Intelligent question and answer method for text travel based on knowledge graph

Abstract

The invention discloses a knowledge graph-based intelligent question and answer method for text travel. The method comprises the following steps of: obtaining and preprocessing related data in a text travel service system; constructing a structured knowledge graph in the text travel field; constructing a candidate entity pair and executing relation extraction and weight updating guided by a behavior sequence; dividing and generating a behavior preference sub-graph according to a behavior mode, and extracting a text travel knowledge slice set according to a theme; natural language questions of tourists are received, and path search and semantic aggregation with behavior weights and path structures as constraints are executed; and generating a text travel intelligent question-answer result through a question-answer generation engine, and recording a text travel question-answer interaction log for iterative updating. According to the method, knowledge graph modeling and behavior guiding relation extraction technologies are fused, dynamic reasoning and personalized question and answer generation of the travel knowledge are achieved, and the method has the advantages of being accurate in semantic understanding, high in answer association degree and continuous in self-adaptive evolution.

Description

Technical Field

[0001] This invention relates to the field of cultural and tourism services, and in particular to a knowledge graph-based intelligent question-answering method for cultural and tourism services. Background Technology

[0002] Existing intelligent question-answering systems for the cultural and tourism industry are mostly based on keyword retrieval, template matching, or fixed question-answer databases. Some systems have introduced knowledge graphs in the tourism field, extracting information such as attractions, hotels, and routes into entities and relationships to achieve structured queries for simple factual questions. However, these technologies largely rely on string matching or shallow semantic similarity, making it difficult to fully understand the colloquial expressions of tourists in natural language questions, multi-turn contextual relationships, and implicit constraints such as time, travel mode, and demographic preferences. This results in limited relevance of answers and insufficient support for complex itinerary planning and cross-attraction combination recommendations.

[0003] With the widespread adoption of online booking platforms, smart scenic area systems, and mobile navigation applications, a large amount of order records, behavior logs, and operational data have accumulated in cultural tourism business systems. Some research has begun to attempt to combine knowledge graphs with recommendation technologies, but these typically remain at the level of simple ranking based on ratings or click counts, lacking in-depth modeling of tourist behavior patterns. Existing knowledge graph question answering systems generally do not uniformly encode and dynamically link tourist behavior sequences with entity relationships in the graph, making it impossible to construct a cultural tourism knowledge structure that reflects behavioral weights and temporal evolution characteristics. Furthermore, it is difficult to perform constrained path reasoning based on behavioral preference subgraphs and knowledge slices, thus exhibiting significant shortcomings in personalized question answering, adaptive updates, and reasoning capabilities in highly complex scenarios.

[0004] Therefore, how to provide a knowledge graph-based intelligent question-answering method for cultural tourism is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a knowledge graph-based intelligent question-answering method for cultural tourism. This invention integrates knowledge graph modeling and behavior-guided relationship extraction technologies to achieve dynamic reasoning and personalized question-answer generation of cultural tourism knowledge. It has the advantages of accurate semantic understanding, high relevance of answers, and continuous adaptive evolution.

[0006] A knowledge graph-based intelligent question-answering method for cultural tourism according to an embodiment of the present invention includes the following steps: Acquire relevant data from the cultural and tourism business system and form a set of cultural and tourism knowledge elements through preprocessing; Based on the set of cultural and tourism knowledge elements, entity modeling and relationship modeling are performed to construct a structured knowledge graph in the cultural and tourism field; Collect tourist behavior sequences to construct candidate entity pairs, combine cultural and tourism texts to guide the relationship extraction and weight update of the candidate entity pairs' execution behavior sequences, and write them into a structured knowledge graph of the cultural and tourism domain, and adjust the edge weights and time records to obtain a structured knowledge graph of the cultural and tourism domain containing behavior weights; Based on a structured knowledge graph in the cultural and tourism field that includes behavioral weights, behavioral preference subgraphs are generated according to behavioral patterns, and cultural and tourism knowledge slice sets are extracted by topic within the behavioral preference subgraphs. The system receives natural language questions from tourists, determines the target behavior preference subgraph and the target cultural and tourism knowledge slice set through semantic understanding and intent recognition, and performs path search and semantic aggregation constrained by behavior weight and path structure to obtain the target candidate cultural and tourism knowledge fragment set. The system inputs tourists' natural language questions and a set of target candidate cultural and tourism knowledge fragments into the question-and-answer generation engine to generate intelligent cultural and tourism question-and-answer results. The results are output in text or voice form, and the interactive log of cultural and tourism question-and-answer is recorded for iterative updates.

[0007] Optionally, the generation of the cultural and tourism knowledge element set specifically includes: Relevant data were obtained from the cultural and tourism business system and categorized according to their source to obtain the original cultural and tourism data set; The time field and spatial location information in the original cultural and tourism data set are standardized by time reference and spatial coordinate, resulting in a cultural and tourism spatiotemporally aligned data set. Based on the data structure characteristics, the cultural and tourism spatiotemporally aligned data set is divided into a structured cultural and tourism data subset, a cultural and tourism text subset, and a tourist behavior log subset. Missing value imputation, outlier removal and duplicate record merging, noise symbol cleaning and encoding unification, abnormal behavior filtering and invalid event removal are performed respectively to obtain the cleaned structured cultural and tourism data subset, cultural and tourism text subset, and tourist behavior log subset. The cleaned cultural and tourism text subset is subjected to sentence segmentation, word segmentation, part-of-speech tagging and preliminary named entity screening. Phrases involving scenic spot names, route descriptions, modes of transportation, accommodation types, catering types, activity names and policy clauses are marked as cultural and tourism text elements, generating a set of cultural and tourism text elements with semantic tags. The cleaned structured cultural and tourism data subset and the cultural and tourism spatiotemporally aligned data set are standardized and semantically aligned. Based on the set of field mapping rules, the ticket price field, opening time field, schedule field, room type field, menu field, activity time field and policy effective time field in different data sources are uniformly mapped to the preset standardized field space to obtain a standardized set of cultural and tourism data elements. The cleaned subset of visitor behavior logs is aggregated according to visitor identifier and time sequence, and search behavior, click behavior, browsing behavior, collection behavior and order placement behavior are arranged into a set of visitor behavior event sequences in time sequence; The set of cultural and tourism text elements with semantic tags, the set of standardized and structured cultural and tourism data elements, and the set of tourist behavior event sequences are uniformly encapsulated. For each cultural and tourism knowledge element, category attributes, semantic content attributes, time attributes, and spatial attributes are constructed simultaneously to form a set of cultural and tourism knowledge elements.

[0008] Optionally, the generation of the structured knowledge graph in the cultural and tourism field specifically includes: Based on the set of cultural and tourism knowledge elements, the category attributes of each cultural and tourism knowledge element are traversed, and cultural and tourism knowledge elements whose category attributes belong to the same category are divided into the same initial entity candidate set, forming multiple initial entity candidate sets, which are defined as entity sets. Extract the semantic content attributes of each cultural and tourism knowledge element from the set of cultural and tourism knowledge elements, input them into a preset semantic encoding function to obtain the corresponding semantic vector representation, and obtain a set of entities with semantic vector representation; Co-occurrence statistics are performed on cultural and tourism text elements that co-occur with the same cultural and tourism text element. Structural association analysis is also performed on fields with foreign key association, primary key association, and enumeration relationship in standardized structured cultural and tourism data elements, and candidate relationship pairs are formed by pairwise combination. For each candidate relationship pair, a set of candidate relationship types is determined based on the combination of the entity categories of the two entities, forming a set of relationship types that includes multiple cultural and tourism relationship types; For each candidate relation pair, combining the semantic vector representations of the two entities and the attribute features extracted from standardized structured cultural and tourism data elements, the relation scoring function is used to calculate the relation score value of the candidate relation pair under each relation type. The relationship score is filtered by a threshold. Candidate relationship pairs that are greater than the preset relationship threshold are determined as valid relationship edges under that relationship type. They are recorded as a combination of the starting entity, relationship type and ending entity in the form of a triple, forming a relationship set. Based on entity sets and relation sets, a structured knowledge graph for the cultural tourism field is constructed, with entity sets as the set of nodes in the graph structure and relation sets as the set of edges in the graph structure.

[0009] Optionally, the generation of the structured knowledge graph in the cultural and tourism field, which includes behavioral weights, specifically includes: Based on the set of tourist behavior event sequences, tourist behavior events are aggregated according to tourist identifier and time order, and search behavior, click behavior, browsing behavior, collection behavior and order placement behavior are arranged in time order to form a set of tourist behavior sequences; The set of tourist behavior sequences is vectorized and mapped to the corresponding tourist behavior representation. Clustering operation is used to obtain a set of behavior prototypes representing different cultural and tourism behavior patterns. The behavior prototypes are then mapped to the behavior patterns of parent-child preferences, food preferences, cultural preferences, and photography preferences. The similarity between the tourist behavior sequence and each behavior prototype in the behavior prototype set is calculated. The behavior prototypes are then weighted and fused according to the similarity weight to obtain the weighted behavior representation of the behavior prototype. This weighted behavior representation is then matched with the entity set in the structured knowledge graph of the cultural and tourism field to select the behavior interest entity set related to the behavior prototype weighted behavior representation. Finally, the relationship set in the structured knowledge graph of the cultural and tourism field is combined with the entity pair combination in the behavior interest entity set to form a candidate entity pair set for guiding the behavior sequence. For each candidate entity pair in the candidate entity pair set, extract the cultural and tourism text containing the candidate entity pair from the cultural and tourism text subset, obtain the entity context semantic vector and the cultural and tourism text semantic vector through a preset semantic encoding function, and combine the cultural and tourism text semantic vector with the behavior prototype weighted behavior representation to obtain the cultural and tourism text semantic representation that integrates behavior information. During the behavior guidance relationship extraction training phase, the behavior prototype weighted behavior representation in the real sample is replaced with the behavior prototype weighted behavior representation generated by other tourists' behavior sequences. Inconsistent comparison samples are constructed, and the relationship scoring function is called to calculate the behavior guidance relationship score. Based on this, the behavior consistency loss is obtained, so that the behavior guidance relationship score of the real sample is greater than the behavior guidance relationship score of the inconsistent sample. The behavioral consistency loss and the basic relation extraction loss are weighted and combined, and an adaptive weighting coefficient for uncertainty is introduced. The weight of the behavioral consistency loss in the total loss is dynamically adjusted according to the uncertainty of relation prediction, resulting in a total loss that combines basic relation extraction and behavioral consistency. Based on the loss value, behavioral-guided relation extraction training is completed. The trained relation scoring function is used to calculate the behavior guidance relation score of the candidate entity pair set under each cultural and tourism relation type. Threshold filtering and normalization processing are performed on the behavior guidance relation score to obtain the relation type selection result under each cultural and tourism relation type and the edge weight update amount corresponding to the relation type. The edge weight update is accumulated and added to the edge weight of the corresponding relation edge in the structured knowledge graph of the cultural and tourism field. At the same time, the time label of the relation edge is updated according to the current time, so as to obtain a structured knowledge graph of the cultural and tourism field containing behavioral weights.

[0010] Optionally, the generation of the cultural and tourism knowledge slice set specifically includes: Based on a structured knowledge graph in the cultural and tourism field that includes behavioral weights, the entity set, the relationship set, and the behavioral weights and time tags of each relationship edge are read. Each behavioral prototype in the behavioral prototype set is mapped to a behavioral pattern of parent-child preference, food preference, culture preference, and photography preference, thus forming a behavioral pattern set. For each behavior pattern in the behavior pattern set, the relation set is traversed in the structured knowledge graph of the cultural tourism field containing behavior weights. The behavior relevance of the entities connected to the relation edge is accumulated according to the behavior weight of the relation edge in the behavior pattern. Entities with a behavior relevance greater than the behavior relevance threshold are selected into the entity set of the corresponding behavior pattern. The relation edge connecting any two entities in the entity set and having a behavior weight greater than zero is selected into the relation set of the corresponding behavior pattern. The entity set and the relation set constitute the behavior preference subgraph. In the behavioral preference subgraph, entity sets are divided according to the category attributes of entities in the cultural and tourism knowledge element set, and the relational edges connecting entities within the same theme entity set and connecting different theme entity sets are retained in the behavioral preference subgraph to form a behavioral preference subgraph with a theme structure. For behavioral preference subgraphs with thematic structures, within each thematic entity set, the entity and its internal relation edges are divided into several connected components based on graph connectivity, which are regarded as a thematic candidate cultural tourism knowledge slice. At the same time, the relation edges across thematic entity sets are assigned according to the thematic entity sets they connect to, forming cross-thematic candidate cultural tourism knowledge slices. The relationship edges of the candidate cultural and tourism knowledge slices are traversed. A comprehensive importance score is generated based on the behavior weight and time label of the relationship edge. Candidate cultural and tourism knowledge slices with a comprehensive importance score greater than the preset cultural and tourism knowledge slice screening threshold are selected as the set of cultural and tourism knowledge slices.

[0011] Optionally, the generation of the target candidate cultural and tourism knowledge fragment set specifically includes: Receive natural language questions from tourists' terminals, perform sentence segmentation, word segmentation, part-of-speech tagging and dependency parsing, extract semantic elements, and generate a semantic structure of tourist questions containing slot tags and slot values; Based on the semantic structure of tourist questions, an intent recognition model is invoked to classify the intent of tourist natural language questions. Based on the relevant words appearing in the questions and the tourist's behavior patterns in the historical tourist behavior event sequence set, the target behavior pattern corresponding to the tourist's natural language questions in this round is determined. Based on the target behavior pattern, select the behavior preference subgraph corresponding to the target behavior pattern in the structured knowledge graph of the cultural and tourism field that includes behavior weights, and retrieve the target cultural and tourism knowledge slice set that matches both the target behavior pattern and the topic category in the cultural and tourism knowledge slice set based on the topic information involved in the semantic structure of tourist questions. In the behavior preference subgraph, the starting entity identified in the semantic structure of the tourist's question is taken as the starting point of the path, and the target entity in the target cultural and tourism knowledge slice set is taken as the ending point of the path. Under the premise of limiting the maximum path length and the number of node visits, a path search is performed with behavioral weight and path structure as constraints. The behavioral weight of the relation edges on the path, the category of the topic entity passed through the path, and the path length are combined as the path score. The candidate path set is selected based on the path score ranking. Semantically organize the entities and relationships involved in each path in the candidate path set, filter out paths that do not meet the constraints based on the time constraints, travel direction constraints and crowd constraints in the semantic structure of tourist questions, and generate candidate cultural tourism knowledge fragments by semantic aggregation of the remaining paths. Based on the semantic similarity, behavioral pattern matching degree, and cultural and tourism entity coverage between tourists' natural language questions and candidate cultural and tourism knowledge fragments, a weighted ranking is performed, and the top-ranked candidate cultural and tourism knowledge fragments are selected as the target candidate cultural and tourism knowledge fragment set.

[0012] Optionally, the generation of the intelligent question-and-answer results for cultural tourism specifically includes: The natural language questions from tourists and the target set of candidate cultural and tourism knowledge fragments are encapsulated and processed. The slot labels, slot values, and cultural and tourism entities and relationship information corresponding to the target set of candidate cultural and tourism knowledge fragments in the semantic structure of the tourists' questions are extracted to form the input data for question and answer generation. The target candidate cultural and tourism knowledge fragment set is organized, the structured information of relevant entities and their relationships is extracted, and merged with the question-and-answer generation input data; The merged question-and-answer generation input data is fed into the question-and-answer generation engine to generate a set of candidate intelligent questions and answers for cultural tourism that include different language styles, detail granularity and expression methods, and the generation confidence level corresponding to each candidate intelligent questions and answers for cultural tourism is retained; The knowledge consistency of the candidate intelligent cultural and tourism question and answer result set is verified. The answer content is verified based on the entity name, time tag, ticketing rules, transportation mode and policy clause appearing in the target candidate cultural and tourism knowledge fragment set, and the cultural and tourism intelligent question and answer result set that passes the consistency verification is obtained. A quality score is performed on the set of cultural and tourism intelligent question-and-answer results that pass the consistency check. A comprehensive score is calculated based on semantic fluency, knowledge coverage and target behavior pattern matching degree. The cultural and tourism intelligent question-and-answer result with the highest comprehensive score is selected as the final cultural and tourism intelligent question-and-answer result. Depending on the presentation method of the tourist terminal, the final cultural and tourism intelligent Q&A results will be displayed in text form, or input into the speech synthesis engine to generate speech broadcast content and output the cultural and tourism intelligent Q&A results in speech form. The system iteratively updates the cultural tourism Q&A interaction log by recording tourists' natural language questions, the set of target candidate cultural tourism knowledge fragments, the final intelligent cultural tourism Q&A results, and tourist feedback.

[0013] The beneficial effects of this invention are: This invention constructs a structured knowledge system for the cultural tourism field, encompassing entity modeling, relationship modeling, behavior weight updates, behavior preference subgraph generation, and knowledge slice extraction. This transforms the cultural tourism knowledge graph from a static, homogeneous semantic network into one that dynamically evolves with tourist behavior patterns, forming a knowledge structure highly consistent with actual travel preferences. This significantly improves the targeting and accuracy of knowledge retrieval in question-and-answer sessions. Upon receiving natural language questions from tourists, this invention selects target behavior preference subgraphs based on semantic understanding results and completes path search and semantic aggregation under the combined constraints of behavior weights and path structure. This results in candidate cultural tourism knowledge fragments possessing personalized, contextually relevant, and strategically related features, solving the problem that traditional systems cannot deeply understand tourist needs.

[0014] In the final question-and-answer generation stage, this invention utilizes a knowledge-enhanced question-and-answer input structure and a consistency verification mechanism. This ensures that the generative question-and-answer results maintain natural language while strictly adhering to the entities, relationships, and time constraints within the cultural and tourism knowledge slices, preventing false, biased, or non-compliant answers from appearing. Simultaneously, this invention feeds back the interaction logs from each round to the knowledge element update, knowledge graph update, and behavior prototype update processes, enabling the system to continuously adapt and evolve. It can continuously improve the quality of responses based on changes in user behavior, updates to cultural and tourism resources, and policy adjustments, achieving high accuracy, high robustness, and strong personalization in intelligent cultural and tourism question-and-answer systems. Overall performance is significantly superior to existing technologies. Attached Figure Description

[0015] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0016] Figure 1 This is a flowchart of a knowledge graph-based intelligent question-answering method for cultural tourism proposed in this invention; Figure 2 This diagram illustrates the generation of a structured knowledge graph in the cultural tourism field, including behavioral weights, for a knowledge graph-based intelligent question-answering method for cultural tourism proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figures 1-2 A knowledge graph-based intelligent question-answering method for cultural tourism includes the following steps: Acquire relevant data from the cultural and tourism business system and form a set of cultural and tourism knowledge elements through preprocessing; Based on the set of cultural and tourism knowledge elements, entity modeling and relationship modeling are performed to construct a structured knowledge graph in the cultural and tourism field; Collect tourist behavior sequences to construct candidate entity pairs, combine cultural and tourism texts to guide the relationship extraction and weight update of the candidate entity pairs' execution behavior sequences, and write them into a structured knowledge graph of the cultural and tourism domain, and adjust the edge weights and time records to obtain a structured knowledge graph of the cultural and tourism domain containing behavior weights; Based on a structured knowledge graph in the cultural and tourism field that includes behavioral weights, behavioral preference subgraphs are generated according to behavioral patterns, and cultural and tourism knowledge slice sets are extracted by topic within the behavioral preference subgraphs. The system receives natural language questions from tourists, determines the target behavior preference subgraph and the target cultural and tourism knowledge slice set through semantic understanding and intent recognition, and performs path search and semantic aggregation constrained by behavior weight and path structure to obtain the target candidate cultural and tourism knowledge fragment set. The system inputs tourists' natural language questions and a set of target candidate cultural and tourism knowledge fragments into the question-and-answer generation engine to generate intelligent cultural and tourism question-and-answer results. The results are output in text or voice form, and the interactive log of cultural and tourism question-and-answer is recorded for iterative updates.

[0019] In this embodiment, the generation of the cultural and tourism knowledge element set specifically includes: Relevant data were obtained from the cultural and tourism business system and categorized according to their source to obtain the original cultural and tourism data set; The original cultural and tourism data set includes subsets of scenic spot data, route data, opening time data, ticket data, transportation data, accommodation data, catering data, activity data, policy data, and tourist behavior data. The time field and spatial location information in the original cultural and tourism data set are standardized by time reference and spatial coordinate, resulting in a cultural and tourism spatiotemporally aligned data set. Based on the data structure characteristics, the cultural and tourism spatiotemporally aligned data set is divided into a structured cultural and tourism data subset, a cultural and tourism text subset, and a tourist behavior log subset. Missing value imputation, outlier removal and duplicate record merging, noise symbol cleaning and encoding unification, abnormal behavior filtering and invalid event removal are performed respectively to obtain the cleaned structured cultural and tourism data subset, cultural and tourism text subset, and tourist behavior log subset. The cleaned cultural and tourism text subset is subjected to sentence segmentation, word segmentation, part-of-speech tagging and preliminary named entity screening. Phrases involving scenic spot names, route descriptions, modes of transportation, accommodation types, catering types, activity names and policy clauses are marked as cultural and tourism text elements, generating a set of cultural and tourism text elements with semantic tags. The cleaned structured cultural and tourism data subset and the cultural and tourism spatiotemporally aligned data set are standardized and semantically aligned. Based on the set of field mapping rules, the ticket price field, opening time field, schedule field, room type field, menu field, activity time field and policy effective time field in different data sources are uniformly mapped to the preset standardized field space to obtain a standardized set of cultural and tourism data elements. The cleaned subset of visitor behavior logs is aggregated according to visitor identifier and time sequence, and search behavior, click behavior, browsing behavior, collection behavior and order placement behavior are arranged into a set of visitor behavior event sequences in time sequence; The set of cultural and tourism text elements with semantic tags, the set of standardized and structured cultural and tourism data elements, and the set of tourist behavior event sequences are uniformly encapsulated. For each cultural and tourism knowledge element, category attributes, semantic content attributes, time attributes, and spatial attributes are constructed simultaneously to form a set of cultural and tourism knowledge elements.

[0020] In this embodiment, the generation of the structured knowledge graph in the cultural and tourism field specifically includes: Based on the set of cultural and tourism knowledge elements, the category attributes of each cultural and tourism knowledge element are traversed, and cultural and tourism knowledge elements whose category attributes belong to the same category are divided into the same initial entity candidate set, forming multiple initial entity candidate sets, which are defined as entity sets. Category attributes include attraction categories, transportation categories, accommodation categories, dining categories, activity categories, policy categories, route categories, and other categories; Extract the semantic content attributes of each cultural and tourism knowledge element from the set of cultural and tourism knowledge elements, input them into a preset semantic encoding function to obtain the corresponding semantic vector representation, and obtain a set of entities with semantic vector representation; Co-occurrence statistics are performed on cultural and tourism text elements that co-occur with the same cultural and tourism text element. Structural association analysis is also performed on fields with foreign key association, primary key association, and enumeration relationship in standardized structured cultural and tourism data elements, and candidate relationship pairs are formed by pairwise combination. For each candidate relationship pair, a set of candidate relationship types is determined based on the combination of the entity categories of the two entities, forming a set of relationship types that includes multiple cultural and tourism relationship types; For each candidate relation pair, combining the semantic vector representations of the two entities and the attribute features extracted from standardized structured cultural and tourism data elements, the relation scoring function is used to calculate the relation score value of the candidate relation pair under each relation type. The relationship score is filtered by a threshold. Candidate relationship pairs that are greater than the preset relationship threshold are determined as valid relationship edges under that relationship type. They are recorded as a combination of the starting entity, relationship type and ending entity in the form of a triple, forming a relationship set. Based on entity sets and relation sets, a structured knowledge graph for the cultural tourism field is constructed, with entity sets as the set of nodes in the graph structure and relation sets as the set of edges in the graph structure.

[0021] In this embodiment, the generation of the structured knowledge graph in the cultural and tourism field, which includes behavioral weights, specifically includes: Based on the set of tourist behavior event sequences, tourist behavior events are aggregated according to tourist identifier and time order, and search behavior, click behavior, browsing behavior, collection behavior and order placement behavior are arranged in time order to form a set of tourist behavior sequences; The set of tourist behavior sequences is vectorized and mapped to the corresponding tourist behavior representation. Clustering operation is used to obtain a set of behavior prototypes representing different cultural and tourism behavior patterns. The behavior prototypes are then mapped to the behavior patterns of parent-child preferences, food preferences, cultural preferences, and photography preferences. The similarity between the tourist behavior sequence and each behavior prototype in the behavior prototype set is calculated. The behavior prototypes are then weighted and fused according to the similarity weight to obtain the weighted behavior representation of the behavior prototype. This weighted behavior representation is then matched with the entity set in the structured knowledge graph of the cultural and tourism field to select the behavior interest entity set related to the behavior prototype weighted behavior representation. Finally, the relationship set in the structured knowledge graph of the cultural and tourism field is combined with the entity pair combination in the behavior interest entity set to form a candidate entity pair set for guiding the behavior sequence. For each candidate entity pair in the candidate entity pair set, extract the cultural and tourism text containing the candidate entity pair from the cultural and tourism text subset, obtain the entity context semantic vector and the cultural and tourism text semantic vector through a preset semantic encoding function, and combine the cultural and tourism text semantic vector with the behavior prototype weighted behavior representation to obtain the cultural and tourism text semantic representation that integrates behavior information. During the behavior guidance relationship extraction training phase, the behavior prototype weighted behavior representation in the real sample is replaced with the behavior prototype weighted behavior representation generated by other tourists' behavior sequences. Inconsistent comparison samples are constructed, and the relationship scoring function is called to calculate the behavior guidance relationship score. Based on this, the behavior consistency loss is obtained, so that the behavior guidance relationship score of the real sample is greater than the behavior guidance relationship score of the inconsistent sample. The behavioral consistency loss and the basic relation extraction loss are weighted and combined, and an adaptive weighting coefficient for uncertainty is introduced. The weight of the behavioral consistency loss in the total loss is dynamically adjusted according to the uncertainty of relation prediction, resulting in a total loss that combines basic relation extraction and behavioral consistency. Based on the loss value, behavioral-guided relation extraction training is completed. The trained relation scoring function is used to calculate the behavior guidance relation score of the candidate entity pair set under each cultural and tourism relation type. Threshold filtering and normalization processing are performed on the behavior guidance relation score to obtain the relation type selection result under each cultural and tourism relation type and the edge weight update amount corresponding to the relation type. The edge weight update is accumulated and added to the edge weight of the corresponding relation edge in the structured knowledge graph of the cultural and tourism field. At the same time, the time label of the relation edge is updated according to the current time, so as to obtain a structured knowledge graph of the cultural and tourism field containing behavioral weights.

[0022] In this embodiment, the generation of the cultural and tourism knowledge slice set specifically includes: Based on a structured knowledge graph in the cultural and tourism field that includes behavioral weights, the entity set, the relationship set, and the behavioral weights and time tags of each relationship edge are read. Each behavioral prototype in the behavioral prototype set is mapped to a behavioral pattern of parent-child preference, food preference, culture preference, and photography preference, thus forming a behavioral pattern set. For each behavior pattern in the behavior pattern set, the relation set is traversed in the structured knowledge graph of the cultural tourism field containing behavior weights. The behavior relevance of the entities connected to the relation edge is accumulated according to the behavior weight of the relation edge in the behavior pattern. Entities with a behavior relevance greater than the behavior relevance threshold are selected into the entity set of the corresponding behavior pattern. The relation edge connecting any two entities in the entity set and having a behavior weight greater than zero is selected into the relation set of the corresponding behavior pattern. The entity set and the relation set constitute the behavior preference subgraph. In the behavioral preference subgraph, entity sets are divided according to the category attributes of entities in the cultural and tourism knowledge element set, and the relational edges connecting entities within the same theme entity set and connecting different theme entity sets are retained in the behavioral preference subgraph to form a behavioral preference subgraph with a theme structure. The entity set includes the following themes: scenic spot entity set, route entity set, transportation entity set, accommodation entity set, catering entity set, activity entity set, and policy entity set. For behavioral preference subgraphs with thematic structures, within each thematic entity set, the entity and its internal relation edges are divided into several connected components based on graph connectivity, which are regarded as a thematic candidate cultural tourism knowledge slice. At the same time, the relation edges across thematic entity sets are assigned according to the thematic entity sets they connect to, forming cross-thematic candidate cultural tourism knowledge slices. The relationship edges of the candidate cultural and tourism knowledge slices are traversed. A comprehensive importance score is generated based on the behavior weight and time label of the relationship edge. Candidate cultural and tourism knowledge slices with a comprehensive importance score greater than the preset cultural and tourism knowledge slice screening threshold are selected as the set of cultural and tourism knowledge slices.

[0023] In this embodiment, the generation of the target candidate cultural and tourism knowledge fragment set specifically includes: Receive natural language questions from tourists' terminals, perform sentence segmentation, word segmentation, part-of-speech tagging and dependency parsing, extract semantic elements, and generate a semantic structure of tourist questions containing slot tags and slot values; Based on the semantic structure of tourist questions, an intent recognition model is invoked to classify the intent of tourist natural language questions. Based on the relevant words appearing in the questions and the tourist's behavior patterns in the historical tourist behavior event sequence set, the target behavior pattern corresponding to the tourist's natural language questions in this round is determined. Based on the target behavior pattern, select the behavior preference subgraph corresponding to the target behavior pattern in the structured knowledge graph of the cultural and tourism field that includes behavior weights, and retrieve the target cultural and tourism knowledge slice set that matches both the target behavior pattern and the topic category in the cultural and tourism knowledge slice set based on the topic information involved in the semantic structure of tourist questions. In the behavior preference subgraph, the starting entity identified in the semantic structure of the tourist's question is taken as the starting point of the path, and the target entity in the target cultural and tourism knowledge slice set is taken as the ending point of the path. Under the premise of limiting the maximum path length and the number of node visits, a path search is performed with behavioral weight and path structure as constraints. The behavioral weight of the relation edges on the path, the category of the topic entity passed through the path, and the path length are combined as the path score. The candidate path set is selected based on the path score ranking. Semantically organize the entities and relationships involved in each path in the candidate path set, filter out paths that do not meet the constraints based on the time constraints, travel direction constraints and crowd constraints in the semantic structure of tourist questions, and generate candidate cultural tourism knowledge fragments by semantic aggregation of the remaining paths. Based on the semantic similarity, behavioral pattern matching degree, and cultural and tourism entity coverage between tourists' natural language questions and candidate cultural and tourism knowledge fragments, a weighted ranking is performed, and the top-ranked candidate cultural and tourism knowledge fragments are selected as the target candidate cultural and tourism knowledge fragment set.

[0024] In this embodiment, the generation of the intelligent question-and-answer results for cultural tourism specifically includes: The natural language questions from tourists and the target set of candidate cultural and tourism knowledge fragments are encapsulated and processed. The slot labels, slot values, and cultural and tourism entities and relationship information corresponding to the target set of candidate cultural and tourism knowledge fragments in the semantic structure of the tourists' questions are extracted to form the input data for question and answer generation. The target candidate cultural and tourism knowledge fragment set is organized, the structured information of relevant entities and their relationships is extracted, and merged with the question-and-answer generation input data; The merged question-and-answer generation input data is fed into the question-and-answer generation engine to generate a set of candidate intelligent questions and answers for cultural tourism that include different language styles, detail granularity and expression methods, and the generation confidence level corresponding to each candidate intelligent questions and answers for cultural tourism is retained; The knowledge consistency of the candidate intelligent cultural and tourism question and answer result set is verified. The answer content is verified based on the entity name, time tag, ticketing rules, transportation mode and policy clause appearing in the target candidate cultural and tourism knowledge fragment set, and the cultural and tourism intelligent question and answer result set that passes the consistency verification is obtained. A quality score is performed on the set of cultural and tourism intelligent question-and-answer results that pass the consistency check. A comprehensive score is calculated based on semantic fluency, knowledge coverage and target behavior pattern matching degree. The cultural and tourism intelligent question-and-answer result with the highest comprehensive score is selected as the final cultural and tourism intelligent question-and-answer result. Depending on the presentation method of the tourist terminal, the final cultural and tourism intelligent Q&A results will be displayed in text form, or input into the speech synthesis engine to generate speech broadcast content and output the cultural and tourism intelligent Q&A results in speech form. The system iteratively updates the cultural tourism Q&A interaction log by recording tourists' natural language questions, the set of target candidate cultural tourism knowledge fragments, the final intelligent cultural tourism Q&A results, and tourist feedback.

[0025] Example 1: To verify the feasibility of this invention in practice, it was applied to a smart tourism platform built by the cultural and tourism department of a certain province. This platform covers major scenic spots, hotels, transportation hubs, and cultural venues in the area, and receives a large number of tourist inquiries daily, such as route planning, attraction opening hours, dining options, and suitable demographics. However, it has long relied on keyword matching and static question-and-answer databases, failing to understand the true needs of tourists. Especially during the peak summer tourist season, issues such as inaccurate answers, unsuitable recommendations, and disjointed itinerary planning frequently arise, leading to a decline in tourist experience. The application of this invention on the platform is based on real tourist inquiry data, operational data, and behavioral logs, achieving a full-chain intelligent upgrade from knowledge modeling to question-and-answer generation.

[0026] In this scenario, the present invention first accesses multi-source cultural tourism data provided by scenic area management, transportation services, and accommodation and catering services. It then combines this data with tourists' search, browsing, and booking behaviors on mobile devices to form cultural tourism knowledge elements suitable for knowledge modeling. Based on the behavior-guided relationship extraction method proposed in this invention, data from different sources is transformed into structured entities, and relationship edges are dynamically weighted through behavioral patterns. This ensures that the strength of the association between attractions and transportation, catering, and activities reflects the actual level of tourist interest. When tourists submit natural language inquiries on mobile devices, the present invention can automatically parse the semantics, identify implicit travel preferences, companion groups, and time constraints in the question, accurately segment it into the corresponding behavioral preference subgraph, and then select relevant cultural tourism knowledge slices from this subgraph for subsequent reasoning.

[0027] In a real-world platform operation scenario, a tourist with a child repeatedly searched for "Which places are suitable for children?", "How to enjoy the Summer Palace?", and "What are some easily accessible places nearby" during a summer visit. This invention automatically categorized their behavior into a parent-child preference mode. In subsequent voice consultations, it combined behavioral preference subgraphs and knowledge slices to perform path reasoning, outputting comprehensive answers covering transportation convenience, child suitability, and activity scheduling, making the answers more closely aligned with their actual needs. According to data records from local cultural and tourism management departments, during peak tourist consultation periods, the answers generated by this invention showed significant improvements in operational data such as tourist dwell time, subsequent click-through rates, and the number of secondary inquiries. Especially in complex consultation scenarios, tourists' acceptance of the answers was significantly higher than with previous matching-based question-and-answer methods, indicating that the behavior-weighted knowledge reasoning mechanism effectively improved the quality of question-and-answer responses.

[0028] Analysis of operational data reveals that, despite adjustments to cultural and tourism policies, updates to attraction activities, and shifts in tourist focus, this invention leverages accumulated question-and-answer interaction logs to continuously update behavioral prototypes and knowledge slices, ensuring the knowledge structure evolves in sync with actual operations. Staff stated that the implementation of this invention significantly reduces the difficulty of manually maintaining knowledge content, making responses more intelligent and personalized. Even under high-pressure conditions during peak seasons, it maintains stable and accurate information services, fully validating the effectiveness and advancement of this invention in practical applications.

[0029] Table 1. Performance comparison between knowledge graph-based intelligent question answering methods for cultural tourism and traditional methods.

[0030] As shown in Table 1, this invention achieves significant performance improvements in key aspects of intelligent question answering in the cultural tourism sector. Firstly, in terms of semantic understanding, the accuracy of tourist intent recognition increases from 82.3% to 93.1% compared to traditional methods, and the slot filling accuracy increases from 78.6% to 90.4%. This demonstrates that the behavioral pattern-based semantic understanding mechanism of this invention can better combine tourists' historical behavior with current semantic cues, enabling the model to more accurately understand the true needs behind the questions. For example, when tourists with children inquire about routes, this invention can utilize behavioral preference subgraphs to pinpoint their preferred family routes, making intent recognition no longer limited to literal meaning and thus significantly reducing comprehension bias.

[0031] In terms of knowledge reasoning ability, the success rate of multi-hop path reasoning increased from 63.5% to 88.7%, representing the largest improvement among all indicators. This is mainly due to the fact that when constructing a structured knowledge graph in the cultural tourism field that includes behavioral weights, this invention uses behavior-guided relationship extraction and dynamic adjustment of edge weights to make the reasoning path closer to the selection habits of real tourists, thereby improving the accessibility and stability of reasoning. At the same time, the accuracy rate of behavior preference matching increased by 22.8%, fully demonstrating that the behavior pattern clustering mechanism of this invention effectively captures tourists' implicit dependence on preferences such as "suitable for children," "likes food," and "in-depth cultural tours," making the reasoning path and answer content more in line with personalized needs.

[0032] In terms of answer quality, the knowledge coverage rate increased from 74.1% to 91.6%, indicating that the present invention, through its thematic cultural tourism knowledge slice generation mechanism, can extract more complete and relevant knowledge segments from the behavioral preference subgraph, avoiding the information fragmentation problem that occurs in traditional methods. The answer consistency verification pass rate also improved by 14.3%, proving that the knowledge verification mechanism introduced in the final question-and-answer generation stage effectively reduces common errors such as error timing and ticketing or policy information, making the results more credible and standardized.

[0033] Visitor experience metrics further validated the actual effectiveness. The secondary question rate decreased from 31.7% to 18.5%, indicating that visitors no longer frequently ask follow-up questions after receiving answers generated by this invention; the click-through rate increased by 15.8% after answers, reflecting a higher degree of matching between the answers and actual interests; and the 21.3% increase in dwell time indicates a significant improvement in experience quality and enhanced platform attractiveness. These results all demonstrate the significant value of behavior-weighted knowledge reasoning and intelligent question-answering mechanisms in practical cultural and tourism service scenarios.

[0034] In summary, this invention achieves intelligent enhancement of the entire process from data modeling to answer generation through a chain structure of "behavioral weight knowledge graph construction + behavioral preference subgraph + knowledge slicing + constraint path search + semantic generation verification". This makes cultural tourism Q&A more accurate, reliable and personalized, and its overall performance, which is superior to existing technologies, has been fully verified in tabular data.

[0035] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A knowledge graph-based intelligent question and answer method for tourism, characterized in that, The method comprises the following steps: Obtaining relevant data in the travel business system, and forming a travel knowledge element set through preprocessing; Based on the travel knowledge element set, entity modeling and relationship modeling are performed to construct a structured knowledge graph in the travel field; Collecting the behavior sequence of tourists to construct a candidate entity pair, combining the travel text to perform behavior sequence guided relationship extraction and weight updating on the candidate entity pair, and writing into the structured knowledge graph in the travel field and adjusting the edge weight and time record to obtain a structured knowledge graph in the travel field containing behavior weight; Based on the structured knowledge graph in the travel field containing behavior weight, behavior preference subgraphs are generated according to behavior patterns, and travel knowledge slice sets are extracted according to themes in the behavior preference subgraphs; Receiving a natural language question of a tourist, determining a target behavior preference subgraph and a target travel knowledge slice set through semantic understanding and intent recognition, performing path search and semantic aggregation constrained by behavior weight and path structure to obtain a target candidate travel knowledge fragment set; The natural language question of the tourist and the target candidate travel knowledge fragment set are input into a question and answer generation engine to generate a travel intelligent question and answer result, and the travel intelligent question and answer result is output in the form of text or voice, and a travel question and answer interaction log is recorded for iterative updating.

2. The knowledge graph-based intelligent question answering method for tourism according to claim 1, characterized in that, The generation of the travel knowledge element set specifically comprises: Respectively obtaining relevant data from the travel business system, classifying according to data sources, and obtaining an original travel data set; Standardizing the time field and spatial location information in the original travel data set to obtain a travel space-time aligned data set; According to the data structure characteristics, the travel space-time aligned data set is divided into a structured travel data subset, a travel text subset and a tourist behavior log subset, and respectively performs missing value filling, outlier removal and repeated record merging, noise symbol cleaning and coding unification, abnormal behavior filtering and invalid event removal to obtain the cleaned structured travel data subset, travel text subset and tourist behavior log subset; Performing sentence segmentation, word segmentation, part-of-speech tagging and preliminary named entity screening on the cleaned travel text subset, marking phrases related to scenic spot names, route descriptions, transportation methods, accommodation types, catering types, activity names and policy clauses as travel text elements, and generating a travel text element set with semantic labels; Performing field standardization and semantic alignment on the cleaned structured travel data subset and the travel space-time aligned data set, and according to the field mapping rule set, uniformly mapping the ticket price field, opening time field, class field, room type field, dish field, activity time field and policy effective time field in different data sources to the preset standardized field space to obtain a standardized structured travel data element set; Aggregating the cleaned tourist behavior log subset according to the tourist identifier and time sequence, arranging the search behavior, click behavior, browsing behavior, collection behavior and order behavior into a tourist behavior event sequence set according to time sequence; The text elements of the travel and culture field with semantic labels, the standardized structured data elements of the travel and culture field, and the sequence set of tourist behavior events are uniformly encapsulated, and the category attribute, the semantic content attribute, the time attribute, and the space attribute of each travel and culture knowledge element are constructed to form a travel and culture knowledge element set.

3. The knowledge graph-based intelligent question answering method for tourism according to claim 1, characterized in that, The generation of the structured knowledge graph in the travel and culture field specifically includes: Based on the travel and culture knowledge element set, the category attribute of each travel and culture knowledge element is traversed, and the travel and culture knowledge elements with the same category attribute are divided into the same initial entity candidate set to form a plurality of initial entity candidate sets, which are defined as an entity set; The semantic content attribute of each travel and culture knowledge element is extracted from the travel and culture knowledge element set, and a corresponding semantic vector representation is obtained by inputting a preset semantic encoding function to obtain an entity set with a semantic vector representation; The co-occurrence of the text elements of the travel and culture field in the same travel and culture text element is statistically analyzed, and the fields with foreign key association, primary key association, and enumeration relationship in the standardized structured data elements of the travel and culture field are analyzed to form a candidate relationship pair set; For each candidate relationship pair, the candidate relationship type set is determined according to the entity category combination of the two entities to form a relationship type set containing multiple travel and culture relationship types; For each candidate relationship pair, the relationship score value of the candidate relationship pair under each relationship type is calculated by a relationship scoring function in combination with the semantic vector representation of the two entities and the attribute features extracted from the standardized structured data elements of the travel and culture field; The relationship score value is threshold filtered, and the candidate relationship pair with a relationship score value greater than a preset relationship threshold is determined as an effective relationship edge under the relationship type, recorded in the form of a triple combination of a starting entity, a relationship type, and a terminating entity, and a relationship set is formed; Based on the entity set and the relationship set, a structured knowledge graph in the travel and culture field is constructed, the entity set is taken as a node set in the graph structure, and the relationship set is taken as an edge set in the graph structure.

4. The knowledge graph-based intelligent question answering method for cultural tourism according to claim 1, characterized in that, The generation of the structured knowledge graph in the travel and culture field containing behavior weights specifically includes: Based on the sequence set of tourist behavior events, the tourist behavior events are aggregated according to the tourist identifier and the time sequence to form a tourist behavior sequence set in which the search behavior, the click behavior, the browse behavior, the collection behavior, and the order behavior are arranged in time sequence; The tourist behavior sequence set is vectorized and encoded to map a corresponding tourist behavior representation, and a behavior prototype set representing different travel behavior modes is obtained through clustering operation, and the behavior prototype is matched with the behavior modes of the parent-child preference, the food preference, the culture preference, and the photography preference; The similarity between the tourist behavior sequence and each behavior prototype in the behavior prototype set is calculated, each behavior prototype is weighted and fused according to the similarity weight, a behavior prototype weighted behavior representation is obtained, and the behavior prototype weighted behavior representation is matched with the entity set in the structured knowledge graph in the travel and culture field, a behavior attention entity set related to the behavior prototype weighted behavior representation is screened out, and the behavior attention entity set is combined with the relationship set in the structured knowledge graph in the travel and culture field to form a candidate entity pair set for behavior sequence guidance. For each candidate entity pair in the candidate entity pair set, extract a travel and tourism text containing the candidate entity pair from the travel and tourism text subset, obtain the entity context semantic vector and the travel and tourism text semantic vector by using the preset semantic encoding function, and combine the travel and tourism text semantic vector with the behavior prototype weighted behavior representation to obtain a travel and tourism text semantic representation fused with behavior information; In the behavior guidance relationship extraction training phase, the corresponding behavior prototype weighted behavior representation in the real sample is replaced by the behavior prototype weighted behavior representation generated by other tourist behavior sequences to construct a behavior inconsistency comparison sample corresponding thereto, the relationship scoring function is called respectively to calculate the behavior guidance relationship score, and the behavior consistency loss is obtained according to the behavior guidance relationship score of the real sample being greater than the behavior guidance relationship score of the behavior inconsistency sample; The behavior consistency loss and the basic relationship extraction loss are combined by weighting, an uncertainty adaptive weighting coefficient is introduced, the weight of the behavior consistency loss in the total loss is dynamically adjusted according to the relationship prediction uncertainty, the total loss combining the basic relationship extraction and the behavior consistency is obtained, and the behavior guidance relationship extraction training is completed based on the loss value; The behavior guidance relationship score of the candidate entity pair set under each travel and tourism relationship type is calculated by using the trained relationship scoring function, threshold screening and normalization processing are performed on the behavior guidance relationship score to obtain the relationship type selection result under each travel and tourism relationship type and the edge weight update amount corresponding to the relationship type; The edge weight update amount is added to the edge weight of the corresponding relationship edge in the travel and tourism field structured knowledge graph, and the time tag of the relationship edge is updated at the current time to obtain the travel and tourism field structured knowledge graph containing behavior weights.

5. The knowledge graph-based intelligent question answering method for tourism according to claim 1, characterized in that, The generation of the travel and tourism knowledge slice set specifically includes: Based on the travel and tourism field structured knowledge graph containing behavior weights, the entity set, the relationship set, and the behavior weight and the time tag of each relationship edge are read, each behavior prototype in the behavior prototype set is corresponded to the behavior mode of parental preference, food preference, cultural preference, and photography preference to form a behavior mode set; For each behavior mode in the behavior mode set, the relationship set in the travel and tourism field structured knowledge graph containing behavior weights is traversed, the behavior relevance of the entities connected to the relationship edge is accumulated according to the behavior weight of the relationship edge under the behavior mode, the entities greater than the behavior relevance threshold are selected into the entity set of the corresponding behavior mode, and the relationship edges connecting any two entities in the entity set and having a behavior weight greater than zero are selected into the relationship set of the corresponding behavior mode, and the entity set and the relationship set form a behavior preference subgraph; In the behavior preference subgraph, the entity set is divided according to the category attribute of the entity in the travel and tourism knowledge element set, and the relationship edges connecting the entities in the same theme entity set and the relationship edges connecting different theme entity sets in the behavior preference subgraph are retained to form a behavior preference subgraph with a theme structure; In the behavior preference subgraph, the entity set is divided according to the category attribute of the entity in the travel and tourism knowledge element set, and the relationship edges connecting the entities in the same theme entity set and the relationship edges connecting different theme entity sets in the behavior preference subgraph are retained to form a behavior preference subgraph with a theme structure; For the behavior preference subgraph with a theme structure, the entities and their internal relationship edges are divided into several connected components according to the graph connectivity within each theme entity set, and are regarded as a theme candidate travel knowledge slice. Meanwhile, the relationship edges across theme entity sets are attributed to the theme entity sets connected by the relationship edges, forming cross-theme candidate travel knowledge slices. The relationship edges in the candidate travel knowledge slice are traversed, and a comprehensive importance score is generated according to the behavior weight and time label of the relationship edge. The candidate travel knowledge slice with a comprehensive importance score greater than a preset travel knowledge slice screening threshold is selected as a travel knowledge slice set.

6. The knowledge graph-based intelligent question answering method for cultural tourism according to claim 1, characterized in that, The generation of the target candidate travel knowledge fragment set specifically includes: Receiving a natural language question from a tourist terminal, performing sentence segmentation, word segmentation, part-of-speech tagging, and dependency syntax analysis, extracting semantic elements, and generating a tourist question semantic structure containing slot labels and slot values; Based on the tourist question semantic structure, an intent recognition model is called to classify the natural language question of the tourist, and according to the related words in the question, the behavior mode of the tourist in the historical tourist behavior event sequence set is determined, and the target behavior mode corresponding to the natural language question of the tourist is determined; According to the target behavior mode, the behavior preference subgraph corresponding to the target behavior mode is selected from the travel field structured knowledge graph containing behavior weights, and the target travel knowledge slice set matching the target behavior mode and the theme category is retrieved from the travel knowledge slice set according to the theme information involved in the tourist question semantic structure; In the behavior preference subgraph, the starting entity identified in the tourist question semantic structure is taken as the starting point of the path, and the target entity in the target travel knowledge slice set is taken as the end point of the path. Under the premise of limiting the maximum path length and the number of node visits, the path search is performed with the behavior weight and the path structure as the constraints. The behavior weight of the relationship edge on the path, the theme entity category passed through the path, and the path length are comprehensively considered as the path score. Based on the path score ranking, the candidate path set is obtained; The entities and relationships involved in each path in the candidate path set are semantically organized, and the paths that do not meet the constraint conditions are filtered out according to the time constraint, travel direction constraint and crowd constraint in the tourist question semantic structure. The remaining paths are semantically aggregated to generate candidate travel knowledge fragments; Based on the semantic similarity between the natural language question of the tourist and the candidate travel knowledge fragments, the behavior mode matching degree, and the travel entity coverage, the candidate travel knowledge fragments are weighted and sorted, and the top several candidate travel knowledge fragments are selected as the target candidate travel knowledge fragment set.

7. The knowledge graph-based intelligent question answering method for tourism according to claim 1, characterized in that, The generation of the travel intelligent question answering result specifically includes: Encapsulating the natural language question of the tourist and the target candidate travel knowledge fragment set, extracting the slot labels, slot values in the tourist question semantic structure, and the travel entities and relationship information corresponding to the target candidate travel knowledge fragment set, and forming the question and answer generation input data; Organizing the target candidate travel knowledge fragment set, extracting the structured information of the related entities and their relationships, and merging with the question and answer generation input data; The combined question and answer generation input data is input into a question and answer generation engine to generate a set of alternative intelligent travel question and answer results in different language styles, detail granularities, and expression manners, and to retain the generation confidence corresponding to each alternative intelligent travel question and answer result; The set of alternative intelligent travel question and answer results is subjected to knowledge consistency verification, and the answer content is verified according to the entity names, time labels, ticket rules, transportation modes, and policy provisions appearing in the set of target candidate travel knowledge segments to obtain a set of intelligent travel question and answer results that pass the consistency verification; The set of intelligent travel question and answer results that pass the consistency verification is subjected to quality scoring, a comprehensive score is calculated according to semantic fluency, knowledge coverage, and target behavior pattern matching degree, and the intelligent travel question and answer result with the highest comprehensive score is selected as the final intelligent travel question and answer result; According to the presentation mode of the tourist terminal, the final intelligent travel question and answer result is displayed in text form, or input into a speech synthesis engine to generate speech broadcast content, and the intelligent travel question and answer result is output in the form of speech; The tourist natural language question, the set of target candidate travel knowledge segment identifiers, the final intelligent travel question and answer result, and the tourist feedback record are iteratively updated as a travel question and answer interaction log.

Images