Mathematical test question knowledge point automatic labeling method based on deep learning

Abstract

The invention discloses a mathematical test question knowledge point automatic labeling method based on deep learning, and belongs to the technical field of mathematical intelligent teaching, and the method comprises the steps of data preparation, text preprocessing, data structure conversion, automatic labeling model construction, model optimization and labeling result generation. According to the method, the two-way long-short-term network model is adopted for automatic labeling model construction, context information in mathematical test questions is more comprehensively understood through a two-way structure, the long-distance dependency problem is effectively processed, complex knowledge points are accurately labeled, and the overall reliability of the method is enhanced; the improved particle swarm optimization algorithm is adopted for parameter tuning, local optimum is effectively avoided by introducing the distribution estimation algorithm, the global search capability and the convergence speed are improved, the model is better in performance on diversified mathematical test questions, the adaptability of parameter tuning is enhanced, and therefore the model performance is improved, and the method is suitable for large-scale popularization and application. The automatic marking of the knowledge points of the mathematical test questions is more efficient and accurate, and the overall practicability of the method is enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of mathematical intelligent teaching, and specifically refers to a method for automatically labeling knowledge points of mathematical test questions based on deep learning. Background Art

[0002] The automatic annotation of knowledge points in mathematics test questions uses deep learning technology to analyze mathematics test questions, thereby automatically annotating the knowledge points in mathematics test questions. It aims to help teachers quickly understand the knowledge points involved in each question, saving time and energy for manual annotation, and helping to improve the efficiency of building a mathematics question bank and ensure the accuracy of question classification, thereby improving teaching efficiency and promoting the intelligent development of mathematics teaching.

[0003] However, in the existing process of automatic labeling of knowledge points in mathematics test questions, there is a technical problem that mathematics test questions have complex contextual information, and the context in mathematics test questions involves long-distance dependencies. Traditional models are prone to lose important information when processing mathematics test questions, which affects the accuracy of knowledge point labeling; there is a technical problem that in the existing process of automatic labeling of knowledge points in mathematics test questions, the traditional particle swarm optimization algorithm is prone to fall into local optimality when performing parameter tuning, resulting in unsatisfactory parameter tuning effects, affecting model performance, and the traditional parameter tuning method has poor adaptability to new types of questions. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a method for automatic annotation of knowledge points in mathematics test questions based on deep learning. In view of the technical problem that in the existing automatic annotation process of knowledge points in mathematics test questions, the mathematics test questions have complex contextual information, and the context in the mathematics test questions involves long-distance dependencies. The traditional model is prone to lose important information when processing mathematics test questions, which affects the accuracy of knowledge point annotation. This scheme creatively adopts a bidirectional long-short-term network model to construct an automatic annotation model. Through the bidirectional structure, the contextual information in the mathematics test questions is more comprehensively understood, and the long-distance dependency problem is effectively handled, and important information at a long distance is remembered, so as to accurately annotate complex knowledge points and enhance the overall accuracy of the method. The reliability of the whole system; in view of the technical problems that in the process of automatic labeling of knowledge points in existing mathematics test questions, the traditional particle swarm optimization algorithm is prone to fall into the local optimum when performing parameter tuning, resulting in unsatisfactory parameter tuning effect, affecting model performance, and the traditional parameter tuning method has poor adaptability to new types of questions, this solution creatively adopts the improved particle swarm optimization algorithm for parameter tuning, and effectively avoids falling into the local optimum by introducing the distribution estimation algorithm, thereby improving the global search capability and convergence speed, and making the model perform better on a variety of mathematics test questions, enhancing the adaptability of parameter tuning, thereby further improving the model performance, making the automatic labeling of knowledge points in mathematics test questions more efficient and accurate, and enhancing the overall practicality of the method.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides a method for automatically labeling knowledge points of mathematics test questions based on deep learning, and the method comprises the following steps:

[0006] Step S1: data preparation;

[0007] Step S2: text preprocessing;

[0008] Step S3: data structure conversion;

[0009] Step S4: automatic annotation model construction;

[0010] Step S5: model optimization;

[0011] Step S6: Generate annotation results.

[0012] Furthermore, in step S1, the data preparation is used to collect the original data required for automatic annotation of math test question knowledge points, specifically to collect math test questions and corresponding knowledge point labels from a math test question library to obtain math test question text data and knowledge point labels.

[0013] Furthermore, in step S2, the text preprocessing is performed by removing duplicate questions, removing stop words and performing word segmentation processing on the math test question text data to obtain math test question optimized data.

[0014] Furthermore, in step S3, the data structure conversion is used to convert the text into a vector representation, specifically using a pre-trained BERT model to perform word embedding on the math test question optimization data to obtain a math test question word embedding vector.

[0015] Further, in step S4, the automatic annotation model is constructed to construct an automatic annotation model for mathematical test knowledge points, specifically using a bidirectional long-short term network model to construct the automatic annotation model to obtain an automatic annotation model for mathematical test knowledge points;

[0016] The bidirectional long-short-term network model includes an input layer, a bidirectional long-short-term network layer, a fully connected layer and an output layer;

[0017] The bidirectional long-term and short-term network layer is used to capture bidirectional context information and improve the annotation accuracy;

[0018] The fully connected layer is used to further extract features;

[0019] The automatic annotation model construction includes the following steps:

[0020] Step S41: construct an input layer for receiving model input, specifically receiving the math test word embedding vector as the model input through the input layer, and taking the knowledge point label as the model label;

[0021] Step S42: constructing a bidirectional long-term and short-term network layer, including the following steps:

[0022] Step S421: construct a forward long-term and short-term network layer, and calculate through the forward long-term and short-term network layer to obtain the forward long-term and short-term hidden state. The calculation formula is:

[0023] ;

[0024] Where k is the time step index, is the forward long-term and short-term hidden state at the kth time step, specifically the output of the forward long-term and short-term network layer at the kth time step, is the forward long-term and short-term network layer calculation function, X k is the model input at the kth time step;

[0025] Step S422: construct a backward long-term and short-term network layer, and calculate through the backward long-term and short-term network layer to obtain the backward long-term and short-term hidden state. The calculation formula is:

[0026] ;

[0027] In the formula, is the backward long-short-term hidden state at the kth time step, specifically the output of the backward long-short-term network layer at the kth time step, It is the backward long-term and short-term network layer calculation function;

[0028] Step S423: By performing feature concatenation on the forward long-term and short-term hidden state and the backward long-term and short-term hidden state, a bidirectional long-term and short-term hidden state is obtained. The calculation formula is:

[0029] ;

[0030] In the formula, H k is the bidirectional long-short-term hidden state at the k-th time step, specifically refers to the output result of the bidirectional long-short-term network layer at the k-th time step, and concat(·) is the feature concatenation function;

[0031] Step S43: construct a fully connected layer, calculate the bidirectional long and short-term hidden states through the fully connected layer, and obtain the fully connected features. The calculation formula is:

[0032] ;

[0033] In the formula, z k is the fully connected feature at the kth time step, W z is the weight of the fully connected layer, b z is the fully connected layer bias term;

[0034] Step S44: construct an output layer to output the probability distribution of each knowledge point, specifically using a softmax function in the output layer to calculate the fully connected features to obtain the model output result;

[0035] Step S45: constructing a bidirectional long-short-term network model, specifically, constructing a bidirectional long-short-term network model by constructing an input layer, constructing a bidirectional long-short-term network layer, constructing a fully connected layer, and constructing an output layer to obtain a bidirectional long-short-term network model;

[0036] Step S46: initial training, specifically, initial training of the bidirectional long-short term network model to obtain an automatic annotation model for knowledge points of mathematics test questions.

[0037] Further, in step S5, the model optimization is used to optimize the model parameters, specifically using an improved particle swarm optimization algorithm to tune the parameters of the automatic annotation model for mathematics test questions and knowledge points, and obtain the optimal model for automatic annotation of mathematics test questions and knowledge points, including the following steps:

[0038] Step S51: improving the particle swarm optimization algorithm, specifically by introducing a distribution estimation algorithm into the particle swarm optimization algorithm to improve the particle swarm optimization algorithm to obtain an improved particle swarm optimization algorithm;

[0039] Step S52: constructing an improved particle swarm optimization algorithm, including the following steps:

[0040] Step S521: Initialize the particle swarm, specifically, randomly initialize the position of each particle in the interval [0,1], and perform particle swarm optimization. The particle position is used to represent the candidate solution of the model parameters;

[0041] Step S522: Set the maximum number of iterations and initialize the probability vector. The calculation formula of the initialized probability vector is:

[0042] ;

[0043] Where j is the particle dimension index, P j is the probability that a particle in the jth dimension takes the value 1, M is the number of particles, i is the particle index, and p ij is the individual optimal position of the ith particle in the jth dimension;

[0044] Step S523: Update the particle swarm position, the calculation formula is:

[0045] ;

[0046] Where t is the iteration index, s ij(t+1) is the position of the ith particle in the jth dimension at the t+1th iteration, R is a random number in the range [0,1], is a control parameter used to control the particle swarm position update method, p gj (t) is the global optimal position in the jth dimension at the tth iteration;

[0047] Step S524: using the cross entropy loss function as a fitness function to calculate the particle fitness for evaluating the performance of the particle position;

[0048] Step S525: Update the individual optimal position, the calculation formula is:

[0049] ;

[0050] In the formula, p ij (t+1) is the individual optimal position of the ith particle in the jth dimension in the t+1th iteration, p ij (t) is the individual optimal position of the ith particle in the jth dimension in the tth iteration, f(·) is the fitness function, specifically the cross entropy loss function;

[0051] Step S526: Update the global optimal position, the calculation formula is:

[0052] ;

[0053] In the formula, p gj (t+1) is the global optimal position in the jth dimension at the t+1th iteration;

[0054] Step S527: Update the probability vector, the calculation formula is:

[0055] ;

[0056] Step S528: iterative training, specifically, repeatedly executing steps S523 to S527 until the maximum number of iterations is reached, and the global optimal position is obtained as the optimal parameter of the model;

[0057] Step S53: parameter tuning, specifically, tuning the parameters of the automatic labeling model for math test questions and knowledge points by using an improved particle swarm optimization algorithm to obtain the optimal model for the automatic labeling of math test questions and knowledge points.

[0058] Further, in step S6, the annotation result generation is used to generate the automatic annotation result of the knowledge points of the mathematics test questions, specifically, the mathematics test questions are automatically annotated through the optimal model for automatic annotation of the knowledge points of the mathematics test questions, and the automatic annotation result of the knowledge points of the mathematics test questions is obtained.

[0059] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0060] (1) In view of the technical problem that in the existing automatic annotation process of knowledge points in mathematics test questions, mathematics test questions have complex context information, and the context in mathematics test questions involves long-distance dependencies. Traditional models are prone to lose important information when processing mathematics test questions, which affects the accuracy of knowledge point annotation. This solution creatively uses a bidirectional long-short-term network model to construct an automatic annotation model. Through the bidirectional structure, it can more comprehensively understand the context information in mathematics test questions, effectively handle long-distance dependencies, remember important information at a long distance, and thus accurately annotate complex knowledge points, enhancing the overall reliability of the method.

[0061] (2) In view of the technical problems that the traditional particle swarm optimization algorithm is prone to fall into the local optimum when performing parameter tuning in the process of automatic labeling of knowledge points in existing mathematics test questions, resulting in unsatisfactory parameter tuning results and affecting model performance, and the traditional parameter tuning method has poor adaptability to new types of questions, this scheme creatively adopts an improved particle swarm optimization algorithm for parameter tuning. By introducing a distribution estimation algorithm, it effectively avoids falling into the local optimum, improves the global search capability and convergence speed, and makes the model perform better on a variety of mathematics test questions, enhances the adaptability of parameter tuning, and thus further improves the model performance, making the automatic labeling of knowledge points in mathematics test questions more efficient and accurate, and enhancing the overall practicality of the method. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 A schematic diagram of a flow chart of a method for automatically labeling knowledge points of mathematics test questions based on deep learning provided by the present invention;

[0063] Figure 2 is a schematic flow chart of step S4;

[0064] Figure 3 is a schematic flow chart of step S5;

[0065] Figure 4 It is a flowchart diagram of step S52.

[0066] 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. DETAILED DESCRIPTION

[0067] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0068] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0069] Example 1, see Figure 1 The present invention provides a method for automatically labeling knowledge points of mathematics test questions based on deep learning, and the method comprises the following steps:

[0070] Step S1: data preparation;

[0071] Step S2: text preprocessing;

[0072] Step S3: data structure conversion;

[0073] Step S4: automatic annotation model construction;

[0074] Step S5: model optimization;

[0075] Step S6: Generate annotation results.

[0076] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the data preparation is used to collect the original data required for automatic annotation of math test knowledge points, specifically to collect math test questions and corresponding knowledge point labels from a math test question library to obtain math test question text data and knowledge point labels.

[0077] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, the text is preprocessed by removing duplicate questions, eliminating stop words and performing word segmentation on the math test text data to obtain math test optimized data.

[0078] Example 4, see Figure 1 This embodiment is based on the above embodiment. In step S3, the data structure conversion is used to convert the text into a vector representation, specifically using a pre-trained BERT model to perform word embedding on the math test optimization data to obtain a math test word embedding vector.

[0079] Example 5, see Figure 1 and Figure 2This embodiment is based on the above embodiment. In step S4, the automatic annotation model is constructed to construct an automatic annotation model for mathematical test knowledge points. Specifically, a bidirectional long-short term network model is used to construct the automatic annotation model to obtain an automatic annotation model for mathematical test knowledge points.

[0080] The bidirectional long-short-term network model includes an input layer, a bidirectional long-short-term network layer, a fully connected layer and an output layer;

[0081] The bidirectional long-term and short-term network layer is used to capture bidirectional context information and improve the annotation accuracy;

[0082] The fully connected layer is used to further extract features;

[0083] The automatic annotation model construction includes the following steps:

[0084] Step S41: construct an input layer for receiving model input, specifically receiving the math test word embedding vector as the model input through the input layer, and taking the knowledge point label as the model label;

[0085] Step S42: constructing a bidirectional long-term and short-term network layer, including the following steps:

[0086] Step S421: construct a forward long-term and short-term network layer, and calculate through the forward long-term and short-term network layer to obtain the forward long-term and short-term hidden state. The calculation formula is:

[0087] ;

[0088] Where k is the time step index, is the forward long-term and short-term hidden state at the kth time step, specifically the output of the forward long-term and short-term network layer at the kth time step, is the forward long-term and short-term network layer calculation function, X k is the model input at the kth time step;

[0089] Step S422: construct a backward long-term and short-term network layer, and calculate through the backward long-term and short-term network layer to obtain the backward long-term and short-term hidden state. The calculation formula is:

[0090] ;

[0091] In the formula, is the backward long-short-term hidden state at the kth time step, specifically the output of the backward long-short-term network layer at the kth time step, It is the backward long-term and short-term network layer calculation function;

[0092] Step S423: By performing feature concatenation on the forward long-term and short-term hidden state and the backward long-term and short-term hidden state, a bidirectional long-term and short-term hidden state is obtained. The calculation formula is:

[0093] ;

[0094] In the formula, H k is the bidirectional long-short-term hidden state at the k-th time step, specifically refers to the output result of the bidirectional long-short-term network layer at the k-th time step, and concat(·) is the feature concatenation function;

[0095] Step S43: construct a fully connected layer, calculate the bidirectional long and short-term hidden states through the fully connected layer, and obtain the fully connected features. The calculation formula is:

[0096] ;

[0097] In the formula, z k is the fully connected feature at the kth time step, W z is the weight of the fully connected layer, b z is the fully connected layer bias term;

[0098] Step S44: construct an output layer to output the probability distribution of each knowledge point, specifically using a softmax function in the output layer to calculate the fully connected features to obtain the model output result;

[0099] Step S45: constructing a bidirectional long-short-term network model, specifically, constructing a bidirectional long-short-term network model by constructing an input layer, constructing a bidirectional long-short-term network layer, constructing a fully connected layer, and constructing an output layer to obtain a bidirectional long-short-term network model;

[0100] Step S46: initial training, specifically, initial training of the bidirectional long-term and short-term network model to obtain a mathematical test question knowledge point automatic labeling model;

[0101] By performing the above operations, in order to solve the technical problem that in the process of automatic labeling of knowledge points in existing mathematics test questions, the mathematics test questions have complex contextual information, and the context in the mathematics test questions involves long-distance dependencies. The traditional model is prone to lose important information when processing mathematics test questions, which affects the accuracy of knowledge point labeling, this solution creatively adopts a bidirectional long-short-term network model to construct an automatic labeling model. Through the bidirectional structure, it can more comprehensively understand the contextual information in the mathematics test questions, effectively handle the long-distance dependency problem, remember important information at a long distance, thereby accurately labeling complex knowledge points and enhancing the overall reliability of the method.

[0102] Example 6, see Figure 1 , Figure 3 and Figure 4This embodiment is based on the above embodiment. In step S5, the model optimization is used to optimize the model parameters. Specifically, the improved particle swarm optimization algorithm is used to tune the parameters of the automatic labeling model for mathematical test questions and knowledge points, and the optimal model for automatic labeling of mathematical test questions and knowledge points is obtained, including the following steps:

[0103] Step S51: improving the particle swarm optimization algorithm, specifically by introducing a distribution estimation algorithm into the particle swarm optimization algorithm to improve the particle swarm optimization algorithm to obtain an improved particle swarm optimization algorithm;

[0104] Step S52: constructing an improved particle swarm optimization algorithm, including the following steps:

[0105] Step S521: Initialize the particle swarm, specifically, randomly initialize the position of each particle in the interval [0,1], and perform particle swarm optimization. The particle position is used to represent the candidate solution of the model parameters;

[0106] Step S522: Set the maximum number of iterations and initialize the probability vector. The calculation formula of the initialized probability vector is:

[0107] ;

[0108] Where j is the particle dimension index, P j is the probability that a particle in the jth dimension takes the value 1, M is the number of particles, i is the particle index, and p ij is the individual optimal position of the ith particle in the jth dimension;

[0109] Step S523: Update the particle swarm position, the calculation formula is:

[0110] ;

[0111] Where t is the iteration index, s ij (t+1) is the position of the ith particle in the jth dimension at the t+1th iteration, R is a random number in the range [0,1], is a control parameter used to control the particle swarm position update method, p gj (t) is the global optimal position in the jth dimension at the tth iteration;

[0112] Step S524: using the cross entropy loss function as a fitness function to calculate the particle fitness for evaluating the performance of the particle position;

[0113] Step S525: Update the individual optimal position, the calculation formula is:

[0114] ;

[0115] In the formula, pij (t+1) is the individual optimal position of the ith particle in the jth dimension in the t+1th iteration, p ij (t) is the individual optimal position of the ith particle in the jth dimension in the tth iteration, f(·) is the fitness function, specifically the cross entropy loss function;

[0116] Step S526: Update the global optimal position, the calculation formula is:

[0117] ;

[0118] In the formula, p gj (t+1) is the global optimal position in the jth dimension at the t+1th iteration;

[0119] Step S527: Update the probability vector, the calculation formula is:

[0120] ;

[0121] Step S528: iterative training, specifically, repeatedly executing steps S523 to S527 until the maximum number of iterations is reached, and the global optimal position is obtained as the optimal parameter of the model;

[0122] Step S53: parameter tuning, specifically, tuning the parameters of the automatic labeling model for math test questions knowledge points through an improved particle swarm optimization algorithm to obtain the optimal model for automatic labeling of math test questions knowledge points;

[0123] By performing the above operations, in order to solve the technical problems that in the process of automatic labeling of knowledge points in existing mathematics test questions, the traditional particle swarm optimization algorithm is prone to fall into the local optimum when performing parameter tuning, resulting in unsatisfactory parameter tuning effect, affecting model performance, and the traditional parameter tuning method has poor adaptability to new types of questions, this solution creatively adopts the improved particle swarm optimization algorithm for parameter tuning, and effectively avoids falling into the local optimum by introducing the distribution estimation algorithm, thereby improving the global search capability and convergence speed, making the model perform better on a variety of mathematics test questions, enhancing the adaptability of parameter tuning, thereby further improving the model performance, making the automatic labeling of knowledge points in mathematics test questions more efficient and accurate, and enhancing the overall practicality of the method.

[0124] Embodiment 7, see Figure 1 This embodiment is based on the above embodiment. In step S6, the annotation result is generated to generate the automatic annotation result of the knowledge points of the mathematics test questions. Specifically, the mathematics test questions are automatically annotated through the optimal model for automatic annotation of the knowledge points of the mathematics test questions to obtain the automatic annotation result of the knowledge points of the mathematics test questions.

[0125] 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.

[0126] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0127] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A method for automatic annotation of knowledge points in mathematics test questions based on deep learning, characterized by: The method comprises the following steps: Step S1: data preparation; Step S2: text preprocessing; Step S3: data structure conversion; Step S4: automatic annotation model construction, which is used to construct an automatic annotation model for math test knowledge points, specifically using a bidirectional long-short term network model to construct the automatic annotation model to obtain an automatic annotation model for math test knowledge points; The long-short-term network model includes an input layer, a bidirectional long-short-term network layer, a fully connected layer and an output layer; The long-term and short-term network layer is used to capture bidirectional context information; The fully connected layer is used to further extract features; Step S5: Model optimization, using an improved particle swarm optimization algorithm to tune the parameters of the automatic labeling model for math test questions and obtain the optimal model for automatic labeling of math test questions and knowledge points; Step S6: Generate annotation results.

2. The method for automatic annotation of knowledge points of mathematics test questions based on deep learning according to claim 1, characterized in that: In step S4, the automatic annotation model is constructed, including the following steps: Step S41: construct an input layer for receiving model input, specifically receiving the math test word embedding vector as the model input through the input layer, and taking the knowledge point label as the model label; Step S42: constructing a bidirectional long-term and short-term network layer, including the following steps: Step S421: construct a forward long-term and short-term network layer, and calculate through the forward long-term and short-term network layer to obtain the forward long-term and short-term hidden state. The calculation formula is: ； Where k is the time step index, is the forward long-term and short-term hidden state at the kth time step, specifically the output of the forward long-term and short-term network layer at the kth time step, is the forward long-term and short-term network layer calculation function, X k is the model input at the kth time step; Step S422: construct a backward long-term and short-term network layer, and calculate through the backward long-term and short-term network layer to obtain the backward long-term and short-term hidden state. The calculation formula is: ； In the formula, is the backward long-short-term hidden state at the kth time step, specifically the output of the backward long-short-term network layer at the kth time step, It is the backward long-term and short-term network layer calculation function; Step S423: By performing feature concatenation on the forward long-term and short-term hidden state and the backward long-term and short-term hidden state, a bidirectional long-term and short-term hidden state is obtained. The calculation formula is: ； In the formula, H k is the bidirectional long-short-term hidden state at the k-th time step, specifically refers to the output result of the bidirectional long-short-term network layer at the k-th time step, and concat(·) is the feature concatenation function; Step S43: construct a fully connected layer, calculate the bidirectional long and short-term hidden states through the fully connected layer, and obtain the fully connected features. The calculation formula is: ； In the formula, z k is the fully connected feature at the kth time step, W z is the weight of the fully connected layer, b z is the fully connected layer bias term; Step S44: construct an output layer to output the probability distribution of each knowledge point, specifically using a softmax function in the output layer to calculate the fully connected features to obtain the model output result; Step S45: constructing a bidirectional long-short-term network model, specifically, constructing a bidirectional long-short-term network model by constructing an input layer, constructing a bidirectional long-short-term network layer, constructing a fully connected layer, and constructing an output layer to obtain a bidirectional long-short-term network model; Step S46: initial training, specifically, initial training of the bidirectional long-short term network model to obtain an automatic annotation model for knowledge points of mathematics test questions.

3. The method for automatic annotation of knowledge points of mathematics test questions based on deep learning according to claim 2 is characterized in that: In step S5, the model optimization is used to optimize the model parameters, specifically using an improved particle swarm optimization algorithm to tune the parameters of the automatic labeling model for mathematics test questions and knowledge points, and obtain the optimal model for automatic labeling of mathematics test questions and knowledge points, including the following steps: Step S51: improving the particle swarm optimization algorithm, specifically by introducing a distribution estimation algorithm into the particle swarm optimization algorithm to improve the particle swarm optimization algorithm to obtain an improved particle swarm optimization algorithm; Step S52: constructing an improved particle swarm optimization algorithm, including the following steps: Step S521: Initialize the particle swarm, specifically, randomly initialize the position of each particle in the interval [0,1], and perform particle swarm optimization. The particle position is used to represent the candidate solution of the model parameters; Step S522: Set the maximum number of iterations and initialize the probability vector. The calculation formula of the initialized probability vector is: ； Where j is the particle dimension index, P j is the probability that a particle in the jth dimension takes the value 1, M is the number of particles, i is the particle index, and p ij is the individual optimal position of the ith particle in the jth dimension; Step S523: Update the particle swarm position, the calculation formula is: ； Where t is the iteration index, s ij (t+1) is the position of the ith particle in the jth dimension at the t+1th iteration, R is a random number in the range [0,1], is a control parameter used to control the particle swarm position update method, p gj (t) is the global optimal position in the jth dimension at the tth iteration; Step S524: using the cross entropy loss function as a fitness function to calculate the particle fitness for evaluating the performance of the particle position; Step S525: Update the individual optimal position, the calculation formula is: ； In the formula, p ij (t+1) is the individual optimal position of the ith particle in the jth dimension in the t+1th iteration, p ij (t) is the individual optimal position of the ith particle in the jth dimension in the tth iteration, f(·) is the fitness function, specifically the cross entropy loss function; Step S526: Update the global optimal position, the calculation formula is: ； In the formula, p gj (t+1) is the global optimal position in the jth dimension at the t+1th iteration; Step S527: Update the probability vector, the calculation formula is: ； Step S528: iterative training, specifically, repeatedly executing steps S523 to S527 until the maximum number of iterations is reached, and the global optimal position is obtained as the optimal parameter of the model; Step S53: parameter tuning, specifically, tuning the parameters of the automatic labeling model for math test questions and knowledge points by using an improved particle swarm optimization algorithm to obtain the optimal model for the automatic labeling of math test questions and knowledge points.

4. The method for automatic annotation of knowledge points of mathematics test questions based on deep learning according to claim 3 is characterized in that: In step S1, the data preparation is used to collect the original data required for automatic annotation of math test question knowledge points, specifically to collect math test questions and corresponding knowledge point labels from a math test question library to obtain math test question text data and knowledge point labels.

5. The method for automatic annotation of knowledge points of mathematics test questions based on deep learning according to claim 4 is characterized in that: In step S2, the text is preprocessed by removing duplicate questions, removing stop words and performing word segmentation on the math test question text data to obtain math test question optimized data.

6. The method for automatic annotation of knowledge points of mathematics test questions based on deep learning according to claim 5, characterized in that: In step S3, the data structure conversion is used to convert the text into a vector representation, specifically using a pre-trained BERT model to perform word embedding on the math test question optimization data to obtain a math test question word embedding vector.

7. The method for automatic annotation of knowledge points of mathematics test questions based on deep learning according to claim 6 is characterized by: In step S6, the annotation result is generated to generate the automatic annotation result of the knowledge points of the mathematics test questions, specifically, the mathematics test questions are automatically annotated through the optimal model for automatic annotation of the knowledge points of the mathematics test questions to obtain the automatic annotation result of the knowledge points of the mathematics test questions.

Images