Multi-dimensional work order supervision method based on differential status labels

By extracting features and calculating similarity from historical problem work orders on the work order platform, and combining the error rate and time of manual review, a standard similarity threshold was determined. This solved the problem of decreased platform efficiency caused by users submitting unclear work orders, and achieved more efficient work order review and allocation.

CN118734086BActive Publication Date: 2024-11-29上海市大数据中心
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202410869860.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-01
Publication Date
2024-11-29
Estimated Expiration
2044-07-01

AI Technical Summary

Technical Problem

In the work order platform, the problem work orders submitted by users lack clear system, menu, and module identifiers, which leads to a decrease in platform efficiency. A multi-dimensional supervision method based on differential status tags is needed for review.

Method used

By extracting features from historical problem work orders on the work order platform and calculating similarity to the feature information database, a random similarity threshold is selected. Combined with the error rate and time of manual review, a standard ranking value is calculated to determine the standard similarity threshold, which is used for the review and allocation of work orders in real time.

Benefits of technology

This improved the accuracy and efficiency of the work order platform's review process, reduced the probability of non-compliant work orders entering the platform, and prevented a decline in the work efficiency of reviewers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118734086B_ABST
    Figure CN118734086B_ABST
Patent Text Reader

Abstract

The present invention discloses a multi-dimensional work order supervision method based on differential state labels, and belongs to the field of automated supervision technology. The present invention extracts features from problem work orders to obtain feature information of each historical problem work order; calculates random similarity thresholds of various feature information; calculates the error rate of each random similarity threshold, and collects the ranking value of each random similarity threshold in the error rate set; calculates the time set spent on manual review of each random similarity threshold, and collects the ranking value of each random similarity threshold in the time set spent on manual review; obtains the standard similarity threshold according to the standard position; performs text classification on the real-time problem work order in the work order platform, calculates the real-time similarity of the real-time problem work order, and compares the real-time similarity with the standard similarity threshold. Reduce the number of non-compliant problem work orders entering the work order platform.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of automated supervision, and in particular to a multi-dimensional work order supervision method based on differential state labels. Background Art

[0002] Build a complete closed loop of work order management, and provide "I want to ask questions" service for budget integration users, covering service work orders, implementation work orders, and R&D work orders. Each district can accept and process work orders on its own, and work orders that require assistance from the municipal bureau can be submitted step by step.

[0003] In the work order platform, when users submit problem work orders, they need to clarify the system, menu, and module to which the current problem belongs. However, some users are not very clear about the system, menu, and module to which they belong. If the problem work orders submitted by users are not supervised, the efficiency of the work order platform will decrease. Therefore, a multi-dimensional work order supervision method based on differential status labels is needed to review the problem work orders submitted by users first, avoid non-compliant problem work orders from entering the work order platform, and improve the efficiency of the work order platform. Summary of the invention

[0004] The purpose of the present invention is to provide a multi-dimensional work order supervision method based on differential status labels to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the present invention provides the following technical solution: a multi-dimensional work order supervision method based on differential state labels, the method comprising:

[0006] Step S100: pre-processing the historical problem work orders in the work order platform, extracting features from the processed problem work orders to obtain feature information of each historical problem work order, and calculating similarity between the feature information and a feature information database established for feature information of various problem work orders in the work order platform;

[0007] Step S200: randomly taking values ​​for similarity weights of various types of feature information in the feature information database, and calculating random similarity thresholds of the various types of feature information;

[0008] Step S300: comparing the similarity between the feature information of each historical problem work order and each type of feature information in the feature information database and the random similarity threshold of each type of feature information in the feature information database, to obtain the number of historical problem work orders that meet each type of feature information and the number of historical problem work orders that do not meet each type of feature information;

[0009] Step S400: Manually review each historical problem work order, obtain the number of historical problem work orders that meet various feature information after manual review, calculate the error rate of each random similarity threshold, summarize the error rate, and collect the ranking value of each random similarity threshold in the error rate set;

[0010] Step S500: collecting the time spent on manual review of each historical problem work order, and calculating the time spent on manual review of each random similarity threshold;

[0011] Step S600: Summarize the time spent on manual review of each random similarity threshold, and collect the ranking value of the time spent on manual review of each random similarity threshold;

[0012] Step S700: Calculate and obtain standard ranking values ​​according to the ranking values ​​of each random similarity threshold in the error rate set and the time set spent on manual review, and sort the standard ranking values ​​to obtain standard similarity thresholds for various types of feature information in the feature information database;

[0013] Step S800: Calculate the real-time similarity between the real-time problem work order in the work order platform and various types of feature information in the feature information library, and compare the real-time similarity with the standard similarity threshold of various types of feature information in the feature information library.

[0014] Furthermore, step S100 includes:

[0015] Step S101: collecting several historical problem work orders initiated by users on the work order platform, segmenting and cleaning the text in the historical problem work orders to obtain processed problem work orders, extracting features from the processed problem work orders to obtain feature information of the i-th historical problem work order;

[0016] Step S102: Retrieve the feature information of the i-th historical problem work order, and calculate the feature information of the i-th historical problem work order with the j-th feature information of the feature information library in the work order platform by using the cosine similarity method to obtain the similarity between the feature information of the i-th historical problem work order and the j-th feature information.

[0017] In the above steps, the work order platform refers to providing the "I want to ask questions" service for users in the budget management integrated system, collecting the problem work orders entered by users into the work order platform, and allocating and processing the problem work orders on the work order platform;

[0018] Sentence segmentation and text cleaning are performed, and natural language processing tools are used to obtain processed standard problem tickets;

[0019] Use the BERT model to extract and classify the processed standard problem ticket text to obtain the feature information of the historical problem ticket;

[0020] There is a feature information database inside the work order platform, which is composed of feature information of various problem work orders;

[0021] The feature information of historical problem work orders and the feature information of various problem work orders are calculated, and the cosine similarity method in the BERT model is used to obtain the similarity between the semantic representation vectors of the feature information of historical problem work orders and the feature information of various problem work orders, which supports the subsequent analysis of the similarity threshold.

[0022] Further, step S200 includes:

[0023] Step S201: Summarize the similarities between the feature information of each historical problem ticket and the feature information of the jth category to obtain the similarity set of the feature information of the jth category, and calculate the standard deviation of the similarity of the feature information of the jth category as Q j and the average value is P j ;

[0024] Step S202: Obtain the similarity threshold X of the j-th type of feature information by calculation j , according to the similarity threshold formula: X j =P j -Q j ×N j , where N j It is expressed as the similarity weight range of the j-th type of feature information;

[0025] Step S203: within the similarity weight range of the j-th type of feature information, randomly select the similarity weight, input the randomly selected similarity weight into the similarity threshold formula to obtain the random similarity threshold of the j-th type of feature information, and summarize them to obtain the random similarity threshold set of the j-th type of feature information.

[0026] In the above steps, it is equivalent to calculating the similarity threshold of each type of feature information. In order to obtain a more accurate similarity threshold, when calculating the similarity threshold, the similarity weight is added for calculation. According to the similarity weight range, the similarity weight is randomly selected to obtain a random similarity threshold set;

[0027] Specifying a similarity weight range helps to obtain several similarity thresholds. Different similarity thresholds will result in different numbers of work orders that meet historical issues and different review times. In order to find a better similarity threshold, the similarity weights must be randomly selected. If the similarity weights are not randomly selected, the similarity threshold will be too single, which will reduce the accuracy and efficiency of the work order platform when classifying.

[0028] Furthermore, step S300 includes:

[0029] Step S301: Compare the similarity F between the feature information of the i-th historical problem ticket and the feature information of the j-th category ir , and the rth random similarity threshold X in the random similarity threshold set of the jth feature information jr For comparison, when F ir ≥X jr , the review result of the i-th historical problem work order is marked as a historical problem work order that meets the j-th type of feature information. ir <X jr , the review result of the i-th historical problem work order is marked as a historical problem work order that does not meet the j-th type of feature information;

[0030] Step S302: When the random similarity threshold of the j-th type of feature information is X jr When , the number of historical problem tickets that meet and do not meet the j-th type of feature information are collected respectively, and the random similarity threshold of the j-th type of feature information is obtained as X jr When the number of historical problem tickets that meet and do not meet the j-th feature information is A jr and B jr .

[0031] In the above steps, the data in the random similarity threshold set of each type of feature information is used as a judgment standard, and the similarity between the feature information of the historical problem work order and each type of feature information is compared to determine whether the feature information of the historical problem work order belongs to each type of feature information, and the number of historical problem work orders that meet and do not meet each type of feature information is collected;

[0032] By using different random similarity thresholds as judgment criteria, the number of compliance and non-compliance under different random similarity thresholds can be obtained. If different random similarity thresholds are not used as judgment criteria, the data will be uniform and the purpose of improving accuracy will not be achieved.

[0033] Furthermore, step S400 includes:

[0034] Step S401: Manually review all historical problem work orders, and obtain the number of historical problem work orders A that meet the j-th type of characteristic information after manual review.j ;

[0035] Step S402: When the random similarity threshold of the j-th type of feature information is X jr When the error rate of the j-th feature information is W jr Calculate according to the error rate formula: The error rates corresponding to the random similarity thresholds of the j-th type of feature information are sorted and summarized from low to high to obtain an error rate set of the j-th type of feature information;

[0036] Step S403: The random similarity threshold for the j-th type of feature information is X jr The error rate of the jth type feature information is obtained by collecting the sorted values ​​in the error rate set to obtain a random similarity threshold value of X jr When the ranking value in the error rate set is Y jr .

[0037] In the above steps, because it is easy to make errors when judging whether the historical problem tickets are in compliance with the random similarity threshold, the historical problem tickets are manually reviewed to obtain the number of historical problem tickets that meet various feature information and calculate the error rate;

[0038] Because different random similarity thresholds result in different error rates, all error rates need to be sorted and summarized, and the position of each error rate in the set is marked to obtain the ranking value of the error rate of each random similarity threshold in the set.

[0039] Furthermore, step S500 includes:

[0040] Step S501: The time spent on manual review of all historical problem work orders is collected, and the average time spent on the historical problem work orders with a review result of conforming is calculated as C, and the average time spent on the historical problem work orders with a review result of non-conforming is calculated as D;

[0041] Step S502: When the random similarity threshold of the j-th type of feature information is X jr When the number of historical problem tickets that meet and do not meet the j-th feature information is calculated with the average time spent on the audit, the random similarity threshold of the j-th feature information is obtained as X jr The time spent on reviewing the historical problem ticket of the j-th feature information is T jr , according to the time formula: T jr =A jr ×C+B jr ×D.

[0042] In the above steps, because the work efficiency of the work order platform must be considered, the time spent must be collected.jr ×C is the time spent on the historical problem tickets whose audit results are in compliance. jr ×D is used to calculate the time spent on historical problem work orders with non-conformance results. jr =A jr ×C+B jr ×D obtains the time spent on reviewing various types of feature information, which supports the subsequent analysis of the impact of various random similarity thresholds on manual review.

[0043] Furthermore, step S600 includes:

[0044] Step S601: sorting and summarizing the time spent on the review corresponding to each random similarity threshold of the j-th type of feature information from short to long, and obtaining a set of time spent on the review corresponding to the j-th type of feature information;

[0045] Step S602: The random similarity threshold for the j-th type of feature information is X jr The time spent on the corresponding audit is collected by collecting the ranking values ​​in the time spent on the audit, and the random similarity threshold of the j-th type of feature information is obtained as X jr When the ranking value in the set of time spent on the audit is Z jr .

[0046] The above steps are equivalent to obtaining the corresponding position of each random similarity threshold in the time set spent on manual review, which supports the subsequent analysis to obtain the standard similarity threshold.

[0047] Furthermore, step S700 includes:

[0048] Step S701: Retrieve the random similarity threshold of the j-th type of feature information as X jr When the ranking value in the error rate set and the ranking value in the audit time set are used, the random similarity threshold of the j-th feature information is calculated as X jr The standard ranking value K jr , according to the formula:

[0049] Step S702: Summarize the standard ranking values ​​of the random similarity thresholds of the j-th type of feature information to obtain a standard ranking value set of the j-th type of feature information, sort the ranking value set from small to large, and use the random similarity threshold corresponding to the minimum value as the standard similarity threshold to obtain the standard similarity threshold of the j-th type of feature information;

[0050] Step S703: input the standard similarity threshold of the j-th type of feature information into the feature information library in the work order platform as a judgment standard for the j-th type of feature information.

[0051] In the above steps, Y jr +Z jr It is equivalent to calculating according to the error rate of each random similarity threshold and the ranking value of the time spent on the review. The higher the ranking value, the more reasonable the similarity threshold represented, because such calculation will obtain a standard similarity threshold with a smaller error rate and a shorter time. If the standard ranking value is not calculated, the error rate will become lower, resulting in lower efficiency, or the efficiency will become higher, resulting in a higher error rate. Taking both into consideration, a more appropriate standard similarity threshold can be obtained.

[0052] Furthermore, step S800 includes:

[0053] Step S801: pre-process the real-time problem work order in the work order platform, and calculate the real-time similarity between the real-time problem work order and various feature information;

[0054] Step S802: compare the real-time similarity with the standard similarity threshold of each type of feature information. When the real-time similarity is greater than or equal to the standard similarity threshold, mark the real-time work order as an accepted work order and proceed to the next step of allocation.

[0055] In the above steps, the real-time problem work orders of the work order platform are collected, and the real-time similarity between the real-time problem work orders and various feature information is calculated by using the BERT model. The real-time similarity is compared with the standard similarity threshold of various feature information. When the real-time similarity is greater than or equal to the standard similarity threshold of various feature information, the real-time problem work order is a compliant work order, and the real-time problem work order is marked as an accepted work order, and is entered into the work order platform for the next step of allocation and processing.

[0056] Compared with the prior art, the beneficial effect achieved by the present invention is as follows: the present invention performs feature extraction on historical problem work orders, calculates random similarity thresholds for various types of feature information, calculates the error rate of each random similarity threshold and the time spent on manual review, and collects the ranking values ​​of each random similarity threshold in the error rate set and the ranking values ​​of each random similarity threshold in the time spent on manual review set. In order to avoid an increase in the time spent on manual review due to an error rate that is too low, or an increase in the error rate due to a reduction in the time spent on manual review, a standard ranking value is calculated based on the ranking value, and the random similarity threshold corresponding to the minimum value of the standard ranking value is used as the standard similarity threshold, thereby reducing the number of problem work orders that do not meet the requirements from entering the work order platform, which would reduce the work efficiency of reviewers. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0058] Figure 1 It is a flow chart of the multi-dimensional work order supervision method based on differential status labels of the present invention. DETAILED DESCRIPTION

[0059] See also Figure 1 The present invention provides a technical solution: a multi-dimensional work order supervision method based on differential state labels, the method comprising:

[0060] Step S100: pre-processing the historical problem work orders in the work order platform, extracting features from the processed problem work orders to obtain feature information of each historical problem work order, and calculating similarity between the feature information and a feature information database established for feature information of various problem work orders in the work order platform;

[0061] Wherein, step S100 includes:

[0062] Step S101: collecting several historical problem work orders initiated by users on the work order platform, segmenting and cleaning the text in the historical problem work orders to obtain processed problem work orders, extracting features from the processed problem work orders to obtain feature information of the i-th historical problem work order;

[0063] Step S102: Retrieve the feature information of the i-th historical problem work order, and calculate the feature information of the i-th historical problem work order with the j-th feature information of the feature information library in the work order platform by using the cosine similarity method to obtain the similarity between the feature information of the i-th historical problem work order and the j-th feature information.

[0064] Step S200: randomly taking values ​​for similarity weights of various types of feature information in the feature information database, and calculating random similarity thresholds of the various types of feature information;

[0065] Wherein, step S200 includes:

[0066] Step S201: Summarize the similarities between the feature information of each historical problem ticket and the feature information of the jth category to obtain the similarity set of the feature information of the jth category, and calculate the standard deviation of the similarity of the feature information of the jth category as Q j and the average value is P j ;

[0067] Step S202: Obtain the similarity threshold X of the j-th type of feature information by calculation j , according to the similarity threshold formula: X j =Pj -Q j ×N j , where N j It is expressed as the similarity weight range of the j-th type of feature information;

[0068] Step S203: within the similarity weight range of the j-th type of feature information, randomly select the similarity weight, input the randomly selected similarity weight into the similarity threshold formula to obtain the random similarity threshold of the j-th type of feature information, and summarize them to obtain the random similarity threshold set of the j-th type of feature information.

[0069] Step S300: comparing the similarity between the feature information of each historical problem work order and each type of feature information in the feature information database and the random similarity threshold of each type of feature information in the feature information database, to obtain the number of historical problem work orders that meet each type of feature information and the number of historical problem work orders that do not meet each type of feature information;

[0070] Wherein, step S300 includes:

[0071] Step S301: Compare the similarity F between the feature information of the i-th historical problem ticket and the feature information of the j-th category ir , and the rth random similarity threshold X in the random similarity threshold set of the jth feature information jr For comparison, when F ir ≥X jr , the review result of the i-th historical problem work order is marked as a historical problem work order that meets the j-th type of feature information. ir <X jr , the review result of the i-th historical problem work order is marked as a historical problem work order that does not meet the j-th type of feature information;

[0072] Step S302: When the random similarity threshold of the j-th type of feature information is X jr When , the number of historical problem tickets that meet and do not meet the j-th type of feature information are collected respectively, and the random similarity threshold of the j-th type of feature information is obtained as X jr When the number of historical problem tickets that meet and do not meet the j-th feature information is A jr and B jr .

[0073] For example, for the first type of feature information, the first random similarity threshold is 0.3, and the number of qualified work orders in the first type of feature information is 8 and the number of unqualified work orders is 2;

[0074] The first type of feature information, the second random similarity threshold is 0.6, and the number of work orders that meet the first type of feature information is 5 and the number of work orders that do not meet the first type of feature information is 5;

[0075] The first type of feature information, the third random similarity threshold is 0.8, and the number of work orders that meet the first type of feature information is 3 and the number of work orders that do not meet the first type of feature information is 7.

[0076] Step S400: Manually review each historical problem work order, obtain the number of historical problem work orders that meet various feature information after manual review, calculate the error rate of each random similarity threshold, summarize the error rate, and collect the ranking value of each random similarity threshold in the error rate set;

[0077] Wherein, step S400 includes:

[0078] Step S401: Manually review all historical problem work orders, and obtain the number of historical problem work orders A that meet the j-th type of characteristic information after manual review. j ;

[0079] Step S402: When the random similarity threshold of the j-th type of feature information is X jr When the error rate of the j-th feature information is W jr Calculate according to the error rate formula: The error rates corresponding to the random similarity thresholds of the j-th type of feature information are sorted and summarized from low to high to obtain an error rate set of the j-th type of feature information;

[0080] Step S403: The random similarity threshold for the j-th type of feature information is X jr The error rate of the jth type feature information is obtained by collecting the sorted values ​​in the error rate set to obtain a random similarity threshold value of X jr When the ranking value in the error rate set is Y jr .

[0081] For example, a manual review is conducted on each historical problem ticket, and the manual review result shows that the number of historical problem tickets that meet the first type of characteristic information is 6;

[0082] The error rate of the first random similarity threshold is 33.3%, the error rate of the second random similarity threshold is 16.7%, and the error rate of the third random similarity threshold is 50%;

[0083] The error rate set is {16.7%, 33.3%, 50%}, the ranking value of the first random similarity threshold in the error rate set is 2, the ranking value of the second random similarity threshold in the error rate set is 1, and the ranking value of the third random similarity threshold in the error rate set is 3.

[0084] Step S500: collecting the time spent on manual review of each historical problem work order, and calculating the time spent on manual review of each random similarity threshold;

[0085] Wherein, step S500 includes:

[0086] Step S501: The time spent on manual review of all historical problem work orders is collected, and the average time spent on the historical problem work orders with a review result of conforming is calculated as C, and the average time spent on the historical problem work orders with a review result of non-conforming is calculated as D;

[0087] Step S502: When the random similarity threshold of the j-th type of feature information is X jr When the number of historical problem tickets that meet and do not meet the j-th feature information is calculated with the average time spent on the audit, the random similarity threshold of the j-th feature information is obtained as X jr The time spent on reviewing the historical problem ticket of the j-th feature information is T jr , according to the time formula: T jr =A jr ×C+B jr ×D.

[0088] For example, the average time spent on auditing a work order with a compliance result is calculated to be 3 minutes, and the average time spent on auditing a work order with a non-compliance result is 5 minutes;

[0089] For the first random similarity threshold, the time spent on reviewing historical problem tickets for the first type of feature information is 34 minutes;

[0090] For the second random similarity threshold, the time spent on reviewing historical problem tickets for the first type of feature information is 40 minutes;

[0091] At the third random similarity threshold, the time spent on reviewing historical problem tickets for the first type of feature information is 44 minutes.

[0092] Step S600: Summarize the time spent on manual review of each random similarity threshold, and collect the ranking value of the time spent on manual review of each random similarity threshold;

[0093] Wherein, step S600 includes:

[0094] Step S601: sorting and summarizing the time spent on the review corresponding to each random similarity threshold of the j-th type of feature information from short to long, and obtaining a set of time spent on the review corresponding to the j-th type of feature information;

[0095] Step S602: The random similarity threshold for the j-th type of feature information is Xjr The time spent on the corresponding audit is collected by collecting the ranking values ​​in the time spent on the audit, and the random similarity threshold of the j-th type of feature information is obtained as X jr When the ranking value in the set of time spent on the audit is Z jr .

[0096] For example, the time set spent on manual review is {34 minutes, 40 minutes, 44 minutes}, the ranking value of the first random similarity threshold in the error rate set is 1, the ranking value of the second random similarity threshold in the error rate set is 2, and the ranking value of the third random similarity threshold in the error rate set is 3.

[0097] Step S700: Calculate and obtain standard ranking values ​​according to the ranking values ​​of each random similarity threshold in the error rate set and the time set spent on manual review, and sort the standard ranking values ​​to obtain standard similarity thresholds for various types of feature information in the feature information database;

[0098] Wherein, step S700 includes:

[0099] Step S701: Retrieve the random similarity threshold of the j-th type of feature information as X jr When the ranking value in the error rate set and the ranking value in the audit time set are used, the random similarity threshold of the j-th feature information is calculated as X jr The standard ranking value K jr , according to the formula:

[0100] Step S702: Summarize the standard ranking values ​​of the random similarity thresholds of the j-th type of feature information to obtain a standard ranking value set of the j-th type of feature information, sort the ranking value set from small to large, and use the random similarity threshold corresponding to the minimum value as the standard similarity threshold to obtain the standard similarity threshold of the j-th type of feature information;

[0101] Step S703: input the standard similarity threshold of the j-th type of feature information into the feature information library in the work order platform as a judgment standard for the j-th type of feature information.

[0102] Step S800: Calculate the real-time similarity between the real-time problem work order in the work order platform and various types of feature information in the feature information library, and compare the real-time similarity with the standard similarity threshold of various types of feature information in the feature information library.

[0103] Wherein, step S800 includes:

[0104] Step S801: pre-process the real-time problem work order in the work order platform, and calculate the real-time similarity between the real-time problem work order and various feature information;

[0105] Step S802: compare the real-time similarity with the standard similarity threshold of each type of feature information. When the real-time similarity is greater than or equal to the standard similarity threshold, mark the real-time work order as an accepted work order and proceed to the next step of allocation.

[0106] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0107] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-dimensional work order supervision method based on differential state labels, characterized in that: Methods include: Step S100: pre-processing the historical problem work orders in the work order platform, extracting features from the processed problem work orders to obtain feature information of each historical problem work order, and calculating similarity between the feature information and a feature information database established for feature information of various problem work orders in the work order platform; Step S200: randomly taking values ​​for similarity weights of various types of feature information in the feature information database, and calculating random similarity thresholds of the various types of feature information; Step S300: comparing the similarity between the feature information of each historical problem work order and each type of feature information in the feature information database and the random similarity threshold of each type of feature information in the feature information database, to obtain the number of historical problem work orders that meet each type of feature information and the number of historical problem work orders that do not meet each type of feature information; Step S400: Manually review each historical problem work order, obtain the number of historical problem work orders that meet various feature information after manual review, calculate the error rate of each random similarity threshold, summarize the error rate, and collect the ranking value of each random similarity threshold in the error rate set; Step S500: collecting the time spent on manual review of each historical problem work order, and calculating the time spent on manual review of each random similarity threshold; Step S600: Summarize the time spent on manual review of each random similarity threshold, and collect the ranking value of the time spent on manual review of each random similarity threshold; Step S700: Calculate and obtain standard ranking values ​​according to the ranking values ​​of each random similarity threshold in the error rate set and the time set spent on manual review, and sort the standard ranking values ​​to obtain standard similarity thresholds for various types of feature information in the feature information database; Step S800: Calculate the real-time similarity between the real-time problem ticket in the ticket platform and various types of feature information in the feature information database, and compare the real-time similarity with the standard similarity threshold of various types of feature information in the feature information database; The step S800 includes: Step S801: pre-process the real-time problem work order in the work order platform, and calculate the real-time similarity between the real-time problem work order and various feature information; Step S802: compare the real-time similarity with the standard similarity threshold of each type of feature information. When the real-time similarity is greater than or equal to the standard similarity threshold, mark the real-time work order as an accepted work order and proceed to the next step of allocation.

2. The multi-dimensional work order supervision method based on differential state labels according to claim 1 is characterized in that: The step S100 includes the following steps: Step S101: collecting several historical problem work orders initiated by users on the work order platform, segmenting and cleaning the text in the historical problem work orders to obtain processed problem work orders, extracting features from the processed problem work orders to obtain feature information of the i-th historical problem work order; Step S102: Retrieve the feature information of the i-th historical problem work order, and calculate the feature information of the i-th historical problem work order with the j-th feature information of the feature information library in the work order platform by using the cosine similarity method to obtain the similarity between the feature information of the i-th historical problem work order and the j-th feature information.

3. The multi-dimensional work order supervision method based on differential state labels according to claim 2 is characterized in that: The step S200 includes the following steps: Step S201: Summarize the similarities between the feature information of each historical problem ticket and the feature information of the jth category to obtain the similarity set of the feature information of the jth category, and calculate the standard deviation of the similarity of the feature information of the jth category as Q j and the average value is P j ; Step S202: Obtain the similarity threshold X of the j-th type of feature information by calculation j , according to the similarity threshold formula: X j =P j -Q j ×N j , where N j It is expressed as the similarity weight range of the j-th type of feature information; Step S203: within the similarity weight range of the j-th type of feature information, randomly select the similarity weight, input the randomly selected similarity weight into the similarity threshold formula to obtain the random similarity threshold of the j-th type of feature information, and summarize them to obtain the random similarity threshold set of the j-th type of feature information.

4. The multi-dimensional work order supervision method based on differential state labels according to claim 3 is characterized in that: The step S300 includes the following steps: Step S301: Compare the similarity F between the feature information of the i-th historical problem ticket and the feature information of the j-th category ir , and the rth random similarity threshold X in the random similarity threshold set of the jth feature information jr For comparison, when F ir ≥X jr , the review result of the i-th historical problem work order is marked as a historical problem work order that meets the j-th type of feature information. ir <X jr , the review result of the i-th historical problem work order is marked as a historical problem work order that does not meet the j-th type of feature information; Step S302: When the random similarity threshold of the j-th type of feature information is X jr When , the number of historical problem tickets that meet and do not meet the j-th type of feature information are collected respectively, and the random similarity threshold of the j-th type of feature information is obtained as X jr When the number of historical problem tickets that meet and do not meet the j-th feature information are X jr and B jr .

5. The multi-dimensional work order supervision method based on differential state labels according to claim 4 is characterized in that: The step S400 includes the following steps: Step S401: Manually review all historical problem work orders, and obtain the number of historical problem work orders A that meet the j-th type of characteristic information after manual review. j ; Step S402: When the random similarity threshold of the j-th type of feature information is X jr When the error rate of the j-th feature information is W jr Calculate according to the error rate formula: The error rates corresponding to the random similarity thresholds of the j-th type of feature information are sorted and summarized from low to high to obtain an error rate set of the j-th type of feature information; Step S403: The random similarity threshold for the j-th type of feature information is X jr The error rate of the jth type feature information is obtained by collecting the sorted values ​​in the error rate set to obtain a random similarity threshold value of X jr When the ranking value in the error rate set is Y jr .

6. The multi-dimensional work order supervision method based on differential state labels according to claim 5 is characterized in that: The step S500 includes the following steps: Step S501: The time spent on manual review of all historical problem work orders is collected, and the average time spent on the historical problem work orders with a review result of conforming is calculated as C, and the average time spent on the historical problem work orders with a review result of non-conforming is calculated as D; Step S502: When the random similarity threshold of the j-th type of feature information is X jr When the number of historical problem tickets that meet and do not meet the j-th feature information is calculated with the average time spent on the audit, the random similarity threshold of the j-th feature information is obtained as X jr The time spent on reviewing the historical problem ticket of the j-th feature information is T jr , according to the time spent formula: T jr =A jr ×C+B jr ×D.

7. The multi-dimensional work order supervision method based on differential state labels according to claim 6 is characterized in that: The step S600 includes the following steps: Step S601: sorting and summarizing the time spent on the review corresponding to each random similarity threshold of the j-th type of feature information from short to long, and obtaining a set of time spent on the review corresponding to the j-th type of feature information; Step S602: The random similarity threshold for the j-th type of feature information is X jr The time spent on the corresponding audit is collected by collecting the ranking values ​​in the time spent on the audit, and the random similarity threshold of the j-th type of feature information is obtained as X jr When the ranking value in the set of time spent on the audit is Z jr .

8. The multi-dimensional work order supervision method based on differential state labels according to claim 7 is characterized in that: The step S700 includes the following steps: Step S701: Retrieve the random similarity threshold of the j-th type of feature information as X jr When the ranking value in the error rate set and the ranking value in the audit time set are used, the random similarity threshold of the j-th feature information is calculated as X jr The standard ranking value K jr , according to the formula: Step S702: Summarize the standard ranking values ​​of the random similarity thresholds of the j-th type of feature information to obtain a standard ranking value set of the j-th type of feature information, sort the ranking value set from small to large, and use the random similarity threshold corresponding to the minimum value as the standard similarity threshold to obtain the standard similarity threshold of the j-th type of feature information; Step S703: input the standard similarity threshold of the j-th type of feature information into the feature information library in the work order platform as a judgment standard for the j-th type of feature information.

Citation Information

Patent Citations

  • Abnormal work order identification method and device, electronic equipment, and readable storage medium

    CN113626607A

  • Image matching method, system, device and medium

    CN114581853A