Automatic evaluation method based on user behavior measure in virtual experiment environment

Abstract

The invention discloses an automatic evaluation method based on user behavior measurement in a virtual experiment environment, and the method comprises the steps: monitoring user behavior information in a virtual laboratory in real time, and generating behavior sequence data; the monitoring system records an experiment result in real time and generates a result log; identifying experimental steps of the user through the intelligent evaluation model; standard results of the corresponding experiment steps are obtained through a preset knowledge base, and the experiment results of the experiment steps are automatically evaluated based on the standard results, the behavior sequence data and the result logs; according to the method, the behavior sequence of the student is finely analyzed in combination with the deep neural network, and the experimental steps are accurately identified by using the trained deep learning framework, so that not only is the scoring accuracy improved, but also the mastering degree of the student on each step of the experiment is understood through fine scoring, and the virtual experiment scoring efficiency and accuracy are improved.

Description

Technical Field

[0001] The invention relates to the field of educational technology, and in particular to an automatic evaluation method based on user behavior measurement in a virtual experiment environment. Background Art

[0002] The scoring of students' experiments is an important part of the school's experimental courses. With the development of virtual machines and simulation experiment technology, more and more virtual laboratories have been designed. Compared with traditional laboratories, virtual laboratories only virtualize real experiments. The experiments are still scored using paper experiment reports. However, the experimental results fed back by paper experiment reports have great limitations and cannot fully reflect students' practical abilities. At the same time, it is easy for teachers to make mistakes when scoring based on experimental reports, which reduces the accuracy of experimental scoring. Moreover, it is difficult for experimental reports to reflect students' experimental process and it is difficult to understand students' mastery of each step of the experiment.

[0003] Therefore, the present invention provides an automatic evaluation method based on user behavior measurement in a virtual experiment environment to solve the above problems. Summary of the invention

[0004] The purpose of the present invention is to provide an automatic evaluation method based on user behavior measurement in a virtual experiment environment to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] An automatic evaluation method based on user behavior measurement in a virtual experiment environment comprises the following steps:

[0007] Step S100: real-time monitoring of user behavior information in the virtual laboratory, and adding valid information to the storage queue in sequence, wherein: the behavior information includes the behavior type, the time point when the behavior occurs, the observed behavior sequence number and the behavior parameters, and the behavior parameters include specific parameters of different behaviors;

[0008] Step S200: Real-time detection of the result log of the monitoring system. If the result log is not updated, continue to add new behavior information to the storage queue; if the result log is updated, obtain the behavior sequence data from the storage queue, and store the behavior sequence and result log in the user's work log;

[0009] Step S300: According to the virtual experiment sample, the behavior sequence data and result logs generated by multiple users in each experimental step are collected in advance, the data sample of the experiment is generated, the intelligent evaluation model is constructed, and the experimental steps of the user are identified;

[0010] Step S400: according to the identified experimental steps, obtain the standard experimental results corresponding to the experimental steps from the preset knowledge base, obtain the user's behavior sequence data and result log, and generate the result score of the experimental steps according to the preset evaluation template;

[0011] Step S500: query the number of times the user has completed the experimental step. If the number of times the experimental step has been completed is greater than a set threshold, the highest score of the experimental step is taken as the result score of the experimental step and added to the total score; if the number of times the experimental step has been completed does not reach the threshold, the new result score is added to the total score;

[0012] Step S600: After the result scoring of the experimental step is completed, determine whether the user experiment is over according to the preset judgment template. If the judgment result is that the experiment is over, directly output the total score; if the judgment result is that the experiment is not over, clear the storage queue and continue monitoring and scoring.

[0013] Furthermore, the specific implementation process of step S100 includes:

[0014] The user behavior information of the virtual laboratory is monitored in real time through the keyboard, mouse, microphone and system environment configuration software, where the behavior information includes the behavior type, the time point when the behavior occurs, the observed behavior sequence number and the behavior parameters, and the behavior parameters include the specific parameters of different behaviors;

[0015] Valid information is extracted from the behavior information according to a preset judgment template, and the valid information is added to the storage queue in sequence, wherein: the preset judgment template includes information screening rules for screening valid information, and the information screening rules are based on the correlation between the user behavior type and the experimental steps to screen out the behavior information related to the experiment.

[0016] It is necessary to explain that monitoring equipment is installed in the virtual laboratory to monitor the operation of the experimenter. The monitoring equipment includes: hardware facilities such as keyboard, mouse and microphone, and system environment configuration software. For the above monitored information, the behavior information is stored in a queue, and the monitored behavior information will be added to the queue in sequence; the user behavior is further analyzed and filtered through the preset judgment template, including: judging whether the mouse click object is related to the experiment, judging whether each keyboard tap is effective for the experiment, and judging whether other behaviors are related to the experiment. These behavioral information related to the experiment and effective for the experiment is valid information. During monitoring, behaviors that are invalid for the experiment will not be added to the queue.

[0017] Furthermore, the specific implementation process of step S200 includes:

[0018] The result log of the monitoring system is checked in real time. If the result log is not updated, new behavior information is continuously added to the storage queue. The monitoring system is used to determine in real time whether a new user experimental step result is generated. When a new user experimental step result is generated, the new user experimental step result is written into the result log and updated.

[0019] If the result log has been updated, get the behavior sequence data based on the behavior information of the storage queue:<type,time,index,attributes> , where: type represents the behavior type; time represents the time point when the behavior occurs; index represents the sequence number of the observed behavior; attributes represents the behavior parameters; and the behavior sequence and result log are stored in the user's work log.

[0020] It needs to be explained that when the user is doing an experiment, once the result of a certain experimental step is generated, the monitoring system will record it in real time and generate a result log; behavior types include mouse, keyboard input, file operations, etc.; behavior parameters, such as the position of mouse operation, keyboard input string sequence, etc.

[0021] Furthermore, the specific implementation process of step S300 includes:

[0022] Step S301: According to the virtual experiment sample, collect the behavior sequence data and result log generated by the user in each experimental step to generate the data sample of the experiment. The data sample is stored in the form of tuples in the format of: <eid,<sid, <actions>,target>>, where: eid represents the number of the experiment; sid represents the number of the experimental step; target represents the result of the experimental step; <actions>Represents the behavior sequence data corresponding to the current experimental step;

[0023] Step S302: Building an intelligent evaluation model based on the experimental data samples:

[0024] Count the number of times and frequency of use of hardware facilities in each behavior sequence, and use the statistical results as global features to form a feature vector G i =[N i,1 ,F i,1 ,N i,2 ,F i,2 ,...,N i,j ,F i,j ], where i represents the experimental step; j represents the total number of hardware facilities; N i,j represents the number of times hardware facility j is used in experimental step i; F i,j It indicates the frequency of use of hardware facility j in experimental step i;

[0025] Based on the experimental data samples, the state machine is used to model the behavior sequence data, the state of the LSTM network is updated, and the LSTM state vector H is generated. i ;

[0026] The feature vector G i and the LSTM state vector H i As input data to the input layer of a fully connected neural network;

[0027] The feature vector and LSTM state vector are fused and extracted through the fully connected layer to generate a new feature vector K i =[G i ;H i ]; where: K i Represents the feature vector of the fully connected layer when the experimental step is i;

[0028] Use the softmax function as the activation function and output the probability distribution of the predicted experimental steps: p i =softmax(W*K i +b); where: W and b are the weight matrix and bias vector of the fully connected layer respectively; softmax() is used to map each element of the input vector to the interval [0, 1], and the sum of all elements is 1;

[0029] Step S303: extract the real-time behavior sequence data and the real-time result log, count the number of times the hardware facilities in the real-time behavior sequence data are used and the frequency of use, and input the real-time behavior sequence data, the real-time result log and the statistical results into the intelligent evaluation model to identify the user's experimental steps.

[0030] Furthermore, the specific implementation process of generating the LSTM state vector includes:

[0031] Map different behavior types to different states in the state machine and construct a state set S = {s d |d∈[1,2,…,D]}, where: s d represents the state of the dth behavior type; D represents the number of behavior types; the transition between states is triggered by the occurrence of behavior;

[0032] Construct the transfer set T = {t mn |m,n∈[1,2,…,D]}, where: t mn Indicates that from state s m To state s n transfer;

[0033] Represent each action as an input vector to the LSTM:

[0034] x h =[embedding(type h ),encode(time h ),encode(attribute h ]; where: h represents the time step; x h Represents the input vector of the h-th time step LSTM; type h Indicates the behavior type of the hth time step; time h Indicates the time point at which the behavior of the hth time step occurs; attribute h represents the behavior parameters of the hth time step; embedding is an embedding function used to map the behavior type to a low-dimensional vector space; encode is an encoding function used to convert time and behavior parameters into a format suitable for LSTM input;

[0035] LSTM network state update: f h =σ(W f *[y h-1 ,x h ]+b f );

[0036] e h =σ(W e *[y h-1 ,x h ]+b e );

[0037]

[0038] o h =σ(W o *[y h-1 ,x h ]+b o );

[0039] y h =o h *tanh(c h );where: f h represents the forget gate of the hth time step; σ represents the sigmoid function; e h represents the input gate of the hth time step; represents the candidate cell state at the hth time step; c h represents the cell state at the hth time step; o h represents the output gate of the hth time step; y h represents the hidden state of the hth time step; tanh represents the hyperbolic tangent function; W f , W e , W c , W o Represents the preset weight matrix; b f , b i , b c , b o Represents the preset bias item;

[0040] Process each action in turn and update the hidden state, which is recorded in the LSTM state vector H i middle.

[0041] It needs to be explained that target is the result of the experimental step, and the result is saved in the form of a text file.

[0042] Furthermore, the specific implementation process of step S400 includes:

[0043] According to the identified experimental steps, the standard experimental results corresponding to the experimental steps are obtained from a preset knowledge base, wherein: the preset knowledge base stores the experimental steps and the corresponding standard experimental results, and one experimental step corresponds to one standard experimental result; based on the standard experimental results, behavior sequence data and result logs, a result score for the experimental steps is generated through a preset evaluation template, wherein: the preset evaluation template includes result score generation rules for experimental scoring.

[0044] Furthermore, the specific implementation process of step S500 includes:

[0045] Query the number of times the user completes the experimental steps in the experiment. If the number of completions of the experimental steps is greater than a preset threshold, take the highest score among the result scores of the experimental steps as the result score of the experimental steps and add it to the total score; if the number of completions of the experimental steps does not reach the threshold, add the new result score to the total score.

[0046] It should be explained that the scoring method designed by the present invention allows students to repeat a single experimental step and score the best one. In this way, the teacher can not only know the number of times each experimental step is repeated when the students do the experiment, but also understand the changes in the score of each result of the experimental step; combined with the above information, it can be judged which steps of the experiment the students have a low level of mastery, thereby achieving precise teaching, saving teaching resources, and improving scoring efficiency and accuracy.

[0047] Furthermore, the specific implementation process of step S600 includes:

[0048] When the result scoring of the experimental step is completed, the preset judgment template is used to determine whether the user experiment is over. If the judgment result is that the experiment is over, the total score is directly output; if the judgment result is that the experiment is not over, the storage queue is cleared, and monitoring and scoring continue, where: the preset judgment template includes judgment rules for determining whether the user experiment is over.

[0049] It needs to be explained that the scores of paper lab reports need to be entered manually into the database, and personal scores can be easily leaked; the virtual laboratory automatic scoring system uses computers to generate scores and record them in real time, making it difficult for data to be leaked, which helps protect students' privacy. At the same time, the scores generated by computers are difficult to be tampered with manually with the help of some technologies, which enhances data security.

[0050] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: the automatic evaluation method based on user behavior measurement in a virtual experimental environment provided by the present invention includes real-time monitoring of user behavior information in a virtual laboratory, obtaining behavior sequence data and result logs; identifying the user's experimental steps through an intelligent evaluation model; obtaining standard results of corresponding experimental steps from a preset knowledge base, and automatically evaluating the experimental results of the experimental steps based on the standard results, behavior sequence data and result logs; the present invention combines a deep neural network to conduct a detailed analysis of the student's behavior sequence, and uses a trained deep learning framework to accurately identify the experimental steps, which not only improves the accuracy of the scoring, but also realizes detailed scoring to understand the student's mastery of each step of the experiment, thereby improving the efficiency and accuracy of virtual experiment scoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0052] Figure 1 is an automatic scoring flow chart of an embodiment of the present invention;

[0053] Figure 2 It is a diagram of the evaluation model architecture of an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] See also Figure 1 In the first embodiment, an automatic evaluation method based on user behavior measurement in a virtual experiment environment includes the following steps:

[0056] Step S100: real-time monitoring of user behavior information in the virtual laboratory, and adding valid information to the storage queue in sequence, wherein: the behavior information includes the behavior type, the time point when the behavior occurs, the observed behavior sequence number and the behavior parameters, and the behavior parameters include specific parameters of different behaviors;

[0057] Step S200: Real-time detection of the result log of the monitoring system. If the result log is not updated, continue to add new behavior information to the storage queue; if the result log is updated, obtain the behavior sequence data from the storage queue, and store the behavior sequence and result log in the user's work log;

[0058] Step S300: According to the virtual experiment sample, the behavior sequence data and result logs generated by multiple users in each experimental step are collected in advance, the data sample of the experiment is generated, the intelligent evaluation model is constructed, and the experimental steps of the user are identified;

[0059] Step S400: according to the identified experimental steps, obtain the standard experimental results corresponding to the experimental steps from the preset knowledge base, obtain the user's behavior sequence data and result log, and generate the result score of the experimental steps according to the preset evaluation template;

[0060] Step S500: query the number of times the user has completed the experimental step. If the number of times the experimental step has been completed is greater than the set threshold, the highest score of the experimental step is taken as the result score of the experimental step and added to the total score; if the number of times the experimental step has been completed does not reach the threshold, the new result score is added to the total score;

[0061] Step S600: After the result scoring of the experimental step is completed, determine whether the user experiment is over according to the preset judgment template. If the judgment result is that the experiment is over, directly output the total score; if the judgment result is that the experiment is not over, clear the storage queue and continue monitoring and scoring.

[0062] Specifically, the specific implementation process of step S100 includes:

[0063] The user behavior information of the virtual laboratory is monitored in real time through the keyboard, mouse, microphone and system environment configuration software, where the behavior information includes the behavior type, the time point when the behavior occurs, the observed behavior sequence number and the behavior parameters, and the behavior parameters include the specific parameters of different behaviors;

[0064] Valid information is extracted from the behavior information according to a preset judgment template, and the valid information is added to the storage queue in sequence, wherein: the preset judgment template includes information screening rules for screening valid information, and the information screening rules are based on the correlation between the user behavior type and the experimental steps to screen out the behavior information related to the experiment.

[0065] Specifically, the specific implementation process of step S200 includes:

[0066] The result log of the monitoring system is checked in real time. If the result log is not updated, new behavior information is continuously added to the storage queue. The monitoring system is used to determine in real time whether a new user experimental step result is generated. When a new user experimental step result is generated, the new user experimental step result is written into the result log and updated.

[0067] If the result log has been updated, get the behavior sequence data based on the behavior information of the storage queue:<type,time,index,attributes> , where: type represents the behavior type; time represents the time point when the behavior occurs; index represents the sequence number of the observed behavior; attributes represents the behavior parameters; and the behavior sequence and result log are stored in the user's work log.

[0068] Specifically, the specific implementation process of step S300 includes:

[0069] Step S301: According to the virtual experiment sample, collect the behavior sequence data and result log generated by the user in each experimental step to generate the data sample of the experiment. The data sample is stored in the form of tuples in the format of: <eid,<sid, <actions>,target>>, where: eid represents the number of the experiment; sid represents the number of the experimental step; target represents the result of the experimental step; <actions>Represents the behavior sequence data corresponding to the current experimental step;

[0070] Step S302: Building an intelligent evaluation model based on the experimental data samples:

[0071] Count the number of times and frequency of use of hardware facilities in each behavior sequence, and use the statistical results as global features to form a feature vector G i =[N i,1 ,F i,1 ,N i,2 ,F i,2 ,...,N i,j ,F i,j ], where i represents the experimental step; j represents the total number of hardware facilities; N i,j represents the number of times hardware facility j is used in experimental step i; F i,j It indicates the frequency of use of hardware facility j in experimental step i;

[0072] Based on the experimental data samples, the state machine is used to model the behavior sequence data, the state of the LSTM network is updated, and the LSTM state vector H is generated. i ;

[0073] The feature vector G i and the LSTM state vector H i As input data to the input layer of a fully connected neural network;

[0074] The feature vector and LSTM state vector are fused and extracted through the fully connected layer to generate a new feature vector K i =[G i ;H i ]; where: K i Represents the feature vector of the fully connected layer when the experimental step is i;

[0075] Use the softmax function as the activation function and output the probability distribution of the predicted experimental steps: p i =softmax(W*K i +b); where: W and b are the weight matrix and bias vector of the fully connected layer respectively; softmax() is used to map each element of the input vector to the interval [0, 1], and the sum of all elements is 1;

[0076] Step S303: extract the real-time behavior sequence data and the real-time result log, count the number of times the hardware facilities in the real-time behavior sequence data are used and the frequency of use, and input the real-time behavior sequence data, the real-time result log and the statistical results into the intelligent evaluation model to identify the user's experimental steps.

[0077] Specifically, the specific implementation process of generating the LSTM state vector includes:

[0078] Map different behavior types to different states in the state machine and construct a state set S = {s d |d∈[1,2,…,D]}, where: s d represents the state of the dth behavior type; D represents the number of behavior types; the transition between states is triggered by the occurrence of behavior;

[0079] Construct the transfer set T = {t mn |m,n∈[1,2,…,D]}, where: t mn Indicates that from state s m To state s n transfer;

[0080] Represent each action as an input vector to the LSTM:

[0081] x h =[embedding(type h ),encode(time h ),encode(attribute h ]; where: h represents the time step; x h Represents the input vector of the h-th time step LSTM; type h Indicates the behavior type of the hth time step; time h Indicates the time point at which the behavior of the hth time step occurs; attribute h represents the behavior parameters of the hth time step; embedding is an embedding function used to map the behavior type to a low-dimensional vector space; encode is an encoding function used to convert time and behavior parameters into a format suitable for LSTM input;

[0082] LSTM network state update: f h =σ(W f *[y h-1 ,x h ]+b f );

[0083] e h =σ(W e *[y h-1 ,x h ]+b e );

[0084]

[0085] o h =σ(W o *[y h-1 ,x h ]+b o );

[0086] y h =o h *tanh(c h );where: f h represents the forget gate of the hth time step; σ represents the sigmoid function; e h represents the input gate of the hth time step; represents the candidate cell state at the hth time step; c h represents the cell state at the hth time step; o h represents the output gate of the hth time step; y h represents the hidden state of the hth time step; tanh represents the hyperbolic tangent function; W f , W e , W c , W o Represents the preset weight matrix; b f 、b e 、b c 、b o Represents the preset bias item;

[0087] Process each action in turn and update the hidden state, which is recorded in the LSTM state vector H i middle.

[0088] Specifically, the specific implementation process of step S400 includes:

[0089] According to the identified experimental steps, the standard experimental results corresponding to the experimental steps are obtained from the preset knowledge base, wherein: the preset knowledge base stores the experimental steps and the corresponding standard experimental results, and one experimental step corresponds to one standard experimental result; based on the standard experimental results, behavior sequence data and result logs, the result scores of the experimental steps are generated through the preset evaluation template, wherein: the preset evaluation template includes result score generation rules for experimental scoring.

[0090] Specifically, the specific implementation process of step S500 includes:

[0091] Query the number of times the user completes the experimental steps in the experiment. If the number of completions of the experimental steps is greater than the preset threshold, take the highest score in the result scores of the experimental steps as the result scores of the experimental steps and add it to the total score; if the number of completions of the experimental steps does not reach the threshold, add the new result score to the total score.

[0092] Specifically, the specific implementation process of step S600 includes:

[0093] When the result scoring of the experimental step is completed, the preset judgment template is used to determine whether the user experiment is over. If the judgment result is that the experiment is over, the total score is directly output; if the judgment result is that the experiment is not over, the storage queue is cleared, and monitoring and scoring continue, where: the preset judgment template includes judgment rules for determining whether the user experiment is over.

[0094] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0095] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.< / actions> < / actions> < / actions> < / actions>

Claims

1. An automatic evaluation method based on user behavior measurement in a virtual experiment environment, characterized in that: The method comprises the following steps: Step S100: real-time monitoring of user behavior information in the virtual laboratory, and adding valid information to the storage queue in sequence, wherein: the behavior information includes the behavior type, the time point when the behavior occurs, the observed behavior sequence number and the behavior parameters, and the behavior parameters include specific parameters of different behaviors; Step S200: Real-time detection of the result log of the monitoring system. If the result log is not updated, continue to add new behavior information to the storage queue; if the result log is updated, obtain the behavior sequence data from the storage queue, and store the behavior sequence and result log in the user's work log; Step S300: According to the virtual experiment sample, the behavior sequence data and result logs generated by multiple users in each experimental step are collected in advance, the data sample of the experiment is generated, the intelligent evaluation model is constructed, and the experimental steps of the user are identified; Step S400: according to the identified experimental steps, obtain the standard experimental results corresponding to the experimental steps from the preset knowledge base, obtain the user's behavior sequence data and result log, and generate the result score of the experimental steps according to the preset evaluation template; Step S500: query the number of times the user has completed the experimental step. If the number of times the experimental step has been completed is greater than a set threshold, the highest score of the experimental step is taken as the result score of the experimental step and added to the total score; if the number of times the experimental step has been completed does not reach the threshold, the new result score is added to the total score; Step S600: After the result scoring of the experimental step is completed, determine whether the user experiment is over according to the preset judgment template. If the judgment result is that the experiment is over, directly output the total score; if the judgment result is that the experiment is not over, clear the storage queue and continue monitoring and scoring.

2. The automatic evaluation method based on user behavior measurement in a virtual experiment environment according to claim 1 is characterized in that: The specific implementation process of step S100 includes: The user behavior information of the virtual laboratory is monitored in real time through the keyboard, mouse, microphone and system environment configuration software, where the behavior information includes the behavior type, the time point when the behavior occurs, the observed behavior sequence number and the behavior parameters, and the behavior parameters include the specific parameters of different behaviors; Valid information is extracted from the behavior information according to a preset judgment template, and the valid information is added to the storage queue in sequence, wherein: the preset judgment template includes information screening rules for screening valid information, and the information screening rules are based on the correlation between the user behavior type and the experimental steps to screen out the behavior information related to the experiment.

3. The automatic evaluation method based on user behavior measurement in a virtual experiment environment according to claim 2 is characterized in that: The specific implementation process of step S200 includes: The result log of the monitoring system is checked in real time. If the result log is not updated, new behavior information is continuously added to the storage queue. The monitoring system is used to determine in real time whether a new user experimental step result is generated. When a new user experimental step result is generated, the new user experimental step result is written into the result log and updated. If the result log has been updated, get the behavior sequence data based on the behavior information of the storage queue:<type,time,index,attributes> , where: type represents the behavior type; time represents the time point when the behavior occurs; index represents the sequence number of the observed behavior; attributes represents the behavior parameters; and the behavior sequence and result log are stored in the user's work log.

4. The automatic evaluation method based on user behavior measurement in a virtual experiment environment according to claim 3 is characterized in that: The specific implementation process of step S300 includes: Step S301: According to the virtual experiment sample, collect the behavior sequence data and result log generated by the user in each experimental step to generate the data sample of the experiment. The data sample is stored in the form of tuples in the format of: <eid,<sid, <actions>,target>>, where: eid represents the number of the experiment; sid represents the number of the experimental step; target represents the result of the experimental step; <actions> Represents the behavior sequence data corresponding to the current experimental step;< / actions> < / actions> Step S302: Building an intelligent evaluation model based on the experimental data samples: Count the number of times and frequency of use of hardware facilities in each behavior sequence, and use the statistical results as global features to form a feature vector G i =[N i,1 ,F i,1 ,N i,2 ,F i,2 ,...,N i,j ,F i,j ], where i represents the experimental step; j represents the total number of hardware facilities; N i,j represents the number of times hardware facility j is used in experimental step i; F i,j It indicates the frequency of use of hardware facility j in experimental step i; Based on the experimental data samples, the state machine is used to model the behavior sequence data, the state of the LSTM network is updated, and the LSTM state vector H is generated. i ; The feature vector G i and the LSTM state vector H i As input data to the input layer of a fully connected neural network; The feature vector and LSTM state vector are fused and extracted through the fully connected layer to generate a new feature vector K i =[G i ;H i ]; where: K i Represents the feature vector of the fully connected layer when the experimental step is i; Use the softmax function as the activation function and output the probability distribution of the predicted experimental steps: p i =softmax(W*K i +b); where: W and b are the weight matrix and bias vector of the fully connected layer respectively; softmax() is used to map each element of the input vector to the interval [0, 1], and the sum of all elements is 1; Step S303: extract the real-time behavior sequence data and the real-time result log, count the number of times the hardware facilities in the real-time behavior sequence data are used and the frequency of use, and input the real-time behavior sequence data, the real-time result log and the statistical results into the intelligent evaluation model to identify the user's experimental steps.

5. The automatic evaluation method based on user behavior measurement in a virtual experiment environment according to claim 4 is characterized in that: The specific implementation process of generating the LSTM state vector includes: Map different behavior types to different states in the state machine and construct a state set S = {s d |d∈[1,2,…,D]}, where: s d represents the state of the dth behavior type; D represents the number of behavior types; the transition between states is triggered by the occurrence of behavior; Construct the transfer set T = {t mn |m,n∈[1,2,…,D]}, where: t mn Indicates that from state s m To state s n transfer; Represent each action as an input vector to the LSTM: x h =[embedding(type h ),encode(time h ),encode(attribute h ]; where: h represents the time step; x h Represents the input vector of the h-th time step LSTM; type h Indicates the behavior type of the hth time step; time h Indicates the time point at which the behavior of the hth time step occurs; attribute h represents the behavior parameters of the hth time step; embedding is an embedding function used to map the behavior type to a low-dimensional vector space; encode is an encoding function used to convert time and behavior parameters into a format suitable for LSTM input; LSTM network state update: f h =σ(W f *[y h-1 ,x h ]+b f ); e h =σ(W e *[y h-1 ,x h ]+b e ); the h =σ(W o *[y h-1 ,x h ]+b o ); y h =o h *tanh(c h );where: f h represents the forget gate of the hth time step; σ represents the sigmoid function; e h represents the input gate of the hth time step; represents the candidate cell state at the hth time step; c h represents the cell state at the hth time step; o h represents the output gate of the hth time step; y h represents the hidden state of the hth time step; tanh represents the hyperbolic tangent function; W f , W e , W c , W o represents the preset weight matrix; b f 、b e 、b c 、b o Represents the preset bias item; Process each action in turn and update the hidden state, which is recorded in the LSTM state vector H i middle.

6. The automatic evaluation method based on user behavior measurement in a virtual experiment environment according to claim 4 is characterized in that: The specific implementation process of step S400 includes: According to the identified experimental steps, the standard experimental results corresponding to the experimental steps are obtained from a preset knowledge base, wherein: the preset knowledge base stores the experimental steps and the corresponding standard experimental results, and one experimental step corresponds to one standard experimental result; based on the standard experimental results, behavior sequence data and result logs, a result score for the experimental steps is generated through a preset evaluation template, wherein: the preset evaluation template includes result score generation rules for experimental scoring.

7. The automatic evaluation method based on user behavior measurement in a virtual experiment environment according to claim 6 is characterized in that: The specific implementation process of step S500 includes: Query the number of times the user completes the experimental steps in the experiment. If the number of completions of the experimental steps is greater than a preset threshold, take the highest score among the result scores of the experimental steps as the result score of the experimental steps and add it to the total score; if the number of completions of the experimental steps does not reach the threshold, add the new result score to the total score.

8. The automatic evaluation method based on user behavior measurement in a virtual experiment environment according to claim 7 is characterized in that: The specific implementation process of step S600 includes: When the result scoring of the experimental step is completed, the preset judgment template is used to determine whether the user experiment is over. If the judgment result is that the experiment is over, the total score is directly output; if the judgment result is that the experiment is not over, the storage queue is cleared, and monitoring and scoring continue, where: the preset judgment template includes judgment rules for determining whether the user experiment is over.

Images