A personalized recommendation system for traditional chinese medicine blood-activating and stasis-removing prescriptions for coronary heart disease

Abstract

The present application relates to the technical field of drug recommendation, in particular to a personalized recommendation system for Chinese medicine blood-activating and stasis-removing prescriptions for coronary heart disease. According to the matching reliability of the description word vector, and the semantic universality and deviation from the standard semantic degree of the description word vector, a comprehensive risk coefficient is obtained to select the to-be-corrected word vector. According to the semantic dispersion degree of the to-be-corrected word vector and the description word vector relative to the same standard word vector in the standard word vector set, the change trend of the association degree of the to-be-corrected word vector and the standard word vector is fed back to obtain the spatiotemporal correction reasonable coefficient of the to-be-corrected word vector and the standard word vector. Then, the matching degree of the corresponding two word vectors is corrected to obtain the final matching degree, and an associated recommendation list of the current interaction record is generated. The present application dynamically integrates the internal risk and external evidence to iteratively adjust the matching degree, effectively eliminating colloquial and ambiguous expressions while retaining valuable personalized features.

Description

Technical Field

[0001] This invention relates to the field of drug recommendation technology, specifically to a personalized recommendation system for traditional Chinese medicine formulas for promoting blood circulation and removing blood stasis in patients with coronary heart disease. Background Technology

[0002] Coronary heart disease (CHD) is caused by the formation of plaques on the walls of the coronary arteries due to the deposition of cholesterol and other deposits, leading to narrowing or blockage of the blood vessels and obstructing blood flow. It falls under the category of "blood stasis" in Traditional Chinese Medicine (TCM). TCM formulas for promoting blood circulation and removing blood stasis are numerous, with varying medicinal properties, meridian tropism, and emphases. Clinically, doctors need to personalize the prescriptions and dosages based on individual patient differences and the severity of their condition to achieve precise treatment and improve efficacy.

[0003] However, high-level TCM resources are limited. To improve diagnostic and treatment efficiency, online consultation platforms often use natural language processing technology to automatically match patients' colloquial descriptions of their symptoms into standard medical terms, assisting non-professional users in mapping colloquial descriptions to standard medical terminology. However, patients generally lack medical knowledge, and their descriptions are often inaccurate, leading to errors in the matching results. This can not only mislead doctors but also increase their workload in verification and correction.

[0004] To address the matching errors caused by inaccurate descriptions, existing technologies typically employ clustering algorithms to group patient descriptions and directly remove matches at the cluster edges to improve overall accuracy. However, cluster edge nodes may contain unique, atypical information about the patient's actual condition (such as rare but crucial physical signs), possessing potential reference value. Simply removing edge nodes may result in the loss of key information reflecting individual differences, leading to a lack of personalization and comprehensiveness in the recommendation results. Summary of the Invention

[0005] To address the technical problems of missing key information and low matching accuracy caused by directly removing edge nodes through clustering to process vague, colloquial user descriptions, the present invention aims to provide a personalized recommendation system for traditional Chinese medicine formulas for promoting blood circulation and removing blood stasis in patients with coronary heart disease. The specific technical solution adopted is as follows: This invention proposes a personalized recommendation system for traditional Chinese medicine formulas for promoting blood circulation and removing blood stasis in patients with coronary heart disease. The system includes: The data acquisition module is used to obtain the set of descriptive word vectors of the user in the current interaction record, the set of feedback word vectors of several historical interaction records, and the set of standard word vectors of the standard semantic term library. The risk analysis module is used to obtain the comprehensive risk coefficient of the descriptive word vector based on the matching reliability of the descriptive word vector, as well as the semantic universality and deviation from the standard semantics of the descriptive word vector, and to select the word vector to be corrected. The spatiotemporal gain analysis module is used to obtain the spatiotemporal correction rationality coefficient between each word vector to be corrected and each standard word vector based on the semantic dispersion of each word vector to be corrected and the other descriptive word vectors relative to the same standard word vector in the standard word vector set, and the changing trend of the correlation between each feedback word vector set and the corresponding standard word vector. The correction module is used to iteratively adjust the matching degree between each word vector to be corrected and each standard word vector according to the spatiotemporal correction rationality coefficient and the comprehensive risk coefficient, until the preset convergence condition is met, and obtain the final matching degree. The recommendation module is used to generate a list of related recommendations for the current interaction record based on the final matching degree.

[0006] Further, the step of obtaining the comprehensive risk coefficient of the descriptive word vectors and selecting the word vectors to be corrected includes: Select the standard word vector corresponding to the maximum matching degree between the description word vector and all standard word vectors in the standard word vector set, and denote it as the matching vector of the description word vector; obtain the matching error degree based on the reliability of the matching of the description word vector to its matching vector; The descriptive word vectors in the descriptive word vector set are divided into different clusters; the cluster edge degree is obtained based on the relative position information of the descriptive word vectors in their respective clusters and the semantic prevalence of the descriptive word vectors. The arithmetic mean of the matching error and the clustering edge degree is used as the comprehensive risk coefficient for describing word vectors; Select descriptive word vectors from the descriptive word vector set whose comprehensive risk coefficient is greater than the preset risk threshold, and denote them as word vectors to be corrected.

[0007] Further, obtaining the matching error degree includes: Calculate the degree of matching between the description word vector and the remaining standard word vectors in the standard word vector set, excluding its matching vector. Select the standard word vector with the highest degree of matching and denote it as the reliable reference vector. The ratio of the degree of matching between the descriptive word vector and its reliable reference vector, and the degree of matching between the matching vectors, is denoted as the similarity ratio. The degree of matching between the matching vector of the descriptive word vector and the reliable reference vector is denoted as semantic proximity. The semantic proximity is negatively correlated, and the product of the mapping result and the similarity ratio is used as the matching error of the descriptive word vector.

[0008] Furthermore, obtaining the cluster edge degree includes: The ratio of the distance between the descriptive word vector and the cluster center of its cluster is used as the numerator, and the cluster radius of the cluster in which the descriptive word vector is located is used as the intra-cluster eccentricity of the descriptive word vector. A negative correlation mapping is performed on the cluster size of the cluster in which the descriptive word vector is located, and the product of the mapping result and the eccentricity within the cluster is used as the cluster edge degree of the descriptive word vector.

[0009] Furthermore, obtaining the spatiotemporal correction coefficients for each descriptive word vector and each standard word vector includes: Choose one word vector to be corrected and denote it as the description example vector; choose one word vector from the standard word vector set and denote it as the standard example vector. Calculate the matching degree between the description word vectors (excluding the description example vectors) and the standard example vectors in the description word vector set. Average the differences between all matching degrees and the matching degree between the description example vectors and the standard example vectors to obtain the contextual dispersion of the description example vectors and the standard example vectors. Based on the changing trend of the correlation between each set of feedback word vectors and the standard example vectors, the historical trend inertia of the standard example vectors is obtained; The context dispersion is negatively correlated and mapped, and the arithmetic mean of the mapping result and the historical trend inertia is used as the spatiotemporal correction coefficient describing the example vector and the standard example vector.

[0010] Furthermore, obtaining the historical trend inertia of the standard example vector includes: The average matching degree between all feedback word vectors in each feedback word vector set and the standard example vectors is used to obtain the feedback relevance. Arrange all feedback word vector sets with the feedback correlation degree of the standard example vectors in reverse chronological order according to the historical interaction records to obtain the feedback correlation sequence; Obtain the first-order difference sequence of the feedback correlation sequence, and select the element with the smallest subscript value from the two adjacent elements corresponding to the first non-negative number in the feedback correlation sequence as the effective feedback cutoff element; The ratio of the difference between the first element in the feedback association sequence and the effective feedback cutoff element to the time interval between the corresponding historical interaction records of the two elements is used as the historical trend inertia of the standard example vector.

[0011] Furthermore, the step of dividing the descriptive word vectors in the descriptive word vector set into different clusters includes: The feature n-tuple of the description word vector is composed of the description word vector, the matching vector of the description word vector, and the degree of matching between the description word vector and its matching vector; Based on the distance between the feature n-tuples of different descriptive word vectors, the descriptive word vectors in the descriptive word vector set are clustered to obtain different clusters.

[0012] Furthermore, obtaining the final matching degree includes: The normalized result of the product of the comprehensive risk coefficient and the spatiotemporal correction reasonable coefficient is used as the matching correction coefficient between each word vector to be corrected and each standard word vector. Calculate the sum of the matching correction coefficient and constant 1 between each word vector to be corrected and each standard word vector, and multiply the sum by the matching degree of the corresponding two word vectors as the corrected matching degree of the corresponding two word vectors; Based on the matching degree between the word vector to be corrected and all standard word vectors after correction, the matching vector and feature n-tuple of the word vector to be corrected are updated; and based on the updated feature n-tuple of the word vector to be corrected, the set of descriptive word vectors is re-divided into different clusters to obtain the updated clusters. Based on the updated clusters, the comprehensive risk coefficient of the descriptive word vectors is re-obtained, and the updated word vectors to be corrected are selected. When the updated word vectors to be corrected do not meet the convergence condition and the number of iterations has not reached the preset iteration threshold, the matching correction coefficient between the updated word vectors to be corrected and each standard word vector is recalculated, and the matching degree between the corresponding two word vectors is further corrected. This process continues until the updated word vectors to be corrected meet the convergence condition or the number of iterations reaches the preset iteration threshold. The current corrected matching degree is then used as the final matching degree between each descriptive word vector and each standard word vector. The convergence condition includes: the proportion of the updated word vectors to be corrected in the set of descriptive word vectors is less than a preset threshold.

[0013] Further, generating a related recommendation list for the current interaction record based on the final matching degree includes: For each descriptive word vector, the maximum value among the final matching degrees of the descriptive word vector with all standard word vectors is selected. When the maximum value is greater than the preset effective matching threshold, the drug associated with the standard word vector corresponding to the maximum value is used as the recommended drug. The recommended drugs in the descriptive word vector set are sorted from high to low according to the maximum value corresponding to the descriptive word vector, and an associated recommendation list for the current interaction record is generated.

[0014] Furthermore, the degree of matching between a word vector and a standard word vector is described by the cosine similarity between the two word vectors.

[0015] The present invention has the following beneficial effects: In this embodiment of the invention, the matching reliability characterizes the ambiguity of the semantic mapping of the descriptive word vectors, while the semantic universality and deviation from the standard semantics assess the outlier nature of the semantic distribution of the descriptive word vectors. A comprehensive analysis yields a comprehensive risk coefficient that represents the internal risk of the descriptive word vectors, enabling the identification and retention of potentially important information rather than simple removal, thus avoiding information loss due to the accidental deletion of atypical but authentic personalized descriptions. The spatiotemporal gain analysis module generates a reasonable spatiotemporal correction coefficient by fusing spatial-dimensional group semantic consistency with historical demand trends in the temporal dimension. This provides a reasonable correction direction and evidence support for the matching of word vectors to be corrected, guiding the originally ambiguous semantic associations to the correct matching target, significantly improving the accuracy and interpretability of the correction. The correction module iteratively adjusts the matching degree by dynamically fusing the comprehensive risk coefficient representing the inherent risk with the spatiotemporal correction coefficient of external evidence. While retaining valuable personalized features, it effectively eliminates noise introduced by colloquial and ambiguous expressions, resulting in the most stable and accurate final matching result. The technical indicators after deep semantic cleaning and correction are then transformed into an intuitively perceptible list of related recommendations, enhancing the reference value of personalized recommendations. Attached Figure Description

[0016] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a system structure diagram of a personalized recommendation system for traditional Chinese medicine prescriptions for promoting blood circulation and removing blood stasis in coronary heart disease, provided in one embodiment of the present invention; Figure 2 This is a structural diagram of a risk analysis module provided in one embodiment of the present invention; Figure 3 This is a schematic diagram of a computer device for a personalized recommendation system of traditional Chinese medicine formulas for promoting blood circulation and removing blood stasis in coronary heart disease, provided in one embodiment of the present invention. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a personalized recommendation system for traditional Chinese medicine formulas for promoting blood circulation and removing blood stasis in coronary heart disease based on the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0020] The following description, in conjunction with the accompanying drawings, details the specific scheme of the personalized recommendation system for traditional Chinese medicine prescriptions for promoting blood circulation and removing blood stasis in coronary heart disease provided by this invention.

[0021] Example 1: Please see Figure 1 The diagram illustrates a system block diagram of a personalized recommendation system for traditional Chinese medicine formulas for promoting blood circulation and removing blood stasis in coronary heart disease, provided by an embodiment of the present invention. The system includes: a data acquisition module 110, a risk analysis module 120, a spatiotemporal gain analysis module 130, a correction module 140, and a recommendation module 150.

[0022] The data acquisition module 110 is used to acquire the set of descriptive word vectors of the user in the current interaction record, the set of feedback word vectors of several historical interaction records, and the set of standard word vectors of the standard semantic term library.

[0023] This solution is suitable for primary triage or general health consultation scenarios, where user input quality is relatively low. It acquires the user's current online consultation interaction record, i.e., the user's spoken description of their symptoms via text input, and records this as the current interaction record. The text of this record is preprocessed, including removing irrelevant characters, punctuation, and stop words. Using natural language processing tools, combined with a custom dictionary of coronary heart disease and traditional Chinese medicine symptomology, the preprocessed text is segmented into words. Keywords expressing symptoms, signs, and feelings are selected and recorded as symptom keywords. A pre-trained deep semantic model (such as the ClinicalBERT model) is used to generate semantic embedding vectors corresponding to the symptom keywords, recorded as descriptive word vectors. All descriptive word vectors generated from the current interaction record constitute a descriptive word vector set.

[0024] Retrieve the user's most recent three completed historical interaction records. For each historical interaction record: extract the text describing the user's symptoms in that record, as well as the blood-activating and stasis-removing prescription made by the doctor in that record; for this text, apply the same text preprocessing, word segmentation, and vectorization process as the current interaction record, and record the semantic embedding vectors corresponding to the symptom keywords generated from the historical interaction records as feedback word vectors; the set of feedback word vectors is composed of the semantic embedding vectors corresponding to all symptom keywords. Simultaneously, establish a mapping relation database to associate each symptom keyword with its corresponding Chinese medicine name, and associate the drugs contained in the prescriptions made by the TCM doctor in the historical interaction records with the feedback word vectors through the relation database.

[0025] A standard semantic terminology database covering the field of traditional Chinese medicine for promoting blood circulation and removing blood stasis in coronary heart disease was constructed. The original terms in this database were derived from clinical practice guidelines and recognized medical knowledge graphs. The core content included: (1) commonly used single herbs (such as Danshen, Chuanxiong, and Honghua) and compound herbs in formulas for promoting blood circulation and removing blood stasis; (2) standardized efficacy, symptom and syndrome terms related to the pathogenesis of "blood stasis" in coronary heart disease and the treatment of promoting blood circulation and removing blood stasis (such as "promoting blood circulation and removing blood stasis", "unblocking meridians and relieving pain", "chest pain", "heart pain", and "qi stagnation and blood stasis"). Each term in the standard semantic terminology database was vectorized using a pre-trained deep semantic model (such as the ClinicalBERT model) to obtain the standard word vector of the corresponding term. The standard word vectors of all terms in the standard semantic terminology database constituted the standard word vector set.

[0026] The risk analysis module 120 is used to obtain the comprehensive risk coefficient of the descriptive word vector based on the matching reliability of the descriptive word vector, as well as the semantic universality and deviation from the standard semantics of the descriptive word vector, and to select the word vector to be corrected.

[0027] Matching reliability characterizes the ambiguity of the semantic mapping of word vectors. Semantic universality and deviation from standard semantics assess the outlier nature of the semantic distribution of word vectors from two dimensions: local specificity and global rarity. Combining these factors to analyze the uncertainty, i.e., internal risk, of word vectors yields a comprehensive risk coefficient. Based on this comprehensive risk coefficient, word vectors that require correction due to vague or erroneous colloquial expressions can be accurately identified, providing a clear target for subsequent refined processing and fundamentally avoiding the loss of crucial information that may result from blindly eliminating them.

[0028] The spatiotemporal gain analysis module 130 is used to obtain the spatiotemporal correction rationality coefficient between each word vector to be corrected and each standard word vector based on the semantic dispersion of each word vector to be corrected and the other descriptive word vectors relative to the same standard word vector in the standard word vector set, and the changing trend of the correlation between each feedback word vector set and the corresponding standard word vector.

[0029] Based on clinical understanding, multiple symptom descriptions from a user in a single interaction typically point to an intrinsically related pathological state. Therefore, the descriptive word vectors should exhibit a certain clustering tendency in the semantic space, manifested as follows: when any descriptive word vector is highly correlated with a specific medical concept (i.e., a standard word vector), the correlation between other descriptive word vectors and that medical concept tends to be within a predictable range. A user's medication history during past treatments can reflect the patient's condition to some extent, containing temporal information about the disease's development; when the user's symptom description in the current interaction is semantically ambiguous, this temporal information can provide a reference-based evidence for discerning their true intentions. This mechanism aims to simulate a doctor's clinical thinking: when faced with ambiguous patient descriptions, doctors will assess the possibility of various interpretations by combining medical history trends, rather than forcibly distorting the current facts with history.

[0030] The word vector to be corrected and other descriptive word vectors show spatial consistency with the same standard word vector in the standard word vector set. The changing trend of the correlation between the feedback word vector set and the standard word vector shows the evolution of demand behavior in the time dimension. By introducing external evidence in the spatial and temporal dimensions, a clear direction of correction is provided for the matching degree between the word vector to be corrected and the standard word vector in the fuzzy state, i.e., the rationality of correction. The spatiotemporal correction rationality coefficient is obtained, which significantly improves the accuracy of the correction direction.

[0031] The correction module 140 is used to iteratively adjust the matching degree between each word vector to be corrected and each standard word vector based on the spatiotemporal correction reasonable coefficient and the comprehensive risk coefficient, until the preset convergence condition is met, and obtain the final matching degree.

[0032] The comprehensive risk coefficient, as an internal factor, quantifies the uncertainty of the word vector to be corrected, determining the required degree of correction for the matching degree between the word vector to be corrected and the standard word vector. The spatiotemporal correction rationality coefficient, as an external factor, provides correct directional guidance based on group consensus and historical trends. The combination of these two factors ensures targeted enhancement of nodes that are both unreliable and supported by clear spatiotemporal evidence, resulting in the final matching degree. This preserves personalized features while accurately eliminating matching biases caused by users' colloquial or ambiguous expressions. A single semantic correction is often insufficient to achieve optimal results; multiple fine-tuning steps gradually eliminate semantic oscillations and reduce the impact of colloquial and ambiguous expressions on the matching of the word vector to be corrected and the standard word vector.

[0033] The recommendation module 150 is used to generate a list of related recommendations for the current interaction record based on the final matching degree.

[0034] The final matching degree can present standard terms that best reflect the user's true intent. The technical indicators after deep semantic cleaning and correction are transformed into an intuitive and perceptible list of related recommendations, which improves the efficiency and accuracy of decision support and enhances the reference value of personalized recommendations.

[0035] Please see Figure 2 The diagram illustrates a structural diagram of a risk analysis module provided in an embodiment of the present invention. The risk analysis module includes: a matching error unit 121, a clustering edge analysis unit 122, and a risk analysis unit 123.

[0036] The matching error unit 121 is used to select the standard word vector corresponding to the maximum value from the matching degree between the description word vector and all standard word vectors in the standard word vector set, and denot it as the matching vector of the description word vector; and obtain the matching error degree according to the reliability of the description word vector matching its matching vector.

[0037] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the matching error degree includes: calculating the matching degree between the description word vector and the remaining standard word vectors in the standard word vector set excluding its matching vector, selecting the standard word vector corresponding to the maximum matching degree, and denoting it as the reliable reference vector; denoting the ratio of the matching degree of the description word vector with its reliable reference vector and the matching degree of the matching vector as the similarity ratio; denoting the matching degree of the matching vector of the description word vector and the reliable reference vector as the semantic proximity; performing a negative correlation mapping on the semantic proximity, and using the product of the mapping result and the similarity ratio as the matching error degree of the description word vector.

[0038] In this embodiment of the invention, the cosine similarity between the descriptive word vector and the standard word vector is used as the matching degree between the two word vectors; the larger the cosine similarity, the more semantically similar the two word vectors are. It should be noted that although the theoretical range of the cosine similarity between two word vectors is -1 to 1, when using certain specific word embedding models, such as the Word2Vec algorithm and the Bidirectional Encoder Representations from Transformers (BERT) model, all dimensions in the text vector are usually non-negative, and the cosine similarity between the two text vectors ranges from 0 to 1. Even with dense vectors, in the semantic matching scenario of this scheme, only positive relevance is usually considered; when the cosine similarity between two text vectors is negative, the cosine similarity is directly set to zero.

[0039] It's important to note that the matching vector describes the best match of the word vectors; the reliable reference vector represents the standard word vector most likely to confuse the matching vector, belonging to the second-best match. The similarity ratio measures the ambiguity of the matching results, while semantic closeness measures the semantic distinguishability between the best and second-best matches. A higher similarity ratio indicates greater semantic similarity between the best and second-best matches, meaning the matching vectors lack distinguishability and are more prone to ambiguity. Conversely, a lower semantic closeness indicates a greater semantic difference between the matching vector and the reliable reference vector, leading to completely different interpretations in case of mismatches. This results in more severe consequences for mismatches, making the matching vector less reliable and thus increasing the matching error. Therefore, the similarity ratio is positively correlated with the matching error, while semantic closeness is negatively correlated. The matching error reflects the reliability of the word vector matching; a higher value indicates a larger and less reliable matching error.

[0040] In this embodiment of the invention, since the cosine similarity, i.e. the degree of matching, between the descriptive word vector and the standard word vector ranges from 0 to 1, the degree of semantic matching also ranges from 0 to 1. Therefore, the semantic closeness can be negatively correlated by subtracting the constant 1 from the semantic closeness.

[0041] The clustering edge analysis unit 122 is used to divide the descriptive word vectors in the descriptive word vector set into different clusters; and to obtain the clustering edge degree based on the relative position information of the descriptive word vectors in their respective clusters and the semantic prevalence of the descriptive word vectors.

[0042] In this embodiment of the invention, the feature n-tuple of the description word vector is composed of the description word vector, the matching vector of the description word vector, and the degree of matching between the description word vector and its matching vector; based on the distance between the feature n-tuples of different description word vectors, the description word vectors in the set of description word vectors are clustered to obtain different clusters. It should be noted that the method for obtaining the distance between two feature n-tuples includes: calculating the difference between the constant 1 and the cosine similarity of the descriptive word vectors and the cosine similarity of the matching vectors in the feature n-tuples of the two feature word vectors, respectively, and denoting them as the descriptive word vector distance and the matching vector distance; denoting the absolute difference of the matching degree in the feature n-tuples of the two feature word vectors as the word vector distance and the matching vector distance; based on the word vector distance, matching vector distance, and absolute difference of the matching degree between each descriptive word vector in the descriptive word vector set and each standard word vector in the standard word vector set, the word vector distance, matching vector distance, and absolute difference of the matching degree corresponding to each pair of feature word vectors are normalized respectively, to obtain the normalized word vector distance, normalized matching vector distance, and normalized matching distance, respectively, and the arithmetic mean of the three is calculated as the distance between the feature n-tuples of the corresponding two feature word vectors. Where the number of elements in the feature n-tuple is 3, then n is 3.

[0043] In one implementation of this invention, the DBSCAN algorithm is used to cluster the descriptive word vectors.

[0044] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the cluster edge degree includes: using the distance between the descriptive word vector and the cluster center of its cluster as the numerator and the cluster radius of the cluster in which the descriptive word vector is located as the denominator, the ratio obtained is used as the intra-cluster eccentricity of the descriptive word vector; performing a negative correlation mapping on the cluster size of the cluster in which the descriptive word vector is located, and using the product of the mapping result and the intra-cluster eccentricity as the cluster edge degree of the descriptive word vector.

[0045] It should be noted that patients' overall descriptions of their symptoms are based on their own physical sensations and are generally not significantly inaccurate. However, words that are colloquial or ambiguous may differ considerably from the overall clustering results, potentially occupying less prominent and isolated positions. Intra-cluster eccentricity measures the relative deviation of the descriptive word vector within a local group. A higher value indicates a greater deviation from the standard semantics (i.e., the mainstream semantic center), suggesting a higher likelihood of atypical features or misrepresentations. Cluster size reflects the semantic prevalence of the cluster to which the descriptive word vector belongs. A lower value indicates a less common overall semantic pattern within the cluster, further deviating from mainstream semantics. Intra-cluster eccentricity and cluster size assess the deviation of the descriptive word vector from mainstream semantics from two dimensions: local specificity and global rarity, respectively, yielding cluster marginality. A higher value indicates a greater deviation from mainstream semantics, suggesting a higher likelihood of atypical features or misrepresentations. Therefore, intra-cluster eccentricity is positively correlated with cluster marginality, while cluster size and cluster marginality are negatively correlated.

[0046] In this embodiment of the invention, the ratio of the cluster size of the cluster to which each descriptive word vector belongs to the maximum cluster size among all clusters is calculated. By subtracting a constant 1 from this ratio, the cluster sizes of the descriptive word vectors are negatively correlated and normalized to obtain the standard ratio. The product of the standard ratio and the intra-cluster eccentricity is then normalized to obtain the cluster marginality of the descriptive word vectors. Specifically, based on the product of the standard ratio and the intra-cluster eccentricity of each descriptive word vector in the descriptive word vector set, the Z-score normalization method is used to normalize the product of the standard ratio and the intra-cluster eccentricity of each descriptive word vector to obtain the cluster marginality.

[0047] It is important to note that the cluster radius is the distance from the cluster center to the farthest element within that cluster, and the cluster size is the number of elements within the cluster. When a cluster contains only one element or all element vectors overlap and the cluster radius is 0, the descriptor vectors within that cluster are extremely outliers, and the descriptor vectors within the cluster can be directly set to the maximum value of 1.

[0048] Risk analysis unit 123 is used to take the arithmetic mean of matching error degree and cluster edge degree as the comprehensive risk coefficient of the descriptive word vector.

[0049] It should be noted that the matching error degree characterizes the ambiguity of the semantic mapping, while the clustering edge degree characterizes the outlier nature of the semantic distribution. Both reflect the uncertainty of the descriptive word vector from different dimensions, resulting in a comprehensive risk coefficient. The larger this value is, the greater the error of the descriptive word vector in matching standard terms, and the more it deviates from the mainstream expression in the group semantic distribution. In other words, the lower its credibility as an effective input feature, the more it needs subsequent correction processing.

[0050] It is important to note that before analyzing the overall risk value, the matching error of each descriptive word vector in the descriptive word vector set needs to be normalized using the Z-score standardization method to obtain the standard cluster marginality; and the overall risk coefficient is then calculated using the standard cluster marginality.

[0051] In this embodiment of the invention, descriptive word vectors with a comprehensive risk coefficient greater than a preset risk threshold are selected from the descriptive word vector set and denoted as word vectors to be corrected. It should be noted that, based on historically labeled and expert-verified clinical conversation data, a specific higher-order quantile (e.g., the 90th or 95th percentile) of the comprehensive risk coefficient of each descriptive word vector is calculated, and this quantile is used as the preset risk threshold.

[0052] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the spatiotemporal correction rationality coefficient includes: randomly selecting a word vector to be corrected as the description example vector, and randomly selecting a word vector from the standard word vector set as the standard example vector; calculating the matching degree between the remaining description word vectors in the description word vector set (excluding the description example vector) and the standard example vector, and averaging the differences between all matching degrees and the matching degree between the description example vector and the standard example vector to obtain the contextual dispersion of the description example vector and the standard example vector; obtaining the historical trend inertia of the standard example vector based on the changing trend of the correlation between each feedback word vector set and the standard example vector; performing a negative correlation mapping on the contextual dispersion, and using the product of the mapping result and the historical trend inertia as the spatiotemporal correction rationality coefficient between the description example vector and the standard example vector.

[0053] It should be noted that the context discrete quantification describes the degree of deviation of the example vector from the group consensus when mapping the standard example vector. The larger this value is, the lower the consistency between the example vector and the overall descriptive context of the current interaction record when mapping the standard example vector. It is very likely that there is noise or incorrect association. In order to prevent the introduction of new deviations by forcibly increasing the matching degree due to noise interference, the correction should be reduced. Conversely, if the matching direction is confirmed by the group consensus, it is more appropriate to increase the correction.

[0054] In this embodiment of the invention, the method for obtaining historical trend inertia includes: averaging the matching degree between all feedback word vectors in each feedback word vector set and the standard example vector to obtain the feedback correlation degree; arranging the feedback correlation degrees of all feedback word vector sets and the standard example vector in reverse chronological order of historical interaction records to obtain a feedback correlation sequence; obtaining the first-order difference sequence of the feedback correlation sequence, and selecting the element with the smallest subscript value from the two adjacent elements corresponding to the first non-negative number in the first-order difference sequence as the effective feedback cutoff element; and using the ratio of the difference between the first element in the feedback correlation sequence and the effective feedback cutoff element to the time interval between the corresponding historical interaction records of the two elements as the historical trend inertia of the standard example vector.

[0055] It should be noted that the feedback correlation reflects the overall semantic relevance between the prescriptions written by doctors for users in a single historical interaction and the standard example vector. Sorting by reverse chronological order ensures that the analysis traces back from the most recent to the historical, aligning with the clinical understanding that "recent information is more important" in treatment decisions. The effective feedback cutoff element dynamically extracts the most recent monotonic interval of change. Between the two interaction records corresponding to the first element and the effective feedback cutoff element in the feedback correlation sequence, the aforementioned semantic relevance gradually increases. A greater historical trend inertia indicates a clearer and more consistent demand for the medication associated with the standard example vector in historical treatments. Based on the principle of user behavior continuity, this demand is more likely to continue into the current interaction, increasing the likelihood that the descriptive example vector aims to express the standard example vector. Therefore, there is greater justification for forcibly correcting the ambiguous descriptive example vector to the standard example vector, thereby accurately capturing the user's implicit true needs.

[0056] Contextual dispersion and historical trend inertia, from the perspectives of group consistency in the spatial dimension and behavioral evolution in the temporal dimension, respectively, indicate that the greater the magnitude of both, the more reasonable the correction of the matching degree between the descriptive example vector and the standard example vector is, resulting in a larger spatiotemporal correction rationality coefficient. Therefore, contextual dispersion and spatiotemporal correction rationality coefficient are negatively correlated, while historical trend inertia and spatiotemporal correction rationality coefficient are positively correlated. In this embodiment of the invention, based on the contextual dispersion and historical trend inertia of each descriptive word vector and each standard word vector, the Z-score standardization method is used to normalize the contextual dispersion and historical trend inertia of the descriptive example vector and the standard example vector to obtain standard dispersion and standard inertia; the difference between constant 1 and standard dispersion and the arithmetic mean of standard inertia are calculated as the spatiotemporal correction rationality coefficient between the descriptive example vector and the standard example vector.

[0057] It should be noted that if all elements in the first-order difference sequence of the feedback correlation sequence are negative, the first element in the feedback correlation sequence is recorded as the effective feedback cutoff element; if there are no non-negative elements in the first-order difference sequence of the feedback correlation sequence, indicating that there is no positive trend for reference in the historical stage, the historical trend inertia is set to 0.

[0058] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the final matching degree includes: normalizing the product of the comprehensive risk coefficient and the spatiotemporal correction reasonable coefficient as the matching correction coefficient between each word vector to be corrected and each standard word vector; calculating the sum of the matching correction coefficient between each word vector to be corrected and each standard word vector and the constant 1, and multiplying the sum by the matching degree of the corresponding two word vectors as the corrected matching degree of the corresponding two word vectors; updating the matching vector and feature n-tuple of the word vector to be corrected based on the corrected matching degree of the word vector to be corrected with all standard word vectors respectively; and re-dividing the descriptive word vector set into different clusters based on the updated feature n-tuple of the word vector to be corrected. The class is updated to obtain the cluster class; based on the updated cluster class, the comprehensive risk coefficient of the descriptive word vectors is re-obtained and the updated word vectors to be corrected are selected; when the updated word vectors to be corrected do not meet the convergence condition and the number of iterations has not reached the preset iteration threshold, the matching correction coefficient between the updated word vectors to be corrected and each standard word vector is recalculated, and the matching degree between the corresponding two word vectors is further corrected; until the updated word vectors to be corrected meet the convergence condition or the number of iterations reaches the preset iteration threshold, the current corrected matching degree is taken as the final matching degree between each descriptive word vector and each standard word vector; the convergence condition includes: the proportion of the updated word vectors to be corrected in the set of descriptive word vectors is less than the preset number threshold.

[0059] It should be noted that the higher the comprehensive risk coefficient and the spatiotemporal correction reasonableness coefficient, the greater the degree of matching between the word vector to be corrected (which is vague in description but conforms to historical habits and does not conflict with the context) and the standard word vector should be in order to accurately hit the user's potential intent. Therefore, the required degree of correction for this matching degree is greater, resulting in a larger matching correction coefficient. Conversely, for descriptive word vectors that are clear, vague in description, lack historical support, and have significant contextual differences, the required degree of correction is smaller in order to maintain a low matching degree. Adjusting the matching degree using the matching correction coefficient achieves a targeted boost to ambiguous semantics.

[0060] In this embodiment of the invention, based on the product of the comprehensive risk coefficient and the spatiotemporal correction reasonable coefficient of each descriptive word vector and each standard word vector, the Z-score standardization method is used to normalize the product of the comprehensive risk coefficient and the spatiotemporal correction reasonable coefficient of each descriptive word vector and each standard word vector.

[0061] As an example, suppose there are three descriptive word vectors (A1, A2, A3) and two standard word vectors (B1, B2). The initial matching degree of A1 with B1 and B2 is as follows: , The initial matching vector of A1 is B1, and the initial feature n-tuple of A1 is... The matching vectors for A2 and A3 are both B2, and the initial matching degrees of A2 and A3 with B2 are respectively: , A2 initial feature n-tuple A3 initial feature n-tuple The initial set of word vectors to be corrected .

[0062] First iteration: Calculate the matching correction coefficients between the word vector A1 to be corrected and B1 and B2 respectively. , . use , In turn , After correction, the matching degree between the word vector A1 to be corrected and B1 and B2 is obtained. , Assuming Update the matching vector of A1 to B2. The updated feature n-tuple of A1 The original feature n-tuples of A2 and A3 are directly used as the updated feature n-tuples, that is, the updated feature n-tuples of A2. A3 initial feature n-tuple .based on The descriptive word vectors are re-clustered. The new clustering results may cause changes in the relative positions of A2 and A3 within their respective clusters. Based on the new clusters, the system recalculates the comprehensive risk coefficients of A1, A2, and A3, obtaining the following results sequentially. ; Assumption The updated set of word vectors to be corrected is greater than the preset risk threshold. At this point, let's assume the number of word vectors to be corrected after the update is... If the number of elements in the middle does not meet the convergence condition and the number of iterations is 1, the preset iteration threshold has not been reached, so proceed to the second round of iteration.

[0063] It should be noted that since A2 and A3 were not word vectors to be corrected in the first iteration, Let A2 be the initial matching degree with B1 and B2 respectively, and then let A2 be the revised matching degree with B1 and B2 respectively. , The degree of matching between A3 and the corrected versions of B1 and B2, respectively. , This also corresponds to the initial matching degree of the two word vectors.

[0064] Second iteration: Calculation Matching correction coefficients of word vector A2 to be corrected with B1 and B2 respectively , . use , In turn After correction, the matching degree between the word vector A2 to be corrected and B1 and B2 is obtained. Similar to the first iteration, the matching degrees of A1 and A3 calculated in the first iteration with the corrected standard word vectors B1 and B2 are directly used as the corrected matching degrees of A1 and A3 with the corrected standard word vectors B1 and B2 in the second iteration. Assume... The matching vector for A2 remains B2. Although the matching relationship has not fundamentally changed, the matching degree has been reasonably calibrated. The updated feature n-tuples for A2... The feature n-tuples obtained from A1 and A3 in the first iteration are used as the updated n-tuples in the second iteration, and clustering continues. The system recalculates the comprehensive risk coefficients of A1, A2, and A3, and obtains the results sequentially. Assuming If all values ​​exceed the preset risk threshold, the new set of word vectors to be corrected becomes an empty set. At this point, the convergence condition is met, and the loop terminates.

[0065] In the second iteration, the corrected matching degrees of A1, A2, and A3 with B1 and B2, respectively, are taken as the final matching degrees of the corresponding word vectors. In this example, the matching result of A1 may be corrected from B1 and stabilized as B2, while the matching result of A2 is strengthened to B2. The matching relationship of the entire system reaches an internally stable state through the consistency test of context and historical trends. It should be noted that the methods for obtaining the word vectors to be corrected, the matching correction coefficients, and the final matching degrees are the same in the first iteration and subsequent iterations.

[0066] In one implementation of this invention, the method of this solution is applied to a large number of historical interaction records for backtracking testing to observe the convergence speed and result stability under different thresholds. The selection of a preset quantity threshold (e.g., set to 5%-20% of the number of descriptive word vectors in the descriptive word vector set) aims to balance the thoroughness of correction with computational efficiency, ensuring that the system can reduce the proportion of word vectors to be corrected to a low level within an acceptable timeframe in most cases, thereby achieving stable output. The setting of a preset iteration threshold (e.g., set to 5 times) is to prevent excessively long loops in rare abnormal cases, serving as a safe upper limit for ensuring system response time. The above thresholds can be adaptively adjusted according to the computing power and real-time requirements of the actual deployment environment.

[0067] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the associated recommendation list includes: for each descriptive word vector, selecting the maximum value among the final matching degrees of the descriptive word vector and all standard word vectors; when the maximum value is greater than a preset effective matching threshold, taking the drug associated with the standard word vector corresponding to the maximum value as the recommended drug; sorting the recommended drugs of the descriptive word vectors in the descriptive word vector set from high to low according to the maximum value corresponding to the descriptive word vector, and generating the associated recommendation list of the current interaction record.

[0068] It should be noted that the recommended drugs for a single descriptive word vector correspond to the most prominent symptom features. Setting a preset effective matching threshold can effectively filter out matching noise that is locally optimal but still has low overall confidence, preventing interference from irrelevant drugs that may be caused by forced recommendations. The higher a drug ranks in the associated recommendation list, the higher the confidence of the associated symptom description, meaning that the symptom is more typical, and it should be given higher consideration when formulating personalized prescriptions.

[0069] In one implementation of this invention, the maximum value of the corrected matching degree between a large number of correct descriptive word vectors confirmed by medical experts and all standard word vectors is calculated, and the distribution range of the value is analyzed. The preset effective matching threshold can be set as the lower confidence interval boundary of the above distribution range or a certain percentile (such as the 10th percentile) to ensure that the matching degree reaches the minimum standard that can be clinically referenced.

[0070] The resulting associated recommendation list is an objective data reference output after quantitatively matching the semantic features of the patient's description text with a standard medical knowledge base. It utilizes natural language processing and machine learning techniques to extract and quantify correlation indicators from the text, aiming to assist physicians in more efficiently sorting through patient complaints and focusing on potential key pathological features. The associated recommendation list only provides structured and interpretable references for doctors to make professional judgments; the final prescription for promoting blood circulation and removing blood stasis must be independently prescribed by a doctor based on a comprehensive clinical examination.

[0071] This invention is now complete.

[0072] Example 2: Figure 3 This is a schematic diagram of a computer device for a personalized recommendation system of traditional Chinese medicine formulas for promoting blood circulation and removing blood stasis in coronary heart disease, provided in one embodiment of the present invention. For example, as shown... Figure 3 As shown, the computer device includes: a memory 201, a processor 202, and a computer program 203 stored in the memory 201 and running on the processor 202, wherein when the processor 202 executes the computer program 203, the computer device can execute any of the aforementioned personalized recommendation systems for traditional Chinese medicine prescriptions for promoting blood circulation and removing blood stasis in coronary heart disease.

[0073] Furthermore, this application also protects an apparatus that may include a memory and a processor, wherein the memory stores executable program code, and the processor is used to call and execute the executable program code to execute the personalized recommendation system for traditional Chinese medicine blood-activating and stasis-removing prescriptions for coronary heart disease provided in this application.

[0074] This embodiment can divide the device into functional modules based on the above method example. For example, each module can correspond to a separate function, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware. It should be noted that the module division in this embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0075] It should be understood that the device provided in this embodiment is used to execute the above-mentioned personalized recommendation system for traditional Chinese medicine prescriptions for promoting blood circulation and removing blood stasis in coronary heart disease, and therefore can achieve the same effect as the above-mentioned implementation method.

[0076] When using integrated units, the device may include a processing module and a storage module. When applied to a workpiece, the processing module can be used to control and manage the workpiece's operations. The storage module can be used to support the execution of program code by the workpiece.

[0077] The processing module may be a processor or a controller, which can implement or execute various exemplary logic blocks, modules, and circuits as disclosed in this application. The processor may also be a combination of computing functions, such as a combination of one or more microprocessors, a combination of digital signal processing (DSP) and microprocessors, etc., and the storage module may be a memory.

[0078] Example 3: This embodiment also provides a computer-readable storage medium storing computer program code. When the computer program code is run on a computer, the computer executes the above-described related method steps to implement the personalized recommendation system for traditional Chinese medicine prescriptions for promoting blood circulation and removing blood stasis in coronary heart disease provided in the above embodiment.

[0079] In this embodiment, the device and computer-readable storage medium are used to execute the corresponding system provided above. Therefore, the beneficial effects that can be achieved can be referred to the beneficial effects of the corresponding system provided above, and will not be repeated here.

[0080] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0081] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0082] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A personalized recommendation system for traditional Chinese medicine formulas for promoting blood circulation and removing blood stasis in patients with coronary heart disease, characterized in that, The system includes: The data acquisition module is used to obtain the set of descriptive word vectors of the user in the current interaction record, the set of feedback word vectors of several historical interaction records, and the set of standard word vectors of the standard semantic term library. The risk analysis module is used to obtain the comprehensive risk coefficient of the descriptive word vector based on the matching reliability of the descriptive word vector, as well as the semantic universality and deviation from the standard semantics of the descriptive word vector, and to select the word vector to be corrected. The spatiotemporal gain analysis module is used to obtain the spatiotemporal correction rationality coefficient between each word vector to be corrected and each standard word vector based on the semantic dispersion of each word vector to be corrected and the other descriptive word vectors relative to the same standard word vector in the standard word vector set, and the changing trend of the correlation between each feedback word vector set and the corresponding standard word vector. The correction module is used to iteratively adjust the matching degree between each word vector to be corrected and each standard word vector according to the spatiotemporal correction rationality coefficient and the comprehensive risk coefficient, until the preset convergence condition is met, and obtain the final matching degree. The recommendation module is used to generate a list of related recommendations for the current interaction record based on the final matching degree.

2. The personalized recommendation system for traditional Chinese medicine formulas for promoting blood circulation and removing blood stasis in coronary heart disease according to claim 1, characterized in that, The process of obtaining the comprehensive risk coefficient of the descriptive word vectors and selecting the word vectors to be corrected includes: Select the standard word vector corresponding to the maximum matching degree between the description word vector and all standard word vectors in the standard word vector set, and denote it as the matching vector of the description word vector; obtain the matching error degree based on the reliability of the matching of the description word vector to its matching vector; The descriptive word vectors in the descriptive word vector set are divided into different clusters; the cluster edge degree is obtained based on the relative position information of the descriptive word vectors in their respective clusters and the semantic prevalence of the descriptive word vectors. The arithmetic mean of the matching error and the clustering edge degree is used as the comprehensive risk coefficient for describing word vectors; Select descriptive word vectors from the descriptive word vector set whose comprehensive risk coefficient is greater than the preset risk threshold, and denote them as word vectors to be corrected.

3. The personalized recommendation system for traditional Chinese medicine formulas for promoting blood circulation and removing blood stasis in coronary heart disease according to claim 2, characterized in that, The process of obtaining the matching error degree includes: Calculate the degree of matching between the description word vector and the remaining standard word vectors in the standard word vector set, excluding its matching vector. Select the standard word vector with the highest degree of matching and denote it as the reliable reference vector. The ratio of the degree of matching between the descriptive word vector and its reliable reference vector, and the degree of matching between the matching vectors, is denoted as the similarity ratio. The degree of matching between the matching vector of the descriptive word vector and the reliable reference vector is denoted as semantic proximity. The semantic proximity is negatively correlated, and the product of the mapping result and the similarity ratio is used as the matching error of the descriptive word vector.

4. The personalized recommendation system for traditional Chinese medicine formulas for promoting blood circulation and removing blood stasis in coronary heart disease according to claim 2, characterized in that, The process of obtaining the cluster edge degree includes: The ratio of the distance between the descriptive word vector and the cluster center of its cluster is used as the numerator, and the cluster radius of the cluster in which the descriptive word vector is located is used as the intra-cluster eccentricity of the descriptive word vector. A negative correlation mapping is performed on the cluster size of the cluster in which the descriptive word vector is located, and the product of the mapping result and the eccentricity within the cluster is used as the cluster edge degree of the descriptive word vector.

5. A personalized recommendation system for traditional Chinese medicine formulas for promoting blood circulation and removing blood stasis in coronary heart disease according to claim 1, characterized in that, The process of obtaining the spatiotemporal correction coefficients for each descriptive word vector and each standard word vector includes: Choose one word vector to be corrected and denote it as the description example vector; choose one word vector from the standard word vector set and denote it as the standard example vector. Calculate the matching degree between the description word vectors (excluding the description example vectors) and the standard example vectors in the description word vector set. Average the differences between all matching degrees and the matching degree between the description example vectors and the standard example vectors to obtain the contextual dispersion of the description example vectors and the standard example vectors. Based on the changing trend of the correlation between each set of feedback word vectors and the standard example vectors, the historical trend inertia of the standard example vectors is obtained; The context dispersion is negatively correlated and mapped, and the arithmetic mean of the mapping result and the historical trend inertia is used as the spatiotemporal correction coefficient describing the example vector and the standard example vector.

6. A personalized recommendation system for traditional Chinese medicine formulas for promoting blood circulation and removing blood stasis in coronary heart disease according to claim 5, characterized in that, The process of obtaining the historical trend inertia of the standard example vectors includes: The average matching degree between all feedback word vectors in each feedback word vector set and the standard example vectors is used to obtain the feedback relevance. Arrange all feedback word vector sets with the feedback correlation degree of the standard example vectors in reverse chronological order according to the historical interaction records to obtain the feedback correlation sequence; Obtain the first-order difference sequence of the feedback correlation sequence, and select the element with the smallest subscript value from the two adjacent elements corresponding to the first non-negative number in the feedback correlation sequence as the effective feedback cutoff element; The ratio of the difference between the first element in the feedback association sequence and the effective feedback cutoff element to the time interval between the corresponding historical interaction records of the two elements is used as the historical trend inertia of the standard example vector.

7. A personalized recommendation system for traditional Chinese medicine formulas for promoting blood circulation and removing blood stasis in coronary heart disease according to claim 2, characterized in that, The step of dividing the descriptive word vectors in the descriptive word vector set into different clusters includes: The feature n-tuple of the description word vector is composed of the description word vector, the matching vector of the description word vector, and the degree of matching between the description word vector and its matching vector; Based on the distance between the feature n-tuples of different descriptive word vectors, the descriptive word vectors in the descriptive word vector set are clustered to obtain different clusters.

8. The personalized recommendation system for traditional Chinese medicine formulas for promoting blood circulation and removing blood stasis in coronary heart disease according to claim 1, characterized in that, Obtaining the final matching degree includes: The normalized result of the product of the comprehensive risk coefficient and the spatiotemporal correction reasonable coefficient is used as the matching correction coefficient between each word vector to be corrected and each standard word vector. Calculate the sum of the matching correction coefficient and constant 1 between each word vector to be corrected and each standard word vector, and multiply the sum by the matching degree of the corresponding two word vectors as the corrected matching degree of the corresponding two word vectors; Based on the matching degree between the word vector to be corrected and all standard word vectors after correction, the matching vector and feature n-tuple of the word vector to be corrected are updated; and based on the updated feature n-tuple of the word vector to be corrected, the set of descriptive word vectors is re-divided into different clusters to obtain the updated clusters. Based on the updated clusters, the comprehensive risk coefficient of the descriptive word vectors is re-obtained, and the updated word vectors to be corrected are selected. When the updated word vectors to be corrected do not meet the convergence condition and the number of iterations has not reached the preset iteration threshold, the matching correction coefficient between the updated word vectors to be corrected and each standard word vector is recalculated, and the matching degree between the corresponding two word vectors is further corrected. This process continues until the updated word vectors to be corrected meet the convergence condition or the number of iterations reaches the preset iteration threshold. The current corrected matching degree is then used as the final matching degree between each descriptive word vector and each standard word vector. The convergence condition includes: the proportion of the updated word vectors to be corrected in the set of descriptive word vectors is less than a preset threshold.

9. A personalized recommendation system for traditional Chinese medicine formulas for promoting blood circulation and removing blood stasis in coronary heart disease according to claim 1, characterized in that, The step of generating a related recommendation list for the current interaction record based on the final matching degree includes: For each descriptive word vector, the maximum value among the final matching degrees of the descriptive word vector with all standard word vectors is selected. When the maximum value is greater than the preset effective matching threshold, the drug associated with the standard word vector corresponding to the maximum value is used as the recommended drug. The recommended drugs in the descriptive word vector set are sorted from high to low according to the maximum value corresponding to the descriptive word vector, and an associated recommendation list for the current interaction record is generated.

10. A personalized recommendation system for traditional Chinese medicine formulas for promoting blood circulation and removing blood stasis in coronary heart disease according to claim 1, characterized in that, The degree of matching between a word vector and a standard word vector is described by the cosine similarity between the two word vectors.

Images