An exhibition hall scene self-adaptive intelligent switching control system and method

Abstract

The application belongs to the technical field of exhibition hall control, and specifically discloses an exhibition hall scene self-adaptive intelligent switching control system and method, which comprises a visitor information acquisition module, an individual vector establishing module, a group matching analysis module, a service mode determining module and a control instruction generating and executing terminal. The system obtains identity background, declaration preference and real-time behavior data, dynamically generates an individual interest vector through the individual vector establishing module, identifies a real-time intention state, calculates an individual-group matching degree through the individual deviation analysis module, and finally realizes intelligent matching of a group tour, light individualization and deep individualization three-level service modes through the service mode determining module. The application realizes a technical leap from static prediction to real-time perception, establishes a quantitative matching mechanism of individual and group interests, solves the problem of service mode hierarchical decision-making, and effectively improves the intelligent level of exhibition hall service and visitor satisfaction.

Description

Technical Field

[0001] This invention belongs to the field of exhibition hall control technology, specifically, it relates to an adaptive intelligent switching control system and method for exhibition hall scenes. Background Technology

[0002] As an important venue for cultural display and knowledge dissemination, exhibition halls, with the development of digital technology, rely on intelligent scene control not only to optimize the display effect, but also to directly affect the visitor's immersion and satisfaction, thus highlighting the importance of their control and management.

[0003] Existing technologies, such as the exhibition hall scene adaptive switching control method, device, equipment and storage medium disclosed in Chinese invention patent application No. 202411764637.8, generate waiting queues and personal interest rankings for each exhibition area by comprehensively considering visitor check-in, appointment time and personal association information. This combines interest matching with queue management, enabling like-minded people to queue together and maintaining a balance between personalization and efficiency.

[0004] Current technologies for adaptive switching of exhibition hall scenes mainly follow a scheduling-centric paradigm, treating visitors as schedulable units to achieve optimal macro-efficiency. However, this paradigm has the following fundamental flaws in pursuing a balance between group efficiency and overall experience: 1. It relies on prior information such as visitor reservations and check-ins before entering the park, lacking dynamic behavioral data generated during the visit. It cannot dynamically adjust the guided tour content or scene presentation based on real-time changes in visitors' interests and emotional feedback, resulting in a rigid guided tour experience.

[0005] 2. The reliability and timeliness of interest rankings generated solely based on prior information are difficult to guarantee. The lack of cross-validation through the comprehensive use of historical visit records and real-time interactive data leads to inaccuracies in interest assessment and matching.

[0006] 3. The focus on maximizing the common interests of the existing target group completely ignores the intensity of individual visitors' specific preferences and their urgent need for personalized services. There is no situational management mechanism to distinguish which visitors are suitable for group tours and which require more personalized and targeted services. Summary of the Invention

[0007] In view of this, in order to solve the above problems, an adaptive intelligent switching control system and method for exhibition hall scenes are proposed.

[0008] The objective of this invention can be achieved through the following technical solution: This invention provides an adaptive intelligent switching control system for exhibition hall scenes. The system includes: a visitor information collection module, which collects visitor identity background, declared preference data, real-time behavior data and location information, and generates a visitor information dataset.

[0009] The individual vector generation module, based on the dataset and combined with the temporal analysis of behavioral data, outputs dynamically evolving individual interest vectors and identifies the real-time intent state of visitors.

[0010] The group matching analysis module dynamically clusters visitors based on their real-time location and individual interest vectors to generate group preference vectors. It then calculates the similarity between each visitor's individual interest vector and the corresponding group preference vector using cosine similarity to generate the individual-group matching degree.

[0011] The service mode determination module compares the individual-group matching degree with a preset threshold and matches a preset service mode based on the comparison result.

[0012] The control command generation and execution terminal generates corresponding scene control commands based on the matched service mode and distributes them to the corresponding exhibition area equipment controllers to execute scene switching control.

[0013] The present invention also provides a method for an adaptive intelligent switching control system for exhibition hall scenes, the method comprising: collecting visitor identity background, declared preference data, real-time behavior data and location information, and generating a visitor information dataset.

[0014] Based on the dataset, dynamic evolutionary individual interest vectors are output by combining temporal analysis of behavioral data, and the real-time intent state of visitors is identified.

[0015] Visitors are dynamically clustered to generate group preference vectors. The similarity between each visitor's individual interest vector and the corresponding group preference vector is calculated using cosine similarity to generate individual-group matching degree.

[0016] The individual-group matching degree is compared with a preset threshold, and a preset service mode is matched according to the comparison result.

[0017] Based on the matched service mode, corresponding scene control instructions are generated and distributed to the corresponding exhibition area equipment controllers to execute scene switching control.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention continuously collects real-time behavior data of visitors through a sensor network deployed in the exhibition area, and combines dynamic clustering analysis to realize the instant adjustment of the guide content and scene presentation, effectively overcoming the response lag problem caused by the reliance on static information in traditional systems, and enabling the exhibition hall scene to be updated adaptively with changes in visitor status.

[0019] (2) When conducting individual interest analysis, this invention combines historical visit records with real-time interactive information for cross-validation, which greatly improves the accuracy and completeness of interest representation, thereby truly reflecting the visitor's current focus and interest changes, significantly improving the accuracy and timeliness of interest assessment, and solving the problem of insufficient credibility of a single data source.

[0020] (3) Through primary and secondary mapping classification and differentiated weight allocation, this invention can accurately distinguish core interests and secondary concerns in interest vectors, so that interest evaluation can effectively identify and separate stable preferences and temporary interests, ensuring the accuracy and pertinence of interest classification, and also laying a refined data foundation for subsequent personalized services.

[0021] (4) In the analysis of group preference vectors, the present invention uses the synergistic effect of dynamic clustering algorithm and multi-dimensional weight adjustment to make the generated group preference vectors accurately capture group consensus and effectively retain individual differences. Combined with the real-time updated individual interests and intentions, the present invention fully considers the intensity of individual visitors’ special preferences and their urgent need for personalized services, ensuring that the group preference vectors always reflect the real group dynamics.

[0022] (5) This invention achieves a key breakthrough in transforming from extensive group management to refined diversion by establishing a quantitative evaluation of individual-group matching. At the same time, it intelligently divides visitors into three service modes: group tour, light personalization, and deep personalization, automatically identifying visitors suitable for group tour and individuals who need special services. This ensures the efficiency of basic services while meeting the personalized needs of different levels. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of the system module connections of the present invention.

[0025] Figure 2 This is a schematic diagram of the overall implementation process of the method of the present invention.

[0026] Figure 3 This is a schematic diagram of the individual interest vector output process of the present invention.

[0027] Figure 4 This is a schematic diagram of the preference category mapping rules of the present invention.

[0028] Figure 5This is a schematic diagram of the group preference vector generation process of the present invention. Detailed Implementation

[0029] 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, and 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.

[0030] Please see Figure 1 As shown, the present invention provides an adaptive intelligent switching control system for exhibition hall scenes. The system includes: a visitor information collection module, an individual vector establishment module, a group matching analysis module, a service mode determination module, and a control command generation and execution terminal.

[0031] In the above, the individual vector establishment module is connected to the visitor information collection module and the group matching analysis module, respectively, and the service mode determination module is connected to the group matching analysis module and the control command generation and execution terminal, respectively.

[0032] The visitor information collection module collects visitor identity background, declared preference data, real-time behavior data, and location information to generate a visitor information dataset.

[0033] In practice, the visitor registration system acquires background information and declared preferences. Simultaneously, a multi-source sensor network deployed throughout the exhibition area, including location beacons, visual sensors, and interactive devices, collects real-time visitor movement trajectories, dwell time, interactive actions, and location information. This collected information is then spatiotemporally aligned to form a complete visitor information dataset, providing multi-dimensional data support for subsequent analysis.

[0034] All data collection processes adhere to privacy protection principles, employing data anonymization and encrypted transmission technologies to ensure information security. Furthermore, the spatiotemporal alignment of data is a current technology, and its specific implementation process will not be described in detail.

[0035] The individual vector building module, based on the dataset and combined with the time-series analysis of behavioral data, outputs dynamically evolving individual interest vectors and identifies the visitor's real-time intent state.

[0036] Considering that the declared preference data entered by visitors is equivalent to visitors' initial expectations of the exhibition hall, the intensity of visitors' initial interest can be obtained by analyzing the declared preference data.

[0037] Furthermore, considering that visitors' interests change with the content of the exhibition during the actual exhibition process, and that these changes are directly reflected in visitors' actual behavior during the exhibition, it is possible to obtain the intensity of visitors' behavioral interests in the current update cycle by analyzing behavioral data.

[0038] Based on this, the updated interest intensity value is obtained by fusing the behavioral interest intensity with the initial interest intensity value of the corresponding preference category, and an individual interest vector is constructed based on the updated interest intensity value.

[0039] Based on the above and referring to [see also] Figure 3 As shown in the specific embodiment, the individual interest vector output process includes three parts: processing of declared preference data and calculation of initial interest intensity, collection of behavioral data and calculation of behavioral interest intensity, and fusion of interest intensity and generation of individual interest vector, which are represented by A1-A3.

[0040] Furthermore, the specific implementation of the A1 part of the declared preference data processing and initial interest intensity calculation includes: A1-1, extracting the user-inputted free text or selected preference tags from the declared preference data.

[0041] A1-2. Calculate the semantic similarity between the free text or preference tag and the semantic extension set of each preset preference category, and map the preference category according to the calculation result.

[0042] The specific implementation process for semantic similarity calculation involves converting free text or preference tags into semantic vectors using text processing techniques, and simultaneously converting the semantic extension sets of each preset preference category, including synonyms, near-synonyms, and related concept descriptions, into vector representations. A similarity score is obtained by calculating the cosine similarity between two vectors. The text processing techniques used are existing technologies and will not be elaborated upon here.

[0043] The specific implementation process for mapping preference categories based on the calculation results can be found in [reference needed]. Figure 4 As shown, this specifically includes: determining whether there is a preset preference category whose calculated similarity score is higher than a preset matching threshold.

[0044] If it does not exist, mark the stated preference as an unidentified item and store it in a temporary interest pool.

[0045] If it exists, determine whether the number of preference categories whose similarity scores exceed the matching threshold is 1.

[0046] If the value is 1, the declared preference is mapped to that category; otherwise, the preset preference category with the highest similarity score is used as the primary mapping category, and the rest are used as secondary mapping categories, and corresponding labels are applied.

[0047] As a preferred embodiment, the matching threshold is set by analyzing the semantic similarity distribution between successfully mapped declared preferences and preset categories in historical data, and selecting a specific quantile, such as the 25th percentile.

[0048] A1-3. Assign an initial interest intensity value to each selected preference label according to the preset allocation rules.

[0049] In specific implementation, the preset allocation rules include: A1-31, extracting the form type to which the preference tag belongs, wherein the form type is one or more of single-choice form, multiple-choice form, and rating questionnaire.

[0050] A1-32. If the preference label comes from a single-choice form, the selected unique preference label will receive the preset highest initial interest intensity value. .

[0051] Considering that subsequent data comparison and analysis need to be on the same data dimension, the initial interest value range can be set to 0 to 1.

[0052] Considering that setting the minimum initial interest intensity value to 0 would prematurely filter out certain preference categories, in order to prevent premature filtering, the minimum initial interest intensity value is set to 0.1, and the maximum value is selected as the upper limit of the initial interest value range to participate in the subsequent calculation of the final initial interest intensity value.

[0053] A1-33. If the preference label comes from a multi-select form, assign the same initial interest intensity value to each selected preference label based on the preset highest initial interest intensity value.

[0054] In practice, when a preference tag is detected to originate from a multi-select form, the preset maximum initial interest intensity value is proportionally allocated to the total number of preference tags actually selected by the user, i.e., the ratio of the maximum initial interest intensity value to the total number of preference tags is taken.

[0055] A1-34. If the preference tags are derived from the rating questionnaire, perform the following steps: obtain the rating labels selected by the visitor and the highest rating label, and calculate the number of interval ratings between them.

[0056] The relative positional proportion is obtained by calculating the complement of the ratio of the number of interval levels to the total number of levels.

[0057] The initial interest intensity value is calculated using a preset nonlinear mapping function. , , The preset minimum initial interest intensity value, The sensitivity coefficient is configured.

[0058] Among them, the sensitivity coefficient The value of is used to adjust the shape of the mapping curve, specifically: when ... When the value is greater than 1, the mapping curve is a concave function, which imparts a stronger excitation effect to high relative position ratios. When = 1, the mapping is a linear relationship. When the position is between 0 and 1, the mapping curve is a convex function and is more sensitive to low relative position ratios.

[0059] In one specific embodiment, the sensitivity coefficient can be configured in a targeted manner according to the current operational strategy. That is, the exhibition hall administrator can dynamically configure the sensitivity coefficient through the management interface according to the current exhibition hall operation strategy: when it is necessary to focus on promoting a specific themed exhibition area, such as a temporary special exhibition, the sensitivity coefficient can be... Setting it to a value greater than 1 significantly enhances the interest intensity of highly rated visitors, thus prioritizing their visit to the target exhibition area. When a more balanced distribution of visitor traffic is needed, set it to... A score less than 1 compresses the intensity differences between different ratings, resulting in a smoother distribution of visitor interests. This differentiated configuration can flexibly adapt to operational goals at different stages, optimizing resource allocation while implementing a differentiated scoring strategy.

[0060] Taking a 5-level questionnaire as an example, if the visitor selects level 4, the preset highest level is level 5, and the relative position ratio is 0.8, when the sensitivity coefficient is set to 2, the intensity value is calculated as follows: .

[0061] To verify its fine-grained differentiation effect, we compared it with a linear strategy ( =1) Compare the results. When one visitor selects level 5 and another selects level 4: Under the linear strategy, the initial interest intensity values ​​calculated for the two are 1 and 0.8 respectively, with a difference of 0.2.

[0062] In nonlinear strategies ( Under condition 2), the initial interest intensity values ​​calculated by the two methods are 1 and 0.64, respectively, with the difference widening to 0.36.

[0063] Based on the above comparison, it can be seen that by introducing The nonlinear mapping greater than 1 effectively widens the intensity gap between ordinary positive reviews and highly enthusiastic reviews. This designed amplification effect ensures clearer and more decisive data support when allocating scarce resources such as personalized tours and in-depth interactions, thereby enabling refined differentiation and precise resource matching for visitors with different levels of enthusiasm.

[0064] It should also be noted that the nonlinear mapping function formula considers the minimum initial interest intensity value, ensuring that even if a visitor gives the lowest rating, their preference intensity will not be zero, thus retaining a record of that preference category in the visitor registration system. This design prevents the preference category from being completely ignored due to a single low rating, preserving the possibility and data channel for subsequent updates and corrections to the initial judgment through real-time behavioral data, thereby improving the robustness of the system.

[0065] A14. Query the visitor's historical visit records. If there is an intersection between the visitor's historical valid preference set and the current declared preference, multiply the initial interest intensity value of the preference category in the intersection by a preset enhancement coefficient and output the final initial interest intensity value.

[0066] In practice, the preset reinforcement coefficient is typically selected based on the historical frequency of occurrence of historical valid preference categories and currently declared preference data. For example, by analyzing the historical records of all visitors, the probability is calculated that if a preference category tag appears with a specific frequency in the first N visits, it will still be a valid preference tag in the (N+1)th visit. When the frequency of a preference category tag is less than 30%, historical frequency data shows that the probability of it continuing in subsequent visits is approximately 55%, only about 10% higher than the 50% random probability. In this case, a reinforcement coefficient of 1.1 is selected. When the frequency of a preference tag is between 30% and 70%, historical frequency data shows that the probability of the preference category tag continuing significantly increases to approximately 75%. Compared to the 50% random probability, its stability is improved by approximately 30%. In this case, a reinforcement coefficient of 1.3 is selected. When the frequency of a preference tag is greater than 70%, historical frequency data shows that the probability of the preference category tag continuing exceeds 90%. Compared to the 50% random probability, its stability is improved by approximately 50%. In this case, a reinforcement coefficient of 1.5 is selected. Furthermore, the final reinforcement coefficient value can be changed as the data in the historical effective preference set changes.

[0067] Furthermore, the specific implementation of behavioral data collection and behavioral interest intensity in section A2 includes: A2-1, periodically collecting and quantifying visitors' temporal behavioral data through a sensor network deployed in the exhibition area, and statistically analyzing the normalized dwell time and normalized interaction frequency of related interactive devices under each preference category's associated exhibits.

[0068] In one specific embodiment, the normalized dwell time is the ratio of the actual dwell time to the average dwell time of the exhibit, and the normalized interaction frequency is the ratio of the actual interaction frequency to the average interaction frequency of the interactive device.

[0069] A2-2. For each preference category, calculate the behavioral interest intensity of the current update cycle by linearly weighted summing and normalizing the dwell time and interaction frequency.

[0070] In practice, weights are set by analyzing the correlation between visitor behavior data and final interest intensity under different exhibit types in historical data. The correlation can be statistically analyzed using the Pearson coefficient. For example, for educational exhibits, data analysis shows that the correlation coefficient between dwell time and interest intensity is 0.82, while the correlation coefficient for interaction frequency is only 0.31; therefore, a weight of (0.8, 0.2) is configured. For interactive exhibits, the correlation coefficient between interaction frequency and interest intensity is 0.78, and the correlation coefficient for dwell time is 0.35; therefore, a weight of (0.2, 0.8) is used. This ensures that the optimal weight combination can be automatically selected based on the exhibit characteristics, making the calculation of behavioral interest intensity more consistent with actual exhibition viewing patterns. The Pearson coefficient calculation formula is an existing formula and will not be displayed here.

[0071] As a preferred embodiment, the calculation of behavioral interest intensity further includes weight combination matching based on preference category, specifically including: if the currently calculated preference category is the main mapping, then the first weight combination is adopted, wherein the weight of normalized dwell time is greater than the weight of normalized interaction frequency.

[0072] If it is a secondary mapping, the second weight combination is used, where the weight of normalized dwell time is less than the weight of normalized interaction frequency.

[0073] The primary mapping category represents visitors' core interests, the intensity of which is mainly reflected in immersive viewing behaviors such as prolonged dwell time; therefore, dwell time is given higher weight. Secondary mapping categories reflect secondary interests, typically manifested as tentative interactions; therefore, interaction frequency is given higher weight to capture their exploratory characteristics. This design, based on the correlation between interest levels and behavioral patterns, ensures the accuracy of assessing different types of interests.

[0074] In one specific embodiment, a first weight combination of (0.7, 0.3) is used for the primary mapping category, significantly increasing the contribution ratio of normalized dwell time. A second weight combination of (0.3, 0.7) is used for the secondary mapping category, focusing on strengthening the influence weight of normalized interaction frequency. Furthermore, the specific values ​​can be adjusted according to the exhibition hall's display needs and are not fixed.

[0075] This embodiment of the invention, through such differentiated configuration, can ensure the accurate distinction between deeply immersive behaviors and exploratory interactive behaviors. For example, lingering in the classical art exhibition area of ​​the main mapping versus briefly experiencing the interactive technology exhibits of the secondary mapping. By quantifying parameters, the precise conversion from viewing behavior to interest intensity is achieved, effectively improving the accuracy of interest updates.

[0076] To facilitate understanding, let's take classical art and digital technology as examples. Assume visitor A's primary mapping preference category is classical art, and their secondary mapping category is digital technology. In the classical art exhibition area, they linger and appreciate the art. By using a weight of (0.7, 0.3), the high normalized dwell time is amplified, increasing the calculated intensity of their behavioral interest and correctly reinforcing their primary interest. In front of the digital technology interactive screen, they click frequently but have a short dwell time. By using a weight of (0.3, 0.7), the impact of the short normalized dwell time is reduced, while the normalized interaction frequency is increased, reasonably maintaining their secondary interest without neglecting or overestimating it.

[0077] As demonstrated by the examples, a differentiated weighting strategy can accurately analyze the behavioral representation patterns of different interest categories, ensuring in-depth mining of core interests while avoiding the overemphasis on secondary interests. This makes the calculation of behavioral interest intensity more consistent with cognitive logic, thereby providing more discriminative data input for dynamic interest vectors and ultimately improving the accuracy and rationality of personalized scene recommendations.

[0078] Furthermore, the A3 part of interest intensity fusion and individual interest vector generation specifically includes: A3-1 performing linear weighted fusion of the behavioral interest intensity with the initial interest intensity value of the corresponding preference category to generate an updated interest intensity value.

[0079] In one specific embodiment, the fusion formula for linearly weighting the behavioral interest intensity and the initial interest intensity value of the corresponding preference category is as follows: In the formula, Corresponding to the updated interest intensity value, and These correspond to the initial interest intensity values ​​for behavioral interest and the corresponding preference category, respectively. The weights corresponding to the intensity of behavioral interest. The weights corresponding to the initial interest intensity values.

[0080] Considering the positive linear correlation between visitor interest intensity and the novelty of exhibition hall events—meaning that as visitors repeatedly visit the same type of exhibition hall, their curiosity about the exhibits gradually decreases, and the more times they participate in the same interactive event, the lower the probability of their subsequent active participation—a higher weighting is assigned to first-time visitors. The possible value is 0.8. For repeat customers with M visit records, then... The calculation formula is used to calculate the corresponding value. m is the reference number of visits, where the reference visit coefficient is obtained by statistically analyzing the historical visitor data of the exhibition hall. For example, it is the average number of visits when the intensity of visitors' interest in similar exhibits first decreases by more than 20%. This dynamic weight setting ensures that interest profiles can be quickly established for new visitors, while maintaining the intensity of interest for mature visitors.

[0081] A3-2 generates dynamically evolving individual interest vectors based on all preference categories and their corresponding updated interest intensity values.

[0082] It's important to note that by continuously integrating all preference categories and their dynamically updated interest intensity values, an individual interest vector with temporal evolution characteristics is constructed. This vector accurately reflects the complete evolution trajectory of a visitor's preferences from initial to real-time. It can simultaneously capture short-term behavioral feedback and long-term preference characteristics, reflecting both immediate interest fluctuations and preserving stable preference tendencies. The dynamically generated interest vector not only provides accurate basis for scenario matching but also adapts to the natural changes in visitor interests through a continuous evolution mechanism, ultimately achieving precise synchronization between personalized services and the visitor's real-time status.

[0083] More specifically, identifying the visitor's real-time intent state includes: B1, acquiring the visitor's real-time behavior data stream.

[0084] Specifically, the real-time behavior data stream includes, but is not limited to, real-time location coordinate sequence, head orientation angle data, visual focus coordinate data, and interaction data with the interactive device. The interaction data includes, but is not limited to, the interaction frequency and the timestamps of each operation corresponding to each interaction.

[0085] In practice, real-time location coordinate sequences are collected through UWB positioning beacons, and head orientation angle data and visual focus coordinate data are captured using a camera array. Combined with interactive device logs, a complete behavioral data stream is generated.

[0086] B2. Based on the behavioral data stream, the total continuous dwell time, average movement speed, the proportion of gaze on the exhibits, the operation interval of each interactive device, and the tortuosity of the movement path are statistically analyzed before the exhibits are set up.

[0087] Specifically, the continuous dwell time is obtained by using positioning sensors to measure the continuous time a visitor stays within the polygonal area of ​​the exhibit area. When a visitor leaves the area for more than a set interval threshold, such as 5 seconds, it is considered the end of a single dwell time. The total continuous dwell time is obtained by accumulating all single dwell times.

[0088] The average moving speed per unit time is calculated based on the real-time position coordinate sequence. The instantaneous speed within each sampling period is calculated, and after removing abnormal fluctuations, the average speed within a sliding time window, such as 30 seconds, is taken.

[0089] The proportion of time the gaze lingers on the exhibit is calculated by combining visual focus coordinate data with head orientation angle data, intersecting the gaze vector with the exhibit's spatial area, and statistically analyzing the percentage of time the gaze falls on the core area of ​​the exhibit during the observation period.

[0090] Operation interval time statistics are calculated by recording the timestamps of consecutive operations performed by visitors on the same interactive device, calculating the time difference sequence of adjacent operations, and taking the median of the sequence as the operation interval time.

[0091] The path tortuosity is calculated by generating a moving trajectory based on a real-time location coordinate sequence, calculating the line segment lengths between adjacent trajectory points and summing them to obtain the total trajectory length, and simultaneously calculating the Euclidean distance between the trajectory start and end points to obtain the straight-line distance. Finally, the ratio of the total trajectory length to the straight-line distance is used as the path tortuosity.

[0092] B3. Compare the statistical feature indicators with the preset intent state threshold and output the real-time intent state, wherein the intent state includes at least one of in-depth visit, fast passage, exploration and search, and interactive experience.

[0093] In practice, the thresholds are set based on statistical analysis of typical exhibition behaviors. After a large amount of data statistical analysis, the thresholds for the in-depth visit state are set as follows: the dwell time is greater than or equal to 120 seconds and the proportion of eye contact is greater than or equal to 70%; the fast passage state is set as the movement speed is greater than or equal to 0.8 meters / second and the single-point dwell time is less than or equal to 30 seconds; the exploration and search state is set as the path tortuosity is greater than or equal to 1.2 times / meter and the proportion of eye contact is less than or equal to 40%; and the interactive experience state is set as the operation interval is less than or equal to 15 seconds and the dwell time is in the range of 30 to 180 seconds.

[0094] It should be noted that the threshold can be dynamically adjusted within ±20% according to the characteristics of different exhibition areas. For example, the dwell threshold for in-depth visits can be increased to 150 seconds in key exhibition areas, while the movement speed standard for fast passage can be appropriately relaxed in rest areas.

[0095] B4. Associate the status identifier with the timestamp and store it.

[0096] It should be noted that this design is based on cognitive behavioral science theory and is used to capture the dynamic changes in visitors' exhibition needs through the fusion of multi-dimensional behavioral features. This transforms discrete behavioral data, such as movement trajectories and gaze focus, into continuous intent states, providing a basis for scene adaptive control and solving the service lag problem caused by misjudgment of intent.

[0097] By combining real-time intent status identifiers with timestamp-based storage, scenario-based service strategies can be adjusted promptly, improving resource allocation efficiency. This, in turn, provides data support for personalized service optimization, ultimately achieving a service model upgrade from passive response to proactive adaptation.

[0098] The group matching analysis module dynamically clusters visitors based on their real-time location and individual interest vectors to generate group preference vectors. It then calculates the similarity between each visitor's individual interest vector and the corresponding group preference vector using cosine similarity to generate the individual-group matching degree.

[0099] Specifically, please refer to Figure 5 As shown. The specific generation of the group preference vector includes: D1, dynamically grouping visitors based on location coordinates using the DBSCAN clustering algorithm to form temporary tour guide groups.

[0100] It should be noted that the DBSCAN clustering algorithm is an existing algorithm, and its specific grouping process will not be described in detail here.

[0101] D2. Extract the individual interest vectors of all members within each group, and calculate the average and variance of the interest intensity of the same preference category within the group.

[0102] D3. Set the initial values ​​of member weights based on the average value and variance of the interest intensity, and determine whether the member triggers the following conditions: Condition 1: The interest intensity value of this category in the member is lower than the group average and the variance is higher than the discrete threshold.

[0103] Condition 2: The member's dwell time at the current location exceeds the group's average dwell time.

[0104] Condition 3: The member is identified as being in deep visit status.

[0105] D4. If condition 1 is triggered, reduce the initial value of the member weight to the first weight value.

[0106] D5. If condition 2 is triggered, the initial value of the member weight will be increased to the second weight value.

[0107] D6. If condition 3 is triggered, the initial value of the member weight is increased to the third weight value, which is greater than the second weight value.

[0108] In practice, the dispersion threshold is set based on the 75th percentile of the dispersion of group interests in historical data. When the variance exceeds this value, it indicates that there are significant differences in interest in this preference within the group. At this time, the initial weight of low-interest members is halved, and 0.5 is used as the first weight value, which can effectively suppress the noise interference of individual members who follow the crowd and ensure the representativeness of the group preference vector.

[0109] Divide the group into a stay duration exceeding the mean group and a normal group. Convert the intensity of behavioral interest into a behavioral interest vector. Calculate the cosine similarity between the individual interest vector and the behavioral interest vector for both the stay duration exceeding the mean group and the normal group. The relative increase of the average cosine similarity of the stay duration exceeding the mean group compared to the average cosine similarity of the normal group is calculated to be 20%, which means the second weight value is set to 1.2.

[0110] Behavioral data from both the above-average and normal groups were collected simultaneously. This data was then input into a binary logistic regression model for validation, using the final effective interest vector derived from historical data as the validation benchmark. Logistic regression analysis showed that members marked as deeply engaged achieved a 90% accuracy rate in predicting true interest intensity based on their eye movement and behavioral characteristics, significantly higher than the 60% benchmark for normal members, representing a relative improvement of 50%. Therefore, a third weight of 1.5 was set. Since the binary logistic regression model is an existing model, its implementation will not be described further.

[0111] D7. If the condition is not triggered, keep the initial value of the member weight unchanged.

[0112] D8. Perform a linear weighted fusion calculation on the final member weight value and its interest intensity value, and divide by the total number of group members to obtain the final group interest intensity value.

[0113] D9. Generate a group preference vector by combining the final group interest intensity values ​​of all preference categories.

[0114] It should be noted that the dynamic weight adjustment mechanism primarily addresses the bias issue in traditional group preference modeling, where a minority of users dominate or in-depth visitors are overlooked. By setting three judgment conditions, it effectively identifies and suppresses noise interference from members with ambiguous interests, while strengthening the influence of members with clear behavioral orientations, and assigns the highest weight to high-value exhibition needs that have been validated through multimodal analysis. This hierarchical processing ensures that the group preference vector represents both mainstream tendencies and incorporates high-quality individual characteristics. This further guarantees that scene control commands based on this vector can meet the needs of most visitors while effectively responding to in-depth interest characteristics, providing accurate and reliable group behavior data for intelligent switching of exhibition scenes, and significantly improving the personalization of group services and the efficiency of resource allocation.

[0115] The service mode determination module compares the individual-group matching degree with a preset threshold and matches a preset service mode based on the comparison result.

[0116] Specifically, the matching process of the preset service mode includes: when the individual-group matching degree is greater than or equal to the first threshold, matching the group tour mode.

[0117] When the individual-group matching degree is between the first threshold and the second threshold, a slightly personalized matching pattern is applied, where the first threshold is less than the second threshold.

[0118] When the individual-group matching degree is less than the second threshold, the matching depth is personalized.

[0119] In one specific embodiment, the first and second thresholds are determined by analyzing the balance point between group satisfaction and individual satisfaction in historical data: First, individual-group matching data and corresponding satisfaction scores for all visitors over at least three months are collected. The average satisfaction level for different matching intervals is calculated using a sliding window statistical method. The minimum matching degree where the average satisfaction rate exceeds 90% is set as the first threshold. When the decrease in an adjacent interval first exceeds 15 percentage points, the lower limit of the matching degree corresponding to the interval where the decrease occurs is set as the second threshold. This process is automatically updated quarterly to ensure that the threshold settings always conform to the actual operational status.

[0120] The control command generation and execution terminal generates corresponding scene control commands based on the matched service mode and distributes them to the corresponding exhibition area equipment controller to execute scene switching control.

[0121] Specifically, the generation of scene control instructions includes: H1. If it is a group navigation mode, retrieve a matching scene configuration scheme from the scene template library based on the group preference vector, and generate the corresponding integrated control instructions.

[0122] In practice, the scene template library is constructed as follows: based on historical operational data, combinations of scene equipment parameters that correspond to different group preference vectors and elicit high satisfaction feedback are extracted and stored as scene templates. Each template contains a target group preference vector and a standardized set of equipment instructions bound to it. This set of instructions explicitly specifies the parameters for light color temperature and brightness, background music ID, multimedia content ID, and projection mapping scheme.

[0123] When it is necessary to generate instructions for the group navigation mode, the cosine similarity between the currently calculated group preference vector and the target group preference vector in the scene template library can be calculated to retrieve the scene template with the highest similarity and use the standardized device instruction set contained therein as the integrated control instruction.

[0124] H2. In the case of a mildly personalized mode, supplementary scene instructions are generated by extracting the differentiated preference dimension from the individual interest vector based on the integrated control instructions.

[0125] In the lightly personalized mode, the following steps are performed to generate supplementary scene instructions: calculate the intensity difference between the individual interest vector and the group preference vector in each preference category dimension.

[0126] Dimensions with intensity differences exceeding a preset threshold are selected and marked as significantly different dimensions.

[0127] Each significantly different dimension is mapped to a subset of devices that it can influence without affecting the main scene of the group. The mapping is based on predefined rules; for example, a significant positive difference in the intensity of the classical art dimension can be mapped to instructions to add classical music audio streams to individual terminals and adjust the color temperature of spotlights in their local area to a warm yellow light.

[0128] Based on the mapping relationship, device control instructions are generated specifically for the visitor terminal or its micro-area, serving as supplementary scene instructions.

[0129] In practice, the preset threshold for filtering dimensions with significant differences is set based on the statistical distribution of the intensity differences between individual interest vectors and group preference vectors across each dimension. For example, the top 25% quantile of the absolute values ​​of the intensity differences across all dimensions in historical data can be used as this threshold.

[0130] H3. In the case of a deeply personalized mode, scene switching sequence instructions containing a unique visit path are generated based on the individual's interest vector, and personalized configuration parameters are preloaded for the relevant exhibition area equipment to execute instructions.

[0131] In the deeply personalized mode, the following steps are executed to generate scene switching sequence instructions: Based on the individual interest vector, all exhibits are sorted in descending order according to the intensity value of the associated preference category to form an initial interest point sequence.

[0132] By combining the real-time location of visitors with the geographical location of exhibits, the initial sequence of points of interest is optimized based on the shortest path algorithm (Dijkstra's algorithm) to generate a physically feasible exclusive visitor path.

[0133] Based on the interest intensity value of the preference category associated with each exhibit, a pre-trained regression model is used to predict the preset dwell time of visitors at that exhibit.

[0134] For each exhibit on the path, a scene switching sub-instruction is generated. The sub-instruction contains the device preload parameters of the target exhibit and a trigger timestamp calculated based on the previous dwell time. The preload parameters are derived from the scene template that matches the preference category most relevant to the exhibit.

[0135] All scene switching sub-instructions are assembled into a complete scene switching sequence instruction according to the path order and timestamp.

[0136] In specific implementation, the pre-trained regression model takes the interest intensity values ​​of each dimension in the individual interest vector, real-time behavioral data, the attributes of the exhibit itself, and real-time external conditions as input, and the preset dwell time as output. The specific pre-trained regression model can adopt the gradient boosting decision tree model. The gradient boosting decision tree model is an existing model that can be trained using historical data. The model is also a conventional method, and its specific training process will not be described further.

[0137] It should be added that after receiving the instruction, the equipment controllers in each exhibition area complete the state switch within the specified time window. The scene switch status is fed back to the central processing unit of the exhibition hall in real time, and the scene execution effect is continuously monitored and dynamically adjusted according to the real-time feedback from visitors.

[0138] Please see Figure 2 As shown, the present invention provides an adaptive intelligent switching control method for exhibition hall scenes. The method includes: collecting visitor identity background, declared preference data, real-time behavior data and location information, and generating a visitor information dataset.

[0139] Based on the dataset, dynamic evolutionary individual interest vectors are output by combining temporal analysis of behavioral data, and the real-time intent state of visitors is identified.

[0140] Visitors are dynamically clustered to generate group preference vectors. The similarity between each visitor's individual interest vector and the corresponding group preference vector is calculated using cosine similarity to generate individual-group matching degree.

[0141] The individual-group matching degree is compared with a preset threshold, and a preset service mode is matched according to the comparison result.

[0142] Based on the matched service mode, corresponding scene control instructions are generated and distributed to the corresponding exhibition area equipment controllers to execute scene switching control.

[0143] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. An adaptive intelligent switching control system for exhibition hall scenes, characterized in that, The system includes: Visitor information collection module Collect visitor identity background, declared preference data, real-time behavior data, and location information to generate a visitor information dataset; The individual vector generation module, based on the dataset and combined with temporal analysis of behavioral data, outputs dynamically evolving individual interest vectors. Specifically, it generates an initial interest intensity value based on declared preference data, form type, and historical visit records; calculates behavioral interest intensity by statistically normalizing dwell time and interaction frequency based on visitor temporal behavioral data; linearly weights and fuses the behavioral interest intensity with the initial interest intensity value to generate an updated interest intensity value; outputs a dynamically evolving individual interest vector based on all preference categories and the updated interest intensity value; and identifies the visitor's real-time intent state. The identification of the visitor's real-time intent status includes: Obtain real-time behavioral data streams from visitors; Based on the behavioral data stream statistics, the total continuous dwell time, average movement speed, the proportion of gaze on the exhibits, the operation interval time of each interactive device, and the tortuosity of the movement path are statistically analyzed before the exhibits are set up. The statistical feature indicators are compared with the preset intent state thresholds to output the real-time intent state, which includes at least one of in-depth visit, fast passage, exploration and search, and interactive experience. Store the status identifier associated with the timestamp; The group matching analysis module dynamically clusters visitors based on their real-time location and individual interest vectors to generate group preference vectors. It then calculates the similarity between each visitor's individual interest vector and the corresponding group preference vector using cosine similarity to generate the individual-group matching degree. The service mode determination module compares the individual-group matching degree with a preset threshold and matches a preset service mode based on the comparison result. The control command generation and execution terminal generates corresponding scene control commands based on the matched service mode and distributes them to the corresponding exhibition area equipment controllers to execute scene switching control.

2. The exhibition hall scene adaptive intelligent switching control system as described in claim 1, characterized in that: The individual interest vector output includes: Extract free text or selected preference labels from the declared preference data; The semantic similarity between the free text or preference tag and the semantic extension set of each preset preference category is calculated, and the preference category is mapped according to the calculation result; Each selected preference label is assigned an initial interest intensity value according to a preset allocation rule; Query the visitor's historical visit records. If there is an intersection between the visitor's historical valid preference set and the current declared preference, multiply the initial interest intensity value of the preference category in the intersection by a preset enhancement coefficient and output the final initial interest intensity value. By periodically collecting and quantifying visitor behavior data through a sensor network deployed in the exhibition area, the normalized dwell time and normalized interaction frequency of related interactive devices under each preference category are statistically analyzed. For each preference category, the behavioral interest intensity in the current update cycle is calculated by linearly weighted summing and normalizing the dwell time and the interaction frequency. The behavioral interest intensity is linearly weighted and fused with the initial interest intensity value of the corresponding preference category to generate an updated interest intensity value; A dynamically evolving individual interest vector is generated based on all preference categories and their corresponding updated interest intensity values.

3. The exhibition hall scene adaptive intelligent switching control system as described in claim 2, characterized in that: The specific content of the preset allocation rule includes: Extract the form type to which the preference tags belong; If the preference label comes from a single-choice form, the selected unique preference label will receive the preset highest initial interest intensity value. ; If the preference label comes from a multi-select form, assign the same initial interest intensity value to each selected preference label based on the preset highest initial interest intensity value; If the preference tags are derived from a rating questionnaire, then perform the following steps: Obtain the visitor's selected level identifier and the highest level identifier, and calculate the number of level intervals between them; The relative positional proportion is obtained by calculating the complement of the ratio of the number of interval levels to the total number of levels; The initial interest intensity value is calculated using a preset nonlinear mapping function. , , The preset minimum initial interest intensity value, The sensitivity coefficient is configured.

4. The exhibition hall scene adaptive intelligent switching control system as described in claim 2, characterized in that: The mapping process for the preference categories includes: Determine whether there are any preset preference categories whose calculated similarity scores are higher than preset matching thresholds; If it does not exist, mark the stated preference as an unidentified item and store it in a temporary interest pool; If they exist, determine whether the number of preference categories whose similarity scores exceed the matching threshold is 1. If the value is 1, the declared preference is mapped to that category; otherwise, the preset preference category with the highest similarity score is used as the primary mapping category, and the rest are used as secondary mapping categories, and corresponding labels are applied.

5. The exhibition hall scene adaptive intelligent switching control system as described in claim 4, characterized in that: The calculation of behavioral interest intensity also includes weighted combination matching based on preference categories, specifically including: If the current preference category is the main mapping, then the first weight combination is used, where the weight of normalized dwell time is greater than the weight of normalized interaction frequency. If it is a secondary mapping, the second weight combination is used, where the weight of normalized dwell time is less than the weight of normalized interaction frequency.

6. The exhibition hall scene adaptive intelligent switching control system as described in claim 1, characterized in that: The specific generation of the group preference vector includes: Visitors are dynamically grouped using the DBSCAN clustering algorithm based on their location coordinates to form temporary guided tour groups; Extract the individual interest vectors of all members within each group, and calculate the mean and variance of the interest intensity of the same preference category within the group; Based on the average and variance of the interest intensity, set the initial values ​​of the member weights, and determine whether the member triggers the following conditions: Condition 1: The interest intensity value of this category among members is lower than the group average and the variance is higher than the discrete threshold; Condition 2: The member's dwell time at the current location exceeds the group's average dwell time; Condition 3: The member is identified as being in an in-depth visit status; If condition 1 is triggered, the initial value of the member weight will be reduced to the first weight value; If condition 2 is triggered, the initial value of the member's weight will be increased to the second weight value; If condition 3 is triggered, the initial value of the member weight will be increased to the third weight value, which is greater than the second weight value; If the condition is not triggered, the initial values ​​of the member weights remain unchanged; The final member weight value and its interest intensity value are linearly weighted and fused together, and then divided by the total number of group members to obtain the final group interest intensity value; The final group interest intensity value, which combines all preference categories, is used to generate a group preference vector.

7. The exhibition hall scene adaptive intelligent switching control system as described in claim 1, characterized in that: The matching process for the preset service mode includes: When the individual-group matching degree is greater than or equal to the first threshold, the matching group navigation mode is activated. When the individual-group matching degree is between the first threshold and the second threshold, a slightly personalized matching pattern is applied, where the first threshold is less than the second threshold. When the individual-group matching degree is less than the second threshold, the matching depth is personalized.

8. The exhibition hall scene adaptive intelligent switching control system as described in claim 7, characterized in that: The generation of the scene control commands includes: In the group navigation mode, the system retrieves a matching scene configuration scheme from the scene template library based on the group preference vector and generates the corresponding integrated control commands. In the case of a lightly personalized mode, supplementary scene instructions are generated by extracting the differentiated preference dimension from the individual interest vector based on the integrated control instructions. In the case of a deeply personalized mode, scene switching sequence instructions containing a unique visitor path are generated based on the individual's interest vector, and personalized configuration parameters are preloaded for the relevant exhibition area equipment to execute instructions.

9. A method for adaptive intelligent switching control of exhibition hall scenes, characterized in that: The method includes: Collect visitor identity background, declared preference data, real-time behavior data, and location information to generate a visitor information dataset; Based on the dataset, and combined with time-series analysis of behavioral data, a dynamically evolving individual interest vector is output. Specifically, an initial interest intensity value is generated based on declared preference data, form type, and historical visit records; the behavioral interest intensity is calculated by statistically normalizing dwell time and interaction frequency based on visitor time-series behavioral data; the behavioral interest intensity and the initial interest intensity value are linearly weighted and fused to generate an updated interest intensity value; and a dynamically evolving individual interest vector is output based on all preference categories and the updated interest intensity value; and the visitor's real-time intent state is identified. The identification of the visitor's real-time intent status includes: Obtain real-time behavioral data streams from visitors; Based on the behavioral data stream statistics, the total continuous dwell time, average movement speed, the proportion of gaze on the exhibits, the operation interval time of each interactive device, and the tortuosity of the movement path are statistically analyzed before the exhibits are set up. The statistical feature indicators are compared with the preset intent state thresholds to output the real-time intent state, which includes at least one of in-depth visit, fast passage, exploration and search, and interactive experience. Store the status identifier associated with the timestamp; Visitors are dynamically clustered to generate group preference vectors. The similarity between each visitor's individual interest vector and the corresponding group preference vector is calculated using cosine similarity to generate individual-group matching degree. The individual-group matching degree is compared with a preset threshold, and a preset service mode is matched according to the comparison result; Based on the matched service mode, corresponding scene control instructions are generated and distributed to the corresponding exhibition area equipment controllers to execute scene switching control.

Images