User feature determination and cancer user classification method, medium and equipment

Abstract

The invention relates to a method for determining user features and classifying cancer users, a medium and equipment, and relates to the technical field of machine learning. The method comprises the steps of calculating feature importance degrees of original cell features of a target user based on a preset weight value calculation model, and sorting the original cell features based on the feature importance degrees to obtain first target cell features; constructing a first original feature map of the target user according to the first target cell feature, and inputting the first original feature map into a preset feature map processing model to obtain a first target feature vector; performing clustering processing on the target user based on the first target feature vector to obtain a first user clustering result, and determining a first contour coefficient according to the first user clustering result; and determining a target contour coefficient according to the first contour coefficient, and determining a target user feature corresponding to the target user according to the target contour coefficient. The calculation efficiency is improved.

Description

Technical Field

[0001] This disclosure relates to the field of machine learning technology, and more specifically, to a method for determining user characteristics, a device for determining user characteristics, a method for classifying cancer users, a computer-readable storage medium, and an electronic device. Background Technology

[0002] In existing methods, when classifying cancer users, the user's cellular characteristics are directly input into the user classification model.

[0003] However, this method does not process cell features, which leads to low computational efficiency due to the excessively large feature dimensions.

[0004] It should be noted that the information in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this disclosure is to provide a method for determining user characteristics, a device for determining user characteristics, a method for classifying cancer users, a computer-readable storage medium, and an electronic device, thereby overcoming, at least to some extent, the problem of low computational efficiency caused by the limitations and defects of related technologies.

[0006] According to one aspect of this disclosure, a method for determining user characteristics is provided, comprising:

[0007] The feature importance of the original cell features of the target user is calculated based on a preset weight value calculation model, and the original cell features are sorted according to the feature importance to obtain the first target cell features;

[0008] Based on the first target cell features, a first original feature map of the target user is constructed, and the first original feature map is input into a preset feature map processing model to obtain a first target feature vector;

[0009] Clustering is performed on the target users based on the first target feature vector to obtain the first user clustering result, and the first contour coefficient is determined based on the first user clustering result.

[0010] The target contour coefficient is determined based on the first contour coefficient, and the target user feature corresponding to the target user is determined based on the target contour coefficient; wherein the dimension of the target user feature is lower than the dimension of the original cell feature.

[0011] In one exemplary embodiment of this disclosure, the feature importance of the target user's original cellular features is calculated based on a preset weight value calculation model, including:

[0012] Obtain the original cell features of the target user, and calculate the average feature value of the original feature values ​​of the original cell features;

[0013] Based on the average feature value and the original feature value, the feature variance of the original cell feature is determined, and the original cell feature is input into the preset weight value calculation model to obtain the feature weight value of the original cell feature;

[0014] The feature importance of the original cell features is determined based on the feature variance and feature weight values.

[0015] In one exemplary embodiment of this disclosure, the original cell features are sorted based on their importance to obtain first target cell features, including:

[0016] Iterate through the importance of the features and sort them in descending order of importance to obtain the importance ranking result.

[0017] Based on the position of the feature importance in the importance ranking result, determine the position of the original cell feature corresponding to the feature importance in the feature ranking result;

[0018] Based on the position of the original cell features in the feature sorting results, the original cell features are sorted sequentially to obtain the first target cell features.

[0019] In one exemplary embodiment of this disclosure, constructing a first original feature map of the target user based on the first target cell features includes:

[0020] Based on the first target cell features, a first original feature similarity between the target users is determined, and a first original feature similarity matrix between the target users is constructed based on the first original feature similarity.

[0021] Traverse the first original feature similarity in the first original feature similarity matrix, and extract the first feature similarity to be replaced from the first original feature similarity matrix whose first original feature similarity is less than a preset similarity threshold;

[0022] The first feature similarity to be replaced is replaced based on a preset similarity value to obtain a first target similarity matrix. Based on the first target feature similarity matrix and the first target cell feature, the first original feature map of the target user is constructed.

[0023] In one exemplary embodiment of this disclosure, constructing a first original feature map of the target user based on the first target feature similarity matrix and the first target cell features includes:

[0024] Based on the target user, a first set of child nodes is constructed, and based on the first target cell feature of the target user, a first set of feature nodes is constructed.

[0025] Based on the first target feature similarity matrix, determine the first connection relationship between the first child nodes in the first child node set, and construct the first connection edge set based on the first connection relationship;

[0026] The first original feature map of the target user is constructed based on the first set of child nodes, the first set of feature nodes, and the first set of connecting edges.

[0027] In one exemplary embodiment of this disclosure, the first original feature map is input into a preset feature map processing model to obtain a first target feature vector, including:

[0028] Based on the preset feature map processing model, the first local feature information and the first global feature information of the first original feature map are calculated, and based on the first local feature information and the first global feature information, the first original feature map is transformed to obtain the first target feature vector.

[0029] In one exemplary embodiment of this disclosure, clustering the target users based on the first target feature vector to obtain a first user clustering result includes:

[0030] A preset number of first current user samples are randomly selected from the target users as the first initial cluster centers;

[0031] Based on the first initial cluster center, the first remaining user samples among the target users, excluding the first current user sample, are randomly assigned to obtain the first initial clustering result;

[0032] Based on the first target feature vector, calculate the distance from the first remaining user sample in the first initial clustering result to the first sample point of the first initial cluster center;

[0033] The first initial clustering result and the first initial cluster center are updated based on the distance between the first sample points to obtain the first user clustering result.

[0034] In one exemplary embodiment of this disclosure, determining the first silhouette coefficient based on the first user clustering result includes:

[0035] Randomly select a first target user cluster result from the first user cluster result, and select a first target sample user from the first target user cluster result;

[0036] Calculate the first sample cohesion of the first target sample user in the clustering result of the first target user;

[0037] Calculate the first sample equal distance between the first target sample user and the first other sample users in the first other clustering results other than the first target user clustering result;

[0038] The first profile coefficient is determined based on the average distance of the first samples and the cohesion of the first samples.

[0039] In one exemplary embodiment of this disclosure, determining the target contour coefficient based on the first contour coefficient includes:

[0040] The first target cell features are used to perform dimensionality reduction to obtain the second target cell features, and the second contour coefficient is determined based on the second target cell features.

[0041] Repeat the steps for determining the second profile coefficient in sequence to obtain the third profile coefficient, ..., the Gth profile coefficient;

[0042] Extract the maximum value of the contour coefficient from the first contour coefficient, the second contour coefficient, the third contour coefficient, ..., the Gth contour coefficient to obtain the target contour coefficient.

[0043] In one exemplary embodiment of this disclosure, dimensionality reduction processing is performed based on the first target cell features to obtain second target cell features, including:

[0044] The features in the first target cell features that are located at preset positions are deleted in reverse order to obtain the second target cell features.

[0045] In one exemplary embodiment of this disclosure, determining a second contour coefficient based on second target cell features includes:

[0046] Based on the second target cell features, a second original feature map of the target user is constructed, and the second original feature map is input into a preset feature map processing model to obtain a second target feature vector;

[0047] Clustering is performed on the target users based on the second target feature vector to obtain the second user clustering result, and the second silhouette coefficient is determined based on the second user clustering result.

[0048] In one exemplary embodiment of this disclosure, determining the target user features corresponding to the target user based on the target contour coefficient includes:

[0049] The target cell features corresponding to the target contour coefficient are determined, and these target cell features are used as the target user features.

[0050] In one exemplary embodiment of this disclosure, the preset weight value calculation model is obtained in the following manner:

[0051] The historical cell features of historical users and the feature labels corresponding to historical users are obtained, and the historical cell features are input into the network model to be trained to obtain the model prediction results.

[0052] Based on the model prediction results and feature labels, a first loss function and a second loss function are constructed, and the network model to be trained is trained based on the first loss function and the second loss function to obtain the preset weight value calculation model;

[0053] The first loss function includes an L1 regularization loss function, and the second loss function includes an L2 regularization loss function.

[0054] According to one aspect of this disclosure, a method for classifying cancer patients is provided, comprising:

[0055] Obtain the original cell features of the user to be predicted;

[0056] The target cell feature to be predicted is determined from the original cell features to be predicted; wherein the target cell feature to be predicted is obtained by processing the original cell features to be predicted using any of the user feature determination methods described above;

[0057] The target cell features to be predicted are input into a preset user classification model to obtain the user classification result of the user to be predicted.

[0058] According to one aspect of this disclosure, a user characteristic determination apparatus is provided, comprising:

[0059] The feature importance determination module is used to calculate the feature importance of the original cell features of the target user based on a preset weight value calculation model, and to sort the original cell features based on the feature importance to obtain the first target cell features;

[0060] The first feature vector determination module is used to construct a first original feature map of the target user based on the first target cell features, and input the first original feature map into a preset feature map processing model to obtain a first target feature vector;

[0061] The first contour coefficient determination module is used to perform clustering processing on the target users based on the first target feature vector to obtain the first user clustering result, and determine the first contour coefficient based on the first user clustering result;

[0062] The user feature determination module is used to determine a target contour coefficient based on a first contour coefficient, and to determine the target user feature corresponding to the target user based on the target contour coefficient; wherein the dimension of the target user feature is lower than the dimension of the original cell feature.

[0063] According to one aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method for determining user characteristics as described in any one of the preceding claims, and the method for classifying cancer users as described in any one of the preceding claims.

[0064] According to one aspect of this disclosure, an electronic device is provided, comprising:

[0065] Processor; and

[0066] Memory for storing the executable instructions of the processor;

[0067] The processor is configured to execute the user characteristic determination method and the cancer user classification method described in any one of the preceding claims via executing the executable instructions.

[0068] This disclosure provides a method for determining user features. On one hand, it calculates the feature importance of the original cell features of a target user based on a preset weight value calculation model, and sorts the original cell features according to their importance to obtain a first target cell feature. Then, based on the first target cell feature, it constructs a first original feature map of the target user and inputs it into a preset feature map processing model to obtain a first target feature vector. Next, it performs clustering processing on the target user based on the first target feature vector to obtain a first user clustering result, and determines a first contour coefficient based on the first user clustering result. Finally, it determines a target contour coefficient based on the first contour coefficient, and determines the target user feature corresponding to the target user based on the target contour coefficient. Since the dimension of the target user feature is lower than that of the original cell feature, it solves the problem of low computational efficiency caused by excessively large feature dimensions in the prior art. On the other hand, since the target user feature corresponding to the target user can be determined based on the target contour coefficient, the accuracy of the determined target user feature is improved.

[0069] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0070] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0071] Figure 1 The flowchart illustrates a method for determining user characteristics according to an example embodiment of the present disclosure.

[0072] Figure 2 The illustration shows a scenario example diagram of a Sigmoid function according to an exemplary embodiment of the present disclosure.

[0073] Figure 3 The flowchart illustrates an example of a training method for a weight value calculation model according to an exemplary embodiment of the present disclosure.

[0074] Figure 4 The illustration schematically shows a scenario example diagram of a first original feature map according to an exemplary embodiment of the present disclosure.

[0075] Figure 5 The diagram schematically illustrates a user node according to an example embodiment of the present disclosure and an example diagram showing the user node connected to its corresponding feature node.

[0076] Figure 6 The illustration shows a scenario example diagram illustrating a first user clustering result according to an example embodiment of the present disclosure.

[0077] Figure 7 The diagram illustrates a flowchart of a method for classifying cancer users according to an exemplary embodiment of this disclosure.

[0078] Figure 8 The diagram schematically illustrates a structural example of a user characteristic determination device according to an exemplary embodiment of the present disclosure.

[0079] Figure 9 The diagram schematically illustrates a structural example of a cancer user classification device according to an exemplary embodiment of the present disclosure.

[0080] Figure 10 An electronic device is illustrated schematically according to an example embodiment of the present disclosure for implementing a method for determining user characteristics and / or a method for classifying cancer users. Detailed Implementation

[0081] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this disclosure more comprehensive and complete, and to fully convey the concept of the example embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced with one or more of the specific details omitted, or other methods, components, apparatus, steps, etc., can be employed. In other instances, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0082] Furthermore, the accompanying drawings are merely illustrative of this disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0083] Currently, tumor diagnosis and treatment have entered the era of precision medicine. In actual diagnosis and treatment, tumors can be classified into multiple subtypes according to different detection methods such as gene mutation, gene expression, and tumor microenvironment. At the same time, doctors will provide corresponding treatment plans for different subtypes, which is conducive to personalized treatment for patients. Furthermore, gene-based subtyping can better explain the biological heterogeneity of tumors and the differences in treatment response. Moreover, in the process of diagnosis and treatment, comprehensive consideration of immune cells plays an important role in the physiological function of tumors.

[0084] In one example embodiment, the proportion of immune cells, as an immunophenotyping feature representation method, can be used to represent tumor cell subtypes. However, the data features of the deconvolutioned immune cell proportion have a large dimensionality, necessitating the identification of features related to immunophenotyping and survival risk. This is beneficial for identifying important features and improving detection efficiency. Therefore, how to extract key features from immune cells for cancer patient classification has become an urgent problem to be solved.

[0085] Based on this, this exemplary embodiment first provides a method for determining user characteristics, which can run on a server, server cluster, or cloud server, etc. Of course, those skilled in the art can also run the method disclosed herein on other platforms as needed, and this exemplary embodiment does not impose any special limitations on this. Specifically, refer to... Figure 1 As shown, the method for determining this user characteristic may include the following steps:

[0086] Step S110. Calculate the feature importance of the original cell features of the target user based on a preset weight value calculation model, and sort the original cell features based on the feature importance to obtain the first target cell features;

[0087] Step S120. Based on the first target cell features, construct the first original feature map of the target user, and input the first original feature map into a preset feature map processing model to obtain the first target feature vector;

[0088] Step S130. Perform clustering processing on the target users based on the first target feature vector to obtain the first user clustering result, and determine the first contour coefficient based on the first user clustering result;

[0089] Step S140. Determine the target contour coefficient based on the first contour coefficient, and determine the target user feature corresponding to the target user based on the target contour coefficient; wherein the dimension of the target user feature is lower than the dimension of the original cell feature.

[0090] In the aforementioned method for determining user features, on the one hand, the feature importance of the original cell features of the target user is calculated based on a preset weight value calculation model, and the original cell features are sorted according to the feature importance to obtain the first target cell features; then, based on the first target cell features, a first original feature map of the target user is constructed, and the first original feature map is input into a preset feature map processing model to obtain a first target feature vector; then, based on the first target feature vector, the target user is clustered to obtain the first user clustering result, and a first contour coefficient is determined based on the first user clustering result; finally, a target contour coefficient is determined based on the first contour coefficient, and the target user features corresponding to the target user are determined based on the target contour coefficient; since the dimension of the target user features is lower than that of the original cell features, the problem of low computational efficiency caused by excessively large feature dimensions in the prior art can be solved; on the other hand, since the target user features corresponding to the target user can be determined based on the target contour coefficient, the accuracy of the determined target user features is improved.

[0091] The method for determining user characteristics described in the exemplary embodiments of this disclosure will be further explained and illustrated below with reference to the accompanying drawings.

[0092] First, the application scenarios of the exemplary embodiments of this disclosure will be explained and described. Specifically, the method for determining user characteristics described in the exemplary embodiments of this disclosure can be used to accurately classify the subtype to which tumor patients belong; that is, based on the target user characteristics determined by the method for determining user characteristics described in the exemplary embodiments of this disclosure, the subtype to which the target user belongs can be classified, thereby reducing the dimensionality of the features and improving computational efficiency while ensuring the accuracy of the user classification results.

[0093] Furthermore, in the user feature determination method described in the exemplary embodiments of this disclosure, supervised and unsupervised feature selection methods can be used for feature selection. The results are then subjected to subsequent cluster analysis, and the stopping condition is determined based on the cluster evaluation results. The supervised feature selection method described here refers to training a weight value calculation model in a supervised manner, and then determining the weight values ​​of cell features based on the trained weight value calculation model. The unsupervised feature selection method described here refers to training a graph convolutional neural network in an unsupervised manner, then determining the target feature vector based on the trained graph convolutional neural network, and finally clustering the target feature vector to determine the final user features based on the silhouette coefficients of the clustering results. Meanwhile, the user feature determination method described in the exemplary embodiments of this disclosure has a certain degree of interpretability in the process of selecting cell features (deleting features with low importance), and improves computational efficiency compared to exhaustive methods.

[0094] Secondly, the preset weight calculation model involved in the exemplary embodiments of this disclosure will be explained and described. Specifically, the preset weight calculation model described herein can be implemented based on a logistic regression model; that is, the preset weight calculation model can be obtained by training the logistic regression model based on the historical cell features of historical users. In practical applications, feature mapping can be performed based on the Sigmoid function, and the weight parameters in the logistic regression model can be adjusted based on loss functions with L1 and L2 regularization; a specific example diagram of the Sigmoid function described herein can be found in [reference needed]. Figure 2 As shown.

[0095] The following will combine Figure 3 This section explains and illustrates the specific training process of the weight calculation model. For details, please refer to [link / reference]. Figure 3 As shown, the specific training process of the weight value calculation model may include the following steps:

[0096] Step S310: Obtain the historical cell features of historical users and the feature labels corresponding to the historical users, and input the historical cell features into the network model to be trained to obtain the model prediction results;

[0097] Step S320: Construct a first loss function and a second loss function based on the model prediction results and feature labels; wherein, the first loss function includes an L1 regularized loss function, and the second loss function includes an L2 regularized loss function;

[0098] Step S330: Train the network model to be trained based on the first loss function and the second loss function to obtain the preset weight value calculation model.

[0099] The following will further explain and illustrate steps S310-S330. Specifically, in the actual training process, it is assumed that the feature dimension of the historical cell features of historical users is m-dimensional (the first cell feature, the second cell feature, ..., the m-th cell feature, which are concatenated together to obtain an m-dimensional cell feature); at the same time, the feature labels recorded here can refer to the current survival status of the user, for example, label 0 represents the dead state, and label 1 represents the living state. Furthermore, the specific expressions of the L1 regularized loss function and the L2 regularized loss function recorded above can be referred to as formulas (1) and (2) below:

[0100]

[0101]

[0102] Where L1 is the first loss function, L2 is the second loss function, N1 is the number of historical users, and y (i) Let h(x) be the feature label of the i-th historical user. (i) Let be the model prediction result for the i-th historical user, α be the model penalty coefficient, which is a known constant; w be the model parameters, i.e., the weight values ​​corresponding to different features; ||w||1 represents the sum of the absolute values ​​of the model parameters; It represents the square root of the sum of squares of the model parameters.

[0103] It should be added that after the preset weight value calculation model is trained, when cell features are input into the weight value calculation model, different cell features will have corresponding feature weights.

[0104] The following will combine Figure 2 as well as Figure 3 right Figure 1 The method for determining user characteristics shown will be further explained and illustrated. Specifically:

[0105] In step S110, the feature importance of the original cell features of the target user is calculated based on a preset weight value calculation model, and the original cell features are sorted based on the feature importance to obtain the first target cell features.

[0106] In this example embodiment, the feature importance of the original cell features needs to be calculated first. Specifically, the calculation process for feature importance can be implemented as follows: First, obtain the original cell features of the target user and calculate the average feature value of the original feature values ​​of the original cell features; second, determine the feature variance of the original cell features based on the average feature value and the original feature value, and input the original cell features into the preset weight value calculation model to obtain the feature weight value of the original cell features; then, determine the feature importance of the original cell features based on the feature variance and the feature weight value.

[0107] The target users mentioned above can be understood as users who need to undergo tumor subtyping, and the number of target users can be N². The original cell characteristics can be the cellular characteristics of the immune cells used to characterize the target user. For example, taking lymphoma immunotyping as an example, immune cells, as an important component of the tumor microenvironment (TME), play a crucial role in the physiological function of tumors and have certain research value. Therefore, calculating the proportion of immune cells as a characteristic of the sample (i.e., the target user characteristic of the target user) and then conducting subsequent survival analysis and prognostic risk assessment based on this target user characteristic is of great significance. Furthermore, the main purpose of selecting target user characteristics is to find the most representative characteristics or combinations of characteristics as the diagnostic basis for immunotyping. Further, the average characteristic value mentioned above can be obtained by weighted summation based on each original characteristic value, or by direct calculation. The specific calculation process of the characteristic variance is consistent with the specific calculation process of the ordinary variance, and will not be elaborated further here.

[0108] Furthermore, in the process of calculating feature weights, the original cell features can be directly input into the weight calculation model. Then, based on the arrangement of the original cell features and the linear mapping expression of the weight model, the feature weights can be obtained (features with larger weights are more important for subsequent cluster analysis). At the same time, in the process of calculating feature importance, the feature weights and feature variances can be directly added together or weighted summation can be performed. This example does not impose any special restrictions on this.

[0109] Secondly, after obtaining the importance of the features, the original cell features can be sorted. Specifically, the feature sorting process can be implemented as follows: traverse the importance of the features, sorting them in descending order of importance to obtain the importance ranking result; determine the ranking position of the original cell features corresponding to the importance of the features in the feature ranking result based on their ranking position; and sort the original cell features sequentially based on their ranking position to obtain the first target cell feature. In other words, the positions of the original cell features can be rearranged based on feature importance to obtain the first target cell feature. The commonality between the first target cell feature and the original cell feature is that the feature dimension is the same (m-dimensional), but the difference lies in the arrangement position of the features (the original cell features are randomly arranged, while the first target cell features are arranged according to feature importance).

[0110] In step S120, a first original feature map of the target user is constructed based on the first target cell features, and the first original feature map is input into a preset feature map processing model to obtain a first target feature vector.

[0111] In this example embodiment, firstly, a first original feature map is constructed. Specifically, the construction process of the first original feature map can be implemented as follows: based on the first target cell features, the first original feature similarity between the target users is determined, and a first original feature similarity matrix between the target users is constructed based on the first original feature similarity; the first original feature similarities in the first original feature similarity matrix are traversed, and a first feature similarity to be replaced is extracted from the first original feature similarity matrix with a first original feature similarity less than a preset similarity threshold; the first feature similarity to be replaced is replaced based on a preset similarity value to obtain a first target similarity matrix, and the first original feature map of the target user is constructed based on the first target feature similarity matrix and the first target cell features.

[0112] Specifically, the calculation process for the first original feature similarity can be achieved by calculating the dot product of the current feature vectors of each first target cell feature. The current feature vector can be calculated based on the feature arrangement position of the first target cell features. It should be noted that since the feature arrangement results of the first target cell features are different for different target users, but the feature dimensions and feature names of the original cell features for each target user are the same, the current feature vector can be determined based on the arrangement position of each cell feature in the first target cell features, and then the first original feature similarity can be determined based on this current feature vector. Simultaneously, in the process of constructing the first original feature similarity matrix, the target user can be directly assigned a position in the matrix, and then the first original feature similarity matrix can be obtained based on the user's position in the matrix and the first original feature similarity between users. This first original feature similarity matrix is ​​a diagonal matrix. Furthermore, to prevent information redundancy, the first original feature similarity values ​​in the first original feature similarity matrix below a certain threshold (i.e., a preset similarity threshold, specifically 0.62) can be replaced with 0 (i.e., a preset similarity value). It should be noted that the preset similarity threshold can be obtained through repeated trials or predicted using a corresponding similarity threshold prediction model; this example does not impose any special restrictions on this. Furthermore, this similarity threshold prediction model can be implemented based on neural networks, such as convolutional neural networks, recurrent neural networks, etc.; this example also does not impose any special restrictions on this.

[0113] In one example embodiment, constructing the first original feature map of the target user based on the first target feature similarity matrix and the first target cell features can be achieved as follows: constructing a first set of child nodes based on the target user, and constructing a first set of feature nodes based on the first target cell features of the target user; determining a first connection relationship between the first child nodes in the first set of child nodes based on the first target feature similarity matrix, and constructing a first set of connection edges based on the first connection relationship; constructing the first original feature map of the target user based on the first set of child nodes, the first set of feature nodes, and the first set of connection edges. In other words, during the construction of the first original feature map, each target user can be abstracted into a first child node, thus constructing a set of first child nodes. Then, the first target cell feature corresponding to that target user is abstracted into multiple first feature nodes, thus constructing a set of first feature nodes. Further, based on the first target feature similarity matrix between users, other users connected to the first child nodes are determined to obtain the first connection relationship, and a first set of connection edges is constructed based on the first connection relationship. The weight of the first connection edge is the target feature similarity between the two users (i.e., the updated feature similarity). Of course, a second set of connection edges also needs to be constructed based on the second connection relationship between each target user and the first target cell feature. The weight of the second connection edge can be the feature importance of each cell feature. Finally, based on the combination of the first child nodes, the first feature node set, the first connection edge set, and the second connection edge set, the first original feature map is constructed. The obtained first original feature map can be referenced... Figure 4 As shown.

[0114] In one example embodiment, reference is made to... Figure 4 As shown, the first original feature map may include user nodes and feature nodes; each user node may be connected to multiple user nodes, and each user node is connected to multiple feature nodes; an example diagram of a user node and its connections to its corresponding feature nodes can be found in [reference needed]. Figure 5 As shown.

[0115] Secondly, after obtaining the first original feature map, the first original feature map can be input into the preset feature map processing model to obtain the first target feature vector. Specifically, the specific calculation process of the first target feature vector can be implemented in the following way: calculate the first local feature information and the first global feature information of the first original feature map based on the preset feature map processing model, and transform the first original feature map based on the first local feature information and the first global feature information to obtain the first target feature vector. Specifically, in the process of calculating the first target feature vector, after inputting the first original feature map into the model, a graph convolution algorithm can be used for vector embedding to obtain the first target feature vector. At the same time, in the process of determining the first local feature information and the first global feature information, it can be implemented based on random walk or hierarchical clustering; in the process of transforming the original features, it can be implemented based on autoencoder or matrix decomposition. Among them, the specific calculation process of graph convolution can be implemented by the following formula (3):

[0116]

[0117] Among them, H l+1 Let σ(·) be the first target feature vector, D be the degree matrix (i.e., the number of other first child nodes connected to the first child node), M be the first target similarity matrix, H be the first original feature map, and W be the network parameters of the graph convolutional neural network. The reason for using a graph convolutional neural network to calculate the first target feature vector is that existing techniques cannot fully consider the potential connections between cells, resulting in low accuracy of the obtained target feature vector. Furthermore, graph-based analysis methods can leverage neighbor information to mine latent feature representation capabilities, thereby further improving the accuracy of the target feature vector.

[0118] In step S130, the target users are clustered based on the first target feature vector to obtain the first user clustering result, and the first contour coefficient is determined based on the first user clustering result.

[0119] In this example embodiment, the first step is to determine the first user clustering result. Specifically, the first user clustering result can be determined as follows: A preset number of first current user samples are randomly selected from the target users as the first initial cluster centers; based on the first initial cluster centers, the first remaining user samples (excluding the first current user samples) from the target users are randomly assigned to obtain the first initial clustering result; according to the first target feature vector, the distance between the first remaining user samples in the first initial clustering result and the first sample point of the first initial cluster center is calculated; the first initial clustering result and the first initial cluster center are updated based on the first sample point distance to obtain the first user clustering result. That is, in practical applications, the entire clustering process can be implemented based on K-means; the preset number K can be 3, or other values, and this example does not impose any special restrictions; furthermore, the specific value of K can be determined based on multiple trials or a prediction model, and this example does not impose any special restrictions. The obtained first user clustering result can be referenced... Figure 6 As shown.

[0120] Secondly, the first contour coefficient is calculated. Specifically, the determination process of the first contour coefficient can be implemented as follows: randomly select a first target user cluster result from the first user cluster result, and select a first target sample user from the first target user cluster result; calculate the first sample cohesion of the first target sample user in the first target user cluster result; calculate the first sample average distance between the first target sample user and the first other sample user in the first other cluster result other than the first target user cluster result; determine the first contour coefficient based on the first sample average distance and the first sample cohesion. The calculation formula of the first contour coefficient can be implemented by the following formula (4):

[0121]

[0122] Where S(i) represents the first contour coefficient of the i-th first target user; a(i) represents the cohesion of the sample point (the i-th first target user) (the first sample cohesion), and the specific calculation method can be shown in the following formula (5):

[0123]

[0124] Where j represents other sample points in the same class (first target user clustering result) as sample i, and distance represents the distance between i and j; dis(i,j) can be determined by the first target feature similarity in the first target feature similarity matrix; for example, the first sample cohesion can be obtained by using 1 - the first target feature similarity; therefore, the smaller a(i) is, the more compact the class is. Furthermore, the calculation method of the first sample average distance b(i) is similar to that of a(i), except that it is necessary to traverse other clusters (first other clustering results) to obtain multiple values ​​{b1(i), b2(i), b3(i), ..., b m (i)}; and then select the smallest value from the multiple values ​​obtained as the final result (first sample average distance).

[0125] In step S140, a target contour coefficient is determined based on a first contour coefficient, and a target user feature corresponding to the target user is determined based on the target contour coefficient; wherein the dimension of the target user feature is lower than the dimension of the original cell feature.

[0126] In this example embodiment, firstly, the target contour coefficient is determined. Specifically, the determination process of the target contour coefficient can be implemented as follows: dimensionality reduction is performed based on the first target cell features to obtain the second target cell features, and the second contour coefficient is determined based on the second target cell features; the determination steps of the second contour coefficient are repeated sequentially to obtain the third contour coefficient, ..., the Gth contour coefficient; the maximum value of the contour coefficient is extracted from the first contour coefficient, the second contour coefficient, the third contour coefficient, ..., the Gth contour coefficient to obtain the target contour coefficient. That is to say, in practical applications, multiple contour coefficients need to be calculated, and then the target contour coefficient is determined from the multiple contour coefficients; at the same time, since a larger contour coefficient indicates a better clustering effect, the calculation steps are repeated cyclically. If the contour coefficient is greater than the previous calculation result, the loop continues until the contour coefficient is less than the previous result, at which point the loop stops; that is, the value of G can be determined based on the change of the contour coefficient; for example, after each contour coefficient is obtained, it can be compared with the previous contour coefficient. If it is constantly increasing, it needs to be continuously calculated; if it starts to decrease, the previous contour coefficient is determined as the target contour coefficient.

[0127] In one example embodiment, in order to perform dimensionality reduction on the first target cell features (i.e., select target user features from the original cell features) and then calculate other contour coefficients based on the dimensionality-reduced target cell features, the following steps need to be performed: Dimensionality reduction is performed on the first target cell features to obtain second target cell features; specifically, the dimensionality reduction process can be implemented as follows: features at preset positions in the first target cell features are deleted in reverse order to obtain the second target cell features. That is, one or more dimensions of cell features are deleted from the very end of the first target cell features (i.e., the last or second-to-last cell features are deleted) to obtain the second target features; the specific number of features to be deleted each time can be determined according to actual needs, and this example does not impose any special restrictions on this.

[0128] In one example embodiment, after obtaining the second target cell features, the second contour coefficient can be determined based on the second target cell features. Specifically, this can be achieved as follows: based on the second target cell features, a second original feature map of the target user is constructed, and the second original feature map is input into a preset feature map processing model to obtain a second target feature vector; based on the second target feature vector, the target user is clustered to obtain a second user clustering result, and the second contour coefficient is determined based on the second user clustering result. That is, the above steps for determining the first contour coefficient can be repeated to obtain the second contour coefficient; the difference between the second original feature map and the first original feature map is that the second original feature map has fewer feature nodes than the first original feature map, and of course, the first connecting edge in the second original feature map will also change.

[0129] Secondly, the target user features are determined. Specifically, this can be achieved by identifying the target cell features corresponding to the target contour coefficient and using these target cell features as the target user features. That is, the target cell features in the dimension containing the target contour coefficient can be used as the target user features. It should be noted that removing cell features with the lowest feature importance through a cyclic deletion process, and then using the remaining cell features as the target user features, can improve computational efficiency while ensuring the accuracy of the user classification results.

[0130] Thus, the method for determining user features described in the exemplary embodiments of this disclosure has been fully implemented. Based on the foregoing description, it can be understood that the method for determining user features described in the exemplary embodiments of this disclosure, on the one hand, can achieve the final feature selection based on a feature selection method combining unsupervised and supervised feature selection, thereby improving the accuracy of feature selection; on the other hand, the data clustering method based on graph convolution can also improve computational efficiency.

[0131] This disclosure also provides an example embodiment of a method for classifying cancer patients. Specifically, refer to... Figure 7 As shown, the method for classifying cancer users may include the following steps:

[0132] Step S710: Obtain the original cell features of the user to be predicted;

[0133] Step S720: Determine target cell features from the original cell features to be predicted; wherein the target cell features to be predicted are obtained by processing the original cell features to be predicted using a user feature determination method.

[0134] Step S730: Input the target cell features to be predicted into a preset user classification model to obtain the user classification result of the user to be predicted.

[0135] In the aforementioned cancer user classification method, because the dimensionality of the target cell feature is lower than that of the original cell feature, it solves the problem of low computational efficiency caused by excessively large feature dimensions in existing technologies, thus improving user classification efficiency. Furthermore, by classifying users based on the target cell feature, subsequent survival analysis and risk assessment can be performed on the user based on the classification results, thereby providing specific treatment options.

[0136] The following are embodiments of the apparatus disclosed herein, which can be used to execute embodiments of the method disclosed herein. For details not disclosed in the apparatus embodiments of this disclosure, please refer to the embodiments of the method disclosed herein.

[0137] This disclosure also provides an apparatus for determining user characteristics in exemplary embodiments. Specifically, refer to... Figure 8 As shown, the user feature determination device may include a feature importance determination module 810, a first feature vector determination module 820, a first contour coefficient determination module 830, and a user feature determination module 840. Wherein:

[0138] The feature importance determination module 810 can be used to calculate the feature importance of the original cell features of the target user based on a preset weight value calculation model, and sort the original cell features based on the feature importance to obtain the first target cell features;

[0139] The first feature vector determination module 820 can be used to construct a first original feature map of the target user based on the first target cell features, and input the first original feature map into a preset feature map processing model to obtain a first target feature vector;

[0140] The first contour coefficient determination module 830 can be used to perform clustering processing on the target user based on the first target feature vector to obtain the first user clustering result, and determine the first contour coefficient based on the first user clustering result;

[0141] The user feature determination module 840 can be used to determine a target contour coefficient based on a first contour coefficient, and to determine a target user feature corresponding to the target user based on the target contour coefficient; wherein the dimension of the target user feature is lower than the dimension of the original cell feature.

[0142] In one exemplary embodiment of this disclosure, calculating the feature importance of a target user's original cell features based on a preset weight value calculation model includes: acquiring the target user's original cell features and calculating the average feature value of the original feature values ​​of the original cell features; determining the feature variance of the original cell features based on the average feature value and the original feature value, and inputting the original cell features into the preset weight value calculation model to obtain the feature weight value of the original cell features; and determining the feature importance of the original cell features based on the feature variance and the feature weight value.

[0143] In one exemplary embodiment of this disclosure, sorting the original cell features based on the importance of the features to obtain a first target cell feature includes: traversing the importance of the features and sorting the features in descending order to obtain an importance ranking result; determining the position of the original cell feature corresponding to the importance of the feature in the feature ranking result based on the position of the feature importance in the importance ranking result; and sorting the original cell features sequentially based on the position of the original cell feature in the feature ranking result to obtain the first target cell feature.

[0144] In an exemplary embodiment of this disclosure, constructing a first original feature map of the target user based on the first target cell features includes: determining a first original feature similarity between the target users based on the first target cell features, and constructing a first original feature similarity matrix between the target users based on the first original feature similarity; traversing the first original feature similarities in the first original feature similarity matrix, extracting a first feature similarity to be replaced from the first original feature similarity matrix whose first original feature similarity is less than a preset similarity threshold; replacing the first feature similarity to be replaced based on a preset similarity value to obtain a first target similarity matrix, and constructing the first original feature map of the target user based on the first target feature similarity matrix and the first target cell features.

[0145] In one exemplary embodiment of this disclosure, constructing a first original feature map of the target user based on the first target feature similarity matrix and the first target cell feature includes: constructing a first set of child nodes based on the target user, and constructing a first set of feature nodes based on the first target cell feature of the target user; determining a first connection relationship between the first child nodes in the first set of child nodes based on the first target feature similarity matrix, and constructing a first set of connection edges based on the first connection relationship; and constructing the first original feature map of the target user based on the first set of child nodes, the first set of feature nodes, and the first set of connection edges.

[0146] In one exemplary embodiment of this disclosure, inputting the first original feature map into a preset feature map processing model to obtain a first target feature vector includes: calculating first local feature information and first global feature information of the first original feature map based on the preset feature map processing model, and transforming the first original feature map based on the first local feature information and the first global feature information to obtain the first target feature vector.

[0147] In one exemplary embodiment of this disclosure, clustering the target users based on the first target feature vector to obtain a first user clustering result includes: randomly selecting a preset number of first current user samples from the target users as first initial cluster centers; randomly assigning first remaining user samples (excluding the first current user samples) from the target users based on the first initial cluster centers to obtain a first initial clustering result; calculating the distance between the first remaining user samples in the first initial clustering result and the first sample point of the first initial cluster center according to the first target feature vector; and updating the first initial clustering result and the first initial cluster center based on the first sample point distance to obtain the first user clustering result.

[0148] In one exemplary embodiment of this disclosure, determining the first silhouette coefficient based on the first user clustering result includes: randomly selecting a first target user clustering result from the first user clustering result, and selecting a first target sample user from the first target user clustering result; calculating a first sample cohesion of the first target sample user in the first target user clustering result; calculating a first sample average distance between the first target sample user and a first other sample user in a first other clustering result other than the first target user clustering result; and determining the first silhouette coefficient based on the first sample average distance and the first sample cohesion.

[0149] In an exemplary embodiment of this disclosure, determining the target contour coefficient based on the first contour coefficient includes: performing dimensionality reduction processing based on the first target cell features to obtain second target cell features, and determining the second contour coefficient based on the second target cell features; repeating the determination step of the second contour coefficient in sequence to obtain the third contour coefficient, ..., the Gth contour coefficient; and extracting the maximum value of the contour coefficient from the first contour coefficient, the second contour coefficient, the third contour coefficient, ..., the Gth contour coefficient to obtain the target contour coefficient.

[0150] In one exemplary embodiment of this disclosure, the second target cell feature is obtained by performing dimensionality reduction processing based on the first target cell feature, including: deleting features in the first target cell feature that are located at a preset position in reverse order to obtain the second target cell feature.

[0151] In one exemplary embodiment of this disclosure, determining the second contour coefficient based on the second target cell features includes: constructing a second original feature map of the target user based on the second target cell features, and inputting the second original feature map into a preset feature map processing model to obtain a second target feature vector; performing clustering processing on the target user based on the second target feature vector to obtain a second user clustering result, and determining the second contour coefficient based on the second user clustering result.

[0152] In one exemplary embodiment of this disclosure, determining the target user feature corresponding to the target user based on the target contour coefficient includes: determining the target cell feature corresponding to the target contour coefficient, and using the target cell feature as the target user feature.

[0153] In an exemplary embodiment of this disclosure, the preset weight value calculation model is obtained as follows: historical cell features of historical users and feature labels corresponding to historical users are obtained, and the historical cell features are input into the network model to be trained to obtain the model prediction result; a first loss function and a second loss function are constructed based on the model prediction result and the feature labels, and the network model to be trained is trained based on the first loss function and the second loss function to obtain the preset weight value calculation model; wherein, the first loss function includes an L1 regularized loss function, and the second loss function includes an L2 regularized loss function.

[0154] This disclosure also provides a device for classifying cancer patients. Specifically, see [link to relevant documentation]. Figure 9 As shown, the cancer user classification device may include a raw cell feature acquisition module 910, a target cell feature determination module 920, and a cancer user classification module 930. Wherein:

[0155] The original cell feature acquisition module 910 can be used to acquire the original cell features of the user to be predicted.

[0156] The target cell feature determination module 920 can be used to determine the target cell feature from the original cell features to be predicted; wherein the target cell feature is obtained by processing the original cell features to be predicted through any one of the user feature determination methods described above.

[0157] The cancer user classification module 930 can be used to input the target cell features to be predicted into a preset user classification model to obtain the user classification result of the user to be predicted.

[0158] The specific details of each module in the aforementioned user characteristic determination device and cancer user classification device have been described in detail in the corresponding user characteristic determination method and cancer user classification method, so they will not be repeated here.

[0159] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0160] Furthermore, although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0161] In an exemplary embodiment of this disclosure, an electronic device capable of implementing the above-described method is also provided.

[0162] Those skilled in the art will understand that various aspects of this disclosure can be implemented as a system, method, or program product. Therefore, various aspects of this disclosure can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software aspects, collectively referred to herein as a "circuit," "module," or "system."

[0163] The following reference Figure 10 To describe an electronic device 1000 according to such an embodiment of the present disclosure. Figure 10 The electronic device 1000 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0164] like Figure 10 As shown, the electronic device 1000 is manifested in the form of a general-purpose computing device. The components of the electronic device 1000 may include, but are not limited to: at least one processing unit 1010, at least one storage unit 1020, a bus 1030 connecting different system components (including storage unit 1020 and processing unit 1010), and a display unit 1040.

[0165] The storage unit stores program code that can be executed by the processing unit 1010, causing the processing unit 1010 to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure. For example, the processing unit 1010 can perform actions such as... Figure 1Step S110: Calculate the feature importance of the original cell features of the target user based on a preset weight value calculation model, and sort the original cell features based on the feature importance to obtain the first target cell feature; Step S120: Construct the first original feature map of the target user based on the first target cell feature, and input the first original feature map into a preset feature map processing model to obtain the first target feature vector; Step S130: Perform clustering processing on the target user based on the first target feature vector to obtain the first user clustering result, and determine the first contour coefficient based on the first user clustering result; Step S140: Determine the target contour coefficient based on the first contour coefficient, and determine the target user feature corresponding to the target user based on the target contour coefficient; wherein, the dimension of the target user feature is lower than the dimension of the original cell feature.

[0166] For example, the processing unit 1010 can also perform actions such as Figure 7 The steps shown are as follows: Step S710: Obtain the original cell features to be predicted for the user to be predicted; Step S720: Determine the target cell features to be predicted from the original cell features to be predicted; wherein the target cell features to be predicted are obtained by processing the original cell features to be predicted through any of the user feature determination methods described above; Step S730: Input the target cell features to be predicted into a preset user classification model to obtain the user classification result of the user to be predicted.

[0167] Storage unit 1020 may include readable media in the form of volatile storage units, such as random access memory (RAM) 10201 and / or cache memory 10202, and may further include read-only memory (ROM) 10203.

[0168] Storage unit 1020 may also include a program / utility 10204 having a set (at least one) program module 10205, such program module 10205 including but not limited to: operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0169] Bus 1030 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the multiple bus structures.

[0170] Electronic device 1000 can also communicate with one or more external devices 1100 (e.g., keyboard, pointing device, Bluetooth device, etc.), one or more devices that enable a user to interact with electronic device 1000, and / or any device that enables electronic device 1000 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 1050. Furthermore, electronic device 1000 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 1060. As shown, network adapter 1060 communicates with other modules of electronic device 1000 via bus 1030. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 1000, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0171] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal device, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0172] In exemplary embodiments of this disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the methods described above is stored. In some possible implementations, various aspects of this disclosure may also be implemented as a program product including program code that, when the program product is run on a terminal device, causes the terminal device to perform the steps of the various exemplary embodiments of this disclosure described in the "Exemplary Methods" section above.

[0173] The program product for implementing the above-described method according to embodiments of the present disclosure may employ a portable compact disc read-only memory (CD-ROM) and include program code, and may run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto. In this document, the readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0174] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0175] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting programs for use by or in conjunction with an instruction execution system, apparatus, or device.

[0176] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0177] Program code for performing the operations of this disclosure can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0178] Furthermore, the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of this disclosure and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0179] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention described herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not invented by this disclosure. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

Claims

1. A method of determining a user characteristic, the method comprising: The method comprises the following steps: calculating the feature importance degree of the original cell feature of the target user based on a preset weight value calculation model, and sorting the original cell feature based on the feature importance degree to obtain a first target cell feature; constructing a first original feature map of the target user according to the first target cell feature, and inputting the first original feature map into a preset feature map processing model to obtain a first target feature vector; performing clustering processing on the target user based on the first target feature vector to obtain a first user clustering result, and determining a first contour coefficient according to the first user clustering result; determining a target contour coefficient according to the first contour coefficient, and determining a target user feature corresponding to the target user according to the target contour coefficient; wherein the dimension of the target user feature is lower than the dimension of the original cell feature.

2. The method of determining user characteristics according to claim 1, wherein, The method comprises the following steps: obtaining the original cell feature of the target user, and calculating the average feature value of the original feature value of the original cell feature; determining the feature variance of the original cell feature according to the average feature value and the original feature value, and inputting the original cell feature into the preset weight value calculation model to obtain the feature weight value of the original cell feature; determining the feature importance degree of the original cell feature according to the feature variance and the feature weight value.

3. The method of claim 1, wherein, The method comprises the following steps: traversing the feature importance degree, and sequentially sorting the feature importance degree in descending order to obtain an importance degree sorting result; determining the arrangement position of the original cell feature corresponding to the feature importance degree in the feature sorting result according to the arrangement position of the feature importance degree in the importance degree sorting result; sequentially sorting the original cell feature based on the arrangement position of the original cell feature in the feature sorting result to obtain the first target cell feature.

4. The method of claim 1, wherein, The method comprises the following steps: determining the first original feature similarity between the target users according to the first target cell feature, and constructing a first original feature similarity matrix between the target users according to the first original feature similarity; traversing the first original feature similarity in the first original feature similarity matrix, and extracting a first to-be-replaced feature similarity with a first original feature similarity less than a preset similarity threshold from the first original feature similarity matrix; replacing the first to-be-replaced feature similarity based on a preset similarity value to obtain a first target similarity matrix, and constructing the first original feature map of the target user according to the first target feature similarity matrix and the first target cell feature.

5. The method of claim 4, wherein, The method comprises the following steps: According to the target user, a first sub-node set is constructed, and according to a first target cell feature of the target user, a first feature node set is constructed; According to the first target feature similarity matrix, a first connection relationship between first sub-nodes in the first sub-node set is determined, and a first connection edge set is constructed according to the first connection relationship; According to the first sub-node set, the first feature node set and the first connection edge set, a first original feature map of the target user is constructed.

6. The method of claim 1, wherein, The first original feature map is input into a preset feature map processing model to obtain a first target feature vector, including: Based on the preset feature map processing model, first local feature information and first global feature information of the first original feature map are calculated, and the first original feature map is converted based on the first local feature information and the first global feature information to obtain the first target feature vector.

7. The method of claim 1, wherein, Based on the first target feature vector, the target user is clustered to obtain a first user clustering result, including: A preset number of first current user samples are randomly selected from the target user as first initial clustering centers; Based on the first initial clustering center, the first remaining user samples in the target user except the first current user samples are randomly assigned to obtain a first initial clustering result; According to the first target feature vector, a first sample point distance from the first remaining user samples in the first initial clustering result to the first initial clustering center is calculated; Based on the first sample point distance, the first initial clustering result and the first initial clustering center are updated to obtain the first user clustering result.

8. The method of claim 1, wherein, According to the first user clustering result, a first contour coefficient is determined, including: A first target user clustering result is randomly selected from the first user clustering result, and a first target sample user is selected from the first target user clustering result; A first sample internal cohesion degree of the first target sample user in the first target user clustering result is calculated; A first sample tie distance between the first target sample user and first other sample users in first other clustering results except the first target user clustering result is calculated; According to the first sample average distance and the first sample internal cohesion degree, the first contour coefficient is determined.

9. The method of claim 1, wherein, According to the first contour coefficient, a target contour coefficient is determined, including: According to the first target cell feature, a dimension reduction processing is performed to obtain a second target cell feature, and a second contour coefficient is determined according to the second target cell feature; The determination step of the second contour coefficient is repeated in turn to obtain a third contour coefficient, …, a Gth contour coefficient; From the first contour coefficient, the second contour coefficient, the third contour coefficient, …, the Gth contour coefficient, a contour coefficient maximum value is extracted to obtain the target contour coefficient.

10. The method of claim 9, wherein, According to the first target cell feature, a dimension reduction processing is performed to obtain a second target cell feature, including: In a reverse order, the features in the first target cell feature arranged at a preset position are deleted to obtain the second target cell feature.

11. The method of claim 9, wherein, According to the second target cell feature, a second contour coefficient is determined, including: construct a second original feature map of the target user according to the second target cell feature, and input the second original feature map into a preset feature map processing model to obtain a second target feature vector; perform clustering processing on the target user based on the second target feature vector to obtain a second user clustering result, and determine a second contour coefficient according to the second user clustering result.

12. The method of claim 1, wherein, determine a target user feature corresponding to the target user according to the target contour coefficient, including: determine a target cell feature corresponding to the target contour coefficient, and take the target cell feature as the target user feature.

13. The method of claim 1, wherein: The preset weight value calculation model is obtained by the following method: obtain historical cell features of historical users and feature labels corresponding to the historical users, and input the historical cell features into a network model to be trained to obtain a model prediction result; construct a first loss function and a second loss function according to the model prediction result and the feature labels, and train the network model to be trained based on the first loss function and the second loss function to obtain the preset weight value calculation model.

14. A method of classifying a cancer user, characterized by, including: obtain original predicted cell features of a user to be predicted; determine target predicted cell features from the original predicted cell features; wherein the target predicted cell features are obtained by processing the original predicted cell features by the method for determining a user feature according to any one of claims 1-12; input the target predicted cell features into a preset user classification model to obtain a user classification result of the user to be predicted.

15. An apparatus for determining a user characteristic, the apparatus comprising: including: a feature importance degree determination module configured to calculate a feature importance degree of original cell features of a target user based on a preset weight value calculation model, and sort the original cell features based on the feature importance degree to obtain first target cell features; a first feature vector determination module configured to construct a first original feature map of the target user according to the first target cell features, and input the first original feature map into a preset feature map processing model to obtain a first target feature vector; a first contour coefficient determination module configured to perform clustering processing on the target user based on the first target feature vector to obtain a first user clustering result, and determine a first contour coefficient according to the first user clustering result; a user feature determination module configured to determine a target contour coefficient according to the first contour coefficient, and determine a target user feature corresponding to the target user according to the target contour coefficient; wherein a dimension of the target user feature is lower than a dimension of the original cell features.

16. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the method for determining a user feature according to any one of claims 1-13, and the method for classifying a cancer user according to claim 14.

17. An electronic device, comprising: including: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute the method for determining a user feature according to any one of claims 1-13, and the method for classifying a cancer user according to claim 14 by executing the executable instructions.

Images