CN113627382B - User behavior identification method and system for video conference system and storage medium - Google Patents
User behavior identification method and system for video conference system and storage medium Download PDFInfo
- Publication number
- CN113627382B CN113627382B CN202110971140.3A CN202110971140A CN113627382B CN 113627382 B CN113627382 B CN 113627382B CN 202110971140 A CN202110971140 A CN 202110971140A CN 113627382 B CN113627382 B CN 113627382B
- Authority
- CN
- China
- Prior art keywords
- goblet
- individual
- ascidian
- user behavior
- machine learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Two-Way Televisions, Distribution Of Moving Picture Or The Like (AREA)
Abstract
The invention relates to a user behavior identification method and system for a video conference system and a storage medium. According to the user behavior identification method facing the video conference system, Cauchy distributed random numbers and logistic stetty chaotic coefficients are introduced into the goblet sea squirt group algorithm, and the diversity of population individuals in the searching process of the goblet sea squirt group algorithm can be increased. In addition, the data characteristics and the machine learning algorithm hyper-parameters are simultaneously optimized and selected by adopting the enhanced goblet sea squirt group algorithm based on the K-fold cross validation strategy, and after the selection result of the data characteristics and the selection result of the machine learning algorithm hyper-parameters are obtained, the user behavior identification model is established based on the selection result of the data characteristics and the selection result of the machine learning algorithm hyper-parameters, so that the identification performance of the user behavior identification model established based on the enhanced goblet sea squirt group algorithm can be improved, and the user behavior identification result in the video conference system is more accurate.
Description
Technical Field
The invention relates to the technical field of user behavior identification, in particular to a user behavior identification method and system for a video conference system and a storage medium.
Background
With the rapid development of network communication and economy, the demand of the society for the video conference system is increasing, and the video conference system is widely applied to various scenes due to the advantages of accelerating information transfer, improving working efficiency, reducing cooperation cost and the like. Particularly, the video conference system can not be supported in remote live broadcasting announcement, remote conference cooperation, remote online teaching and the like. The core of the video conference system is to provide stable and smooth communication service for users, and the stable and smooth communication service cannot reasonably distribute resources such as networks and the like. In order to realize reasonable allocation of resources such as networks, the actual behavior of the user needs to be identified, and different resources are allocated according to different user behaviors. Therefore, how to identify the actual behavior of the user by using the network traffic information in the video conference has important research value.
For user behavior recognition based on network traffic data, machine learning techniques are widely used due to their flexible and intelligent characteristics. For the machine learning method, the input data characteristics and the machine learning algorithm hyper-parameters are two important factors influencing the machine learning modeling performance, and the redundant and irrelevant input characteristics and the improper algorithm hyper-parameters reduce the machine learning modeling performance to a certain extent. The data characteristics used for describing the network flow information are numerous, the algorithm hyper-parameter selection space is large, the data characteristics and the algorithm hyper-parameter space are mutually influenced and closely connected, and the optimization algorithm is performed on the data characteristics and the algorithm hyper-parameters separately, so that the optimal modeling performance is difficult to realize, and therefore, the optimization selection needs to be performed on the input data characteristics and the algorithm hyper-parameters simultaneously to realize the high-precision user behavior identification effect.
In recent years, the group intelligent optimization algorithm has attracted extensive attention in academia and industry due to its good global optimization capability and optimization efficiency, and is applied to solve the problem of feature selection of a machine learning method. The goblet sea squirt group algorithm is an optimization algorithm for simulating the predation behavior of the goblet sea squirts, and is applied to various fields due to the characteristics of simple implementation and high convergence speed. However, when the method is used for solving a high-dimensional complex optimization problem, the algorithm may be trapped in local optimization, so that the modeling performance of the machine learning algorithm is reduced, and the user behavior recognition result is inaccurate. Therefore, a new optimization algorithm is urgently needed to reduce the possibility of falling into local optimization, so that a feature set and algorithm hyper-parameters capable of obtaining optimal modeling performance are selected in a unified space of data features and algorithm hyper-parameters, the modeling performance of a machine learning algorithm is effectively improved, and the accuracy of user behavior identification is improved.
Disclosure of Invention
The invention aims to provide a user behavior identification method, a user behavior identification system and a storage medium for a video conference system, which can effectively improve the modeling performance of a machine learning algorithm and improve the accuracy of user behavior identification.
In order to achieve the purpose, the invention provides the following scheme:
a user behavior identification method facing a video conference system comprises the following steps:
acquiring network flow data under a user video conference; the traffic data includes: network state data, protocol analysis data, and user behavior information corresponding to the network state data and the protocol analysis data; the network status data includes: throughput, packet traffic, and delay jitter; the protocol analysis data includes: protocol type, protocol packet length, connection duration, port information, and IP information; the user behavior information includes: a video communication conference behavior, a voice chat conference behavior, an online lecture speaking behavior and an online listening and speaking behavior;
acquiring a machine learning algorithm according to the flow data, and simultaneously performing optimization selection on the data characteristics and the hyper-parameters of the machine learning algorithm by adopting an enhanced goblet sea squirt group algorithm based on a K-fold cross validation strategy to obtain a selection result of the data characteristics and a selection result of the hyper-parameters of the machine learning algorithm; the enhanced goblet sea squirt group algorithm is a goblet sea squirt group algorithm introduced with Cauchy distributed random numbers and logistic stetty chaotic coefficients;
establishing a user behavior recognition model based on the selection result of the data characteristics and the selection result of the hyper-parameters of the machine learning algorithm;
and inputting the network state data and the protocol analysis data of the user to be detected into the user behavior recognition model to obtain the behavior information of the user to be detected output by the user behavior recognition model.
Preferably, the obtaining of the machine learning algorithm according to the flow data, and performing optimization selection on the data features and the hyper-parameters of the machine learning algorithm by using the enhanced goblet sea squirt group algorithm based on the K-fold cross validation strategy to obtain the selection results of the data features and the selection results of the hyper-parameters of the machine learning algorithm specifically include:
acquiring a machine learning algorithm;
initializing parameters of the enhanced goblet sea squirt group algorithm; parameters of the enhanced goblet sea squirt group algorithm comprise the number of groups, the maximum iteration times and Cauchy distribution adjusting factors;
randomly initializing N goblet ascidian individuals; the goblet ascidian individuals are selected as a result of data characteristic selection and a result of machine learning algorithm hyper-parameter selection;
obtaining fitness parameters according to the flow data based on the machine learning algorithm and a K-fold cross validation strategy; the fitness parameter comprises: the accuracy index of the user behavior recognition, the average value of the user behavior recognition and the variance of the user behavior recognition are obtained;
determining the fitness of each individual goblet ascidian according to the fitness parameters;
determining the optimal goblet ascidian individuals of the current population according to the fitness and updating the positions of all goblet ascidian individuals in the population; the optimal cask ascidian individual is the individual with the maximum fitness;
determining the fitness of all bottle ascidian individuals after the position updating, and updating the historical optimal position of each bottle ascidian individual in the population and the global historical optimal position of all the bottle ascidian individuals;
judging whether the iteration times reach preset iteration times or not;
if the iteration times reach the preset iteration times, determining the optimal goblet ascidian individual as the selection result of the data characteristics and the selection result of the hyper-parameters of the machine learning algorithm;
and if the iteration times do not reach the preset iteration times, returning to execute the steps of determining the optimal goblet and sea squirt individuals of the current population according to the fitness and updating the positions of all the goblet and sea squirt individuals in the population until the iteration times reach the preset iteration times.
Preferably, the updating of the location of all individual ascidians in the population specifically comprises:
updating the position of the first goblet ascidian individual in the population by adopting a first updating formula; the first update formula is:
wherein the content of the first and second substances,x ,j1the location of the first individual of a goblet of ascidians,tfor the current number of iterations,F b historical optimal ascidian individuals searched for the population,c 1to comply with Cauchy: (u 10.1) a random number of the distribution,c 2to comply with Cauchy: (u 20.1) a random number of the distribution,u 1=(1-c)·u 1+c·meanL(c 1)，u 2=(1-c)·u 2+c·meanL(c 2) And meanL (x) is the mean Lehmer,cis Cauchy distribution regulating factor;
updating the positions of the individual goblet ascidians except the first individual goblet ascidians in the population by adopting a second updating formula; the second update formula is:
wherein the content of the first and second substances,w t =4w t-1·(1-w t-1)，w t is as followstThe logistic chaotic coefficient of the sub-iteration,x i,j the position of the individual of goblet sea squirt,
is in position ofx i,j The historical optimal position searched by the individual of goblet sea squirt,c 3to comply with Cauchy: (u 30.1) a random number of the distribution,u 3=(1-c)·u 3+c·meanL(c 3)。
preferably, the machine learning algorithm comprises: support vector machines, decision trees and neural networks.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
according to the user behavior identification method facing the video conference system, the Cauchy distributed random number and the logistic chaos coefficient are introduced into the goblet sea squirt group algorithm, so that the diversity of population individuals in the searching process of the goblet sea squirt group algorithm can be increased. In addition, the data characteristics and the machine learning algorithm hyper-parameters are simultaneously optimized and selected by adopting the enhanced goblet sea squirt group algorithm based on the K-fold cross validation strategy, and after the selection result of the data characteristics and the selection result of the machine learning algorithm hyper-parameters are obtained, the user behavior identification model is established based on the selection result of the data characteristics and the selection result of the machine learning algorithm hyper-parameters, so that the identification performance of the user behavior identification model established based on the enhanced goblet sea squirt group algorithm can be improved, and the user behavior identification result in the video conference system is more accurate.
Corresponding to the user behavior identification method for the video conference system, the invention also provides the following implementation devices:
one of them is a user behavior recognition system facing a video conference system, the user behavior recognition system includes:
the network flow data acquisition module is used for acquiring network flow data under a user video conference; the traffic data includes: network state data, protocol analysis data, and user behavior information corresponding to the network state data and the protocol analysis data; the network status data includes: throughput, packet traffic, and delay jitter; the protocol analysis data includes: protocol type, protocol packet length, connection duration, port information, and IP information; the user behavior information includes: a video communication conference behavior, a voice chat conference behavior, an online lecture speaking behavior and an online listening and speaking behavior;
the selection result determining module is used for acquiring a machine learning algorithm according to the flow data, and performing optimization selection on the data characteristics and the hyper-parameters of the machine learning algorithm by adopting an enhanced halymenia group algorithm based on a K-fold cross validation strategy to obtain a selection result of the data characteristics and a selection result of the hyper-parameters of the machine learning algorithm; the enhanced goblet sea squirt group algorithm is a goblet sea squirt group algorithm introduced with Cauchy distributed random numbers and logistic stetty chaotic coefficients;
the user behavior identification model establishing module is used for establishing a user behavior identification model based on the selection result of the data characteristics and the selection result of the hyperparameter of the machine learning algorithm;
and the user behavior identification module is used for inputting the network state data and the protocol analysis data of the user to be detected into the user behavior identification model to obtain the behavior information of the user to be detected output by the user behavior identification model.
Preferably, the selection result determining module includes:
a machine learning algorithm acquisition unit for acquiring a machine learning algorithm;
the parameter initialization unit is used for initializing parameters of the enhanced goblet sea squirt group algorithm; parameters of the enhanced goblet sea squirt group algorithm comprise the number of groups, the maximum iteration times and Cauchy distribution adjusting factors;
the goblet sea squirt individual initialization unit is used for initializing N goblet sea squirt individuals in the population randomly; the goblet ascidian individuals are selected as a result of data characteristic selection and a result of machine learning algorithm hyper-parameter selection;
the fitness parameter determining unit is used for obtaining a fitness parameter according to the flow data based on the machine learning algorithm and a K-fold cross validation strategy; the fitness parameter comprises: the accuracy index of the user behavior recognition, the average value of the user behavior recognition and the variance of the user behavior recognition are obtained;
the fitness determining unit is used for determining the fitness of each individual goblet ascidian according to the fitness parameters;
the goblet ascidian individual position updating unit is used for determining the optimal goblet ascidian individual of the current population according to the fitness and updating the positions of all goblet ascidian individuals in the population; the optimal cask ascidian individual is the individual with the maximum fitness;
the individual historical optimal position updating unit of the goblet ascidians is used for determining the fitness of all individual goblet ascidians after the position updating and updating the historical optimal position of each individual goblet ascidian in the population and the global historical optimal position of all individual goblet ascidians;
the iteration frequency judging unit is used for judging whether the iteration frequency reaches the preset iteration frequency or not;
the optimal goblet ascidian individual determination unit is used for determining the optimal goblet ascidian individual as the selection result of the data characteristics and the selection result of the hyper-parameters of the machine learning algorithm when the iteration times reach the preset iteration times;
and the iteration unit is used for returning to execute the steps of determining the optimal goblet ascidian individuals of the current population according to the fitness and updating the positions of all the goblet ascidian individuals in the population until the iteration times reach the preset iteration times when the iteration times do not reach the preset iteration times.
Preferably, the said individual location updating unit of goblet sea squirt comprises:
the first goblet ascidian individual position updating subunit is used for updating the position of the first goblet ascidian individual in the population by adopting a first updating formula; the first update formula is:
wherein the content of the first and second substances,x ,j1the location of the first individual of a goblet of ascidians,tfor the current number of iterations,F b historical optimal ascidian individuals searched for the population,c 1to comply with Cauchy: (u 10.1) a random number of the distribution,c 2to comply with Cauchy: (u 20.1) a random number of the distribution,u 1=(1-c)·u 1+c·meanL(c 1)，u 2=(1-c)·u 2+c·meanL(c 2) And meanL (x) is the mean Lehmer,cis Cauchy distribution regulating factor;
the second goblet ascidian individual position updating subunit is used for updating the positions of the goblet ascidian individuals except the first goblet ascidian individual in the population by adopting a second updating formula; the second update formula is:
wherein the content of the first and second substances,w t =4w t-1·(1-w t-1)，w t is as followstThe logistic chaotic coefficient of the sub-iteration,x i,j the position of the individual of goblet sea squirt,
is in position ofx i,j The historical optimal position searched by the individual of goblet sea squirt,c 3to comply with Cauchy: (u 30.1) a random number of the distribution,u 3=(1-c)·u 3+c·meanL(c 3)。
the other type is a computer readable storage medium, wherein a computer running program is stored in the computer readable storage medium; the computer running program is used for executing the user behavior identification method facing the video conference system.
The technical effects achieved by the user behavior identification system and the computer readable storage medium for the video conference system provided by the invention are the same as those achieved by the user behavior identification method for the video conference system provided by the invention, and therefore, the description is not repeated herein.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
Fig. 1 is a flowchart of a user behavior identification method for a video conference system according to the present invention;
fig. 2 is a flowchart of a user network traffic data feature selection method based on the enhanced cask ascidian group algorithm according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a user behavior recognition system for a video conference system according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide a user behavior identification method, a user behavior identification system and a storage medium for a video conference system, which can effectively improve the modeling performance of a machine learning algorithm and improve the accuracy of user behavior identification.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
As shown in fig. 1, the method for identifying user behavior for a video conference system provided by the present invention includes:
step 100: and acquiring network flow data under the video conference of the user. The flow data includes: network status data, protocol analysis data, and user behavior information corresponding to the network status data and the protocol analysis data. The network status data includes: throughput, packet traffic, and delay jitter, but is not limited to such. The protocol analysis data includes: protocol type, protocol packet length, connection duration, port information, and IP information, but is not limited thereto. The user behavior information includes: video communication conferencing behavior, voice chat conferencing behavior, online lecture speaking behavior, and online listening and speaking behavior, but are not limited to such.
Step 101: and obtaining a machine learning algorithm according to the flow data, and simultaneously carrying out optimization selection on the data characteristics and the hyper-parameters of the machine learning algorithm by adopting an enhanced goblet sea squirt group algorithm based on a K-fold cross validation strategy to obtain a selection result of the data characteristics and a selection result of the hyper-parameters of the machine learning algorithm. The enhanced goblet sea squirt group algorithm is a goblet sea squirt group algorithm introduced with Cauchy distributed random numbers and logistic chaos coefficients.
Step 102: and establishing a user behavior recognition model based on the selection result of the data characteristics and the selection result of the hyper-parameters of the machine learning algorithm.
Step 103: and inputting the network state data and the protocol analysis data of the user to be detected into the user behavior recognition model to obtain the behavior information of the user to be detected output by the user behavior recognition model.
As shown in fig. 2, the specific implementation process of step 101 includes:
step 1010: a machine learning algorithm is obtained. The machine learning algorithm includes: support vector machines, decision trees, and neural networks, but are not so limited.
Step 1011: parameters of the enhanced goblet sea squirt group algorithm are initialized. The parameters of the algorithm for enhancing the sea squirt group of goblet comprise the number of the groupNMaximum number of iterationsTAnd Cauchy distribution regulatorc。
Step 1012: random initializationNIndividual goblet ascidian, individual goblet ascidianx i =[x i,1 , x i,2 , …, x i,D ,x i,D+1 , …, x i,M ](i=1,2,..,N). The individual goblet ascidians are the result of selecting the data characteristics and the result of selecting the hyper-parameters of the machine learning algorithm. Wherein the content of the first and second substances,x i,1 andx i,D respectively representing the 1 st and the 1 st of the flow dataDSelecting intensity by roundingThe policy determines whether a flow data feature is selected or not,x i,D+1 , …, x i,M respectively representing the setting conditions of each hyper-parameter of the machine learning algorithm.
Step 1013: and obtaining a fitness parameter according to the flow data based on a machine learning algorithm and a K-fold cross validation strategy. The fitness parameters include: accuracy index for user behavior recognitionAUCAverage value of user behavior recognitionAUC m Variance of user behavior recognitionAUC sd 。
Step 1014: and determining the individual fitness of each bottle of ascidians according to the fitness parameters. For example, according to a formulaf i =AUC m /(1+AUC sd ) Calculating the fitness of each individualf i 。
Step 1015: determining the optimal individual of the current population according to the fitnessFAnd updating the positions of all the individual goblet sea squirts in the population. The optimal cask ascidian individual is the individual with the highest fitness.
Wherein, the position of the first goblet ascidian individual in the population is updated by adopting a first updating formula. The first update formula is:
wherein the content of the first and second substances,x ,j1the location of the first individual of a goblet of ascidians,tfor the current number of iterations,F b historical optimal ascidian individuals searched for the population,c 1to comply with Cauchy: (u 10.1) a random number of the distribution,c 2to comply with Cauchy: (u 20.1) a random number of the distribution,u 1=(1-c)·u 1+c·meanL(c 1)，u 2=(1-c)·u 2+c·meanL(c 2) And meanL (x) is the mean Lehmer,cis Cauchy distribution regulating factor.
And updating the positions of the individual goblet ascidians except the first individual goblet ascidians in the population by adopting a second updating formula. The second update formula is:
wherein the content of the first and second substances,w t =4w t-1·(1-w t-1)，w t is as followstThe logistic chaotic coefficient of the sub-iteration,x i,j the position of the individual of goblet sea squirt,
is in position ofx i,j The historical optimal position searched by the individual of goblet sea squirt,c 3to comply with Cauchy: (u 30.1) a random number of the distribution,u 3=(1-c)·u 3+c·meanL(c 3)。
step 1016: and determining the fitness of all bottle ascidian individuals after the position updating, and updating the historical optimal position of each bottle ascidian individual in the population and the global historical optimal position of all the bottle ascidian individuals.
Step 1017: and judging whether the iteration times reach the preset iteration times.
Step 1018: and if the iteration times reach the preset iteration times, determining the optimal individual of the halymenia goblet as a selection result of the data characteristics and a selection result of the hyperparameter of the machine learning algorithm.
Step 1019: if the iteration number does not reach the preset iteration number, the process returns to step 1015 until the iteration number reaches the preset iteration number.
And training and generating a user behavior recognition model on the acquired network traffic data based on the feature selection result obtained in the specific implementation process of the step 101 and the acquired hyper-parameter optimization result of the machine learning algorithm.
In order to further improve the accuracy of the identification result, after the step 100, the method for identifying the user behavior of the video conference system further includes:
and carrying out standardization processing on the flow data. The flow data can be expressed by the formulap j ’=(p j -p avg )/σThe normalization method shown. Wherein the content of the first and second substances,p j is shown asjThe original value of the data on a certain feature,p avg represents the average of all data over the feature,σrepresenting the variance of all data over the feature. The user behavior information data can be standardized by adopting a single hot code and other modes.
In conclusion, the user behavior identification method for the video conference system, provided by the invention, adopts Cauchy distribution aiming at the position update parameter of the first individual of the population, so that the diversity of the individual of the population is favorably kept, and the premature convergence of the population is avoided. Secondly, the use of the Lehmer mean value enables adaptive search to focus on parameters with larger values, the global search capability of the population is enhanced, and in addition, the introduction of the historical optimal position of the population can accelerate the search process. And for the position updating of other individuals in the population, a logistic chaotic coefficient is adopted, and individual history is introduced to search for an optimal position, so that the local searching capability of the population is improved.
Based on the overall implementation framework of the user behavior identification method for the video conference system, provided by the invention, after corresponding network data is obtained, other optimization selection methods can be adopted to select the optimal data feature set and the algorithm hyper-parameters for establishing the user behavior identification model.
Corresponding to the user behavior identification method for the video conference system, the invention also provides the following implementation devices:
one of them is a user behavior recognition system facing a video conference system, as shown in fig. 3, the user behavior recognition system includes: the system comprises a network flow data acquisition module 1, a selection result determination module 2, a user behavior identification model building module 3 and a user behavior identification module 4.
The network traffic data obtaining module 1 is configured to obtain network traffic data in a user video conference. The flow data includes: network status data, protocol analysis data, and user behavior information corresponding to the network status data and the protocol analysis data. The network status data includes: throughput, packet traffic, and delay jitter. The protocol analysis data includes: protocol type, protocol packet length, connection duration, port information, and IP information. The user behavior information includes: video communication conference behavior, voice chat conference behavior, online lecture speaking behavior, and online listening and speaking behavior.
The selection result determining module 2 is used for obtaining a machine learning algorithm according to the flow data, and performing optimization selection on the data characteristics and the hyper-parameters of the machine learning algorithm by adopting an enhanced halymenia group algorithm based on a K-fold cross validation strategy to obtain a selection result of the data characteristics and a selection result of the hyper-parameters of the machine learning algorithm. The enhanced goblet sea squirt group algorithm is a goblet sea squirt group algorithm introduced with Cauchy distributed random numbers and logistic chaos coefficients.
The user behavior identification model establishing module 3 is used for establishing a user behavior identification model based on the selection result of the data characteristics and the selection result of the hyperparameter of the machine learning algorithm.
The user behavior recognition module 4 is configured to input the network state data and the protocol analysis data of the user to be detected into the user behavior recognition model, and obtain behavior information of the user to be detected output by the user behavior recognition model.
The selection result determining module 2 may include: the system comprises a machine learning algorithm obtaining unit, a parameter initializing unit, a goblet and ascidian individual initializing unit, a fitness parameter determining unit, a fitness determining unit, a goblet and ascidian individual position updating unit, a goblet and ascidian individual history optimal position updating unit, an iteration number judging unit, an optimal goblet and ascidian individual determining unit and an iteration unit.
The machine learning algorithm obtaining unit is used for obtaining a machine learning algorithm.
The parameter initialization unit is used for initializing parameters of the enhanced goblet sea squirt group algorithm. Parameters of the enhanced goblet sea squirt group algorithm include the number of groups, the maximum iteration number and Cauchy distribution adjustment factors.
The goblet ascidian individual initialization unit is used for randomly initializing N goblet ascidian individuals in the population. The individual goblet ascidians are the result of selecting the data characteristics and the result of selecting the hyper-parameters of the machine learning algorithm.
And the fitness parameter determining unit is used for obtaining a fitness parameter according to the flow data based on a machine learning algorithm and a K-fold cross validation strategy. The fitness parameters include: an accuracy index of the user behavior recognition, an average value of the user behavior recognition, and a variance of the user behavior recognition.
The fitness determining unit is used for determining the fitness of each individual bottle of ascidians according to the fitness parameters.
And the goblet ascidian individual position updating unit is used for determining the optimal goblet ascidian individual of the current population according to the fitness and updating the positions of all goblet ascidian individuals in the population. The optimal cask ascidian individual is the individual with the highest fitness.
And the historical optimal position updating unit of the individual goblet ascidians is used for determining the fitness of all goblet ascidian individuals after the position updating and updating the historical optimal position of each individual goblet ascidian in the population and the global historical optimal position of all individual goblet ascidians.
The iteration frequency judging unit is used for judging whether the iteration frequency reaches the preset iteration frequency.
The optimal goblet ascidian individual determination unit is used for determining the optimal goblet ascidian individual as a selection result of the data characteristics and a selection result of the hyperparameter of the machine learning algorithm when the iteration times reach the preset iteration times.
The iteration unit is used for returning and executing the steps of determining the optimal goblet and sea squirt individuals of the current population according to the fitness and updating the positions of all the goblet and sea squirt individuals in the population when the iteration times do not reach the preset iteration times until the iteration times reach the preset iteration times.
Further, the aforementioned individual location updating unit for the goblet ascidian may further comprise: the individual position updating subunit of the first goblet ascidian and the individual position updating subunit of the second goblet ascidian.
The first goblet ascidian individual position updating subunit is used for updating the position of the first goblet ascidian individual in the population by adopting a first updating formula. The first update formula is:
wherein the content of the first and second substances,x ,j1the location of the first individual of a goblet of ascidians,tfor the current number of iterations,F b historical optimal ascidian individuals searched for the population,c 1to comply with Cauchy: (u 10.1) a random number of the distribution,c 2to comply with Cauchy: (u 20.1) a random number of the distribution,u 1=(1-c)·u 1+c·meanL(c 1)，u 2=(1-c)·u 2+c·meanL(c 2) And meanL (x) is the mean Lehmer,cis Cauchy distribution regulating factor.
And the second goblet ascidian individual position updating subunit is used for updating the positions of the goblet ascidian individuals except the first goblet ascidian individual in the population by adopting a second updating formula. The second update formula is:
wherein the content of the first and second substances,w t =4w t-1·(1-w t-1)，w t is as followstThe logistic chaotic coefficient of the sub-iteration,x i,j the position of the individual of goblet sea squirt,
is in position ofx i,j The historical optimal position searched by the individual of goblet sea squirt,c 3to comply with Cauchy: (u 30.1) a random number of the distribution,u 3=(1-c)·u 3+c·meanL(c 3)。
the other is a computer-readable storage medium in which a computer-executable program is stored. The computer runs a program for executing the user behavior recognition method for the video conference system provided above.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (6)
1. A user behavior identification method facing a video conference system is characterized by comprising the following steps:
acquiring network flow data under a user video conference; the traffic data includes: network state data, protocol analysis data, and user behavior information corresponding to the network state data and the protocol analysis data; the network status data includes: throughput, packet traffic, and delay jitter; the protocol analysis data includes: protocol type, protocol packet length, connection duration, port information, and IP information; the user behavior information includes: a video communication conference behavior, a voice chat conference behavior, an online lecture speaking behavior and an online listening and speaking behavior;
acquiring a machine learning algorithm according to the flow data, and simultaneously performing optimization selection on the data characteristics and the hyper-parameters of the machine learning algorithm by adopting an enhanced goblet sea squirt group algorithm based on a K-fold cross validation strategy to obtain a selection result of the data characteristics and a selection result of the hyper-parameters of the machine learning algorithm; the enhanced goblet sea squirt group algorithm is a goblet sea squirt group algorithm introduced with Cauchy distributed random numbers and logistic stetty chaotic coefficients;
establishing a user behavior recognition model based on the selection result of the data characteristics and the selection result of the hyper-parameters of the machine learning algorithm; the user behavior recognition model takes network state data and protocol analysis data characteristics as input and takes user behavior information as output;
taking network state data and protocol analysis data of a user to be detected as input, and obtaining behavior information of the user to be detected by adopting the user behavior identification model;
the acquiring of the machine learning algorithm according to the flow data, and performing optimization selection on the data characteristics and the hyper-parameters of the machine learning algorithm by using the enhanced goblet sea squirt group algorithm based on the K-fold cross validation strategy to obtain the selection results of the data characteristics and the hyper-parameters of the machine learning algorithm specifically include:
acquiring a machine learning algorithm;
initializing parameters of the enhanced goblet sea squirt group algorithm; parameters of the enhanced goblet sea squirt group algorithm comprise the number of groups, the maximum iteration times and Cauchy distribution adjusting factors;
randomly initializing N goblet ascidian individuals; the goblet ascidian individuals are selected as a result of data characteristic selection and a result of machine learning algorithm hyper-parameter selection;
obtaining fitness parameters according to the flow data based on the machine learning algorithm and a K-fold cross validation strategy; the fitness parameter comprises: the accuracy index of the user behavior recognition, the average value of the user behavior recognition and the variance of the user behavior recognition are obtained;
determining the fitness of each individual goblet ascidian according to the fitness parameters;
determining the optimal goblet ascidian individuals of the current population according to the fitness and updating the positions of all goblet ascidian individuals in the population; the optimal cask ascidian individual is the individual with the maximum fitness;
determining the fitness of all bottle ascidian individuals after the position updating, and updating the historical optimal position of each bottle ascidian individual in the population and the global historical optimal position of all the bottle ascidian individuals;
judging whether the iteration times reach preset iteration times or not;
if the iteration times reach the preset iteration times, determining the optimal goblet ascidian individual as the selection result of the data characteristics and the selection result of the hyper-parameters of the machine learning algorithm;
and if the iteration times do not reach the preset iteration times, returning to execute the steps of determining the optimal goblet and sea squirt individuals of the current population according to the fitness and updating the positions of all the goblet and sea squirt individuals in the population until the iteration times reach the preset iteration times.
2. The method for identifying user behaviors oriented to a video conference system as claimed in claim 1, wherein updating the positions of all bottle sea squirts in a population specifically comprises:
updating the position of the first goblet ascidian individual in the population by adopting a first updating formula; the first update formula is:
wherein x is1,jIs the position of the first individual of goblet sea squirt, t is the current iteration number, FbFor historically optimal ascidian individuals searched for by the population, c1To comply with Cauchy (u)10.1) random numbers of distribution, c2To comply with Cauchy (u)20.1) random numbers of distribution, u1＝(1-c)·u1+c·meanL(c1)，u2＝(1-c)·u2+c·meanL(c2) Mean (,) is the Lehmer mean value, c is the Cauchy distribution regulating factor, and T is the maximum iteration number;
updating the positions of the individual goblet ascidians except the first individual goblet ascidians in the population by adopting a second updating formula; the second update formula is:
wherein, wt＝4wt-1·(1-wt-1)，wtLogistic chaotic coefficient, x, for the t-th iterationi,jThe position of the individual of goblet sea squirt,
is at a position xi,jHistorical optimal position searched by individual goblet sea squirt, c3To comply with Cauchy (u)30.1) random numbers of distribution, u3＝(1-c)·u3+c·meanL(c3)，xi-1,jThe position of the i-1 st individual of the ascidian goblet.
3. The videoconference system-oriented user behavior recognition method of claim 1, wherein the machine learning algorithm comprises: support vector machines, decision trees and neural networks.
4. A system for identifying user behavior for a video conferencing system, comprising:
the network flow data acquisition module is used for acquiring network flow data under a user video conference; the traffic data includes: network state data, protocol analysis data, and user behavior information corresponding to the network state data and the protocol analysis data; the network status data includes: throughput, packet traffic, and delay jitter; the protocol analysis data includes: protocol type, protocol packet length, connection duration, port information, and IP information; the user behavior information includes: a video communication conference behavior, a voice chat conference behavior, an online lecture speaking behavior and an online listening and speaking behavior;
the selection result determining module is used for acquiring a machine learning algorithm according to the flow data, and performing optimization selection on the data characteristics and the hyper-parameters of the machine learning algorithm by adopting an enhanced halymenia group algorithm based on a K-fold cross validation strategy to obtain a selection result of the data characteristics and a selection result of the hyper-parameters of the machine learning algorithm; the enhanced goblet sea squirt group algorithm is a goblet sea squirt group algorithm introduced with Cauchy distributed random numbers and logistic stetty chaotic coefficients;
the user behavior identification model establishing module is used for establishing a user behavior identification model based on the selection result of the data characteristics and the selection result of the hyperparameter of the machine learning algorithm; the user behavior recognition model takes network state data and protocol analysis data characteristics as input and takes user behavior information as output;
the user behavior recognition module is used for taking the network state data and the protocol analysis data of the user to be detected as input and obtaining the behavior information of the user to be detected by adopting the user behavior recognition model;
the selection result determining module includes:
a machine learning algorithm acquisition unit for acquiring a machine learning algorithm;
the parameter initialization unit is used for initializing parameters of the enhanced goblet sea squirt group algorithm; parameters of the enhanced goblet sea squirt group algorithm comprise the number of groups, the maximum iteration times and Cauchy distribution adjusting factors;
the goblet sea squirt individual initialization unit is used for initializing N goblet sea squirt individuals in the population randomly; the goblet ascidian individuals are selected as a result of data characteristic selection and a result of machine learning algorithm hyper-parameter selection;
the fitness parameter determining unit is used for obtaining a fitness parameter according to the flow data based on the machine learning algorithm and a K-fold cross validation strategy; the fitness parameter comprises: the accuracy index of the user behavior recognition, the average value of the user behavior recognition and the variance of the user behavior recognition are obtained;
the fitness determining unit is used for determining the fitness of each individual goblet ascidian according to the fitness parameters;
the goblet ascidian individual position updating unit is used for determining the optimal goblet ascidian individual of the current population according to the fitness and updating the positions of all goblet ascidian individuals in the population; the optimal cask ascidian individual is the individual with the maximum fitness;
the individual historical optimal position updating unit of the goblet ascidians is used for determining the fitness of all individual goblet ascidians after the position updating and updating the historical optimal position of each individual goblet ascidian in the population and the global historical optimal position of all individual goblet ascidians;
the iteration frequency judging unit is used for judging whether the iteration frequency reaches the preset iteration frequency or not;
the optimal goblet ascidian individual determination unit is used for determining the optimal goblet ascidian individual as the selection result of the data characteristics and the selection result of the hyper-parameters of the machine learning algorithm when the iteration times reach the preset iteration times;
and the iteration unit is used for returning to execute the steps of determining the optimal goblet ascidian individuals of the current population according to the fitness and updating the positions of all the goblet ascidian individuals in the population until the iteration times reach the preset iteration times when the iteration times do not reach the preset iteration times.
5. The system as claimed in claim 4, wherein the cask ascidian location updating unit comprises:
the first goblet ascidian individual position updating subunit is used for updating the position of the first goblet ascidian individual in the population by adopting a first updating formula; the first update formula is:
wherein x is1,jIs the position of the first individual of goblet sea squirt, t is the current iteration number, FbFor historically optimal ascidian individuals searched for by the population, c1To comply with Cauchy (u)10.1) random numbers of distribution, c2To comply with Cauchy (u)20.1) random numbers of distribution, u1＝(1-c)·u1+c·meanL(c1)，u2＝(1-c)·u2+c·meanL(c2) Mean (,) is the Lehmer mean value, c is the Cauchy distribution regulating factor, and T is the maximum iteration number;
the second goblet ascidian individual position updating subunit is used for updating the positions of the goblet ascidian individuals except the first goblet ascidian individual in the population by adopting a second updating formula; the second update formula is:
wherein, wt＝4wt-1·(1-wt-1)，wtLogistic chaotic coefficient, x, for the t-th iterationi,jThe position of the individual of goblet sea squirt,
is at a position xi,jHistorical optimal position searched by individual goblet sea squirt, c3To comply with Cauchy (u)30.1) random numbers of distribution, u3＝(1-c)·u3+c·meanL(c3)，xi-1,jThe position of the i-1 st individual of the ascidian goblet.
6. A computer-readable storage medium in which a computer-executable program is stored; the computer running program is used for executing the user behavior recognition method facing the video conference system according to any one of claims 1 to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110971140.3A CN113627382B (en) | 2021-08-24 | 2021-08-24 | User behavior identification method and system for video conference system and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110971140.3A CN113627382B (en) | 2021-08-24 | 2021-08-24 | User behavior identification method and system for video conference system and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113627382A CN113627382A (en) | 2021-11-09 |
| CN113627382B true CN113627382B (en) | 2022-02-22 |
Family
ID=78387318
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110971140.3A Active CN113627382B (en) | 2021-08-24 | 2021-08-24 | User behavior identification method and system for video conference system and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113627382B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109873810A (en) * | 2019-01-14 | 2019-06-11 | 湖北工业大学 | A kind of phishing detectin method based on cup ascidian group's algorithm support vector machines |
| CN110738070A (en) * | 2018-07-02 | 2020-01-31 | 中国科学院深圳先进技术研究院 | Behavior identification method and behavior identification device based on video and terminal equipment |
| CN111027663A (en) * | 2019-11-12 | 2020-04-17 | 天津大学 | Method for improving algorithm of goblet sea squirt group |
| CN111126549A (en) * | 2019-12-24 | 2020-05-08 | 昆明理工大学 | Double-star spectrum fitting method based on strategy improved goblet and sea squirt intelligent algorithm |
| CN111506036A (en) * | 2020-05-25 | 2020-08-07 | 北京化工大学 | Multivariate Hammerstein model identification method and system under heavy tail noise interference |
| CN112383738A (en) * | 2020-11-11 | 2021-02-19 | 浙江讯盟科技有限公司 | Multi-user audio and video conference method and system with low traffic and resource consumption |
| CN112396135A (en) * | 2021-01-21 | 2021-02-23 | 北京电信易通信息技术股份有限公司 | Method and system for detecting abnormal traffic of converged communication network |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11122240B2 (en) * | 2017-09-11 | 2021-09-14 | Michael H Peters | Enhanced video conference management |
| CN113110490A (en) * | 2021-05-07 | 2021-07-13 | 金陵科技学院 | Robot multi-target path planning based on improved goblet sea squirt group algorithm |
| CN113255138B (en) * | 2021-05-31 | 2023-05-23 | 河北工业大学 | Load distribution optimization method for power system |
-
2021
- 2021-08-24 CN CN202110971140.3A patent/CN113627382B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110738070A (en) * | 2018-07-02 | 2020-01-31 | 中国科学院深圳先进技术研究院 | Behavior identification method and behavior identification device based on video and terminal equipment |
| CN109873810A (en) * | 2019-01-14 | 2019-06-11 | 湖北工业大学 | A kind of phishing detectin method based on cup ascidian group's algorithm support vector machines |
| CN111027663A (en) * | 2019-11-12 | 2020-04-17 | 天津大学 | Method for improving algorithm of goblet sea squirt group |
| CN111126549A (en) * | 2019-12-24 | 2020-05-08 | 昆明理工大学 | Double-star spectrum fitting method based on strategy improved goblet and sea squirt intelligent algorithm |
| CN111506036A (en) * | 2020-05-25 | 2020-08-07 | 北京化工大学 | Multivariate Hammerstein model identification method and system under heavy tail noise interference |
| CN112383738A (en) * | 2020-11-11 | 2021-02-19 | 浙江讯盟科技有限公司 | Multi-user audio and video conference method and system with low traffic and resource consumption |
| CN112396135A (en) * | 2021-01-21 | 2021-02-23 | 北京电信易通信息技术股份有限公司 | Method and system for detecting abnormal traffic of converged communication network |
Non-Patent Citations (2)
| Title |
|---|
| Feature Selection Using Salp Swarm Algorithm with Chaos;Sobhi Ahmed等;《ISMSI "18:Proceedings of the 2nd International Conference on Intelligent Systems, Metaheuristics & Swarm Intelligence》;20180324;第65-69页 * |
| 混沌映射与动态学习的自适应樽海鞘群算法;卓然 等;《计算机工程与设计》;20210716;第42卷(第7期);第1963-1972页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113627382A (en) | 2021-11-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | Deep reinforcement learning based resource management for DNN inference in industrial IoT | |
| CN111563275B (en) | Data desensitization method based on generation countermeasure network | |
| CN106453495B (en) | A kind of information centre's network-caching method based on content popularit prediction | |
| CN112181971A (en) | Edge-based federated learning model cleaning and equipment clustering method, system, equipment and readable storage medium | |
| CN110968426B (en) | Edge cloud collaborative k-means clustering model optimization method based on online learning | |
| CN112396135B (en) | Method and system for detecting abnormal traffic of converged communication network | |
| CN110290077B (en) | Industrial SDN resource allocation method based on real-time service configuration | |
| CN110234155A (en) | A kind of super-intensive network insertion selection method based on improved TOPSIS | |
| Chen et al. | Deep-broad learning system for traffic flow prediction toward 5G cellular wireless network | |
| Zhu et al. | FedOVA: one-vs-all training method for federated learning with non-IID data | |
| CN107133268B (en) | Collaborative filtering method for Web service recommendation | |
| CN116208567A (en) | Method and system for flow scheduling of SDN network resources of cross-domain data center | |
| Mestoukirdi et al. | User-centric federated learning | |
| CN113627382B (en) | User behavior identification method and system for video conference system and storage medium | |
| CN106789349B (en) | Quality of experience modeling analysis and conversation flow classification based method | |
| CN117829307A (en) | Federal learning method and system for data heterogeneity | |
| Wang et al. | Multi-objective joint optimization of communication-computation-caching resources in mobile edge computing | |
| CN116915746A (en) | Network addressing method | |
| US20230087774A1 (en) | Parameter optimization method, electronic device, and storage medium | |
| WO2023065640A1 (en) | Model parameter adjustment method and apparatus, electronic device and storage medium | |
| Tang et al. | Tackling system induced bias in federated learning: Stratification and convergence analysis | |
| Wu et al. | Model-heterogeneous federated learning with partial model training | |
| Wang et al. | A PSO-based multicast routing algorithm | |
| CN115766140A (en) | Distributed denial of service (DDoS) attack detection method and device | |
| CN115695429A (en) | Non-IID scene-oriented federal learning client selection method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: User behavior recognition method, system and storage medium for video conferencing system Effective date of registration: 20221102 Granted publication date: 20220222 Pledgee: Beijing technology intellectual property financing Company limited by guarantee Pledgor: BEIJING TELECOMMUNICATION YITONG INFORMATION TECHNOLOGY Co.,Ltd. Registration number: Y2022990000766 |
|
| PE01 | Entry into force of the registration of the contract for pledge of patent right |