Smart home user behavior prediction method based on semantic analysis

Abstract

The invention discloses a smart home user behavior prediction method based on semantic analysis, and relates to the technical field of natural language processing. The method comprises the following specific steps: data acquisition: respectively acquiring user interaction data, equipment operation historical data and user interaction feedback data through a voice acquisition device, a user input interface, a smart home gateway and a feedback interface, and uniformly storing the data after formatting processing; the speech acoustic features and the text features are extracted on the emotional semantic recognition level, the pre-training model is used for outputting the user emotional state, the defect that in the prior art, emotional perception is lacked is overcome, in the semantic ambiguity resolution aspect, a learning mechanism with user interaction feedback data as an optimization signal is adopted, and the user experience is improved. When the same number of times of correction of the same fuzzy instruction by the user reaches a preset threshold value, the semantic mapping rule is automatically generated and written into the rule base, and the semantic analysis priority can be adjusted according to the family structure change.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, specifically to a method for predicting smart home user behavior based on semantic analysis. Background Technology

[0002] With the rapid development of smart home technology, smart home systems are playing an increasingly important role in people's daily lives. Through interaction with users, smart home systems enable intelligent control of various home devices, improving the convenience and comfort of users' lives. Natural language processing technology, as one of the key technologies for human-computer interaction, has been widely used in the smart home field, allowing users to communicate with the smart home system and issue operating commands in natural language. At the same time, user behavior analysis technology is of great significance for smart home systems to understand user needs and predict user behavior, helping the system to prepare in advance and provide more personalized services.

[0003] However, existing smart home systems suffer from significant technical deficiencies in semantic analysis. On one hand, the parsing of user interaction data is limited to explicit command recognition. When processing voice and text data, they lack the ability to extract emotional features and fail to consider the temporal patterns of user actions. This results in the system's inability to fully and accurately understand user intentions. For example, the same command issued by a user in different emotional states may imply different underlying needs, but current technology cannot capture these differences. Furthermore, users' operating habits vary at different times, making it difficult for current technology to accurately predict based on temporal patterns. On the other hand, existing technologies rely on fixed ontology libraries or preset rules to achieve semantic consistency. This approach is inadequate when facing dynamic changes in the home environment or ambiguous commands. When the family structure changes, such as adding family members or devices, or when users issue ambiguous commands, the system requires manual reconfiguration of semantic mapping. This not only consumes significant human and material resources but also severely limits adaptability and accuracy, failing to meet the needs of smart home systems in practical applications. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method for predicting smart home user behavior based on semantic analysis. This method can collect user interaction data, device operation history data, and user interaction feedback data in real time through a data acquisition module. A multi-dimensional feature analysis module performs emotional semantic recognition and semantic ambiguity resolution on the user interaction data, outputting the user's emotional state and explicit semantic analysis results. A temporal behavior mining module processes the device operation history data, mines the patterns in device operation sequences, and establishes a user temporal behavior database. A correlation graph construction module constructs a four-dimensional correlation graph and dynamically updates it based on user behavior and feedback. A behavior prediction and control module matches and generates device control sequences in the correlation graph based on real-time data and executes them. Simultaneously, it records user feedback data to reverse-optimize semantic mapping rules and the correlation graph. This solves the technical problems of insufficient semantic analysis adaptability and lack of emotional perception and temporal correlation capabilities in existing technologies, providing smart home systems with adaptive semantic analysis and behavior prediction capabilities.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a method for predicting smart home user behavior based on semantic analysis, comprising the following specific steps: Data collection: User interaction data, device operation history data, and user interaction feedback data are collected through voice acquisition devices, user input interfaces, smart home gateways, and feedback interfaces, and then stored uniformly after being formatted. Multi-dimensional feature analysis: Performs two-layer analysis on user interaction data, extracts speech and text features and inputs them into a pre-trained model to complete emotional semantic recognition, and resolves semantic ambiguity through fuzzy command recognition, rule generation and scene adaptation, and finally outputs the user's emotional state and explicit semantic analysis results. Temporal behavior mining: After preprocessing the device operation history data, the data is divided into segments according to the time dimension and temporal behavior patterns are extracted. The data is then classified and stored according to emotional state and time period to build a user temporal behavior database. Association graph construction: Define four types of core nodes and establish multi-dimensional mapping relationships to construct a four-dimensional association graph, and complete the two-dimensional dynamic update of the graph based on new user operation habits and interaction feedback; Behavior prediction and control: Based on real-time user sentiment semantics, semantic parsing results and time information matching association graph, generate and execute device control sequences, and record user feedback to optimize the rule base and graph in reverse.

[0006] Furthermore, in the data acquisition step, user voice data is acquired through a voice acquisition device, text data is acquired through a user input interface to form user interaction data, device operation information including the type of operating device, operation timestamp, operation parameters, and operation sequence is recorded in real time through a smart home gateway to form device operation history data, and user interaction feedback data including the user's correction behavior for ambiguous commands, confirmation operation and rejection operation for device operation are collected through the interactive interface and device operation feedback interface. The three types of data collected are formatted and then stored uniformly.

[0007] Furthermore, in the multi-dimensional feature analysis step, a two-layer analysis process is performed on the user interaction data. At the emotional semantic recognition level, the tone change rate, average speech rate, and volume fluctuation amplitude features of the speech data are extracted, as well as the frequency of use of interjections, punctuation marks, and sentence structure features of the text data. A comprehensive emotional feature vector is generated through an emotional feature fusion formula, and this vector is input into a pre-trained deep learning-based hybrid model. This model includes network units for feature extraction and network units for classification. Its training data consists of speech samples and text samples labeled with emotion tags. The model outputs the user's current emotional state through computation. At the semantic ambiguity resolution level, ambiguous commands in the user input are identified through keyword matching and semantic integrity detection. A threshold formula is generated based on semantic rules to determine the threshold. When a correction behavior is detected to reach the threshold, a semantic mapping rule is automatically generated and written into the rule base. At the same time, changes in family structure are detected in real time, and the semantic analysis priority is adjusted accordingly. In the case of new device access, the functional semantic features of the new device are extracted through transfer learning, and the operation time and environmental association information of the user's previous similar devices are associated to generate exclusive semantic analysis rules. Finally, a clear semantic analysis result is output.

[0008] Furthermore, in the multi-dimensional feature parsing step, a comprehensive sentiment feature vector is generated through a sentiment feature fusion formula, the calculation formula of which is: ,in, It is the fused sentiment feature vector It is a speech acoustic feature vector with dimension . Includes pitch variation rate, average speech rate, and volume fluctuation range. Each acoustic feature component It is a text semantic feature vector with dimension 1. This includes the frequency of use of modal particles, the frequency of use of punctuation marks, and sentence structure features. Each text feature component It is the acoustic feature weighting coefficient.

[0009] Furthermore, in the multi-dimensional feature parsing step, a threshold is determined based on a threshold formula generated according to semantic rules. The formula is as follows: ,in, It is the first A threshold for generating rules for fuzzy instructions is set; when this threshold is reached, mapping rules are automatically generated. It is the first The same number of corrections for a fuzzy instruction. It is the first The feedback confidence of a fuzzy instruction. It is the weighting coefficient for the number of corrections.

[0010] Furthermore, in the time-series behavior mining step, the device operation history data is first preprocessed by deduplication and completion, then the data is divided into segments according to time dimensions such as day, week, and month. A fixed time window is set using a time window division method, and high-frequency operation sequences and operation parameter correlation patterns within each time window are extracted using a sequence pattern extraction method. The confidence of the operation sequence is calculated using a time-series pattern confidence formula, and the time-series behavior patterns that meet the confidence threshold are classified and stored according to emotional state and time period to construct a user time-series behavior database.

[0011] Furthermore, the confidence score of the operation sequence is calculated using the time-series pattern confidence formula, and the formula is as follows: ,in, In the time window and emotional state The confidence level of a certain operation sequence. It is a time window and emotional state The frequency of occurrence of this operation sequence. It is a time window The total frequency of occurrence of all operation sequences within the sequence. It is a time window The periodicity coefficient.

[0012] Furthermore, in the association graph construction step, four types of core nodes are defined: emotional semantic nodes, semantic rule nodes, temporal nodes, and device control nodes. Among them, emotional semantic nodes correspond to the user's emotional state, semantic rule nodes correspond to the mapping relationship between fuzzy instructions and target devices, temporal nodes correspond to time periods and time windows, and device control nodes correspond to device types and operating parameters. Based on the above nodes, multi-dimensional mapping relationships are established, including the mapping between emotional semantic nodes and temporal nodes, the mapping between temporal nodes and device control nodes, the mapping between semantic rule nodes and device control nodes, and the mapping between emotional semantic nodes and device control nodes, thereby constructing a four-dimensional association graph. This graph supports dual-dimensional dynamic updates. On the one hand, the mapping relationship between emotion, temporal sequence, and operation sequence is updated based on new user operating habits or emotion-behavior association patterns. On the other hand, the mapping relationship between semantic rule nodes and semantic-device control is updated based on user interaction feedback data.

[0013] Furthermore, in the behavior prediction and control step, the system acquires the sentiment semantics corresponding to the user's current interaction data, the explicit semantic parsing results after ambiguity resolution, and the current time information in real time. This information is used as retrieval conditions to match the corresponding temporal behavior patterns and semantic rules in the four-dimensional association graph. Based on the matching results, a device control sequence containing the device operation order and specific control parameters is generated. This control sequence is sent to the smart home gateway, which drives the corresponding device to perform control operations. At the same time, the system records the user's feedback data on the control operations in real time, including modifications to control parameters and rejections of operations. Based on the feedback data, the system iteratively optimizes the semantic mapping rule base and the node mapping relationships in the association graph, continuously improving the system's behavior prediction capabilities.

[0014] Compared with existing technologies, this semantic analysis-based method for predicting smart home user behavior has the following advantages: I. This invention extracts speech acoustic features and text features at the level of emotional semantic recognition, and uses a pre-trained model to output the user's emotional state, thus overcoming the deficiency of existing technologies in emotional perception. In terms of semantic ambiguity resolution, it adopts a learning mechanism that uses user interaction feedback data as optimization signals. When the number of times a user makes the same correction to the same ambiguous command reaches a preset threshold, semantic mapping rules are automatically generated and written into the rule base. It can also adjust the semantic parsing priority according to changes in family structure. When a new device is connected, it generates exclusive semantic parsing rules through transfer learning. This solves the problems of existing technologies relying on a fixed ontology library and having limited adaptability and accuracy when facing dynamic changes in family scenarios or ambiguous commands, greatly improving the accuracy of semantic parsing.

[0015] Second, this invention preprocesses and performs time-series analysis on historical device operation data to uncover patterns in device operation sequences under different time periods and emotional states. It establishes a user time-series behavior database, using emotional states, semantic parsing results, and time-series information as core nodes to construct a four-dimensional association graph, storing multiple mapping relationships. Based on the association graph, it generates and executes device control sequences, while simultaneously recording user feedback data to inversely optimize semantic mapping rules and the association graph. This allows for adaptive adjustments based on user behavior habits and scene changes, providing powerful adaptive semantic parsing and behavior prediction capabilities for smart home systems.

[0016] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0018] Figure 1 This is a flowchart illustrating the operation of a semantic analysis-based smart home user behavior prediction method. Figure 2 This is a flowchart of a method for predicting smart home user behavior based on semantic analysis. Detailed Implementation

[0019] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0020] This invention provides a method for predicting user behavior in smart homes based on semantic analysis. A data acquisition module collects user interaction data, device operation history data, and user interaction feedback data in real time. A multi-dimensional feature analysis module performs sentiment semantic recognition and semantic ambiguity resolution on the user interaction data, outputting the user's emotional state and explicit semantic analysis results. A temporal behavior mining module processes the device operation history data, mines patterns in device operation sequences, and establishes a user temporal behavior database. A correlation graph construction module constructs a four-dimensional correlation graph and dynamically updates it based on user behavior and feedback. A behavior prediction and control module matches and generates device control sequences in the correlation graph based on real-time data and executes them. Simultaneously, it records user feedback data to inversely optimize semantic mapping rules and the correlation graph. This method solves the technical problems of insufficient semantic analysis adaptability and lack of emotional perception and temporal correlation capabilities in existing technologies, providing smart home systems with adaptive semantic analysis and behavior prediction capabilities.

[0021] Example 1 The user is a working professional living alone with a regular daily routine. They have a habit of operating smart home devices during fixed times in the morning on weekdays (such as within 30 minutes after waking up). The system needs to rely on historical behavior data and real-time interaction information to achieve automatic predictive control according to the method described in the document, and complete the orderly start-up and parameter adaptation of commonly used devices such as curtains, lights, and coffee machines in the morning.

[0022] like Figure 1As shown, user voice data is acquired through a bedroom voice acquisition device, and text data related to morning light adjustment manually entered by the user in the past is acquired through the mobile APP input interface. The two together constitute user interaction data. The smart home gateway records the user's device operation history data at fixed times during weekday mornings over the past 1-3 months. This data includes the type of device operated, operation timestamp, operation parameters, and operation sequence. The APP feedback interface collects the user's past interaction feedback data, including corrections to ambiguous commands, confirmations of device operations, and rejections that did not occur. These three types of data are formatted and stored in the database.

[0023] The document describes a two-layer parsing process for user interaction data. At the sentiment semantic recognition level, it extracts the pitch change rate, average speech rate, and volume fluctuation amplitude from the speech data, and simultaneously extracts the frequency of modal particles, punctuation usage, and sentence structure features from the text data, based on the sentiment feature fusion formula. (in, The fused sentiment feature vector This is a speech acoustic feature vector that includes the above three speech features. This is a text semantic feature vector that includes the above three text features. A comprehensive emotion feature vector is generated by calculating the acoustic feature weight coefficients. This vector is then input into a pre-trained deep learning-based hybrid model, which includes network units for extracting deep features and network units for emotion classification. The training data consists of speech and text samples labeled with emotion tags. The model outputs that the user's current emotional state is calm. At the semantic ambiguity resolution level, ambiguous commands are identified through keyword matching and semantic integrity detection, and a threshold formula is generated based on semantic rules in the document. (in, Generate a threshold for the rule of the k-th fuzzy instruction. This refers to the number of previous identical modifications performed by this instruction. The confidence level of the feedback for this instruction, such as the user's confirmation rate of the light adjustment result. To determine the threshold for calculating the weighting coefficient of the correction frequency, because and The weighted result reached It automatically generates semantic mapping rules and writes them into the rule base. Since the family structure has not changed, the semantic parsing priority remains unchanged. In addition, there are no new devices connected to the system, so there is no need to extract the semantic features of new device functions through transfer learning. Finally, it outputs explicit semantic parsing results.

[0024] In the time-series behavior mining step, the historical data of equipment operation is first preprocessed, and the data is divided into segments according to the daily and weekly time dimensions (e.g., daily morning data is one segment, and data from the five working days of a week is one weekly segment). A fixed time window segmentation method is adopted to set a fixed time window covering the morning operation period. Combined with the sequence pattern extraction method, the operation sequences and operation parameter association patterns that recur frequently within the time window are selected. Based on the time-series pattern confidence formula (in, Morning time window Emotional state The confidence level of this high-frequency operation sequence under (calm) conditions. For this sequence in the time window Emotional state Frequency of occurrence of the following For time window The total frequency of occurrence of all operation sequences within the sequence. For time window The periodicity coefficient is due to the strong periodicity of morning operations. If the value is close to 1, calculate the confidence level of the operation sequence. When the result meets the system's preset confidence level threshold, store the time sequence behavior pattern in the user's time sequence behavior database according to the emotional state of calm and the time cycle of daily workdays, and improve the morning behavior samples in the database.

[0025] Four core node types are defined: sentiment semantic nodes, semantic rule nodes, temporal nodes, and device control nodes. Based on these nodes, a multi-dimensional mapping relationship is established for the four-dimensional association graph: first, the association between sentiment semantic nodes and temporal nodes; second, the mapping between temporal nodes and device control nodes; third, the correspondence between semantic rule nodes and lighting control nodes; and fourth, the direct association between sentiment semantic nodes and the three types of device control nodes. After the graph is built, since no new user operating habits have been observed recently, the feedback confidence of semantic rule nodes is updated based solely on past interaction feedback data, thus completing the two-dimensional dynamic update of the graph.

[0026] The system acquires real-time emotional semantics, unambiguous semantic parsing results, and current time information corresponding to users' voice interaction data during fixed morning time periods. These three types of information are used as retrieval conditions to match corresponding temporal behavior patterns and semantic rules in a four-dimensional association graph. Based on the matching results, a device control sequence is generated, clarifying the operation order and control logic of each device. This control sequence is sent to the smart home gateway, which drives the devices to perform operations in sequence. During the operation, if the user does not modify parameters or reject the operation, but only provides feedback "operation normal" through the APP, this confirmation feedback data is recorded in real-time. Based on this, iterative optimization is performed: first, the feedback confidence of the semantic mapping rule base is updated; second, the mapping strength between emotional semantic nodes and device control nodes in the association graph is strengthened, further improving the accuracy of behavior prediction in subsequent morning scenarios.

[0027] Example 2 The user's family consists of two elderly people who habitually watch TV in the living room at a fixed time every evening. The elderly people's instructions are often vague, and they are highly sensitive to temperature and volume. It is necessary to optimize and predict based on the elderly people's behavior and feedback to achieve precise control of the TV, air conditioner, and lights.

[0028] Voice data from the elderly is collected through a voice acquisition device in the living room, such as "Turn on the TV, it's a bit cold." Text data related to adjusting the TV volume, which the elderly manually entered, is also collected through the touch input interface of the living room control panel, forming user interaction data. The smart home gateway records the elderly's operation information during their regular evening leisure time over a period of time, including turning on the TV, adjusting the volume, adjusting the air conditioner, adjusting the lights, and the corresponding operation sequence, forming device operation history data. Feedback from the user is collected through the control panel feedback interface, including corrections to ambiguous commands and confirmations for each operation, forming user interaction feedback data. These three types of data are then formatted and stored uniformly.

[0029] like Figure 2As shown, a two-layer parsing process is performed on user interaction data. At the emotional semantic recognition level, the tone change rate, average speech rate, and volume fluctuation amplitude of the elderly's speech data are extracted, as well as the frequency of use of interjections, punctuation marks, and sentence structure features of the text data. A comprehensive emotional feature vector is generated through an emotional feature fusion formula. This vector is input into a pre-trained deep learning-based hybrid model, which outputs that the user's current emotional state is "comfortable". At the semantic ambiguity resolution level, keyword matching and semantic integrity detection identify "a little cold" as a vague command. A threshold formula is generated based on semantic rules to determine the threshold. When the number of identical corrections and feedback confidence of the command reach the threshold, the corresponding semantic mapping rule is automatically generated and written into the rule base. The family structure consists of two elderly people, and the semantic parsing priority is tilted towards air conditioner temperature adjustment and TV volume adjustment. No new devices are connected, and no transfer learning is required. The final output is a clear semantic parsing result: turn on the TV, adjust the air conditioner, and adjust the lights to soft mode.

[0030] First, the historical data of device operation is preprocessed by deduplication and completion. The data is divided into segments according to the daily and weekly time dimensions. A fixed time window is set, and the high-frequency operation sequences and operation parameter correlation patterns within the time window are extracted by combining the sequence pattern extraction method. The confidence of the operation sequence is calculated by the time sequence pattern confidence formula. If the result meets the confidence threshold, the time sequence behavior pattern is classified and stored according to the emotional state of comfort and the time period of daily evening, so as to improve the user time sequence behavior database.

[0031] Four core node categories are defined: emotional semantic nodes (comfort), semantic rule nodes (a slight chill - adjust air conditioning, sound-related commands - adjust TV volume), temporal nodes (fixed leisure time window each evening), and device control nodes (TV (turn on, adjust volume), air conditioning (adjust), lighting (adjust to soft mode)). Based on these nodes, multi-dimensional mapping relationships are established, including mappings between comfort and temporal nodes, temporal nodes and device control nodes, two semantic rule nodes and their corresponding device control nodes, and comfort and the three types of device control nodes. A four-dimensional association graph is constructed. Based on the elderly's recent corrective lighting habits, the emotional-temporal-operation sequence mapping relationship is updated; based on recent confirmation feedback data, the feedback confidence of semantic rule nodes is updated, completing a two-dimensional dynamic update of the graph.

[0032] The system acquires real-time data on the emotional and semantic comfort of elderly users' voice interactions during their fixed evening leisure time, along with the clarification of semantics after disambiguation and current time information. This data is then used as search criteria to match the data in a four-dimensional association graph, identifying corresponding temporal behavior patterns and semantic rules. Based on the matching results, a device control sequence is generated and sent to the smart home gateway. The gateway then drives the corresponding devices to perform operations sequentially. The elderly user does not need to make corrections to the operations; they only need to confirm the operation through the control panel. The system records this feedback data in real time and iteratively optimizes the semantic mapping rule base and the node mapping relationships in the association graph based on this feedback, continuously adapting to the elderly user's habits.

[0033] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for predicting smart home user behavior based on semantic analysis, characterized in that, The method includes the following specific steps: Data collection: User interaction data, device operation history data, and user interaction feedback data are collected through voice acquisition devices, user input interfaces, smart home gateways, and feedback interfaces, and then stored uniformly after being formatted. Multi-dimensional feature analysis: Performs two-layer analysis on user interaction data, extracts speech and text features and inputs them into a pre-trained model to complete emotional semantic recognition, and resolves semantic ambiguity through fuzzy command recognition, rule generation and scene adaptation, and finally outputs the user's emotional state and explicit semantic analysis results. Temporal behavior mining: After preprocessing the device operation history data, the data is divided into segments according to the time dimension and temporal behavior patterns are extracted. The data is then classified and stored according to emotional state and time period to build a user temporal behavior database. Association graph construction: Define four types of core nodes and establish multi-dimensional mapping relationships to construct a four-dimensional association graph. Based on new user operating habits and interaction feedback, complete the two-dimensional dynamic update of the graph. Behavior prediction and control: Based on real-time user sentiment semantics, semantic parsing results and time information matching association graph, generate and execute device control sequences, and record user feedback to optimize the rule base and graph in reverse.

2. The method for predicting smart home user behavior based on semantic analysis according to claim 1, characterized in that, In the data acquisition step, user voice data is acquired through a voice acquisition device, text data is acquired through a user input interface to form user interaction data, and device operation information, including the type of operating device, operation timestamp, operation parameters, and operation sequence, is recorded in real time through a smart home gateway to form device operation history data. User interaction feedback data, including the user's correction behavior for ambiguous commands, confirmation operation and rejection operation of device operation, is collected through the interactive interface and device operation feedback interface. The three types of data are formatted and then stored uniformly.

3. The method for predicting smart home user behavior based on semantic analysis according to claim 1, characterized in that, In the multi-dimensional feature analysis step, a two-layer analysis process is performed on the user interaction data. At the emotional semantic recognition level, the pitch change rate, average speech rate, and volume fluctuation amplitude features of the speech data are extracted, as well as the frequency of use of interjections, punctuation marks, and sentence structure features of the text data. A comprehensive emotional feature vector is generated through an emotional feature fusion formula, and this vector is input into a pre-trained deep learning-based hybrid model. This model includes network units for feature extraction and network units for classification. Its training data consists of speech samples and text samples labeled with emotion tags. The model outputs the user's current emotional state through computation. At the semantic ambiguity resolution level, ambiguous commands in the user input are identified through keyword matching and semantic integrity detection. A threshold formula is generated based on semantic rules to determine the threshold. When a correction behavior is detected to reach the threshold, a semantic mapping rule is automatically generated and written into the rule base. At the same time, changes in family structure are detected in real time, and the semantic analysis priority is adjusted accordingly. In the case of new device access, the functional semantic features of the new device are extracted through transfer learning, and the operation time and environmental association information of the user's previous similar devices are associated to generate exclusive semantic analysis rules. Finally, a clear semantic analysis result is output.

4. The method for predicting smart home user behavior based on semantic analysis according to claim 3, characterized in that, In the multi-dimensional feature parsing step, a comprehensive sentiment feature vector is generated through the sentiment feature fusion formula, the calculation formula of which is as follows: ,in, It is the fused sentiment feature vector It is a speech acoustic feature vector with dimension . Includes pitch variation rate, average speech rate, and volume fluctuation range. Each acoustic feature component It is a text semantic feature vector with dimension 1. This includes the frequency of use of modal particles, the frequency of use of punctuation marks, and sentence structure features. Each text feature component It is the acoustic feature weighting coefficient.

5. The method for predicting smart home user behavior based on semantic analysis according to claim 3, characterized in that, In the multi-dimensional feature parsing step, the threshold is determined based on a threshold formula generated according to semantic rules. The formula is as follows: ,in, It is the first A threshold for generating rules for fuzzy instructions is set; when this threshold is reached, mapping rules are automatically generated. It is the first The same number of corrections for a fuzzy instruction. It is the first The feedback confidence of a fuzzy instruction. It is the weighting coefficient for the number of corrections.

6. The method for predicting smart home user behavior based on semantic analysis according to claim 1, characterized in that, In the aforementioned temporal behavior mining step, the device operation history data is first preprocessed by deduplication and completion. Then, the data is divided into segments according to time dimensions such as day, week, and month. Fixed time windows are set using a time window division method. High-frequency operation sequences and operation parameter correlation patterns within each time window are extracted using a sequence pattern extraction method. The confidence of the operation sequence is calculated using a temporal pattern confidence formula. Temporal behavior patterns that meet the confidence threshold are classified and stored according to emotional state and time period to construct a user temporal behavior database.

7. The method for predicting smart home user behavior based on semantic analysis according to claim 6, characterized in that, The confidence score of the operation sequence is calculated using the time-series pattern confidence formula, which is as follows: ,in, In the time window and emotional state The confidence level of a certain operation sequence. It is a time window and emotional state The frequency of occurrence of this operation sequence. It is a time window The total frequency of occurrence of all operation sequences within the sequence. It is a time window The periodicity coefficient.

8. The method for predicting smart home user behavior based on semantic analysis according to claim 1, characterized in that, In the association graph construction step, four types of core nodes are defined: emotional semantic nodes, semantic rule nodes, temporal nodes, and device control nodes. Emotional semantic nodes correspond to the user's emotional state, semantic rule nodes correspond to the mapping relationship between fuzzy instructions and target devices, temporal nodes correspond to time periods and time windows, and device control nodes correspond to device types and operating parameters. Based on the above nodes, multi-dimensional mapping relationships are established, including the mapping between emotional semantic nodes and temporal nodes, the mapping between temporal nodes and device control nodes, the mapping between semantic rule nodes and device control nodes, and the mapping between emotional semantic nodes and device control nodes, thereby constructing a four-dimensional association graph. This graph supports dual-dimensional dynamic updates. On the one hand, the mapping relationship between emotion, temporal sequence, and operation sequence is updated based on new user operating habits or emotion-behavior association patterns. On the other hand, the mapping relationship between semantic rule nodes and semantic-device control is updated based on user interaction feedback data.

9. The method for predicting smart home user behavior based on semantic analysis according to claim 1, characterized in that, In the behavior prediction and control step, the system acquires the sentiment semantics corresponding to the user's current interaction data, the explicit semantic parsing results after ambiguity resolution, and the current time information in real time. This information is used as retrieval conditions to match the corresponding temporal behavior patterns and semantic rules in the four-dimensional association graph. Based on the matching results, a device control sequence containing the device operation order and specific control parameters is generated. This control sequence is sent to the smart home gateway, which drives the corresponding device to perform control operations. At the same time, the system records the user's feedback data on the control operations in real time, including modifications to control parameters and rejections of operations. Based on the feedback data, the system iteratively optimizes the semantic mapping rule base and the node mapping relationships in the association graph, continuously improving the system's behavior prediction capabilities.

Images