Work order identification method and apparatus, electronic device, and computer program product
By using a target classification model for work order identification, and employing LDA and LSTM models for feature extraction, combined with an RNN model for classification, the problem of low efficiency in manual identification is solved. This enables automatic and rapid classification and accurate identification of work orders, reducing the impact of human factors.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- INNER MONGOLIA MOBILE
- Filing Date
- 2022-03-07
- Publication Date
- 2026-06-09
AI Technical Summary
Manual identification of work orders is inefficient, leading to untimely processing and impacting system operation. This is especially true when there are a large number of alarms and the interface refreshes frequently, where human factors have a significant impact.
A target classification model is used for work order identification. Through feature extraction, feature fusion and text classification, the work order classification is automatically identified. LDA topic model and LSTM model are used for feature extraction, and RNN model is combined for classification to reduce redundancy and improve accuracy.
It enables automatic and rapid classification of work orders, reduces the impact of human factors, improves identification efficiency and accuracy, ensures timely processing of work orders, and reduces the possibility of mis-assignment and missed assignment.
Smart Images
Figure CN116775861B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of information processing technology, specifically to a work order identification method, device, electronic device, and computer program product. Background Technology
[0002] Currently, before processing work orders, backend staff need to categorize and identify them based on their descriptions and other information, and then dispatch them according to the categorization results. With a large number of work orders, manual dispatching is extremely inefficient. Failure to process work orders promptly can lead to serious consequences.
[0003] For example, during the alarm work order dispatch process, on-duty personnel need to manually locate the corresponding monitoring responsible person based on the alarm content, alarm level, and alarm type, and then notify the maintenance personnel. Mobile communication networks have hundreds of thousands to millions of monitoring alarm points, resulting in a large volume of alarms and rapid interface refresh rates. Manual work order identification is inefficient and significantly influenced by human factors. If on-duty personnel fail to discover or notify alarms in a timely manner, the resulting alarms cannot be processed promptly, leading to serious consequences. Summary of the Invention
[0004] This application provides a work order identification method, device, electronic device, and computer program product to solve the technical problem of low efficiency in manual work order identification.
[0005] In a first aspect, embodiments of this application provide a work order identification method, including:
[0006] The target feature vector corresponding to the work order to be identified is input into the feature extraction layer of the target classification model for feature extraction, so as to obtain the global feature vector and local feature vector corresponding to the target feature vector;
[0007] The global feature vector and the local feature vector are input into the feature fusion layer of the target classification model to perform feature fusion and obtain a fused feature vector.
[0008] The fused feature vector is input into the text classification layer of the target classification model for classification to obtain the target classification of the work order to be identified.
[0009] The target classification model is trained based on sample work orders carrying classification labels.
[0010] In one embodiment, the target classification model is trained as follows:
[0011] Based on the average difference and average character length of the candidate keywords in the sample work orders, target candidate keywords are determined; the candidate keywords are keywords that appear in the sample work orders with a frequency exceeding a frequency threshold.
[0012] Remove the target candidate keywords from the sample work order to generate the target text sequence;
[0013] The initial classification model is trained based on the feature vector corresponding to the target text sequence and the classification label to obtain the target classification model.
[0014] In one embodiment, the candidate keywords include multiple key-value pairs;
[0015] The average degree of difference is determined in the following manner:
[0016] Determine the word vectors corresponding to any two values in the plurality of key-value pairs;
[0017] The difference between any two values is determined based on the cosine similarity between the word vectors corresponding to any two values.
[0018] The average difference is determined based on the difference between any two values.
[0019] The average character length is determined in the following way:
[0020] The average character length is determined based on the character length of each value in the plurality of key-value pairs.
[0021] In one embodiment, before inputting the target feature vector corresponding to the work order to be identified into the feature extraction layer of the target classification model for feature extraction, the method further includes:
[0022] Based on the target candidate keywords, the work orders to be identified are deduplicated;
[0023] The deduplicated work orders to be identified are vectorized to obtain the target feature vector.
[0024] In one embodiment, the step of inputting the target feature vector corresponding to the work order to be identified into the feature extraction layer of the target classification model for feature extraction, to obtain the global feature vector and local feature vector corresponding to the target feature vector, includes:
[0025] The target feature vector is extracted using the LDA topic model in the feature extraction layer to obtain the global feature vector;
[0026] The target feature vector is extracted using the Long Short-Term Memory (LTSM) network model in the feature extraction layer to obtain the local feature vector.
[0027] In one embodiment, the step of inputting the global feature vector and the local feature vector into the feature fusion layer of the target classification model for feature fusion to obtain a fused feature vector includes:
[0028] The global feature vector and the local feature vector are fused by means of bitwise concatenation, bitwise addition, or bitwise multiplication to obtain the fused feature vector.
[0029] In one embodiment, inputting the fused feature vector into the text classification layer of the target classification model for classification to obtain the target classification corresponding to the work order to be identified includes:
[0030] The fused feature vector is input into the recurrent neural network (RNN) model in the text classification layer for classification to obtain the target classification.
[0031] Secondly, embodiments of this application provide a work order identification device, comprising:
[0032] The feature extraction module is used to input the target feature vector corresponding to the work order to be identified into the feature extraction layer of the target classification model for feature extraction, so as to obtain the global feature vector and local feature vector corresponding to the target feature vector;
[0033] The feature fusion module is used to input the global feature vector and the local feature vector into the feature fusion layer in the target classification model to perform feature fusion and obtain a fused feature vector.
[0034] The recognition module is used to input the fused feature vector into the text classification layer of the target classification model for classification, so as to obtain the target classification corresponding to the work order to be recognized.
[0035] The target classification model is trained based on sample work orders carrying classification labels.
[0036] Thirdly, embodiments of this application provide an electronic device, including a processor and a memory storing a computer program, wherein the processor executes the program to implement the work order identification method described in the first aspect.
[0037] Fourthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the work order identification method described in the first aspect.
[0038] The work order identification method, device, electronic device, and computer program product provided in this application extract and fuse features from the target feature vector corresponding to the work order to be identified through a target classification model. This can obtain fused feature vectors with different meanings and levels, effectively reducing the loss of work order information and enabling automatic classification. This eliminates the influence of human factors on work order processing, ensuring that the target classification label of the work order can be automatically and quickly identified without human intervention, while also improving the accuracy of work order identification. Attached Figure Description
[0039] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0040] Figure 1 This is a flowchart illustrating the work order identification method provided in the embodiments of this application;
[0041] Figure 2 This is a schematic diagram of the model structure of the work order identification method provided in the embodiments of this application;
[0042] Figure 3 This is a schematic diagram illustrating the feature fusion of the work order identification method provided in the embodiments of this application;
[0043] Figure 4 This is a schematic diagram of the preprocessing flow of the work order identification method provided in the embodiments of this application;
[0044] Figure 5 This is an overall structural diagram of the work order identification method provided in the embodiments of this application;
[0045] Figure 6 This is a schematic diagram of the work order identification device provided in the embodiments of this application;
[0046] Figure 7 This is a schematic diagram of the physical structure of the electronic device provided in the embodiments of this application. Detailed Implementation
[0047] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0048] The execution subject of the work order identification method provided in this application can be an electronic device, a component in an electronic device, an integrated circuit, or a chip. The electronic device can be a mobile electronic device or a non-mobile electronic device. For example, a mobile electronic device can be a mobile phone, tablet computer, laptop computer, PDA, in-vehicle electronic device, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc., while a non-mobile electronic device can be a server, network attached storage (NAS), personal computer (PC), ATM, or self-service machine, etc. This invention does not impose specific limitations.
[0049] The following example, using a computer executing the work order recognition method provided in this application, illustrates the technical solution of this application in detail.
[0050] Figure 1 This is a flowchart illustrating the work order identification method provided in an embodiment of this application. (Refer to...) Figure 1 This application provides a work order identification method, which may include steps 110, 120 and 130.
[0051] Step 110: Input the target feature vector corresponding to the work order to be identified into the feature extraction layer of the target classification model for feature extraction to obtain the global feature vector and local feature vector corresponding to the target feature vector.
[0052] It should be noted that the work orders to be identified are those that need to be classified. A work order can be a simple maintenance or manufacturing plan based on work tasks consisting of one or more operations.
[0053] For example: alarm work orders or maintenance work orders in mobile communication networks; work orders used to describe important issues in an enterprise's internal systems, such as work orders that report process problems in an enterprise's office system; work orders used to record customer needs, such as work orders generated by customers in customer service systems in the financial or other fields regarding order reminders or complaints.
[0054] In practice, a target classification model can include a text preprocessing layer, a text input layer, a feature extraction layer, a feature fusion layer, a text classification layer, and an output layer, such as... Figure 2 As shown.
[0055] In the text preprocessing layer, before inputting the target feature vector into the trained target feature model, the text information corresponding to the work order to be identified needs to be preprocessed, including deleting irrelevant and duplicate information from the original text information, initially filtering out information that is not related to classification, and then inputting the deduplicated text information as the input text sequence into the target classification model for classification.
[0056] In the text input layer, embedding can be used to convert the input text sequence corresponding to the work order to be identified into a vector, thus obtaining the target feature vector. This target feature vector is then used as the input to the feature extraction layer.
[0057] Embedding is a vectorization technique that compresses the dimensionality of discrete features and generalizes feature representations. Embedding vectors can represent certain features of a corresponding object, and the distance between vectors reflects the similarity between objects. Embedding vectors can incorporate almost any information for encoding, making them inherently rich in valuable information. Furthermore, embedding vectors are often concatenated with other features and fed into subsequent deep learning networks for training.
[0058] In the feature extraction layer, global and local feature extraction can be performed on the target feature vector to obtain feature vectors with different meanings and levels, which can include global feature vectors representing global semantics and local feature vectors representing local semantics.
[0059] Step 120: Input the global feature vector and local feature vector into the feature fusion layer in the target classification model to perform feature fusion and obtain the fused feature vector.
[0060] In this step, feature fusion is represented as an optimized combination of different feature vectors. Feature fusion can obtain the most differentiated information from multiple original feature sets involved in the fusion, and can eliminate redundant information caused by the correlation between different feature sets.
[0061] In one embodiment, the global feature vector and the local feature vector are input into the feature fusion layer of the target classification model for feature fusion to obtain a fused feature vector, including:
[0062] The global feature vector and the local feature vector are fused by using bitwise concatenation, bitwise addition, or bitwise multiplication to obtain the fused feature vector.
[0063] It should be noted that feature fusion can be performed in any of the following ways: bitwise summation, bitwise multiplication, bitwise concatenation or other transformation methods. The embodiments of this application do not impose specific limitations.
[0064] like Figure 3As shown, feature vector 1 is a global feature vector, and feature vector 2 is a local feature vector. Feature vector 1 and feature vector 2 can be combined by bitwise summation, bitwise multiplication, bitwise concatenation or other transformations to obtain a fused feature vector.
[0065] Preferably, the global feature vector and the local feature vector can be concatenated bit by bit to obtain the fused feature vector. The fused feature vector is the text feature corresponding to the work order to be identified. The text feature refers to the vectorized modeling of text data to reflect the features in the text that are highly representative and computationally valuable.
[0066] The work order recognition method provided in this application embodiment can ensure the integrity of the fused feature vector through feature fusion, which can greatly improve the performance of the target classification model.
[0067] Step 130: Input the fused feature vector into the text classification layer of the target classification model for classification to obtain the target classification corresponding to the work order to be identified.
[0068] In the text classification layer, the fused feature vector is input into the text classification layer, which uses the Softmax function for classification.
[0069] In the output layer, the final output is the target classification predicted by the target classification model. The target classification is the category corresponding to the work order to be identified. For example, the target classification of an alarm work order can be equipment alarm, environmental alarm, processing error alarm, service quality alarm, communication alarm, and software error alarm, etc.
[0070] The target classification model is trained based on sample work orders carrying classification labels.
[0071] It is understandable that sample work orders can include work orders carrying classification labels. Sample work orders can be divided into training and test sets according to a target ratio. For example, the target ratio could be 8:2.
[0072] In practice, the training set from the sample work orders is input into the initial classification model for training until the initial classification model converges, resulting in the first classification model. This first classification model is then applied to the test set from the sample work orders for prediction, thereby determining the best-performing target classification model.
[0073] After determining the target category, the corresponding work order handler can be assigned based on the target category. Using Robotic Process Automation (RPA) technology, the notification process is triggered according to the pre-set notification level and the option to notify by phone. The notification process is then initiated based on the pre-set work order handler and their corresponding contact information.
[0074] For example, a mobile phone number can be used to automatically trigger dialing. After the call is connected, the text-to-speech system can be used to automatically inform the person handling the work order of the work order content.
[0075] The work order recognition method provided in this application uses a target classification model to extract and fuse features from the target feature vector corresponding to the work order to be recognized. This can obtain fused feature vectors with different meanings and levels, effectively reducing the loss of work order information and enabling automatic classification. This eliminates the influence of human factors on work order processing, ensuring that the target classification label of the work order can be automatically and quickly identified without human intervention. At the same time, it can also improve the accuracy of work order recognition.
[0076] In one embodiment, the target classification model is trained as follows:
[0077] Based on the average difference and average character length of the candidate keywords in the sample work orders, target candidate keywords are determined; the candidate keywords are keywords that appear in the sample work orders with a frequency exceeding a frequency threshold.
[0078] Remove target candidate keywords from sample work orders to generate target text sequences;
[0079] The initial classification model is trained based on the feature vectors and classification labels corresponding to the target text sequence to obtain the target classification model.
[0080] Before performing work order recognition, it is necessary to determine the target classification model. First, sample work orders need to be collected and preprocessed.
[0081] It is understood that a sample work order may include information from several sample work orders. This information may include keywords from several sample work orders; different types of work order information may include different types of keywords, which are not specifically limited here.
[0082] For example, sample alarm work orders may contain sample alarm information such as alarm occurrence time, alarm discovery time, alarm type, alarm level, alarm location information, device type, and manufacturer. Sample maintenance work orders may contain sample maintenance information such as maintenance location, maintenance items, repair request time, maintenance time, and maintenance cost.
[0083] The sample work order preprocessing process is as follows:
[0084] Keywords that appear frequently in each sample work order are identified as candidate keywords. For example, if 20 sample work orders are obtained, all keywords are extracted, and the frequency threshold is set to 10, then keywords that appear more than 10 times in each of the 20 sample work orders are identified as candidate keywords.
[0085] The frequency threshold can be set according to user needs, or it can be the system's default value or range value, and no specific limitation is made here.
[0086] For each candidate keyword, calculate its average dissimilarity and average character length. Then, based on the average dissimilarity and average character length of each candidate keyword, determine the target candidate keywords. Target candidate keywords are those that are difficult to distinguish between sample work orders, and therefore need to be eliminated.
[0087] The average dissimilarity is determined based on the cosine similarity of the candidate keywords. Cosine similarity is calculated by taking the cosine of the angle between two vectors to assess their similarity. The average character length is determined based on the character length of each candidate keyword.
[0088] Divide the average dissimilarity of each candidate keyword by the corresponding average character length to obtain the ratio of the average dissimilarity to the average character length for each candidate keyword.
[0089] For each candidate keyword, if the ratio is less than the first threshold, it indicates that the candidate keyword is difficult to distinguish between different samples, and thus the candidate keyword is determined as a target candidate keyword. If the ratio is greater than the first threshold, it indicates that the candidate keyword can distinguish between different samples, meaning that the candidate keyword is not a target candidate keyword.
[0090] The first threshold can be set according to user needs, or it can be a system default value or range value, and no specific limitation is made here.
[0091] After removing target candidate keywords from all keywords included in the sample work order, a target text sequence can be generated based on the remaining keywords. This target text sequence is then converted into a vector using embedding, yielding the target feature vector. This target feature vector is then used as input to the target classification model.
[0092] The feature vector corresponding to the target text sequence and the classification label carried by the sample work order are input into the initial classification model for training, and the output result of the initial classification model is obtained.
[0093] The sample work orders can be divided into training and testing sets according to a target ratio. For example, the target ratio could be 8:2.
[0094] In practice, the training set from the sample work orders is input into the initial classification model for training until the initial classification model converges, resulting in the first classification model. This first classification model is then applied to the test set from the sample work orders for prediction. During training, a loss function can be used to calculate the error, thereby continuously updating the model parameters of the first classification model until the expected goal is achieved, ultimately completing the training and obtaining the target classification model.
[0095] The work order recognition method provided in this application can reduce the redundancy of training samples and reduce the input dimension of the target classification model by preprocessing the sample work orders. By inputting the preprocessed target text sequence into the target classification model for training, the training time and training effect of the target classification model can be shortened, thereby improving the accuracy of the target classification model in classifying and recognizing work orders.
[0096] In one embodiment, candidate keywords include multiple key-value pairs;
[0097] The average degree of difference is determined as follows:
[0098] Determine the word vectors corresponding to any two values in a set of key-value pairs;
[0099] The degree of difference between any two values is determined based on the cosine similarity between the word vectors corresponding to any two values.
[0100] Determine the average degree of difference based on the degree of difference between any two values;
[0101] The average character length is determined as follows:
[0102] The average character length is determined based on the character length of each value in multiple key-value pairs.
[0103] Understandably, before determining target candidate keywords, it is necessary to first determine the average difference and average character length of the candidate keywords.
[0104] The following example, applied to an alarm work order recognition scenario, illustrates the process of determining the average difference of candidate keywords, the average character length of candidate keywords, and the target candidate keywords.
[0105] In practice, candidate keywords include multiple key-value pairs, which are generally represented as "key:value".
[0106] The process of determining target candidate keywords mainly consists of two aspects: first, determining the average difference of candidate keywords, and second, determining the average character length of candidate keywords.
[0107] like Figure 4 As shown, step 410 involves obtaining multiple sample alarm messages. Each sample alarm message includes multiple key-value pairs. For each sample alarm message, each key-value pair is extracted.
[0108] For example, each sample alarm information key-value pair may include at least one of the following:
[0109] "Alarm occurrence time: 2021-03-19 17:32:02", "Alarm discovery time: 2021-03-19 17:32:07", "Alarm type: original device alarm", "Alarm level: Level 2", "Alarm location information: Inner Mongolia Autonomous Region", "Manufacturer: Company Z", and "Device type: Access and Mobility Management Function (AMF) device, etc."
[0110] Then, the keys of keywords that appear most frequently from all key-value pairs are selected as candidate keywords. Examples include keywords such as alarm occurrence time, alarm level, manufacturer, and device type. Assuming a frequency threshold of 10, if a keyword appears in more than 10 sample alarm messages, then these keywords can be identified as candidate keywords.
[0111] Step 420: For each candidate keyword, calculate the average difference of the value of that candidate keyword.
[0112] For example, if the candidate keyword "alarm occurrence time" appears 18 times, there are 18 possible values for "alarm occurrence time". For each of these 18 values, the pairwise difference is calculated. Then, the average of these multiple differences is calculated to obtain the average difference.
[0113] To calculate the difference between any two values, you can first extract the word vectors (i.e., features) corresponding to each of the two values, and then calculate the cosine similarity between the two word vectors. Finally, subtract the cosine similarity value from 1 to obtain the difference between any two values.
[0114] Step 430: For each candidate keyword, calculate the average character length of the candidate keyword's value.
[0115] For example, if the candidate keyword "alarm occurrence time" appears 18 times, there are a total of 18 values for "alarm occurrence time". For each of these 18 values, the character length of each value is calculated, and the average of these 18 character lengths is taken to obtain the average character length of the candidate keyword "alarm occurrence time".
[0116] Step 440: For each candidate keyword, determine the ratio of the average difference degree to the average character length of each candidate keyword.
[0117] Divide the average difference of each candidate keyword by the corresponding average character length to obtain the target ratio of each candidate keyword.
[0118] For example, for the candidate keyword "alarm occurrence time", the target ratio of "alarm occurrence time" is obtained by dividing the average difference of "alarm occurrence time" by the average character length of "alarm occurrence time".
[0119] Then, for the candidate keyword "alarm level", the target ratio for "alarm level" is obtained by dividing the average difference of "alarm level" by the average character length of "alarm level". This process is repeated to obtain the target ratio for each candidate keyword.
[0120] Step 450: For each candidate keyword, if the average difference in its value is small, it means that the candidate keyword is not easy to distinguish different samples. If the average difference in its value is small, but the average character length of its value is also small, then the candidate keyword can distinguish different samples.
[0121] For example, for the candidate keyword "alarm occurrence time", the corresponding values are usually: 2021-03-19 17:32:02, 2021-03-19 17:36:45, 2021-03-21 18:54:09, etc. The average difference between these values is small, and the average character length of these values is also very long. Therefore, "alarm occurrence time" is not easy to distinguish different samples.
[0122] For example, for the keyword "alarm level," the corresponding values are usually: Level 1, Level 2, Level 3, and Level 4, etc. The average difference between these values is small, but the average character length of these values is also very short. Therefore, "alarm level" can relatively easily distinguish different samples.
[0123] Therefore, after determining the ratio of the average difference of candidate keywords to the average character length, a first threshold needs to be set to judge the magnitude of the ratio. When the ratio is less than the first threshold, it is considered that the candidate keyword is difficult to distinguish between different samples, and thus the candidate keyword can be identified as the target keyword, and the target candidate keyword and its corresponding key-value pair are removed. If the ratio is greater than the first threshold, it indicates that the candidate keyword can distinguish between different samples, that is, the candidate keyword is not a target candidate keyword.
[0124] In practice, step 460 is also included, whereby after determining the target keywords, the target keywords and their corresponding key-value pairs can be stored in a dictionary for subsequent text preprocessing. This dictionary can be specifically used to record deleted target keywords and their corresponding key-value pairs.
[0125] The work order recognition method provided in this application can reduce the redundancy of training samples, reduce the input dimension of the target classification model, shorten the training time and training effect of the target classification model, and thus improve the accuracy of the target classification model in classifying and recognizing work orders by removing target keywords from the sample work orders.
[0126] In one embodiment, the target feature vector corresponding to the work order to be identified is input into the feature extraction layer of the target classification model for feature extraction, to obtain the global feature vector and local feature vector corresponding to the target feature vector, including:
[0127] The LDA topic model in the feature extraction layer is used to extract features from the target feature vector to obtain the global feature vector;
[0128] The Long Short-Term Memory (LTSM) network model in the feature extraction layer is used to extract features from the target feature vector, thus obtaining local feature vectors.
[0129] When using deep learning algorithms to classify text, the quality of feature extraction has a significant impact on the performance of the target classification model. Traditional extraction methods cannot effectively solve the text classification problem. In this embodiment, the feature extraction layer of the target classification model can include the following two models: Latent Dirichlet Allocation (LDA) topic model and Long Short-Term Memory (LSTM) model. The LDA topic model and the LSTM model can constitute a dual-channel deep topic feature extraction model, that is, the LDA topic model and the LSTM model are used as two channels for feature extraction respectively.
[0130] By using the LDA topic model to model the global features of the target feature vector and the LSTM model to model the local features of the target feature vector, the target classification model can simultaneously express the global and local features of the text corresponding to the work order to be identified. This can achieve an effective combination of supervised and unsupervised learning and obtain text feature extraction at different levels.
[0131] In natural language processing, LSTM models acquire local semantic information of a single text through supervised learning, while LDA topic models acquire implicit topic information in the text through unsupervised learning, thereby obtaining global semantic information for a set of documents.
[0132] LDA (Low-Level Analytical Topic Model) is a feature extraction method based on the bag-of-words model. It ignores word order and contextual information, extracting a global feature vector composed of probability distributions at three levels: document, topic, and word. LDA constructs the topic distribution of a text at the text-level granularity through a probabilistic model, emphasizing the overall semantic expression of the text.
[0133] The LSTM model mines word meanings at the word granular level to perform fine semantic expression of text, and can extract local feature vectors corresponding to the target feature vector.
[0134] like Figure 2 As shown, the LSTM model can also extract classification labels from sample work orders. These classification labels are used to combine with local feature vectors to provide a reference for the classification of fused feature vectors.
[0135] The work order recognition method provided in this application can extract local and global features through deep feature extraction, thereby enabling more in-depth mining of text features corresponding to work orders from different levels. At the same time, it can greatly improve the classification performance of the model, eliminate the interference of human factors, save labor costs, and greatly reduce the possibility of mis-assignment and missed assignment of work orders.
[0136] In one embodiment, the fused feature vector is input into the text classification layer of the target classification model for classification to obtain the target classification corresponding to the work order to be identified, including:
[0137] The fused feature vectors are input into a recurrent neural network (RNN) model in the text classification layer for classification, thus obtaining the target classification.
[0138] Understandably, a text classifier can be used to classify the fused feature vectors to obtain the target classification label. The text classifier can use a recurrent neural network (RNN) model to classify text.
[0139] The target classification is determined based on the trained target classification model.
[0140] For example, if the target classification model is used to classify alarm work orders, the target classification can be: equipment alarm, environmental alarm, processing error alarm, service quality alarm, communication alarm, and software error alarm, etc.
[0141] After determining the target classification, the attribute information of the work order to be identified, such as the work order type, work order content, and work order level, can be determined based on the target classification. In this way, the relevant attribute information of the work order can be dispatched to the corresponding work order handler.
[0142] For example, alarm work orders are categorized and assigned to responsible personnel. Using RPA technology, an SMS notification is sent upon alarm occurrence, based on pre-set alarm levels, with the option to notify by phone. The system automatically triggers dialing based on the pre-set responsible person and mobile phone number. After the call is connected, the text-to-speech system automatically reads the alarm content to the responsible person.
[0143] It should be noted that RNNs are a type of neural network with short-term memory capabilities. Neurons in an RNN can receive information from other neurons as well as information from themselves, thus forming a neural network with loops.
[0144] RNN refers to a structure that repeats itself over time. It has wide applications in various fields of Natural Language Processing (NLP), including speech and image processing.
[0145] The biggest difference between RNNs and other networks is that RNNs can perform a certain "memory function," making them the best choice for time series analysis. RNNs retain some memory of the information they have processed, unlike other types of neural networks that cannot retain memory of processed information.
[0146] The work order recognition method provided in this application uses RNN to classify the fused feature vector, which can automatically and quickly determine the target classification label.
[0147] In one embodiment, before inputting the target feature vector corresponding to the work order to be identified into the feature extraction layer of the target classification model for feature extraction, the method further includes:
[0148] Based on the target candidate keywords, the work orders to be identified are deduplicated;
[0149] The deduplicated work orders to be identified are vectorized to obtain the target feature vector.
[0150] In practice, for ordinary text classification scenarios, text segmentation and stop word removal are essential steps in the preprocessing process.
[0151] Regarding the work order recognition application scenario involved in this application embodiment, since the work order information is text information automatically generated by the system, the work orders have the same format and similar content. Different categories of work orders will have a large amount of repeated word information relative to natural language text, and this repeated information cannot be used as a basis for classification. For example, in the alarm work order recognition application scenario involved in this application embodiment, alarm work orders contain a large amount of repeated word information.
[0152] Commonly used text preprocessing methods are not suitable for the application scenario of work order recognition. Therefore, a text preprocessing method for removing duplicate words is proposed based on this specific application scenario.
[0153] The text preprocessing method involves comparing keywords across all sample work orders after text segmentation to identify target candidate keywords with a repetition rate β greater than a certain threshold. The repetition rate β can be adjusted based on classification accuracy.
[0154] The repetition rate β can be determined using the following formula:
[0155] β = Number of samples containing word x / Total number of samples × 100%.
[0156] In the above embodiments, the target keywords have been stored in the dictionary. For newly generated work orders to be identified, preprocessing is required before inputting them into the target classification model.
[0157] After performing text segmentation on the work order to be identified, the target candidate keywords removed during training are compared with the dictionary saved in the segmented work order text to remove duplicate words. The deduplicated work order to be identified is then input into the target classification model for classification.
[0158] Then, the input text sequence corresponding to the work order to be identified is converted into a vector through Embedding to obtain the target feature vector.
[0159] The work order recognition method provided in this application embodiment can reduce the redundancy of the work orders to be recognized, thereby facilitating the rapid acquisition of recognition results for the work orders to be recognized.
[0160] In one embodiment, the work order identification method provided in this application can be applied to an alarm monitoring automatic dispatch system.
[0161] The work order identification method is explained below with reference to specific embodiments.
[0162] In the telecommunications industry, mobile communication network failures are inevitable. Timely detection of alarms and notification of relevant personnel are prerequisites for quickly resolving these failures.
[0163] The sources of data for mobile communication network faults are extremely diverse. Due to the sheer size and complexity of the system, the maintenance workload is enormous, involving thousands of servers and tens of thousands of software programs. Tens of thousands of alarms are generated daily, with thousands requiring telephone notification. Alarms are generated around the clock; they can occur at any time. Maintenance personnel receive hundreds of alarms daily, and often some important alarms are buried among the general ones. This provides a data foundation for automated dispatching based on big data.
[0164] One of the key components of an automated alarm monitoring and dispatch system is the collection and labeling of historical alarm data from various channels. This known alarm data, along with corresponding target classification labels, is used to train a model, enabling the target classification model to automatically categorize alarm work orders. By combining the alarm classification labels predicted by the target classification model, the system can assign the appropriate responsible party and automatically send a notification.
[0165] The prerequisite for automatic dispatch of alarm work orders is the ability to automatically classify alarm work orders.
[0166] The work order identification method provided in this application embodiment classifies alarm work orders based on a target classification model using deep topic feature extraction and RNN algorithm, and provides classified reminders according to alarm level.
[0167] In practice, the alarm categories and levels of mobile communication network incidents can be freely defined, making them highly adaptable.
[0168] like Figure 5 As shown, Figure 5 This is an overall structural diagram of the work order identification method provided in the embodiments of this application.
[0169] Reference Figure 5 By utilizing a large number of existing alarm work orders, the text sets corresponding to the alarm work orders are processed by LDA topic model and LSTM model to extract global and local features, and the target text features are obtained by concatenating them bit by bit.
[0170] RNNs are used to classify the target text features, resulting in the corresponding target category. Each target category is then associated with a responsible person. RPA technology is then used to send an SMS notification to the responsible person, based on a pre-set alarm level, and with the option of a phone call. The system automatically triggers a call based on the pre-set responsible person and phone number. Once the call is connected, a text-to-speech system automatically reads the alarm content to the responsible person.
[0171] This application embodiment can automatically match the responsible person and send SMS, email and telephone notifications to relevant operation and maintenance personnel according to the alarm content corresponding to the target category. This greatly improves the efficiency of notifying the relevant responsible person after the alarm is generated, and provides the prerequisite for timely handling of faults.
[0172] The embodiments of this application can also be applied to the following scenarios:
[0173] In the field of intelligent voice, based on artificial intelligence RPA technology, text-to-speech conversion can be achieved, and talking robots can replace manual monitoring and work order dispatch.
[0174] Regarding storm filtering, alarm storm filtering is implemented. Because the system generates a large number of alarms during large-scale failures or system cutovers, the system needs to intelligently analyze and merge alarms to ensure timely processing of alarm work orders.
[0175] In terms of intelligent paging, alarm text information from various systems is collected, alarm information is automatically matched with maintenance personnel's telephone numbers, and automatic calls are made. Based on RPA robots for automatic sending, the system achieves automatic alarm push functionality, exhibiting strong adaptability.
[0176] This application embodiment strengthens the integration of business and system resource usage, enabling the system processing capacity to adapt to the needs of business development, ensuring the existing service level while taking into account the rapid response to business development; on the other hand, due to the integration of resources, human factors are reduced, saving manpower input while effectively reducing environmental operating costs, and further improving the level of operation and maintenance management.
[0177] The work order identification device provided in the embodiments of this application is described below. The work order identification device described below can be referred to in correspondence with the work order identification method described above.
[0178] like Figure 6 As described in this application embodiment, a work order recognition device is provided. This device may include: a feature extraction module 610, a feature fusion module 620, and a recognition module 630.
[0179] The feature extraction module 610 is used to input the target feature vector corresponding to the work order to be identified into the feature extraction layer of the target classification model for feature extraction, so as to obtain the global feature vector and local feature vector corresponding to the target feature vector;
[0180] The feature fusion module 620 is used to input the global feature vector and the local feature vector into the feature fusion layer in the target classification model to perform feature fusion and obtain a fused feature vector.
[0181] The recognition module 630 is used to input the fused feature vector into the text classification layer of the target classification model for classification, so as to obtain the target classification corresponding to the work order to be recognized.
[0182] The target classification model is trained based on sample work orders carrying classification labels.
[0183] The work order recognition device provided in this application uses a target classification model to extract and fuse features from the target feature vector corresponding to the work order to be recognized. This allows for the acquisition of fused feature vectors with different meanings and levels, effectively reducing the loss of work order information and enabling automatic classification. This eliminates the influence of human factors on work order processing, ensuring that the target classification label of the work order can be automatically and quickly identified without human intervention. It also improves the accuracy of work order recognition.
[0184] In one embodiment, the target classification model is trained as follows:
[0185] Based on the average difference and average character length of the candidate keywords in the sample work orders, target candidate keywords are determined; the candidate keywords are keywords that appear in the sample work orders with a frequency exceeding a frequency threshold.
[0186] Remove the target candidate keywords from the sample work order to generate the target text sequence;
[0187] The initial classification model is trained based on the feature vector corresponding to the target text sequence and the classification label to obtain the target classification model.
[0188] In one embodiment, the candidate keywords include multiple key-value pairs;
[0189] The average degree of difference is determined in the following manner:
[0190] Determine the word vectors corresponding to any two values in the plurality of key-value pairs;
[0191] The difference between any two values is determined based on the cosine similarity between the word vectors corresponding to any two values.
[0192] The average difference is determined based on the difference between any two values.
[0193] The average character length is determined in the following way:
[0194] The average character length is determined based on the character length of each value in the plurality of key-value pairs.
[0195] In one embodiment, the work order identification device further includes:
[0196] The preprocessing module is used to deduplicatize the work orders to be identified based on the target candidate keywords;
[0197] The deduplicated work orders to be identified are vectorized to obtain the target feature vector.
[0198] In one embodiment, the feature extraction module 610 is further configured to:
[0199] The target feature vector is extracted using the LDA topic model in the feature extraction layer to obtain the global feature vector;
[0200] The target feature vector is extracted using the Long Short-Term Memory (LTSM) network model in the feature extraction layer to obtain the local feature vector.
[0201] In one embodiment, the feature fusion module 620 is further configured to:
[0202] The global feature vector and the local feature vector are fused by means of bitwise concatenation, bitwise addition, or bitwise multiplication to obtain the fused feature vector.
[0203] In one embodiment, the identification module 630 is further configured to:
[0204] The fused feature vector is input into the recurrent neural network (RNN) model in the text classification layer for classification to obtain the target classification.
[0205] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7 As shown, the electronic device may include: a processor 710, a communication interface 720, a memory 730, and a communication bus 740, wherein the processor 710, the communication interface 720, and the memory 730 communicate with each other via the communication bus 740. The processor 710 can call a computer program in the memory 730 to execute the steps of the work order identification method, such as including:
[0206] The target feature vector corresponding to the work order to be identified is input into the feature extraction layer of the target classification model for feature extraction, so as to obtain the global feature vector and local feature vector corresponding to the target feature vector;
[0207] The global feature vector and the local feature vector are input into the feature fusion layer of the target classification model to perform feature fusion and obtain a fused feature vector.
[0208] The fused feature vector is input into the text classification layer of the target classification model for classification to obtain the target classification of the work order to be identified.
[0209] The target classification model is trained based on sample work orders carrying classification labels.
[0210] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0211] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the steps of the work order identification method provided in the above embodiments, such as:
[0212] The target feature vector corresponding to the work order to be identified is input into the feature extraction layer of the target classification model for feature extraction, so as to obtain the global feature vector and local feature vector corresponding to the target feature vector;
[0213] The global feature vector and the local feature vector are input into the feature fusion layer of the target classification model to perform feature fusion and obtain a fused feature vector.
[0214] The fused feature vector is input into the text classification layer of the target classification model for classification to obtain the target classification of the work order to be identified.
[0215] The target classification model is trained based on sample work orders carrying classification labels.
[0216] On the other hand, embodiments of this application also provide a processor-readable storage medium storing a computer program for causing a processor to perform the steps of the methods provided in the above embodiments, such as including:
[0217] The target feature vector corresponding to the work order to be identified is input into the feature extraction layer of the target classification model for feature extraction, so as to obtain the global feature vector and local feature vector corresponding to the target feature vector;
[0218] The global feature vector and the local feature vector are input into the feature fusion layer of the target classification model to perform feature fusion and obtain a fused feature vector.
[0219] The fused feature vector is input into the text classification layer of the target classification model for classification to obtain the target classification of the work order to be identified.
[0220] The target classification model is trained based on sample work orders carrying classification labels.
[0221] Processor-readable storage media can be any available medium or data storage device that the processor can access, including but not limited to magnetic storage (e.g., floppy disks, hard disks, magnetic tapes, magneto-optical disks (MOs), etc.), optical storage (e.g., CDs, DVDs, BDs, HVDs, etc.), and semiconductor storage (e.g., ROMs, EPROMs, EEPROMs, non-volatile memory (NAND flash), solid-state drives (SSDs)).
[0222] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0223] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.
[0224] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.
Claims
1. A work order identification method, characterized in that, include: The target feature vector corresponding to the work order to be identified is input into the feature extraction layer of the target classification model for feature extraction, so as to obtain the global feature vector and local feature vector corresponding to the target feature vector; The global feature vector and the local feature vector are input into the feature fusion layer of the target classification model to perform feature fusion and obtain a fused feature vector. The fused feature vector is input into the text classification layer of the target classification model for classification to obtain the target classification of the work order to be identified. The target classification model is trained based on sample work orders carrying classification labels; The target classification model is trained as follows: based on the average difference and average character length of the candidate keywords in the sample work orders, target candidate keywords are determined; the candidate keywords are keywords in the sample work orders whose frequency exceeds a frequency threshold. Remove the target candidate keywords from the sample work order to generate a target text sequence; train the initial classification model based on the feature vector corresponding to the target text sequence and the classification label to obtain the target classification model; The candidate keywords include multiple key-value pairs; The average difference is determined as follows: the word vectors corresponding to any two values in the plurality of key-value pairs are determined; the difference between the two values is determined based on the cosine similarity between the word vectors corresponding to the two values; and the average difference is determined based on the difference between the two values. The average character length is determined as follows: the average character length is determined based on the character length of each value in the plurality of key-value pairs.
2. The work order identification method according to claim 1, characterized in that, Before inputting the target feature vector corresponding to the work order to be identified into the feature extraction layer of the target classification model for feature extraction, the method further includes: Based on the target candidate keywords, the work orders to be identified are deduplicated; The deduplicated work orders to be identified are vectorized to obtain the target feature vector.
3. The work order identification method according to claim 1 or 2, characterized in that, The step of inputting the target feature vector corresponding to the work order to be identified into the feature extraction layer of the target classification model for feature extraction, to obtain the global feature vector and local feature vector corresponding to the target feature vector, includes: The target feature vector is extracted using the LDA topic model in the feature extraction layer to obtain the global feature vector; The target feature vector is extracted using the Long Short-Term Memory (LTSM) network model in the feature extraction layer to obtain the local feature vector.
4. The work order identification method according to claim 1 or 2, characterized in that, The step of inputting the global feature vector and the local feature vector into the feature fusion layer of the target classification model for feature fusion to obtain a fused feature vector includes: The global feature vector and the local feature vector are fused by means of bitwise concatenation, bitwise addition, or bitwise multiplication to obtain the fused feature vector.
5. The work order identification method according to claim 1 or 2, characterized in that, The step of inputting the fused feature vector into the text classification layer of the target classification model for classification to obtain the target classification corresponding to the work order to be identified includes: The fused feature vector is input into the recurrent neural network (RNN) model in the text classification layer for classification to obtain the target classification.
6. A work order identification device, characterized in that, include: The feature extraction module is used to input the target feature vector corresponding to the work order to be identified into the feature extraction layer of the target classification model for feature extraction, so as to obtain the global feature vector and local feature vector corresponding to the target feature vector; The feature fusion module is used to input the global feature vector and the local feature vector into the feature fusion layer in the target classification model to perform feature fusion and obtain a fused feature vector. The recognition module is used to input the fused feature vector into the text classification layer of the target classification model for classification, so as to obtain the target classification corresponding to the work order to be recognized. The target classification model is trained based on sample work orders carrying classification labels; The target classification model is trained as follows: based on the average difference and average character length of the candidate keywords in the sample work orders, target candidate keywords are determined; the candidate keywords are keywords in the sample work orders whose frequency exceeds a frequency threshold. Remove the target candidate keywords from the sample work order to generate a target text sequence; train the initial classification model based on the feature vector corresponding to the target text sequence and the classification label to obtain the target classification model; The candidate keywords include multiple key-value pairs; The average difference is determined as follows: the word vectors corresponding to any two values in the plurality of key-value pairs are determined; the difference between the two values is determined based on the cosine similarity between the word vectors corresponding to the two values; and the average difference is determined based on the difference between the two values. The average character length is determined as follows: the average character length is determined based on the character length of each value in the plurality of key-value pairs.
7. An electronic device comprising a processor and a memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the work order identification method according to any one of claims 1 to 5.
8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the work order identification method according to any one of claims 1 to 5.