Artificial intelligence-based exhibition hall interactive display method and system

Abstract

The application discloses an exhibition hall interactive display method and system based on artificial intelligence, and belongs to the technical field of intelligent exhibition halls. The method comprises the following steps: determining a target exhibit currently focused on by a user according to real-time position information; when it is detected that the length of time that the user stays in front of the target exhibit exceeds a preset time threshold, updating a dynamic interest vector associated with the user based on a preset knowledge label vector of the target exhibit and the length of time that the user stays; when the user enters an interactive area corresponding to a new target exhibit, acquiring the updated dynamic interest vector of the user, and determining a target content level to be displayed based on the dynamic interest vector and a knowledge label vector of the new target exhibit; and displaying content on a display terminal associated with the new target exhibit according to the target content level. The application quantifies the implicit behavior of the user into an interest state in real time through the construction of a dynamic feedback loop, and solves the technical problems of high interaction delay, rigid content and incoherent experience in the prior art.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an interactive exhibition method and system for exhibition halls based on artificial intelligence. Background Technology

[0002] Modern exhibition halls, such as museums, science and technology museums, and art galleries, are increasingly adopting digital multimedia technologies to enhance visitor interaction. Traditional interactive displays typically rely on user-initiated actions, such as clicking on touchscreens, scanning QR codes, or using designated applications. While these methods enrich the presentation of information to some extent, they generally suffer from high latency, rigid content, and disjointed experiences. Users must first interrupt their natural visitor flow to perform explicit actions to obtain information, creating an unavoidable delay between the generation of user interest and information acquisition.

[0003] Furthermore, the content provided by the system is often pre-set and homogeneous, unable to dynamically adjust according to individual differences among visitors (such as knowledge background and interests), resulting in inefficient information delivery. More importantly, user interactions are isolated; the system lacks the ability to remember user visit paths and historical interests, failing to create coherent and progressive knowledge guidance, thus diminishing the experiential value of in-depth visits. Therefore, how to design an interactive display solution that can respond to users' implicit interests in real time and seamlessly, and provide dynamic and personalized content, is a pressing technical problem that needs to be solved in the field of smart exhibition halls. Summary of the Invention

[0004] This application aims to improve the technical problems of interactive displays in exhibition halls, such as interaction delays, rigid content, and inconsistent experiences.

[0005] In a first aspect, this application provides an artificial intelligence-based interactive exhibition method, the method comprising: acquiring location information associated with a user, representing the user's real-time location within an exhibition hall; determining, based on the location information, a target exhibit currently being viewed by the user in a preset exhibit interaction area; performing time decay processing on a dynamic interest vector associated with the user during the period the user is viewing the target exhibit; when the user is detected leaving the exhibit interaction area corresponding to the target exhibit, acquiring the duration of the user's viewing of the target exhibit, and if the duration of viewing exceeds a preset duration threshold, updating the dynamic interest vector after time decay processing based on a preset knowledge tag vector of the target exhibit and the duration of viewing; when the user enters an exhibit interaction area corresponding to a new target exhibit, acquiring the updated dynamic interest vector of the user, and determining a target content level to be displayed based on the dynamic interest vector and the knowledge tag vector of the new target exhibit; and displaying content on a display terminal associated with the new target exhibit according to the target content level.

[0006] Optionally, the step of performing time decay processing on a dynamic interest vector associated with the user includes: exponentially decaying each dimension component of the dynamic interest vector according to a preset time step and based on a preset global interest decay coefficient, so as to reduce the influence of historical interests on the current state.

[0007] Optionally, updating the interest gain of the dynamic interest vector after time decay processing based on the preset knowledge tag vector of the target exhibit and the dwell time includes: determining a behavior gain coefficient positively correlated with the dwell time through a nonlinear gain function based on the dwell time; combining the behavior gain coefficient with the knowledge tag vector of the target exhibit to generate an interest gain vector; and summing the interest gain vector with the dynamic interest vector after time decay processing to complete the interest gain update.

[0008] Optionally, the nonlinear gain function is a logarithmic function, and the determination of the behavior gain coefficient includes: performing a logarithmic operation on the ratio of the dwell time to a normalized time constant to obtain the behavior gain coefficient, so that the contribution of the dwell time to the interest gain exhibits a diminishing marginal effect.

[0009] Optionally, determining a target content level to be displayed based on the dynamic interest vector and the knowledge tag vector of the new target exhibit includes: calculating the vector similarity between the dynamic interest vector and the knowledge tag vector of the new target exhibit to generate a matching score; comparing the matching score with multiple preset content level thresholds, and selecting the preset content level corresponding to the interval to which the matching score belongs as the target content level.

[0010] Optionally, the vector similarity is cosine similarity.

[0011] Optionally, determining a target content level to be displayed includes: obtaining the previous exhibit that the user followed before entering the new target exhibit, which serves as the previous interaction node; determining a narrative link strength based on a preset narrative link relationship between the previous exhibit and the new target exhibit; determining an interest matching score based on the vector similarity between the dynamic interest vector and the knowledge tag vector of the new target exhibit; combining the narrative link strength and the interest matching score with preset weights to generate a final recommendation score; and determining the target content level to be displayed based on the final recommendation score.

[0012] Optionally, the method further includes: when the narrative link strength is greater than a preset link threshold, additional guiding information is displayed on the display terminal to explain the narrative link relationship between the previous exhibit and the new target exhibit.

[0013] Secondly, this application provides an artificial intelligence-based interactive exhibition system for exhibition halls. The system includes: a data acquisition module for acquiring location information associated with a user, representing the user's real-time location within the exhibition hall; and a user status management module connected to the data acquisition module, used to determine the target exhibit currently being viewed by the user within a preset exhibit interaction area based on the location information; to perform time decay processing on a dynamic interest vector associated with the user during the period the user views the target exhibit; and, when the user is detected leaving the exhibit interaction area corresponding to the target exhibit, to acquire the duration of the user's attention on the target exhibit; if the duration of attention is greater than... A preset duration threshold is defined, and based on the preset knowledge tag vector of the target exhibit and the dwell time, the dynamic interest vector after time decay processing is updated with interest gain. A content matching and generation module, connected to the user state management module, is used to obtain the updated dynamic interest vector of the user when the user enters the exhibit interaction area corresponding to a new target exhibit, and determine a target content level to be displayed based on the dynamic interest vector and the knowledge tag vector of the new target exhibit. A display terminal module, connected to the content matching and generation module, is used to display content on a display terminal associated with the new target exhibit according to the target content level.

[0014] Optionally, the user state management module is further configured to: exponentially decay each dimension component of the dynamic interest vector according to a preset time step and based on a preset global interest decay coefficient, so as to reduce the influence of historical interests on the current state.

[0015] Optionally, the user status management module is further configured to: determine a behavior gain coefficient positively correlated with the dwell time based on the dwell time using a nonlinear gain function; combine the behavior gain coefficient with the knowledge tag vector of the target exhibit to generate an interest gain vector; and perform vector summation between the interest gain vector and the dynamic interest vector after time decay processing to complete the interest gain update.

[0016] Optionally, the content matching and generation module is further configured to: calculate the vector similarity between the dynamic interest vector and the knowledge tag vector of the new target exhibit to generate a matching score; and compare the matching score with multiple preset content level thresholds, and select the preset content level corresponding to the interval to which the matching score belongs as the target content level.

[0017] Optionally, the system further includes: a narrative path generation module, used to store preset narrative link relationships between multiple exhibits, and when requested by the content matching and generation module, to provide the narrative link strength between the previous exhibit that the user was interested in before entering the new target exhibit and the new target exhibit; and the content matching and generation module is further used to: obtain the narrative link strength, and combine it with an interest matching score determined based on vector similarity using preset weights to generate a final recommendation score, and determine the target content level to be displayed based on the final recommendation score.

[0018] This application's embodiments construct a dynamic feedback loop based on user implicit behaviors (location, dwell time), mapping the user's visit process in real-time and quantitatively onto a dynamic interest vector. This vector simulates the natural forgetting of interests through continuous time decay and is specifically enhanced by the user's dwelling behavior in front of specific exhibits. This mechanism enables the system to track user interest drift in real time and possesses "short-term memory" capabilities across exhibits. When a user approaches a new exhibit, the system can proactively and predictively push the most relevant content levels based on this dynamically updated and personalized interest state, thereby transforming the traditional "request-response" interaction into a "perception-push" interaction. This method solves the interaction delay problem caused by reliance on user-initiated operations in existing technologies, significantly improves information acquisition efficiency through automated and seamless content push, solves the problem of rigid and generic content by dynamically generating content highly matched to user interests, and solves the problem of isolated user interaction history and disjointed experience through the continuous evolution of the interest vector. Ultimately, this application can improve the intelligence and personalization level of interactive displays in exhibition halls and optimize the overall visitor experience. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0020] Figure 1 This is a flowchart illustrating the AI-based interactive exhibition method for exhibition halls as described in Embodiment 1 of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0023] Example 1 This embodiment provides an AI-based interactive exhibition method. In one specific implementation, this method employs a dynamic interest vector model that updates in real time based on users' implicit behaviors. This allows it to seamlessly and automatically detect shifts in user interests and proactively push personalized exhibit content. This method improves upon the technical problems commonly found in existing interactive exhibition systems, such as high interaction latency, rigid content presentation, and inconsistent user experience, thereby enhancing the intelligence and personalization of the exhibition visitor experience.

[0024] Reference Figure 1 The method described in this embodiment may specifically include the following steps: S100: Obtain location information associated with a user, representing the user's real-time location within the exhibition hall; based on the location information, determine the target exhibit that the user is currently interested in within a preset exhibit interaction area.

[0025] In one implementation scenario, when a user enters the exhibition hall, the system assigns them a unique user identifier, such as User_ID, that remains unchanged throughout the visit. This User_ID can be bound to a small, low-power positioning beacon (or positioning tag) worn by the user. This positioning beacon is the physical carrier for achieving high-precision location information acquisition.

[0026] Location information can be acquired using various mature indoor positioning technologies. These include, but are not limited to, Ultra-Wideband (UWB) positioning, Angle of Arrival (AoA) based Bluetooth positioning, or Wi-Fi fingerprint positioning. In a preferred embodiment, UWB positioning technology is employed. This technology involves deploying multiple UWB positioning base stations on the ceiling or walls of the exhibition hall. These base stations accurately measure the time difference (TDoA) between the arrival times of the UWB signal pulses emitted by the user's beacon at different base stations. A central positioning calculation engine processes this time difference data to calculate the precise coordinates of the user's beacon in three-dimensional space—the location information—with extremely low latency. The update frequency of this location information can be set relatively high, for example, 5 times per second (5 Hz), to ensure that the system can capture the user's continuous, smooth movement trajectory, rather than discrete jump points. The data structure for the location information can be a tuple containing a user identifier, three-dimensional coordinates (x, y, z), and a timestamp, for example: (User_ID, x, y, z, timestamp).

[0027] After obtaining the user's real-time location information, the system needs to map it onto specific exhibits. To this end, during the system deployment phase, the digital map of the entire exhibition hall needs to be preprocessed. This involves defining a virtual, geometrically regular or irregular "exhibit interaction area" on the digital map for each exhibit or a group of closely related exhibits. These areas are typically non-overlapping spatially, collectively forming a complete division of the exhibition hall's visitor space. Each exhibit interaction area is associated with a unique exhibit identifier (Exhibit_ID).

[0028] Similarly, to ensure that the "exhibit interaction area" described in S100 accurately reflects the user's attention intent and avoids false triggers (users merely passing by) or missed triggers (users showing interest but not entering the area) due to unreasonable area delineation, its delineation process cannot follow simple geometric boundaries. In a specific implementation scenario, the delineation principles should at least include: First, the principle of full coverage of the best viewing area. The primary goal of the interaction area is to cover one or more of the best viewing positions and ranges for the exhibit. For a painting, this might be a fan-shaped area within 2 to 4 meters directly in front of it; for a sculpture that can be viewed from 360 degrees, it might be a circular or elliptical area centered on the sculpture's base. The boundaries of the area should be determined through actual pedestrian flow simulation or line-of-sight analysis to ensure that users have a good viewing experience in any typical position within the area. Second, the principle of safe buffer zone for pedestrian flow. A clear safe buffer zone must be maintained between the boundary of the interaction area and the main pedestrian passage within the exhibition hall. The width of this buffer zone should be set according to the width of the passage and the density of pedestrian flow, for example, it can be set to 0.5 meters to 1.0 meter. The purpose is to ensure that users who only walk along the main aisle and do not intend to linger do not accidentally "graze" the edges of the interactive area, thus avoiding a large number of invalid trigger events and ensuring the effectiveness of the dwell time calculation. Third, the principle of hierarchical delineation of multi-focal exhibits. For large exhibits with large volumes or scattered information points (e.g., a complete dinosaur fossil skeleton, a complex mechanical device, or a scene reconstruction), a hierarchical or nested approach to delineating interactive areas can be adopted. A large primary interactive area covering the entire exhibit can be set first to trigger an overall introduction to the exhibit. Then, within this primary area, smaller, independent secondary interactive areas can be delineated for key parts of the exhibit (e.g., the head and claws of a dinosaur, the engine of a mechanical device). When a user's trajectory shows that they have lingered in a primary area for a long time and further entered a secondary area, the system can push more targeted, in-depth information related to that specific part. Fourth, the principle of physical environment adaptation. The geometry of the interactive area should not be rigid but must be adaptively adjusted according to the actual physical environment of the exhibit. For example, if an exhibit is adjacent to a wall or large column, the boundary of the interactive area in that direction should be aligned with the physical obstacle. If the exhibit is placed in a glass display case, the boundary of the interactive area should be slightly larger than the outer contour of the display case to match the user's actual accessible location. This adaptability ensures a high degree of consistency between the virtual interactive area and the real physical space, improving the accuracy of location judgment. Therefore, the process of determining the exhibit that the user is currently interested in transforms into a spatial geometry judgment problem: the system continuously matches the user's current location coordinates (x, y, z) with all preset exhibit interactive areas to determine which area the coordinate point falls within.Once the region is determined, the corresponding Exhibit_ID is identified as the exhibit the user is currently interested in. This process ensures that the system can accurately identify the user's focus and serves as the logical starting point for all subsequent personalized calculations.

[0029] For example, suppose there are two exhibits in the exhibition hall: Exhibit A (Exhibit_ID_A) and Exhibit B (Exhibit_ID_B). The system administrator defines a circular interactive area with a radius of 2 meters for Exhibit A on the digital map in the background, with the center coordinates (10.5, 20.3, 1.5), in meters. A rectangular interactive area of ​​3 meters × 4 meters is defined for Exhibit B, with the four vertex coordinates (15.0, 30.0, 1.5), (18.0, 30.0, 1.5), (18.0, 34.0, 1.5), and (15.0, 34.0, 1.5) respectively. When the system's data acquisition module receives a location information from user User_007 as (User_007, 10.8, 20.1, 1.6, 1678886400), the positioning calculation engine will perform a judgment. First, calculate the distance between this point and the center of the circle in area A of the exhibit. The distance is less than the preset radius of 2 meters, so the system determines that user_007 is currently focused on exhibit A. The system starts a timer associated with user_007 and Exhibit_ID_A to record the duration of their attention. If the user later moves to coordinates (16.2, 31.5, 1.7), the system determines that this coordinate point is within the rectangular area of ​​exhibit B, thus switching the user's focus to exhibit B. This process is continuously and automatically executed, providing real-time context-aware input for subsequent steps.

[0030] S200: During the period when the user is paying attention to the target exhibit, a dynamic interest vector associated with the user is subjected to time decay processing; when the user is detected to have left the exhibit interaction area corresponding to the target exhibit, the duration of the user's attention to the target exhibit is obtained; if the duration of ...

[0031] This step constructs a mathematical model for quantifying and tracking users' dynamic interests. The core of this model is a dynamic interest vector maintained independently for each user, denoted as... This vector exists within a high-dimensional knowledge space, where each dimension corresponds to an atomized knowledge tag within the museum's content system. For example, the knowledge space of an art museum could consist of hundreds of knowledge tags such as "Impressionism," "Cubism," "Oil Painting Techniques," "Bronze Ware," and "Renaissance." Therefore, dynamic interest vectors... It is an N-dimensional real vector, where N is the total number of knowledge tags. The nth element in the vector... Each component The magnitude of the value represents the user's opinion on the number 1. The current interest intensity of each knowledge tag. When a user first enters the exhibition hall, their dynamic interest vector is initialized to a zero vector, indicating that the system knows nothing about their interests.

[0032] To ensure the completeness, orthogonality, and scalability of the N-dimensional knowledge space described in this embodiment, the construction of the core carrier knowledge label system itself can follow a systematic construction method to eliminate the arbitrariness of manual settings. In a preferred embodiment, this construction method can adopt a hybrid mode that combines data-driven and expert verification. Specifically, the first stage of this method is the generation of a label candidate set driven by data. In this stage, natural language processing (NLP) techniques are used to perform in-depth text mining on the official text description materials of all exhibits in the exhibition hall (including but not limited to exhibit nameplates, official catalogs, academic research papers, historical documents, etc.). For example, all text materials can be first aggregated into a large corpus. Then, the Term Frequency-Inverse Document Frequency (TF-IDF) algorithm is used to calculate the TF-IDF value for each word or phrase in the corpus. This value can quantify the importance of a word for a certain document (i.e., the description of a certain exhibit), and at the same time filter out stop words (such as "of", "is", etc.) that commonly appear in all documents but have no differentiating meaning. By setting a TF-IDF threshold (for example, selecting the top 10 words with the highest TF-IDF values in each exhibit description), a batch of high-value keywords can be initially screened out. To further explore the relevance at the theme level, a topic model can also be introduced, such as the Latent Dirichlet Allocation (LDA) model. Taking the entire corpus as the input of the LDA model, the model can unsupervised learn several "topics", and each topic is defined by a series of word distributions with high co-occurrence probabilities. For example, the model may automatically cluster a topic with high-probability words such as "brushstroke", "light and shadow", "color", "pigment", and those skilled in the art can manually label it as a high-level label of "oil painting technique". The second stage of this method is expert verification and ontology construction. The label candidate set generated in the first stage (whether it is the TF-IDF-based keywords or the LDA-based topic clusters) is submitted to a review committee composed of multi-domain experts (such as art historians, archaeologists, curators). The experts will review the candidate labels, eliminate redundancies, correct ambiguities, and supplement concepts that may have been missed in the data mining process based on their professional knowledge. More importantly, the experts will be responsible for constructing the hierarchical and associative relationships between these labels, that is, constructing a knowledge ontology (Ontology). For example, they will place "Van Gogh" and "Monet" under the parent label of "Impressionism", and juxtapose "Impressionism" and "Post-Impressionism" under "19th Century Art Schools". Through this stage, a flat list of labels is organized into a structured and logically hierarchical knowledge graph, which is the final knowledge label system. The knowledge label vector of each exhibit This is the result of associating the exhibit with the corresponding node in this map.

[0033] The update mechanism described in this step consists of two parallel and antagonistic sub-processes: a global "time decay" and an event-driven "interest gain".

[0034] The time decay process is continuous, a background computational task that runs uninterrupted throughout the user's entire visit. Specifically, the time decay process continues from the moment the system assigns an identifier to the user and begins location tracking until the user leaves the exhibition hall and the system terminates their visit session. Whether the user remains within an interactive exhibit area, moves between areas, or stays in non-exhibit areas such as rest areas, the system decays the user's entire dynamic interest vector V_user at a fixed time step Δt (e.g., 1 second). This decay aims to simulate the natural forgetting of interests, ensuring that early, potentially irrelevant interests do not permanently dominate the user's interest profile, thus allowing the model to better reflect the user's recent focus. In a specific implementation, time decay can be described by the following formula: in, It is the interest vector at the current moment. It is the interest vector of the previous time step. It is a global interest decay coefficient, a small normal number that controls the rate at which interest is forgotten. The value needs to be adjusted based on the average visit duration and content relevance density of the exhibition hall. A smaller A value indicating slow interest decay and a relatively stable user interest profile is suitable for exhibition halls with closely related content and coherent visitor paths; a larger A low value indicates that interest decays quickly, and the system focuses more on the user's immediate, short-term interests, making it suitable for exhibition halls with diverse themes and highly disjointed content. For example, The value can be set to 0.0005. This means that at a time step Within seconds, the intensity of interest will be multiplied by a decay factor of approximately 0.9995.

[0035] Interest gain is updated when specific conditions are triggered. This process is not continuous but is triggered by specific user behavior, namely, lingering in front of an exhibit for a valid duration. When the spatial judgment logic in S100 detects that a user has moved from one exhibit interaction area to another, or left all areas and entered a public passage, the system determines that the user has ended their attention to the previous exhibit. At this time, the system will obtain the total lingering time of the user in the previous exhibit interaction area. The system will preset a "duration threshold," such as 10 seconds. The purpose of this threshold is to filter out brief interaction events where users are merely passing by and have not generated any real interest. If If the duration exceeds the threshold, the system considers the user to have generated valid interest in the exhibit and initiates the interest gain update process.

[0036] The core basis for interest gain updates is the knowledge tag vector of the exhibit, denoted as... This vector is related to the user's dynamic interest vector. They reside within the same N-dimensional knowledge space. It is a pre-defined, typically sparse vector, assigned to each exhibit by the exhibition experts, to describe the knowledge attributes associated with that exhibit. In a specific embodiment, the knowledge tag vector, i.e. The specific form of this is a multi-hot vector. For ease of understanding, we first explain the "one-hot vector," its foundation. In a classification system with N categories, a one-hot vector is an N-dimensional vector where only one dimension has a value of 1, indicating that the sample belongs to only this one category, while the values ​​of the remaining N-1 dimensions are all 0. The multi-hot vector used in this embodiment is an extension of the one-hot vector in a multi-label scenario. Specifically, the multi-hot vector is also an N-dimensional vector, where N is the total number of knowledge labels in the knowledge label system. Unlike the one-hot vector, the multi-hot vector can have multiple dimensions with a value of 1. A value of 1 in the i-th dimension indicates that the exhibit possesses the knowledge attribute corresponding to the i-th knowledge label; a value of 0 indicates that it does not possess it. By using the multi-hot vector form, we can accurately represent the complex situation where an exhibit may simultaneously belong to multiple knowledge domains or possess multiple technical attributes in a structured and computable way. For example, in a system containing tags such as {Impressionism, Van Gogh, oil painting techniques}, an exhibit that is related to all three would have a multi-heat vector form of [1, 1, 1, 0, ...], where the first three dimensions have a value of 1.

[0037] The update process is accomplished using the following formula: Here, It is the interest vector after the user leaves the exhibit and has undergone the last time decay. It is a global behavior gain learning rate coefficient that controls the intensity of the impact of a single effective dwelling behavior on the overall interest profile. Its role is to transform the dimensionless behavior gain function... and dimensionless knowledge tag vector direction The resulting gain signal is transformed into a signal related to the dynamic interest vector. Adjustments with the same dimensions. Therefore, It inherently possesses a unit of interest intensity and can be considered a basic unit of interest gain. For example, its value can be set to 0.2 units of interest intensity.

[0038] It is a key nonlinear gain function that will determine the original dwell time. This is transformed into a weight that better reflects the intensity of interest. (Directly use...) Using it as a weight is undesirable because it leads to an infinite linear increase in interest intensity. A non-linear function, such as a logarithmic function, can reasonably simulate diminishing marginal returns: that is, the interest gain from increasing dwell time from 10 seconds to 20 seconds is far greater than the gain from increasing it from 100 seconds to 110 seconds. This function can be specifically defined as: in, It is a normalized time constant used to adjust the growth rate of the function, and its unit is __________. The same, both are seconds (s). For example, It can be set to be equal to the duration threshold, i.e., 10 seconds.

[0039] Finally, by summing the user's old interest vector with this weighted gain vector representing the newly learned knowledge, a precise quantitative update of the user's dynamic interest vector is achieved.

[0040] For example, continuing with the example in S100, a 5-dimensional knowledge space is introduced, with labels {Tag_1: Impressionism, Tag_2: Van Gogh, Tag_3: Oil painting techniques, Tag_4: Sculpture, Tag_5: Rodin}. The knowledge label vector for exhibit A (Van Gogh's *Starry Night*) is shown below. User_007's initial interest vector upon entering the exhibition hall. Assume the user stopped in front of exhibit A for a total of [number missing] hours. The duration has exceeded the 10-second threshold. System parameters are set to... , , Assuming a user spends 45 seconds entering and leaving exhibit A, their interest vector will continuously decay during this time, but since the initial value is 0, it will remain 0 after decay. When they leave, an interest gain update is triggered. First, the gain function value is calculated: Then calculate the interest gain vector: Last update of the user's dynamic interest vector: At this moment, the user's interest profile has been updated, clearly pointing to areas related to "Impressionism," "Van Gogh," and "oil painting techniques." This updated vector will serve as the basis for his subsequent visit decisions.

[0041] S300: When the user enters the interactive area of ​​a new target exhibit, the updated dynamic interest vector of the user is obtained, and a target content level to be displayed is determined based on the dynamic interest vector and the knowledge tag vector of the new target exhibit; according to the target content level, the content is displayed on a display terminal associated with the new target exhibit.

[0042] This step translates the user's interest state calculated in S200 into specific personalized exhibit content. This process is triggered in real time: when the positioning system in S100 detects that the user has entered a new exhibit interaction area that is different from the previous one, the logic of this step is activated.

[0043] After activation, the system retrieves the user's latest dynamic interest vector from the user status database. This vector, after continuous time decay and possibly one or more interest gain updates, is the best quantitative representation of the user's interest state up to the current moment.

[0044] Simultaneously, the system will retrieve the knowledge tag vector of the new target exhibit from the exhibit content database based on the Exhibit_ID of the current exhibit interaction area. .

[0045] The system quantifies the degree of matching between two vectors by calculating their vector similarity, and the result is a scalar value, namely the matching score. In a preferred embodiment, cosine similarity is used. The formula for calculating cosine similarity is as follows: in, It is the dot product of two vectors. and These are the L2 norms (i.e., the magnitude or length of the vectors) of the two vectors. The cosine similarity value ranges from -1 to 1 (in this application, since the vector components are non-negative, the actual range is [0, 1]). The closer the value is to 1, the closer the two vectors are in direction, meaning the user's interest direction aligns more closely with the exhibit's knowledge attribute direction; the closer the value is to 0, the less correlated (orthogonal) they are in direction. Cosine similarity only considers the direction of the vectors and ignores their magnitude. This means that even if the user's overall interest intensity (vector magnitude) is low, a high matching score can still be achieved as long as their interest distribution (direction) highly matches the exhibit.

[0046] In obtaining a matching score The system then uses this score to decide what content to display. Each exhibit can have multiple preset content levels with different depths and angles. For example, three levels can be set: L1 (Basic Layer): Provides the most basic and universal information about the exhibit, such as name, author, year, and size. Suitable for all users who are encountering the exhibit for the first time.

[0047] L2 (In-Depth): Introduces the background story, creative techniques, historical significance, or related anecdotes of the exhibits. Suitable for users with some interest in this field.

[0048] L3 (Expert Level): Offers professional academic interpretations, comparisons with other works, or the latest relevant research findings. Suitable for serious enthusiasts or professional researchers.

[0049] For example, the L1 basic layer corresponds to [0, 0.3), the L2 advanced layer corresponds to [0.3, 0.7), and the L3 expert layer corresponds to [0.7, 1.0]. The content level corresponding to the score interval is selected as the final decision.

[0050] if If so, then choose L1.

[0051] if If so, then choose L2.

[0052] if If so, then choose L3.

[0053] Finally, the system calculates... The selected area determines the target content level to be displayed. The content matching and generation module generates an instruction containing an Exhibit_ID and the target content level ID (e.g., L2), which is sent via the internal network to the display terminal module located next to the target exhibit. Upon receiving the instruction, this terminal module (which can be a high-definition display screen, a projector, or an AR glasses content source) immediately retrieves and plays the corresponding content level from its local or cloud content library. The latency throughout the entire process, from user entry into the area to content presentation, can be controlled as needed; for example, a low latency can provide users with a smooth and seamless personalized experience.

[0054] Continuing with the previous example, after user User_007 visits exhibit A (Van Gogh's "Starry Night"), their interest vector is updated to... Suppose he then walks to exhibit C, another Impressionist painting, whose knowledge tag vector is... (Related to "Impressionism" and "Oil Painting Techniques"). As he walks towards exhibit C, assuming 30 seconds have passed, his interest vector will slightly decrease due to time decay. Let the decay factor be... Then his interest vector when he arrives at exhibit C is The system then calculates the matching degree: Since the matching score of 0.817 is greater than the threshold of 0.7, the system determines that the user is a deep interest user. Therefore, it will instruct the display screen next to exhibit C to directly display the L3 (expert level) content, such as a detailed analysis of the similarities and differences between this painting and "Starry Night" in terms of light and shadow treatment.

[0055] However, if the user visits exhibit A and then moves on to exhibit B (Rodin's sculpture "The Thinker"), their knowledge tag vector will be... Therefore, when he arrives at exhibit B, the matching degree is calculated as follows: Since the matching degree is 0, the system determines that the user has no prior interest in this area of ​​exhibits, and therefore will push L1 (basic layer) content to provide the most basic introductory information. This process demonstrates the system's personalization and adaptive capabilities.

[0056] Example 2 This embodiment, based on Embodiment 1, provides an artificial intelligence-based interactive exhibition system for exhibition halls. This system is the physical carrier or functional entity for implementing the aforementioned methods. The system may include one or more processors, a memory, and program modules for implementing specific functions. In this embodiment, the system can be specifically divided into the following core functional modules: Data acquisition module 10 is used to continuously collect raw location data from all users within the exhibition hall in real time. At the hardware level, it consists of positioning base stations (e.g., UWB base stations) deployed throughout the exhibition hall, positioning beacons worn by users, and a data aggregation gateway for collecting and initially processing base station data. At the software level, it includes a positioning resolution server that runs an algorithm that converts raw signal data (such as TDoA values) into three-dimensional coordinates. This module continuously outputs location information with timestamps and user IDs to other modules in the system (primarily the user status management module) in the form of a data stream.

[0057] The user status management module 20 is responsible for maintaining and updating the dynamic interest vector of each user. Coupled with the data acquisition module 10, it may contain a user session manager to handle user entry and exit, and create and destroy corresponding user data instances. At its core is a vector database for real-time storage and rapid reading and writing of the dynamic interest vectors of all online users. It also includes an update calculation engine that executes operations according to the algorithm logic defined in step S200. This engine subscribes to the location information stream of the data acquisition module, determines the user's dwell state and duration based on the location information, periodically performs time decay calculations, and triggers interest gain calculations when the user leaves the exhibition area. All calculation results are written back to the vector database in real time. A figure (not shown) illustrates the core workflow of this module, namely an initial interest vector. After a period of global time decay, and with the specific lingering behavior (duration) and exhibit labels The generated local interest gain vector By combining these elements, an updated interest vector is finally generated. .

[0058] The content matching and generation module 30, connected to the user state management module 20, is used to obtain the latest user interest vectors when needed. This module is activated when a user enters a new exhibit interaction area (an event that can be notified by the user state management module). Internally, it includes an exhibit database interface for quickly retrieving the exhibit's knowledge tag vectors and content at each level based on the Exhibit_ID. A matching score calculator performs vector similarity calculations (such as cosine similarity) as defined in S300. A content decision engine makes the final content level selection based on the calculated matching score and preset threshold rules. After the decision is made, this module generates a standardized content display instruction and sends it to the corresponding display terminal module.

[0059] The display terminal module 40 is the front end that directly interacts with the user. Physically, it consists of interactive devices deployed next to each exhibit, such as touch screens, projectors, or smart speakers. Each terminal device has a unique ID that corresponds to the exhibit ID. Internally, it contains a network communication unit for listening for instructions from the content matching and generation module 30. Once it receives an instruction that matches its own ID, its internal content rendering engine loads the corresponding multimedia materials (text, images, videos, audio) from local storage or a cloud content server according to the content level ID specified in the instruction, and presents them on the display device.

[0060] In a typical application scenario, these four modules work together to form a closed-loop real-time feedback system. The user's unconscious behaviors (movement and stillness) are captured by the data acquisition module, interpreted and quantified into dynamic interest states by the user state management module, and then the content matching and generation module makes intelligent content recommendation decisions based on these states. Finally, personalized information is fed back to the user through the display terminal module. Subsequent user behaviors become new inputs for the system, continuously optimizing its understanding of user interests, thus forming a constantly self-adjusting and adaptive intelligent interactive ecosystem.

[0061] Example 3 Based on the technical solutions disclosed in Embodiments 1 and 2, this embodiment provides an interactive display method and system with predictive guidance capabilities. In this embodiment, the interactive display system further includes a narrative path generation module, and uses narrative relevance as a key dimension for recommendation decisions, thereby enhancing the intelligence level and educational guidance value of the interactive display, and addressing the problems of narrative fragmentation and excessive opportunity costs caused by the lack of macro-planning of visitor paths in existing technologies.

[0062] To achieve the above objectives, this embodiment, based on Embodiment 2, further includes a narrative path generation module 50. The following will describe in detail the parts of this embodiment that differ from other embodiments; parts identical to those in Embodiments 1 and 2 will not be repeated.

[0063] In this embodiment, in addition to assigning a knowledge tag vector to each exhibit, the system also pre-constructs and stores a narrative link matrix. This is an M×M asymmetric matrix, where M is the total number of exhibits in the exhibition hall. Each element in the matrix... Representatives from the exhibits To the exhibits The strength of the one-way narrative link, with a value range of [0, 1], is a dimensionless number. This strength value is set by the curatorial experts based on the inherent logical relationship between the exhibits, for example: : Indicates from the exhibits To the exhibits There is a very strong positive narrative connection. This connection can be the continuation of an artist's creative process (such as from a painting to a letter recording the creative process), the causal relationship of technological development (such as from an early prototype to a mature product), or the direct continuation of historical events.

[0064] : Indicates a moderate degree of connection, such as similarity in style or theme (e.g., the works of two artists of the same school) or indirect relationship between people (e.g., two famous people from different fields in the same historical period).

[0065] This indicates that the two are basically unrelated in terms of narrative logic.

[0066] Because narrative links are usually directional, Generally not equal to This matrix is ​​stored in a database managed by the newly added narrative path generation module 50. Optionally, it can be stored in a graph database to handle complex networks of relationships between exhibits more efficiently.

[0067] Based on this, step S300 in the original method flow will be replaced or deepened by the following step S400.

[0068] S400: Performs content generation and path recommendation based on hybrid guidance decisions.

[0069] This step, based on the original S300, introduces a narrative dimension, and its detailed sub-steps are as follows: S410: Calculate the interest matching score.

[0070] This step is the same as the original S330 execution process. When the user enters a new target exhibit (denoted as...) When the user interacts with the content in the relevant area, the content matching and generation module 30 first obtains the user's latest dynamic interest vector. And calculate its relationship with the knowledge tag vector of the target exhibit. The cosine similarity between them yields the interest matching score. .

[0071] S420: Query narrative link strength.

[0072] In obtaining Afterwards, the content matching and generation module 30 does not immediately make a decision. It retrieves the ID of the last exhibit from the user status management module 20 that generated a valid interaction with the user (i.e., the dwell time exceeded the threshold), and records it as... Subsequently, module 30 sends a query request to the newly added narrative path generation module 50, the request containing the source exhibit ID. and target exhibit ID After receiving the request, the narrative path generation module 50 accesses the narrative link matrix. It returns the corresponding link strength value, denoted as . .

[0073] S430: Final recommended score for fusion computing.

[0074] The content matching and generation module 30 uses a preset linear weighted fusion formula to combine interest matching scores with narrative link strength to generate a more comprehensive final recommendation score. The formula is as follows: in, and These are the weight coefficients for interest matching and narrative linking, respectively. They are all positive real numbers, and their sum is 1 (i.e., ...). The setting of these two weights reflects the system's recommendation strategy. For example, it can be set to... and This indicates that the system, when making decisions, considers the user's immediate interests more, but is also significantly guided by the narrative logic between exhibits. In some more complex implementations, these two weights can even be dynamically adjusted. For example, for users visiting for the first time or whose interest profile is relatively vague (their dynamic interest vector...) (With a relatively small module length), the system can dynamically improve... The value can be adjusted to quickly respond to and cultivate the user's interest; while for expert users whose interest profiles are already very clear, the value can be appropriately increased. The value is used to recommend narrative paths that may be beyond their current interests but are highly inspiring.

[0075] S440: Decisions and displays are based on the final recommendation score.

[0076] The content matching and generation module 30 will use the calculated Instead of the original This is compared with a preset content level threshold to determine the target content level to be displayed.

[0077] Furthermore, in this embodiment, the system presets a narrative link threshold, for example, 0.8. In S420, if the queried link... If the value exceeds this threshold, it means the user is currently exploring along a very strong narrative thread. In this case, the content matching and generation module 30, in addition to generating regular content display instructions, will also generate an additional guiding instruction. This instruction will drive the display terminal module 40 to display a piece of guiding information while showing the core content. This information aims to clearly reveal to the user the profound connection between the current exhibit and the previous exhibit, thereby strengthening their understanding of the knowledge framework.

[0078] For example, the scenario in Embodiment 1 is reproduced and extended: Assuming the system parameters are set to The narrative link threshold is 0.8.

[0079] The user's interest vector is Its previous effective interactive exhibit The exhibit is item A (Van Gogh's "Starry Night").

[0080] When a user approaches exhibit C (Monet's "Impression, Sunrise"): (Medium style association) Therefore, it does not trigger active guidance.

[0081] When a user approaches exhibit D (Van Gogh's letter): (Interest match is not as high as C) (Extremely strong connection between the characters' mental states) This triggers active guidance.

[0082] Decision outcome analysis: In this embodiment, although exhibit D has a lower direct interest match with the user than exhibit C, its final recommendation score is higher due to the strong narrative connection between it and the user's previous point of interest. The score of 0.7462 is higher than that of exhibit C (0.6902). Therefore, the system determines that exhibit D is the next exploration point that is more valuable to the user. When the user enters the interactive area of ​​exhibit D, the system will not only select a higher content level (such as L3 expert level) based on the score of 0.7462, but will also display additional guiding information on the screen, such as: "Following your appreciation of 'The Starry Night', this letter that Van Gogh wrote to his brother Theo during the same period will truly restore his exciting emotions during the creation of the work." In this way, the method of this embodiment is no longer simply a passive response to the user's interests, but a personalized and proactive approach that constructs and recommends different exploration paths with logic and depth for each user, thereby greatly enhancing the value and creativity of the visitor experience.

[0083] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0084] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the descriptions of apparatuses in the above embodiments are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0085] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0086] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0087] When the integrated units described above are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0088] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0089] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit described above can be implemented in hardware.

[0090] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An interactive exhibition method for exhibition halls based on artificial intelligence, characterized in that, The method includes: Obtain location information associated with a user, representing the user's real-time location within the exhibition hall; Based on the location information, determine the target exhibit that the user is currently interested in within the preset exhibit interaction area; During the period when the user is paying attention to the target exhibit, a dynamic interest vector associated with the user is subjected to time decay processing; When it is detected that the user leaves the interactive area of ​​the exhibit corresponding to the target exhibit, the duration of the user's attention on the target exhibit is obtained. If the duration of the attention is greater than a preset duration threshold, the interest gain is updated on the dynamic interest vector after time decay processing based on the preset knowledge tag vector of the target exhibit and the duration of the attention. When the user enters the interactive area of ​​a new target exhibit, the updated dynamic interest vector of the user is obtained, and a target content level to be displayed is determined based on the dynamic interest vector and the knowledge tag vector of the new target exhibit. Based on the target content hierarchy, the content is displayed on a display terminal associated with the new target exhibit.

2. The method according to claim 1, characterized in that, The time decay processing of a dynamic interest vector associated with the user includes: According to a preset time step, based on a preset global interest decay coefficient, each dimension component of the dynamic interest vector is exponentially decayed to reduce the influence of historical interests on the current state.

3. The method according to claim 1, characterized in that, The process of updating the interest gain of the dynamic interest vector after time decay processing based on the preset knowledge tag vector of the target exhibit and the dwell time includes: Based on the dwell time, a behavioral gain coefficient that is positively correlated with the dwell time is determined through a nonlinear gain function; The behavior gain coefficient is combined with the knowledge tag vector of the target exhibit to generate an interest gain vector; The interest gain vector is summed with the dynamic interest vector after time decay processing to complete the interest gain update.

4. The method according to claim 3, characterized in that, The nonlinear gain function is a logarithmic function, and the determination of the behavioral gain coefficient includes: The ratio of the dwell time to a normalized time constant is logarithmically calculated to obtain the behavior gain coefficient, thereby making the contribution of dwell time to interest gain exhibit a diminishing marginal effect.

5. The method according to claim 1, characterized in that, The process of determining a target content level to be displayed based on the dynamic interest vector and the knowledge tag vector of the new target exhibit includes: Calculate the vector similarity between the dynamic interest vector and the knowledge tag vector of the new target exhibit to generate a matching score; The matching score is compared with multiple preset content level thresholds, and the preset content level corresponding to the interval to which the matching score belongs is selected as the target content level.

6. The method according to claim 5, characterized in that, The vector similarity is cosine similarity.

7. The method according to claim 1, characterized in that, Determining a target content level to be displayed includes: Obtain the previous exhibit that the user was interested in before entering the new target exhibit, which serves as the previous interaction node; Based on the pre-defined narrative link relationship between the previous exhibit and the new target exhibit, a narrative link strength is determined; An interest matching score is determined based on the vector similarity between the dynamic interest vector and the knowledge tag vector of the new target exhibit. A final recommendation score is generated by combining the narrative link strength and the interest matching score with preset weights. Based on the final recommendation score, the target content level to be displayed is determined.

8. The method according to claim 7, characterized in that, The method further includes: When the narrative link strength is greater than a preset link threshold, additional guiding information is displayed on the display terminal to explain the narrative link relationship between the previous exhibit and the new target exhibit.

9. An interactive exhibition system for exhibition halls based on artificial intelligence, characterized in that, The system includes: A data acquisition module is used to acquire location information associated with a user, representing the user's real-time location within the exhibition hall; A user status management module, connected to the data acquisition module, is used to determine the target exhibit currently being viewed by the user in a preset exhibit interaction area based on the location information. During the period when the user is viewing the target exhibit, a dynamic interest vector associated with the user is subjected to time decay processing. When the user is detected to have left the exhibit interaction area corresponding to the target exhibit, the module obtains the duration of the user's viewing of the target exhibit. If the duration of viewing exceeds a preset duration threshold, the module updates the interest gain of the dynamic interest vector after time decay processing based on the preset knowledge tag vector of the target exhibit and the duration of viewing. A content matching and generation module, connected to the user state management module, is used to obtain the updated dynamic interest vector of the user when the user enters the interactive area of ​​the exhibit corresponding to a new target exhibit, and to determine a target content level to be displayed based on the dynamic interest vector and the knowledge tag vector of the new target exhibit. A display terminal module, connected to the content matching and generation module, is used to display content on a display terminal associated with the new target exhibit according to the target content level.

10. The system according to claim 9, characterized in that, The system also includes: A narrative path generation module is used to store the preset narrative link relationships between multiple exhibits, and when the content matching and generation module requests it, it provides the narrative link strength between the previous exhibit that the user was interested in before entering the new target exhibit and the new target exhibit. Furthermore, the content matching and generation module is further used for: The narrative link strength is obtained and combined with the interest matching score determined based on vector similarity with a preset weight to generate a final recommendation score, and the target content level to be displayed is determined based on the final recommendation score.

Images