Intelligent furniture scene recommendation method, system and device based on smart center screen

Abstract

The application discloses a smart furniture scene recommendation method, system and device based on a smart middle screen, relates to the technical field of data processing, and comprises the following steps: performing multi-modal semantic fusion on the voice data and the video data of the same user to obtain device-level demand semantic features of each user on each smart furniture device; for a single smart furniture device, performing semantic aggregation on the device-level demand semantic features corresponding to each user on the smart furniture device respectively to obtain corresponding semantic aggregation features; performing semantic deduction on the semantic aggregation features of each smart furniture device based on a device topology graph to obtain device cooperation semantic features; performing semantic decomposition on the device cooperation semantic features based on a pre-generated base scene vector to obtain scene coefficients of each preset base scene; and generating a scene recommendation scheme based on the scene coefficients. The application has the effect of improving the understanding ability of the smart middle screen for multi-user demands.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, system and device for recommending smart furniture scenarios based on a smart central screen. Background Technology

[0002] With the rapid development of smart home technology, smart screens, as the core interactive terminal in the home environment, are gradually evolving from simple device control panels into intelligent hubs with multimodal perception and centralized management capabilities. Smart screens are typically deployed in common areas of the home, such as living rooms and bedrooms. They can not only receive users' voice commands but also perceive users' behavior, posture, and spatial location through vision, thereby understanding user intentions in a more natural interaction process. In practical applications, smart screens face diverse needs arising simultaneously or alternately from multiple family members in the same space. These needs include explicit control requirements clearly expressed by users through voice, as well as implicit needs hidden in user habits, environmental reactions, and multi-user interactions. Different users may have different preferences for the same smart home devices, and their expectations for the linkage of multiple devices may conflict or complement each other, making scene understanding and recommendations particularly complex.

[0003] Currently, handling such scenarios typically relies on single voice command recognition or rule-based automated triggering mechanisms. Voice-based control requires users to issue relatively clear and structured commands, making it difficult to handle the ambiguous expressions and nonverbal intentions commonly found in real-world homes. Rule-based linkage methods, on the other hand, often trigger simply based on preset conditions, lacking a dynamic understanding of the user's actual needs, and are prone to false triggers or delayed responses. More significantly, existing systems generally lack the ability to effectively differentiate and comprehensively process the needs of multiple users. When multiple users have different or even conflicting control needs in the same space, the system often makes decisions based solely on the most recent command or simple user priority, failing to reasonably aggregate and coordinate the needs of each user at the semantic level. This results in inconsistent recommendation results, impacting the user experience.

[0004] Furthermore, smart furniture devices exhibit strong functional interrelationships and collaborative control requirements. For instance, the states of devices such as lights, curtains, air conditioners, and windows often need to coordinate to create a reasonable scene effect. However, currently, most systems treat each device as an independent control object, lacking effective modeling and reasoning regarding the collaborative relationships between devices. Even if some systems support scene modes, their scene definitions often rely on fixed templates preset by manufacturers or manually configured by users, making it difficult to dynamically adjust according to current user combinations, behavioral states, and environmental changes. This results in a significant lack of flexibility and personalization.

[0005] In summary, the uniqueness of smart central screen application scenarios lies in their need to continuously confront the complex environment of interwoven information from multiple users, devices, and modalities. It must accurately identify the explicit control needs of each user while also capturing implicit, unspoken, but objectively existing needs, and then dynamically generate reasonable scenario recommendations based on the collaborative relationships between devices. Existing technologies have significant limitations in these areas, causing smart central screens to remain reliant on passive responses and fixed mode switching in practical use, failing to truly achieve intelligent, personalized, and collaborative scenario recommendations. Therefore, there is an urgent need for a method that can improve the understanding and comprehensive capabilities of multi-user needs, enhance the level of device collaborative reasoning, and support dynamic scenario generation to improve the overall interactive experience and service quality of smart home systems. Summary of the Invention

[0006] To improve the ability of smart central screens to understand the needs of multiple users and thus enhance the accuracy of scene recommendations, this application provides a method, system, and device for smart furniture scene recommendation based on smart central screens.

[0007] Firstly, this application provides a method for recommending smart furniture scenarios based on a smart central screen, employing the following technical solution: The smart furniture scene recommendation method based on the smart central screen includes: Acquire voice and video data of each user in the target space, perform multimodal semantic fusion on the voice and video data of the same user to obtain device-level demand semantic features of each user for each smart furniture device, wherein the device-level demand semantic features are used to characterize the explicit control needs and implicit control needs of the same user for the same smart furniture device. For a single smart furniture device, semantic aggregation is performed on the device-level demand semantic features corresponding to each user for that smart furniture device to obtain corresponding semantic aggregation features, wherein the semantic aggregation features are used to characterize the comprehensive control needs of each user for the same smart furniture device; Based on the pre-generated device topology map, semantic inference is performed on the semantic aggregation features of each smart furniture device to obtain device collaboration semantic features, wherein the device collaboration semantic features are used to characterize the collaborative control requirements between each smart furniture device. Based on the pre-generated base scene vectors, the collaborative semantic features of the devices are semantically decomposed to obtain the scene coefficients of each preset base scene; Based on the aforementioned scene coefficients, a scene recommendation scheme is generated to control the smart central screen and related devices.

[0008] By adopting the above technical solution, voice and video data of each user in the target space are acquired. Multimodal semantic fusion is performed on the voice and video data of the same user to obtain the device-level demand semantic features of each user for each smart furniture device. The device-level demand semantic features are used to represent the explicit and implicit control needs of the same user for the same smart furniture device. Then, for a single smart furniture device, the device-level demand semantic features corresponding to each user for that smart furniture device are semantically aggregated to obtain the corresponding semantic aggregated features. The semantic aggregated features are used to represent the comprehensive control needs of each user for the same smart furniture device. Then, based on the pre-generated device topology map, the semantic aggregated features of each smart furniture device are semantically deduced to obtain device collaboration semantic features. The device collaboration semantic features are used to represent the collaborative control needs between each smart furniture device. Then, based on the pre-generated base scene vector, the device collaboration semantic features are semantically decomposed to obtain the scene coefficients of each preset base scene. Then, based on the scene coefficients, a scene recommendation scheme is generated to control the smart central screen and linked devices. This invention enhances the smart screen's ability to understand the explicit and implicit control needs of multiple users through multimodal semantic fusion of voice and video. By aggregating the semantics of multi-user needs and conducting collaborative reasoning with devices, and dynamically generating scene recommendation schemes based on base scene vectors, the accuracy of scene recommendations is improved, demand conflicts and false triggers are reduced, and the overall interactive experience of the smart home system is enhanced.

[0009] Optionally, the step of performing multimodal semantic fusion on the voice data and video data of the same user to obtain the device-level demand semantic features of each user for each smart furniture device includes: For each user, speech recognition and semantic parsing are performed on the user's speech data to obtain speech semantic features; and human motion recognition and visual target detection are performed on the user's video data to obtain motion semantic features. For the same user, the speech semantic features and the action semantic features are aligned in the time dimension to obtain the user's speech semantic alignment features and action semantic alignment features. Cross-modal fusion is performed on the speech semantic alignment features and the action semantic alignment features to obtain the semantic-action semantic features of the user; The semantic-action semantic features are decoded for device requirements and mapped for intent to obtain the user's device-level requirement semantic features for each smart furniture device.

[0010] By adopting the above technical solution, in order to generate device-level requirement semantic features, for each user, speech recognition and semantic parsing are performed on the user's speech data to obtain speech semantic features; and human motion recognition and visual target detection are performed on the user's video data to obtain motion semantic features. Then, the speech semantic features and motion semantic features of the same user are aligned in the time dimension to obtain the user's speech semantic alignment features and motion semantic alignment features. Then, cross-modal fusion is performed on the speech semantic alignment features and motion semantic alignment features to obtain the user's semantic-action semantic features. Finally, device requirement decoding and intent mapping are performed on the semantic-action semantic features to obtain the user's device-level requirement semantic features for each smart furniture device.

[0011] Optionally, the step of semantically aggregating the device-level demand semantic features corresponding to each user for the smart furniture device to obtain the corresponding semantic aggregated features includes: Obtain the device-level semantic features of each user's requirements for the smart furniture device, and form a group of semantic features of requirements. Conflict detection is performed on any two device-level requirement semantic features in the requirement semantic feature group to obtain a conflict feature pair; and conflict resolution is performed on the conflict feature pair to obtain a conflict resolution feature. The requirement semantic feature group is updated based on the conflict resolution feature to obtain the updated requirement semantic feature group. The features in the updated demand semantic feature group are averaged and pooled to obtain the average demand semantic features. Calculate the semantic similarity between the features in the updated demand semantic feature group and the average demand semantic feature, and normalize each semantic similarity to obtain their respective similarity weights. Based on the respective similarity weights, the features in the updated demand semantic feature group are weighted and aggregated to obtain the semantic aggregate features of the smart furniture device.

[0012] By adopting the above technical solution, in order to generate semantic aggregation features, the device-level demand semantic features corresponding to each user for the smart furniture device are obtained to form a demand semantic feature group. Then, conflict detection is performed on any two device-level demand semantic features in the demand semantic feature group to obtain conflict feature pairs. Conflict resolution is performed on the conflict feature pairs to obtain conflict resolution features. The demand semantic feature group is then updated based on the conflict resolution features to obtain an updated demand semantic feature group. Then, the features in the updated demand semantic feature group are averaged to obtain average demand semantic features. Then, the semantic similarity between the features in the updated demand semantic feature group and the average demand semantic features is calculated, and each semantic similarity is normalized to obtain its corresponding similarity weight. Finally, based on each similarity weight, the features in the updated demand semantic feature group are weighted and aggregated to obtain the semantic aggregation features of the smart furniture device.

[0013] Optionally, the step of resolving the conflict features to obtain conflict-resolved features includes: Obtain the user priority level, historical device control frequency, and historical operating status of the smart furniture device corresponding to the two device-level demand semantic features in the conflict feature pair, respectively. Based on the user priority level, historical device control frequency, and historical operating status, the conflict resolution weights of the two device-level demand semantic features in the conflict feature pair are determined. The conflict resolution weights are positively correlated with the user priority level, positively correlated with the historical device control frequency, and positively correlated with the semantic fit between the historical operating status and the corresponding device-level demand semantic features. Based on the conflict resolution weights, the two device-level requirement semantic features in the conflict feature pair are weighted and averaged to obtain the conflict resolution features.

[0014] By adopting the above technical solution, in order to generate conflict resolution features, the user priority level, historical device control frequency, and historical operating status of the smart furniture device corresponding to the two device-level demand semantic features in the conflict feature pair are obtained respectively. Then, based on the user priority level, historical device control frequency, and historical operating status, the conflict resolution weights of the two device-level demand semantic features in the conflict feature pair are determined. The conflict resolution weights are positively correlated with the user priority level, positively correlated with the historical device control frequency, and positively correlated with the semantic fit between the historical operating status and the corresponding device-level demand semantic features. Then, based on the conflict resolution weights, the two device-level demand semantic features in the conflict feature pair are weighted and averaged to obtain the conflict resolution features.

[0015] Optionally, the step of performing semantic inference on the semantic aggregation features of each smart furniture device based on the pre-generated device topology map to obtain device collaborative semantic features includes: Obtain a pre-generated device topology map, wherein each smart furniture device is a node and the functional linkage relationship or spatial co-occurrence relationship between smart furniture devices is an edge; Based on the device topology graph, the neighboring nodes of each node are determined, and the node features are initialized according to the semantic aggregation features. In the t-th feature propagation, based on the node features after the (t-1)-th feature propagation, the semantic-topological fusion weight between each node and its neighboring nodes is calculated; and based on the semantic-topological fusion weight, the node features of the neighboring nodes are weighted and fused to obtain the corresponding neighbor weighted features; and based on the neighbor weighted features and the node features after the (t-1)-th feature propagation, the node features of the t-th feature propagation are updated to obtain the node features after the t-th feature propagation. When t reaches the preset number of iterations T, the node features of each node are concatenated to obtain the device collaborative semantic features.

[0016] By adopting the above technical solution, in order to generate device collaborative semantic features, a pre-generated device topology map is obtained. The device topology map uses each smart furniture device as a node and the functional linkage relationship or spatial co-occurrence relationship between smart furniture devices as edges. Then, based on the device topology map, the neighbor nodes of each node are determined, and the node features are initialized according to the semantic aggregation features. Then, in the t-th feature propagation, based on the node features after the (t-1)-th feature propagation, the semantic-topology fusion weight between each node and its neighbor nodes is calculated. And, based on the semantic-topology fusion weight, the node features of the neighbor nodes are weighted and fused to obtain the corresponding neighbor weighted features. And, based on the neighbor weighted features and the node features after the (t-1)-th feature propagation, the node features of the t-th feature propagation are updated to obtain the node features after the t-th feature propagation. Then, when t reaches the preset number of iterations T, the node features of each node are concatenated to obtain the device collaborative semantic features.

[0017] Optionally, the step of calculating the semantic-topological fusion weights between each node and its neighboring nodes based on the node features after the (t-1)th feature propagation in the t-th feature propagation includes: For a single node, obtain the node features of the node after the (t-1)th feature propagation, and the node features of the node's neighboring nodes after the (t-1)th feature propagation, and calculate the semantic similarity between the node features of the node and the node features of the neighboring nodes. Obtain the edge weights between the node and its neighboring nodes in the device topology graph, and perform a multiplication operation on the semantic similarity and the edge weights to obtain the corrected weights between the node and its neighboring nodes; The modified weights between the node and its neighboring nodes are normalized to obtain the semantic-topology fusion weights.

[0018] By adopting the above technical solution, in order to generate semantic-topology fusion weights, for a single node, the node features of the node after the (t-1)th feature propagation and the node features of the node's neighboring nodes after the (t-1)th feature propagation are obtained, and the semantic similarity between the node features of the node and the node features of the neighboring nodes is calculated. Then, the edge weights between the node and the neighboring nodes in the device topology graph are obtained, and the semantic similarity and edge weights are multiplied to obtain the corrected weights between the node and the neighboring nodes. Then, the corrected weights between the node and the neighboring nodes are normalized to obtain the semantic-topology fusion weights.

[0019] Optionally, the step of semantically decomposing the device collaborative semantic features based on pre-generated base scene vectors to obtain scene coefficients for each preset base scene includes: Obtain a pre-generated set of base scene vectors, wherein the set of base scene vectors includes base scene vectors corresponding to multiple preset base scenes, and each base scene vector is used to represent the semantic center of a preset base scene; Based on the base scene vector, the device collaborative semantic features are semantically projected to obtain multiple collaborative semantic projection features, wherein the collaborative semantic projection features and the base scene vector are in the same semantic space; Calculate the semantic similarity between the collaborative semantic projection features and each of the base scene vectors; The semantic similarity scores are normalized to obtain the corresponding scene coefficients.

[0020] By adopting the above technical solution, in order to generate scene coefficients, a pre-generated set of base scene vectors is obtained. The set of base scene vectors includes base scene vectors corresponding to multiple preset base scenes. Each base scene vector is used to represent the semantic center of a preset base scene. Then, based on the base scene vectors, the device collaborative semantic features are semantically projected to obtain multiple collaborative semantic projection features. The collaborative semantic projection features and the base scene vectors are in the same semantic space. Then, the semantic similarity between the collaborative semantic projection features and each base scene vector is calculated. Then, each semantic similarity is normalized to obtain the corresponding scene coefficients.

[0021] Optionally, the step of semantically projecting the device collaborative semantic features based on the base scene vector to obtain multiple collaborative semantic projection features includes: Each of the base scene vectors is normalized to obtain the corresponding unit projection vector, wherein the unit projection vector has the same semantic direction as the base scene vector and has a magnitude of 1; Based on the unit projection vector, the device collaborative semantic features are vector-projected to obtain the corresponding intermediate projection features; The intermediate projection features are subjected to nonlinear activation processing to obtain the corresponding collaborative semantic projection features.

[0022] By adopting the above technical solution, in order to generate collaborative semantic projection features, each base scene vector is normalized to obtain the corresponding unit projection vector. The unit projection vector has the same semantic direction as the base scene vector and the magnitude is 1. Then, based on the unit projection vector, the device collaborative semantic features are vector-projected to obtain the corresponding intermediate projection features. Then, the intermediate projection features are nonlinearly activated to obtain the corresponding collaborative semantic projection features.

[0023] Secondly, this application also provides an intelligent furniture scene recommendation system based on a smart central screen, which adopts the following technical solution: A smart furniture scene recommendation system based on a smart central screen includes: The multimodal semantic fusion module is used to acquire voice data and video data of each user in the target space, and to perform multimodal semantic fusion on the voice data and video data of the same user to obtain the device-level demand semantic features of each user for each smart furniture device. The device-level demand semantic features are used to characterize the explicit control needs and implicit control needs of the same user for the same smart furniture device. The semantic aggregation module is used to perform semantic aggregation on the device-level demand semantic features corresponding to each user for a single smart furniture device, and obtain corresponding semantic aggregation features, wherein the semantic aggregation features are used to characterize the comprehensive control needs of each user for the same smart furniture device. The semantic inference module is used to perform semantic inference on the semantic aggregation features of each smart furniture device based on the pre-generated device topology map to obtain device collaboration semantic features, wherein the device collaboration semantic features are used to characterize the collaborative control requirements between each smart furniture device. The semantic decomposition module is used to perform semantic decomposition on the collaborative semantic features of the device based on the pre-generated base scene vectors to obtain the scene coefficients of each preset base scene; The scene recommendation generation module is used to generate scene recommendation schemes based on the scene coefficients, so as to control the smart central screen and linked devices.

[0024] Thirdly, this application also provides a computer device, which adopts the following technical solution: A computer device includes a memory and a processor, the memory storing a computer program executable on the processor, the processor executing the computer program to implement the method described in the first aspect.

[0025] In summary, this application includes at least the following beneficial technical effects: acquiring voice and video data of each user within the target space; performing multimodal semantic fusion on the voice and video data of the same user to obtain device-level demand semantic features for each user's smart furniture devices; wherein, the device-level demand semantic features are used to characterize the explicit and implicit control needs of the same user for the same smart furniture device; then, for a single smart furniture device, semantically aggregating the device-level demand semantic features corresponding to each user for that smart furniture device to obtain corresponding semantic aggregation features; wherein, the semantic aggregation features are used to characterize the comprehensive control needs of each user for the same smart furniture device; then, based on a pre-generated device topology map, semantically deriving the semantic aggregation features of each smart furniture device to obtain device collaboration semantic features; wherein, the device collaboration semantic features are used to characterize the collaborative control needs between each smart furniture device; then, based on a pre-generated base scene vector, semantically decomposing the device collaboration semantic features to obtain scene coefficients for each preset base scene; and then, based on the scene coefficients, generating scene recommendation schemes to control the smart central screen and linked devices. This invention enhances the smart screen's ability to understand the explicit and implicit control needs of multiple users through multimodal semantic fusion of voice and video. By aggregating the semantics of multi-user needs and conducting collaborative reasoning with devices, and dynamically generating scene recommendation schemes based on base scene vectors, the accuracy of scene recommendations is improved, demand conflicts and false triggers are reduced, and the overall interactive experience of the smart home system is enhanced. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the overall process of an embodiment of this application.

[0027] Figure 2 This is a schematic diagram of the system structure of this application.

[0028] Figure 3 This is a structural block diagram of the computer device described in this application. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0030] This application discloses a method for recommending smart furniture scenarios based on a smart central screen.

[0031] Reference Figure 1 The smart furniture scene recommendation method based on the smart central screen includes: Step S11: Acquire the voice data and video data of each user in the target space, perform multimodal semantic fusion on the voice data and video data of the same user, and obtain the device-level demand semantic features of each user for each smart furniture device. The device-level demand semantic features are used to characterize the explicit control needs and implicit control needs of the same user for the same smart furniture device.

[0032] It should be noted that in step S11, by simultaneously acquiring the voice and video data of each user within the target space and fusing the two types of modal information for the same user, the user's control intentions towards various smart furniture devices can be comprehensively understood from a semantic perspective. Since single voice command recognition in the background technology struggles to capture implicit needs not explicitly expressed by the user, while the behavioral, posture, and spatial location information contained in video data can supplement need cues beyond voice, this step, through joint analysis of multimodal data, enables device-level demand semantic features to reflect both explicit control needs explicitly expressed by the user through voice and implicit control needs hidden in user behavior and environmental interactions. This provides a more complete semantic foundation for subsequent multi-user demand synthesis and scenario reasoning.

[0033] Step S12: For a single smart furniture device, perform semantic aggregation on the device-level demand semantic features corresponding to each user for the smart furniture device to obtain the corresponding semantic aggregation features. The semantic aggregation features are used to characterize the comprehensive control needs of each user for the same smart furniture device.

[0034] It should be noted that in step S12, for a single smart furniture device, the semantic features of device-level demands formed by different users for that device are semantically aggregated. The purpose is to form a semantic aggregate feature that can represent the overall control tendency of the device when multiple user demands coexist or even conflict. Background technology points out that existing systems often process multiple user demands based only on the most recent instruction or simple priority, which can easily lead to demand confusion or response deviation. This step, through semantic-level aggregation rather than simple rule selection, can integrate the different demands of various users for the same device, reduce direct conflicts between individual demands, and enable subsequent collaborative reasoning based on device state to be carried out on a more reasonable comprehensive demand basis.

[0035] Step S13: Based on the pre-generated device topology map, semantic inference is performed on the semantic aggregation features of each smart furniture device to obtain device collaboration semantic features, wherein the device collaboration semantic features are used to characterize the collaborative control requirements between each smart furniture device.

[0036] It should be noted that in step S13, semantic inference is performed on the semantic aggregation features of each smart furniture device based on the pre-built device topology graph. The purpose is to infer the cross-device collaborative control requirements by utilizing the relationships between devices. As mentioned in the background technology, existing solutions usually treat each device as an independent control object, lacking modeling of the functional associations and collaborative relationships between devices, making it difficult for scene recommendations to form reasonable linkage effects. This step uses the device connection or functional dependency relationships expressed by the topology graph to propagate and infer the comprehensive requirements of a single device on the graph, thereby obtaining the device collaborative semantic features, so that the recommendation results can reflect the mutual cooperation and constraints of multiple devices in state adjustment.

[0037] Step S14: Based on the pre-generated base scene vector, perform semantic decomposition on the device collaborative semantic features to obtain the scene coefficients of each preset base scene.

[0038] It should be noted that in step S14, the device collaborative semantic features are semantically decomposed using pre-generated base scene vectors to obtain scene coefficients corresponding to each preset base scene. In the background technology, existing scene patterns are mostly fixed templates, making it difficult to dynamically adjust according to the current multi-user needs and environmental conditions. This step, however, projects complex collaborative semantic features onto a set of base scene vectors, expressing the degree of matching between the current needs and each basic scene pattern in coefficient form. This decomposition method allows scene recommendation to move beyond simply selecting a complete preset scene; instead, it allows for the flexible combination of coefficients from multiple base scenes based on actual semantic features, providing a quantitative basis for dynamically generating personalized scene solutions.

[0039] Step S15: Based on the scene coefficient, generate a scene recommendation scheme to control the smart central screen and linked devices.

[0040] It should be noted that in step S15, a specific scene recommendation scheme is generated based on the decomposed scene coefficients, and the smart central screen and linked devices are controlled accordingly. In the background technology, smart central screens in practical use still mainly remain at the level of passive response and fixed mode switching. This step, by mapping scene coefficients to executable device control strategies, can output a recommendation scheme that better fits the current actual state, taking into account the needs of multiple users, device collaboration relationships, and dynamic environments. The final control actions include both adjustments to the content displayed or the interactive state of the smart central screen, as well as the coordinated control of related linked devices, thereby improving the overall responsiveness and user experience of the smart home system.

[0041] In the above implementation, voice and video data of each user within the target space are acquired. Multimodal semantic fusion is performed on the voice and video data of the same user to obtain device-level demand semantic features for each user's smart furniture devices. These device-level demand semantic features represent the explicit and implicit control needs of the same user for the same smart furniture device. Then, for a single smart furniture device, the device-level demand semantic features corresponding to each user for that smart furniture device are semantically aggregated to obtain corresponding semantic aggregated features. These semantic aggregated features represent the comprehensive control needs of each user for the same smart furniture device. Then, based on a pre-generated device topology map, the semantic aggregated features of each smart furniture device are semantically deduced to obtain device collaboration semantic features. These device collaboration semantic features represent the collaborative control needs between each smart furniture device. Then, based on a pre-generated base scene vector, the device collaboration semantic features are semantically decomposed to obtain scene coefficients for each preset base scene. Finally, based on the scene coefficients, a scene recommendation scheme is generated to control the smart central screen and linked devices. This invention enhances the smart screen's ability to understand the explicit and implicit control needs of multiple users through multimodal semantic fusion of voice and video. By aggregating the semantics of multi-user needs and conducting collaborative reasoning with devices, and dynamically generating scene recommendation schemes based on base scene vectors, the accuracy of scene recommendations is improved, demand conflicts and false triggers are reduced, and the overall interactive experience of the smart home system is enhanced.

[0042] As a further implementation of the method, the step of performing multimodal semantic fusion on the voice data and video data of the same user to obtain the device-level demand semantic features of each user for each smart furniture device includes: Step S21: For each user, perform speech recognition and semantic parsing on the user's speech data to obtain speech semantic features; and perform human motion recognition and visual target detection on the user's video data to obtain motion semantic features.

[0043] Step S22: Align the speech semantic features and action semantic features of the same user in the time dimension to obtain the user's speech semantic alignment features and action semantic alignment features.

[0044] Step S23: Perform cross-modal fusion of speech semantic alignment features and action semantic alignment features to obtain the semantic-action semantic features of the user.

[0045] Step S24: Decode the semantic-action semantic features for device requirements and map the intent to obtain the user's device-level semantic features for each smart furniture device.

[0046] It should be noted that the core of this implementation method lies in enhancing the comprehensive understanding of users' explicit and implicit control needs through the synergistic complementarity of multimodal information. Voice data, after recognition and semantic parsing, can accurately capture users' explicitly expressed control intentions, but relying solely on voice is insufficient to uncover unspoken user needs. Video data, through human motion recognition and visual target detection, can capture users' behavioral tendencies, pointing gestures, and interaction postures with devices in space, thus supplementing implicit need cues. Aligning the two types of features in the temporal dimension ensures that verbal expressions and behavioral performances correspond at the same moment, avoiding semantic confusion caused by temporal misalignment between modalities. Based on this, cross-modal fusion is performed, allowing verbal and behavioral intentions to mutually corroborate and reinforce each other in a unified representation, improving the accuracy and robustness of judging users' true intentions. Finally, through device need decoding and intent mapping, the fused user intent is directed and associated with specific smart furniture devices, forming device-level need semantic features. This ensures that the expression of needs is both user-specific and device-oriented, providing a reliable and fine-grained semantic foundation for subsequent multi-user need aggregation and scenario-level collaborative reasoning.

[0047] In the above implementation, in order to generate device-level requirement semantic features, for each user, speech recognition and semantic parsing are performed on the user's speech data to obtain speech semantic features; and human motion recognition and visual target detection are performed on the user's video data to obtain motion semantic features. Then, the speech semantic features and motion semantic features of the same user are aligned in the time dimension to obtain the user's speech semantic alignment features and motion semantic alignment features. Then, cross-modal fusion is performed on the speech semantic alignment features and motion semantic alignment features to obtain the user's semantic-action semantic features. Finally, device requirement decoding and intent mapping are performed on the semantic-action semantic features to obtain the user's device-level requirement semantic features for each smart furniture device.

[0048] As a further implementation of the method, the step of semantically aggregating the device-level demand semantic features corresponding to each user for the smart furniture device to obtain the corresponding semantic aggregated features includes: Step S31: Obtain the device-level demand semantic features corresponding to each user for the smart furniture device, and form a demand semantic feature group.

[0049] Step S32: Detect conflict between any two device-level requirement semantic features in the requirement semantic feature group to obtain a conflict feature pair; and resolve the conflict feature pair to obtain a conflict resolution feature.

[0050] Step S33: Update the demand semantic feature group based on the conflict resolution features to obtain the updated demand semantic feature group.

[0051] Step S34: Perform average pooling on the features in the updated demand semantic feature group to obtain the average demand semantic features.

[0052] Step S35: Calculate the semantic similarity between the features in the updated demand semantic feature group and the average demand semantic features, and normalize each semantic similarity to obtain their respective similarity weights.

[0053] Step S36: Based on each similarity weight, perform weighted aggregation on the features in the updated demand semantic feature group to obtain the semantic aggregated features of the smart furniture device.

[0054] It should be noted that this implementation addresses the issue of inconsistent or even conflicting needs among multiple users for the same smart furniture device. It achieves a reasonable transformation from individual needs to comprehensive device-level needs through semantic conflict detection and resolution, and similarity-based weighted aggregation. First, the semantic features of each user's device-level needs are acquired to form a feature group. Conflict detection is then performed on any pair of features to identify semantically contradictory feature pairs. Conflict resolution is used to reconcile contradictions and avoid directly ignoring a user's needs. The feature group is updated using the resolved features. Then, the average pooling result is used as a reference for the overall demand trend. The semantic similarity between each feature and the average demand is calculated and normalized to obtain weights. Features consistent with the overall trend receive higher weights, while features deviating significantly are appropriately suppressed. Finally, the weighted aggregation of semantic features reflects the common control tendencies of most users while preserving reasonable individual differences, reducing the interference of conflicting needs on the recommendation results, and providing a more reasonable and stable semantic input for subsequent device collaborative reasoning.

[0055] In the above implementation, in order to generate semantic aggregation features, the device-level demand semantic features corresponding to each user for the smart furniture device are obtained to form a demand semantic feature group. Then, conflict detection is performed on any two device-level demand semantic features in the demand semantic feature group to obtain conflict feature pairs. Conflict resolution is performed on the conflict feature pairs to obtain conflict resolution features. The demand semantic feature group is then updated based on the conflict resolution features to obtain an updated demand semantic feature group. The features in the updated demand semantic feature group are then averaged to obtain an average demand semantic feature. The semantic similarity between the features in the updated demand semantic feature group and the average demand semantic feature is then calculated, and each semantic similarity is normalized to obtain its corresponding similarity weight. Finally, based on each similarity weight, the features in the updated demand semantic feature group are weighted and aggregated to obtain the semantic aggregation features of the smart furniture device.

[0056] As a further implementation of the method, the step of resolving conflicts in the conflict feature pairs to obtain conflict-resolved features includes: Step S41: Obtain the user priority level, historical device control frequency, and historical operating status of the smart furniture device corresponding to the two device-level demand semantic features in the conflict feature pair.

[0057] Step S42: Based on the user priority level, historical device control frequency, and historical operating status, determine the conflict resolution weights of the two device-level requirement semantic features in the conflict feature pair. The conflict resolution weights are positively correlated with the user priority level, the historical device control frequency, and the semantic fit between the historical operating status and the corresponding device-level requirement semantic features.

[0058] Step S43: Based on the conflict resolution weight, perform a weighted average of the two device-level requirement semantic features in the conflict feature pair to obtain the conflict resolution features.

[0059] It should be noted that this implementation addresses the issue of inconsistent needs among multiple users for the same smart furniture device. It achieves a reasonable transformation from individual needs to comprehensive device needs through semantic conflict detection and resolution, and similarity-based weighted aggregation. First, the semantic features of different users' device-level needs are grouped into a demand semantic feature group. Conflict detection is then performed on any two features to identify contradictory demand expressions. A conflict resolution mechanism is used to semantically reconcile contradictory parts, avoiding the direct omission of any user's needs. The demand semantic feature group is updated using the conflict-resolved features, eliminating obviously conflicting intentions without losing effective information and improving the stability of subsequent aggregation. Next, average pooling is used to obtain the general trend of the device's needs, and weights are calculated based on the semantic similarity between each feature and the average needs. Features consistent with the overall demand trend receive higher weights, while features deviating significantly are appropriately suppressed. This better reflects the common demands of most users in the aggregation result while preserving reasonable individual differences. Finally, the semantic aggregated features obtained through weighted aggregation comprehensively reflect the overall control tendencies of multiple users for the same smart furniture device, providing a more reasonable and robust semantic input for subsequent collaborative reasoning based on the device topology graph.

[0060] In the above implementation, in order to generate conflict resolution features, the user priority level, historical device control frequency, and historical operating status of the smart furniture device corresponding to the two device-level demand semantic features in the conflict feature pair are obtained respectively. Then, based on the user priority level, historical device control frequency, and historical operating status, the conflict resolution weights of the two device-level demand semantic features in the conflict feature pair are determined. The conflict resolution weights are positively correlated with the user priority level, positively correlated with the historical device control frequency, and positively correlated with the semantic fit between the historical operating status and the corresponding device-level demand semantic features. Then, based on the conflict resolution weights, the two device-level demand semantic features in the conflict feature pair are weighted and averaged to obtain the conflict resolution features.

[0061] As a further implementation of the method, the step of semantically deriving the semantic aggregation features of each smart furniture device based on a pre-generated device topology map to obtain device collaborative semantic features includes: Step S51: Obtain the pre-generated device topology map, wherein each smart furniture device is a node and the functional linkage relationship or spatial co-occurrence relationship between smart furniture devices is an edge.

[0062] Step S52: Based on the device topology map, determine the neighboring nodes of each node, and initialize the node features according to the semantic aggregation features.

[0063] Step S53: In the t-th feature propagation, based on the node features after the (t-1)-th feature propagation, calculate the semantic-topological fusion weight between each node and its neighboring nodes; and, based on the semantic-topological fusion weight, perform weighted fusion on the node features of the neighboring nodes to obtain the corresponding neighbor weighted features; and, based on the neighbor weighted features and the node features after the (t-1)-th feature propagation, update the node features of the t-th feature propagation to obtain the node features after the t-th feature propagation.

[0064] Step S54: When t reaches the preset number of iterations T, the node features of each node are concatenated to obtain the device collaborative semantic features.

[0065] It should be noted that this implementation method achieves effective derivation from the comprehensive needs of a single device to the collaborative needs of multiple devices through multi-round feature propagation on the device topology graph. Background technology points out that existing smart home systems typically treat each smart home device as an independent control object, lacking effective modeling of functional linkages and spatial co-occurrence relationships between devices, making it difficult to achieve reasonable cross-device linkage effects in scene recommendations. This implementation method first constructs a device topology graph with devices as nodes and functional linkage or spatial co-occurrence relationships as edges, allowing the potential collaborative relationships between devices to be explicitly expressed. Based on this, the semantic aggregation features of each device are used as the initial state of the nodes, and the demand information of neighboring nodes is gradually absorbed through multiple feature propagations. In each round of propagation, the fusion weight is dynamically calculated based on the semantic relevance and topological relationship between the current node and its neighboring nodes, enabling neighboring devices that are functionally or spatially closely related to the current device and have consistent demand semantics to exert a stronger semantic influence on the current device. This weighted fusion mechanism considers both the static topological associations between devices and the dynamic matching degree of the current demand semantics, thereby avoiding the blind introduction of irrelevant device information. After multiple iterations, the features of each node not only contain its own requirement information, but also integrate the requirement semantics of the devices directly or indirectly related to it. Finally, the device collaborative semantic features obtained by splicing all the node features can express the mutual cooperation and constraint relationship between the devices in terms of control requirements. This provides a more comprehensive collaborative semantic foundation for the dynamic generation of subsequent base scene decomposition and scene recommendation schemes, and enhances the collaborative rationality and real-time adaptability of the recommendations.

[0066] In the above implementation, in order to generate device collaborative semantic features, a pre-generated device topology graph is obtained. The device topology graph uses each smart furniture device as a node and the functional linkage relationship or spatial co-occurrence relationship between smart furniture devices as edges. Then, based on the device topology graph, the neighboring nodes of each node are determined, and the node features are initialized according to the semantic aggregation features. Then, in the t-th feature propagation, based on the node features after the (t-1)-th feature propagation, the semantic-topology fusion weight between each node and its neighboring nodes is calculated. And, based on the semantic-topology fusion weight, the node features of the neighboring nodes are weighted and fused to obtain the corresponding neighbor weighted features. And, based on the neighbor weighted features and the node features after the (t-1)-th feature propagation, the node features of the t-th feature propagation are updated to obtain the node features after the t-th feature propagation. Then, when t reaches the preset iteration number T, the node features of each node are concatenated to obtain the device collaborative semantic features.

[0067] As a further implementation of the method, in the t-th feature propagation, the step of calculating the semantic-topological fusion weights between each node and its neighboring nodes based on the node features after the (t-1)-th feature propagation includes: Step S61: For a single node, obtain the node features of the node after the (t-1)th feature propagation and the node features of the node's neighboring nodes after the (t-1)th feature propagation, and calculate the semantic similarity between the node features of the node and the node features of the neighboring nodes.

[0068] Step S62: Obtain the edge weights between the node and its neighboring nodes in the device topology graph, and perform a multiplication operation on the semantic similarity and edge weights to obtain the corrected weights between the node and its neighboring nodes.

[0069] Step S63: Normalize the modified weights between the node and its neighboring nodes to obtain the semantic-topology fusion weights.

[0070] It should be noted that this implementation method dynamically combines the inherent topological relationships between devices with the semantics of current requirements to calculate the fusion weights during feature propagation, thereby improving the rationality and accuracy of collaborative inference. As mentioned in the background, existing systems lack effective modeling of functional linkages and spatial co-occurrence relationships between devices. Even when simple linkages exist, they are mostly based on fixed rules, making it difficult to adapt to the dynamic changes in current multi-user needs. If feature propagation relies solely on static edge weights in the device topology graph, the relevance of actual requirement semantics is easily overlooked, leading to the introduction of irrelevant device information. If only semantic similarity is relied upon, the inherent functional cooperation relationships between devices may be ignored, causing collaborative inference to deviate from the actual device associations. Therefore, this step first calculates the semantic similarity between a node and its neighboring nodes under the current feature representation to determine whether the control intentions between the two are related from the requirement perspective. Then, it combines the predefined edge weights in the topology graph to confirm the inherent association strength between the two from the perspective of device function or spatial relationship. By multiplying and correcting semantic similarity and edge weights, only neighboring nodes that have high relevance in both semantic and topological dimensions can have a strong influence on the current node, thereby effectively suppressing interference from irrelevant or weakly related information. The corrected weights are then normalized to ensure a consistent scale for nodes with different numbers of neighbors during fusion, preventing imbalances in feature propagation due to differences in the number of neighbors. This dynamic and dual-constraint weight calculation method allows the derivation of device collaborative semantic features to both conform to the objective relationship structure between devices and flexibly adapt to the actual distribution of current multi-user needs, providing a more accurate and robust collaborative semantic foundation for subsequent scenario recommendations.

[0071] In the above implementation, in order to generate semantic-topology fusion weights, for a single node, the node features of the node after the (t-1)th feature propagation and the node features of the node's neighboring nodes after the (t-1)th feature propagation are obtained, and the semantic similarity between the node features of the node and the node features of the neighboring nodes is calculated. Then, the edge weights between the node and the neighboring nodes in the device topology graph are obtained, and the semantic similarity and edge weights are multiplied to obtain the corrected weights between the node and the neighboring nodes. Then, the corrected weights between the node and the neighboring nodes are normalized to obtain the semantic-topology fusion weights.

[0072] As a further implementation of the method, the step of semantically decomposing the device collaborative semantic features based on pre-generated base scene vectors to obtain scene coefficients for each preset base scene includes: Step S71: Obtain a pre-generated set of base scene vectors, wherein the set of base scene vectors includes base scene vectors corresponding to multiple preset base scenes, and each base scene vector is used to represent the semantic center of a preset base scene.

[0073] Step S72: Based on the base scene vector, perform semantic projection on the device collaborative semantic features to obtain multiple collaborative semantic projection features, wherein the collaborative semantic projection features and the base scene vector are in the same semantic space.

[0074] Step S73: Calculate the semantic similarity between the collaborative semantic projection features and each base scene vector.

[0075] Step S74: Normalize each semantic similarity to obtain the corresponding scene coefficient.

[0076] It should be noted that this implementation method achieves the transformation from device-coordinated semantic features to scene coefficients through semantic projection and similarity calculation of base scene vectors, enabling scene recommendation to break free from the limitations of fixed templates. Device-coordinated semantic features contain the comprehensive control intentions of multiple devices under functional linkage and spatial co-occurrence relationships. Their semantic dimension is high and their structure complex, making direct application to scene decision-making difficult. The pre-generated set of base scene vectors represents the central position of several basic scenes in the semantic space. By projecting device-coordinated semantic features onto the same semantic space, the current collaborative requirement becomes comparable to each preset base scene. Based on this, the semantic similarity between the projected features and each base scene vector can be calculated, quantitatively assessing the degree of matching between the current requirement and each basic scene pattern. Normalizing the similarity yields scene coefficients, allowing each coefficient to reflect the distribution of the current requirement across different base scenes on a uniform scale. This process does not simply select a fixed scene, but rather expresses the degree of matching of the current requirement across multiple base scenes in the form of coefficients. This supports weighted combinations of multiple base scenes, generating more personalized and dynamically adaptive scene recommendation schemes, improving the flexibility and accuracy of recommendations.

[0077] In the above implementation, in order to generate scene coefficients, a pre-generated set of base scene vectors is obtained. The set of base scene vectors includes base scene vectors corresponding to multiple preset base scenes. Each base scene vector is used to represent the semantic center of a preset base scene. Then, based on the base scene vectors, the device collaborative semantic features are semantically projected to obtain multiple collaborative semantic projection features. The collaborative semantic projection features and the base scene vectors are in the same semantic space. Then, the semantic similarity between the collaborative semantic projection features and each base scene vector is calculated. Then, each semantic similarity is normalized to obtain the corresponding scene coefficients.

[0078] As a further implementation of the method, the step of semantically projecting device collaborative semantic features based on base scene vectors to obtain multiple collaborative semantic projection features includes: Step S81: Normalize each base scene vector to obtain the corresponding unit projection vector, wherein the unit projection vector has the same semantic direction as the base scene vector and the magnitude is 1.

[0079] Step S82: Based on the unit projection vector, perform vector projection on the device collaborative semantic features to obtain the corresponding intermediate projection features.

[0080] Step S83: Perform nonlinear activation processing on the intermediate projection features to obtain the corresponding collaborative semantic projection features.

[0081] It should be noted that, in this embodiment, normalizing the base scene vectors eliminates the differences in vector magnitudes between different preset base scenes, ensuring that subsequent projection calculations retain only semantic matching information in the direction, thus avoiding biases introduced by the different scales of the base scene vectors themselves. Projecting device collaboration semantic features based on unit projection vectors essentially decomposes the current device collaboration requirements in the semantic direction of each preset base scene, obtaining the magnitude of the components in that direction, thereby quantitatively reflecting the degree of matching between the current requirements and each base scene pattern. Applying nonlinear activation processing to the intermediate projection features helps to enhance the feature response in significantly matching directions while suppressing interference from weakly correlated or noisy directions, making the obtained collaborative semantic projection features more discriminative and robust. This processing method enables subsequent calculations of the semantic similarity between collaborative semantic projection features and each base scene vector to more accurately reflect the true distribution of current multi-device collaboration requirements in different preset scene directions, thereby providing a more reliable semantic basis for the normalization calculation of scene coefficients and dynamic scene recommendation, improving the accuracy and adaptability of the recommendation results.

[0082] By adopting the above technical solution, in order to generate collaborative semantic projection features, each base scene vector is normalized to obtain the corresponding unit projection vector. The unit projection vector has the same semantic direction as the base scene vector and the magnitude is 1. Then, based on the unit projection vector, the device collaborative semantic features are vector-projected to obtain the corresponding intermediate projection features. Then, the intermediate projection features are nonlinearly activated to obtain the corresponding collaborative semantic projection features.

[0083] This application also discloses an intelligent furniture scene recommendation system based on a smart central screen.

[0084] refer to Figure 2 A smart furniture scene recommendation system based on a smart central screen includes: The multimodal semantic fusion module is used to acquire voice data and video data of each user in the target space, and to perform multimodal semantic fusion on the voice data and video data of the same user to obtain the device-level demand semantic features of each user for each smart furniture device. Among them, the device-level demand semantic features are used to characterize the explicit control needs and implicit control needs of the same user for the same smart furniture device. The semantic aggregation module is used to perform semantic aggregation on the device-level demand semantic features of each user for a single smart furniture device, and obtain the corresponding semantic aggregation features. The semantic aggregation features are used to characterize the comprehensive control needs of each user for the same smart furniture device. The semantic inference module is used to perform semantic inference on the semantic aggregation features of each smart furniture device based on the pre-generated device topology map, so as to obtain the device collaboration semantic features. The device collaboration semantic features are used to characterize the collaborative control requirements between the smart furniture devices. The semantic decomposition module is used to perform semantic decomposition on the device collaborative semantic features based on the pre-generated base scene vectors to obtain the scene coefficients of each preset base scene; The scene recommendation generation module is used to generate scene recommendation schemes based on scene coefficients in order to control the smart central screen and linked devices.

[0085] The intelligent furniture scene recommendation system based on a smart central screen of the present invention can implement any of the methods in the intelligent furniture scene recommendation method based on a smart central screen, and the specific working process of the intelligent furniture scene recommendation system based on a smart central screen of the present invention can refer to the corresponding process in the above-mentioned intelligent furniture scene recommendation method based on a smart central screen.

[0086] This application also discloses a computer device.

[0087] refer to Figure 3 A computer device includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement any of the above-described methods for recommending smart furniture scenarios based on a smart central screen.

[0088] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A method for recommending smart furniture scenarios based on a smart central screen, characterized in that, include: Acquire voice and video data of each user in the target space, perform multimodal semantic fusion on the voice and video data of the same user to obtain device-level demand semantic features of each user for each smart furniture device, wherein the device-level demand semantic features are used to characterize the explicit control needs and implicit control needs of the same user for the same smart furniture device. For a single smart furniture device, semantic aggregation is performed on the device-level demand semantic features corresponding to each user for that smart furniture device to obtain corresponding semantic aggregation features, wherein the semantic aggregation features are used to characterize the comprehensive control needs of each user for the same smart furniture device; Based on the pre-generated device topology map, semantic inference is performed on the semantic aggregation features of each smart furniture device to obtain device collaboration semantic features, wherein the device collaboration semantic features are used to characterize the collaborative control requirements between each smart furniture device. Based on the pre-generated base scene vectors, the collaborative semantic features of the devices are semantically decomposed to obtain the scene coefficients of each preset base scene; Based on the aforementioned scene coefficients, a scene recommendation scheme is generated to control the smart central screen and related devices.

2. The intelligent furniture scene recommendation method based on a smart central screen according to claim 1, characterized in that, The step of performing multimodal semantic fusion on the voice data and video data of the same user to obtain the device-level semantic features of each user's requirements for each smart furniture device includes: For each user, speech recognition and semantic parsing are performed on the user's speech data to obtain speech semantic features; and human motion recognition and visual target detection are performed on the user's video data to obtain motion semantic features. For the same user, the speech semantic features and the action semantic features are aligned in the time dimension to obtain the user's speech semantic alignment features and action semantic alignment features. Cross-modal fusion is performed on the speech semantic alignment features and the action semantic alignment features to obtain the semantic-action semantic features of the user; The semantic-action semantic features are decoded for device requirements and mapped for intent to obtain the user's device-level requirement semantic features for each smart furniture device.

3. The intelligent furniture scene recommendation method based on a smart central screen according to claim 1, characterized in that, The step of semantically aggregating the semantic features of the device-level requirements for each user's smart furniture device to obtain corresponding semantic aggregated features includes: Obtain the device-level semantic features of each user's requirements for the smart furniture device, and form a group of semantic features of requirements. Conflict detection is performed on any two device-level requirement semantic features in the requirement semantic feature group to obtain a conflict feature pair; and conflict resolution is performed on the conflict feature pair to obtain a conflict resolution feature. The requirement semantic feature group is updated based on the conflict resolution feature to obtain the updated requirement semantic feature group. The features in the updated demand semantic feature group are averaged and pooled to obtain the average demand semantic features. Calculate the semantic similarity between the features in the updated demand semantic feature group and the average demand semantic feature, and normalize each semantic similarity to obtain their respective similarity weights. Based on the respective similarity weights, the features in the updated demand semantic feature group are weighted and aggregated to obtain the semantic aggregate features of the smart furniture device.

4. The intelligent furniture scene recommendation method based on a smart central screen according to claim 3, characterized in that, The step of resolving the conflict features to obtain conflict-resolved features includes: Obtain the user priority level, historical device control frequency, and historical operating status of the smart furniture device corresponding to the two device-level demand semantic features in the conflict feature pair, respectively. Based on the user priority level, historical device control frequency, and historical operating status, the conflict resolution weights of the two device-level demand semantic features in the conflict feature pair are determined. The conflict resolution weights are positively correlated with the user priority level, positively correlated with the historical device control frequency, and positively correlated with the semantic fit between the historical operating status and the corresponding device-level demand semantic features. Based on the conflict resolution weights, the two device-level requirement semantic features in the conflict feature pair are weighted and averaged to obtain the conflict resolution features.

5. The intelligent furniture scene recommendation method based on a smart central screen according to claim 1, characterized in that, The step of performing semantic inference on the semantic aggregation features of each smart furniture device based on the pre-generated device topology map to obtain the device collaborative semantic features includes: Obtain a pre-generated device topology map, wherein each smart furniture device is a node and the functional linkage relationship or spatial co-occurrence relationship between smart furniture devices is an edge; Based on the device topology graph, the neighboring nodes of each node are determined, and the node features are initialized according to the semantic aggregation features. In the t-th feature propagation, based on the node features after the (t-1)-th feature propagation, the semantic-topological fusion weight between each node and its neighboring nodes is calculated; and based on the semantic-topological fusion weight, the node features of the neighboring nodes are weighted and fused to obtain the corresponding neighbor weighted features; and based on the neighbor weighted features and the node features after the (t-1)-th feature propagation, the node features of the t-th feature propagation are updated to obtain the node features after the t-th feature propagation. When t reaches the preset number of iterations T, the node features of each node are concatenated to obtain the device collaborative semantic features.

6. The intelligent furniture scene recommendation method based on a smart central screen according to claim 5, characterized in that, The step of calculating the semantic-topological fusion weights between each node and its neighboring nodes based on the node features after the (t-1)th feature propagation in the t-th feature propagation includes: For a single node, obtain the node features of the node after the (t-1)th feature propagation, and the node features of the node's neighboring nodes after the (t-1)th feature propagation, and calculate the semantic similarity between the node features of the node and the node features of the neighboring nodes. Obtain the edge weights between the node and its neighboring nodes in the device topology graph, and perform a multiplication operation on the semantic similarity and the edge weights to obtain the corrected weights between the node and its neighboring nodes; The modified weights between the node and its neighboring nodes are normalized to obtain the semantic-topology fusion weights.

7. The intelligent furniture scene recommendation method based on a smart central screen according to claim 1, characterized in that, The step of semantically decomposing the device collaborative semantic features based on the pre-generated base scene vectors to obtain the scene coefficients of each preset base scene includes: Obtain a pre-generated set of base scene vectors, wherein the set of base scene vectors includes base scene vectors corresponding to multiple preset base scenes, and each base scene vector is used to represent the semantic center of a preset base scene; Based on the base scene vector, the device collaborative semantic features are semantically projected to obtain multiple collaborative semantic projection features, wherein the collaborative semantic projection features and the base scene vector are in the same semantic space; Calculate the semantic similarity between the collaborative semantic projection features and each of the base scene vectors; The semantic similarity scores are normalized to obtain the corresponding scene coefficients.

8. The intelligent furniture scene recommendation method based on a smart central screen according to claim 7, characterized in that, The step of semantically projecting the device collaborative semantic features based on the base scene vector to obtain multiple collaborative semantic projection features includes: Each of the base scene vectors is normalized to obtain the corresponding unit projection vector, wherein the unit projection vector has the same semantic direction as the base scene vector and has a magnitude of 1; Based on the unit projection vector, the device collaborative semantic features are vector-projected to obtain the corresponding intermediate projection features; The intermediate projection features are subjected to nonlinear activation processing to obtain the corresponding collaborative semantic projection features.

9. A smart furniture scene recommendation system based on a smart central screen, characterized in that, include: The multimodal semantic fusion module is used to acquire voice data and video data of each user in the target space, and to perform multimodal semantic fusion on the voice data and video data of the same user to obtain the device-level demand semantic features of each user for each smart furniture device. The device-level demand semantic features are used to characterize the explicit control needs and implicit control needs of the same user for the same smart furniture device. The semantic aggregation module is used to perform semantic aggregation on the device-level demand semantic features corresponding to each user for a single smart furniture device, and obtain corresponding semantic aggregation features, wherein the semantic aggregation features are used to characterize the comprehensive control needs of each user for the same smart furniture device. The semantic inference module is used to perform semantic inference on the semantic aggregation features of each smart furniture device based on the pre-generated device topology map to obtain device collaboration semantic features, wherein the device collaboration semantic features are used to characterize the collaborative control requirements between each smart furniture device. The semantic decomposition module is used to perform semantic decomposition on the collaborative semantic features of the device based on the pre-generated base scene vectors to obtain the scene coefficients of each preset base scene; The scene recommendation generation module is used to generate scene recommendation schemes based on the scene coefficients, so as to control the smart central screen and linked devices.

10. A computer device, characterized in that, The method includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the method of any one of claims 1 to 8.

Images