Postoperative home rehabilitation exercise recommendation method for patient with head and neck tumors
By calculating the vocabulary characteristic value and hierarchical clustering of the medical records of head and neck tumor patients, determining the characteristic words and importance values, and recommending the movement in the optimal clustering cluster, solving the inaccurate cluster caused by complex medical record information in the prior art. Problem, accurate recommendations for postoperative rehabilitation exercises for patients with head and neck tumors have been achieved, and the recovery effect has been improved.
Patent Information
- Application Number
- CN202411142564.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-20
- Publication Date
- 2024-11-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Among the recommended methods for postoperative home rehabilitation exercises for patients with head and neck tumors, the existing technology has inaccurate clustering categories and unreasonable recommended exercises, which affects the patient's recovery effect.
By obtaining the medical record text of patients with head and neck tumors, splitting it into vocabulary and calculating the vocabulary feature value, determining feature words, performing hierarchical clustering, obtaining the importance of clustering trees and feature words, calculating feature words weights, and accurately dividing data samples. , recommend the motion in the optimal cluster cluster.
It improves the accuracy and efficiency of clustering, ensures the accurate dimensionality reduction and clustering effect of characteristic words, provides the most suitable postoperative home rehabilitation exercise recommendations for patients with head and neck tumors, and promotes effective recovery of patients.
Smart Images

Figure CN118942627A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of electronic digital data processing, and in particular to a method for recommending home rehabilitation exercises for patients with head and neck tumors after surgery. Background Art
[0002] Head and neck tumors refer to tumors that occur in the tissues of the head and neck, including tumors in the oral cavity, pharynx, larynx, nasal cavity, paranasal sinuses, maxillofacial area, etc. The tumors are treated through surgical resection, chemotherapy, and targeted therapy. In order to ensure that patients with head and neck tumors get better recovery effects after surgery, they need to perform postoperative home rehabilitation exercises. Because each head and neck tumor patient has different conditions, the choice of postoperative home rehabilitation exercises for each head and neck tumor patient should be in line with the actual situation.
[0003] In order to make reasonable recommendations for postoperative home rehabilitation exercises for patients with head and neck tumors, existing methods cluster the medical records of a large number of head and neck tumor rehabilitation patients, obtain the medical record categories of different head and neck tumor rehabilitation patients, determine the medical record category to which the current head and neck tumor patient's medical record belongs, and then use the postoperative home rehabilitation exercises of head and neck tumor rehabilitation patients within the corresponding medical record category as the postoperative recommended exercises for the current head and neck tumor patient. However, in actual situations, because the medical record information of head and neck tumor rehabilitation patients is too extensive and complex, the medical record categories obtained by clustering are inaccurate, which in turn leads to unreasonable home rehabilitation exercises recommended for postoperative head and neck tumor patients, which makes head and neck tumor patients unable to recover well after surgery and causes harm to their bodies. Summary of the invention
[0004] In order to solve the technical problem that the medical record information of head and neck tumor rehabilitation patients is too extensive and complex, resulting in inaccurate medical record categories obtained by clustering, and then leading to unreasonable home rehabilitation exercises recommended for head and neck tumor patients after surgery, the purpose of the present invention is to provide a method for recommending home rehabilitation exercises for head and neck tumor patients after surgery. The technical scheme adopted is as follows:
[0005] The present invention proposes a method for recommending home rehabilitation exercises for patients with head and neck tumors after surgery, the method comprising the following steps:
[0006] Obtain the medical records of patients who have recovered from head and neck tumors and current patients with head and neck tumors, split the text of each medical record into words, and obtain the word feature vector of each medical record; take the medical record of each patient who has recovered from head and neck tumors as a data sample;
[0007] According to the frequency of each word appearing in each data sample and the frequency of each word appearing in all data samples, the characteristic value of each word in each data sample is obtained to determine the characteristic word of each data sample;
[0008] Perform hierarchical clustering on the data samples according to the characteristic value of each characteristic word in each data sample to obtain a clustering tree for each characteristic word; obtain the importance value of each characteristic word according to the similarity between each node in the clustering tree and each other node in its hierarchy, as well as the similarity between its child nodes;
[0009] According to the importance value of each feature word and the feature value in each data sample, as well as the association between the lexical feature vectors of any two data samples, the weight of each feature word is obtained; according to the weight of each feature word and the number of occurrences in each data sample, the participation vector of each data sample is obtained, and the data samples are hierarchically clustered to obtain the target clustering tree; according to the participation vector of the data sample in each node in each level of the target clustering tree, the optimal clustering cluster is obtained;
[0010] According to the similarity between the lexical feature vectors of the medical records of the current patients with head and neck tumors and the lexical feature vectors of each data sample in each optimal clustering cluster, the recommended exercises for postoperative home rehabilitation of the current patients with head and neck tumors are obtained.
[0011] Furthermore, the calculation formula of the characteristic value is:
[0012]
[0013] Where, T a,b is the feature value of the bth word in the ath data sample; n a,b is the number of times the bth word appears in the ath data sample; N a is the total number of words in the ath data sample; m b is the number of data samples with the bth word; M is the total number of data samples; is the frequency of the bth word in the ath data sample; is the frequency of the bth word in all data samples; norm is the normalization function; e is a natural constant.
[0014] Furthermore, the method for determining the characteristic words of each data sample is:
[0015] When the feature value is greater than a preset feature value threshold, the corresponding word is used as a feature word of the corresponding data sample.
[0016] Furthermore, the method for obtaining the importance value of each feature word according to the similarity between each node in the clustering tree and each other node in the layer where it is located, and the similarity between its child nodes is as follows:
[0017] For the i-th node in the clustering tree of the k-th feature word, other nodes in the level where the i-th node is located are used as reference nodes;
[0018] When the ith node has a child node, and the child node of the ith node has a child node, the vocabulary feature vectors of all data samples in the ith node, the child node of the ith node, and each reference node are processed respectively by the average pooling algorithm to obtain the overall vocabulary feature vectors of the ith node, the child node of the ith node, and each reference node;
[0019] Obtaining a clustering effect value of the i-th node according to the cosine similarity of the overall vocabulary feature vectors between the i-th node and each reference node, and the cosine similarity of the overall vocabulary feature vectors between the child nodes of the i-th node;
[0020] When the ith node has a child node, and the child node of the ith node has no child node, the lexical feature vectors of all data samples in the ith node and each reference node are processed respectively by the average pooling algorithm to obtain the overall lexical feature vectors of the ith node and each reference node;
[0021] Obtaining a clustering effect value of the i-th node according to the cosine similarity of the overall lexical feature vectors between the i-th node and each reference node, and the cosine similarity of the lexical feature vectors between the child nodes of the i-th node;
[0022] When the i-th node does not have a child node, the clustering effect value of the i-th node is set to a first preset constant; wherein the first preset constant is a non-negative number;
[0023] The result of accumulating the clustering effect value of each node in the clustering tree of the k-th feature word is used as the importance value of the k-th feature word.
[0024] Furthermore, the calculation formula of the clustering effect value is:
[0025] When the i-th node in the clustering tree of the k-th feature word has a child node, and the child node of the i-th node has a child node:
[0026]
[0027] When the i-th node in the clustering tree of the k-th feature word has a child node, and the child node of the i-th node does not have a child node:
[0028]
[0029] When the i-th node in the clustering tree of the k-th feature word has no child nodes:
[0030] G k,i =δ
[0031] In the formula, G k,iis the clustering effect value of the i-th node in the clustering tree of the k-th feature word; J is the total number of reference nodes corresponding to the i-th node; X k,i ′ is the overall vocabulary feature vector of the i-th node in the clustering tree of the k-th feature word; X k,j ′ is the overall vocabulary feature vector of the jth reference node corresponding to the i-th node in the clustering tree of the k-th feature word; X k,i1 ′ is the overall vocabulary feature vector of the first child node of the i-th node in the clustering tree of the k-th feature word; X k,i2 ′ is the overall vocabulary feature vector of the second child node of the i-th node in the clustering tree of the k-th feature word; X k,i1 is the vocabulary feature vector of the first child node of the i-th node in the clustering tree of the k-th feature word; X k,i2 is the lexical feature vector of the second child node of the ith node in the clustering tree of the kth feature word; δ is a first preset constant, δ≥0; α is a second preset constant, greater than 0; |||| is the vector modulus symbol.
[0032] Furthermore, the calculation formula of the weight is:
[0033]
[0034] Where W k is the weight of the k-th feature word; S k is the importance value of the kth feature word; Q is the total number of data samples; T q,k is the characteristic value of the k-th characteristic word in the q-th data sample; R q,v is the correlation coefficient of the vocabulary feature vectors of the qth data sample and the vth data sample; β is the third preset constant, which is greater than 0; || is the absolute value function; and norm is the normalization function.
[0035] Furthermore, the method of obtaining the participation vector of each data sample according to the weight of each feature word and the number of times it appears in each data sample, performing hierarchical clustering on the data samples, and obtaining the target clustering tree is as follows:
[0036] Get the mean of the weights of all feature words as the standard weight;
[0037] When the weight is greater than or equal to the standard weight, the feature word of the corresponding category is used as the target feature word;
[0038] The product of the number of occurrences of each target feature word in each data sample and the weight of the corresponding target feature word is used as the adjusted number of each target feature word in each data sample;
[0039] The sum of the adjustment times and the appearance times of each target feature word in each data sample is taken as the correction times of each target feature word in each data sample;
[0040] The number of corrections corresponding to each target feature word in each data sample is constructed into a vector as the participation vector of each data sample;
[0041] According to the cosine similarity between any two participating vectors, the data samples are clustered through the hierarchical clustering algorithm to obtain the target clustering tree.
[0042] Furthermore, the method for obtaining the optimal clustering cluster according to the participation vector of the data sample in each node in each level of the target clustering tree is:
[0043] For any level of the target clustering tree, obtain the difference between the total variance and the intra-class variance of the participation vectors of the data samples in all nodes in the level as the inter-class variance of the participation vectors of the data samples in all nodes in the level;
[0044] Each node in the layer corresponding to the largest inter-class variance is regarded as the optimal clustering cluster.
[0045] Furthermore, the method for obtaining the recommended exercise for postoperative home rehabilitation of the current patient with head and neck tumors based on the similarity between the vocabulary feature vector of the medical record of the current patient with head and neck tumors and the vocabulary feature vector of each data sample in each optimal cluster is:
[0046] According to the cosine similarity between the lexical feature vector of the medical record of the current patient with head and neck tumor and the overall lexical feature vector of each optimal clustering cluster, the belonging degree value of the medical record of the current patient with head and neck tumor and each optimal clustering cluster is obtained;
[0047] The optimal cluster corresponding to the maximum belonging degree value is taken as the target cluster;
[0048] Obtain the cosine similarity between the vocabulary feature vector of the medical record of the current patient with head and neck tumors and the vocabulary feature vector of each data sample in the target cluster as a reference value;
[0049] When the reference value is greater than the preset reference value threshold, the corresponding data sample is used as a reference sample;
[0050] The exercises during the rehabilitation process of the reference sample are used as the recommended exercises for postoperative home rehabilitation of current patients with head and neck tumors.
[0051] Furthermore, the calculation formula of the attribution degree value is:
[0052]
[0053] Where, Lu is the degree of belonging between the medical records of the current patient with head and neck tumor and the uth optimal cluster; X is the vocabulary feature vector of the medical records of the current patient with head and neck tumor; X u ′ is the overall vocabulary feature vector of the u-th optimal cluster; |||| is the vector modulus symbol; γ is the fourth preset constant, which is greater than 0.
[0054] The present invention has the following beneficial effects:
[0055] According to the frequency of each word in each data sample and the frequency of each word in all data samples, the characteristic value of each word in each data sample is obtained, the characteristic word of each data sample is accurately determined, and the accuracy of hierarchical clustering of data samples is improved; in order to avoid too many types of characteristic words, which leads to inaccurate and inefficient results of hierarchical clustering of data samples based on characteristic words, the data samples are hierarchically clustered according to the characteristic value of each characteristic word in each data sample, and the clustering tree of each characteristic word is obtained, so as to facilitate accurate analysis of the clustering effect of each characteristic word separately, and prepare for the subsequent accurate dimensionality reduction of the types of characteristic words; and then according to the similarity between each node in the clustering tree and each other node in its level, as well as the similarity between its child nodes, the clustering effect of each characteristic word is determined, and then the importance value of each characteristic word is accurately obtained, so as to accurately obtain the weight of each characteristic word in the subsequent accurate acquisition; further, according to the importance value of each characteristic word and the characteristic value in each data sample, and The lexical feature vectors of any two data samples are associated, the weight of each feature word is obtained, and the types of feature words are accurately reduced in dimension. Therefore, according to the weight of each feature word and the number of occurrences in each data sample, the participation vector of each data sample is accurately obtained, the characteristics of each data sample are accurately and concisely represented, and the data samples are accurately and efficiently hierarchically clustered to obtain the target clustering tree; in order to determine the best division of the data samples, the optimal clustering clusters are accurately obtained according to the participation vectors of the data samples in each node in each level of the target clustering tree, and the data samples are accurately divided; according to the similarity between the lexical feature vectors of the medical records of the current patients with head and neck tumors and the lexical feature vectors of each data sample in each optimal clustering cluster, the optimal clustering cluster to which the medical records of the current patients with head and neck tumors belong is accurately obtained, and then the data samples with the most reference value for the medical records of the current patients with head and neck tumors are accurately obtained, and the recommended exercises for postoperative home rehabilitation of the current patients with head and neck tumors are accurately obtained, so that the current patients with head and neck tumors can recover better after surgery. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0057] Figure 1 A flowchart of a method for recommending home rehabilitation exercises for patients with head and neck tumors after surgery is provided in one embodiment of the present invention. DETAILED DESCRIPTION
[0058] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of a recommended method of postoperative home rehabilitation exercise for head and neck tumor patients proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.
[0059] Unless defined otherwise, 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 belongs.
[0060] The specific scheme of the recommended method for postoperative home rehabilitation exercise for patients with head and neck tumors provided by the present invention is described in detail below with reference to the accompanying drawings.
[0061] See also Figure 1 , which shows a flow chart of a method for recommending postoperative home rehabilitation exercises for patients with head and neck tumors provided by an embodiment of the present invention, the method comprising the following steps:
[0062] Step S1: Obtain the medical records of patients who have recovered from head and neck tumors and current patients with head and neck tumors, split the text of each medical record into words, and obtain the word feature vector of each medical record; take the medical record of each patient who has recovered from head and neck tumors as a data sample.
[0063] Specifically, a large number of medical records of patients with head and neck tumor rehabilitation are obtained by accessing open databases or purchasing in the data set market, wherein the medical record content includes information such as treatment and rehabilitation process, postoperative home rehabilitation exercises, etc. The embodiment of the present invention sets the number of medical records of patients with head and neck tumor rehabilitation to 500. The implementer can set the number of medical records of patients with head and neck tumor rehabilitation according to actual conditions, which is not limited here. Taking the acquisition of recommended postoperative rehabilitation exercises for current patients with head and neck tumors as an example, the medical records of current patients with head and neck tumors need to be acquired. In order to perform subsequent analysis through medical records, it is first necessary to split the text of each medical record into words through the NLPIR algorithm, and then obtain the vocabulary feature vector of each medical record through the bag-of-words model. Among them, the NLPIR algorithm and the bag-of-words model are both well-known technologies and will not be described in detail. In order to explain the embodiment of the present invention more clearly in the future, the medical record of each patient with head and neck tumor rehabilitation is taken as a data sample.
[0064] The specific scenario of the embodiment of the present invention is that there are no abnormal data samples such as duplicate data samples and blank data samples.
[0065] The purpose of the embodiment of the present invention is: when the medical records of the head and neck tumor rehabilitation patients are divided by the hierarchical clustering algorithm, the medical records of the head and neck tumor rehabilitation patients are not accurately divided due to the complex medical record information of the head and neck tumor rehabilitation patients. When the medical records of the current head and neck tumor patients are divided into a certain medical record category, it is impossible to accurately recommend the postoperative home rehabilitation exercises of the head and neck tumor rehabilitation patients corresponding to the medical record category. In order to accurately recommend the postoperative home rehabilitation exercises of the current head and neck tumor patients, it is necessary to accurately cluster the medical records of the head and neck tumor rehabilitation patients. Therefore, the embodiment of the present invention obtains the characteristic words of the medical records, analyzes each characteristic word, obtains the weight of each characteristic word, and then reduces the dimension of the characteristic words of the medical records. According to the characteristic words after the dimension reduction, the medical records of the head and neck tumor rehabilitation patients are accurately and efficiently clustered. By determining the medical record category to which the medical records of the current head and neck tumor patients belong, the recommendation of the postoperative home rehabilitation exercises of the current head and neck tumor patients is accurately obtained, so that the current head and neck tumor patients can recover better after surgery. Among them, the hierarchical clustering algorithm is a well-known technology and will not be described in detail.
[0066] Step S2: According to the frequency of each word appearing in each data sample and the frequency of each word appearing in all data samples, the feature value of each word in each data sample is obtained to determine the feature word of each data sample.
[0067] Specifically, there are multiple words in each data sample. Hierarchical clustering of the data samples directly based on the words of the data samples is likely to cause errors in the clustering results and generate a large amount of calculations. It is known that when a certain word appears more times in a certain data sample, but the fewer times it appears in all data samples, it means that the higher the degree of expression of the word to the data sample, the more representative the word is of the data sample. Therefore, the embodiment of the present invention counts the number of times each word appears in each data sample and the number of times it appears in all data samples, obtains the frequency of each word appearing in each data sample and the frequency of each word appearing in all data samples, and then determines the characteristic words of each data sample. It should be noted that the number of times a certain word appears in all data samples is the number of data samples containing the word. Then, according to the frequency of each word appearing in each data sample and the frequency of each word appearing in all data samples, the characteristic value of each word in each data sample is obtained. When the characteristic value is larger, it means that the corresponding word is more likely to be the characteristic word of the corresponding data sample.
[0068] As an example, take the bth word in the ath data sample as an example. In order to avoid mistaking common words that appear multiple times in each data sample and are meaningless to the characteristics of the data sample for characteristic words, the frequency of the bth word in all data samples is obtained. The greater the frequency of the bth word in all data samples, the more likely the bth word is to be a common word. Therefore, the calculation formula for obtaining the characteristic value of the bth word in the ath data sample is:
[0069]
[0070] Where, T a,b is the feature value of the bth word in the ath data sample; n a,b is the number of times the bth word appears in the ath data sample; N a is the total number of words in the ath data sample; m b is the number of data samples with the bth word; M is the total number of data samples; is the frequency of the bth word in the ath data sample; is the frequency of the bth word in all data samples; norm is the normalization function; e is a natural constant.
[0071] It should be noted that The bigger, The smaller it is, the more times the b-th word appears in the a-th data sample, and the fewer times it appears in all data samples, the more the b-th word can represent the characteristics of the a-th data sample. The bigger, The larger the Ta,b The larger the a,b The larger the bth word is, the more likely it is that the bth word is the feature word of the ath data sample.
[0072] According to the method for obtaining the feature value of the bth word in the ath data sample, the feature value of each word in each data sample is obtained.
[0073] It is known that the larger the characteristic value, the more likely the corresponding word is the characteristic word of the corresponding data sample. Therefore, the embodiment of the present invention sets the preset characteristic value threshold to 0.6. The implementer can set the size of the preset characteristic value threshold according to the actual situation, which is not limited here. When the characteristic value is greater than the preset characteristic value threshold, the corresponding word is used as the characteristic word of the corresponding data sample. For example, when the characteristic value of the b-th word in the a-th data sample is greater than the preset characteristic value threshold, the b-th word is used as the characteristic word of the a-th data sample. At this point, the characteristic word of each data sample is determined.
[0074] Step S3: hierarchically cluster the data samples according to the characteristic value of each characteristic word in each data sample to obtain a clustering tree for each characteristic word; obtain the importance value of each characteristic word according to the similarity between each node in the clustering tree and each other node in its hierarchy, as well as the similarity between its child nodes.
[0075] Specifically, due to the influence of external factors such as the physical condition, living environment, and work and rest rules of different head and neck tumor rehabilitation patients, there are differences between data samples. Therefore, the feature words between each data sample are not exactly the same. Among them, there are some feature words that can only meet the relatively broad category division, which is not conducive to the accurate division of data samples, and at the same time affects the efficiency of the hierarchical clustering algorithm. Therefore, according to the size of the feature value of each feature word in each data sample, the embodiment of the present invention performs hierarchical clustering of the split method on all data samples to obtain a clustering tree of each feature word, wherein one feature word corresponds to one clustering tree. It should be noted that when a certain data sample does not contain a certain feature word, the feature value of the feature word in the data sample is defaulted to 0. When the difference between a node in the clustering tree and each other node in the level where it is located is greater, it means that the data sample in the node is divided more accurately. At the same time, when the subnodes of a node are more similar, it also means that the data sample in the node is divided more accurately, that is, the clustering effect of the node is better. When the clustering effect of each node in the clustering tree of a certain feature word is better, it means that the feature word is more conducive to accurately clustering the data sample, and the feature word is more important. Therefore, the embodiment of the present invention obtains the importance value of each feature word based on the similarity between each node in the clustering tree and each other node in its level, as well as the similarity between its child nodes. Among them, the larger the importance value, the greater the reference significance of the corresponding feature word for dividing the data sample.
[0076] Preferably, the method for obtaining the importance value is: for the i-th node in the clustering tree of the k-th feature word, other nodes in the level where the i-th node is located are used as reference nodes; since a node may contain multiple data samples, in order to reduce the amount of calculation, the embodiment of the present invention processes the vocabulary feature vectors of all data samples in the node containing at least two data samples through an average pooling algorithm to obtain an overall vocabulary feature vector of the corresponding node. Therefore, when the i-th node has a child node, and the child node of the i-th node has a child node, the vocabulary feature vectors of all data samples in the i-th node, the child node of the i-th node and each reference node are processed respectively by the average pooling algorithm to obtain the overall vocabulary feature vector of the i-th node, the child node of the i-th node and each reference node; it is known that when the cosine similarity of the overall feature vector or the vocabulary feature vector between two nodes is larger, it means that the similarity between the two nodes is higher. Therefore, according to the cosine similarity of the overall lexical feature vectors between the ith node and each reference node, and the cosine similarity of the overall lexical feature vectors between the child nodes of the ith node, the clustering effect value of the ith node is obtained; when the ith node has a child node, and the child node of the ith node does not have a child node, the lexical feature vectors of all data samples in the ith node and each reference node are processed respectively by the average pooling algorithm to obtain the overall lexical feature vector of the ith node and each reference node; according to the cosine similarity of the overall lexical feature vectors between the ith node and each reference node, and the cosine similarity of the lexical feature vectors between the child nodes of the ith node, the clustering effect value of the ith node is obtained; when the ith node does not have a child node, the clustering effect value of the ith node is set to a first preset constant; wherein the first preset constant is a non-negative number; the result of accumulating the clustering effect value of each node in the clustering tree of the kth feature word is used as the importance value of the kth feature word. Among them, the average pooling algorithm and the method for obtaining the cosine similarity are both well-known technologies and will not be described in detail.
[0077] As an example, taking the kth feature word as an example, the specific method for obtaining the importance value of the kth feature word is as follows:
[0078] (1) Obtain the clustering effect value.
[0079] As an example, taking the i-th node in the clustering tree of the k-th feature word as an example, the calculation formula for obtaining the clustering effect value of the i-th node in the clustering tree of the k-th feature word is:
[0080] When the i-th node in the clustering tree of the k-th feature word has a child node, and the child node of the i-th node has a child node:
[0081]
[0082] When the i-th node in the clustering tree of the k-th feature word has a child node, and the child node of the i-th node does not have a child node:
[0083]
[0084] When the i-th node in the clustering tree of the k-th feature word has no child nodes:
[0085] G k,i =δ
[0086] In the formula, G k,i is the clustering effect value of the i-th node in the clustering tree of the k-th feature word; J is the total number of reference nodes corresponding to the i-th node; X k,i ′ is the overall vocabulary feature vector of the i-th node in the clustering tree of the k-th feature word; X k,j ′ is the overall vocabulary feature vector of the jth reference node corresponding to the i-th node in the clustering tree of the k-th feature word; X k,i1 ′ is the overall vocabulary feature vector of the first child node of the i-th node in the clustering tree of the k-th feature word; X k,i2 ′ is the overall vocabulary feature vector of the second child node of the i-th node in the clustering tree of the k-th feature word; X k,i1 is the vocabulary feature vector of the first child node of the i-th node in the clustering tree of the k-th feature word; X k,i2 is the lexical feature vector of the second child node of the ith node in the clustering tree of the kth feature word; δ is a first preset constant, δ≥0; α is a second preset constant, greater than 0; |||| is the vector modulus symbol.
[0087] In the embodiment of the present invention, δ is set to 1. The implementer can set the size of δ according to the actual situation, which is not limited here. α is set to 0.1 to avoid the denominator being 0. The implementer can set the size of α according to the actual situation, which is not limited here.
[0088] It should be noted that The larger it is, the more similar the i-th node in the clustering tree of the k-th feature word is to its corresponding j-th reference node. The smaller, The smaller it is, the smaller the difference between the i-th node in the clustering tree of the k-th feature word and each corresponding reference node is, which indirectly indicates that the data sample division in the i-th node in the clustering tree of the k-th feature word is more inaccurate, and G k,i When the i-th node in the clustering tree of the k-th feature word has a child node, and the child node of the i-th node has a child node, The smaller it is, the smaller the similarity between the child nodes of the ith node in the clustering tree of the kth feature word is, and the less accurate the data sample division in the ith node in the clustering tree of the kth feature word is. k,i When the i-th node in the clustering tree of the k-th feature word has a child node, and the child node of the i-th node does not have a child node, The smaller it is, the smaller the similarity between the child nodes of the ith node in the clustering tree of the kth feature word is, and the less accurate the data sample division in the ith node in the clustering tree of the kth feature word is. k,i Therefore, G k,i The smaller it is, the worse the clustering effect of the i-th node in the clustering tree of the k-th feature word is, which indirectly means that the k-th feature word is less conducive to the accurate division of data samples. In particular, when the i-th node in the clustering tree of the k-th feature word does not have a child node, it is directly defaulted to G k,i is 1.
[0089] According to the method for obtaining the clustering effect value of the i-th node in the clustering tree of the k-th feature word, the clustering effect value of each node in the clustering tree of the k-th feature word is obtained.
[0090] (2) Obtain importance value.
[0091] When the clustering effect value of each node in the clustering tree of the k-th feature word is larger, it means that the k-th feature word can divide the data sample more accurately and the k-th feature word is more important. Therefore, the clustering effect value of each node in the clustering tree of the k-th feature word is accumulated to obtain the calculation formula of the importance value of the k-th feature word:
[0092]
[0093] In the formula, S k is the importance value of the k-th feature word; I is the total number of nodes in the clustering tree of the k-th feature word, excluding the root node; G k,i is the clustering effect value of the i-th node in the clustering tree of the k-th feature word.
[0094] It should be noted that S k The larger it is, the better the clustering effect of the clustering tree of the k-th feature word is, and the more conducive the k-th feature word is to improving the accuracy of data sample division.
[0095] According to the method for obtaining the importance value of the k-th feature word, the importance value of each feature word is obtained.
[0096] Step S4: Obtain the weight of each feature word according to the importance value of each feature word and the feature value in each data sample, as well as the association between the lexical feature vectors of any two data samples; obtain the participation vector of each data sample according to the weight of each feature word and the number of occurrences in each data sample, perform hierarchical clustering on the data samples, and obtain a target clustering tree; obtain the optimal clustering cluster according to the participation vector of the data sample in each node in each level of the target clustering tree.
[0097] Specifically, when the importance value of a certain feature word is larger, it means that the feature word is more conducive to the accurate division of data samples. In order to improve the clustering effect of data samples, the weight of the feature word when participating in the division of data samples is larger. When the characteristic value of a certain feature word in a certain data sample is larger, it means that the feature word can better express the characteristics of the data sample. When the data sample is divided, the weight of the feature word in the data sample is larger, which indirectly means that the feature word is more conducive to the accurate division of data samples. The weight of the feature word when participating in the division of data samples is larger. When the association between a certain data sample and other data samples is smaller, it means that the data sample is more special. In order to ensure that the data sample is accurately divided, the feature word of the data sample participates in the division of the data sample. The greater the weight. Therefore, the embodiment of the present invention obtains the weight of each feature word according to the importance value of each feature word and the characteristic value in each data sample, as well as the association between the vocabulary feature vectors of any two data samples. Among them, the larger the weight, the greater the participation ratio of the feature word of the corresponding type in the accurate division of data samples. In order to reduce the amount of calculation in the hierarchical clustering process and perform accurate and efficient hierarchical clustering on data samples, the embodiment of the present invention reduces the dimension of the types of feature words according to the weight of each feature word, obtains the participation vector of each data sample through the weight of each feature word after dimensionality reduction and the number of occurrences in each data sample, and clusters the data samples through the hierarchical clustering algorithm based on the similarity between the participation vectors, and obtains the final clustering tree, which is the target clustering tree. It is known that when the difference between each node in a certain level of the target clustering tree is greater, but the data samples in each node are more similar, the clustering effect of the level is better, and each node in the level is more likely to be the optimal clustering cluster. Therefore, the embodiment of the present invention analyzes the difference between the lexical feature vectors of each node in each level of the target clustering tree, and the difference between the lexical feature vectors in each node in each level, determines the level with the best clustering effect, and then obtains each optimal clustering cluster with the most accurate division. Among them, the specific method for obtaining the optimal clustering cluster is as follows:
[0098] (1) Obtain weights.
[0099] As an example, taking the kth feature word in step S3 as an example, the embodiment of the present invention obtains the correlation coefficient of the vocabulary feature vectors of any two data samples to represent the association between any two data samples. The method for obtaining the correlation coefficient is a well-known technology and will not be described in detail. The calculation formula for obtaining the weight of the kth feature word is:
[0100]
[0101] Where W k is the weight of the k-th feature word; S k is the importance value of the kth feature word; Q is the total number of data samples; T q,k is the characteristic value of the k-th characteristic word in the q-th data sample; r q,v is the correlation coefficient of the vocabulary feature vectors of the qth data sample and the vth data sample; β is the third preset constant, which is greater than 0; | | is the absolute value function; and norm is the normalization function.
[0102] In the embodiment of the present invention, β is set to 0.1 to avoid the denominator being 0. The implementer may set the value of β according to the actual situation, which is not limited here.
[0103] It should be noted that S k The larger the value, the better the clustering effect of the k-th feature word on the data sample. k The bigger the T q,k The larger the value, the more the k-th feature word can represent the characteristics of the q-th data sample, and the greater the weight of the k-th feature word in the data sample division of the q-th data sample. k The larger; |R q,v The smaller | is, the smaller the association between the qth data sample and the vth data sample is, and the greater the weight of the feature word of the qth data sample when participating in the data sample division. k The larger the k The larger it is, the more important the k-th feature word is in data sample division, and the more conducive it is to accurately divide the data sample.
[0104] According to the method for obtaining the weight of the k-th feature word, the weight of each feature word is obtained.
[0105] (2) Obtain the target clustering tree.
[0106] According to the weight of each feature word, the types of feature words are reduced in dimension, that is, the feature words that have an adverse effect on the accurate division of data samples are removed. The feature words after dimensionality reduction can not only accurately cluster data samples, but also improve the efficiency of the hierarchical clustering algorithm. The embodiment of the present invention obtains the mean of the weights of all kinds of feature words as the standard weight; when the weight is less than the standard weight, the feature words of the corresponding type are removed; when the weight is greater than or equal to the standard weight, the feature words of the corresponding type are used as target feature words; therefore, the target feature words are the feature words after dimensionality reduction. It should be noted that for data samples without target feature words, it is of little significance to obtain the recommended reference for postoperative home rehabilitation exercises for current patients with head and neck tumors. Therefore, the data samples without target feature words are deleted, and the data samples that appear subsequently are all data samples containing at least one target feature word.
[0107] In order to accurately and efficiently cluster data samples by target feature words, the embodiment of the present invention uses the product of the number of occurrences of each target feature word in each data sample and the weight of the corresponding target feature word as the adjusted number of each target feature word in each data sample; and uses the sum of the adjusted number and the number of occurrences of each target feature word in each data sample as the corrected number of each target feature word in each data sample. When a certain target feature word does not exist in a data sample, the number of occurrences of the target feature word in the data sample is set to 0.
[0108] As an example, the calculation formula for obtaining the number of corrections of the f-th target feature word in the a-th data sample is:
[0109] y′ a,f =y a,f +y a,f ×W f
[0110] In the formula, y′ a,f is the number of corrections of the f-th target feature word in the a-th data sample; y a,f is the number of occurrences of the f-th target feature word in the a-th data sample; W f is the weight of the f-th target feature word; y a,f ×W f is the number of adjustments of the f-th target feature word in the a-th data sample.
[0111] According to the method for obtaining the number of corrections of the f-th target feature word in the a-th data sample, the number of corrections of each target feature word in each data sample is obtained.
[0112] The number of corrections corresponding to each target feature word in each data sample is constructed into a vector as the participation vector of each data sample; wherein the length of the participation vector of each data sample is equal, and the elements at the same position in each participation vector correspond to the same target feature word.
[0113] Then the cosine similarity between any two participating vectors is obtained, and the data samples are clustered through the hierarchical clustering algorithm to obtain the target clustering tree.
[0114] (3) Obtain the optimal clustering cluster.
[0115] Preferably, the method for obtaining the optimal clustering cluster is: for any level of the target clustering tree, the difference between the total variance and the intra-class variance of the participation vectors of the data samples in all nodes in the level is obtained as the inter-class variance of the participation vectors of the data samples in all nodes in the level; each node in the level corresponding to the largest inter-class variance is taken as the optimal clustering cluster. Among them, the relationship between the total variance, the intra-class variance and the inter-class variance, and the acquisition method are all well-known technologies and will not be described in detail.
[0116] As an example, taking the dth level of the target clustering tree as an example, the calculation formula for obtaining the inter-class variance of the participation vectors of data samples in all nodes in the dth level is:
[0117]
[0118] In the formula, is the inter-class variance of the participation vector of data samples in all nodes in the dth level; σ 2 d is the total variance of the participation vectors of data samples in all nodes in the dth level; is the intra-class variance of the participation vectors of data samples in all nodes in the dth level.
[0119] It should be noted that The smaller it is, the more similar the participation vectors of the data samples in each node in the dth level are, and the more accurate the division of the data samples in each node in the dth level is. The bigger; The larger it is, the greater the difference in participation vectors between each node in the d-th level is, which indirectly indicates that the clustering effect of the d-th level is better.
[0120] According to the method of obtaining the inter-class variance of the participation vectors of the data samples in all the nodes in the dth level, the inter-class variance of the participation vectors of the data samples in all the nodes in each level of the target clustering tree is obtained.
[0121] When the inter-class variance is larger, the difference between each node in the corresponding level is larger, the data samples in each node are more similar, and the clustering effect of the corresponding level is better. Therefore, the level corresponding to the largest inter-class variance is taken as the optimal clustering level; if the largest inter-class variance corresponds to at least two levels, the level with the least number of nodes is taken as the optimal clustering level. Then each node in the optimal clustering level is taken as the optimal clustering cluster. So far, each optimal clustering cluster with the best clustering effect is obtained.
[0122] Step S5: According to the similarity between the vocabulary feature vector of the medical record of the current patient with head and neck tumor and the vocabulary feature vector of each data sample in each optimal cluster, obtain the recommended exercise for postoperative home rehabilitation of the current patient with head and neck tumor.
[0123] Specifically, the similarity between each optimal cluster and the medical records of the current patient with head and neck tumors is analyzed to determine the optimal cluster to which the medical records of the current patient with head and neck tumors belong. Therefore, the embodiment of the present invention obtains the belonging degree value of the medical records of the current patient with head and neck tumors and each optimal cluster according to the cosine similarity between the vocabulary feature vector of the medical records of the current patient with head and neck tumors and the overall vocabulary feature vector of each optimal cluster. Among them, the larger the belonging degree value, the more the medical records of the current patient with head and neck tumors belong to the corresponding optimal cluster.
[0124] As an example, taking the medical records of the current patient with head and neck tumors and the belonging degree value of the u-th optimal cluster as an example, the calculation formula for obtaining the belonging degree value of the medical records of the current patient with head and neck tumors and the u-th optimal cluster is:
[0125]
[0126] Where, L u is the degree of belonging between the medical records of the current patient with head and neck tumor and the uth optimal cluster; X is the vocabulary feature vector of the medical records of the current patient with head and neck tumor; X u ′ is the overall vocabulary feature vector of the u-th optimal cluster; |||| is the vector modulus symbol; γ is the fourth preset constant, which is greater than 0.
[0127] In the embodiment of the present invention, γ is set to 1 to avoid the denominator being 0. The implementer may set the value of γ according to actual conditions, which is not limited here.
[0128] It should be noted that The larger the value is, the more similar the medical records of the current patients with head and neck cancer are to the data samples in the u-th optimal cluster, and the more the medical records of the current patients with head and neck cancer belong to the u-th optimal cluster. The smaller, The larger the L u The larger the uThe larger it is, the more the medical records of the current patient with head and neck cancer belong to the uth optimal cluster.
[0129] According to the method of obtaining the belonging degree value between the medical record of the current patient with head and neck tumor and the u-th optimal cluster, the belonging degree value between the medical record of the current patient with head and neck tumor and each optimal cluster is obtained.
[0130] It is known that the larger the belonging degree value is, the more the medical record of the current patient with head and neck tumor belongs to the corresponding optimal clustering cluster. Therefore, the embodiment of the present invention uses the optimal clustering cluster corresponding to the maximum belonging degree value as the target clustering cluster. Since there are certain differences in the data samples within the same optimal clustering cluster, in order to ensure that the rehabilitation exercise recommended to the current patient with head and neck tumor is more in line with the actual situation, the similarity between each data sample in the target clustering cluster and the medical record of the current patient with head and neck tumor is analyzed, that is, the cosine similarity between the lexical feature vector of the medical record of the current patient with head and neck tumor and the lexical feature vector of each data sample in the target clustering cluster is obtained as a reference value. When the reference value is greater than the preset reference value threshold, the corresponding data sample is used as a reference sample. The embodiment of the present invention sets the preset reference value threshold to 0.9, and the implementer can set the preset reference value threshold according to the actual situation, which is not limited here. Among them, the reference sample is the data sample with the most reference value. Therefore, the exercise in the rehabilitation process of the reference sample is used as the recommended exercise for postoperative home rehabilitation of the current patient with head and neck tumor, and the postoperative home rehabilitation exercise of the current patient with head and neck tumor is accurately recommended.
[0131] So far, the present invention is completed.
[0132] In summary, the embodiment of the present invention obtains the medical records of rehabilitation patients, one medical record is one data sample; obtains the characteristic words of the data samples; hierarchically clusters the data samples according to the characteristic values of each characteristic word to obtain a clustering tree; obtains the importance value according to the clustering effect of the clustering tree; obtains the weight of each characteristic word according to the importance value, the characteristic value and the association between the data sample, and then obtains the participation vector, clusters the data samples, obtains the target clustering tree, and determines the optimal clustering cluster; obtains the recommended exercise for the patient's postoperative home rehabilitation according to the similarity between the patient's medical record and the data sample in the optimal clustering cluster. The present invention accurately obtains the reference weight of each data sample, and then accurately and efficiently hierarchically clusters the data samples to accurately obtain the recommended exercise for the patient's postoperative home rehabilitation.
[0133] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0134] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.
Claims
1. A method for recommending home rehabilitation exercises for patients with head and neck tumors after surgery, characterized in that: The method comprises the following steps: Obtain the medical records of patients who have recovered from head and neck tumors and current patients with head and neck tumors, split the text of each medical record into words, and obtain the word feature vector of each medical record; take the medical record of each patient who has recovered from head and neck tumors as a data sample; According to the frequency of each word appearing in each data sample and the frequency of each word appearing in all data samples, the characteristic value of each word in each data sample is obtained to determine the characteristic word of each data sample; Perform hierarchical clustering on the data samples according to the characteristic value of each characteristic word in each data sample to obtain a clustering tree for each characteristic word; obtain the importance value of each characteristic word according to the similarity between each node in the clustering tree and each other node in its hierarchy, as well as the similarity between its child nodes; According to the importance value of each feature word and the feature value in each data sample, as well as the association between the lexical feature vectors of any two data samples, the weight of each feature word is obtained; according to the weight of each feature word and the number of occurrences in each data sample, the participation vector of each data sample is obtained, and the data samples are hierarchically clustered to obtain the target clustering tree; according to the participation vector of the data sample in each node in each level of the target clustering tree, the optimal clustering cluster is obtained; According to the similarity between the lexical feature vectors of the medical records of the current patients with head and neck tumors and the lexical feature vectors of each data sample in each optimal clustering cluster, the recommended exercises for postoperative home rehabilitation of the current patients with head and neck tumors are obtained.
2. A method for recommending postoperative home rehabilitation exercises for patients with head and neck tumors as claimed in claim 1, characterized in that: The calculation formula of the characteristic value is: Where, T a,b is the feature value of the bth word in the ath data sample; n a,b is the number of times the bth word appears in the ath data sample; N a is the total number of words in the ath data sample; m b is the number of data samples with the bth word; M is the total number of data samples; is the frequency of the bth word in the ath data sample; is the frequency of the bth word in all data samples; norm is the normalization function; e is a natural constant.
3. A method for recommending home rehabilitation exercises for patients with head and neck tumors after surgery as claimed in claim 1, characterized in that: The method for determining the characteristic words of each data sample is: When the feature value is greater than a preset feature value threshold, the corresponding word is used as a feature word of the corresponding data sample.
4. A method for recommending home rehabilitation exercises for patients with head and neck tumors after surgery as claimed in claim 1, characterized in that: The method for obtaining the importance value of each feature word according to the similarity between each node in the clustering tree and each other node in its level, as well as the similarity between its child nodes, is as follows: For the i-th node in the clustering tree of the k-th feature word, other nodes in the level where the i-th node is located are used as reference nodes; When the ith node has a child node, and the child node of the ith node has a child node, the vocabulary feature vectors of all data samples in the ith node, the child node of the ith node, and each reference node are processed respectively by the average pooling algorithm to obtain the overall vocabulary feature vectors of the ith node, the child node of the ith node, and each reference node; Obtaining a clustering effect value of the i-th node according to the cosine similarity of the overall vocabulary feature vectors between the i-th node and each reference node, and the cosine similarity of the overall vocabulary feature vectors between the child nodes of the i-th node; When the ith node has a child node, and the child node of the ith node has no child node, the lexical feature vectors of all data samples in the ith node and each reference node are processed respectively by the average pooling algorithm to obtain the overall lexical feature vectors of the ith node and each reference node; Obtaining a clustering effect value of the i-th node according to the cosine similarity of the overall lexical feature vectors between the i-th node and each reference node, and the cosine similarity of the lexical feature vectors between the child nodes of the i-th node; When the i-th node does not have a child node, the clustering effect value of the i-th node is set to a first preset constant; wherein the first preset constant is a non-negative number; The result of accumulating the clustering effect value of each node in the clustering tree of the k-th feature word is used as the importance value of the k-th feature word.
5. A method for recommending postoperative home rehabilitation exercises for patients with head and neck tumors as claimed in claim 4, characterized in that: The calculation formula of the clustering effect value is: When the i-th node in the clustering tree of the k-th feature word has a child node, and the child node of the i-th node has a child node: When the i-th node in the clustering tree of the k-th feature word has a child node, and the child node of the i-th node does not have a child node: When the i-th node in the clustering tree of the k-th feature word has no child nodes: G k,i =d In the formula, G k,i is the clustering effect value of the i-th node in the clustering tree of the k-th feature word; J is the total number of reference nodes corresponding to the i-th node; X k,i ' is the overall vocabulary feature vector of the i-th node in the clustering tree of the k-th feature word; X k,j ' is the overall vocabulary feature vector of the jth reference node corresponding to the i-th node in the clustering tree of the k-th feature word; X k,i1 ' is the overall vocabulary feature vector of the first child node of the i-th node in the clustering tree of the k-th feature word; X k,i2 ' is the overall vocabulary feature vector of the second child node of the i-th node in the clustering tree of the k-th feature word; X k,i1 is the vocabulary feature vector of the first child node of the i-th node in the clustering tree of the k-th feature word; X k,i2 is the lexical feature vector of the second child node of the ith node in the clustering tree of the kth feature word; δ is a first preset constant, δ≥0; α is a second preset constant, greater than 0; || || is the vector modulus symbol.
6. A method for recommending home rehabilitation exercises for patients with head and neck tumors after surgery as claimed in claim 1, characterized in that: The calculation formula of the weight is: Where W k is the weight of the k-th feature word; S k is the importance value of the kth feature word; Q is the total number of data samples; T q,k is the characteristic value of the k-th characteristic word in the q-th data sample; R q,v is the correlation coefficient of the vocabulary feature vectors of the qth data sample and the vth data sample; β is the third preset constant, which is greater than 0; || is the absolute value function; and norm is the normalization function.
7. A method for recommending home rehabilitation exercises for patients with head and neck tumors after surgery as claimed in claim 1, characterized in that: The method of obtaining the participation vector of each data sample according to the weight of each feature word and the number of times it appears in each data sample, performing hierarchical clustering on the data samples, and obtaining the target clustering tree is as follows: Get the mean of the weights of all feature words as the standard weight; When the weight is greater than or equal to the standard weight, the feature word of the corresponding category is used as the target feature word; The product of the number of occurrences of each target feature word in each data sample and the weight of the corresponding target feature word is used as the adjusted number of each target feature word in each data sample; The sum of the adjustment times and the appearance times of each target feature word in each data sample is taken as the correction times of each target feature word in each data sample; The number of corrections corresponding to each target feature word in each data sample is constructed into a vector as the participation vector of each data sample; According to the cosine similarity between any two participating vectors, the data samples are clustered through the hierarchical clustering algorithm to obtain the target clustering tree.
8. A method for recommending home rehabilitation exercises for patients with head and neck tumors after surgery as claimed in claim 1, characterized in that: The method for obtaining the optimal clustering cluster according to the participation vector of the data sample in each node in each level of the target clustering tree is: For any level of the target clustering tree, obtain the difference between the total variance and the intra-class variance of the participation vectors of the data samples in all nodes in the level as the inter-class variance of the participation vectors of the data samples in all nodes in the level; Each node in the layer corresponding to the largest inter-class variance is regarded as the optimal clustering cluster.
9. A method for recommending home rehabilitation exercises for patients with head and neck tumors after surgery as claimed in claim 4, characterized in that: The method for obtaining the recommended exercise for postoperative home rehabilitation of the current head and neck tumor patient based on the similarity between the vocabulary feature vector of the medical record of the current head and neck tumor patient and the vocabulary feature vector of each data sample in each optimal cluster is: According to the cosine similarity between the lexical feature vector of the medical record of the current patient with head and neck tumor and the overall lexical feature vector of each optimal clustering cluster, the belonging degree value of the medical record of the current patient with head and neck tumor and each optimal clustering cluster is obtained; The optimal cluster corresponding to the maximum belonging degree value is taken as the target cluster; Obtain the cosine similarity between the vocabulary feature vector of the medical record of the current patient with head and neck tumors and the vocabulary feature vector of each data sample in the target cluster as a reference value; When the reference value is greater than the preset reference value threshold, the corresponding data sample is used as a reference sample; The exercises during the rehabilitation process of the reference sample are used as the recommended exercises for postoperative home rehabilitation of current patients with head and neck tumors.
10. A method for recommending home rehabilitation exercises for patients after surgery for head and neck tumors as claimed in claim 9, characterized in that: The calculation formula of the belonging degree value is: Where, L u is the degree of belonging between the medical records of the current patient with head and neck tumor and the uth optimal cluster; X is the vocabulary feature vector of the medical records of the current patient with head and neck tumor; X u ' is the overall vocabulary feature vector of the uth optimal cluster; || || is the vector modulus symbol; γ is the fourth preset constant, which is greater than 0.
