An attraction AI intelligent tour guide method and system

Abstract

The application belongs to the technical field of intelligent guide technology, and provides a scenic spot AI intelligent guide method and system. The method comprises the following steps: periodically fusing private domain knowledge built by a user and public knowledge graph provided by an organization to obtain a scenic spot knowledge graph corresponding to a target scenic spot; generating a plurality of virtual AI travel companion dolls corresponding to the target scenic spot based on subject object data and companion object data; analyzing a first scenic spot feature data group and a plurality of second scenic spot feature data groups to obtain a tour route of the subject object in the target scenic spot; matching and analyzing the first scenic spot feature data group, the tour route and the scenic spot knowledge graph to obtain target knowledge data, and generating target explanation data based on the target knowledge data; and sending the target explanation data to a mobile terminal and outputting the target explanation data by an AI travel companion doll selected by the user. The application can meet the personalized explanation needs of the user, and can also significantly improve the interactive interest by generating different virtual AI travel companion dolls.

Description

Technical Field

[0001] This invention relates to the field of intelligent tour guide technology, and more specifically, to an AI-powered intelligent tour guide method and system for scenic areas. Background Technology

[0002] Currently, scenic spots still commonly use pre-recorded audio guides to provide guided tours for tourists. This means that by providing tourists with pre-recorded audio guides, when tourists arrive at designated scenic spots, the pre-stored audio or text content is triggered to convey information about the history, culture, architectural features, and natural landscapes of the scenic spots.

[0003] The content of existing audio guides is pre-recorded or configured by scenic area management. This means the content is generated based on public knowledge defined and released by relevant organizations, and parameters such as the knowledge points, narrative style, and information depth remain fixed after equipment deployment, making dynamic adjustments impossible based on individual visitor differences. Furthermore, current guided tour methods lack engagement and visitor participation. The traditional one-way information output model of audio guides lacks interactive design, leaving visitors largely in a passive receiving state, hindering immersive experiences and resulting in low usage and satisfaction rates for guided tours.

[0004] Therefore, in response to the problems of fixed content and lack of interest in existing scenic spot guide technologies, there is an urgent need for an AI-powered intelligent guide method for scenic spots that can deliver personalized content and enhance interactive fun, so as to improve the guide experience for tourists and the quality of scenic spot services. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a scenic area AI intelligent navigation method, system, electronic device, non-transient computer-readable storage medium, and computer program product.

[0006] This invention provides an AI-powered intelligent tour guide method for scenic areas, applied to an intelligent platform, comprising the following steps: periodically fusing user-built private domain knowledge with public domain knowledge graphs provided by institutions to obtain a scenic area knowledge graph corresponding to the target scenic area; extracting subject object data and companion object data from a first real-scene video sent from a mobile terminal, and generating several virtual AI travel companion dolls corresponding to the target scenic area based on the subject object data and companion object data; extracting a first scenic spot feature data group from a second real-scene video sent from a mobile terminal, and analyzing the subject object's tour route within the target scenic area based on the first scenic spot feature data group and several second scenic spot feature data groups; wherein the second scenic spot feature data group is a feature data group previously extracted from other locations within the target scenic area; matching and analyzing the first scenic spot feature data group, the tour route, and the scenic area knowledge graph to obtain target knowledge data, and generating target explanation data based on the target knowledge data; wherein the tour route is related to the tourist's tour preferences; and sending the target explanation data to the mobile terminal and outputting it through the AI ​​travel companion doll selected by the user.

[0007] This invention also provides a scenic area AI intelligent guide system, applied to an intelligent platform, comprising: a knowledge graph generation module, which periodically merges user-built private domain knowledge with public domain knowledge graphs provided by institutions to obtain a scenic area knowledge graph corresponding to the target scenic area; a virtual avatar generation module, which extracts subject object data and companion object data from a first real-scene video sent from a mobile terminal, and generates several virtual AI travel companion dolls corresponding to the target scenic area based on the subject object data and companion object data; a tour route generation module, which extracts a first scenic spot feature data group from a second real-scene video sent from a mobile terminal, and analyzes the tour route of the subject object within the target scenic area based on the first scenic spot feature data group and several second scenic spot feature data groups; wherein, the second scenic spot feature data group is a feature data group previously extracted from other locations within the target scenic area; a narration data generation module, which matches and analyzes the first scenic spot feature data group, the tour route, and the scenic area knowledge graph to obtain target knowledge data, and generates target narration data based on the target knowledge data; wherein, the tour route is related to the tourist's tour preferences; and a narration data sending module, which sends the target narration data to the mobile terminal and outputs it through the AI ​​travel companion doll selected by the user.

[0008] The present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, implements the method described in the preceding invention.

[0009] The present invention also provides a non-transient computer-readable storage medium having instructions embodied thereon that can be executed by one or more processors to perform a method as described in any of the preceding claims.

[0010] The present invention also provides a computer program product comprising an executable computer program that, when executed, implements the method as described in any of the preceding claims.

[0011] This invention enriches the content of explanations by periodically integrating users' self-built private knowledge and public knowledge graphs provided by institutions. It also infers tourists' travel preferences based on their actual tour routes and makes adaptive adjustments to the content of explanations to meet users' personalized needs. At the same time, this invention also generates different virtual AI travel companion dolls for users, which can significantly enhance the interactive fun. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating a scenic area AI intelligent navigation method disclosed in an embodiment of the present invention.

[0013] Figure 2 This is a schematic diagram of the display interface of the navigation APP disclosed in the embodiment of the present invention.

[0014] Figure 3 This is a schematic diagram of the structure of a scenic area AI intelligent guide system disclosed in an embodiment of the present invention. Detailed Implementation

[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. 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.

[0017] To facilitate understanding of the embodiments of the present invention, further explanations and descriptions will be provided below with reference to the accompanying drawings and specific embodiments. These embodiments do not constitute a limitation on the embodiments of the present invention.

[0018] like Figure 1As shown, this embodiment of the invention discloses a scenic area AI intelligent navigation method 100, which is applied to an intelligent platform and includes the following steps: S10, periodically merging user-built private domain knowledge and public domain knowledge graphs provided by institutions to obtain a scenic area knowledge graph corresponding to the target scenic area.

[0019] In this step, the intelligent platform acquires private domain knowledge created by users (including tourists, self-media bloggers, etc.) and public domain knowledge graphs provided by institutions. The private domain knowledge created by users includes personalized information uploaded by tourists in advance, such as personal interests (e.g., preference for details of ancient buildings, interest in the medicinal value of plants), exclusive stories (e.g., family members' experiences of visiting the scenic spot), and special needs (e.g., needing to focus on child-oriented explanations when bringing young children). The public domain knowledge graphs provided by institutions (tourism research institutes, local cultural and tourism bureaus, local universities, etc.) include standardized content such as the official history of the scenic spot, building specifications, and safety tips.

[0020] The intelligent platform periodically integrates the acquired private and public knowledge graphs, such as weekly, bi-weekly, or monthly, to form a scenic area knowledge graph that precisely corresponds to the target scenic area.

[0021] S20 extracts the main object data and companion object data from the first real-scene video sent from the mobile terminal, and generates several virtual AI travel companion dolls corresponding to the target scenic spot based on the main object data and companion object data.

[0022] In this step, the solution of the present invention can be presented as a tour guide app on the tourist's mobile device. After entering the target scenic area, the tourist launches the tour guide app, and the front-facing camera is turned on to capture a first real-scene video. After receiving the first real-scene video, the intelligent platform extracts the main subject data and companion data from it to form tourist characteristics. Among them, the main subject is the tourist holding the mobile device, and the main subject data includes the tourist's age, gender, and identity characteristics (such as a photography enthusiast carrying a camera, an art student holding a drawing board), etc.; the companions are those who appear in the first real-scene video with the tourist holding the mobile device, including the number of companions and their relationship with the main subject (such as parents and children, elderly people traveling together), etc.

[0023] Based on the aforementioned tourist characteristics, the intelligent platform generates several virtual AI travel companion avatars, which are then displayed on the tour guide app screen for users to choose from. Figure 2As shown, after a user selects their preferred virtual AI travel companion, that companion is permanently displayed at the top of the tour guide app's screen. The appearance, personality, and language style of these virtual AI companions are tailored to the characteristics of the tourists. For example, the AI ​​companion for families with children is modeled after Nasreddin Hodja, speaking in Nasreddin-esque language; the AI ​​companion for elderly travelers can be a gentle elder, speaking slowly and offering caring suggestions. This transforms what was originally a one-way information transmission into a human-like interaction, significantly enhancing the enjoyment of the tour.

[0024] S30, extract the first scenic spot feature data group from the second real-scene video sent by the mobile terminal, and analyze the tour route of the main object in the target scenic area based on the first scenic spot feature data group and several second scenic spot feature data groups; wherein, the second scenic spot feature data group is a feature data group extracted earlier from other locations in the target scenic area.

[0025] In this step, while touring the target scenic area, tourists use the rear camera of an activated mobile app to capture a second real-scene video. This second real-scene video includes the actual view of the target scenic area that the tourist is pointing at. For example, when a tourist walks to an ancient pavilion, they use the rear camera of the guide app to capture the pavilion and its surrounding environment, thus generating a second real-scene video.

[0026] After receiving the second real-scene video, the intelligent platform uses AI image recognition technology to extract a set of first-spot feature data that uniquely identifies the current scenic spot from the video footage. This includes architectural features, iconic elements, and spatial environment features. Architectural features include, for example, the roof style of the ancient pavilion (hipped roof / pyramidal roof), the number of columns, and the carved patterns. Iconic elements include, for example, specific sculptures within the scenic spot (such as stone lions or human statues), plaque text (such as the "Lake Pavilion" plaque), and unique vegetation (such as a century-old pine tree that exists only at this scenic spot). Spatial environment features include, for example, the location of the scenic spot (lakeside, hillside, center of a square), and surrounding facilities (such as adjacent benches and signs).

[0027] Meanwhile, the tour guide app can remain open throughout the visit, continuously acquiring second-view video footage. For example, as the visitor starts from the entrance and films the ancient bridge, stone tablet, and pavilion, each segment becomes part of the second-view video. Based on these earlier second-view videos, corresponding feature data sets for the second attractions can be extracted and stored in the database of the intelligent platform and / or the mobile device's storage. In other words, the first attraction feature data set represents the visitor's real-time characteristics, while the second attraction feature data set represents the visitor's historical characteristics.

[0028] The intelligent platform pre-stores high-precision digital maps of the target scenic area. These maps are marked with standard geographical location information (such as latitude and longitude, relative coordinates) and corresponding feature anchor points (i.e., the binding relationship between the standard features of each scenic spot and its geographical location). Then, for the first set of scenic spot feature data, it compares it with the standard feature anchor points of each scenic spot in the digital map to find the scenic spot with the highest feature overlap, determining the current real-time geographical location corresponding to that feature set. For example, the Songhe Pavilion feature set is mapped to the latitude and longitude coordinates of Songhe Pavilion on the map. For the second set of scenic spot feature data, each set of historical features is compared with the standard feature anchor points of the digital map to determine the historical visitor geographical location corresponding to each set of features. For example, the ancient bridge feature set is mapped to the coordinates of the ancient bridge, and the stone tablet feature set is mapped to the coordinates of the stone tablet.

[0029] The intelligent platform connects historical geographical locations (the mapping results of the second scenic spot feature data group) sequentially on a high-precision digital map according to time sequence to form the visited trajectory (such as entrance → ancient bridge → stone tablet); then, combined with the current real-time geographical location (the mapping results of the first scenic spot feature data group, such as Songhe Pavilion), it supplements the current location node to form the latest complete tour route for the tourist.

[0030] S40, the first scenic spot feature data group, the tour route, and the scenic area knowledge graph are matched and analyzed to obtain target knowledge data, and target explanation data is generated based on the target knowledge data; wherein, the tour route is related to the tourists' tour preferences.

[0031] In this step, the intelligent platform first correlates and matches the first scenic spot feature data group (real-time features of the current scenic spot) and the complete tour route (including historical trajectory and current location) obtained in step S30 with the scenic spot knowledge graph constructed in step S10, which integrates private domain personalized information and public domain standardized knowledge, to filter out target knowledge data that is suitable for the current scenic spot, itinerary pace, and tourist needs. Then, combined with the attribute features (appearance, language style, etc.) of the AI ​​travel companion mascot selected by the tourist in step S20, the target knowledge data is transformed into target explanation data that fits the scene and interaction needs.

[0032] It should be noted that target scenic areas usually have many attractions, and different tour routes reflect the tourist's tour preferences. This invention infers the tourist's tour preferences based on different tour routes, and then determines the appropriate target knowledge data, thereby providing tourists with a better tour experience.

[0033] The S50 sends the target explanation data to the mobile device, where it is output by the AI ​​travel companion doll selected by the user.

[0034] In this step, the intelligent platform sends the target explanation data generated in step S40 to the tourist's mobile device, and the virtual AI travel companion doll selected by the tourist in step S20 completes the output. That is, the output format is in line with the characteristics of the AI ​​travel companion (such as the corresponding style of voice and the matching visual presentation). Then, the personalized explanation content after screening and conversion is delivered to the tourist, allowing the tourist to conveniently obtain interactive explanation content that suits their own needs.

[0035] This invention enriches the content of explanations by periodically integrating users' self-built private knowledge and public knowledge graphs provided by institutions. It also infers tourists' travel preferences based on their actual tour routes and makes adaptive adjustments to the content of explanations to meet users' personalized needs. At the same time, this invention also generates different virtual AI travel companion dolls for users, which can significantly enhance the interactive fun.

[0036] As an example, the step of periodically fusing user-built private domain knowledge and public domain knowledge graphs provided by institutions to obtain a scenic area knowledge graph corresponding to the target scenic area includes: S11, extracting multi-dimensional features from the private domain knowledge set to obtain a private domain feature set; wherein, the multi-dimensional features include information creation time features, information relevance features, information integrity features, and user interaction frequency features.

[0037] In this step, for the fusion of private domain knowledge sets and public domain knowledge graphs, the present invention adopts a triggered fusion method to reduce the computational load caused by unnecessary fusion.

[0038] Specifically, the intelligent platform collects user-generated private domain knowledge in real time from channels such as mobile apps, social media APIs, and scenic area interactive terminals, forming a private domain knowledge set. Understandably, the data collected at each stage needs to undergo deduplication, using the SimHash algorithm to eliminate similar content with a duplication rate exceeding 85%; then, a rule engine removes low-quality content containing advertisements, sensitive words, etc.; finally, the heterogeneous data is uniformly converted into RDF triple format.

[0039] For the aforementioned private domain knowledge set, the following four types of aggregation features are calculated: 1. Information creation time feature: used to evaluate the overall freshness of the private domain knowledge set. Newly generated knowledge (such as the latest check-in points uploaded by tourists recently, or interesting events that have just happened) is often more timely and valuable for reference, while outdated private information may no longer be applicable (such as a facility that has been demolished), and its integration priority will be relatively low.

[0040] The aggregation feature is calculated as follows: using an exponential decay or linear decay model, the freshness score of any piece of private domain knowledge is calculated. ,in, The difference between the current time and the time the knowledge was created; It is the decay coefficient, used to control the rate at which freshness decreases. The higher the value, the faster the old knowledge decays. Then, the average or weighted average of the freshness scores of all private knowledge in the entire private knowledge spectrum is calculated and used as the information creation time feature.

[0041] 2. Information Relevance Feature: Used to assess the relevance between the private domain knowledge set and the existing public domain knowledge graph of the target scenic area.

[0042] The aggregation feature is calculated as follows: Natural language processing models (such as Word2Vec and BERT) are used to map key entities, concepts, and contexts in the private domain knowledge, as well as nodes in the public domain knowledge graph, into a high-dimensional vector space. The cosine similarity or Euclidean distance between each private domain knowledge vector and all relevant node vectors in the public domain knowledge graph is calculated. The highest set of similarity scores (e.g., the average of the top-3) is taken as the relevance score for that private domain knowledge. The information relevance feature of the entire private domain knowledge set can be represented as the average or median of the relevance scores of all private domain knowledge. A high average means that this batch of new knowledge is generally highly relevant to the scenic area's theme.

[0043] 3. Information Completeness Characteristics: Used to assess the richness and structure of private domain knowledge. A complete piece of knowledge can contain multiple information elements such as subject, attribute, time, and location (e.g., "My grandfather (subject) took a photo with his comrades (relationship) at the Yellow Crane Tower (location) in 1945 (event)"); conversely, an incomplete piece of information may only contain fragments (e.g., "This place is beautiful"). Knowledge with high completeness has greater integration value and can add three-dimensional details to the scenic area's knowledge base.

[0044] The aggregation feature is calculated as follows: It examines the text length of the private domain knowledge, whether it contains key entities (such as attraction names, people, and times), and the complexity of its syntactic structure; and uses a pre-trained text quality assessment model or summarization model to determine the information density and completeness of the text in the private domain knowledge, calculating a completeness score for each piece of private domain knowledge. The information completeness feature of the entire private domain knowledge set is the average of the completeness scores of all knowledge entries.

[0045] 4. User interaction frequency characteristics: Used to assess the community recognition gained by this private domain knowledge set. A piece of private domain knowledge that receives a large number of likes, comments, and shares usually means that it has gained widespread recognition from other visitors, and its value is higher.

[0046] The aggregation feature is calculated as follows: Data such as likes, comments, and shares for each knowledge item is obtained from the source platform API. A weighted formula is used to calculate the popularity score of each knowledge item, for example: Popularity_score = α*log(likes+1) + β*log(comments+1) + γ*(shares). Here, α, β, and γ are weighting coefficients, and logarithmic processing is used to smooth the impact of extreme values. The user interaction frequency feature of the entire private domain knowledge set can be represented as the average or sum of the Popularity_scores of all knowledge items.

[0047] S12, calculate the update degree evaluation value of the private domain knowledge graph set based on the private domain feature set. When the update degree evaluation value is higher than the update threshold, trigger the fusion of the private domain knowledge set with the public domain knowledge graph or the current scenic area knowledge graph.

[0048] In this step, the update score is a weighted average of the features in the private domain feature set, calculated using the formula: Update_Score = W1 * F_fresh + W2 * F_rel + W3 * F_comp + W4 * F_pop, where Update_Score is the update score, F_fresh is the score for the information creation time feature, F_rel is the score for the information relevance feature, F_comp is the score for the information integrity feature, and F_pop is the score for the user interaction frequency feature. W1, W2, W3, and W4 correspond to the weight coefficients of the four feature scores, and satisfy W1 + W2 + W3 + W4 = 1.

[0049] If the calculated Update_Score is higher than the update threshold, the value of this batch of private domain knowledge sets is determined to be high enough, and the subsequent knowledge graph fusion process is immediately triggered. If it is lower than the update threshold, this batch of private domain knowledge sets will not be fused and can be discarded or temporarily stored in a buffer pool, waiting to be merged with subsequently collected data and recalculated for the evaluation value.

[0050] Understandably, the update threshold is a pre-set critical value determined through experimentation and optimization. It can be a fixed value or a dynamically adjusted value, such as automatically fine-tuning it according to the current computing load of the system. When the load is high, the threshold is slightly raised, and vice versa.

[0051] As an example, the process of generating several virtual AI travel companion dolls corresponding to the target scenic spot based on the subject object data and the peer object data includes: S21, performing group profile analysis on the subject object data and the peer object data to generate a group profile tag set; wherein, the group profile tag set includes group type tags, core knowledge preference tags, and interaction style preference tags.

[0052] In this step, the subject data refers to the characteristic data of the tourist holding the mobile device, including age group (e.g., child, youth, middle-aged, elderly), gender (male, female), and explicit identity characteristics (e.g., photography enthusiasts with professional cameras, art students carrying drawing boards, hikers carrying large backpacks) obtained through analysis of the first real-scene video captured by the front-facing camera. The companion data consists of information about tourists traveling with the subject, obtained through analysis of the first real-scene video, including the total number of people in the group, the age and gender of each person, and group relationships inferred through behavior and posture analysis (e.g., families with children, couples, groups of friends, senior citizen tour groups). The subject data and companion data together constitute an initial perception of the current tourist group.

[0053] The intelligent platform's built-in group profiling analysis module (which can be based on a rule engine or a lightweight classification model) comprehensively processes the above data to generate a structured group profiling tag set. This tag set includes: group type tags: macro-classification of tourist groups, such as "families with young children", "middle school study tour groups", "elderly groups", "couples", "business delegations", etc.

[0054] Core knowledge preference tags: Based on group type, infer the knowledge categories that they may be interested in. For example, the "family with children" group may prefer knowledge about "myths and legends", "interesting animals and plants", and "interactive adventures"; "photography enthusiasts" may be more interested in "architectural composition aesthetics", "best shooting locations", and "historical stories of light and shadow".

[0055] Interaction style preference tags: Define the preferred information presentation methods based on group type. For example, for "families with children", the interaction style should be "lively and cute" and "guided and inspiring"; for "elderly groups", it should be "slow and gentle" and "caring and reminding"; for "young couples", it can be "romantic and poetic" and "gentle and friendly".

[0056] S22, the group profile tag set is matched with the preset AI travel companion database to select a number of candidate AI travel companion dolls; wherein, the AI ​​travel companion database stores multiple AI travel companion doll templates, and each AI travel companion doll template is pre-configured with image identifiers, knowledge domain tags, personality and language tags and scene fit tags.

[0057] In this step, the intelligent platform has a pre-built AI travel companion database, which stores a series of AI travel companion doll templates. Each template is a data entity with the following pre-configured attributes: image identifier: the doll's name (such as "Avanti", "Little Poet Immortal", "Naan Bread", etc.) and its corresponding visual resource index.

[0058] Knowledge domain tag: Define the types of knowledge that this doll is good at explaining, such as [ancient poetry, anecdotes of literati], [architectural history, construction techniques], [botany encyclopedia, geological evolution].

[0059] Personality Language Tags: Define the interaction style of this doll, such as [learned and calm], [playful and lively], [gentle and approachable].

[0060] Scene fit label: Indicates the degree of relevance between the doll and a specific attraction or scene within the scenic area (e.g., the "Beilin Guide" doll is highly compatible with the stone carving scenic area).

[0061] The group profile tag set generated in step S21 is used as a query vector, and multi-dimensional similarity calculations (e.g., using cosine similarity algorithm) are performed between it and the knowledge domain tag, personality and language tag, and scene suitability tag of each doll template in the database. The calculated comprehensive matching scores are then sorted in descending order, and the top N (e.g., ...) are selected. Figure 2 The six AI travel companion dolls shown are candidates for AI travel companion dolls.

[0062] S23, send the plurality of candidate AI travel companion dolls to the mobile terminal, and receive the user selection instruction returned by the mobile terminal to determine the target AI travel companion doll.

[0063] In this step, the intelligent platform sends the shortlisted candidate AI travel companion mascots (including their logos and brief descriptions) to the user's mobile tour guide app. The tour guide app presents these candidate mascots in a visual interface (such as clickable cartoon cards) for users to browse and select. Users choose their favorite mascot through touchscreen clicks or other methods. The mobile app then sends this user selection command back to the intelligent platform. Based on this command, the intelligent platform ultimately determines that the mascot corresponding to that ID is the target AI travel companion mascot used for this tour guide service.

[0064] Subsequently, all narration content will be generated and output based on the personality and language tags and knowledge domain tags of this target AI travel companion doll, thereby achieving a highly personalized interactive tour experience.

[0065] As an example, the first scenic spot feature data group, the tour route and the scenic area knowledge graph are matched and analyzed to obtain target knowledge data, including: S41, based on the first scenic spot feature data group, entity matching is performed in the scenic area knowledge graph to determine the current scenic spot entity and its associated entity set; wherein, the associated entity set includes cultural event entities, historical figure entities, architectural component entities and natural landscape entities that have a direct or indirect relationship with the current scenic spot entity.

[0066] In this step, the first set of scenic spot feature data includes second real-scene videos continuously collected through the mobile device's rear camera, architectural features (such as roof style and number of columns), iconic element features (such as specific sculptures and plaque text), and spatial environment features (such as geographical location and surrounding facilities) extracted using AI image recognition technology. The intelligent platform compares these feature data with the standard feature anchor points of each scenic spot in the pre-stored scenic area knowledge graph for similarity. The scenic spot with the highest matching degree is identified as the current scenic spot entity (such as "Yuehu Pavilion").

[0067] Subsequently, using the current scenic spot entity as the central node, the knowledge graph is traversed in one or more hops along relational edges such as "built in", "involved in", "related people", and "located in", retrieving and collecting all directly or indirectly related entities to form a set of related entities. For example, for "Yuehu Pavilion", its set of related entities includes: cultural event entities: such as "Emperor Qianlong's Southern Inspection Tour and Poetry Inscription Event"; historical figure entities: such as "Emperor Qianlong" and "Designer Yangshi Lei"; architectural component entities: such as "hip roof", "white marble railings", and "relief patterns"; and natural landscape entities: such as "West Lake water area" and "lotus grove".

[0068] S42, based on the tour route, extract the historical tour entity sequence from the scenic area knowledge graph, and calculate the tourist tour rhythm mode by combining real-time location information; wherein, the tour rhythm mode includes fast browsing mode, regular tour mode and in-depth exploration mode.

[0069] In this step, tourist behavior data is analyzed to infer their tour preferences and pace.

[0070] Specifically, the intelligent platform maps the tour route obtained in step S30 (e.g., scenic area entrance → Broken Bridge in Snow → Bai Causeway → Autumn Moon over the Calm Lake → Yuehu Pavilion) back to entities in the scenic area's knowledge graph, resulting in a historical tour entity sequence arranged chronologically. Furthermore, combining real-time location information, it analyzes the dwell time at each attraction in this historical tour entity sequence (calculated based on the timestamps of entry and exit from each attraction area) and the average movement speed between attractions. Based on predefined threshold rules, it dynamically calculates the tourist's tour rhythm pattern, including: a fast browsing mode: short average dwell time (e.g., <3 minutes) and fast movement speed. This indicates that the tourist may be short on time or not very interested in details.

[0071] In-depth exploration mode: Long average stay (e.g., >10 minutes), slow movement speed. This indicates that tourists have a strong interest in this type of attraction and are willing to explore it in depth.

[0072] Typical sightseeing pattern: The duration of stay and the speed of movement are somewhere in between, which is the general state for most tourists.

[0073] The tour rhythm pattern determined above is a key dynamic parameter that determines the depth and breadth of knowledge explanation.

[0074] S43, based on the tour rhythm pattern, select a subset of entities from the associated entity set that match the current tour context, and extract attributes from the scenic area knowledge graph based on the entity subset to generate multi-granularity knowledge fragments; wherein, the multi-granularity knowledge fragments include core knowledge points, extended knowledge points, and interesting knowledge points.

[0075] In this step, based on the tour rhythm mode, the most suitable subset of entities is selected from the set of related entities. Specifically, in the fast browsing mode, only the most essential entities are selected from the set of related entities to form a minimal subset. This subset typically contains only the current attraction entity itself and its most direct and crucial related entity (such as the most important historical event or figure).

[0076] In deep exploration mode, most entities in the associated entity set are retained, forming a rich subset of entities. In addition, this mode triggers deep queries on the knowledge graph, which can retrieve more granular association information of each entity in the entity subset (such as the craftsmanship details of building components, academic controversies about historical events, etc.).

[0077] For the regular tour mode, a compromise strategy is adopted, which selects all directly related entities and some important indirectly related entities from the set of related entities to form a subset of entities that is comprehensive in content and of appropriate length.

[0078] After determining the subset of entities to be explained, each entity in this subset is used as an index key to query the scenic area's knowledge graph, extracting all attribute data corresponding to these entities (such as detailed text descriptions, high-resolution images, historical document summaries, etc.). Subsequently, this extracted raw data is intelligently organized and packaged into knowledge packages of different granularities according to information depth, importance, and interest. Specifically, this includes: Core knowledge points: containing the most basic and crucial factual descriptions of the entities (such as "name," "era," and "function"). This is the basic information that must be output.

[0079] Extended knowledge points: These include detailed background information, deeper meanings, and technical details of the entity (such as "the evolution of architectural styles," "the specific process of historical events," and "the life and deeds of the figures"). The level of detail in the output is dynamically controlled by the tour pace mode.

[0080] Fun Facts: Includes anecdotes, legends, and unsolved mysteries related to the site (such as "local folk tales" or "interesting stories about a famous person"). Primarily used to enhance the appeal of the tour, it can be added to both regular tour and in-depth exploration modes.

[0081] S44, the multi-granularity knowledge fragments are adaptively fused with the knowledge domain tags and personality language tags of the AI ​​travel companion doll to generate the target knowledge data.

[0082] This step ensures that the generated knowledge is not only accurate in content but also perfectly matches the style of the virtual travel companion character selected by the user. The specific implementation is as follows: Knowledge Domain Adaptation: Pre-configured knowledge domain tags (such as [ancient architecture, historical chronology]) are read from the AI ​​travel companion doll template. Subsequently, in the multi-granularity knowledge fragments generated in step S43, content matching these tags is prioritized, highlighted, and even deepened and expanded. For example, for a doll with the knowledge domain of "architectural expert," the focus is on rendering details about architectural construction in the "extended knowledge points"; while for a "storyteller" doll, "fun knowledge points" are prioritized.

[0083] Personality and Language Rendering: Based on the AI ​​travel companion doll's personality and language tags (such as [learned and composed], [playful and lively]), the organized knowledge text undergoes final stylized translation and packaging: For the [playful and lively] type doll, colloquial and emotionally charged sentences are used, incorporating interjections, questions, and internet slang, such as "Wow! Everyone, look at this roof! Isn't it super cool! This is called a 'hipped roof,' doesn't it look like several small hills put together?"

[0084] For the "learned and composed" type of doll, a formal, logically rigorous, and professionally precise language style is used. For example, "Yuehu Pavilion adopts the hip roof form, which is of a high rank in traditional Chinese architecture. Its structural features include one main ridge, four hip ridges, and four gable ridges, and it is often used in important buildings such as palaces and temples."

[0085] After the above two processes of content selection and stylized rendering, the output structured data is the final target knowledge data. This data package not only contains accurately adapted explanatory content, but also defines the final narrative tone and style of the AI ​​travel companion doll.

[0086] As an example, the step of extracting the historical tour entity sequence from the scenic area knowledge graph and calculating the tourist tour rhythm pattern by combining real-time location information includes: S421, performing reverse geocoding matching in the scenic area knowledge graph based on the geographical coordinate sequence contained in the tour route to obtain the historical tour entity sequence arranged in chronological order.

[0087] In this step, the intelligent platform receives the tour route generated in step S30, which consists of a series of latitude and longitude coordinates sorted by timestamps. It then invokes pre-stored geofencing or spatial indexing services in the scenic area's knowledge graph to associate each coordinate point with the nearest or best-matching attraction entity in the knowledge graph (i.e., reverse geocoding). For example, coordinates (120.123, 30.456) are matched to the "Broken Bridge in Snow" entity in the knowledge graph.

[0088] Once a match is successful, a semantically rich sequence of historical tour entities is formed, arranged chronologically, for example: [timestamp 1: Scenic Area Entrance, timestamp 2: Broken Bridge in Snow, timestamp 3: Bai Causeway, timestamp 4: Autumn Moon over the Calm Lake]). This sequence precisely describes the tourist's tour path up to the current moment.

[0089] S422, Analyze the dwell time and movement speed between attractions in the historical tour entity sequence; based on the data of the companion objects in the first real-scene video and each of the second real-scene videos, analyze the continuous appearance of the companion objects.

[0090] In this step, the historical visitor entity sequence is traversed, and the dwell time at each attraction is calculated based on the entry and exit timestamps of each entity (dwell time = departure time - entry time). The average movement speed is calculated based on the geographical distance between two consecutive attractions and the time taken to move between them. The calculated dwell time and movement speed sequences constitute a core set of quantitative indicators describing tourist behavior habits. For example, a visitor might spend 15 minutes at "Yuehu Pavilion" and move from "Bai Causeway" to "Pinghu Qiuyue" at a speed of 1.2 meters per second.

[0091] Simultaneously, the system analyzes the first real-scene video (front-facing camera, primarily capturing the user and their companions) and the second real-scene video (rear-facing camera, primarily capturing scenic spots, but may also capture companions to the side, rear, or surroundings). Using real-time object detection and continuous tracking algorithms integrated into the video stream (such as a deep learning-based YOLO model combined with the Deep SORT algorithm), pedestrians in the footage are identified and tracked. Special attention is paid to the features of companions identified in step S20 (such as clothing color, body shape, facial features, etc.).

[0092] By analyzing consecutive video frames, the frequency of a specific object appearing in the frame or the number / duration of consecutively missing frames can be calculated. For example, if "object A" appears in video frames for a total duration of 55 seconds in the past 60 seconds, it is determined to be "continuously appearing".

[0093] S423, Based on the dwell time, the movement speed, and the continuous occurrence, calculate the tourist tour rhythm pattern using a pre-trained classification model.

[0094] In this step, the statistical features such as average dwell time and average movement speed of multiple attractions obtained in step S422 are combined with the features of continuous presence of companions obtained in step S423 (e.g., number of valid companions, percentage of appearance time in the last 30 seconds, etc.) to construct a multi-dimensional feature vector. This multi-dimensional feature vector is then input into a pre-trained multi-classification model (such as Gradient Boosting Decision Tree (GBDT), Random Forest, or Support Vector Machine (SVM). This model has been trained on massive amounts of historical tourist behavior data and its labeled rhythm patterns, and has learned the complex, non-linear mapping relationship between features and patterns. The multi-classification model outputs any one of the following modes: Fast browsing mode: Features include short dwell time and high movement speed. In this case, the continuous presence of companions is used to help determine whether this is a fast-moving group (companions continuously appear) or an individual leaving in a hurry (companions have disappeared).

[0095] Deep Exploration Mode: Characterized by long dwell times and low movement speed. In this mode, the continuous presence of fellow travelers helps to bolster confidence – if companions also linger nearby for extended periods, it further confirms a group-based deep exploration.

[0096] For example, when a prolonged stay and low movement speed are detected, and video analysis reveals that fellow travelers continue to appear in the frame (i.e., they are also staying in the same area), the current activity is determined to be a group-based in-depth exploration mode with high confidence. However, if a prolonged stay and low movement speed are detected, but video analysis shows that fellow travelers have disappeared from the frame for some time (i.e., other companions have left the area, leaving only the main tourist), this behavior is more likely due to personal reasons for stagnation (such as resting, daydreaming, waiting, or getting separated from the group), rather than genuine in-depth exploration. In this case, a knowledge-pushing strategy that does not trigger in-depth exploration mode or triggers it with low weight will be avoided to prevent inappropriate and lengthy explanations when the user does not need in-depth information. Instead, a lightweight interaction strategy can be adopted, such as having an AI travel companion ask, "Do you need to rest here a little longer?", or providing simple route guidance.

[0097] Standard tour mode: Characteristics fall between the two. The consistent presence of fellow travelers is an important indicator of a normal group tour.

[0098] In this embodiment, by introducing the key context of the status of peers, the system effectively distinguishes between scenarios that are prone to misjudgment, such as a fast-paced team versus an individual falling behind, or a team exploring deeply versus an individual daydreaming, making the pattern recognition results more consistent with reality.

[0099] like Figure 3As shown, this embodiment of the invention also discloses a scenic area AI intelligent guide system 200, which is applied to an intelligent platform and includes: a knowledge graph generation module 11, which periodically merges user-built private domain knowledge and public domain knowledge graphs provided by institutions to obtain a scenic area knowledge graph corresponding to the target scenic area.

[0100] The virtual avatar generation module 12 extracts the main object data and companion object data from the first real-scene video sent from the mobile terminal, and generates several virtual AI travel companion dolls corresponding to the target scenic spot based on the main object data and companion object data.

[0101] The tour route generation module 13 extracts a first scenic spot feature data group from the second real-scene video sent by the mobile terminal, and analyzes the tour route of the main object in the target scenic area based on the first scenic spot feature data group and several second scenic spot feature data groups; wherein, the second scenic spot feature data group is a feature data group extracted earlier from other locations in the target scenic area.

[0102] The explanation data generation module 14 matches and analyzes the first scenic spot feature data group, the tour route, and the scenic area knowledge graph to obtain target knowledge data, and generates target explanation data based on the target knowledge data; wherein, the tour route is related to the tourists' tour preferences.

[0103] The data transmission module 15 transmits the target explanation data to the mobile device, where it is output by the AI ​​travel companion doll selected by the user.

[0104] As an example, the knowledge graph generation module 11 specifically: extracts multi-dimensional features from the private domain knowledge set to obtain a private domain feature set; wherein, the multi-dimensional features include information creation time features, information relevance features, information integrity features, and user interaction frequency features; calculates the update degree evaluation value of the private domain knowledge graph set based on the private domain feature set, and when the update degree evaluation value is higher than the update threshold, triggers the fusion of the private domain knowledge set with the public domain knowledge graph or the current scenic area knowledge graph.

[0105] As an example, the virtual avatar generation module 12 specifically: performs group profile analysis on the subject object data and peer object data to generate a group profile tag set; wherein, the group profile tag set includes group type tags, core knowledge preference tags, and interaction style preference tags; matches the group profile tag set with a preset AI travel companion database to filter out several candidate AI travel companion dolls; wherein, the AI ​​travel companion database stores multiple AI travel companion doll templates, each AI travel companion doll template is pre-configured with image identifiers, knowledge domain tags, personality and language tags, and scene suitability tags; sends the several candidate AI travel companion dolls to the mobile terminal and receives the user selection instruction returned by the mobile terminal to determine the target AI travel companion doll.

[0106] As an example, the explanation data generation module 14 specifically: based on the first scenic spot feature data group, performs entity matching in the scenic area knowledge graph to determine the current scenic spot entity and its associated entity set; wherein, the associated entity set includes cultural event entities, historical figure entities, architectural component entities, and natural landscape entities that have a direct or indirect relationship with the current scenic spot entity; based on the tour route, extracts the historical tour entity sequence from the scenic area knowledge graph, and calculates the tourist tour rhythm mode in combination with real-time location information; wherein, the tour rhythm mode includes a fast browsing mode, a regular tour mode, and an in-depth exploration mode; according to the tour rhythm mode, selects a subset of entities that conform to the current tour context from the associated entity set, and extracts attributes from the scenic area knowledge graph based on the entity subset to generate multi-granularity knowledge fragments; wherein, the multi-granularity knowledge fragments include core knowledge points, extended knowledge points, and interesting knowledge points; adaptively fuses the multi-granularity knowledge fragments with the knowledge domain tags and personality language tags of the AI ​​travel companion doll to generate the target knowledge data.

[0107] As an example, the explanation data generation module 14 specifically: based on the geographical coordinate sequence included in the tour route, performs reverse geocoding matching in the scenic area knowledge graph to obtain a historical tour entity sequence arranged in chronological order; analyzes the dwell time of each attraction in the historical tour entity sequence and the movement speed between attractions; analyzes the continuous appearance of the accompanying objects based on the data of the accompanying objects in the first real-scene video and each of the second real-scene videos; and calculates the tourist tour rhythm pattern through a pre-trained classification model based on the dwell time, the movement speed, and the continuous appearance.

[0108] This invention also discloses an electronic device, including a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, implements the method described in the preceding embodiment.

[0109] This invention also discloses a non-transient computer-readable storage medium having instructions embodied thereon that can be executed by one or more processors to perform the method described in any of the preceding embodiments.

[0110] Although the invention has been specifically shown and described with reference to preferred embodiments, those skilled in the art will understand that various modifications in form and detail may be made without departing from the spirit and scope of the invention. Accordingly, the disclosed invention should be considered merely illustrative and limited only by the scope specified in the appended claims.

Claims

1. A scenic area AI-powered intelligent navigation method, applied to an intelligent platform, characterized in that: Includes the following steps: The user-built private domain knowledge and the public domain knowledge graph provided by the institution are periodically integrated to obtain the scenic area knowledge graph corresponding to the target scenic area. The main object data and companion object data are extracted from the first real-scene video sent by the mobile device through the front-facing camera. Based on the main object data and companion object data, several virtual AI travel companion dolls corresponding to the target scenic spot are generated. The main object is the tourist who owns the mobile device, and the companion object is the person who appears in the first real-scene video with the tourist who owns the mobile device. The first scenic spot feature data group is extracted from the second real-scene video taken by the rear camera sent from the mobile terminal. Based on the first scenic spot feature data group and several second scenic spot feature data groups, the tour route of the main object in the target scenic area is analyzed. The second scenic spot feature data group is a feature data group extracted earlier from other locations in the target scenic area. The first scenic spot feature data group, the tour route, and the scenic area knowledge graph are matched and analyzed to obtain target knowledge data, and target explanation data is generated based on the target knowledge data; wherein, the tour route is related to the tourists' tour preferences; The target explanation data is sent to the mobile device and output by the AI ​​travel companion doll selected by the user; The first scenic spot feature data group, the tour route, and the scenic area knowledge graph are matched and analyzed to obtain target knowledge data, including: Based on the first scenic spot feature data set, entity matching is performed in the scenic area knowledge graph to determine the current scenic spot entity and its associated entity set; wherein, the associated entity set includes cultural event entities, historical figure entities, architectural component entities and natural landscape entities that have a direct or indirect relationship with the current scenic spot entity; Based on the tour route, historical tour entity sequences are extracted from the scenic area knowledge graph, and the tour rhythm pattern of tourists is calculated by combining real-time location information; wherein, the tour rhythm pattern includes fast browsing mode, regular tour mode and in-depth exploration mode; Based on the tour rhythm pattern, a subset of entities that match the current tour context is selected from the set of associated entities, and attributes are extracted from the scenic area knowledge graph based on the entity subset to generate multi-granularity knowledge fragments; wherein, the multi-granularity knowledge fragments include core knowledge points, extended knowledge points, and interesting knowledge points; The multi-granularity knowledge fragments are adaptively fused with the knowledge domain tags and personality language tags of the AI ​​travel companion doll to generate the target knowledge data; Historical visitor entity sequences are extracted from the scenic area knowledge graph, and visitor visit rhythm patterns are calculated by combining real-time location information, including: Based on the sequence of geographic coordinates contained in the tour route, reverse geocoding matching is performed in the scenic area knowledge graph to obtain a sequence of historical tour entities arranged in chronological order; Analyze the dwell time at each attraction and the movement speed between attractions in the historical visit sequence; Based on the data of peer objects in the first real-scene video and each of the second real-scene videos, analyze the continuous occurrence of peer objects; Based on the dwell time, the movement speed, and the continuous occurrence, a tourist tour rhythm pattern is calculated using a pre-trained classification model.

2. The AI-powered intelligent navigation method for scenic areas according to claim 1, characterized in that: The system periodically merges user-built private knowledge graphs with public knowledge graphs provided by institutions to derive a scenic area knowledge graph corresponding to the target scenic area, including: Multi-dimensional features are extracted from the private domain knowledge set to obtain the private domain feature set; wherein, the multi-dimensional features include information creation time features, information relevance features, information integrity features, and user interaction frequency features; The update degree evaluation value of the private domain knowledge graph set is calculated based on the private domain feature set. When the update degree evaluation value is higher than the update threshold, the fusion of the private domain knowledge set with the public domain knowledge graph or the current scenic area knowledge graph is triggered.

3. The AI-powered intelligent tour guide method for scenic areas according to claim 2, characterized in that: Based on the subject object data and peer object data, several virtual AI travel companion dolls corresponding to the target scenic spot are generated, including: Perform group profiling analysis on the main object data and peer object data to generate a group profiling tag set; wherein, the group profiling tag set includes group type tags, core knowledge preference tags, and interaction style preference tags; The group profile tag set is matched with the preset AI travel companion database to filter out a number of candidate AI travel companion dolls; wherein, the AI ​​travel companion database stores multiple AI travel companion doll templates, and each AI travel companion doll template is pre-configured with image identifiers, knowledge domain tags, personality and language tags and scene fit tags. The candidate AI travel companion dolls are sent to the mobile terminal, and the user selection instruction returned by the mobile terminal is received to determine the target AI travel companion doll.

4. A scenic area AI intelligent navigation system, applied to an intelligent platform, characterized in that, include: The knowledge graph generation module periodically merges user-built private domain knowledge with public domain knowledge graphs provided by institutions to generate a scenic area knowledge graph corresponding to the target scenic area. The virtual avatar generation module extracts the main object data and companion object data from the first real-scene video sent from the mobile terminal, and generates several virtual AI travel companion dolls corresponding to the target scenic spot based on the main object data and companion object data. The tour route generation module extracts a first scenic spot feature data group from the second real-scene video sent by the mobile terminal, and analyzes the tour route of the main object in the target scenic area based on the first scenic spot feature data group and several second scenic spot feature data groups; wherein, the second scenic spot feature data group is a feature data group extracted earlier from other locations in the target scenic area. The explanation data generation module matches and analyzes the first scenic spot feature data group, the tour route, and the scenic area knowledge graph to obtain target knowledge data, and generates target explanation data based on the target knowledge data; wherein, the tour route is related to the tourists' tour preferences; The data transmission module sends the target explanation data to the mobile device, where it is output by the AI ​​travel companion doll selected by the user. The data generation module specifically performs the following steps: Based on the first scenic spot feature data group, entity matching is performed in the scenic area knowledge graph to determine the current scenic spot entity and its associated entity set; wherein, the associated entity set includes cultural event entities, historical figure entities, architectural component entities, and natural landscape entities that have a direct or indirect relationship with the current scenic spot entity; based on the tour route, a historical tour entity sequence is extracted from the scenic area knowledge graph, and a tourist tour rhythm pattern is calculated in combination with real-time location information; wherein, the tour rhythm pattern includes a fast browsing mode, a regular tour mode, and an in-depth exploration mode; according to the tour rhythm pattern, a subset of entities that conform to the current tour context is selected from the associated entity set, and attributes are extracted from the scenic area knowledge graph based on the entity subset to generate multi-granularity knowledge fragments; wherein, the multi-granularity knowledge fragments include core knowledge points, extended knowledge points, and interesting knowledge points; the multi-granularity knowledge fragments are adaptively fused with the knowledge domain tags and personality language tags of the AI ​​travel companion doll to generate the target knowledge data; The data generation module specifically performs the following steps: based on the geographical coordinate sequence included in the tour route, it performs reverse geocoding matching in the scenic area knowledge graph to obtain a historical tour entity sequence arranged in chronological order; it analyzes the dwell time at each attraction in the historical tour entity sequence and the movement speed between attractions; based on the data of companion objects in the first real-scene video and each of the second real-scene videos, it analyzes the continuous appearance of companion objects; and based on the dwell time, the movement speed, and the continuous appearance, it calculates the tourist tour rhythm pattern through a pre-trained classification model.

5. The scenic area AI intelligent guide system according to claim 4, characterized in that: The knowledge graph generation module, specifically: Multi-dimensional features are extracted from the private domain knowledge set to obtain the private domain feature set; wherein, the multi-dimensional features include information creation time features, information relevance features, information integrity features, and user interaction frequency features; The update degree evaluation value of the private domain knowledge graph set is calculated based on the private domain feature set. When the update degree evaluation value is higher than the update threshold, the fusion of the private domain knowledge set with the public domain knowledge graph or the current scenic area knowledge graph is triggered.

6. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, implements the method described in any one of claims 1-3.

7. A non-transient computer-readable storage medium having instructions embodied thereon, characterized in that, The instructions can be executed by one or more processors to perform the method for implementing any one of claims 1-3.

8. A computer program product comprising an executable computer program, characterized in that: When the computer program is executed, it implements the method as described in any one of claims 1-3.

Images