CN110705645A - An English teaching quality assessment method based on SOFM neural network - Google Patents

Abstract

The invention relates to an English teaching quality assessment method based on a SOFM neural network, which solves the technical problem of complex output results and constructs a SOFM neural network model formed by connecting two stages of SOFM neural sub-networks in series by adopting step 1; step 2, defining a threshold value R, inputting English teaching quality evaluation scores, carrying out SOFM learning algorithm training on a 1 st-level SOFM neural sub-network, and calculating various initial clustering centers; step 3, judging whether the neuron in the competition layer is deleted or not; step 4, judging whether the neurons in the competition layer are clustered secondarily; and step 5, dividing English teaching object data to be evaluated into c evaluation grades according to the final clustering center value and outputting the evaluation grades, so that the problem is well solved, and the method can be used for English teaching.

Description

An English teaching quality assessment method based on SOFM neural network

technical field

The invention relates to the field of English teaching, in particular to an English teaching quality evaluation method based on SOFM neural network.

Background technique

English teaching refers to the process of teaching English to people whose first language is English or not. English teaching involves a variety of professional theoretical knowledge, including linguistics, second language acquisition, vocabulary, syntax, stylistics, corpus theory, cognitive psychology and so on. English teaching is a step-by-step process, whether for those who speak English as their first language or not, English learning is crucial in today's fast-developing globalization.

Many existing English teaching quality assessments are reflected through scoring, which is complex and not intuitive enough. The invention provides an English teaching quality evaluation method based on SOFM neural network, which further optimizes the learning performance of the neural network on the basis of solving the above problems.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is the technical problem of complex output results existing in the prior art. A new English teaching quality assessment method based on SOFM neural network is provided, and the English teaching quality assessment method based on SOFM neural network has the characteristics of intuitive results.

In order to solve the above technical problems, the technical solutions adopted are as follows:

An English teaching quality assessment method based on SOFM neural network, including

Step 1, construct a SOFM neural network model connected in series by two-level SOFM neural sub-network, the first-level SOFM neural sub-network consists of γ1 SOFM neural network units, and the second-level SOFM neural sub-network consists of γ2 SOFM neural networks in parallel. Unit composition, γ1 < γ2;

Step 2: Define the number c of English teaching quality evaluation grades, define the threshold R for evaluating neuron merging or splitting, input the English teaching quality evaluation score, and train the SOFM learning algorithm for the first-level SOFM neural sub-network, and calculate the various initial cluster center;

Step 3: Determine whether there is one neuron corresponding to a sample in the competition layer that is lower than the threshold of the number of samples, and if so, delete the corresponding competition layer neuron;

Step 4: Determine whether there is one neuron corresponding to more than two rating levels in the neurons of the competition layer. If so, call the second-level SOFM neural sub-network to train the SOFM learning algorithm and output various cluster center values as the final cluster. The center value, otherwise, the initial cluster center value of each type in step 2 is taken as the final cluster center value w _ij ;

Step 5: According to the final cluster center value, the English teaching object data to be evaluated is divided into c evaluation levels and output.

In the above scheme, for optimization, further, step 2 includes

Step (1), calculate the inter-class distances D _j =||m _j -m _j+1 ||, j=1, 2, 3, ...... c-1;

Step (2), compare the inter-class distance D _j and the threshold R, if D _j < R, the number of samples corresponding to neuron j is lower than the threshold of the number of samples, where m _j is the cluster center value of the jth class.

Further, step 3 includes:

Step (3), calculate the average distance within each class

Step (4), compare the average distance d _j and the threshold value R, if d _j > R, then neuron j corresponds to more than 2 rating levels, x _i is the English teaching quality evaluation score value corresponding to the input layer neuron i, n _j is the score within the class.

Further, step 4 also includes correcting the final cluster center value, and the correction amount is:

Δm _ij =m _ij (t+1)−m _ij (t)=β(t)(x _i (t)−m _ij (t));

为t时刻神经网络学习率，t为神经网络时间。Among them, m _ij (t+1) is the weight between the input layer i neuron and the mapping layer j neuron at time t+1, and m _ij (t) is the input layer i neuron and the mapping layer j neuron at time t weight between;

is the learning rate of the neural network at time t, and t is the neural network time.

Further, when the English teaching quality evaluation score is input in step 2, the English teaching quality evaluation score is preprocessed, and the preprocessing is coarse sampling, and normalization is performed after the preprocessing.

Beneficial effects of the present invention: by constructing a double-layer SOFM network model, the present invention adaptively adjusts the scale of SOFM according to the number of input samples and the number of output levels, and can ensure classification accuracy. At the same time, after the weights of the winning neuron and all neurons in its neighborhood are corrected, the learning rate will be adjusted, and the size of the neuron neighborhood that needs to be adjusted with the input layer will gradually change to improve the classification accuracy.

Description of drawings

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

Fig. 1 is a schematic diagram of the English teaching quality evaluation method in Example 1.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

Example 1

This embodiment provides an English teaching quality assessment method based on SOFM neural network, as shown in Figure 1, including:

Step 1. Build a SOFM neural network model connected in series by two-level SOFM neural sub-networks. The first-level SOFM neural sub-network consists of one SOFM neural network unit, and the second-level SOFM neural sub-network consists of two parallel SOFM neural networks. Unit composition; the number of SOFM neural network units here can be adjusted, the principle is that the number of the first layer is much smaller than the number of the second layer;

Step 2: Define the number c of English teaching quality evaluation grades, define the threshold R for evaluating neuron merging or splitting, input the English teaching quality evaluation score, and train the SOFM learning algorithm for the first-level SOFM neural sub-network, and calculate the various initial cluster center;

Step 3: Determine whether there is one neuron corresponding to a sample in the competition layer that is lower than the threshold of the number of samples, and if so, delete the corresponding competition layer neuron;

Step 4: Determine whether there is one neuron corresponding to more than two rating levels in the neurons of the competition layer. If so, call the second-level SOFM neural sub-network to train the SOFM learning algorithm and output various cluster center values as the final cluster. The center value, otherwise, the initial cluster center value of each type in step 2 is taken as the final cluster center value w _ij ;

Step 5: According to the final cluster center value, the English teaching object data to be evaluated is divided into c evaluation levels and output.

The SOFM learning algorithm in this embodiment is completed by using the existing training method.

Specifically, step 2 includes

Step (1), calculate the inter-class distances D _j =||m _j -m _j+1 ||, j=1, 2, 3, ... c-1;

Step (2), compare the inter-class distance D _j and the threshold R, if D _j < R, the number of samples corresponding to neuron j is lower than the threshold of the number of samples, where m _j is the cluster center value of the jth class.

Specifically, step 3 includes:

Step (3), calculate the average distance within each class

Step (4), compare the average distance d _j and the threshold value R, if d _j > R, then neuron j corresponds to more than 2 rating levels, x _i is the English teaching quality evaluation score value corresponding to the input layer neuron i, n _j is the score within the class.

Preferably, step 4 further includes correcting the final cluster center value, and the correction amount is:

Δm _ij =m _ij (t+1)−m _ij (t)=β(t)(x _i (t)−m _ij (t));

为t时刻神经网络学习率，t为神经网络时间。Among them, m _ij (t+1) is the weight between the input layer i neuron and the mapping layer j neuron at time t+1, and m _ij (t) is the input layer i neuron and the mapping layer j neuron at time t weight between;

is the learning rate of the neural network at time t, and t is the neural network time.

Preferably, when the English teaching quality evaluation scores are input in step 2, preprocessing is performed on the English teaching quality evaluation scores, and the preprocessing is coarse sampling, and normalization is performed after the preprocessing. Coarse processing of scores and normalization can reduce learning consumption.

Although the illustrative specific embodiments of the present invention are described above so that those skilled in the art can understand the present invention, the present invention is not limited to the scope of the specific embodiments. As long as such changes fall within the spirit and scope of the present invention as defined and determined by the appended claims, all inventions and creations utilizing the inventive concept are included in the protection list.

Claims (5)

1. an English teaching quality assessment method based on SOFM neural network, is characterized in that: the described English teaching quality assessment method based on SOFM neural network comprises: Step 1, construct a SOFM neural network model connected in series by two-level SOFM neural sub-network, the first-level SOFM neural sub-network consists of γ1 SOFM neural network units, and the second-level SOFM neural sub-network consists of γ2 SOFM neural networks in parallel. Unit composition, γ1 < γ2; Step 2: Define the number c of English teaching quality evaluation grades, define the threshold R for evaluating neuron merging or splitting, input the English teaching quality evaluation score, and train the SOFM learning algorithm for the first-level SOFM neural sub-network, and calculate the various initial cluster center; Step 3: Determine whether there is one neuron corresponding to a sample in the competition layer that is lower than the threshold of the number of samples, and if so, delete the corresponding competition layer neuron; Step 4: Determine whether there is one neuron corresponding to more than two rating levels in the neurons of the competition layer. If so, call the second-level SOFM neural sub-network to train the SOFM learning algorithm and output various cluster center values as the final cluster. The center value, otherwise, the initial cluster center value of each type in step 2 is taken as the final cluster center value w _ij ; Step 5: According to the final cluster center value, the English teaching object data to be evaluated is divided into c evaluation levels and output. 2. the English teaching quality assessment method based on SOFM neural network according to claim 1, is characterized in that: step 2 comprises Step (1), calculate the inter-class distances D _j =||m _j -m _j+1 ||, j=1, 2, 3, ...... c-1; Step (2), compare the inter-class distance D _j and the threshold R, if D _j < R, the number of samples corresponding to neuron i is lower than the threshold of the number of samples, where m _j is the cluster center value of the jth class. 3. the English teaching quality assessment method based on SOFM neural network according to claim 2, is characterized in that: step 3 comprises: Step (3), calculate the average distance within each class

Step (4), compare the average distance d _j and the threshold value R, if d _j > R, then neuron j corresponds to more than 2 rating levels, x _i is the English teaching quality evaluation score value corresponding to the input layer neuron i, n _j is the score within the class. 4. the English teaching quality assessment method based on SOFM neural network according to claim 1, is characterized in that: step 4 also comprises that final cluster center value is corrected, and correction amount is: Δm _ij =m _ij (t+1)−m _ij (t)=β(t)(x _i (t)−m _ij (t)); Among them, m _ij (t+1) is the weight between the input layer i neuron and the mapping layer j neuron at time t+1, and m _ij (t) is the input layer i neuron and the mapping layer j neuron at time t weight between;

is the learning rate of the neural network at time t, and t is the neural network time. 5. the English teaching quality assessment method based on SOFM neural network according to claim 1, is characterized in that: when step 2 inputs the English teaching quality assessment score, carries out preprocessing to the English teaching quality assessment score, and preprocessing is rough sampling, Normalization is performed after preprocessing.

Images