Method and system for constructing civil aviation inspector dictionary cards based on deep learning

Abstract

The present invention discloses a method and system for constructing a civil aviation inspector dictionary card based on deep learning. A polyline classification network is constructed based on the mining degree value. A matrix detection network is constructed based on the word order. The current discriminant statement is obtained. Based on the current discriminant statement, the polyline classification network and the matrix detection network are used to associate the discriminant statement with the performance standard corresponding to the dictionary card to obtain the current performance standard. The influence of different key adjectives in each keyword-value pair on the performance standard is determined. Furthermore, the influence of word order on the performance standard is detected based on the word order.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and system for constructing a civil aviation inspector dictionary card based on deep learning. Background Art

[0002] The goal of constructing the dictionary cards is to further flesh out each OB. These cards present the definition of each OB in the competency model, along with the performance standards and scoring criteria (unqualified, qualified, and excellent) that civil aviation inspectors at each level should achieve under specific conditions. The research team developed the final dictionary cards based on interviews and analysis of civil aviation inspectors at each level, as well as discussions with senior civil aviation inspectors. For example, the OB item "Establishing a Comprehensive Incident Solution" within the problem-solving competency category is presented. This dictionary card presents the definition of the OB, along with the performance standards and scoring criteria achieved by civil aviation inspectors and senior civil aviation inspectors under the corresponding conditions. Quantitative evaluation criteria for each OB item are also determined. To ensure measurable and comparable competency among civil aviation inspectors, further quantitative evaluation criteria for each OB item are necessary.

[0003] However, the performance standards of civil aviation inspectors' behavior (OB) are usually determined manually, which wastes a lot of manpower and resources. It is necessary to find a method to determine performance standards based on big data. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for constructing a civil aviation inspector dictionary card based on deep learning, so as to solve the above-mentioned problems existing in the prior art.

[0005] In a first aspect, an embodiment of the present invention provides a method for constructing a civil aviation inspector dictionary card based on deep learning, comprising:

[0006] Obtaining m discriminant statements; wherein the discriminant statements represent statements that historically record characteristics of civil aviation inspectors of different levels;

[0007] According to m discriminant sentences, phrases with the same noun are detected to obtain n keyword-value pairs and m discriminant adjective phrase sets; the keyword-value pairs include monitoring keywords and corresponding keyword-adjectives; the keyword-adjectives represent adjectives used to judge the keywords; the discriminant adjective phrase set represents phrases that describe phrases with noun parts of speech in the discriminant sentences;

[0008] Based on n keyword key-value pairs and m discriminant adjective phrase sets, n mining degree values ​​are obtained; 1 monitoring keyword corresponds to 1 mining degree value;

[0009] Based on n keyword key-value pairs and corresponding mining degree values, a broken line classification network is constructed; the broken line classification network is used to associate the performance criteria corresponding to the discriminant sentence and the dictionary card using the mining degree value of the monitoring keyword;

[0010] Constructing a matrix detection network based on a set of m discriminant adjective phrases according to word order; the matrix detection network is used to associate the discriminant sentences with the performance criteria corresponding to the dictionary cards using the word order of the monitoring keywords;

[0011] Obtain the current discriminant statement; based on the current discriminant statement, associate the discriminant statement with the performance standard corresponding to the dictionary card through the broken line classification network and the matrix detection network to obtain the current performance standard.

[0012] Optionally, constructing a broken line classification network based on n keyword-value pairs and corresponding mining degree values ​​includes:

[0013] Obtaining a marked performance level value; m discriminant statements corresponding to m marked performance level values; the marked performance level value represents the marked performance level;

[0014] Arrange the monitoring keywords in the n keyword key-value pairs according to the mining degree value from small to large to obtain the arranged monitoring keyword vector;

[0015] Based on the arranged monitoring keyword vector, the annotated performance level value and the mining degree value, a keyword state coordinate is constructed; the keyword state coordinate is a two-dimensional coordinate;

[0016] Find the monitoring keyword and the corresponding performance level value of n keyword key-value pairs in the keyword state coordinate as the keyword position;

[0017] Mark the key adjective corresponding to the keyword value at the keyword position to obtain a monitoring keyword state diagram; the monitoring keyword state diagram has length as the horizontal axis, width as the vertical axis, and the lower left corner as the origin;

[0018] Based on the monitoring keyword state diagram, a plurality of keyword line graphs are obtained; the keyword line graphs represent the influence of key adjectives corresponding to different monitoring keywords on the corresponding performance levels;

[0019] Based on multiple keyword line graphs, a line classification network is constructed.

[0020] Optionally, constructing a broken line classification network based on n keyword-value pairs and corresponding mining degree values ​​includes:

[0021] Get a 2*w two-dimensional convolution kernel; 2 corresponds to the performance level value; w corresponds to the length of the keyword line graph;

[0022] With a step size of 1, a 2*w two-dimensional convolution kernel is convolved with the keyword line graph to obtain a predicted performance level value; the predicted performance level value represents the predicted performance level of the civil aviation inspector;

[0023] The predicted performance level value and the corresponding labeled performance level value are compared to obtain the loss and train the broken line classification network.

[0024] Optionally, constructing keyword state coordinates based on the arranged monitoring keyword vectors, the labeled performance level values, and the mining degree values ​​includes:

[0025] Arrange the monitoring keywords in the monitoring keyword vector as the subscripts of the horizontal axis, mark the performance level value as the vertical axis to construct the keyword state coordinate; the keyword state coordinate is a two-dimensional coordinate;

[0026] The distance between two adjacent monitoring keywords in the arranged monitoring keyword vector on the horizontal axis is used as the adjacent monitoring keyword distance;

[0027] The ratio of the mining degree values ​​corresponding to two adjacent monitoring keywords in the arranged monitoring keyword vector is used as the adjacent monitoring keyword mining ratio;

[0028] The adjacent monitoring keyword distance is equal to the adjacent monitoring keyword mining ratio multiplied by a fixed length value.

[0029] Optionally, the step of obtaining multiple keyword line graphs based on the monitoring keyword status graph includes:

[0030] In the monitoring keyword state diagram, a key adjective is extracted from the horizontal axis corresponding to a monitoring keyword to obtain multiple single-key adjective diagrams;

[0031] Connect the key adjectives corresponding to adjacent horizontal coordinates in the single key adjective graph, construct line segments, and obtain a keyword line graph.

[0032] Optionally, constructing a matrix detection network based on a set of m discriminant adjective phrases according to word order includes:

[0033] Arrange the keyword key-value pairs corresponding to a discriminant statement multiple times, and then obtain the discriminant statement keyword vector;

[0034] Constructing a matrix detection network; connecting two neurons in the input layer of the matrix detection network to one neuron in the adjacent hidden layer to extract features of the monitoring keyword and the corresponding key adjective in the keyword key-value pair; the neurons in the hidden layer of the matrix detection network are fully connected; and the neurons in the output layer of the matrix detection network are fully connected to the neurons in the adjacent hidden layer;

[0035] Input the discriminant sentence keyword vector into the matrix detection network, extract the features between the monitoring keywords and the corresponding key adjectives in the word order, and obtain the second predicted performance level value;

[0036] The second predicted performance level value is compared with the corresponding labeled performance level value to calculate the loss and train the matrix detection network.

[0037] Optionally, the method of obtaining n mining degree values ​​based on n keyword key-value pairs and m discriminant adjective phrase sets includes:

[0038] The keyword adjectives in the n keyword key-value pairs are formed into a set to obtain a similar keyword set;

[0039] Construct a non-similar keyword set from the values ​​other than the keyword adjectives in the m discriminant adjective phrase sets;

[0040] Detect the sentence symbol in the discriminant sentence, split the discriminant sentence, and obtain multiple segmented sentences;

[0041] The number of values ​​appearing in the similar keyword set in the segmented sentence is detected as a similarity correlation value; the similarity correlation values ​​corresponding to multiple values ​​in the similar keyword set appearing in the same segmented sentence are the same;

[0042] The number of values ​​appearing in the non-similar keyword set in the segmented sentence is detected as the non-similar correlation degree value; the non-similar correlation degree values ​​corresponding to multiple values ​​in the non-similar keyword set appearing in the same segmented sentence are the same;

[0043] The similar association degree value is multiplied by the similar keyword mining value, and the non-similar association degree value is multiplied by the non-similar keyword mining value to obtain the mining degree value.

[0044] Optionally, the detecting of phrases with the same noun based on m discriminant statements to obtain n keyword key-value pairs and m discriminant adjective phrase sets includes:

[0045] In the discrimination sentence, phrases are detected to obtain a discrimination noun phrase set and a corresponding discrimination adjective phrase set; the discrimination noun phrase set includes a plurality of discrimination noun phrases; the discrimination adjective phrase set includes a plurality of discrimination adjective phrases; the discrimination adjective phrases represent phrases whose part of speech is adjective; the discrimination noun phrases represent phrases whose part of speech is noun;

[0046] m discriminant sentences correspond to m discriminant noun phrase sets and m discriminant adjective phrase sets;

[0047] All discriminant noun phrases in multiple discriminant noun phrase sets are used as monitoring keywords;

[0048] The discriminant adjective phrase corresponding to the monitoring keyword is used as the keyword adjective.

[0049] Optionally, the step of associating the discriminant statement with the performance standard corresponding to the dictionary card through the broken line classification network and the matrix detection network based on the current discriminant statement to obtain the current performance standard includes:

[0050] Inputting the current discriminant statement into the broken line classification network to obtain a first current performance standard; the first current performance standard represents the performance standard corresponding to the discriminant statement and the dictionary card associated with the mining degree value of the monitoring keyword;

[0051] Inputting the current discriminant sentence into a matrix detection network to obtain a second current performance standard; the second current performance standard represents the performance standard corresponding to the discriminant sentence and the dictionary card associated with the word order of the monitoring keyword;

[0052] If the first current performance standard and the second current performance standard are the same, the first current performance standard or the second current performance standard shall be used as the current performance standard;

[0053] If the first current performance standard and the second current performance standard are different, the higher performance standard between the first current performance standard and the second current performance standard shall be used as the current performance standard.

[0054] In a second aspect, an embodiment of the present invention provides a system for constructing a civil aviation inspector dictionary card based on deep learning, comprising:

[0055] An acquisition module is used to obtain m discriminant statements; the discriminant statements represent statements that historically record the characteristics of civil aviation inspectors of different levels; and obtain a current discriminant statement;

[0056] A keyword module is configured to detect phrases with the same noun based on m discriminant sentences, and obtain n keyword-value pairs and m discriminant adjective phrase sets; the keyword-value pairs include monitoring keywords and corresponding keyword-adjectives; the keyword-adjectives represent adjectives used to judge the keywords; and the discriminant adjective phrase sets represent phrases that describe phrases with noun parts of speech in the discriminant sentences;

[0057] The mining module is used to obtain n mining degree values ​​based on n keyword key-value pairs and m discriminant adjective phrase sets; one monitoring keyword corresponds to one mining degree value;

[0058] A broken line module is used to construct a broken line classification network based on n keyword key-value pairs and corresponding mining degree values; the broken line classification network is used to associate the discriminant sentence with the performance standard corresponding to the dictionary card using the mining degree value of the monitoring keyword;

[0059] A word order module is used to construct a matrix detection network based on the m discriminant adjective phrase sets according to the word order; the matrix detection network is used to associate the discriminant sentences with the performance standards corresponding to the dictionary cards using the word order of the monitoring keywords;

[0060] The current module is used to associate the discriminant statement with the performance standard corresponding to the dictionary card based on the current discriminant statement through the broken line classification network and the matrix detection network to obtain the current performance standard.

[0061] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:

[0062] Embodiments of the present invention provide a method and system for constructing a civil aviation inspector dictionary card based on deep learning.

[0063] In the present invention, the monitoring keywords and the key adjectives describing the monitoring keywords in multiple discriminant sentences are found to constitute keyword-value pairs. The key adjectives can indicate the qualification of the monitoring keywords. A discriminant sentence includes multiple keyword-value pairs, and these keyword-value pairs will generate one performance standard. Different key adjectives result in different performance standards, but when marking, only the multiple keyword-value pairs that make up one discriminant sentence can be marked together. Therefore, a keyword line graph is constructed to find the influence of different key adjectives of each keyword-value pair on the judgment of the performance standard. And according to the word order, the influence of the word order on the judgment of the performance standard is detected. The technical effect of more accurately inputting the discriminant sentence and automatically detecting the performance standard is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 This is a flow chart of a method for constructing a civil aviation inspector dictionary card based on deep learning provided by an embodiment of the present invention.

[0065] Figure 2 It is a structural diagram of a broken line classification network and a matrix detection network in a method for constructing a civil aviation inspector dictionary card based on deep learning provided by an embodiment of the present invention.

[0066] Figure 3 This is a supervisory keyword status diagram in a method for constructing a civil aviation inspector dictionary card based on deep learning provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0067] The present invention will be described in detail below with reference to the accompanying drawings.

[0068] Example 1

[0069] like Figure 1 As shown, an embodiment of the present invention provides a method for constructing a civil aviation inspector dictionary card based on deep learning, the method comprising:

[0070] S101: Obtain m discriminant statements; the discriminant statements represent statements that historically record the characteristics of civil aviation inspectors of different levels.

[0071] S102: Based on m discriminant sentences, detect phrases with the same noun to obtain n keyword key-value pairs and m discriminant adjective phrase sets; the keyword key-value pairs include monitoring keywords and corresponding keyword adjectives; the keyword adjectives represent adjectives used to judge the monitoring keywords; the discriminant adjective phrase set represents phrases that describe phrases with noun parts of speech in the discriminant sentences.

[0072] Wherein, m and n are positive integers, and n is greater than or equal to m.

[0073] The number of keyword-value pairs is equal to the number of all discriminant noun phrases in the multiple discriminant sentences.

[0074] S103: Based on the n keyword key-value pairs and the m discriminant adjective phrase sets, n mining degree values ​​are obtained; one monitoring keyword corresponds to one mining degree value.

[0075] S104: constructing a broken line classification network based on n keyword key-value pairs and corresponding mining degree values; the broken line classification network is used to associate the discriminant sentences with the performance standards corresponding to the dictionary cards using the mining degree values ​​of the monitoring keywords.

[0076] S105: Constructing a matrix detection network based on the m discriminant adjective phrase sets according to the word order; the matrix detection network is used to associate the discriminant sentences with the performance criteria corresponding to the dictionary cards using the word order of the monitoring keywords.

[0077] The word order refers to the order of phrases recorded from early to late at different time points.

[0078] S106: Obtain the current discriminant statement; based on the current discriminant statement, associate the discriminant statement with the performance standard corresponding to the dictionary card through the broken line classification network and the matrix detection network to obtain the current performance standard.

[0079] Among them, after obtaining the current performance standards, it is possible to judge whether the current civil aviation inspector is qualified and assign points based on his or her level.

[0080] The structural diagram of the line classification network and the matrix detection network is as follows: Figure 2 shown.

[0081] The dictionary cards are shown in Table 1 below:

[0082] Table 1

[0083]

[0084] Optionally, constructing a broken line classification network based on n keyword-value pairs and corresponding mining degree values ​​includes:

[0085] Obtaining a marked performance level value; m discriminant statements corresponding to m marked performance level values; the marked performance level value represents the marked performance level.

[0086] Among them, as in this embodiment, the performance standard is "able to actively participate in various business trainings and self-study, master and update the technical knowledge and skills of this profession", and 6 performance levels are constructed based on this standard. The first performance level indicates "unable to actively participate in various business trainings and self-study, and master the technical knowledge and skills of this profession", the second performance level indicates "able to actively participate in various business trainings and self-study, and master the technical knowledge and skills of this profession", the third performance level indicates "able to actively participate in various business trainings and self-study, and master and update the technical knowledge and skills of this profession", the fourth performance level indicates "unable to actively participate in various business trainings and self-study, and master and update the technical knowledge and skills of this profession", the fifth performance level indicates "able to actively participate in various business trainings and self-study, master and update the technical knowledge and skills of this profession and integrate them", and the sixth performance level indicates "able to actively participate in various business trainings and self-study, master and update the technical knowledge and skills of this profession and actively apply them in combination with actual practice".

[0087] Among them, if the discriminant statement A is "understanding the common types of diseases of cement concrete and asphalt pavements", then the discriminant statement A corresponds to the second performance level. If the discriminant statement B is "familiar with the common types of diseases of cement concrete and asphalt pavements and the airworthiness determination standards", then the discriminant statement B corresponds to the third performance level. If the discriminant statement C is "familiar with the common types of diseases of cement concrete and asphalt pavements, the airworthiness determination standards and treatment methods", then the discriminant statement C corresponds to the fifth performance level. If the discriminant statement D is "familiar with the common types of diseases of cement concrete and asphalt pavements, the airworthiness determination standards and treatment methods. At the same time, it can analyze the causes of their occurrence and the deficiencies that need to be improved in pavement management, and guide the airport to optimize and improve the management mechanism. At the same time, it can publicize the latest advanced concepts and achievements of pavement management", then the discriminant statement C corresponds to the sixth performance level.

[0088] According to the mining degree value from small to large, the monitoring keywords in the n keyword key-value pairs are arranged to obtain the arranged monitoring keyword vector.

[0089] Based on the arranged monitoring keyword vectors, the labeled performance level values ​​and the mining degree values, keyword state coordinates are constructed; the keyword state coordinates are two-dimensional coordinates.

[0090] Find the positions of the monitoring keywords and corresponding annotated performance level values ​​of n keyword key-value pairs in the keyword state coordinates as the keyword positions.

[0091] Among them, the monitoring keyword of the keyword-value pair corresponds to the horizontal coordinate of the keyword status coordinate, and the marked performance level value corresponds to the vertical coordinate of the keyword status coordinate.

[0092] The key adjective corresponding to the keyword value is marked on the keyword position to obtain a monitoring keyword status diagram; the monitoring keyword status diagram uses length as the horizontal coordinate, width as the vertical coordinate, and the lower left corner as the origin.

[0093] In this embodiment, the schematic diagram of the monitoring keyword state diagram is as follows: Figure 3 shown.

[0094] The monitoring keyword status diagram is a grayscale image, and grayscale values ​​are used to represent key adjectives.

[0095] Based on the monitoring keyword status diagram, a plurality of keyword line graphs are obtained; the keyword line graphs represent the degree of influence of key adjectives corresponding to different monitoring keywords on the corresponding performance levels.

[0096] Based on multiple keyword line graphs, a line classification network is constructed.

[0097] Optionally, constructing a broken line classification network based on n keyword-value pairs and corresponding mining degree values ​​includes:

[0098] Obtain a 2*w two-dimensional convolution kernel; the 2 represents the labeling performance level value; and w represents a keyword.

[0099] Among them, 2 corresponds to the length of the keyword line graph. w corresponds to the length of the keyword line graph.

[0100] Wherein, w is a positive integer, and a 2*w two-dimensional convolution kernel is convolved with the keyword line graph with a step size of 1 to obtain a predicted performance level value; the predicted performance level value represents the predicted performance level of the civil aviation inspector.

[0101] In this embodiment, the broken line classification network is a convolutional neural network (CNN).

[0102] The predicted performance level value and the corresponding labeled performance level value are compared to obtain the loss and train the broken line classification network.

[0103] Among them, the cross entropy loss function is used to calculate the loss.

[0104] Among them, the performance level corresponding to the discriminant sentence that matches all the monitoring keywords and the corresponding key adjectives in the keyword line graph is found as the marked performance level value corresponding to the predicted performance level value.

[0105] Optionally, constructing keyword state coordinates based on the arranged monitoring keyword vectors, the labeled performance level values, and the mining degree values ​​includes:

[0106] Arrange the monitoring keywords in the monitoring keyword vector as the subscripts of the horizontal axis, mark the performance level value as the vertical axis to construct the keyword state coordinate; the keyword state coordinate is a two-dimensional coordinate;

[0107] The distance between two adjacent monitoring keywords in the arranged monitoring keyword vector on the horizontal axis is used as the adjacent monitoring keyword distance;

[0108] The ratio of the mining degree values ​​corresponding to two adjacent monitoring keywords in the arranged monitoring keyword vector is used as the adjacent monitoring keyword mining ratio.

[0109] The adjacent monitoring keyword distance is equal to the adjacent monitoring keyword mining ratio multiplied by a fixed length value.

[0110] In this embodiment, the fixed length value determines the distance between adjacent monitoring keywords. In this embodiment, the fixed length value is 10, and when performing a multiplication operation, it is rounded up.

[0111] Optionally, the step of obtaining multiple keyword line graphs based on the monitoring keyword status graph includes:

[0112] In the monitoring keyword status diagram, a key adjective is extracted from the horizontal axis corresponding to a monitoring keyword to obtain multiple single-key adjective diagrams.

[0113] The number of the single-key adjective graphs depends on the number of adjective keywords corresponding to each monitoring keyword. In this embodiment, the number of the single-key adjective graphs is m to the power of n.

[0114] The multiple keyword line graphs obtained in this step are used to train the line classification network. When testing, there is only one single-keyword adjective graph.

[0115] Connect the key adjectives corresponding to adjacent horizontal coordinates in the single key adjective graph, construct line segments, and obtain a keyword line graph.

[0116] Among them, the arrangement order of the monitoring keywords in the monitoring keyword status diagram and the corresponding position of the marked performance level values ​​are used to construct multiple line segments with different positions and slopes to determine the correlation between the keywords and the corresponding changes in performance levels.

[0117] Optionally, constructing a matrix detection network based on a set of m discriminant adjective phrases according to word order includes:

[0118] Arrange the keyword key-value pairs corresponding to a discriminant statement multiple times, and then obtain the discriminant statement keyword vector;

[0119] A matrix detection network is constructed; two neurons in the input layer of the matrix detection network are connected to one neuron in the adjacent hidden layer to extract features of monitoring keywords and corresponding key adjectives in keyword key-value pairs; the neurons in the hidden layer of the matrix detection network are fully connected; and the neurons in the output layer of the matrix detection network are fully connected to the neurons in the adjacent hidden layer.

[0120] Wherein, the matrix detection network is a deep neural network (DNN).

[0121] Input the discriminant sentence keyword vector into the matrix detection network, extract the features between the monitoring keywords and the corresponding key adjectives in the word order, and obtain the second predicted performance level value;

[0122] The second predicted performance level value is compared with the corresponding labeled performance level value to calculate the loss and train the matrix detection network.

[0123] In this embodiment, a cross entropy loss function is used to calculate the loss and backward train the matrix detection network.

[0124] In this embodiment, the labeled performance level value corresponding to the second predicted performance level value indicates that the discriminant sentence keyword vector corresponds to the labeled performance level value corresponding to the discriminant sentence.

[0125] Optionally, the method of obtaining n mining degree values ​​based on n keyword key-value pairs and m discriminant adjective phrase sets includes:

[0126] The keyword adjectives in the n keyword key-value pairs are formed into a set to obtain a similar keyword set;

[0127] The values ​​other than the keyword adjectives in the m discriminant adjective phrase sets are used to construct a non-similar keyword set.

[0128] Among them, if it is a value in the similar keyword set, it means that all civil aviation inspectors need to use it to judge the performance level, and if it is a value in the non-similar keyword set, it means that it is a supplement for some civil aviation inspectors to judge the performance level.

[0129] Detect the sentence symbols in the discriminant sentence, split the discriminant sentence, and obtain multiple segmented sentences.

[0130] In this embodiment, the sentence symbol is a period.

[0131] The number of values ​​appearing in the similar keyword set in the segmented sentence is detected as a similarity correlation value; the similarity correlation values ​​corresponding to multiple values ​​in the similar keyword set appearing in the same segmented sentence are the same;

[0132] The number of values ​​appearing in the non-similar keyword set in the segmented sentence is detected as the non-similar association degree value; the non-similar association degree values ​​corresponding to multiple values ​​in the non-similar keyword set appearing in the same segmented sentence are the same.

[0133] Among them, the number of keywords in the segmented sentence is detected. If there are more keywords, it means that the multiple keywords in the segmented sentence are more correlated, and the mining degree is higher.

[0134] The similar association degree value is multiplied by the similar keyword mining value, and the non-similar association degree value is multiplied by the non-similar keyword mining value to obtain the mining degree value.

[0135] Among them, in this embodiment, the similar keyword mining value is 1, and the non-similar keyword mining value is 0.5, which means that the keywords shared by all civil aviation inspectors are mainly used to judge the performance level, while the keywords shared by some civil aviation inspectors are used for supplementary judgment of the performance level.

[0136] Optionally, the detecting of phrases with the same noun based on m discriminant statements to obtain n keyword key-value pairs and m discriminant adjective phrase sets includes:

[0137] In the discrimination sentence, phrases are detected to obtain a discrimination noun phrase set and a corresponding discrimination adjective phrase set; the discrimination noun phrase set includes multiple discrimination noun phrases; the discrimination adjective phrase set includes multiple discrimination adjective phrases; the discrimination adjective phrase represents a phrase whose part of speech is adjective; the discrimination noun phrase represents a phrase whose part of speech is noun.

[0138] Wherein, multiple discriminant noun phrases in the discriminant noun phrase set are identical. If a discriminant sentence has the same noun phrase in the detection, first check whether there is an adjective phrase describing the noun phrase at the front end or back end, and retain the phrase containing the adjective. If there are multiple identical noun phrases, the similarity of the corresponding adjective phrases is tested. If similar, the first noun phrase is retained as the discriminant noun phrase, and the adjective phrase corresponding to the discriminant noun phrase is used as the discriminant adjective phrase.

[0139] In this embodiment, numbers are used to represent the distinguishing noun phrases and the distinguishing adjective phrases. Similar phrases are marked with numbers. In this embodiment, the numbers are 5 digits.

[0140] m discriminant sentences correspond to m discriminant noun phrase sets and m discriminant adjective phrase sets;

[0141] All discriminant noun phrases in the set of m discriminant noun phrases are used as monitoring keywords;

[0142] The discriminant adjective phrase corresponding to the monitoring keyword is used as the keyword adjective.

[0143] Optionally, the step of associating the discriminant statement with the performance standard corresponding to the dictionary card through the broken line classification network and the matrix detection network based on the current discriminant statement to obtain the current performance standard includes:

[0144] Inputting the current discriminant statement into the broken line classification network to obtain a first current performance standard; the first current performance standard represents the performance standard corresponding to the discriminant statement and the dictionary card associated with the mining degree value of the monitoring keyword;

[0145] Inputting the current discriminant sentence into a matrix detection network to obtain a second current performance standard; the second current performance standard represents the performance standard corresponding to the discriminant sentence and the dictionary card associated with the word order of the monitoring keyword;

[0146] If the first current performance standard and the second current performance standard are the same, the first current performance standard or the second current performance standard shall be used as the current performance standard;

[0147] If the first current performance standard and the second current performance standard are different, the higher performance standard between the first current performance standard and the second current performance standard shall be used as the current performance standard.

[0148] In this embodiment, the performance standard for the fourth performance level is higher than that for the first performance level. The performance standard for the second performance level is higher than that for the fourth performance level. The performance standard for the third performance level is higher than that for the second performance level. The performance standard for the fifth performance level is higher than that for the third performance level. The performance standard for the sixth performance level is higher than that for the fifth performance level.

[0149] Example 2

[0150] Based on the above-mentioned method for constructing civil aviation inspector dictionary cards based on deep learning, an embodiment of the present invention also provides a system for constructing civil aviation inspector dictionary cards based on deep learning, wherein the system includes an acquisition module, a keyword module, a mining module, a broken line module, a word order module and a current module.

[0151] An acquisition module is used to obtain m discriminant statements; the discriminant statements represent statements that historically record the characteristics of civil aviation inspectors of different levels; and obtain a current discriminant statement;

[0152] A keyword module is configured to detect phrases with the same noun based on m discriminant sentences, and obtain n keyword-value pairs and m discriminant adjective phrase sets; the keyword-value pairs include monitoring keywords and corresponding keyword-adjectives; the keyword-adjectives represent adjectives used to judge the keywords; and the discriminant adjective phrase sets represent phrases that describe phrases with noun parts of speech in the discriminant sentences;

[0153] The mining module is used to obtain n mining degree values ​​based on n keyword key-value pairs and m discriminant adjective phrase sets; one monitoring keyword corresponds to one mining degree value;

[0154] A broken line module is used to construct a broken line classification network based on n keyword key-value pairs and corresponding mining degree values; the broken line classification network is used to associate the discriminant sentence with the performance standard corresponding to the dictionary card using the mining degree value of the monitoring keyword;

[0155] A word order module is used to construct a matrix detection network based on the m discriminant adjective phrase sets according to the word order; the matrix detection network is used to associate the discriminant sentences with the performance standards corresponding to the dictionary cards using the word order of the monitoring keywords;

[0156] The current module is used to associate the discriminant statement with the performance standard corresponding to the dictionary card based on the current discriminant statement through the broken line classification network and the matrix detection network to obtain the current performance standard.

[0157] The algorithm and display provided herein are not inherently related to any particular computer, virtual system or other device. Various general-purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing this type of system. In addition, the present invention is not directed to any specific programming language. It should be understood that various programming languages ​​can be utilized to realize the content of the present invention described herein, and the above description of specific languages ​​is for the purpose of disclosing the best mode of the present invention.

[0158] In the description provided herein, numerous specific details are described. However, it is understood that embodiments of the present invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques are not shown in detail so as not to obscure the understanding of this description.

[0159] The various component embodiments of the present invention may be implemented in hardware, as software modules running on one or more processors, or as a combination thereof. Those skilled in the art will appreciate that, in practice, a microprocessor or digital signal processor (DSP) may be used to implement some or all of the functionality of some or all of the components of the apparatus according to the embodiments of the present invention. The present invention may also be implemented as an apparatus or device program (e.g., a computer program or computer program product) for performing part or all of the methods described herein. Such a program implementing the present invention may be stored on a computer-readable medium or in the form of one or more signals. Such signals may be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

Claims

1. A method for constructing a civil aviation inspector dictionary card based on deep learning, characterized in that: include: Obtaining m discriminant statements; wherein the discriminant statements represent statements that historically record characteristics of civil aviation inspectors of different levels; According to m discriminant sentences, phrases with the same noun are detected to obtain n keyword-value pairs and m discriminant adjective phrase sets; the keyword-value pairs include monitoring keywords and corresponding keyword-adjectives; the keyword-adjectives represent adjectives used to judge the keywords; the discriminant adjective phrase set represents phrases that describe phrases with noun parts of speech in the discriminant sentences; Based on n keyword key-value pairs and m discriminant adjective phrase sets, n mining degree values ​​are obtained; 1 monitoring keyword corresponds to 1 mining degree value; Based on n keyword key-value pairs and corresponding mining degree values, a broken line classification network is constructed; the broken line classification network is used to associate the performance criteria corresponding to the discriminant sentence and the dictionary card using the mining degree value of the monitoring keyword; Constructing a matrix detection network based on a set of m discriminant adjective phrases according to word order; the matrix detection network is used to associate the discriminant sentences with the performance criteria corresponding to the dictionary cards using the word order of the monitoring keywords; Obtain the current discriminant statement; based on the current discriminant statement, associate the discriminant statement with the performance standard corresponding to the dictionary card through the broken line classification network and the matrix detection network to obtain the current performance standard; Based on the n keyword key-value pairs and the m discriminant adjective phrase sets, n mining degree values ​​are obtained, including: The keyword adjectives in the n keyword key-value pairs are formed into a set to obtain a similar keyword set; Construct a non-similar keyword set from the values ​​other than the keyword adjectives in the m discriminant adjective phrase sets; Detect the sentence symbol in the discriminant sentence, split the discriminant sentence, and obtain multiple segmented sentences; The number of values ​​appearing in the similar keyword set in the segmented sentence is detected as a similarity correlation value; the similarity correlation values ​​corresponding to multiple values ​​in the similar keyword set appearing in the same segmented sentence are the same; The number of values ​​appearing in the non-similar keyword set in the segmented sentence is detected as the non-similar correlation degree value; the non-similar correlation degree values ​​corresponding to multiple values ​​in the non-similar keyword set appearing in the same segmented sentence are the same; Multiplying the similarity association degree value by the similar keyword mining value, and multiplying the non-similarity association degree value by the non-similar keyword mining value to obtain the mining degree value; The broken line classification network is a convolutional neural network; the matrix detection network is a deep neural network.

2. The method for constructing a civil aviation inspector dictionary card based on deep learning according to claim 1, characterized in that: The method of constructing a polyline classification network based on n keyword key-value pairs and corresponding mining degree values ​​includes: Obtaining a marked performance level value; m discriminant statements corresponding to m marked performance level values; the marked performance level value represents the marked performance level; Arrange the monitoring keywords in the n keyword key-value pairs according to the mining degree value from small to large to obtain the arranged monitoring keyword vector; Based on the arranged monitoring keyword vector, the annotated performance level value and the mining degree value, a keyword state coordinate is constructed; the keyword state coordinate is a two-dimensional coordinate; Find the monitoring keyword and the corresponding performance level value of n keyword key-value pairs in the keyword state coordinate as the keyword position; Mark the key adjective corresponding to the keyword value at the keyword position to obtain a monitoring keyword state diagram; the monitoring keyword state diagram has length as the horizontal axis, width as the vertical axis, and the lower left corner as the origin; Based on the monitoring keyword state diagram, a plurality of keyword line graphs are obtained; the keyword line graphs represent the influence of key adjectives corresponding to different monitoring keywords on the corresponding performance levels; Based on multiple keyword line graphs, a line classification network is constructed.

3. The method for constructing a civil aviation inspector dictionary card based on deep learning according to claim 2, characterized in that: The method of constructing a polyline classification network based on n keyword key-value pairs and corresponding mining degree values ​​includes: Get a 2*w two-dimensional convolution kernel; 2 corresponds to the performance level value; w corresponds to the length of the keyword line graph; With a step size of 1, a 2*w two-dimensional convolution kernel is convolved with the keyword line graph to obtain a predicted performance level value; the predicted performance level value represents the predicted performance level of the civil aviation inspector; The predicted performance level value and the corresponding labeled performance level value are compared to obtain the loss and train the broken line classification network.

4. The method for constructing a civil aviation inspector dictionary card based on deep learning according to claim 2, characterized in that: The step of constructing keyword state coordinates based on the arranged monitoring keyword vectors, the labeled performance level values, and the mining degree values ​​includes: Arrange the monitoring keywords in the monitoring keyword vector as the subscripts of the horizontal axis, mark the performance level value as the vertical axis to construct the keyword state coordinate; the keyword state coordinate is a two-dimensional coordinate; The distance between two adjacent monitoring keywords in the arranged monitoring keyword vector on the horizontal axis is used as the adjacent monitoring keyword distance; The ratio of the mining degree values ​​corresponding to two adjacent monitoring keywords in the arranged monitoring keyword vector is used as the adjacent monitoring keyword mining ratio; The adjacent monitoring keyword distance is equal to the adjacent monitoring keyword mining ratio multiplied by a fixed length value.

5. The method for constructing a civil aviation inspector dictionary card based on deep learning according to claim 2, characterized in that: Based on the monitoring keyword state diagram, a plurality of keyword line graphs are obtained, including: In the monitoring keyword state diagram, a key adjective is extracted from the horizontal axis corresponding to a monitoring keyword to obtain multiple single-key adjective diagrams; Connect the key adjectives corresponding to adjacent horizontal coordinates in the single key adjective graph, construct line segments, and obtain a keyword line graph.

6. The method for constructing a civil aviation inspector dictionary card based on deep learning according to claim 1, characterized in that: The matrix detection network is constructed based on a set of m discriminative adjective phrases according to word order, including: Arrange the keyword key-value pairs corresponding to a discriminant statement multiple times, and then obtain the discriminant statement keyword vector; Constructing a matrix detection network; connecting two neurons in the input layer of the matrix detection network to one neuron in the adjacent hidden layer to extract features of the monitoring keyword and the corresponding key adjective in the keyword key-value pair; the neurons in the hidden layer of the matrix detection network are fully connected; and the neurons in the output layer of the matrix detection network are fully connected to the neurons in the adjacent hidden layer; Input the discriminant sentence keyword vector into the matrix detection network, extract the features between the monitoring keywords and the corresponding key adjectives in the word order, and obtain the second predicted performance level value; The second predicted performance level value is compared with the corresponding labeled performance level value to calculate the loss and train the matrix detection network.

7. The method for constructing a civil aviation inspector dictionary card based on deep learning according to claim 1, characterized in that: The method detects phrases with the same noun based on m discriminant sentences, and obtains n keyword key-value pairs and m discriminant adjective phrase sets, including: In the discrimination sentence, phrases are detected to obtain a discrimination noun phrase set and a corresponding discrimination adjective phrase set; the discrimination noun phrase set includes a plurality of discrimination noun phrases; the discrimination adjective phrase set includes a plurality of discrimination adjective phrases; the discrimination adjective phrases represent phrases whose part of speech is adjective; the discrimination noun phrases represent phrases whose part of speech is noun; m discriminant sentences correspond to m discriminant noun phrase sets and m discriminant adjective phrase sets; All discriminant noun phrases in multiple discriminant noun phrase sets are used as monitoring keywords; The discriminant adjective phrase corresponding to the monitoring keyword is used as the keyword adjective.

8. The method for constructing a civil aviation inspector dictionary card based on deep learning according to claim 1, characterized in that: The current performance standard is obtained by associating the discriminant statement with the performance standard corresponding to the dictionary card through the broken line classification network and the matrix detection network based on the current discriminant statement, including: Inputting the current discriminant statement into the broken line classification network to obtain a first current performance standard; the first current performance standard represents the performance standard corresponding to the discriminant statement and the dictionary card associated with the mining degree value of the monitoring keyword; Inputting the current discriminant sentence into a matrix detection network to obtain a second current performance standard; the second current performance standard represents the performance standard corresponding to the discriminant sentence and the dictionary card associated with the word order of the monitoring keyword; If the first current performance standard and the second current performance standard are the same, the first current performance standard or the second current performance standard shall be used as the current performance standard; If the first current performance standard and the second current performance standard are different, the higher performance standard between the first current performance standard and the second current performance standard shall be used as the current performance standard.

9. A system for constructing civil aviation inspector dictionary cards based on deep learning, characterized in that: include: An acquisition module, configured to obtain m discriminant statements, wherein the discriminant statements represent statements that historically record the characteristics of civil aviation inspectors of different levels; Get the current judgment statement; A keyword module is configured to detect phrases with the same noun based on m discriminant sentences, and obtain n keyword-value pairs and m discriminant adjective phrase sets; the keyword-value pairs include monitoring keywords and corresponding keyword-adjectives; the keyword-adjectives represent adjectives used to judge the keywords; and the discriminant adjective phrase sets represent phrases that describe phrases with noun parts of speech in the discriminant sentences; The mining module is used to obtain n mining degree values ​​based on n keyword key-value pairs and m discriminant adjective phrase sets; one monitoring keyword corresponds to one mining degree value; A broken line module is used to construct a broken line classification network based on n keyword key-value pairs and corresponding mining degree values; the broken line classification network is used to associate the discriminant sentence with the performance standard corresponding to the dictionary card using the mining degree value of the monitoring keyword; A word order module is used to construct a matrix detection network based on the word order of m discriminant adjective phrases; the matrix detection network is used to associate the discriminant sentence with the performance index corresponding to the dictionary card using the word order of the monitoring keyword; The current module is used to associate the discriminant statement with the performance standard corresponding to the dictionary card through the broken line classification network and the matrix detection network based on the current discriminant statement to obtain the current performance standard; Based on the n keyword key-value pairs and the m discriminant adjective phrase sets, n mining degree values ​​are obtained, including: The keyword adjectives in the n keyword key-value pairs are formed into a set to obtain a similar keyword set; Construct a non-similar keyword set from the values ​​other than the keyword adjectives in the m discriminant adjective phrase sets; Detect the sentence symbol in the discriminant sentence, split the discriminant sentence, and obtain multiple segmented sentences; The number of values ​​appearing in the similar keyword set in the segmented sentence is detected as a similarity correlation value; the similarity correlation values ​​corresponding to multiple values ​​in the similar keyword set appearing in the same segmented sentence are the same; The number of values ​​appearing in the non-similar keyword set in the segmented sentence is detected as the non-similar correlation degree value; the non-similar correlation degree values ​​corresponding to multiple values ​​in the non-similar keyword set appearing in the same segmented sentence are the same; Multiplying the similarity association degree value by the similar keyword mining value, and multiplying the non-similarity association degree value by the non-similar keyword mining value to obtain the mining degree value; The broken line classification network is a convolutional neural network; the matrix detection network is a deep neural network.

Images