Association Set Extraction Method Based on Element Description Template for Airborne Field

Abstract

The present invention provides a method for extracting an associated set based on an element description template for the airborne field. First, entity recognition is performed on the requirement statement based on a conditional random field, and entities including operation objects, operation attributes, conditional objects, and conditional attributes are extracted. Then, the entities are replaced with neutral characters, and the isolated Be verbs in the requirement statement are replaced with "is equal to" to obtain a replacement statement, recording the replaced entities and their positions. Next, syntactic analysis techniques are used to perform element recognition on the requirement main clause and conditional clause obtained by splitting the replacement statement respectively, so as to recognize the requirement elements and their requirement relationships, and the requirement elements are restored to entities through NP phrase restoration. Finally, the entities and the corresponding requirement relationships are stored as an associated set in the data set. Therefore, the accuracy of element extraction from requirement statements in the airborne field under unsupervised conditions is improved, and a data set is provided for realizing the mechanization of extracted elements in the airborne field.

Description

Technical Field

[0001] The present invention relates to an extraction method based on a requirement element description model for the airborne field, which is applied to the field of semantic expression of objective information and belongs to a technology for describing requirement elements in the airborne field. Background Art

[0002] In the airworthiness certification activities of international mainstream airborne system manufacturers and the most representative domestic airborne system manufacturers, for the natural language requirement descriptions in the civil airborne field, the review work of requirements and design data and the specific forms of natural language requirement descriptions of airborne systems are often analyzed by combining the current domestic and foreign models under review, so as to clarify the general characteristics of natural language requirement descriptions in the civil airborne field.

[0003] It is found through analysis that compared with the natural language used in human daily communication, the grammatical structures and sentence patterns adopted in the requirement description methods in the airborne field have significantly fewer ranges and changes. However, there are still defects in analyzing the changes of grammatical sentence patterns and the combination forms of grammatical structures in the natural language requirements of airborne systems, where some complex problems cannot be solved by equipment when analyzing natural language requirements, and manual understanding is still required when analyzing natural language requirements, which increases the labor cost and is not conducive to mechanization. Summary of the Invention

[0004] To solve the above problems, an extraction method for an associated set based on an element description template for the airborne field is provided, and the present invention adopts the following technical solutions.

[0005] The present invention provides an extraction method for an associated set based on an element description template for the airborne field, which is characterized in that it includes: Step 1, entity recognition is performed on requirement statements based on a conditional random field, and entities including operation objects, operation attributes, conditional objects, and conditional attributes are extracted; Step 2, the entities are replaced with neutral characters, and the isolated Be verbs in the requirement statements are replaced with is equal to, so as to obtain a replacement statement and record the replaced entities and the positions of the entities; Step 3, the element description template is combined with the entities and the positions of the entities to split the replacement statement into a requirement main clause and a conditional clause; Step 4, element recognition is respectively performed on the requirement main clause and the conditional clause based on grammatical analysis technology to respectively identify multiple requirement elements in the requirement main clause and the conditional clause and the requirement relationships between the requirement elements; Step 5, the requirement elements are restored to entities through NP phrase restoration; Step 6, the entities and the corresponding requirement relationships are stored as an associated set in the data set.

[0006] Functions and Effects of the Invention

[0007] A method for extracting an association set based on an element description template for the airborne field according to the present invention. First, entity recognition is performed on the requirement statement based on a conditional random field, and entities including operation objects, operation attributes, conditional objects, and conditional attributes are extracted. Then, the entities are replaced with neutral characters, and isolated Be verbs in the requirement statement are replaced with "is equal to", so as to obtain a replacement statement and record the replaced entities and their positions. The element description template is combined with the entity and its position to split the replacement statement into a requirement main clause and a conditional clause. Then, element recognition is respectively performed on the requirement main clause and the conditional clause based on grammar analysis technology to respectively recognize multiple requirement elements in the requirement main clause and the conditional clause and the requirement relationships between the requirement elements. Finally, the requirement elements are restored to entities through NP phrase restoration, and the entities and the corresponding requirement relationships are stored as an association set in the data set.

[0008] Therefore, the method for extracting an association set based on an element description template for the airborne field provided by the present invention takes into account the influence of special part-of-speech words in professional term phrases in the civil airborne field on grammar in the requirement statement. By means of the method of "neutralizing" replacement of the recognized "object" and "attribute" type elements, conditional clause pruning, and Be verb replacement, the analysis of the requirement statement is prevented from affecting the model. On the basis of the "neutralizing" replacement of the entire element in the professional terms of the civil airborne field with a meaningless string, the use of NLP grammar analysis technology to analyze the professional vocabulary and the relationships between elements in the professional vocabulary that appear in the civil airborne field can better realize the extraction of natural language requirement elements. The method for extracting an association set based on a requirement element description model for the airborne field provided by the present invention has a significant effect on the analysis of requirement statement elements in the airborne field, improves the accuracy of element extraction in an unsupervised situation, has a good effect on realizing the automation of element extraction in the airborne field, and the data set obtained by extraction through the present invention provides a data set for the airborne professional field that can be used for future element extraction in the airborne field. Brief Description of the Drawings

[0009] Figure 1 is a flowchart of the method for extracting an association set based on an element description template for the airborne field in an embodiment of the present invention;

[0010] Figure 2 is a result graph of Experiment 1 of the method for extracting an association set based on an element description template for the airborne field in an embodiment of the present invention;

[0011] Figure 3 is a labeled result graph of the training set and the test set respectively in Experiment 2 of the method for extracting an association set based on an element description template for the airborne field in an embodiment of the present invention;

[0012] Figure 4 It is the extraction effect of requirement elements when training the conditional random field in Experiment 2 of the association set extraction method based on the element description template for the airborne field in the embodiments of the present invention;

[0013] Figure 5 It is the extraction result graph of requirement elements in Experiment 3 of the association set extraction method based on the element description template for the airborne field in the embodiments of the present invention when not applying conditional random field NER and only using syntactic analysis; and

[0014] Figure 6 It is the extraction result graph of syntactic analysis requirement elements in Experiment 3 of the association set extraction method based on the element description template for the airborne field in the embodiments of the present invention when using conditional random field NER. Detailed implementation manners

[0015] In order to make the technical means, creative features, achieved purposes and functions implemented by the present invention easy to understand, the following specifically describes an association set extraction method based on the element description template for the airborne field of the present invention in combination with embodiments and the accompanying drawings.

[0016] <Embodiment>

[0017] Figure 1 It is the flowchart of the association set extraction method based on the element description template for the airborne field in the embodiments of the present invention.

[0018] As Figure 1 shown, the process of the association set extraction method based on the element description template for the airborne field includes Step 1 to Step 6.

[0019] Step 1, perform entity recognition on the requirement statement based on the conditional random field and extract entities including operation objects, operation attributes, conditional objects, and conditional attributes.

[0020] In this embodiment, the requirement statement is the statement used in the airworthiness requirement standard document.

[0021] Step 2, replace the entity with a neutral character and replace the isolated Be verb in the requirement statement with is equalto, so as to obtain a replacement statement and record the replaced entity and the position of the entity.

[0022] Step 3, combine the element description template with the entity and the position of the entity to split the replacement statement into a requirement main clause and a conditional clause.

[0023] Step 4: Based on syntactic analysis technology, identify elements in the main clause of the requirement and the conditional clause respectively, so as to identify multiple requirement elements in the main clause of the requirement and the conditional clause and the requirement relationships between the respective requirement elements.

[0024] Step 5: Restore the requirement elements to entities through NP phrase restoration.

[0025] Step 6: Store the entities and the corresponding requirement relationships as an association set in the data set.

[0026] The element description template classifies requirement statements into requirement languages without conditional clauses and requirement languages with conditional clauses according to whether there are conditional clauses. Among them, the requirement language without conditional clauses includes at least an object and an operation, while the requirement language with conditional clauses includes at least a conditional keyword and a conditional attribute. Specifically, the description patterns and example templates of the requirement language without conditional clauses include Definitions 1 to 6.

[0027] Definition 1. Element relationship:

Object

Operation

[0028] Typical example: Each Messages Received In message shall be used for the generation of log data.

[0029] Definition 2. Element relationship:

Object

Operation

Attribute

[0030] Typical example: The state of the Aviation System Fault Discrete shall be read at a minimum every 150ms.

[0031] Definition 3. Element relationship:

Object

Operation

Qualification

[0032] Typical example: The audio output shall be muted by at least 42dB below the nominal audio output.;

[0033] Definition 4. Element relationship:

Object

Operation

Qualification

Attribute

[0034] Typical example: The main processing channel shall extract the 91Hz and 121Hz connection tones individually from the 'Prime Signal'.

[0035] The operations in the above templates are generally verbs. However, in practice, there is a special type of verb, namely the Be verb. For the cases of using the Be verb, we have also summarized two specific templates for reference.

[0036] Definition 5. Be as a verb operation in element relationships

[0037] Typical example: The package Sequence Number domain shall be the package Sequence Number domain of the previous number incremented by one.

[0038] Definition 6. Be as a passive auxiliary in element relationships

[0039] Typical example: The state of the Aviation System Fault Discrete discrete shall be read at a minimum every 102ms.

[0040] Specifically, the demand language description patterns without conditional clauses and the examples of description templates include Definitions 7 to 16.

[0041] Definition 7. Element relationship

Condition keyword

Condition attribute

[0042] Typical example: With a -13dB radio signal modulated 10% at 1002Hz, the audio output signal ratio shall be 2.3dB or higher.

[0043] Definition 8. Element relationship

Condition keyword

Condition object

Condition determination

Condition attribute

[0044] Typical example: Signal Out shall be sent to the Signal Pre-Processing System receiver within four seconds of the reception of Aviation System Configuration when Fail / Safe Configuration Required is True.

[0045] Definition 9. Element relationship <

Condition keyword

Condition attribute

Combination

[0046] Typical example: After restarted and tested, the Test Signal Switch Test Failure Status shall be considered as "Failed".

[0047] Definition 10. Element relationship

Condition keyword

Condition object

Condition determination

[0048] Typical example: The change in counter domain shall be considered to be monotonic if the value decreases due to the counter roll-over.

[0049] Definition 11. Element relationship <

Condition keyword

Condition object

Condition determination

Condition attribute

Combination

[0050] Typical example: When Ethernet Interface Configuration Required is True and status of package#1 response is True, then the package has been successfully sent and Ethernet Interface Configuration Required shall be reset to False.

[0051] Definition 12. Element relationship

Condition keyword

Condition object

Condition determination

Condition attribute

Condition limitation

[0052] Typical example: If the Error_Detecting_functionality works properly and passes two tests twice, then the Error_Detecting_Status shall indicate Available.

[0053] Definition 13. Element relationship <

Condition keyword

Condition object

Condition determination

Condition attribute

Condition limitation

Combination

[0054] Typical example: When A429_IN_Status has been received within 100 ms of the transmission of A429_OUT_Status and the data contents of A429_IN_Status agree with the data contents of A429_OUT_Status, then the status of package #1 response shall indicate True.

[0055] Definition 14. Alternative expression of the condition keyword in the element relationship at the end of

[0056] Typical example: At the end of the Switch Test, the Select Failure Status shall be considered as Not Failed when the conditions causing the failure are absent.

[0057] Definition 15. Alternative expression of the condition keyword in the element relationship where

[0058] Typical example: The memory location where the processor begins execution after an interrupt event shall contain the initialization components.

[0059] Definition 16. Multiple limitations in the main clause in the element relationship for..until..

[0060] Typical example: If Aviation System / Processing Platform Failed is True, the reset process shall be repeated every 30 seconds until Aviation System / Processing Platform Failed does not indicate True.

[0061] According to the above method, a more accurate airborne domain dataset is obtained by processing the input required language through a conditional random field and the method of replacing the Be verb.

[0062] Next, Experiment 1 of extracting requirement elements only based on the conditional random field model, Experiment 2 of combining the conditional random field and syntactic analysis to extract requirement elements, and Experiment 3 of using Be verb substitution and conditional random field recognition to assist syntactic analysis are respectively carried out for the association set extraction method based on the element description template for the airborne domain provided by the present invention. Through these experiments, it is verified that the present invention can improve the accuracy of extracting requirement statement elements in the airborne domain under unsupervised conditions.

[0063] In Experiment 1, the process of performing named entity recognition on all entities in the natural language requirement only based on the conditional random field is used as the conditional random field named entity model.

[0064] The requirement statements input in this embodiment are from the Definition requirement set. The requirement statements in the Definition requirement set are manually annotated according to the element description template, and the annotated requirement set is split into a training set and a test set. The training set is input into the Stanford NER tool to train the requirement conditional random field named entity model, and the trained conditional random field named entity model is used to identify elements in the requirement statements in the test set. The identified requirement elements are compared with the entities manually annotated in the test set to analyze their accuracy; the elements containing variable and attribute definition information in the Definition requirement set are used to replace the annotation information in the training set and the conditional random field named entity model is retrained. The retrained conditional random field named entity model is used to identify elements in the requirements in the test set, and the results of the second identification are compared with the entity elements manually annotated in the training set and their accuracy is analyzed.

[0065] Figure 2 It is a result diagram of Experiment 1 of the association set extraction method based on the element description template for the airborne domain in the embodiment of the present invention.

[0066] As Figure 2 shownFigure 2 Shows the precision, recall, and F1 values of named entity recognition based on conditional random fields.

[0067] Among them, P represents the precision of named entity recognition based on conditional random fields, R represents the recall of named entity recognition based on conditional random fields, TP is the number of correctly identified named entities, FP is the number of non-identified named entities, and FN is the number of wrongly identified named entities.

[0068] As Figure 2 shown, there are a total of 45 requirement statements in the manually annotated test set. By combining the method of conditional random fields and syntactic analysis to extract requirement elements, 42 elements are accurately identified, and 3 elements are not identified. In addition, there are 9 "conditional judgment" type elements that are not identified; similar to other element categories, among all element categories in these requirement statements, the average recognition accuracy of all element categories is 84.12%, the average recall rate is 56.00%, and the average F1 value is 67.24%.

[0069] In the analysis of the cases of misidentified elements in the experimental results, since the "conditional attributes" and "operation attributes", as well as the "conditional objects" and "operation objects", are relatively close in the context of the statement, and at the same time the training samples used are relatively small, the discrimination of some attributes and objects in the test samples is slightly lower than that of other elements. In the context of real engineering cases, most of the requirement elements can be basically extracted for elements of noun nature. The accuracy of individual element categories is average and cannot yet meet the needs of automated requirement inspection. It can be seen that the method used in Experiment 1, which only uses conditional random fields, cannot accurately extract all elements in natural language requirements. The conditional random field method should be combined with the syntactic analysis method, so Experiment 2 was carried out.

[0070] In Experiment 2, the process of combining conditional random fields and syntactic analysis to extract requirement elements is used as the conditional random field syntactic analysis named entity model.

[0071] Figure 3 It is the annotation result diagrams of the training set and the test set respectively in Experiment 2 of the method for extracting the associated set based on the element description template for the airborne field in the embodiments of the present invention.

[0072] As Figure 3As shown, the requirements sets 4, 5, and 6 are manually annotated by the same experts and combined with the 2b and 3b sets in Experiment 1 to form a test set. An experimental program implemented in the VS.Net environment using a named entity model for conditional random field grammar analysis is used to extract elements one by one from the requirements in the test sets 2b, 3b, 4, 5, and 6, and the elements identified by the algorithm are compared with the results manually annotated in the original 2b, 3b, 4, 5, and 6 sets. After determining the accuracy of element extraction by the pure grammar analysis method, the conditional random field NER model trained by the manually annotated training set in Experiment 1 is continued to be used to identify elements in the requirements sets 2b, 3b, 4, 5, and 6. The noun phrases of the "object" and "attribute" categories in the identified elements are preprocessed and replaced. According to the experimental program implemented in the VS.Net environment using a named entity model for conditional random field grammar analysis, the elements in the requirements sets 2b, 3b, 4, 5, and 6 are extracted, and the preprocessed and replaced noun phrase elements are restored to obtain the requirement element identification results. Finally, the results of this requirement element identification are compared with the manually annotated results to analyze the accuracy.

[0073] By using the information of the Definition data to replace part of the manual annotation and using the unified identification of the "object attribute" class elements, under the condition of the same test set, the effect of applying conditional random fields for named entity recognition is as Figure 4 shown.

[0074] Figure 4 This is the requirement element extraction effect when training the conditional random field in Experiment 2 of the association set extraction method based on the element description template for the airborne field in the embodiments of the present invention.

[0075] As Figure 4 shown, compared with the training results of the "object attribute" class entities in Experiment 1, the training results in this experiment show a slight decrease but can still identify most entities when using part of the Definition data to replace the annotation data for training. Therefore, using the data in the Definition requirements set for training as a method without manual supervision is still feasible. This makes it possible to apply this method to a wider range in the later stage. At the same time, since the information contained in the general requirement Definition requirements set is not sufficient to distinguish "objects" and "attributes" and is also difficult to distinguish the context environment, it brings inconvenience to the construction of the later element model. Therefore, a step of Be verb replacement is added based on Experiment 2 and Experiment 3 is carried out.

[0076] In Experiment 3, the process of using the Be verb substitution preprocessing method on the basis of the conditional random field grammar analysis named entity model is used as the named entity model for Be verb substitution conditional random field language analysis.

[0077] Preprocess and replace the noun phrases in the test requirement set based on Experiment 3; use the Be verb replacement preprocessing method to replace the Be verbs in the test requirement set; use the conditional random field language analysis algorithm program to perform syntactic analysis and element extraction on each requirement entry in Sets 2b, 3b, 4, 5, and 6, and restore the noun phrase replacement and Be verb phrase preprocessing; compare the results of the above requirement element recognition with the results of manual annotation to analyze the accuracy.

[0078] Figure 5 It is a diagram of the extraction result of requirement elements in Experiment 3 of the association set extraction method based on the element description template for the airborne field in the embodiment of the present invention when only syntactic analysis is used without applying conditional random field NER.

[0079] Figure 6 It is a diagram of the syntactic analysis requirement element extraction result in Experiment 3 of the association set extraction method based on the element description template for the airborne field in the embodiment of the present invention when using conditional random field NER.

[0080] Such as Figure 5 and Figure 6 As shown, when using syntactic analysis before and after using conditional random field NER for noun phrase replacement, the accuracy rate of "operation attributes" increased from 79.88% to 84.51%. There were also obvious improvements in other categories. The overall accuracy increased from 90.80% to 94.37%, and the recall rate increased from 85.21% to 89.05%. In addition, for these non-noun elements such as "conditional association", "conditional determination", and "operation", their accuracy generally increased by 2 - 7 percentage points.

[0081] According to Figure 5 and Figure 6 From the experimental results, it is not difficult to obtain that using the conditional random field recognition method to assist syntactic analysis not only improves the extraction accuracy of "object" and "attribute" type elements, but also, due to the use of the noun phrase replacement method, avoids the interference of "objects" and "attributes" with multiple parts of speech on syntactic analysis, simplifies the syntactic relationship of requirement descriptions, and also plays a very good role in improving the extraction of other elements.

[0082] From the above experiments, it can be seen that the method for extracting the associated set based on the element description template for the airborne field can extract requirement elements in actual engineering projects. At the same time, it combines the element recognition and preprocessing of the conditional random field NER method and methods such as Be verb replacement to eliminate some deficiencies that are difficult to overcome by simply relying on grammar analysis methods, further improving the effectiveness of requirement element extraction. The method provided in this embodiment extracts elements from requirement statements, avoiding deeper grammar and semantic analysis of the semantics of the entire natural language requirement entry, thereby realizing partial structuring of natural language requirements, and providing a basis for establishing requirement traceability at the element level and further requirement compliance detection based on element relationships.

[0083] Functions and effects of the embodiment

[0084] According to an associated set extraction method based on an element description template for the airborne field provided in this embodiment, first, entity recognition is performed on requirement statements based on the conditional random field, and entities including operation objects, operation attributes, conditional objects, and conditional attributes are extracted. Then, the entities are replaced with neutral characters, and the isolated Be verbs in the requirement statements are replaced with is equal to, so as to obtain a replacement statement and record the replaced entities and the positions of the entities. The element description template is combined with the entities and the positions of the entities to split the replacement statement into a requirement main clause and a conditional clause. Then, element recognition is respectively performed on the requirement main clause and the conditional clause based on grammar analysis techniques to respectively identify multiple requirement elements in the requirement main clause and the conditional clause and the requirement relationships between the respective requirement elements. Finally, the requirement elements are restored to entities through NP phrase restoration, and the entities and the corresponding requirement relationships are stored as an associated set in the data set.

[0085] Therefore, the association set extraction method based on the element description template for the airborne field provided by the present invention takes into account the influence of special part-of-speech words in professional term phrases in the civilian airborne field on grammar in the requirement statements. By means of the "neutralization" replacement method for the identified "object" and "attribute" class elements, conditional clause pruning, and Be verb substitution, the analysis of the requirement statements is prevented from affecting the model. On the basis of the "neutralization" replacement that uses meaningless strings to replace the entire element in the professional terms of the civilian airborne field, the use of NLP grammar analysis technology to analyze the professional vocabulary and the relationships between various elements in the professional vocabulary that appear in the civilian airborne field can better achieve the extraction of natural language requirement elements. The extraction method based on the requirement element description model for the airborne field provided by the present invention has a significant effect on the analysis of the requirement statement elements in the airborne field, improves the accuracy of element extraction in the unsupervised situation, has a good effect on realizing the automation of element extraction in the airborne field, and the data set obtained by the extraction of the present invention provides a data set for the corresponding airborne professional field that can be used for future element extraction in the airborne field.

[0086] The above embodiments are only used to illustrate the specific implementation manners of the present invention, and the present invention is not limited to the description scope of the above embodiments.

Claims

1. An extraction method for an association set based on an element description template for the airborne field, which is used to extract and store the association set of the input requirement statements, and is characterized in that Including: Step 1: Based on the conditional random field, perform entity recognition on the requirement statement and extract entities including operation objects, operation attributes, conditional objects, and conditional attributes; Step 2: Replace the entities with neutralized characters and replace the isolated Be verbs in the requirement statement with "is equal to", so as to obtain a replacement statement and record the replaced entities and their positions; Step 3: Combine the element description template with the entity and its position to split the replacement statement into a requirement main clause and a conditional clause; Step 4: Based on syntactic analysis technology, perform element recognition on the requirement main clause and the conditional clause respectively to identify multiple requirement elements in the requirement main clause and the conditional clause and the requirement relationships between the requirement elements; Step 5: Restore the requirement elements to the entities through NP phrase restoration; Step 6: Store the entities and the corresponding requirement relationships as an association set in the data set.

Images