A design drawing fusion retrieval method considering label reliability and electronic equipment

Abstract

The application discloses a kind of design drawing fusion search methods and electronic equipment considering labeling reliability, including that label marking is carried out to drawing by label based on drawing attribute structure, constructs the mapping matrix of " drawing-attribute ", uses information entropy to depict the relative weight between attributes, collects the user feedback information received by each labeling, calculates the reliability distribution of labeler;Using the method of bayesian update, the reliability of labeling is calculated;Input search condition, based on the reliability of labeling and the relative weight between attributes, the similarity degree of each drawing and given search condition is calculated, and the search result, i.e.the drawing sequence sorted in descending order of similarity degree, is obtained.The method improves the retrieval efficiency of design drawing by considering the reliability of drawing labeling, and realizes the rapid identification of potential inaccurate labeling information in the case of few user feedback by analyzing the reliability of labeler.

Description

Technical Field

[0001] This invention belongs to the field of information technology and relates to a design drawing fusion retrieval method and electronic device that takes into account the reliability of annotations. Background Technology

[0002] In the personalized customization industry, the diverse and small-batch production of product parts creates a significant demand for non-standard part design drawings. To save human resource costs and enable cross-project reuse of non-standard part design drawings, tag-based similar drawing fusion retrieval is gaining importance. However, this retrieval method is affected by the reliability of drawing annotations; unreliable annotation information can lead to inefficiencies in drawing retrieval.

[0003] Therefore, a method is needed that can predict the reliability of annotation information on drawings and retrieve similar design drawings while taking into account annotation reliability, in order to solve the problem of inefficiency in drawing retrieval. Summary of the Invention

[0004] To address the aforementioned deficiencies in existing technologies, the present invention aims to provide a similar drawing fusion retrieval method and electronic device that considers annotation reliability. This method collects user feedback on drawing annotations, calculates the reliability distribution of the annotation personnel's attribute annotations, uses the personnel reliability distribution as a prior distribution, and takes the user feedback received as evidence. It employs a Bayesian update method to calculate the reliability of the annotation information. Considering both the reliability of the annotation information and attribute weights, it calculates the similarity between each drawing and the given search criteria, and then returns a sequence of drawings similar to the search criteria, thus obtaining the retrieval results.

[0005] The present invention is achieved through the following technical solution.

[0006] According to one aspect of the present invention, a design drawing fusion retrieval method considering annotation reliability is provided, comprising:

[0007] Annotators annotate the drawings based on their attribute structure;

[0008] Construct a mapping matrix of drawings and attributes for the annotated drawings;

[0009] Based on the drawing-attribute mapping matrix, information entropy is used to characterize the relative weights between drawing attributes;

[0010] Collect user feedback on each annotation in the drawing-attribute mapping matrix;

[0011] Based on the collected user feedback information, calculate the reliability distribution of the labelers;

[0012] Based on the collected user feedback information and the reliability distribution of the annotators, the reliability of the annotations is calculated using the Bayesian update method.

[0013] Enter search criteria that match the drawing's attribute structure;

[0014] Based on the reliability of the annotations and the relative weights between the drawing attributes, the similarity between each drawing and the given search criteria is calculated.

[0015] Based on the degree of similarity between each drawing and the given search criteria, the search results are obtained, which is a sequence of drawings sorted in descending order of similarity.

[0016] Preferably, the drawing attribute structure and the drawing itself include a set of drawing attributes and a set of drawings, respectively.

[0017] Preferably, constructing the drawing-attribute mapping matrix includes:

[0018] Based on the drawing set and the drawing attribute set, the drawing annotation results are transformed into a drawing-attribute matrix.

[0019] Preferably, user feedback information on each annotation in the drawing-attribute mapping matrix is ​​collected, including:

[0020] Feedback from users who view the annotations on drawings includes voting information and user view counts.

[0021] The system collects user voting information on tag labeling by using the "This label is unreasonable" voting button, and collects user pageview information based on user browsing behavior.

[0022] Preferably, based on the collected user feedback information, the reliability distribution of the annotators is calculated, including:

[0023] Based on the collected user feedback information, the reliability distribution of the annotators is calculated, including:

[0024] Calculate the annotation score based on the feedback received from the annotators regarding the attributes of the drawings;

[0025] Calculate the mean and standard deviation based on the annotation scores calculated by the annotators;

[0026] The reliability distribution of the labeled personnel was calculated based on the mean and standard deviation.

[0027] Preferably, the reliability of the annotation is calculated based on the collected feedback information and the reliability distribution of the annotators, including:

[0028] For annotations made by annotation personnel to the attributes of drawings, using the reliability distribution of the annotation personnel as prior information and the user feedback information received by the annotation as evidence, a Bayesian update method is used to obtain the posterior distribution of the reliability of the annotation:

[0029] The reliability of the label is measured using the mean of the posterior distribution to which the label reliability follows.

[0030] Preferably, based on the reliability of the annotations and the relative weights between drawing attributes, the similarity degree between each drawing and the given search criteria is calculated, including:

[0031] For a given search criteria, determine the similarity between the drawing and the annotations of the search criteria on the drawing attributes;

[0032] Calculate the similarity between all drawings and the given search criteria to obtain a similarity set.

[0033] Preferably, the search results are obtained based on the degree of similarity between each drawing and the given search criteria, that is, the sequence of drawings sorted in descending order of similarity.

[0034] In another aspect, the present invention provides an electronic device, including a processor and a memory;

[0035] The memory is used to store computer programs, the computer programs including program instructions;

[0036] The processor is used to call the program stored in the memory to execute the design drawing fusion retrieval method that takes into account the reliability of annotations.

[0037] This invention collects user feedback on drawing annotations, calculates the distribution of annotation reliability among annotators, and then calculates the reliability of the annotations. Combining the weights of each attribute calculated based on the annotation result matrix of "drawing-attribute", it calculates the similarity between each drawing and the given search conditions, and then provides feedback on drawings similar to the search conditions.

[0038] The present invention, by adopting the above technical solution, has the following beneficial effects:

[0039] 1. Improved efficiency in retrieving similar drawings. This invention considers the reliability of annotation information in the similarity calculation between drawings and search criteria, assigning high weight to highly reliable annotations and low weight to low-reliability annotations. This reduces the interference of unreliable annotation information on search results, thereby improving the efficiency of retrieving similar drawings.

[0040] 2. Improved identification of potentially inaccurate annotations in situations where user feedback is scarce. In calculating annotation reliability, this invention fully utilizes user feedback information received from other annotations by the same annotator, achieving joint annotation quality analysis centered on the annotator. Compared to methods that only use user feedback to assess current annotation reliability, this invention also considers the reliability of the annotator when user feedback is scarce, thus improving the identification of potentially inaccurate annotations.

[0041] 3. It has strong scalability. This method can be applied to other fields such as electrical and architectural drawings by changing the structure of the drawing attributes, and it has flexible portability. Attached Figure Description

[0042] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, do not constitute an undue limitation of the invention. In the drawings:

[0043] Figure 1 This is a flowchart of a drawing similarity calculation method considering annotation reliability, as described in an embodiment of the present invention.

[0044] Figure 2 This is a schematic diagram of the electronic device structure shown in an embodiment of the present invention. Detailed Implementation

[0045] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention, but are not intended to limit the present invention.

[0046] Figure 1 This is a flowchart of a design drawing similarity calculation method considering the reliability of annotations according to an embodiment of the present invention.

[0047] The similar drawing fusion retrieval method of the present invention includes the following steps:

[0048] Step 1: The annotation personnel annotate the drawing based on the drawing's attribute structure.

[0049] Annotation personnel refers to those who annotate drawings according to their attribute structure. This represents the set of annotators, where Refers to the first h One labeler, H This represents the total number of annotators. The number of annotators and the rules for assigning annotation tasks are preset based on the requirements of the engineering analysis.

[0050] use This represents a set of drawings, in which for n Zhang's blueprint ,N This represents the total number of drawings.

[0051] use This represents the set of attributes of the drawing, where M The number of attributes. For the first... m Drawing attributes The set of all possible values ​​for this attribute can be represented as ,in Attributes The Several possible values. The drawing attribute structure is preset according to the requirements of engineering analysis.

[0052] Step 2: Based on the drawing annotation results from Step 1, construct the "drawing-attribute" matrix.

[0053] Based on the two dimensions of "drawings" and "attributes", according to the drawing set and drawing attribute set The results of the drawing annotations are converted into a "drawing-attribute" matrix, denoted as the mapping matrix. ,in Refers to blueprints In attributes The values ​​that can be taken on are: .

[0054] Step 3: Based on the "drawing-attribute" matrix constructed in Step 2, calculate the relative weights between attributes.

[0055] The relative weights between attributes are calculated using the information entropy formula:

[0056] in, For attributes The set of values The first in k Each value is in the mapping matrix The ratio of the frequency of occurrence in the text to the total number of drawings. For a given set of drawings In attributes The distribution of values ​​on has .

[0057] Step 4: Collect user feedback on each annotation in the drawing-attribute mapping matrix.

[0058] Users refer to those who browse the annotation information on drawings. User feedback on a particular annotation includes user votes and pageview counts. During the browsing process, user votes on annotations are collected via a "This annotation is unreasonable" voting button, and pageview counts are collected based on user browsing behavior. Indicates by personnel Annotated drawings Attributes The number of votes received for "this label is unreasonable" is used as follows: This indicates the total number of people who viewed this label.

[0059] Step 5: Based on the feedback information collected in Step 4, calculate the reliability distribution of the annotators.

[0060] 51) According to the labeling personnel For drawings Attributes Feedback information received from the annotations and The score for this annotation is calculated as follows:

[0061]

[0062] 52) According to the labeling personnel Calculate the annotation score, calculate The mean and standard deviation of are denoted as . and ;

[0063] 53) Based on the mean and standard deviation Calculate the reliability distribution of the annotators, annotators The reliability distribution is represented by a beta distribution, and the parameters of the beta distribution are used. and Indicates. Annotator The reliability distribution is expressed as . and The setting method meets the following conditions:

[0064] like ,but , .

[0065] like ,but , .

[0066] Step 6: Based on the feedback information collected in Step 4 and the reliability distribution of the annotators calculated in Step 5, calculate the reliability of the annotation.

[0067] Annotation reliability refers to the accuracy with which annotations reflect the attribute information of drawings; user feedback on drawing annotations includes the number of "unreasonable" votes for annotations and the number of times annotations are viewed; the reliability of annotation personnel refers to the ability of annotation personnel to accurately annotate the attribute information of drawings during the annotation process.

[0068] For labelers For drawings Attributes The annotations made are based on the annotation personnel. Reliability distribution This is prior information, based on user feedback received using this annotation ( and Using this as evidence, a Bayesian update method is employed to obtain the posterior distribution that the reliability of this label follows: .

[0069] Given a posterior distribution that the reliability of this label follows, the mean of that distribution is used to measure the reliability of the label. The calculation formula is:

[0070] .

[0071] Step 7: Input the search criteria. The search criteria are consistent with the drawing attribute structure, and are expressed as follows: ;

[0072] Step 8: Based on the annotation reliability calculated in Step 6 and the relative weights between drawing attributes calculated in Step 3, calculate the degree of similarity between each drawing and the search conditions collected in Step 7.

[0073] For a given search criteria ,drawing Search criteria The formula for calculating similarity is: .

[0074] in, Refers to blueprints With conditions In attributes Are the labels the same? .

[0075] Calculate the similarity between all drawings and the given search criteria to obtain a similarity set. , .

[0076] Step 9: Based on the similarity set between each drawing and the given search criteria calculated in Step 8. Result search According to the degree of similarity A sequence of drawings sorted in descending order.

[0077] The implementation process of the present invention will be further illustrated by an example below.

[0078] This embodiment selects 20 non-standard parts mechanical design drawings from a mechanical assembly line as the research object. The following demonstrates the application of the method of this invention in a drawing retrieval scenario, following the sequence of the nine steps described above.

[0079] As per the requirements of step 1, the drawing attribute structure was set to "Design Duration, Outline Shape, Designer, Creation Date, File Size, Technical Features, Number of Holes, Hole Shape, Symmetry, Other Features," the number of annotators was set to 3, and the annotation task allocation rule was set to "Random Assignment." The drawing annotation task allocation results were as follows: Annotator 1 annotated drawings 1 / 2 / 3 / 4 / 5 / 6 / 7, Annotator 2 annotated drawings 8 / 9 / 10 / 11 / 12 / 13, and Annotator 3 annotated drawings 14 / 15 / 16 / 17 / 18 / 19 / 20. This resulted in the annotation results of 3 annotators on 20 drawings across 10 attributes.

[0080] According to the requirements of step 2, the constructed "drawing-attribute" matrix is ​​shown in Table 1.

[0081] Table 1 “Drawing-Attributes” Matrix

[0082]

[0083] According to the requirements of step 3, calculate the relative weights between attributes, as shown in Table 2.

[0084] Table 2 Relative weights between attributes

[0085]

[0086] According to the requirements of step 4, the user feedback information received for each annotation is collected, as shown in Table 3.

[0087] Table 3 User feedback information received for each annotation ( , )

[0088]

[0089] According to the requirements of step 5, calculate the reliability distribution of the three annotators. The score set of all annotations by annotator 1 is {0.667, 0.5, 0, 1……0.75, 0}, the score set of all annotations by annotator 2 is {1, 1, 0……0.857, 0.333}, and the score set of all annotations by annotator 3 is {1, 0.375, 0.333……0.769, 0.8}.

[0090] For annotator 1, the average annotation score is The standard deviation is According to the calculation rules ,get and The reliability distribution of annotator 1 is as follows: .

[0091] For annotator 2, the average annotation score is The standard deviation is According to the calculation rules, we get... and The reliability distribution of labeler 2 is as follows .

[0092] For annotator 3, the average annotation score was The standard deviation is According to the calculation rules, we get... and The reliability distribution of labeler 3 is as follows: .

[0093] Based on the requirements of step 6, and using the user feedback information collected in step 4 and the reliability distribution of annotation personnel calculated in step 5, the reliability of the annotations is calculated. The calculation results are shown in Table 4.

[0094] Table 4 Reliability of each label

[0095]

[0096] According to the requirements of step 7, enter the search criteria. The search criteria are consistent with the drawing attribute structure and are represented as follows: ;

[0097] According to the requirements of step 8, based on the annotation reliability calculated in step 6 and the relative weights between attributes calculated in step 3, the similarity between each drawing and the search criteria input in step 7 is calculated. The similarity calculation results between each drawing and the search criteria are shown in Table 5.

[0098] Table 5. Similarity between each drawing and the search criteria

[0099]

[0100] According to the requirements of step 9, the drawing sequence is presented in descending order of similarity. The search results are: "Drawing 19 | Drawing 13 | Drawing 16 | Drawing 9 | Drawing 18 | Drawing 3 | Drawing 8 | Drawing 14 | Drawing 12 | Drawing 2 | Drawing 1 | Drawing 17 | Drawing 5 | Drawing 15 | Drawing 7 | Drawing 20 | Drawing 4 | Drawing 10 | Drawing 6 | Drawing 11".

[0101] The following simulation experiment demonstrates the advantages of the present invention (method 1) in terms of drawing retrieval performance compared to the drawing retrieval method (method 2) that only considers annotation feedback information and the drawing retrieval method (method 3) that does not consider annotation quality information.

[0102] The simulation process includes the following 11 steps.

[0103] Step 1. Randomly generate 100 annotation personnel. Annotators h quality parameters Follows a normal distribution The quality of annotation by the annotators is used to... express, .

[0104] Step 2. Randomly generate 1000 drawings. and 20 drawing attributes For each drawing attribute Randomly generate a set of values The maximum number of possible values ​​is... ,in, .

[0105] Step 3. Randomly generate a "drawing-feature" matrix The generation method is as follows: for drawings... Attributes ,from A value is randomly selected and assigned to The "drawing-feature" matrix here reflects the real relationship between drawings and attributes, which is not visible in reality.

[0106] Step 4. Generate the "Drawing-Attributes" matrix The generation method is as follows: for drawings... From the labelers One person is randomly selected from the pool (using...) This indicates that the attributes of the drawing are labeled. For the drawing... Attributes Annotators by The probability labeling is correct, that is... ;by The probability labeling is incorrect, that is The "drawing-attribute" matrix here reflects the annotation results of the drawings and attributes, which are visible in reality.

[0107] Step 5. Based on the "Drawing-Attributes" matrix Calculate the weights of each attribute. .

[0108] Step 6. Randomly generate user feedback information on the annotations. and For annotations Total number of votes , .in, This indicates rounding to the nearest integer. , , .

[0109] Step 7. Based on user feedback on annotations, calculate the reliability distribution characteristics of annotation personnel. and .

[0110] Step 8. Calculate the reliability of the annotations based on the reliability distribution characteristics of the annotation personnel and user feedback information. In Method 1, the calculation of label reliability considers both personnel reliability and user feedback, i.e. In Method 2, the calculation of label reliability only considers user feedback information, i.e. Method 3 does not consider the reliability of the annotation, i.e. .

[0111] Step 9. Randomly generate search criteria .

[0112] Step 10. Based on the "drawing-feature" matrix Calculate the similarity between the drawings and the search criteria to obtain the top 20 ideal drawing feedback sequences, denoted as... .

[0113] Based on the "drawing-attribute" matrix The similarity between the drawings and the search criteria is calculated, with annotation reliability obtained using a Bayesian update method (using feedback information and annotation personnel reliability information). Based on the obtained drawing feedback sequence, the first 20 are extracted and denoted as... .

[0114] Based on the "drawing-attribute" matrix The similarity between the drawings and the search criteria is calculated, without considering the reliability of the annotations. Based on the obtained drawing feedback sequence, the first 20 are extracted and denoted as... .

[0115] Based on the "drawing-attribute" matrix The similarity between the drawings and the search criteria is calculated, with the labeling reliability obtained using a ratio method (using only feedback information). Based on the obtained drawing feedback sequence, the first 20 are extracted and denoted as _____. .

[0116] Step 11. Calculate the performance score. The calculation method is as follows: .in, Different sorting methods , refer to Sort the sequence i The blueprints.

[0117] The simulation experiment was conducted following the steps outlined above. In step 9, 10,000 search criteria were randomly generated, and the scores of the three methods were obtained across 10,000 searches. Descriptive statistics of the scores for the three methods are shown in Table 6.

[0118] Table 6. Descriptive statistics of scores from the three calculation methods

[0119]

[0120] Based on the scoring results, the mean score of Method 1 (this invention) is higher than that of Methods 2 and 3. Table 7 shows the significance of the mean scores of different methods in the paired tables. t The results of the test showed that the mean score of Method 1 was significantly higher than that of Methods 2 and 3. t Statistic > 0 and p (Value < 0.01), the mean score of method 2 was significantly higher than that of method 3. t Statistic > 0 and p (Value < 0.01). This result reveals the advantage of the present invention (method 1) in terms of the efficiency of retrieving similar drawings.

[0121] Table 7. Results of paired t-tests on the mean scores of the three calculation methods.

[0122]

[0123] like Figure 2 As shown, the present invention provides an electronic device 100 for implementing a method for fusion retrieval of design drawings considering the reliability of annotations, comprising: a data acquisition card 110, a memory 120, a communication bus 130, and a processor 140.

[0124] The computer device includes a processor 140 and a memory 120. The memory 120 stores a computer program, which includes program instructions. The processor 140 executes the program instructions stored in the computer storage medium. The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), or field-programmable gate arrays (FPGAs). The processor described in this embodiment of the invention can be used to implement a design drawing fusion retrieval method that considers the reliability of annotations.

[0125] This invention is not limited to the above embodiments. Based on the technical solutions disclosed in this invention, those skilled in the art can make some substitutions and modifications to some of the technical features without creative effort, and all such substitutions and modifications are within the protection scope of this invention.

Claims

1. A design drawing fusion retrieval method considering the reliability of annotations, characterized in that, include: Annotators annotate the drawings based on their attribute structure; Construct a mapping matrix of drawings and attributes for the annotated drawings; Based on the drawing-attribute mapping matrix, information entropy is used to characterize the relative weights between drawing attributes; Collect user feedback on each annotation in the drawing-attribute mapping matrix; Based on the collected user feedback information, calculate the reliability distribution of the labelers; Based on the collected user feedback information and the reliability distribution of the annotators, the reliability of the annotations is calculated using the Bayesian update method. Enter search criteria that match the drawing's attribute structure; Based on the reliability of the annotations and the relative weights between the drawing attributes, the similarity between each drawing and the given search criteria is calculated. Based on the degree of similarity between each drawing and the given search criteria, a sequence of drawings is obtained, sorted in descending order of similarity. Based on the collected user feedback information, the reliability distribution of the annotators is calculated, including: Calculate the annotation score based on the feedback received from the annotators regarding the attributes of the drawings; Calculate the average score based on the annotation scores calculated by the annotators. and standard deviation ; According to the mean and standard deviation Calculate the reliability distribution of the annotators, following a beta distribution. and This means that the following conditions are met: like ,but , ; like ,but , ; The reliability of the annotations is calculated using a Bayesian update method, including: For annotations made by annotation personnel to the attributes of drawings, using the reliability distribution of the annotation personnel as prior information and the user feedback information received by the annotation as evidence, a Bayesian update method is used to obtain the posterior distribution of the reliability of the annotation: The reliability of the label is measured using the mean of the posterior distribution to which the label reliability follows.

2. The design drawing fusion retrieval method considering annotation reliability according to claim 1, characterized in that, The drawing attribute structure and the drawing itself include the drawing attribute set and the drawing set, respectively.

3. The design drawing fusion retrieval method considering annotation reliability according to claim 2, characterized in that, Construct the drawing-attribute mapping matrix, including: Based on the drawing set and the drawing attribute set, the drawing annotation results are transformed into a drawing-attribute matrix.

4. The design drawing fusion retrieval method considering annotation reliability according to claim 1, characterized in that, Information entropy is used to characterize the relative weights between attributes, calculated using the following formula: in, For attributes The k Each value is in the mapping matrix The ratio of the frequency of occurrence in the text to the total number of drawings. For a given set of drawings In the m Attributes The distribution of values ​​on, M The number of attributes.

5. The design drawing fusion retrieval method considering annotation reliability according to claim 1, characterized in that, Collect user feedback on each annotation in the drawing-attribute mapping matrix, including: Feedback from users who view the annotations on drawings includes voting information and user view counts. The system collects user voting information on tag labeling by using the "This label is unreasonable" voting button, and collects user pageview information based on user browsing behavior.

6. The design drawing fusion retrieval method considering annotation reliability according to claim 1, characterized in that, Based on the reliability of the annotations and the relative weights between drawing attributes, the similarity between each drawing and the given search criteria is calculated, including: For a given search criteria, determine the similarity between the drawing and the annotations of the search criteria on the drawing attributes; Calculate the similarity between all drawings and the given search criteria to obtain a similarity set.

7. The design drawing fusion retrieval method considering annotation reliability according to claim 1, characterized in that, Based on the degree of similarity between each drawing and the given search criteria, the search results are a sequence of drawings, that is, a sequence of drawings arranged in descending order according to the value of the similarity set.

8. An electronic device, characterized in that, Including the processor and memory; The memory is used to store computer programs, the computer programs including program instructions; The processor is used to invoke a program stored in the memory to execute the retrieval method as described in any one of claims 1-7.

Images