragment. The two texts are then input into a semantic encoding model fine-tuned by a hierarchical semantic parsing task for token-level encoding, generating high-dimensional semantic vectors containing contextual semantic information and hierarchical association features. The cosine similarity algorithm is used to calculate the cosine value of the angle between the two semantic vectors. Simultaneously, a weighted fusion is performed by combining the overlap of the core semantic concepts of the two texts. The core semantic concepts are extracted from the text fragments based on semantic framework rules. Finally, a second similarity value is obtained that can comprehensively reflect the surface matching degree and deep semantic association degree of the text.

[0035] In one embodiment, constructing thought chain guidance information based on the semantic decoding sequence includes: A semantic tree is constructed based on the semantic decoding sequence. Non-terminal nodes of the semantic tree are extracted as initial semantic segments. After filtering out redundant segments in the candidate semantic segments, the filtered initial semantic segments are traversed and parsed using a preset target traversal algorithm to obtain a hierarchical semantic sequence. Based on the hierarchical relationship of the semantic tree, the semantic type identifiers of each initially selected semantic segment in the hierarchical semantic sequence, the association relationship of the text segments, and the traversal order are identified to obtain the sequence structure feature information; Based on structural feature information, the semantic decoding sequence is decomposed into ordered steps to obtain an ordered step set; The ordered set of steps is combined according to the traversal order in the sequence structure feature information to obtain the thought chain guidance information.

[0036] In one implementation, the preset target traversal algorithm follows the logic of traversing from the core semantic nodes to the subordinate semantic nodes according to the hierarchical relationship of the semantic tree. Through the node access function built into the traversal algorithm, the semantic tree nodes corresponding to the candidate semantic fragments are read one by one, and the hierarchical affiliation of each node and the relationship between nodes are recorded simultaneously. Finally, the ordered set of nodes output according to this traversal logic is determined as the semantic decoding sequence.

[0037] In one implementation, taking the user query "My 68-year-old father, 10-year-old child, and I plan to go on an in-depth tour of Zhangjiajie National Forest Park next Wednesday to Friday. I have a veteran's preferential treatment card and want to choose a package ticket including cable car. Which ticket should we buy?" as an example, the implementation process of this method is as follows: First, construct a semantic tree based on the original hierarchical semantic representation of the query. Extract the non-terminal nodes of the semantic tree as candidate semantic fragments, including event fragments of ticket recommendations, scenic spot location fragments of Zhangjiajie National Forest Park, time interval fragments from next Wednesday to Friday, in-depth tour type fragments, fragments of accompanying persons for 2 adults and 1 child, fragments of veteran's preferential status, and fragments of package ticket requirements including cable car. After filtering redundant content, parse these fragments using a depth-first traversal algorithm to obtain a semantic decoding sequence, the content of which is event fragments of ticket recommendations, scenic spot location fragments of Zhangjiajie National Forest Park, time interval fragments from next Wednesday to Friday, in-depth tour type fragments, fragments of accompanying persons for 2 adults and 1 child, fragments of veteran's preferential status, and fragments of package ticket requirements including cable car. Based on the hierarchical relationship of the semantic tree, the semantic type identifier, text association relationship, and traversal order of each segment in the sequence are identified to obtain sequence structure feature information. This information records the semantic type of each segment as ticket recommendation, scenic spot location, time interval, type of tour, accompanying persons, discount status, and package ticket demand. The corresponding text is: ticket purchase suggestion, Zhangjiajie National Forest Park, next Wednesday to Friday, in-depth tour, 2 adults and 1 child, veterans, including cable car. The traversal order is: ticket recommendation segment first, scenic spot location segment second, time interval segment in the middle, type of tour segment next, accompanying persons segment next, discount status segment last. Afterwards, the package ticket demand segment ends; then, based on this feature information, the semantic decoding sequence is decomposed into an ordered set of steps, the content of which is: Step 1 corresponds to the ticket purchase suggestion segment of the ticket recommendation type, Step 2 corresponds to the Zhangjiajie National Forest Park segment of the scenic spot location type, Step 3 corresponds to the next Wednesday to Friday segment of the time interval type, Step 4 corresponds to the in-depth tour segment of the play type, Step 5 corresponds to the 2 adults and 1 child segment of the accompanying persons type, Step 6 corresponds to the veteran segment of the preferential status type, and Step 7 corresponds to the cable car segment of the package ticket demand type; finally, the set of steps is combined according to the above traversal order to obtain the thought chain guiding information; The guiding information in the thought chain is as follows: Step 1, semantic type identifier: ticket recommendation, text fragment: ticket purchase suggestion; Step 2, semantic type identifier: scenic spot location, text fragment: Zhangjiajie National Forest Park; Step 3, semantic type identifier: time range, text fragment: next Wednesday to Friday; Step 4, semantic type identifier: type of tour, text fragment: in-depth tour; Step 5, semantic type identifier: accompanying persons, text fragment: 2 adults and 1 child; Step 6, semantic type identifier: preferential status, text fragment: veteran; Step 7, semantic type identifier: package ticket requirement, text fragment: includes cable car. This information will then be combined with candidate semantic fragments, demonstration samples, and query sentences.

[0038] In one embodiment, inputting the target semantic fragment and thought chain guidance information into a pre-defined large language model for semantic fragment decoding to obtain a set of key elements includes: Based on the guiding information of the thought chain, the hierarchical association rules and priorities in the semantic decoding sequence corresponding to the target semantic fragment are parsed to obtain semantic parsing information; The parsing order of the target semantic segment is determined based on the semantic parsing information. Based on the parsing order, the parsing task of the target semantic segment is decomposed into multiple sequentially executed semantic parsing subtasks to obtain a subtask set. The target semantic segment is tokenized to obtain an encoded sequence; Obtain the contextual semantic information corresponding to the target semantic fragment; In the target subtask, a candidate token set is obtained by filtering valid tokens that conform to the semantic rules of the corresponding level of the subtask from the encoded sequence based on contextual semantic information; the target subtask is any one of the subtasks in the set. For the candidate token set, hierarchical semantic information is obtained by semantic combination and relation construction in the order that the core semantics take precedence over the subordinate semantics. Key elements are obtained by extracting key information and removing redundant information from the hierarchical semantic information of all subtasks.

[0039] In one implementation, the contextual semantic information is the text preceding and following the target semantic fragment in the original text; according to the parsing order rules, it is determined whether the semantic type corresponding to the current subtask is core semantic or subordinate semantic; if it is core semantic, the core semantic elements are extracted and fixed; if it is subordinate semantic, the corresponding subordinate semantic elements are associated and extracted based on the fixed core semantic elements.

[0040] In one implementation, the pre-defined large language model works by first guiding the information reception of the semantic decoding sequence through a thought chain with hierarchical association rules and parsing priority guidance. This simulates the human multi-step reasoning process, decomposing the complex semantic parsing task into ordered intermediate reasoning steps. Then, the input target semantic fragment is semantically encoded at the token level. Combining the contextual semantic information and the parsing logic in the thought chain guidance information, a constraint decoding mechanism is used to filter valid tokens that conform to the semantic hierarchy rules at each generation step. The core connotation of the target semantic fragment is deduced step by step in the order of core semantics followed by subordinate semantics. Key information corresponding to each semantic type is extracted and redundant and repetitive content is removed. Finally, structured key elements are output.

[0041] In one embodiment, generating a ticketing scheme set based on an initial verification path set includes: Construct a verification state matrix based on the verification path set; the rows of the verification state matrix correspond to the initial verification path, and the columns correspond to the verification nodes; For each verification node in the verification status matrix, the corresponding inventory status data is obtained in real time, and the inventory status data is filled into the corresponding position in the matrix to obtain the inventory status matrix. The feasibility score set is obtained by calculating the feasibility score of each initial verification path in the initial verification path set based on the inventory status matrix; the feasibility score is calculated based on the proportion of verification nodes that satisfy the constraints in the path and the key constraint satisfaction status. The effective verification path set is obtained by selecting the initial verification path whose feasibility score is greater than a preset threshold from the feasibility score set; For each valid verification path in the valid verification path set, a ticketing scheme is generated based on the constraints of each verification node and real-time inventory data, ultimately resulting in a ticketing scheme set.

[0042] In one implementation, the preset threshold is set by a technician and can be 0.8.

[0043] In one implementation, an example of generating a ticketing solution set is a user inquiring about ticket purchases for a family of four during the National Day holiday. Based on the user's needs, the system generates three verification paths for ticket purchase strategies: the first path corresponds to a family package; the second path corresponds to a combination of adult and child tickets; and the third path corresponds to group discount tickets. The system establishes a three-row, four-column verification table, where each row corresponds to a ticket purchase strategy, and each column corresponds to a condition that needs to be checked: the first column checks if there are enough tickets; the second column checks if the usage time is suitable; the third column checks if the price rules are met; and the fourth column checks if the purchaser's identity matches the criteria. The system queries the current ticket inventory in real time and fills the data into a table. For example, the first path shows no tickets in the first column and meets the requirements in the other columns. The second path shows meet the requirements in all columns. The third path shows the requirements in the fourth column. The system calculates a score based on the compliance status of each path in the table. The second path has the highest score, so the system selects the second path and generates a detailed ticket purchase plan for two adult tickets, one child ticket, and a child ticket, based on the requirements of this path, such as the number of adult tickets, the number of child tickets, and the available time period during the National Day holiday. The system also reminds the user of a detailed ticket purchase plan for a child who is exempt from the ticket.

[0044] In one implementation, the verification state matrix construction process is as follows: First, after inputting the set of key user intent elements into the business rule network, instead of pre-generating a complete verification path, a verification initial node is dynamically activated for each key element, and all initial nodes constitute a verification seed set. Second, based on real-time inventory data and rule dependencies, graph traversal is performed in parallel for each verification seed node. During the traversal, its dependent nodes are dynamically expanded and real-time verification data is populated, forming multiple interconnected local verification subgraphs. Finally, the system maps each subgraph to a column vector of the verification state matrix and constructs a complete verification state matrix by dynamically analyzing the overlap and conflict relationships between subgraphs. This method, by dynamically constructing dependency graphs and mapping them to a matrix structure in real time during the verification process, achieves parallel generation of verification paths and early detection of conflicts, breaking through the efficiency bottleneck of traditional linear verification path construction methods in complex rule scenarios.

[0045] In one implementation, through Calculate the feasibility score of the initial verification path; where, N represents the number of verification nodes that have passed constraint verification in the initial verification path, and N is the total number of verification nodes included in the initial verification path. This is the quantified value of the key constraint verification result, which is achieved when all key constraint verification nodes in the initial verification path pass verification. Use the first preset value; otherwise, use the second preset value. and These are the weighting coefficients. >0, >0, It can be 0.3. It can be 0.7, the first preset value is 1, and the second preset value is 0. This zero-to-one jump performance effectively achieves a veto effect: once a critical constraint is not satisfied, regardless of how high the satisfaction rate of ordinary constraints is, the total score will be reduced due to missing constraints. This results in a large portion of the score being very low, thus ensuring that the path is effectively filtered out.

[0046] The foregoing has described one embodiment of the present invention in detail, but this content is merely a preferred embodiment and should not be considered as limiting the scope of the present invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the scope of the claims of this invention.

Claims

1. A management method for scenic area ticket inventory linked with intelligent question answering, characterized in that, The method includes: Receive query statements input by the user; The query statement is divided into multiple semantic segments to obtain a semantic segment set; the semantic segments in the semantic segment set include semantic type identifiers and their corresponding text segments; Based on the target semantic segment, a matching search is performed on a preset training semantic segment set, and the semantic segment with the highest similarity to the target semantic segment is selected to obtain candidate semantic segments; the target semantic segment is any one of the semantic segments in the set of semantic segments; The candidate semantic segments are parsed hierarchically to obtain the semantic decoding sequence; The thought chain guidance information is constructed based on the semantic decoding sequence. The thought chain guidance information is a semantic parsing guidance method that decomposes the query statement into local semantic fragments and inputs the local semantic fragments into a preset large language model for processing according to priority. This guidance information is used to reduce the complexity of semantic parsing and enable the preset large language model to focus on the recognition and combination of local semantic fragments. The target semantic fragment and the thought chain guidance information are input into a preset large language model to decode the semantic fragment and obtain the key elements; A key element set is obtained by statistically analyzing the key elements of all semantic segments. Based on the key element set, multiple initial verification paths are generated in a preset business rule network to obtain an initial verification path set. A ticketing scheme set is generated based on the initial verification path set; The ticketing scheme set is output to the user interface, responding to the user's selection instruction for a specific alternative ticketing scheme, and performing inventory linkage operation based on the verification path corresponding to the scheme.

2. The management method for scenic spot ticket inventory linked with intelligent question answering as described in claim 1, characterized in that, Based on the target semantic segment, a matching search is performed on a preset training semantic segment set, and the semantic segment with the highest similarity to the target semantic segment is selected to obtain candidate semantic segments, including: Extract semantic type identifiers from the target semantic segment to obtain a semantic type identifier set; Based on the semantic type identifier set, all training semantic segments containing the same semantic type identifier are selected from the preset training semantic segment set to obtain the initial training semantic segment set; Extract the semantic type identifiers of the target initial training semantic fragments to obtain an initial semantic type identifier set; the target initial training semantic fragment is any one of the initial training semantic fragments in the initial training semantic fragment set; Calculate the Jaccard similarity between the semantic type identifier set and the initial semantic type identifier set to obtain a first similarity value; A second similarity value is obtained by calculating the semantic similarity between the text segment corresponding to the target semantic segment and the training text segment corresponding to the target initial training semantic segment; A weighted average of the first similarity value and the second similarity value is obtained to obtain a comprehensive similarity value. The overall similarity value between all initial training semantic segments and the target semantic segment is calculated, and the training semantic segment with the highest overall similarity value is selected as the candidate semantic segment.

3. The management method for scenic spot ticket inventory linked with intelligent question answering as described in claim 1, characterized in that, The thought chain guidance information constructed based on the semantic decoding sequence includes: A semantic tree is constructed based on the semantic decoding sequence. Non-terminal nodes of the semantic tree are extracted as initial semantic segments. After filtering out redundant segments in the candidate semantic segments, a hierarchical semantic sequence is obtained by performing hierarchical traversal parsing on the filtered initial semantic segments through a preset target traversal algorithm. Based on the hierarchical relationship of the semantic tree, the semantic type identifier, the association relationship of the text segments, and the traversal order of each initially selected semantic segment in the hierarchical semantic sequence are identified to obtain the sequence structure feature information; Based on structural feature information, the semantic decoding sequence is decomposed into ordered steps to obtain an ordered step set; The ordered set of steps is combined according to the traversal order in the sequence structure feature information to obtain the thought chain guidance information.

4. The management method for scenic spot ticket inventory linked with intelligent question answering as described in claim 1, characterized in that, The target semantic fragment and the thought chain guidance information are input into a preset large language model to decode the semantic fragment and obtain a set of key elements, including: Based on the thought chain guidance information, the hierarchical association rules and priorities in the semantic decoding sequence corresponding to the target semantic segment are parsed to obtain semantic parsing information; The parsing order of the target semantic segment is determined based on the semantic parsing information, and the parsing task of the target semantic segment is decomposed into multiple sequentially executed semantic parsing subtasks to obtain a subtask set based on the parsing order; The target semantic segment is tokenized to obtain an encoded sequence; Obtain the contextual semantic information corresponding to the target semantic fragment; In the target subtask, a candidate token set is obtained by filtering valid tokens that conform to the semantic rules of the corresponding level of the subtask from the encoded sequence based on the context semantic information; the target subtask is any one of the subtasks in the set. For the candidate token set, hierarchical semantic information is obtained by semantic combination and relation construction in the order that the core semantics take precedence over the subordinate semantics; Key elements are obtained by extracting key information and removing redundant information from the hierarchical semantic information of all subtasks.

5. The management method for scenic spot ticket inventory linked with intelligent question answering as described in claim 1, characterized in that, The ticketing scheme set generated based on the initial verification path set includes: A verification state matrix is ​​constructed based on the set of verification paths; the rows of the verification state matrix correspond to the initial verification paths, and the columns correspond to the verification nodes. For each verification node in the verification status matrix, the corresponding inventory status data is obtained in real time, and the inventory status data is filled into the corresponding position in the matrix to obtain the inventory status matrix. The feasibility score set is obtained by calculating the feasibility score of each initial verification path in the initial verification path set based on the inventory status matrix. The initial verification paths with a feasibility score greater than a preset threshold in the feasibility score set are selected as valid verification paths to obtain a set of valid verification paths. For each valid verification path in the set of valid verification paths, a ticketing scheme is generated based on the constraints of each verification node and real-time inventory data, ultimately resulting in a set of ticketing schemes.

6. A management system that links scenic area ticket inventory with intelligent question-and-answer functionality, characterized in that: The system includes: The query receiving module is used to receive query statements input by the user; The segmentation module is used to divide the query statement into multiple semantic segments to obtain a semantic segment set; the semantic segments in the semantic segment set include semantic type identifiers and their corresponding text segments; The filtering module is used to perform matching and retrieval on a preset training semantic segment set based on the target semantic segment, and filter the semantic segment with the highest similarity to the target semantic segment to obtain candidate semantic segments; the target semantic segment is any semantic segment in the semantic segment set; The hierarchical parsing module is used to perform hierarchical traversal parsing on the candidate semantic segments to obtain the semantic decoding sequence; The thought chain information construction module is used to construct thought chain guidance information based on the semantic decoding sequence. The thought chain guidance information is a semantic parsing guidance method that decomposes the query statement into local semantic fragments and inputs the local semantic fragments into a preset large language model for processing according to priority. This guidance information is used to reduce the complexity of semantic parsing and enable the preset large language model to focus on the recognition and combination of local semantic fragments. The decoding module is used to input the target semantic fragment and the thought chain guidance information into a preset large language model to decode the semantic fragment and obtain key elements. The statistics module is used to count the key elements of all semantic segments to obtain a key element set, and to generate multiple initial verification paths in the preset business rule network based on the key element set to obtain an initial verification path set. The scheme generation module is used to generate a ticketing scheme set based on the initial verification path set; The inventory linkage module is used to output the ticketing scheme set to the user interface, respond to the user's selection instruction for a specific alternative ticketing scheme, and perform inventory linkage operation based on the verification path corresponding to the scheme.

7. The management system for scenic area ticket inventory and intelligent question-and-answer linkage according to claim 6, characterized in that, The filtering module includes: The first extraction module is used to extract semantic type identifiers from the target semantic segment to obtain a semantic type identifier set; The first filtering module is used to filter all training semantic segments containing the same semantic type identifier from the preset training semantic segment set according to the semantic type identifier set to obtain an initial training semantic segment set; The second extraction module is used to extract the semantic type identifier of the target initial training semantic segment to obtain an initial semantic type identifier set; the target initial training semantic segment is any one of the initial training semantic segments in the initial training semantic segment set; The first calculation module is used to calculate the Jaccard similarity between the semantic type identifier set and the initial semantic type identifier set to obtain a first similarity value; The second calculation module is used to calculate the semantic similarity between the text segment corresponding to the target semantic segment and the training text segment corresponding to the target initial training semantic segment to obtain a second similarity value; The weighting module is used to weight the first similarity value and the second similarity value to obtain a comprehensive similarity value; The candidate semantic segment generation module is used to calculate the comprehensive similarity value between all initial training semantic segments and the target semantic segment, and select the training semantic segment with the highest comprehensive similarity value as the candidate semantic segment.

8. The management system for scenic area ticket inventory and intelligent question-and-answer linkage according to claim 6, characterized in that, The thought chain information construction module includes: The hierarchical semantic generation module is used to construct a semantic tree based on the semantic decoding sequence, extract non-terminal nodes of the semantic tree as initial semantic fragments, filter redundant fragments in the candidate semantic fragments, and then perform hierarchical traversal parsing on the filtered initial semantic fragments through a preset target traversal algorithm to obtain a hierarchical semantic sequence. The identification module is used to identify the semantic type identifier, text segment association relationship and traversal order of each initially selected semantic segment in the hierarchical semantic sequence according to the hierarchical relationship of the semantic tree to obtain sequence structure feature information; The step decomposition module is used to decompose the semantic decoding sequence into ordered steps based on structural feature information to obtain an ordered step set; The combination module is used to combine the ordered set of steps according to the traversal order in the sequence structure feature information to obtain the thought chain guidance information.

9. A management system for scenic area ticket inventory and intelligent question-and-answer linkage according to claim 6, characterized in that, The decoding module includes: The semantic information generation module is used to parse the hierarchical association rules and priorities in the semantic decoding sequence corresponding to the target semantic segment according to the thought chain guidance information to obtain semantic parsing information; The subtask generation module is used to determine the parsing order of the target semantic segment based on the semantic parsing information, and to decompose the parsing task of the target semantic segment into multiple sequentially executed semantic parsing subtasks to obtain a subtask set based on the parsing order. The encoding module is used to perform token encoding on the target semantic segment to obtain an encoded sequence; The information acquisition module is used to acquire the contextual semantic information corresponding to the target semantic fragment; The second filtering module is used to filter valid tokens that conform to the corresponding hierarchical semantic rules of the corresponding subtask from the encoded sequence based on the context semantic information to obtain a candidate token set; the target subtask is any one of the subtasks in the set. The semantic information generation module is used to perform semantic combination and relationship construction on the candidate token set in the order that the core semantics are superior to the subordinate semantics to obtain hierarchical semantic information; The key element generation module is used to extract key information and remove redundant information from the hierarchical semantic information of all subtasks to obtain key elements.

10. A management system for linking scenic area ticket inventory with intelligent question answering as described in claim 6, characterized in that, The scheme generation module includes: A matrix construction module is used to construct a verification state matrix based on the verification path set; the rows of the verification state matrix correspond to the initial verification path, and the columns correspond to the verification nodes. The inventory status matrix generation module is used to obtain the corresponding inventory status data in real time for each verification node in the verification status matrix, and fill the inventory status data into the corresponding position of the matrix to obtain the inventory status matrix. The scoring module is used to calculate the feasibility score of each initial verification path in the initial verification path set based on the inventory status matrix to obtain a feasibility score set; the feasibility score is calculated based on the proportion of verification nodes that satisfy constraints and the key constraint satisfaction status in the path. The effective path filtering module is used to select the initial verification paths with a feasibility score greater than a preset threshold from the feasibility score set as effective verification paths to obtain an effective verification path set. The ticketing scheme generation module is used to generate a ticketing scheme for each valid verification path in the valid verification path set, based on the constraints of each verification node and real-time inventory data, and finally obtain a ticketing scheme set.

Images