Load feature determination method, semantic model training method, device and equipment

Abstract

The embodiment of the application provides a load feature determination method, a semantic model training method, a device and equipment, the method comprises determining N label words, N is an integer greater than 1;The semantic analysis and clustering processing are carried out on the N label words, to determine a plurality of application categories, and one application category corresponds to at least one label word;According to the plurality of application categories and a plurality of instance specifications, a plurality of scene categories are determined, and the instance specification indicates the resource size of the instance, and the instance is used to run the application;Load data corresponding to the scene category is obtained, and the load feature corresponding to the scene category is determined according to the load data corresponding to the scene category. The accuracy of determining the load feature in different scenes is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and particularly relates to a load feature determination method, a semantic model training method, a device and equipment. BACKGROUND

[0002] The application can be run in an instance to realize digital intelligentization of the application. For example, the instance can be a cloud server, a virtual machine, etc. The application can be an e-commerce application, a big data application, etc.

[0003] A combination of an application category and an instance specification can correspond to a scene category, and the instance specification can be represented by a Central Processing Unit (CPU) resource and a memory resource in the instance. In actual application, whether the load of the instance is normal can be determined according to a load feature (load portrait) corresponding to the scene category. The load of the instance can be a CPU occupancy rate and a memory occupancy rate of the instance. In related technologies, the load feature corresponding to the scene category is usually determined according to expert experience. However, the load feature of the instance in some specific scenes can only be roughly determined according to the expert experience, which leads to low accuracy of determining the load feature in different scenes. SUMMARY

[0004] Aspects of the present application provide a load feature determination method, a semantic model training method, a device and equipment to improve the accuracy of determining the load feature in different scenes.

[0005] In a first aspect, an embodiment of the present application provides a load feature determination method, comprising:

[0006] determining N label words, wherein N is an integer greater than 1;

[0007] performing semantic analysis and clustering processing on the N label words to determine a plurality of application categories, and one application category corresponding to at least one label word;

[0008] determining a plurality of scene categories according to the plurality of application categories and a plurality of instance specifications, wherein the instance specification indicates a resource size of an instance, and the instance is used to run an application;

[0009] obtaining load data corresponding to the scene category, and determining a load feature corresponding to the scene category according to the load data corresponding to the scene category.

[0010] In a possible implementation, the semantic analysis and clustering processing on the N label words to determine a plurality of application categories comprises:

[0011] processing the N label words by a semantic model to obtain N semantic feature vectors corresponding to the N label words;

[0012] perform clustering processing on the N semantic feature vectors to determine the plurality of application categories.

[0013] In a possible implementation, the clustering processing on the N semantic feature vectors to determine the plurality of application categories includes:

[0014] perform clustering processing on the N semantic feature vectors by a preset algorithm to obtain a plurality of cluster sets, and each cluster set includes at least one semantic feature vector;

[0015] determine an application category corresponding to each cluster set to obtain the plurality of application categories; wherein the at least one label word corresponding to the application category includes label words corresponding to the semantic feature vectors in the cluster set corresponding to the application category.

[0016] In a possible implementation, the plurality of scene categories is determined according to the plurality of application categories and a plurality of instance specifications, including:

[0017] obtain the plurality of instance specifications;

[0018] perform combination processing on the plurality of application categories and the plurality of instance specifications to obtain the plurality of scene categories, wherein one scene category includes one application category and one instance specification.

[0019] In a possible implementation, the load data corresponding to the scene category is obtained, including:

[0020] determine a plurality of first instances according to the scene category, the instance specification of the first instance being the same as the instance specification in the scene category, and the label word corresponding to a first application running in the first instance including at least one of the label words corresponding to the application category in the scene category;

[0021] determine the load data of the plurality of first instances as the load data corresponding to the scene category.

[0022] In a possible implementation, the load feature corresponding to the scene category is determined according to the load data corresponding to the scene category, including:

[0023] determine a plurality of types of load data in the load data corresponding to the scene category;

[0024] determine a data statistical value corresponding to each type of load data;

[0025] determine the load feature corresponding to the scene category according to the data statistical value corresponding to each type of load data.

[0026] In a possible implementation, after determining the load feature corresponding to the scene category according to the load data corresponding to the scene category, the method further includes:

[0027] storing the scene category and the load feature corresponding to the scene category in a preset database.

[0028] In a possible implementation, the method further includes:

[0029] obtaining a load feature query request, wherein the load feature query request includes a first instance specification and a first label word;

[0030] querying whether a target scene category matching the first instance specification and the first label word exists in a preset database, wherein the target scene category includes the first instance specification and the first label word;

[0031] if yes, determining a load feature corresponding to the target scene category as a target load feature; and displaying the target load feature or sending the target load feature.

[0032] In a second aspect, an embodiment of the present application provides a semantic model training method, including:

[0033] obtaining instance label data, wherein the instance label data includes label data marked by a plurality of users on a plurality of instances;

[0034] determining X label words corresponding to each user according to the instance label data, wherein X is an integer greater than 1;

[0035] generating sample data corresponding to each user according to the X label words corresponding to each user, wherein the sample data includes an N-dimensional vector of each label word corresponding to the user and an N*X-dimensional vector of the X label words corresponding to the user;

[0036] performing model training according to the sample data corresponding to the plurality of users to obtain a semantic model, wherein the semantic model is used to determine a semantic feature vector of a label word.

[0037] In a possible implementation, for any one of the plurality of users, determining the X label words corresponding to the user according to the instance label data includes:

[0038] determining, in the instance label data, a plurality of label words marked by the user and a marking frequency of each label word marked by the user;

[0039] sorting the plurality of label words marked by the user in a descending order of the marking frequency of each label word marked by the user;

[0040] The first X label words in the sorted plurality of label words are determined as the X label words corresponding to the user.

[0041] In a possible implementation, for any one of the plurality of users, the sample data corresponding to the user is generated according to the X label words corresponding to the user, including:

[0042] The X label words corresponding to the user are encoded according to a preset encoding rule, to obtain X N-dimensional vectors;

[0043] The N*X-dimensional vector of the X label words is determined according to the X N-dimensional vectors.

[0044] The sample data includes the X N-dimensional vectors and the N*X-dimensional vector.

[0045] In a possible implementation, the semantic model includes an input layer, an intermediate layer and an output layer; the intermediate layer is configured to output a semantic feature vector of a label word.

[0046] In a third aspect, an embodiment of the present application provides a load feature determination apparatus, including: a first determination module, a processing module, a second determination module, a first acquisition module and a third determination module, wherein,

[0047] The first determination module is configured to determine N label words, the N being an integer greater than 1.

[0048] The processing module is configured to perform semantic analysis and clustering processing on the N label words, to determine a plurality of application categories, and one application category corresponds to at least one label word.

[0049] The second determination module is configured to determine a plurality of scene categories according to the plurality of application categories and a plurality of instance specifications, the instance specification indicating a resource size of an instance, and the instance being configured to run an application.

[0050] The first acquisition module is configured to acquire load data corresponding to the scene categories.

[0051] The third determination module is configured to determine a load feature corresponding to the scene categories according to the load data corresponding to the scene categories.

[0052] In a possible implementation, the processing module is specifically configured to:

[0053] The N label words are processed by a semantic model, to obtain N semantic feature vectors corresponding to the N label words.

[0054] The N semantic feature vectors are processed by clustering, to determine the plurality of application categories.

[0055] In a possible implementation, the processing module is specifically configured to:

[0056] perform clustering processing on the N semantic feature vectors by a preset algorithm to obtain a plurality of clustering sets, and each of the clustering sets includes at least one semantic feature vector;

[0057] determine an application category corresponding to each of the clustering sets to obtain the plurality of application categories; and wherein at least one label word corresponding to the application category includes label words corresponding to the semantic feature vectors in the clustering set corresponding to the application category.

[0058] In a possible implementation, the second determining module is specifically configured to:

[0059] obtain the plurality of instance specifications;

[0060] perform combination processing on the plurality of application categories and the plurality of instance specifications to obtain the plurality of scene categories, wherein one of the scene categories includes one of the application categories and one of the instance specifications.

[0061] In a possible implementation, the first obtaining module is specifically configured to:

[0062] determine a plurality of first instances according to the scene category, wherein an instance specification of the first instance is the same as an instance specification in the scene category, and label words corresponding to a first application running in the first instance include at least one of label words corresponding to an application category in the scene category;

[0063] determine load data of the plurality of first instances as load data corresponding to the scene category.

[0064] In a possible implementation, the third determining module is specifically configured to:

[0065] determine a plurality of types of load data in the load data corresponding to the scene category;

[0066] determine a data statistical value corresponding to each type of load data;

[0067] determine a load feature corresponding to the scene category according to the data statistical value corresponding to each type of load data.

[0068] In a possible implementation, the load feature determining apparatus further includes a storage module,

[0069] the storage module is configured to store the scene category and the load feature corresponding to the scene category in a preset database.

[0070] In a possible implementation, the load feature determination apparatus further includes a second acquisition module, a query module,

[0071] The second acquisition module is configured to acquire a load feature query request, wherein the load feature query request includes a first instance specification and a first label word.

[0072] The query module is configured to query whether a target scene category matching the first instance specification and the first label word exists in a preset database, wherein the target scene category includes the first instance specification and the first label word.

[0073] If yes, a load feature corresponding to the target scene category is determined as a target load feature; and the target load feature is displayed or sent.

[0074] In a fourth aspect, an embodiment of the present application provides a semantic model training apparatus, including an acquisition module, a determination module, a generation module, and a training module, wherein,

[0075] The acquisition module is configured to acquire instance label data, wherein the instance label data includes label data marked by a plurality of users on a plurality of instances.

[0076] The determination module is configured to determine X label words corresponding to each user according to the instance label data, wherein X is an integer greater than 1.

[0077] The generation module is configured to generate sample data corresponding to each user according to the X label words corresponding to each user, wherein the sample data includes an N-dimensional vector of each label word corresponding to the user and an N*X-dimensional vector of the X label words corresponding to the user.

[0078] The training module is configured to perform model training according to the sample data corresponding to the plurality of users to obtain a semantic model, wherein the semantic model is used to determine a semantic feature vector of a label word.

[0079] In a possible implementation, the determination module is specifically configured to:

[0080] determine, in the instance label data, a plurality of label words marked by the user and a marking frequency of each label word marked by the user;

[0081] sort the plurality of label words marked by the user in descending order of the marking frequency of each label word marked by the user;

[0082] determine the first X label words in the sorted plurality of label words as the X label words corresponding to the user.

[0083] In a possible implementation, the generating module is specifically configured to:

[0084] encode the X label words corresponding to the user according to a preset encoding rule to obtain X N-dimensional vectors;

[0085] determine an N*X dimensional vector of the X label words according to the X N-dimensional vectors;

[0086] The sample data includes the X N-dimensional vectors and the N*X dimensional vector.

[0087] In a possible implementation, the semantic model includes an input layer, an intermediate layer and an output layer; the intermediate layer is configured to output a semantic feature vector of a label word.

[0088] In a fifth aspect, an embodiment of the present application provides an electronic device, including a memory and a processor.

[0089] The memory stores computer execution instructions.

[0090] The processor executes the computer execution instructions stored in the memory, so that the processor executes the load feature determination method in any one of the first aspect.

[0091] In a sixth aspect, an embodiment of the present application provides an electronic device, including a memory and a processor.

[0092] The memory stores computer execution instructions.

[0093] The processor executes the computer execution instructions stored in the memory, so that the processor executes the semantic model training method in any one of the second aspect.

[0094] In a seventh aspect, an embodiment of the present application provides a load feature determination system, including an electronic device and a data storage device, wherein,

[0095] The electronic device is configured to execute the method in any one of the first aspect to obtain load features corresponding to a plurality of scene categories, or to query a target load feature corresponding to a target scene category in the data storage device.

[0096] The data storage device is configured to store load features corresponding to the plurality of scene categories; the data storage device can be a preset database in the first aspect.

[0097] In an eighth aspect, an embodiment of the present application provides a computer readable storage medium, the computer readable storage medium stores computer execution instructions, when the computer execution instructions are executed by a processor, the computer execution instructions are configured to implement the load feature determination method in any one of the first aspect.

[0098] In a ninth aspect, an embodiment of the present application provides a computer readable storage medium, wherein the computer readable storage medium stores computer execution instructions, and the computer execution instructions are used for implementing the model training method in any one of the second aspect when executed by a processor.

[0099] In a tenth aspect, an embodiment of the present application provides a computer program product, comprising a computer program, and the computer program is used for implementing the load feature determination method in any one of the first aspect when executed by a processor.

[0100] In an eleventh aspect, an embodiment of the present application provides a computer program product, comprising a computer program, and the computer program is used for implementing the model training method in any one of the second aspect when executed by a processor.

[0101] In the embodiment of the present application, N label words can be determined, and semantic analysis can be performed on the N label words through a semantic model, and then clustering processing is performed to determine a plurality of application categories, and then a plurality of scene categories can be determined according to the plurality of application categories and a plurality of instance specifications. A plurality of first instances can be determined according to the instance specifications in the scene categories, and load data of the first instances is obtained, and then load data corresponding to the scene categories is determined according to the load data of the plurality of first instances, and further, load features corresponding to the scene categories are determined according to the load data of each category. Since semantic analysis can be performed on the label words through the semantic model, and the label words are clustered through a preset algorithm, automatic classification of the application is realized, and the load features corresponding to each scene category can be further accurately determined, which improves the accuracy of determining the load features in different scenes compared with the rough determination of the load features of the instance in certain specific scenes according to the expert experience. BRIEF DESCRIPTION OF DRAWINGS

[0102] The accompanying drawings, which are included to provide a further understanding of the present application, constitute a part of the present application, and the illustrative embodiments of the present application and their description serve to explain the present application, and do not constitute improper limitations on the present application. In the drawings:

[0103] Figure 1 A schematic diagram of an application scenario provided for the illustrative embodiments of the present application;

[0104] Figure 2 A flowchart of a semantic model training method provided for the illustrative embodiments of the present application;

[0105] Figure 3 A structure diagram of a semantic model provided for the illustrative embodiments of the present application;

[0106] Figure 4 A flowchart of a load feature determination method provided for the illustrative embodiments of the present application;

[0107] Figure 5 A schematic diagram of a cluster set provided for an exemplary embodiment of the present application;

[0108] Figure 6 A process schematic diagram of a load feature determination method provided for an exemplary embodiment of the present application;

[0109] Figure 7 A flow schematic diagram of a query load feature provided for an exemplary embodiment of the present application;

[0110] Figure 8 A process schematic diagram of a query load feature provided for an exemplary embodiment of the present application;

[0111] Figure 9 A schematic diagram of a load feature determination system provided for an exemplary embodiment of the present application;

[0112] Figure 10 A structural schematic diagram of a load feature determination apparatus provided for an exemplary embodiment of the present application;

[0113] Figure 11 A structural schematic diagram of another load feature determination apparatus provided for an exemplary embodiment of the present application;

[0114] Figure 12 A structural schematic diagram of a semantic model training apparatus provided for an exemplary embodiment of the present application;

[0115] Figure 13 A structural schematic diagram of an electronic device provided for an exemplary embodiment of the present application;

[0116] Figure 14 A structural schematic diagram of another electronic device provided for an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0117] In order to make the objectives, technical solutions and advantages of the present application clearer, the technical solutions of the present application will be described below with reference to the accompanying drawings and specific embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0118] In order to facilitate the understanding of the technical solutions of the present application, the concepts involved in the present application will be explained and described first.

[0119] Instance: a basic unit of cloud service provided by a cloud service provider, and a basic unit of cloud service purchased by a user. The instance can include resources such as CPU, memory, network, disk, etc. For example, the instance can be a cloud server, a virtual machine, etc.

[0120] Instance specification: can refer to the basic attributes of the instance in terms of computing performance, storage performance, network performance, etc. For example, the instance specification can be 4-core CPU, 16 GB memory, 1M bandwidth, 40G system disk, etc.

[0121] Instance label data: refers to the label data marked by the user according to the actual running situation of the application to the currently used instance.

[0122] Application category: refers to the category of the application. If different applications have the same function, the application categories of the different applications are the same. For example, the application category can include a big data category, an e-commerce category, etc.

[0123] Scenario category: can be determined by the application category and the instance specification. For example, application category 1 and instance specification 1 can correspond to scenario category 1, and application category 1 and instance specification 2 can correspond to scenario category 2.

[0124] Load characteristics: can be used to evaluate the running state of the instance. For example, the load characteristics can be CPU occupancy rate 80%, memory occupancy rate 75%.

[0125] Figure 1 A schematic diagram of an application scenario provided by an exemplary embodiment of the present application is shown in FIG. 1. As shown in FIG. 1, it includes a cloud system and a preset database. Figure 1

[0126] The cloud system can include instance performance data, instance specification data, and instance label data. Based on the data in the cloud system, a plurality of scenario categories can be determined, and the load characteristics corresponding to each scenario category can be determined, and the scenario categories and the corresponding load characteristics are stored in the preset database. For example, the preset database can include scenario category 1 and corresponding load characteristics 1, scenario category 2 and corresponding load characteristics 2, scenario category 3 and corresponding load characteristics 3, …, scenario category n and corresponding load characteristics n. In actual application, the load characteristics corresponding to a certain scenario category can also be queried in the preset database.

[0127] In the related art, the load characteristics corresponding to the scenario category are usually determined according to expert experience. However, according to the expert experience, the load characteristics of the instance in some specific scenarios can only be roughly determined, resulting in low accuracy of determining the load characteristics in different scenarios.

[0128] ​In the embodiment of the present application, the label words can be processed by the semantic model and the clustering algorithm to determine a plurality of application categories, and then a plurality of scene categories can be determined according to the plurality of application categories and a plurality of instance specifications. In addition, the load characteristics corresponding to each scene category can be determined according to the load data corresponding to each scene category. Since the label words can be processed by the semantic model and the clustering algorithm, the automatic classification and clustering of the application are realized, and the load characteristics corresponding to each scene category can be accurately determined. Compared with the rough determination of the load characteristics of the instance in some specific scenes according to the expert experience, the accuracy of determining the load characteristics in different scenes is improved.

[0129] The technical solutions shown in the present application will be described in detail below through specific embodiments. It should be noted that the following embodiments can exist independently or be combined with each other. For the same or similar content, the description will not be repeated in different embodiments.

[0130] In the technical solutions of the present application, three processing processes can be included, which are the process of training the semantic model, the process of generating the load characteristics corresponding to the scene category, and the process of load characteristic query. Next, the process of training the semantic model will be described in detail. Figure 2

[0131] Figure 2 A flowchart of a semantic model training method provided by an exemplary embodiment of the present application is shown. Please refer to Figure 2 The method can include the following steps.

[0132] S201, obtaining instance label data.

[0133] The execution subject of the embodiment of the present application can be an electronic device, or a semantic model training device arranged in the electronic device. The semantic model training device can be realized by software, or realized by the combination of software and hardware. The semantic model training device can be a processor in the electronic device. For example, the electronic device can be a computer, a server, or the like. For ease of understanding, the execution subject will be taken as an electronic device in the following description.

[0134] The electronic device can obtain instance label data in a cloud system. The instance label data can include label data marked by a plurality of users on a plurality of instances, and the label data can include label words.

[0135] The label word refers to a label created by an enterprise or an individual for an instance. The instance can be classified and searched according to the label word. The label word can include Chinese, English, and the like.

[0136] ​For example, if the label marked by user 1 on instance 1 is "East China", the label marked by user 1 on instance 2 is "lightweight server", the label marked by user 2 on instance 1 is "supervisor authority", and the label marked by user 2 on instance 2 is "company", the instance label data can include "East China", "lightweight server", "supervisor authority", and "company".

[0137] S202, determining X label words corresponding to each user according to the instance label data.

[0138] In an optional embodiment, the electronic device can preprocess the instance label data, and then determine X label words corresponding to each user according to the instance label data after preprocessing. X can be a value pre-set by the staff, and X is an integer greater than 1.

[0139] Preprocessing can include data cleaning, word segmentation processing, synonym merging, etc.

[0140] Data cleaning refers to removing abnormal characters, null values, etc. in the instance label data.

[0141] Word segmentation processing can refer to processing a label word composed of more than one word through word segmentation software to obtain a label word composed of one word. For example, "running mode" can be processed by word segmentation to obtain the label words "running" and "mode".

[0142] Synonym merging refers to merging words with the same meaning. For example, if label word-1 is "company" and label word-2 is "company", label word-1 and label word-2 have the same meaning, and label word-1 and label word-2 can be merged into "company". For example, if label word-1 is "high load" and label word-2 is "high load", label word-1 and label word-2 have the same meaning, and label word-1 and label word-2 can be merged into "high load".

[0143] Since the instance label data includes label data marked by multiple users on multiple instances, for any one user, multiple label words marked by the user and the number of times each label word is marked by the user can be determined in the instance label data; the multiple label words marked by the user are sorted in descending order of the number of times each label word is marked by the user; and the first X label words in the sorted multiple label words are determined as the X label words corresponding to the user.

[0144] For example, suppose User 1 has tagged multiple instances, and the tags User 1 used for these tags include "Supervisor Permissions," "Management System," and "Orders," with "Supervisor Permissions" tagged 5 times, "Management System" tagged 8 times, and "Orders" tagged 20 times. Then, the tags tagged by this user can be sorted according to the number of times each tag is tagged. The sorting result would be: Orders, Management System, Supervisor Permissions. If X is set to 2, then the first two tags tagged by this user can be determined as the two tags corresponding to that user, namely "Orders" and "Management System."

[0145] S203. Generate sample data for each user based on the X tags corresponding to each user.

[0146] Sample data for each user can be generated as follows: X tag words corresponding to the user are encoded according to a preset encoding rule to obtain X N-dimensional vectors; based on the X N-dimensional vectors, the N*X-dimensional vectors of the X tag words are determined; wherein, the sample data includes X N-dimensional vectors and N*X-dimensional vectors.

[0147] The preset encoding rule can be one-hot encoding, which can be expressed in binary and used to represent discrete, unordered feature values.

[0148] If the total number of tag types is N, then for any tag, it can be encoded according to a preset encoding rule to obtain a corresponding N-dimensional vector. The N-dimensional vector can be represented as (n1, n2, n3, ..., nn...). n ).

[0149] For example, if the total number of tag words is 5, and the tag word "order" belongs to one of them, then the tag word "order" can be encoded using one-hot encoding, and the corresponding 5-dimensional vector can be [0, 1, 0, 0, 0].

[0150] After generating the N-dimensional vector corresponding to each tag word, an N*X-dimensional vector can be generated based on the N-dimensional vectors corresponding to the X tag words. The N*X-dimensional vector can be represented as:

[0151]

[0152] For example, if the total number of tag types is 5, the N-dimensional vector corresponding to tag-1 is [1, 0, 0, 0, 0], the N-dimensional vector corresponding to tag-2 is [0, 1, 0, 0, 0], and the N-dimensional vector corresponding to tag-3 is [0, 0, 1, 0, 0], then a 5*3 dimensional vector can be generated based on the 5-dimensional vectors corresponding to these 3 tag types, which can be represented as:

[0153]

[0154] S204, training a model according to the sample data corresponding to the plurality of users to obtain a semantic model.

[0155] Next, combined with Figure 3 , the semantic model is described.

[0156] Figure 3 The structural diagram of the semantic model provided for the exemplary embodiments of the present application is shown. Please refer to Figure 3 , the semantic model can include 3 layers, which are input layer, intermediate layer and input layer.

[0157] Among them, the input layer can be used to input X N-dimensional vectors corresponding to X label words. For example, if the label words corresponding to the user have X, among them, label word-1 corresponds to N-dimensional vector-1, label word-2 corresponds to N-dimensional vector-2, label word-3 corresponds to N-dimensional vector-3, …, label word-x corresponds to N-dimensional vector-x, the input layer can be used to input N-dimensional vector-1, N-dimensional vector-2, N-dimensional vector-3, …, N-dimensional vector-x.

[0158] The intermediate layer can be used to output the V-dimensional vector corresponding to each N-dimensional vector (V is less than N), that is, the semantic feature vector corresponding to each label word can be output. For example Figure 3 , through the intermediate layer, N-dimensional vector-1 corresponds to V-dimensional vector-1, N-dimensional vector-2 corresponds to V-dimensional vector-2, N-dimensional vector-3 corresponds to V-dimensional vector-3, …, N-dimensional vector-x corresponds to V-dimensional vector-x. Among them, V-dimensional vector-1, V-dimensional vector-2, V-dimensional vector-3, …, V-dimensional vector-x are the semantic feature vectors corresponding to label word-1, label word-2, label word-3, …, label word-x respectively.

[0159] The output layer can process X V-dimensional vectors to obtain and output N*X-dimensional vectors corresponding to X label words.

[0160] In the process of training the semantic model, the semantic model can be trained by a deep learning algorithm. For example, the deep learning algorithm can include Stochastic Gradient Descent (SGD) and Back propagation (BP) algorithm. The SGD can find the minimum value of the model by iteration. The BP algorithm can be based on the SGD, and the model is repeatedly iterated until the error of the model reaches the expected value, and the training of the model is ended.

[0161] After obtaining the semantic model, the N-dimensional vector corresponding to each label word can be processed by the semantic model to obtain the V-dimensional vector corresponding to each label word, that is, the semantic feature vector corresponding to each label word.

[0162] Alternatively, other algorithm models can be used to extract semantic features of each label word to generate a semantic feature vector corresponding to each label word. For example, a continuous bag-of-words (CBOW) model can be used to generate a semantic feature vector corresponding to each label word.

[0163] In the embodiments of the present application, instance label data can be obtained, and X label words corresponding to each user can be determined according to the instance label data. N-dimensional vectors corresponding to each label word and N*X-dimensional vectors corresponding to X label words can be generated according to the X label words corresponding to each user. The N-dimensional vectors and the N*X-dimensional vectors corresponding to the X label words corresponding to each user can be used as sample data corresponding to each user. The semantic model can be obtained by training the semantic model according to the sample data corresponding to multiple users. The semantic feature vector corresponding to each label word can be obtained by extracting semantic features of the label word through the semantic model. The accuracy of obtaining the semantic feature vector is improved.

[0164] After the semantic model is trained, the following describes the process of determining the load feature corresponding to the generated scene category. Figure 4 The load feature corresponding to the generated scene category is described.

[0165] Figure 4 A flowchart of a load feature determination method provided by an exemplary embodiment of the present application is shown. Please refer to Figure 4 The method can include the following steps.

[0166] S401, determining N label words.

[0167] The execution subject of the embodiments of the present application can be an electronic device, or a load feature determination apparatus arranged in the electronic device. The load feature determination apparatus can be implemented by software, or by a combination of software and hardware. The load feature determination apparatus can be a processor in the electronic device. The electronic device can be a computer, a server, or the like. For ease of understanding, the following describes the execution subject as an electronic device.

[0168] In an optional embodiment, N label words can be set by a staff in advance, so that the electronic device can determine the N label words.

[0169] S402, performing semantic analysis and clustering processing on the N label words to determine multiple application categories.

[0170] In an optional embodiment, the N label words can be subjected to semantic analysis by the semantic model to obtain N semantic feature vectors corresponding to the N label words; and the N semantic feature vectors can be subjected to clustering processing to determine the plurality of application categories.

[0171] For any one label word, the corresponding N-dimensional vector can be input into the semantic model, and the semantic model can be used to process the N-dimensional vector to obtain a semantic feature vector corresponding to the label word, and the semantic feature vector is a V-dimensional vector. For example, if the N-dimensional vector-1 corresponding to the label word-1 is [0, 0, 0, 0, 1], the [0, 0, 0, 0, 1] can be input into the semantic model, and after the semantic model processing, the semantic feature vector corresponding to the label word-1 can be obtained, for example, the semantic feature vector can be [0, 0, 1].

[0172] In an optional embodiment, the N semantic feature vectors can be subjected to clustering processing to determine the plurality of application categories by the following manner: the N semantic feature vectors can be subjected to clustering processing by a preset algorithm to obtain a plurality of clustering sets, and each clustering set includes at least one semantic feature vector; the application category corresponding to each clustering set can be determined to obtain the plurality of application categories; and the at least one label word corresponding to the application category includes the label word corresponding to each semantic feature vector in the clustering set corresponding to the application category.

[0173] The preset algorithm can be a clustering algorithm. For example, the preset algorithm can be a K-means algorithm, and by the K-means algorithm, the sample data can be divided into K groups, then K semantic feature vectors are randomly selected as initial seed clustering centers, and then the distance between each semantic feature vector and each seed clustering center is calculated, and each semantic feature vector is assigned to the seed clustering center closest to it. The seed clustering centers and the semantic feature vectors assigned to them represent a clustering set.

[0174] In the following, the clustering set is described in combination with Figure 5 The clustering set is described.

[0175] Figure 5 The schematic diagram of the clustering set provided by the exemplary embodiments of the present application is shown in FIG. 2. Figure 5For example, if the electronic device determines that there are 10 semantic feature vectors through the semantic model, the preset algorithm can be a K-means algorithm, the sample data can be divided into 2 groups, semantic feature vector-3 and semantic feature vector-7 can be selected as the seed clustering centers, the distances between the remaining 7 semantic feature vectors and the two seed clustering centers can be calculated, each semantic feature vector is assigned to the seed clustering center closest to it to form a cluster set. For example, semantic feature vector-1, semantic feature vector-2, semantic feature vector-4, semantic feature vector-5, semantic feature vector-8 can be assigned to semantic feature vector-3 to form cluster set-1, and semantic feature vector-6, semantic feature vector-9, semantic feature vector-10 can be assigned to semantic feature vector-3 to form cluster set-2.

[0176] Alternatively, the preset algorithm can be a Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm, an Expectation maximization (EM) algorithm based on a Gaussian Mixed Model (GMM), or the like.

[0177] After determining the plurality of cluster sets, the application categories corresponding to each cluster set can be determined to obtain a plurality of application categories.

[0178] For example, if cluster set-1 includes semantic feature vector-1, semantic feature vector-2, and semantic feature vector-3, semantic feature vector-1 corresponds to the semantic “data”, semantic feature vector-2 corresponds to the semantic “calculation”, and semantic feature vector-2 corresponds to the semantic “analysis”, the application category-1 corresponding to cluster set-1 can be determined to be a data calculation application, and the label words corresponding to application category-1 can include “data”, “calculation”, “analysis”, and the like. If cluster set-2 includes semantic feature vector-4, semantic feature vector-5, semantic feature vector-6, and semantic feature vector-7, the label words corresponding to semantic feature vector-4 are “short video”, the label words corresponding to semantic feature vector-5 are “play”, the label words corresponding to semantic feature vector-6 are “live broadcast”, and the label words corresponding to semantic feature vector-6 are “attention amount”, the application category-2 corresponding to cluster set-2 can be determined to be a short video application, and the label words corresponding to application category-2 can include “short video”, “play”, “live broadcast”, “attention amount”, and the like.

[0179] S403, determining a plurality of scene categories according to the plurality of application categories and the plurality of instance specifications.

[0180] An instance can be used to run an application. For example, a data computing application can be run on instance-1.

[0181] Each instance has a corresponding instance specification. The electronic device can obtain a plurality of instance specifications, which can indicate the resource size of the instance. For example, the instance specification can be a 4-core CPU, 16 GB of memory, etc.

[0182] The plurality of application categories and the plurality of instance specifications can be combined to obtain a plurality of scene categories, wherein a scene category includes an application category and an instance specification.

[0183] For example, if there are 3 application categories, namely application category-1, application category-2, and application category-3, and there are 4 instance specifications, namely instance specification-1, instance specification-2, instance specification-3, instance specification-4, and instance specification-5, then application category-1 can be combined with the 5 instance specifications respectively to obtain 5 scene categories, namely scene category-1, …, and scene category-5; application category-2 can be combined with the 5 instance specifications respectively to obtain 5 scene categories, namely scene category-6, …, and scene category-10; and application category-3 can be combined with the 5 instance specifications respectively to obtain 5 scene categories, namely scene category-11, …, and scene category-15.

[0184] S404, obtain load data corresponding to the scene category, and determine load characteristics corresponding to the scene category according to the load data corresponding to the scene category.

[0185] In an optional embodiment, according to the scene category, a plurality of first instances can be determined, the instance specification of the first instance is the same as the instance specification in the scene category, and the label words corresponding to the first application running in the first instance include at least one of the label words corresponding to the application category in the scene category; the load data of the plurality of first instances is determined as the load data corresponding to the scene category.

[0186] For example, if the application category-1 in the scenario category-1 is a big data computing application, the corresponding label words are big data, computing, analysis, statistics, data processing, etc.; the instance specification-1 included in the scenario category-1 is 4-core CPU, 16 GB memory, 1M bandwidth, and 40G system disk, then according to the instance specification-1, the first instance with the same instance specification can be determined in all instances. Assuming that the first instance has 5 instances, namely instance-1, instance-2, instance-3, instance-4, and instance-5, the instance specifications of the 5 instances are all 4-core CPU, 16 GB memory, 1M bandwidth, and 40G system disk. The instance-1, instance-2, instance-3, instance-4, and instance-5 respectively include at least one of the label words (big data, computing, analysis, statistics, and data processing), for example, the label words corresponding to the first application running in the instance-1 can include big data and computing; the label words corresponding to the first application running in the instance-2 can include big data, analysis, and statistics, etc.

[0187] For any one first instance, when the first application is running in the first instance, the generated load data is different. For example, when the instance-1 runs the first application, the load data can be CPU occupancy rate 80% and memory occupancy rate 75%; when the instance-2 runs the first application, the load data can be CPU occupancy rate 90% and memory occupancy rate 80%. If the first instance corresponding to the scenario category-1 includes the instance-1 and the instance-2, the load data of the instance-1 and the instance-2 can be determined as the load data corresponding to the scenario category-1.

[0188] In an optional embodiment, multiple types of load data can be determined in the load data corresponding to the scenario category; a data statistical value corresponding to each type of load data is determined; and a load feature corresponding to the scenario category is determined according to the data statistical value corresponding to each type of load data. The data statistical value can be an average value, a maximum value, a minimum value, or a percentile value, etc.

[0189] For example, assuming that the data statistical value is an average value, the load data corresponding to the scenario category-1 includes the load data of the instance-1 (CPU occupancy rate 80% and memory occupancy rate 75%) and the load data of the instance-2 (CPU occupancy rate 90% and memory occupancy rate 80%), then multiple types of load data can be determined in the load data. For example, 2 types of load data can be determined, wherein the first type of load data is CPU occupancy rate and the second type of load data is memory occupancy rate. The average values of the 2 types of load data can be determined, that is, the average value of the CPU occupancy rate can be determined as 85% and the average value of the memory occupancy rate can be determined as 77.5%. The CPU occupancy rate 85% and the memory occupancy rate 77.5% can be determined as the load feature corresponding to the scenario category-1.

[0190] Optionally, the load characteristics corresponding to the scene category can also be determined according to the data that can be counted, such as the maximum value, the minimum value, and the change period of each type of load data.

[0191] After the scene category and the load characteristics corresponding to the scene category are determined, the scene category and the load characteristics corresponding to the scene category can be stored in a preset database.

[0192] In the embodiments of the present application, N label words can be determined, and the N label words are subjected to semantic analysis through a semantic model, and then clustering processing is performed to determine a plurality of application categories, and then a plurality of scene categories can be determined according to the plurality of application categories and a plurality of instance specifications. A plurality of first instances can be determined according to the instance specifications in the scene categories, and load data of the first instances is obtained, and then the load data of the plurality of first instances is used to determine the load data corresponding to the scene category, and further the load characteristics corresponding to the scene category are determined according to each type of load data. Since the label words can be subjected to semantic analysis through the semantic model, and the label words are clustered through a preset algorithm, the automatic classification of the applications is realized, and the load characteristics corresponding to each scene category can be further accurately determined. Compared with the rough determination of the load characteristics of the instances in certain specific scenes according to the expert experience, the accuracy of determining the load characteristics in different scenes is improved.

[0193] Next, the load characteristic determination method will be further described through specific examples in combination with Figure 6

[0194] Figure 6 The process schematic diagram of the load characteristic determination method provided by the exemplary embodiments of the present application is shown in FIG. 1. Figure 6 Suppose that the electronic device can determine 5 label words, which are label word-1, label word-2, label word-3, label word-4, and label word-5. The 5 label words can be encoded through one-hot encoding to obtain corresponding N-dimensional vectors, which are N-dimensional vector-1, N-dimensional vector-2, N-dimensional vector-3, N-dimensional vector-4, and N-dimensional vector-5.

[0195] The electronic device can perform semantic analysis on the 5 N-dimensional vectors through a semantic model to obtain corresponding semantic feature vectors, which are semantic feature vector-1, semantic feature vector-2, semantic feature vector-3, semantic feature vector-4, and semantic feature vector-5. The 5 semantic feature vectors can be subjected to clustering processing through a preset algorithm. Semantic feature vector-1 and semantic feature vector-2 can be classified into one category to determine application category-1, and semantic feature vector-3, semantic feature vector-4, and semantic feature vector-5 can be classified into one category to determine application category-2.

[0196] ​If the set instance specifications are three, instance specification-1, instance specification-2, and instance specification-3, the application category-1 can be combined with the three instance specifications respectively to obtain three scene categories, scene category-1, scene category-2, and scene category-3. The application category-2 can be combined with the three instance specifications respectively to obtain three scene categories, scene category-4, scene category-5, and scene category-6.

[0197] For any one scene category, the first instance with the same instance specification can be determined in the data from the cloud system according to the instance specification included in the scene category. For example, the scene category-3 is combined by the application category-1 and the instance specification-3, and the label words corresponding to the application category-1 in the scene category-3 include the label word-1 and the label word-2. The electronic device can determine the first instance with the same instance specification in the data of the cloud system according to the instance specification-3, which are instance-2, instance-3, and instance-4. The multiple label words corresponding to the application running in the instance-2 include the label word-1, the multiple label words corresponding to the application running in the instance-3 include the label word-2, and the multiple label words corresponding to the application running in the instance-4 include the label word-1. Figure 6

[0198] After determining the instance-2, instance-3, and instance-4, the electronic device can obtain the load data corresponding to the three instances respectively. For example, the load data-2 corresponding to the instance-2 is CPU occupancy rate 90% and memory occupancy rate 80%; the load data-3 corresponding to the instance-3 is CPU occupancy rate 88% and memory occupancy rate 76%; and the load data-4 corresponding to the instance-4 is CPU occupancy rate 80% and memory occupancy rate 72%.

[0199] The electronic device can determine the load data-2, load data-3, and load data-4 as the load data corresponding to the scene category-3, that is, the load data corresponding to the scene category-3 includes CPU occupancy rate 90%, memory occupancy rate 80%, CPU occupancy rate 88%, memory occupancy rate 76%, CPU occupancy rate 80%, and memory occupancy rate 72%. Two types of load data can be determined in the load data corresponding to the scene category-3, which are CPU occupancy rate and memory occupancy rate. Further, assuming that the data statistical value is the average value, the data average value of the CPU occupancy rate can be determined as 86%, and the data average value of the memory occupancy rate can be determined as 76%. The data average value of the CPU occupancy rate 86% and the data average value of the memory occupancy rate 76% can be determined as the load characteristics corresponding to the scene category-3.

[0200] ​In the embodiment of the present application, N label words can be determined, and the N label words are subjected to semantic analysis through a semantic model, and then subjected to clustering processing to determine a plurality of application categories, and then a plurality of scene categories can be determined according to the plurality of application categories and a plurality of instance specifications. A plurality of first instances can be determined according to the instance specifications in the scene categories, and load data of the first instances is acquired, and then load data corresponding to the scene categories is determined according to the load data of the plurality of first instances, and further, load features corresponding to the scene categories are determined. Since the label words are subjected to semantic analysis through the semantic model, and the label words are subjected to clustering through a preset algorithm, automatic classification of the applications is realized, and the load features corresponding to each scene category can be further accurately determined, which improves the accuracy of determining the load features in different scenes compared with roughly determining the load features of the instances in certain specific scenes according to expert experience.

[0201] After determining each scene category and the corresponding load feature, the user can also query the load feature. Next, the process of querying the load feature is described in combination with Figure 7 the process of querying the load feature is described in combination with

[0202] Figure 7 a flowchart of querying the load feature provided by the exemplary embodiments of the present application. Please refer to Figure 7 The method can include the following steps.

[0203] S701, acquiring a load feature query request.

[0204] The load feature query request input by the user can be acquired, and the load feature query request can include a first instance specification and a first label word. For example, the load feature query request can include "4-core CPU" and "data processing".

[0205] S702, querying whether a target scene category matching the first instance specification and the first label word exists in a preset database.

[0206] The preset database stores a plurality of scene categories and load features corresponding to the scene categories. For any one scene category, one application category and one instance specification are included in one scene category. Each application category has a corresponding label word, that is, each scene category has a corresponding label word.

[0207] The first instance specification and the first label word in the load feature query request can be used to query whether a target scene category that can be matched exists in the preset database, and the target scene category can include the first instance specification and the first label word.

[0208] If yes, S703 is performed; if no, S704 is performed.

[0209] For example, if the load feature query request can include "4-core CPU" and "data processing", the preset database can be matched according to "4-core CPU", and 10 scenario categories with the same instance specification are queried out; then, the scenario categories with the same label word are queried out from the 10 scenario categories according to "data processing", and 1 scenario category is queried out, the scenario category can be determined as the target scenario category, and S703 can be executed; if there is no target scenario category, S704 can be executed.

[0210] S703, determining the load feature corresponding to the target scenario category as the target load feature.

[0211] After the target scenario category is determined, the load feature corresponding to the target scenario category can be determined as the target load feature. For example, if the target scenario category is determined as scenario category-1, and the load feature corresponding to scenario category-1 can be CPU occupancy rate of 86% and memory occupancy rate of 76%, the target load feature can be determined as CPU occupancy rate of 86% and memory occupancy rate of 76%.

[0212] Optionally, after the target load feature is determined, the target load feature can be displayed or sent.

[0213] For example, if the execution subject of the embodiment shown in Figure 7 is a computer, after the target load feature is determined, the computer can display the target load feature. If the execution subject of the embodiment shown in Figure 7 is a server, the server can send the target load feature to the client (for example, a computer).

[0214] S704, determining that there is no target load feature.

[0215] If no target scenario category is matched in the preset database according to the load feature query request, it can be determined that there is no target load feature. For example, if the execution subject of the embodiment shown in Figure 7 is a computer, the computer can also display "no matching to your query request" or "no target load feature". If the execution subject of the embodiment shown in Figure 7 is a server, the server can also send prompt information to the client (for example, a computer), and the prompt information can be "no matching to your query request" or "no target load feature".

[0216] In the embodiment of the present application, the load feature query request can be acquired, and whether there is a target scene category matching the first instance specification and the first label word in the preset database is queried according to the first instance specification and the first label word in the load feature query request. If yes, the load feature corresponding to the target scene category is determined as the target load feature; if no, it is displayed that there is no target load feature. Since whether there is a target scene category that can be matched in the preset database is queried according to the load feature query request, and the target load feature is determined, the accuracy and efficiency of the user to determine the target load feature are improved.

[0217] In the following, the query load feature is further described through specific examples. Figure 8

[0218] Figure 8 The process diagram of the query load feature provided by the exemplary embodiments of the present application is shown. Please refer to Figure 8 , which includes interface 801, a preset database and interface 802. The interface 801 and the interface 802 are display interfaces of electronic devices (for example, computers).

[0219] Please refer to the interface 801, the user can input the first instance specification and the first label word, so that the electronic device acquires the load feature query request, such as Figure 8 In the embodiment, the user can input “4-core CPU” and “data processing” so that the electronic device can query in the preset database according to “4-core CPU” and “data processing”.

[0220] The preset database stores a plurality of scene categories and load features corresponding to the scene categories. For any one scene category, a scene category includes an application category and an instance specification. Each application category has a corresponding label word, that is, each scene category has a corresponding label word. For example, the electronic device can query that there is a target scene category that can be matched in the preset database according to “4-core CPU” and “data processing”, which is scene category-3, and scene category-3 is a data analysis scene, and the corresponding load feature-3 includes CPU utilization of 86%, memory occupation rate of 76%, bandwidth utilization of 80%, etc. Therefore, the electronic device can determine the load feature-3 corresponding to the scene category-3 as the target load feature, and display as shown in the interface 802.

[0221] ​In the embodiment of the present application, the load feature query request can be acquired, and whether there is a target scene category matching the first instance specification and the first label word in the preset database is queried according to the first instance specification and the first label word in the load feature query request. If yes, the target load feature corresponding to the target scene category is determined as the target load feature; if no, it is displayed that there is no target load feature. Since whether there is a target scene category that can be matched in the preset database is queried according to the load feature query request, and the target load feature is determined, the accuracy of the user in determining the target load feature is improved.

[0222] Based on any one of the above embodiments, the present application provides a load feature determination system. In the following, the load feature determination system is described in combination with Figure 9 .

[0223] Figure 9 A schematic diagram of a load feature determination system provided by an exemplary embodiment of the present application is shown in FIG. 1. Referring to FIG. 1, the load feature determination system includes an electronic device and a data storage device, and the electronic device and the data storage device can communicate with each other. Figure 9

[0224] The electronic device can be used to receive data from a cloud system, including instance label data, instance specification data, and instance load feature data. The electronic device can process the data of the cloud system to obtain load features corresponding to a plurality of scene categories.

[0225] Optionally, the electronic device can be a computer, a server, or the like. The electronic device can perform the semantic model training method, the load feature determination method, and the load feature query method in the above method embodiments.

[0226] The data storage device can be used to store a plurality of scene categories and corresponding load features.

[0227] It should be noted that the data storage device can be the preset database in any one of the above embodiments.

[0228] Figure 10 A structural schematic diagram of a load feature determination apparatus provided by an exemplary embodiment of the present application is shown in FIG. 2. Referring to FIG. 2, the load feature determination apparatus 10 includes a first determination module 11, a processing module 12, a second determination module 13, a first acquisition module 14, and a third determination module 15, wherein Figure 10

[0229] The first determination module 11 is configured to determine N label words, and N is an integer greater than 1.

[0230] The processing module 12 is configured to perform semantic analysis and clustering processing on the N label words to determine a plurality of application categories, and one application category corresponds to at least one label word.​​

[0231] The second determining module 13 is configured to determine a plurality of scene categories according to the plurality of application categories and a plurality of instance specifications, the instance specification indicating a resource size of an instance, the instance being used to run an application;

[0232] The first obtaining module 14 is configured to obtain load data corresponding to the scene category.

[0233] The third determining module 15 is configured to determine a load feature corresponding to the scene category according to the load data corresponding to the scene category.

[0234] The load feature determining apparatus provided by the embodiments of the present application can implement the technical solutions shown in the method embodiments, and the implementation principles and beneficial effects are similar, which will not be repeated here.

[0235] In a possible implementation, the processing module 12 is specifically configured to:

[0236] The N label words are processed through a semantic model to obtain N semantic feature vectors corresponding to the N label words.

[0237] The N semantic feature vectors are processed through clustering to determine the plurality of application categories.

[0238] In a possible implementation, the processing module 12 is specifically configured to:

[0239] The N semantic feature vectors are processed through clustering through a preset algorithm to obtain a plurality of clustering sets, at least one semantic feature vector being included in each clustering set.

[0240] An application category corresponding to each clustering set is determined to obtain the plurality of application categories, wherein at least one label word corresponding to the application category includes a label word corresponding to each semantic feature vector in the clustering set corresponding to the application category.

[0241] In a possible implementation, the second determining module 13 is specifically configured to:

[0242] The plurality of instance specifications are obtained.

[0243] The plurality of application categories and the plurality of instance specifications are processed through combination to obtain the plurality of scene categories, wherein one scene category includes one application category and one instance specification.

[0244] In a possible implementation, the first obtaining module 14 is specifically configured to:

[0245] According to the scene category, a plurality of first instances are determined, instance specifications of the first instances are same as instance specifications in the scene category, and a first application running in the first instances corresponds to at least one label word in label words corresponding to an application category in the scene category.

[0246] Load data of the plurality of first instances is determined as load data corresponding to the scene category.

[0247] In a possible implementation, the third determining module 15 is specifically configured to:

[0248] In the load data corresponding to the scene category, a plurality of types of load data are determined.

[0249] Data statistical values corresponding to each type of load data are determined.

[0250] According to the data statistical values corresponding to each type of load data, load features corresponding to the scene category are determined.

[0251] The load feature determining apparatus provided by the embodiments of the present application can perform the technical solutions shown in the method embodiments, and the implementation principles and beneficial effects are similar, which will not be repeated here.

[0252] Figure 11 Another load feature determining apparatus provided by the exemplary embodiments of the present application is shown in a structural schematic diagram. Please refer to Figure 11 On the basis of the embodiments shown in Figure 10 The load feature determining apparatus further includes a storage module 16:

[0253] The storage module 16 is configured to store the scene category and the load features corresponding to the scene category in a preset database.

[0254] In a possible implementation, the load feature determining apparatus further includes a second obtaining module 17 and a querying module 18.

[0255] The second obtaining module 17 is configured to obtain a load feature query request, and the load feature query request includes a first instance specification and a first label word.

[0256] The querying module 18 is configured to query whether there is a target scene category matching the first instance specification and the first label word in the preset database, and the target scene category includes the first instance specification and the first label word.

[0257] If yes, load features corresponding to the target scene category are determined as target load features; the target load features are displayed or sent.

[0258] The load feature determination apparatus provided in the embodiments of the present application can execute the technical solutions shown in the method embodiments, and has similar implementation principles and beneficial effects, which will not be repeated here.

[0259] Figure 12 A structural schematic diagram of a semantic model training apparatus provided for the exemplary embodiments of the present application is provided. Please refer to Figure 12 The semantic model training apparatus 20 comprises an acquisition module 21, a determination module 22, a generation module 23, and a training module 24, wherein

[0260] The acquisition module 21 is configured to acquire instance label data, wherein the instance label data comprises label data marked by a plurality of users on a plurality of instances.

[0261] The determination module 22 is configured to determine X label words corresponding to each user according to the instance label data, wherein X is an integer greater than 1.

[0262] The generation module 23 is configured to generate sample data corresponding to each user according to the X label words corresponding to each user, wherein the sample data comprises an N-dimensional vector of each label word corresponding to the user and an N*X-dimensional vector of the X label words corresponding to the user.

[0263] The training module 24 is configured to perform model training according to the sample data corresponding to the plurality of users to obtain a semantic model, wherein the semantic model is configured to determine a semantic feature vector of a label word.

[0264] The semantic model training apparatus provided in the embodiments of the present application can execute the technical solutions shown in the method embodiments, and has similar implementation principles and beneficial effects, which will not be repeated here.

[0265] In a possible implementation, the determination module 22 is specifically configured to:

[0266] determine a plurality of label words marked by the user and the number of times of marking each label word by the user in the instance label data;

[0267] sort the plurality of label words marked by the user according to the number of times of marking each label word by the user from large to small;

[0268] determine the first X label words in the sorted plurality of label words as the X label words corresponding to the user.

[0269] In a possible implementation, the generation module 23 is specifically configured to:

[0270] encode the X label words corresponding to the user according to a preset encoding rule to obtain X N-dimensional vectors;

[0271] determining an N*X dimensional vector of the X label words according to the X N dimensional vectors;

[0272] wherein the sample data comprises the X N dimensional vectors and the N*X dimensional vector.

[0273] In a possible implementation, the semantic model comprises an input layer, an intermediate layer and an output layer; wherein the intermediate layer is configured to output a semantic feature vector of a label word.

[0274] The semantic model training apparatus provided by the embodiments of the present application can implement the technical solutions shown in the method embodiments, and the implementation principles and beneficial effects are similar, which will not be repeated here.

[0275] An example structural diagram of an electronic device is provided in the example embodiments of the present application, please refer to Figure 13 The electronic device 30 can include a processor 31 and a memory 32. Exemplarily, the processor 31, the memory 32, and each part are connected to each other through a bus 33.

[0276] The memory 32 stores computer execution instructions;

[0277] The processor 31 executes the computer execution instructions stored in the memory 32, so that the processor 31 executes the load feature determination method shown in the above method embodiments.

[0278] The electronic device can also execute the load feature query method shown in the above method embodiments.

[0279] Another example structural diagram of an electronic device is provided in the example embodiments of the present application, please refer to Figure 14 The electronic device 40 can include a processor 41 and a memory 42. Exemplarily, the processor 41, the memory 42, and each part are connected to each other through a bus 43.

[0280] The memory 42 stores computer execution instructions;

[0281] The processor 41 executes the computer execution instructions stored in the memory 42, so that the processor 31 executes the semantic model training method shown in the above method embodiments.

[0282] Figures 13-14 The electronic devices shown in the embodiments can be the same electronic device or different electronic devices.

[0283] Correspondingly, the example embodiments of the present application provide a computer readable storage medium, which stores computer execution instructions, and when the computer execution instructions are executed by a processor, the computer execution instructions are used to implement the load feature determination method shown in the above method embodiments.

[0284] Accordingly, the embodiments of the present application can also provide a computer program product comprising a computer program which, when executed by a processor, can implement the semantic model training method shown in the above method embodiments.

[0285] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) containing computer-usable program code.

[0286] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as a combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus generate a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure One one or more flows and / or blocks Figure One means for performing the functions specified in the flowchart

[0287] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction means, which implements the functions specified in the flowcharts and / or block diagrams. Figure One one or more flows and / or blocks Figure One means for performing the functions specified in the flowchart

[0288] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable data processing apparatus to produce a computer-implemented process, so that the instructions executed on the computer or other programmable data processing apparatus provide a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure One one or more flows and / or blocks Figure One means for performing the functions specified in the flowchart

[0289] In a typical configuration, the computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memories.

[0290] Memory can include non-persistent memory, such as volatile random access memory (RAM) and / or non-volatile memory, such as read only memory (ROM), electrically erasable read only memory (EEPROM), flash memory, or the like. Memory is an example of computer readable media.

[0291] Computer readable media includes permanent and non-permanent, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital versatile discs (DVDs) or other optical storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer readable media does not include transitory media, such as modulated data signals and carrier waves.

[0292] It should also be noted that the terms "comprising", "containing", or any other similar term are intended to encompass non-exclusive inclusions such that a process, method, article, or apparatus that comprises a list of elements does not include those elements solely, but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0293] The above description is merely illustrative of the embodiments of the present application and is not intended to limit the scope of the application. Any modifications, equivalent replacements, improvements, and the like of the present application, which are within the spirit and principle of the present application, should be included in the scope of claims of the present application.

Claims

1. A load signature determination method, characterized by, The method comprises the following steps: determining N label words, N being an integer greater than 1; performing semantic analysis and clustering processing on the N label words to determine a plurality of application categories, one of the application categories corresponding to at least one label word; determining a plurality of scenario categories according to the plurality of application categories and a plurality of instance specifications, the instance specification indicating a resource size of an instance used for running an application; obtaining load data corresponding to the scenario category, and determining load characteristics corresponding to the scenario category according to the load data corresponding to the scenario category; the semantic analysis and clustering processing on the N label words to determine a plurality of application categories comprises: processing the N label words through a semantic model to obtain N semantic feature vectors corresponding to the N label words; performing clustering processing on the N semantic feature vectors to determine the plurality of application categories; the clustering processing on the N semantic feature vectors to determine the plurality of application categories comprises: performing clustering processing on the N semantic feature vectors through a preset algorithm to obtain a plurality of clustering sets, at least one semantic feature vector being included in each clustering set; determining an application category corresponding to each clustering set to obtain the plurality of application categories; wherein the at least one label word corresponding to the application category comprises label words corresponding to the semantic feature vectors in the clustering set corresponding to the application category.

2. The method of claim 1, wherein, determining a plurality of scenario categories according to the plurality of application categories and a plurality of instance specifications comprises: obtaining the plurality of instance specifications; performing combination processing on the plurality of application categories and the plurality of instance specifications to obtain the plurality of scenario categories, wherein one of the scenario categories comprises one of the application categories and one of the instance specifications.

3. The method according to claim 1 or 2, characterized in that, obtaining load data corresponding to the scenario category comprises: determining a plurality of first instances according to the scenario category, the instance specification of the first instance being the same as the instance specification in the scenario category, and the label word corresponding to a first application running in the first instance comprising at least one of the label words corresponding to the application categories in the scenario category; determining the load data of the plurality of first instances as the load data corresponding to the scenario category.

4. The method according to claim 1 or 2, characterized in that, determining load characteristics corresponding to the scenario category according to the load data corresponding to the scenario category comprises: determining a plurality of types of load data in the load data corresponding to the scenario category; determining a data statistical value corresponding to each type of load data; determining the load characteristics corresponding to the scenario category according to the data statistical value corresponding to each type of load data.

5. The method according to claim 1 or 2, characterized in that, After determining the load characteristics corresponding to the scenario category according to the load data corresponding to the scenario category, the method further comprises: storing the scenario category and the load characteristics corresponding to the scenario category in a preset database.

6. The method of claim 1 or 2, wherein, The method further comprises: obtaining a load characteristic query request, the load characteristic query request comprising a first instance specification and a first label word; querying whether a target scenario category matching the first instance specification and the first label word exists in the preset database, the target scenario category comprising the first instance specification and the first label word; If yes, a load feature corresponding to the target scene category is determined as a target load feature; and the target load feature is displayed or transmitted. 7.A semantic model training method, characterized in that, The method comprises: obtaining instance label data, the instance label data comprising label data marked by a plurality of users on a plurality of instances; determining X label words corresponding to each user according to the instance label data, X being an integer greater than 1; According to the X label words corresponding to each user, sample data corresponding to each user is generated, the sample data including an N-dimensional vector of each label word corresponding to the user and an N-dimensional vector of the X label words corresponding to the user X-dimensional vector training a semantic model according to sample data corresponding to the plurality of users, the semantic model being used to determine a semantic feature vector of a label word in the method of any one of claims 1-6.

8. The method of claim 7, wherein, For any one of the plurality of users, determining X label words corresponding to the user according to the instance label data comprises: determining, in the instance label data, a plurality of label words marked by the user and a marking frequency of each label word marked by the user; sorting the plurality of label words marked by the user in descending order of the marking frequency of each label word marked by the user; determining the first X label words in the sorted plurality of label words as the X label words corresponding to the user.

9. The method according to claim 7 or 8, characterized in that, For any one of the plurality of users, generating sample data corresponding to the user according to the X label words corresponding to the user comprises: encoding the X label words corresponding to the user according to a preset encoding rule to obtain X N-dimensional vectors; determining an N-dimensional vector of the X tags based on the X N-dimensional vectors X-dimensional vector; wherein the sample data comprises the X N-dimensional vectors and the N X-dimensional vector.

10. The method according to claim 7 or 8, characterized in that, the semantic model comprises an input layer, an intermediate layer and an output layer; wherein the intermediate layer is used to output a semantic feature vector of a label word.

11. An electronic device, comprising: The method comprises: a memory and a processor; the memory stores computer execution instructions; the processor executes the computer execution instructions stored in the memory, so that the processor executes the load feature determination method of any one of claims 1-6 or the semantic model training method of any one of claims 7-10.

12. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer execution instructions, which, when executed by the processor, are used to implement the load feature determination method of any one of claims 1-6 or the semantic model training method of any one of claims 7-10.

Images