An artificial intelligence-based health record personalized recommendation method and system

Abstract

This invention discloses a personalized health record recommendation method and system based on artificial intelligence, belonging to the field of intelligent recommendation technology. The method includes constructing an energy flow graph and a heterogeneous graph, calculating the channel's conserved residuals and health attention, combining the conserved residuals and normalized health attention to calculate the channel driving strength, screening significant keywords, generating a set of significant keywords, screening matching recommendation items, and generating a recommended health record. This invention combines short-time Fourier transform with synchronous squeezing and redistribution technology to improve the analytical accuracy of multi-channel signal energy flow, combines energy flow graph construction with heterogeneous graph correlation analysis to enhance the targeting of personalized recommendations, and enhances the sensitivity of health record recommendation results to individual state changes by combining energy envelope cross-correlation with health keyword matching.

Description

Technical Field

[0001] This invention relates to the field of intelligent recommendation technology, and in particular to a personalized recommendation method and system for health records based on artificial intelligence. Background Technology

[0002] With the widespread adoption of wearable devices and mobile healthcare, coupled with advancements in artificial intelligence, we are now able to manage each person's health status in a more refined and personalized way. Current health management systems can monitor bodily signals such as heart rate, blood pressure, and respiration in real time. They can also combine voice, behavioral, and even environmental information to help us dynamically understand a person's health changes. When analyzing these physiological signals, researchers often use time-frequency analysis methods to observe the characteristics of systems like the cardiovascular and respiratory systems as they change over time.

[0003] Existing personalized health record recommendation methods still have shortcomings. Traditional methods mainly rely on a single type of data or only look at the questionnaires filled out by users, but ignore the dynamic interaction between different physiological signals. It is difficult to accurately characterize the energy flow characteristics of multi-channel signals. Most analysis methods are still limited to observing the overall energy distribution of signals or extracting fixed statistical features, but fail to capture how energy is transferred between different frequency bands, making it difficult to reveal the underlying physiological driving mechanisms. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a personalized health record recommendation method and system based on artificial intelligence, which solves the problem that traditional methods mainly rely on a single type of data or only look at the questionnaires filled out by users, but ignore the dynamic interaction between different physiological signals, making it difficult to accurately characterize the energy flow characteristics of multi-channel signals. Most analysis methods are still limited to observing the overall energy distribution of signals or extracting fixed statistical features, but fail to capture how energy is transferred between different frequency bands, making it difficult to reveal the potential physiological driving mechanisms.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a personalized health record recommendation method based on artificial intelligence, comprising:

[0008] Collect and preprocess multi-source data, perform short-time Fourier transform on physiological signals, calculate complex spectrum and short-time energy spectrum, redistribute short-time energy spectrum to frequency index corresponding to instantaneous frequency, map frequency index to frequency band, and generate aggregated energy.

[0009] Phase extraction is performed on the complex spectrum to generate joint features, the square root of the aggregated energy is calculated, an energy envelope is generated, the maximum associated time delay of the energy envelope is identified, and the total energy transfer is calculated.

[0010] Construct energy flow graphs and heterogeneous graphs, calculate the channel conservation residuals and health attention, combine the conservation residuals and normalized health attention to calculate the channel driving strength, screen significant keywords, generate a set of significant keywords, screen matching recommendations, and generate recommended health profiles.

[0011] As a preferred embodiment of the personalized health record recommendation method based on artificial intelligence described in this invention, the step of performing a short-time Fourier transform on the physiological signal, calculating the complex spectrum and the short-time energy spectrum, redistributing the short-time energy spectrum to the frequency index corresponding to the instantaneous frequency, mapping the frequency index to the frequency band, and generating aggregated energy includes:

[0012] Speech recognition algorithms are used to convert speech signals into text. The text is then divided into time frames within a time grid to generate frame-level text fragments. Based on these frame-level text fragments, regular expression matching methods are used to extract keywords from the ICD-10 International Classification of Diseases to generate a keyword set.

[0013] The physiological signal is divided into time grids by setting the time grid duration using a sliding window. Based on the time grids, a short-time Fourier transform is performed on the physiological signal to calculate the complex spectrum.

[0014] Based on the complex spectrum, the short-time energy spectrum is calculated, and the instantaneous frequency of the complex spectrum is calculated through the phase derivative. Then, the short-time energy spectrum is redistributed to the frequency index corresponding to the instantaneous frequency using the synchronous squeezing transformation.

[0015] The frequency band set is defined using an empirical method, frequency indices are mapped to frequency bands, and each frequency band is aggregated to generate aggregated energy.

[0016] As a preferred embodiment of the personalized health record recommendation method based on artificial intelligence described in this invention, the step of performing phase extraction on the complex spectrum to generate joint features includes:

[0017] Phase extraction is performed on the complex spectrum to generate short-time phase spectra for each channel at each frequency and time frame. The phase difference of the short-time phase spectra is calculated, and a phase wrapping function is used to limit the phase difference to a specific value. ;

[0018] Calculate the first-order forward difference of the short-time energy spectrum to generate the energy increment, calculate the forward energy increment rate, and combine the short-time phase spectrum and the forward energy increment rate to generate a joint feature.

[0019] As a preferred embodiment of the personalized health record recommendation method based on artificial intelligence described in this invention, the steps of calculating the square root of the aggregated energy, generating an energy envelope, identifying the maximum correlation delay of the energy envelope, and calculating the total energy transfer amount include:

[0020] Calculate the square root of the aggregated energy, generate the energy envelope for each channel, calculate the cross-correlation function of the energy envelope within the time grid, identify the maximum associated time delay of the energy envelope, and calculate the amount of energy transferred.

[0021] Aggregate to the time grid level and calculate the total energy transfer within the time grid.

[0022] As a preferred embodiment of the personalized health record recommendation method based on artificial intelligence described in this invention, the following steps are included: constructing an energy flow graph and a heterogeneous graph, calculating the channel's conserved residuals and health attention, and combining the conserved residuals and normalized health attention to calculate the channel driving strength.

[0023] The aggregated energy is summed at the frame level within the time grid to generate the grid-level total energy;

[0024] Based on the total energy transfer, the net inflow and net outflow are calculated. Physiological signal channels are defined as channel nodes, and the energy flow between channels is used as channel edges to generate an energy flow graph. Combining the grid-level total energy, net inflow, and net outflow, the channel conservation residuals are calculated.

[0025] An explicit mapping table is constructed based on expert knowledge to define the correspondence between each health keyword and physiological signal channels;

[0026] Extract keywords from the keyword set as keyword nodes. Based on the explicit mapping table, filter keywords in the channel mapping table and define them as keyword edges to generate a heterogeneous graph.

[0027] Count the number of times keywords appear in the grid and calculate the edge strength of the heterogeneous graph;

[0028] Based on channel nodes, aggregate channels and related keyword nodes to generate the total association strength of channels, and then perform normalization processing;

[0029] By combining the conserved residuals and the normalized total correlation strength, the health concern of the channel is calculated and normalized. Then, by combining the conserved residuals and the normalized health concern, the channel driving strength is calculated.

[0030] As a preferred embodiment of the personalized health record recommendation method based on artificial intelligence described in this invention, the step of filtering significant keywords, generating a set of significant keywords, filtering matching recommendation items, and generating recommended health records includes:

[0031] Select the channel and frequency band corresponding to the maximum total driving strength to obtain the key driving channel and frequency band. Extract the edge strength related to the key driving channel from the heterogeneous graph, screen significant keywords, and generate a set of significant keywords.

[0032] A health record recommendation database is collected from the health management platform. Based on key driving channels and significant keywords, matching recommendations are screened, and a relevance score is calculated for each recommendation in combination with energy transfer amount.

[0033] Sort the relevance scores in descending order, filter the top O recommended items, and generate recommended health profiles, where O is the number of recommended items.

[0034] As a preferred embodiment of the AI-based personalized health record recommendation method of the present invention, the step of collecting and preprocessing multi-source data includes:

[0035] The system collects multi-source data from users through smart bracelets and smartphones, and performs noise reduction and normalization processing.

[0036] The multi-source data includes physiological signals and speech signals;

[0037] The physiological signals include heart rate, blood pressure, and respiratory rate data.

[0038] Secondly, this invention provides a personalized health record recommendation system based on artificial intelligence, comprising:

[0039] The data collection and processing module is used to collect multi-source data from users and perform noise reduction and normalization processing.

[0040] The energy extraction module is used to extract the complex spectrum and short-time energy spectrum of physiological signals, redistribute the short-time energy spectrum to the frequency index corresponding to the instantaneous frequency, map the frequency index to the frequency band, and generate aggregated energy.

[0041] The feature transfer module is used to combine the short-time phase spectrum and the energy increment rate to extract joint features between channels, identify the maximum associated time delay of the energy envelope, and calculate the total energy transfer amount.

[0042] The channel module is used to construct energy flow graphs and heterogeneous graphs, and to calculate the channel conservation residuals and health concerns.

[0043] The personalized recommendation module is used to filter significant keywords, match recommended items from the health record database, sort them by relevance score, and generate recommended health records.

[0044] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the personalized health record recommendation method based on artificial intelligence as described in the first aspect of the present invention.

[0045] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the personalized health record recommendation method based on artificial intelligence as described in the first aspect of the present invention.

[0046] The beneficial effects of this invention are as follows: This invention achieves fine decomposition and energy redistribution of physiological signals in the time-frequency dimension by combining short-time Fourier transform with synchronous squeezing transform, accurately obtains the energy flow characteristics between different frequency bands, and characterizes the energy transfer direction and associated time delay between channels by combining phase difference analysis with energy envelope cross-correlation calculation, revealing the dynamic coupling relationship between multiple physiological channels. The physiological energy flow map combined with the health semantic heterogeneous map establishes an explicit mapping between keywords and physiological channels, quantifying health attention and driving intensity. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart of the AI-based personalized health record recommendation method in Example 1.

[0049] Figure 2 This is a schematic diagram of the AI-based personalized health record recommendation system in Example 1. Detailed Implementation

[0050] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0051] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0052] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0053] Example 1, referring to Figure 1 and Figure 2 This is the first embodiment of the present invention, which provides a personalized health record recommendation method based on artificial intelligence, including the following steps:

[0054] S1. Collect multi-source data and preprocess it. Perform short-time Fourier transform on the physiological signal, calculate the complex spectrum and short-time energy spectrum, redistribute the short-time energy spectrum to the frequency index corresponding to the instantaneous frequency, map the frequency index to the frequency band, and generate aggregated energy.

[0055] Specifically, this involves collecting and preprocessing multi-source data, including:

[0056] The system collects multi-source data from users through smart bracelets and smartphones, and performs noise reduction and normalization processing.

[0057] The multi-source data includes physiological signals and speech signals;

[0058] The physiological signals include heart rate, blood pressure, and respiratory rate data.

[0059] The combination of noise reduction and normalization improves signal comparability and computational stability, ensuring that data collected by different devices can be directly fused and analyzed on the same scale. The combination of voice data and physiological signals breaks through the limitation of the separation of "physiological-semantic" information in traditional health management, enabling individual subjective descriptions to complement objective measurements.

[0060] Furthermore, a short-time Fourier transform is performed on the physiological signal to calculate the complex spectrum and short-time energy spectrum. The short-time energy spectrum is then redistributed to the frequency index corresponding to the instantaneous frequency, and the frequency index is mapped to the frequency band to generate aggregated energy, including:

[0061] Speech recognition algorithms are used to convert speech signals into text. The text is then divided into time frames within a time grid to generate frame-level text fragments. Based on these frame-level text fragments, regular expression matching methods are used to extract keywords from the ICD-10 International Classification of Diseases to generate a keyword set.

[0062] Using a sliding window to set the time grid duration, physiological signals are divided into time grids, as shown in the formula:

[0063] ,

[0064] in For time grid, The sampling rate of the physiological signal. For time grid duration;

[0065] Based on the time grid, a short-time Fourier transform is performed on the physiological signal to calculate the complex spectrum, as shown in the formula:

[0066] ,

[0067] in Let i be the complex spectrum of channel i at frequency k and frame m. The normalized physiological signal is represented by L, where L is the window length, Q is the frame shift, n is the sampling point index, and j is the imaginary unit. For the Hanning window function;

[0068] Based on the complex spectrum, the short-time energy spectrum is calculated using the following formula:

[0069] ,

[0070] in Let i be the short-time energy spectrum of channel i at frequency k and frame m;

[0071] The instantaneous frequency of the complex spectrum can be calculated using the phase derivative, as follows:

[0072] ,

[0073] in Let i be the instantaneous frequency of channel i at frequency k and frame m. For the phase of the complex spectrum, Let be the partial derivative with respect to time t;

[0074] Using synchronous squeezing transform, the short-time energy spectrum is redistributed to the frequency index corresponding to the instantaneous frequency, as shown in the formula:

[0075] ,

[0076] in For channel i at frequency The time-frequency energy density of frame m The frequency points after redistribution. To achieve frequency resolution, divide the sampling rate by the window length. This is the frequency index after redistribution;

[0077] The frequency band set is defined using an empirical method, including low frequency band, mid frequency band and high frequency band, which correspond to blood pressure, respiratory rate and heart rate respectively;

[0078] The frequency index is mapped to the frequency band, and each frequency band is aggregated to generate aggregated energy, as shown in the formula:

[0079] ,

[0080] in For channel i in the frequency band The aggregated energy of time frame m Let r be the frequency band, where r = 1, 2, 3.

[0081] Short-time Fourier transform combined with synchronous squeezing transform improves time-frequency resolution and frequency relocation accuracy. Phase derivative extraction of instantaneous frequency enhances the time sensitivity of the signal, enabling the system to detect weak physiological fluctuations within a short period. Frequency redistribution combined with aggregated energy improves the physical interpretability of energy features, allowing the energy distribution of different physiological indicators to directly correspond to physiological mechanisms. Frequency band definition and mapping transform continuous frequency space into discrete frequency bands with physiological significance. Speech recognition combined with ICD-10 regularized matching realizes the conversion of unstructured speech to structured medical semantics. The time grid synchronization mechanism ensures that semantic information and physiological signals are analyzed on the same time scale. Frame-level text segmentation improves semantic resolution. Aggregated energy mapping can reduce data dimensionality while maintaining local energy features, improving computational efficiency and preserving physiological meaning. Frequency band aggregation combined with time series analysis enables the system to continuously track specific frequency domains (such as the high-frequency region of heart rate variability), thereby identifying chronic health trends or abnormal events.

[0082] S2. Perform phase extraction on the complex spectrum, generate joint features, calculate the square root of the aggregated energy, generate the energy envelope, identify the maximum correlation delay of the energy envelope, and calculate the total energy transfer.

[0083] Specifically, phase extraction is performed on the complex spectrum to generate joint features, including:

[0084] Phase extraction is performed on the complex spectrum to generate short-time phase spectra for each channel at each frequency and time frame, using the following formula:

[0085] ,

[0086] in For channel i, the short-time phase at frequency k and frame m, the range is... , and These are the imaginary and real parts of the complex spectrum, respectively.

[0087] To calculate the phase difference of the short-time phase spectrum, a phase wrapping function is used to constrain the phase difference to a certain value. The formula is:

[0088] ,

[0089] in Let be the phase difference between channel j and channel i. Let be the short-time phase of channel j at frequency k and frame m;

[0090] Calculate the first-order forward difference of the short-time energy spectrum to generate the energy increment, and calculate the forward energy increment rate using the following formula:

[0091] ,

[0092] in Let be the positive energy increment rate of channel i at frequency k and frame m. Let i be the energy increment of channel i at frequency k and frame m. For time resolution;

[0093] By combining the short-time phase spectrum and the positive energy increment rate, a joint feature is generated, as shown in the formula:

[0094] ,

[0095] in Let J be the phase difference between channel j and channel i at frequency k and frame m.

[0096] Aggregation to the frequency band level, the formula is:

[0097] ,

[0098] in For frequency band The joint features within.

[0099] Phase extraction combined with a phase wrapping function effectively eliminates abrupt changes caused by phase discontinuity, ensuring continuity in cross-frame calculations. Phase difference reflects the synchronicity and time lead relationship between different physiological signals. The positive energy increment rate accurately describes the energy growth rate of the signal over time, avoiding the shortcomings of traditional average energy indicators that cannot reflect instantaneous changes. Only positive energy changes (i.e., the growth part) are retained, which can be used to identify the activation phase of the signal and avoid the influence of energy decay caused by noise on the analysis. By combining energy difference with time resolution, quantitative fusion of energy change and time dynamics is achieved, improving time domain sensitivity. The combination of phase synchronization and energy growth enables directional modeling of energy-driven relationships in the time and frequency domains, which can identify the dominant role between multi-channel signals, avoiding the use of any weighting coefficients or fuzzy parameters. Effective coupling is naturally filtered through logical constraints (phase difference intervals), improving model stability.

[0100] Furthermore, the square root of the aggregated energy is calculated to generate an energy envelope, the maximum associated time delay of the energy envelope is identified, and the total energy transfer is calculated, including:

[0101] Calculate the square root of the pooled energy to generate the energy envelope for each channel, and then calculate the cross-correlation function of the energy envelope within the time grid. The formula is as follows:

[0102] ,

[0103] in For channels i and j, in the frequency band Frame m, delay The cross-correlation value, and These are channels i and j, respectively, in the frequency band. ,frame The energy envelope, For time delay, For frame indexes within the time grid;

[0104] Based on the cross-correlation function, the maximum correlation delay of the energy envelope is identified, and the energy transfer amount is calculated using the following formula:

[0105] ,

[0106] in For channels i to j, in the frequency band The amount of energy transferred in frame m. For channels i to j, in the frequency band The maximum associated delay of frame m, if ,but ;

[0107] Aggregating to the time grid level, the total energy transfer within the time grid is calculated using the following formula:

[0108] ,

[0109] in For channels i to j, in the frequency band Time grid The total amount of energy transferred.

[0110] Energy envelope extraction enhances temporal smoothness, making signal trend characteristics more apparent. Cross-correlation analysis combined with time delay identification can quantitatively calculate the response lag between different channels, thereby inferring the signal propagation sequence within the system. By comparing cross-correlation peaks across frequency bands, the coupling hierarchy between different physiological frequency bands can be revealed. The calculation of energy transfer combined with time delay correction avoids the problem of ignoring time offset in simple covariance analysis, making energy flow estimation more consistent with actual physiological propagation characteristics. Through time grid aggregation, short-term fluctuations can be smoothed, obtaining grid-level stability indicators.

[0111] S3. Construct energy flow graphs and heterogeneous graphs, calculate the channel conservation residuals and health attention, combine the conservation residuals and normalized health attention to calculate the channel driving strength, screen significant keywords, generate a set of significant keywords, screen matching recommendations, and generate recommended health profiles.

[0112] Specifically, energy flow graphs and heterogeneity graphs are constructed, the channel conservation residuals and health concerns are calculated, and the channel driving strength is calculated by combining the conservation residuals and normalized health concerns, including:

[0113] The aggregated energy is summed at the frame level within the time grid to generate the grid-level total energy, as shown in the formula:

[0114] ,

[0115] in For channel i, in the frequency band Time grid Total energy;

[0116] Based on the total energy transfer, the net inflow and net outflow are calculated using the following formulas:

[0117] ,

[0118] ,

[0119] in and For channel i, in frequency band Time grid Net inflow and net outflow, For channel j to channel i, in the frequency band Time grid The total amount of energy transferred;

[0120] Physiological signal channels are defined as channel nodes, and the energy flow between channels is used as channel edges to generate an energy flow graph. Combining the total energy at the grid level, net inflow, and net outflow, the conserved residuals of the channels are calculated using the following formula:

[0121] ,

[0122] in For channel i, in the frequency band Time grid The conserved residual;

[0123] An explicit mapping table is constructed based on expert knowledge, defining the correspondence between each health keyword and physiological signal channel. The formula is as follows:

[0124] ,

[0125] in This is a mapping table for channel i, in the format of keyword-channel pairs. For channel i, the set of keywords is... w is the keyword;

[0126] Keywords are extracted from the keyword set as keyword nodes. Based on the explicit mapping table, keywords are filtered in the channel mapping table and defined as keyword edges to generate a heterogeneous graph. The formula is as follows:

[0127] ,

[0128] in The correlation between keyword w and channel i is represented by 1, indicating correlation and 0, indicating no correlation.

[0129] The number of times a keyword appears in the grid is counted, and the edge strength of the heterogeneous graph is calculated using the following formula:

[0130] ,

[0131] in For channel i and keyword w, in the time grid edge strength, The number of times the keyword 'w' appears in frame 'm' is calculated using a regular expression. Let m be the set of keywords;

[0132] Based on channel nodes, the channels and related keyword nodes are aggregated to generate the total association strength of the channels, which is then normalized using the following formula:

[0133] ,

[0134] in For channel i in the time grid The overall correlation strength;

[0135] Combining the conserved residuals and the normalized total correlation strength, the health concern score of the channel is calculated and normalized. The formula is as follows:

[0136] ,

[0137] in For channel i, in the frequency band Time grid Health awareness This represents the normalized total correlation strength.

[0138] Combining the conserved residuals and normalized health attention, the channel driving strength is calculated using the following formula:

[0139] ,

[0140] in For channel i, in the frequency band Time grid driving strength, To normalize health awareness.

[0141] The combination of total energy and net energy flow comprehensively reflects the energy source and consumption of a channel, avoiding the bias caused by using power or flow alone. The conservation residual reflects the energy conservation deviation of the channel. A positive value indicates that the system input exceeds the output, representing physiological activity or enhanced drive. A negative value indicates excessive energy consumption, possibly indicating physiological fatigue or system overload. The explicit mapping table combined with the regular matching mechanism ensures the interpretability and medical consistency of the correspondence between keywords and channels, avoiding the semantic ambiguity problem of the black box in traditional models. The edge strength calculation reflects the activity level of keywords in the time dimension, indirectly reflecting the changing trend of users' recent health concerns. The combination of conservation residual and semantic strength realizes bidirectional coupled modeling from energy dynamics to health semantics.

[0142] Furthermore, significant keywords are filtered to generate a set of significant keywords, matching recommendations are filtered, and recommended health profiles are generated, including:

[0143] Selecting the channel and frequency band corresponding to the maximum total driving intensity yields the key driving channel and frequency band, as shown in the formula:

[0144] ,

[0145] in and These are the key drive channels and frequency bands, respectively.

[0146] Extract edge strengths related to key driving channels from the heterogeneous graph, filter significant keywords, and generate a set of significant keywords using the following formula:

[0147] ,

[0148] in For channel In the time grid prominent keywords, For channel Edge strength with keyword w The filtering threshold is set using the maximum value ratio method, filtering by multiplying the maximum edge strength by 0.1. A set of prominent keywords;

[0149] We collect a health record recommendation database from health management platforms (such as Fitbit), and filter matching recommendations based on key driving channels and prominent keywords, using the following formula:

[0150] ,

[0151] in For channel In the time grid The recommended items, P is the health record recommendation item library, For the recommended item p and the channel The mapping relationship of keyword w;

[0152] Based on the amount of energy transferred, a relevance score is calculated for each recommendation using the following formula:

[0153] ,

[0154] in For the recommended item p in the time grid The calculation of the relevance score, For channel j to channel In frequency band Time grid Total energy transfer For channel With keyword w, in time grid Edge strength;

[0155] Sort the relevance scores in descending order, filter the top O recommended items, and generate recommended health profiles, where O is the number of recommended items. The number of recommended items is set using a fixed threshold method.

[0156] The driving channel combines prominent keywords to achieve targeted screening of health recommendations, ensuring that the recommendation results are strongly correlated with the user's current physiological state. The joint scoring mechanism of energy transfer and semantic intensity can quantify the matching degree between the recommendation items and the energy dynamics of the physiological system, improving the scientificity and real-time performance of the recommendation results. The maximum value ratio method sets the threshold to avoid human experience intervention.

[0157] This embodiment also provides a personalized health record recommendation system based on artificial intelligence, including:

[0158] The data collection and processing module is used to collect multi-source data from users and perform noise reduction and normalization processing.

[0159] The energy extraction module is used to extract the complex spectrum and short-time energy spectrum of physiological signals, redistribute the short-time energy spectrum to the frequency index corresponding to the instantaneous frequency, map the frequency index to the frequency band, and generate aggregated energy.

[0160] The feature transfer module is used to combine the short-time phase spectrum and the energy increment rate to extract joint features between channels, identify the maximum associated time delay of the energy envelope, and calculate the total energy transfer amount.

[0161] The channel module is used to construct energy flow graphs and heterogeneous graphs, and to calculate the channel conservation residuals and health concerns.

[0162] The personalized recommendation module is used to filter significant keywords, match recommended items from the health record database, sort them by relevance score, and generate recommended health records.

[0163] This embodiment also provides a computer device applicable to the personalized health record recommendation method based on artificial intelligence, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the personalized health record recommendation method based on artificial intelligence as proposed in the above embodiment.

[0164] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0165] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the personalized recommendation method for health records based on artificial intelligence as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0166] In summary, this invention achieves fine decomposition and energy redistribution of physiological signals in the time-frequency dimension through short-time Fourier transform combined with synchronous squeezing transform, accurately obtaining the energy flow characteristics between different frequency bands. By combining phase difference analysis with energy envelope cross-correlation calculation, it characterizes the energy transfer direction and associated time delay between channels, revealing the dynamic coupling relationship between multiple physiological channels. The physiological energy flow map combined with the health semantic heterogeneous map establishes an explicit mapping between keywords and physiological channels, quantifying health attention and driving intensity.

[0167] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A personalized recommendation method for health records based on artificial intelligence, characterized in that: include, Collect and preprocess multi-source data, perform short-time Fourier transform on physiological signals, calculate complex spectrum and short-time energy spectrum, redistribute short-time energy spectrum to frequency index corresponding to instantaneous frequency, map frequency index to frequency band, and generate aggregated energy. The multi-source data includes physiological signals and speech signals; Phase extraction is performed on the complex spectrum to generate joint features, the square root of the aggregated energy is calculated, an energy envelope is generated, the maximum associated time delay of the energy envelope is identified, and the total energy transfer is calculated. Construct energy flow graphs and heterogeneous graphs, calculate the conservation residuals and health attention of channels, combine the conservation residuals and normalized health attention to calculate the channel driving strength, screen significant keywords, generate a set of significant keywords, screen matching recommendations, and generate recommended health profiles. The construction of energy flow graphs and heterogeneous graphs, calculation of channel conservation residuals and health concerns, and calculation of channel driving strength by combining conservation residuals and normalized health concerns include: The aggregated energy is summed at the frame level within the time grid to generate the grid-level total energy; Based on the total energy transfer, the net inflow and net outflow are calculated. Physiological signal channels are defined as channel nodes, and the energy flow between channels is used as channel edges to generate an energy flow graph. Combining the grid-level total energy, net inflow, and net outflow, the channel conservation residuals are calculated. An explicit mapping table is constructed based on expert knowledge to define the correspondence between each health keyword and physiological signal channels; Keywords are extracted from the keyword set as keyword nodes. Based on the explicit mapping table, keywords are filtered in the channel mapping table and defined as keyword edges to generate a heterogeneous graph. The formula is as follows: ， in The relationship between keyword w and channel i is represented by 1 indicating correlation and 0 indicating no correlation. This is the mapping table for channel i; Count the number of times keywords appear in the grid and calculate the edge strength of the heterogeneous graph; Based on channel nodes, aggregate channels and related keyword nodes to generate the total association strength of channels, and then perform normalization processing; By combining the conserved residuals and the normalized total correlation strength, the health concern of the channel is calculated and normalized. Then, by combining the conserved residuals and the normalized health concern, the channel driving strength is calculated. The process of filtering significant keywords, generating a set of significant keywords, filtering matching recommendations, and generating recommended health profiles includes: Select the channel and frequency band corresponding to the maximum total driving intensity to obtain the key driving channels and frequency bands. Extract the edge strengths related to the key driving channels from the heterogeneous graph, screen significant keywords, and generate a set of significant keywords. The formula is as follows: ， in For channel In the time grid prominent keywords, For channel Edge strength with keyword w As the filtering threshold, A set of prominent keywords; A health record recommendation database is collected from the health management platform. Based on key driving channels and significant keywords, matching recommendations are screened, and a relevance score is calculated for each recommendation in combination with energy transfer amount. Sort the relevance scores in descending order, filter the top O recommended items, and generate recommended health profiles, where O is the number of recommended items.

2. The personalized health record recommendation method based on artificial intelligence as described in claim 1, characterized in that: The process of performing a short-time Fourier transform on the physiological signal, calculating the complex spectrum and the short-time energy spectrum, redistributing the short-time energy spectrum to the frequency index corresponding to the instantaneous frequency, mapping the frequency index to the frequency band, and generating aggregated energy includes: Speech recognition algorithms are used to convert speech signals into text. The text is then divided into time frames within a time grid to generate frame-level text fragments. Based on these frame-level text fragments, regular expression matching methods are used to extract keywords from the ICD-10 International Classification of Diseases to generate a keyword set. The physiological signal is divided into time grids by setting the time grid duration using a sliding window. Based on the time grids, a short-time Fourier transform is performed on the physiological signal to calculate the complex spectrum. Based on the complex spectrum, the short-time energy spectrum is calculated, and the instantaneous frequency of the complex spectrum is calculated through the phase derivative. Then, the short-time energy spectrum is redistributed to the frequency index corresponding to the instantaneous frequency using the synchronous squeezing transformation. The frequency band set is defined using an empirical method, frequency indices are mapped to frequency bands, and each frequency band is aggregated to generate aggregated energy.

3. The personalized health record recommendation method based on artificial intelligence as described in claim 2, characterized in that: The phase extraction of the complex spectrum to generate joint features includes: Phase extraction is performed on the complex spectrum to generate short-time phase spectra for each channel at each frequency and time frame. The phase difference of the short-time phase spectra is calculated, and a phase wrapping function is used to limit the phase difference to a specific value. ; Calculate the first-order forward difference of the short-time energy spectrum to generate the energy increment, calculate the forward energy increment rate, and combine the short-time phase spectrum and the forward energy increment rate to generate a joint feature.

4. The personalized health record recommendation method based on artificial intelligence as described in claim 3, characterized in that: The calculation of the square root of the aggregated energy, generation of an energy envelope, identification of the maximum associated time delay of the energy envelope, and calculation of the total energy transfer amount include: Calculate the square root of the aggregated energy, generate the energy envelope for each channel, calculate the cross-correlation function of the energy envelope within the time grid, identify the maximum associated time delay of the energy envelope, and calculate the amount of energy transferred. Aggregate to the time grid level and calculate the total energy transfer within the time grid.

5. The personalized health record recommendation method based on artificial intelligence as described in claim 1, characterized in that: The collection and preprocessing of multi-source data includes: The system collects multi-source data from users through smart bracelets and smartphones, and performs noise reduction and normalization processing. The multi-source data includes physiological signals and speech signals; The physiological signals include heart rate, blood pressure, and respiratory rate data.

6. A personalized health record recommendation system based on artificial intelligence, based on the personalized health record recommendation method based on artificial intelligence as described in any one of claims 1 to 5, characterized in that: include, The data collection and processing module is used to collect multi-source data from users and perform noise reduction and normalization processing. The energy extraction module is used to extract the complex spectrum and short-time energy spectrum of physiological signals, redistribute the short-time energy spectrum to the frequency index corresponding to the instantaneous frequency, map the frequency index to the frequency band, and generate aggregated energy. The feature transfer module is used to combine the short-time phase spectrum and the energy increment rate to extract joint features between channels, identify the maximum associated time delay of the energy envelope, and calculate the total energy transfer amount. The channel module is used to construct energy flow graphs and heterogeneous graphs, and to calculate the channel conservation residuals and health concerns. The personalized recommendation module is used to filter significant keywords, match recommended items from the health record database, sort them by relevance score, and generate recommended health records.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the personalized health record recommendation method based on artificial intelligence as described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the personalized health record recommendation method based on artificial intelligence as described in any one of claims 1 to 5.

Images