Service packet processing method and device, storage medium, and program product

Abstract

The present application relates to the technical field of data processing, and discloses a service packet processing method and device, a storage medium, and a program product. The service packet processing method comprises: acquiring original service packet data; performing text vectorization processing on the original service packet data to obtain vectorized service packet data; and on the basis of a pre-trained large model, performing label prediction on the vectorized service packet data to obtain labeled service packet data, wherein the pre-trained large model is trained on the basis of an annotated data set generated by an unsupervised word segmentation algorithm, a pre-trained word embedding model, and a stack backtracking algorithm. By means of the solution, the present application can achieve precise identification of service information in service packet data, and effectively improve the efficiency of service packet processing.

Description

Business message processing methods, equipment, storage media, and program products

[0001] Cross-references to related applications

[0002] This application claims priority based on Chinese Patent Application No. 202411679018.9, filed on November 21, 2024, the entire contents of which are incorporated herein by reference. Technical Field

[0003] This application relates to the field of data processing technology, and in particular to a business message processing method, device, storage medium, and program product. Background Technology

[0004] Identifying specific business scenarios from massive amounts of business message data enables enterprises to manage and maintain business services, and helps identify potential service quality issues within these scenarios. Existing business message processing methods typically employ keyword extraction techniques or word segmentation algorithms to filter out keywords or phrases related to specific business scenarios. However, several limitations remain: First, current methods still rely on manual tagging of business scenarios and processes for business messages, which is not only inefficient but also prone to errors. Second, traditional word segmentation algorithms often struggle to accurately capture the complex relationships and contextual dependencies between words, leading to segmentation errors or omissions that affect the accuracy of subsequent business message tagging.

[0005] Therefore, existing business message processing methods suffer from low accuracy and efficiency in business message tagging.

[0006] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0007] The main objective of this application is to provide a business message processing method, device, storage medium, and program product, which aims to solve the technical problem of low accuracy and efficiency in the existing business message processing methods for business message tagging.

[0008] To achieve the above objectives, this application proposes a business message processing method, the method comprising:

[0009] Obtain the original business message data;

[0010] The original business message data is processed into text vectorization to obtain vectorized business message data;

[0011] Based on a pre-trained large model, label prediction is performed on the vectorized business message data to obtain labeled business message data. The pre-trained large model is trained on a labeled dataset generated by an unsupervised word segmentation algorithm, a pre-trained word embedding model, and a stack reverse tracing algorithm.

[0012] In addition, to achieve the above objectives, this application also proposes a service message processing device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the operation of the service message processing method as described above.

[0013] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the operation of the business message processing method described above.

[0014] In addition, to achieve the above objectives, this application also proposes a computer program product, which includes a computer program that, when executed by a processor, implements the operation of the business message processing method described above.

[0015] One or more technical solutions proposed in this application have at least the following technical effects:

[0016] The business message processing method, device, storage medium, and program product proposed in this application specifically involve: acquiring original business message data; performing text vectorization processing on the original business message data to obtain vectorized business message data; and performing label prediction on the vectorized business message data based on a pre-trained large model to obtain labeled business message data. The pre-trained large model is trained on a labeled dataset generated by an unsupervised word segmentation algorithm, a pre-trained word embedding model, and a stack reverse tracing algorithm.

[0017] This application acquires raw business message data, first performs text vectorization processing on the raw business message data, and then inputs the text vectorized business message data into a pre-trained large model. The pre-trained large model is used to predict the labels of relevant business information, thereby realizing the automatic labeling processing of relevant business information in the business message data and effectively improving the accuracy and efficiency of business message labeling processing. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of 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, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 is a flowchart of the service message processing method provided in Embodiment 1 of this application;

[0021] Figure 2 is a flowchart of the second embodiment of the business message processing method of this application;

[0022] Figure 3 is a flowchart of the service message processing method provided in Embodiment 3 of this application;

[0023] Figure 4 is a flowchart of the fourth embodiment of the business message processing method of this application;

[0024] Figure 5 is a logical flowchart of the service coloring method for service messages involved in Embodiment 4 of this application;

[0025] Figure 6 is a schematic diagram of the hardware operating environment involved in the service message processing method in this application embodiment.

[0026] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0027] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0028] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0029] The main solution of this application embodiment is: to obtain the original business message data; to perform text vectorization processing on the original business message data to obtain vectorized business message data; and to perform label prediction on the vectorized business message data based on a pre-trained large model to obtain labeled business message data. The pre-trained large model is trained on a labeled dataset generated by an unsupervised word segmentation algorithm, a pre-trained word embedding model, and a stack reverse tracing algorithm.

[0030] Technical terms used in this application:

[0031] Large models: Large models refer to models trained using massive amounts of data and advanced algorithms and technologies, possessing powerful predictive and decision-making capabilities. Large models typically have hundreds of millions of parameters, are able to capture complex patterns in the data, and capture complex language structures and semantic information, often achieving or even surpassing human performance. Large models can be widely applied in many fields, such as natural language processing, image recognition, financial risk control, and intelligent recommendation.

[0032] Word embeddings are a technique that transforms text data into low-dimensional dense vector representations. These vectors capture semantic information within the text, allowing the similarity and differences between texts to be measured by the distance between vectors. Word embeddings are widely used in various natural language processing tasks, such as information retrieval, text classification, and clustering. Through word embeddings, models can better understand and process text data, improving task performance.

[0033] The tracing algorithm is a search algorithm that is both systematic and leapfrog. It searches the solution space tree containing all solutions to the problem, starting from the root node, using a depth-first strategy. When the algorithm reaches any node in the solution space tree, it always first determines whether that node definitely does not contain a solution to the problem. If it definitely does not contain a solution, it skips the systematic search of the subtree rooted at that node and traces back to its ancestor nodes level by level. Otherwise, it enters that subtree and continues the search using the depth-first strategy.

[0034] Stack: A stack is a data structure that follows the Last-In-First-Out (LIFO) principle, allowing insertion and deletion operations only at one end of an ordered linear data set (called the top of the stack). Stacks have wide applications in computer science, such as function calls, recursive implementations, and expression evaluation.

[0035] Depth-First Search (DFS) is an algorithm used to traverse or search a tree or graph. Starting from a vertex in the graph, DFS visits as many nodes as possible along the edges until that node has no unvisited neighbors. Then, it traces back to the previous node and continues searching for other unvisited neighbors until all nodes in the graph have been visited. There are several ways to implement DFS, including recursion and explicit stacks (for iterative implementation). This algorithm is commonly used for pathfinding, connectivity testing, and topological sorting problems.

[0036] As the basic unit of data transmission, business messages typically contain rich business information. Identifying specific business scenarios from massive amounts of business message data enables enterprises to perform operational and maintenance management of business services and helps to discover potential service quality issues within those scenarios.

[0037] Existing business message processing methods typically employ keyword extraction techniques or word segmentation algorithms to filter out keywords or phrases relevant to specific business scenarios. However, different business processes often follow specific message protocols, requiring manual tagging of business messages based on business scenarios and processes. Furthermore, traditional word segmentation algorithms often struggle to accurately capture the complex relationships and contextual dependencies between words, particularly when processing business messages containing numerous technical terms and phrases, leading to segmentation errors or omissions that affect the accuracy of subsequent tagging.

[0038] Therefore, existing business message processing methods suffer from low accuracy and efficiency in business message tagging.

[0039] This application provides a solution that obtains raw business message data, first performs text vectorization processing on the raw business message data, and then inputs the text vectorized business message data into a pre-trained large model. The pre-trained large model is used to predict the labels of relevant business information, thereby realizing the automatic labeling processing of relevant business information in the business message data, effectively improving the accuracy and efficiency of business message labeling processing.

[0040] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or a business message processing system capable of implementing the above functions. The following description uses a business message processing system as an example to illustrate this embodiment and the subsequent embodiments.

[0041] Based on this, the present application provides a service message processing method. Referring to FIG1, FIG1 is a flowchart of the first embodiment of the service message processing method of the present application.

[0042] In this embodiment, the service message processing method includes operations S110 to S130:

[0043] Operate S110 to obtain the original service message data;

[0044] Understandably, since the technical solution proposed in this application is actually applied to the tagging of relevant business information in business messages, operation S110 is required. First, the business message processing system needs to acquire the original business message data and treat it as a business message to be tagged, so that the large model can be used to understand and identify the business information contained in the original business message data and to annotate the original business message with relevant business information. The original business message data refers to the message data generated by the business system in response to user business requests, which typically includes business scenario information, corresponding business steps, response results of business requests, business status information, user information, etc.

[0045] Operation S120: Perform text vectorization processing on the original service message data to obtain vectorized service message data;

[0046] In some embodiments, based on the aforementioned original business message data, the text in the original business message data can be segmented into words and converted into corresponding word vectors using methods such as bag-of-words model, word embedding model, and deep learning-based sentence vector model, thereby forming a set of word vectors as the vectorized business message data.

[0047] In operation S130, based on a pre-trained large model, label prediction is performed on the vectorized business message data to obtain labeled business message data. The pre-trained large model is trained on a labeled dataset generated by an unsupervised word segmentation algorithm, a pre-trained word embedding model, and a stack reverse tracing algorithm.

[0048] It should be noted that the pre-trained large model is obtained by relevant personnel based on actual business message processing needs, in order to make the large model output more accurate label information of business messages. In this embodiment, the large model can be one or more combinations of large models such as Qwen, Meta Large Language Model (Meta AI), and Wenxin Yiyan.

[0049] It should be noted that the pre-trained word embedding model is a word embedding model that is pre-trained by relevant personnel in the corpus of the relevant industry based on the actual word segmentation requirements of the business message. In this embodiment, the word embedding model can be the Word2Vec model.

[0050] Unsupervised word segmentation algorithms are those that do not require manually labeled data. They are algorithms that use statistical or machine learning methods to segment continuous strings into individual words. Unsupervised word segmentation algorithms utilize features such as word frequency, contextual information, and co-occurrence relationships between characters to automatically identify word boundaries, thereby achieving string segmentation.

[0051] In this embodiment, the stack reverse tracing algorithm is an algorithm built based on the actual link location requirements of business messages, utilizing the 'last-in, first-out' data structure characteristics of the stack, and combining the idea of ​​tracing algorithms. It is used to locate the link request path of business messages in order to obtain the message parsing results of the business messages.

[0052] In some embodiments, the vectorized business message data obtained above is input into a pre-trained large model. The large model then predicts and infers relevant business information from the vectorized business message data, providing label information that can be used for annotation. This label information is then used to annotate the business message data, resulting in tagged business message data. The label information includes business scenario information, business process information, and business status information associated with the current business message data.

[0053] Furthermore, in order to use the large model to predict the labels of relevant business information in the business message data, the large model needs to be trained and optimized in advance. Therefore, before the above operation S130, operations A01 to A04 may also be included:

[0054] Operation A01 generates a labeled dataset based on historical business message data using an unsupervised word segmentation algorithm, a pre-trained word embedding model, and a stack reverse tracing algorithm.

[0055] In some embodiments, the business message processing system first needs to collect historical business message data generated in a certain industry, and then use unsupervised word segmentation algorithm, pre-trained word embedding model and stack reverse tracing algorithm to perform text vectorization and automatic annotation on the historical business message data, thereby generating a dataset with label information, i.e., a labeled dataset.

[0056] Operation A02 divides the labeled dataset into a training set, a validation set, and a test set;

[0057] Operation A03, based on the training set, trains the service message tag prediction capability of the large model to obtain a pre-trained large model; and / or

[0058] Operation A04 involves evaluating the performance of the pre-trained large model based on the validation and test sets, and optimizing the model based on the performance evaluation results to obtain a pre-optimized large model.

[0059] Then, in order to train and optimize the large model, the business message processing system needs to divide the labeled dataset into training set, validation set and test set.

[0060] The business message processing system trains the large model's business message tag prediction capability based on the partitioned training set. This allows the large model to learn patterns and features of business message data through the training set. By setting preset recognition accuracy and recognition time metrics, the system fine-tunes the large model's parameters, such as weight parameters, bias parameters, and layer normalization parameters in a neural network model. This results in a large model that meets the preset recognition accuracy and recognition time requirements—a pre-trained large model. It should be understood that the preset recognition accuracy and recognition time can be set by relevant personnel based on actual needs or experience, and are not specifically limited here.

[0061] Furthermore, to further ensure the business message labeling processing capability of the pre-trained large model, the business message processing system can utilize a validation set to evaluate the model's generalization ability on unseen data. This allows for the adjustment of the model's hyperparameters, such as learning rate and batch size, to prevent overfitting and better adapt the model to new data. This ensures the model's recognition accuracy and generalization ability on the test set. The business message processing system then uses the test set to ultimately evaluate the model's performance and generalization ability. Confirming the model's performance on the test set reflects its performance in real-world application scenarios.

[0062] It should be understood that when optimizing and adjusting the large model using the validation set and test set, the performance and generalization ability of the large model can be evaluated by using the loss function or preset indicators such as recognition accuracy and recognition time. By continuously reducing the loss function or reaching the preset indicators, the label prediction ability of the large model for business message data can be continuously optimized, thereby obtaining a pre-optimized large model. The pre-optimized large model can then be used to label the business message data.

[0063] Understandably, when large models are applied to the labeling of business message data in different domains, validation sets can be used to compare the performance of different combinations of models on unseen data, helping to select the best large model.

[0064] This embodiment provides a business message processing method, which involves acquiring original business message data; performing text vectorization processing on the original business message data to obtain vectorized business message data; and performing label prediction on the vectorized business message data based on a pre-trained large model to obtain labeled business message data. The pre-trained large model is trained on a labeled dataset generated by an unsupervised word segmentation algorithm, a pre-trained word embedding model, and a stack reverse tracing algorithm.

[0065] This application, through the above-mentioned scheme, obtains the original business message data, first performs text vectorization processing on the original business message data, and then inputs the text vectorized business message data into a pre-trained large model. The pre-trained large model is used to predict the labels of relevant business information, thereby realizing the automatic labeling processing of relevant business information in the business message data, effectively improving the accuracy and efficiency of business message labeling processing.

[0066] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description and will not be repeated hereafter. Based on this, please refer to Figure 2, the operation A01 further includes operations S210 to S230:

[0067] Operation S210: Based on the unsupervised word segmentation algorithm, perform word segmentation ambiguity processing on the historical service message data to obtain the first service message data;

[0068] In some embodiments, the business message processing system first needs to automatically identify data features and semantic information such as word frequency, context information, and co-occurrence relationship between characters in historical business message data based on an unsupervised word segmentation algorithm, thereby automatically identifying word boundaries, handling ambiguities caused by different word segmentation methods, and further segmenting continuous strings into individual words to achieve word segmentation ambiguity processing of historical business message data, and obtaining the word segmentation ambiguity processed business message data, i.e., the first business message data.

[0069] It should be understood that when using unsupervised word segmentation algorithms to handle word segmentation ambiguity in historical business message data, in order to further improve the word segmentation accuracy, the historical business message data can be preprocessed, such as data cleaning and data normalization, to remove special characters, punctuation marks, Hyper Text Markup Language (HTML) tags, etc. from the business message text.

[0070] Operation S220: Based on the pre-trained word embedding model, the first service message data is processed into text vectorization to obtain the second service message data;

[0071] Understandably, business message data is often in text format or other non-numerical format, while large models usually recognize numerical data. Therefore, it is necessary to convert the labeled dataset into numerical data so that the large model can be trained and optimized using the labeled dataset.

[0072] In some embodiments, the business message processing system can segment the text in the first business message data obtained after word segmentation and convert it into corresponding word vectors through a pre-trained word embedding model, thereby forming a set of word vectors. The vectorized first business message data is then used as the second business message data.

[0073] In one implementable manner, operation S220 includes operations B01 to B03:

[0074] Operation B01 loads a pre-trained word embedding model and a pre-defined set of word vectors.

[0075] Operation B02 performs data preprocessing and text segmentation on the first service message data to obtain the segmented first service message data.

[0076] Understandably, after pre-training a word embedding model, a set of word vectors corresponding to the corpus of the relevant industry can be obtained, i.e., a pre-set set of word vectors. This set of word vectors can be used to achieve more accurate recognition of specific industry terms. Alternatively, the pre-set word vectors can also be a set of word vectors related to an open-source, pre-trained word embedding model, such as the open-source Word2Vec model.

[0077] In some embodiments, the business message processing system first needs to load a pre-set set of word vectors from a pre-trained word embedding model, since word embedding models are usually stored in binary format and can be loaded using a corresponding library (such as the gensim library).

[0078] Then, before performing text vectorization on the first business message data, it needs to be preprocessed to conform to the word embedding model processing standard. In some embodiments, the business message processing system needs to perform data preprocessing operations on the first business message data, including removing stop words, special characters, irrelevant characters, etc. from the business message data, and then perform text segmentation processing on the first business message data to divide the text in the first business message data into words or sub-words, thereby obtaining the segmented first business message data.

[0079] Operation B03 uses a pre-trained word embedding model and a preset word vector set to perform text vectorization processing on the segmented first business message data to obtain the second business message data.

[0080] Finally, through a pre-trained word embedding model, text features are extracted and semantic information is recognized from the words or sub-words in the segmented first business message data. This allows us to find word vectors that are semantically similar or close to the words or sub-words in the segmented first business message data. These word vectors are then used to construct the vectorized business message data, i.e., the second business message data.

[0081] In this embodiment, a pre-trained word embedding model can be used to vectorize business message data, which facilitates subsequent large models to use the vectorized business message data to predict relevant business information of the business message, thereby improving the accuracy and efficiency of business message tagging.

[0082] Operation S230 involves labeling the second service message data based on the stack reverse tracing algorithm and a preset message data vector set to obtain a labeled dataset.

[0083] In some embodiments, the business message processing system performs similarity matching between a preset message data vector set and second business message data. When the similarity reaches a preset threshold, a preliminary predicted label can be obtained. The preset message data vector set is a message data vector set pre-constructed by relevant personnel based on known tagged message data; the preset threshold refers to the vector similarity threshold information set by relevant personnel based on actual needs or experience.

[0084] Simultaneously, a stack reverse tracing algorithm is used to locate the link request path of the business message based on the historical business message data corresponding to the second business message data. Then, relevant business information, including but not limited to business scenario information, business process information, and business status information, is retrieved based on the link request path. The business message processing system uses this relevant business information and the initially predicted labels to further supplement the second business message data with label information, thereby obtaining an automatically labeled dataset.

[0085] This embodiment provides a business message processing method. It involves: performing word segmentation ambiguity processing on historical business message data using an unsupervised word segmentation algorithm to obtain first business message data; performing text vectorization processing on the first business message data based on a pre-trained word embedding model to obtain second business message data; labeling the second business message data using a stack reverse tracing algorithm and a preset message data vector set to obtain a labeled dataset; dividing the labeled dataset into a training set, a validation set, and a test set; training the business message label prediction capability of a large model based on the training set to obtain a pre-trained large model; and / or evaluating the model performance of the pre-trained large model based on the validation set and the test set, and optimizing the model based on the performance evaluation results to obtain a pre-optimized large model; acquiring original business message data; performing text vectorization processing on the original business message data to obtain vectorized business message data; and performing label prediction on the vectorized business message data based on the pre-trained large model to obtain labeled business message data.

[0086] This embodiment constructs a labeled dataset using the above-described scheme, employing an unsupervised word segmentation algorithm, a pre-trained word embedding model, and a stack reverse tracing algorithm. The labeled dataset is then used to train a large model, resulting in a pre-trained large model. This model is then used for the automatic labeling of business message data, addressing the inefficiency of manual labeling of business messages in existing technologies and the low accuracy of traditional word segmentation algorithms. This effectively improves the accuracy and efficiency of business message labeling.

[0087] Based on the second embodiment of this application, in the third embodiment of this application, the content that is the same as or similar to that in the second embodiment described above can be referred to the above description and will not be repeated hereafter. Based on this, please refer to Figure 3, the operation S210 may further include operations S310 to S330:

[0088] Operation S310: Based on the preset dictionary segmentation strategy, perform preliminary segmentation on the historical service message data to obtain several maximum intersection ambiguity fields;

[0089] Operation S320 calculates the mutual information and boundary entropy of the maximum intersection ambiguity field based on the unsupervised word segmentation algorithm, and obtains the mutual information value and boundary entropy value of the maximum intersection ambiguity field.

[0090] In this embodiment, the unsupervised word segmentation algorithm is a word segmentation ambiguity handling algorithm based on multi-character mutual information and left and right boundary entropy as word segmentation measurement indicators. The unsupervised word segmentation algorithm includes a preset dictionary word segmentation strategy and a preset word segmentation ambiguity handling strategy.

[0091] It should be noted that the preset dictionary-based word segmentation strategy is an initial word segmentation strategy pre-set by relevant personnel based on actual word segmentation needs. This strategy is used to find the most overlapping ambiguous fields in historical business message data for initial word segmentation. It can be based on the maximum matching method or bidirectional maximum matching method, etc. In this embodiment, the preset dictionary-based word segmentation strategy is a combination of the forward maximum matching method and a one-word fallback mechanism.

[0092] In some embodiments, the business message processing system first needs to employ a dictionary-based forward maximum matching method. First, based on the length of the longest string in the dictionary, word matching is performed on the business message data in a forward order, such as from left to right. When a match is successful, the string is segmented; when a match fails, a backsliding mechanism is used, reducing the length of the failed match by one character and re-matching until all strings in the message text have been segmented. It should be understood that, in order to identify as many maximum intersection ambiguities as possible, when using the dictionary-based forward maximum matching method, if a word is successfully segmented from a string, the system can backslid one character from the left end of the successfully matched string and continue searching for words using the dictionary-based forward maximum matching method.

[0093] When there are overlapping sections in the strings after the above segmentation, the overlapping sections are combined to obtain several maximum overlap ambiguous sections.

[0094] It should be understood that the dictionaries mentioned above may be terminology dictionaries for specific industry fields pre-set by relevant personnel.

[0095] The maximum intersection ambiguity field refers to the longest intersection-type segmentation field in a text that cannot be further decomposed. The definition of the maximum intersection ambiguity field is as follows: Let string S = c1 c2…cn (S is not a word; c1, c2, …, cn are Chinese characters), string Smax = ci…cj is a substring of S (1≤i≤j≤n), and Smax is an intersection-type ambiguity segmentation field. If there is no larger intersection-type ambiguity segmentation field in S that contains Smax, then Smax is called the maximum intersection ambiguity field of S. For example, in the text "He devoted himself to promoting Putonghua," "Putonghua," "phone fees," "dedicated himself," "dedicated himself," and "dedication" can all be words and overlap. Therefore, "Putonghua phone fees" and "Putonghua devoted himself to dedicating ...

[0096] Then, it is determined that there may be strings corresponding to different segmentation methods in the maximum intersection ambiguity field. Based on the calculation formulas for multi-word mutual information and left and right boundary entropy, the multi-word mutual information and left and right boundary entropy of the strings corresponding to the aforementioned possible segmentation methods are calculated, resulting in mutual information values ​​and left and right boundary entropy values ​​corresponding to several different segmentation methods of the maximum intersection ambiguity field. It should be understood that the mutual information and boundary entropy of the maximum intersection ambiguity field in operation S320 refer to the mutual information values ​​and left and right boundary entropy values ​​corresponding to different segmentation methods of the maximum intersection ambiguity field. The multi-word mutual information calculation formula is as follows:

[0097] Where, x1…n It refers to a string of length n.

[0098] In the above formula, this application extends the method for calculating mutual information of multi-character strings. Traditional mutual information calculation methods are only applicable to the evaluation of the associativity of two Chinese characters. In this application, the calculation of multi-character mutual information requires first calculating the mutual information of each pair of adjacent characters in a multi-character word, and then taking the average of the total mutual information. Using the above method for calculating multi-character mutual information can more accurately reflect the word formation probability of multi-character words in a corpus, thereby improving the accuracy and reliability of word segmentation.

[0099] The formulas for calculating the entropy of the left and right boundaries are as follows:

[0100] In the above formula, when x is x i-1 When x is x i…j The set of left adjacent characters is called the conditional entropy of the left character (or simply left entropy, denoted as hl(x)). i…j When x is x j+1 When x is x i…j The set of right adjacent words is called the conditional entropy of the right word (or simply right entropy, denoted as hr(x)). i…j In the above formula, lb refers to the logarithm with base 2, and p(x|x) i…j ) represents the conditional probability.

[0101] In the above formula, this application introduces boundary entropy to measure the probability that a string is a word in the corpus. Considering the diversity of the left and right neighbor sets of a string, by calculating the left and right entropies of each string in the corpus and taking their average as a metric, the potential probability of the string being a word can be assessed. The larger the left and right entropies of a string, the more likely it is to be a word.

[0102] It should be understood that the corpus mentioned above is a corpus of a specific industry or professional field that has been pre-set by relevant personnel.

[0103] Operation S330: Based on the preset word segmentation ambiguity handling strategy, the maximum intersection ambiguity field is segmented according to the mutual information value and the boundary entropy value to obtain the first service message data.

[0104] It should be noted that, in this embodiment, the preset word segmentation ambiguity handling strategy refers to a combination of word segmentation methods, including string frequency segmentation, mutual information value, and left and right boundary entropy values.

[0105] In some embodiments, a comprehensive index of mutual information value and left and right boundary entropy value corresponding to different segmentation methods of the maximum intersection ambiguity field is first determined. The comprehensive index is an index obtained by weighting the mutual information value and left and right boundary entropy value of the string according to a certain weight ratio. The weight ratio is set by relevant personnel based on actual word segmentation needs or experience.

[0106] Then, it is determined whether the comprehensive index corresponding to different segmentation methods of the maximum intersection ambiguity field exceeds the preset index threshold. When the comprehensive index exceeds the preset index threshold, the word segmentation method based on mutual information value and left and right boundary entropy value is used. Specifically, the segmentation method with the highest comprehensive index is found from the comprehensive indexes corresponding to different segmentation methods of the maximum intersection ambiguity field, and this is used as the final text segmentation method.

[0107] It should be noted that the preset indicator threshold refers to the threshold set by relevant personnel based on actual conditions or expert experience, or based on empirical values ​​from multiple large-scale model validations, and this application does not impose any specific limitations.

[0108] When the overall index does not exceed the preset threshold, the string frequency segmentation method is used. Specifically, first, the words belonging to the dictionary in the strings corresponding to different segmentation methods of the maximum intersection ambiguity field are determined. Then, the sum of the frequencies of each word in the corpus under different segmentation methods is counted, i.e., the string frequency under different segmentation methods. Finally, a decision on how to segment the words is made based on the string frequency corresponding to the different segmentation methods of the maximum intersection ambiguity field. When the string frequency of a segmentation method is significantly higher than that of other possible segmentation methods, the method with the higher string frequency is adopted as the final segmentation result.

[0109] For example, consider the maximum intersection ambiguity field "abc", where a, b, and c are all non-empty strings, and ab and bc are words in the dictionary. Then "abc" can be segmented in two ways: ab / c and a / bc. If the frequency of "ab" in the corpus is greater than the frequency of "bc", then the segmentation result is ab / c. It should be understood that the frequency of individual characters is not included in the statistics.

[0110] This embodiment provides a business message processing method. It performs preliminary word segmentation on historical business message data using a preset dictionary word segmentation strategy to find the field with the largest intersection ambiguity. Then, it combines mutual information, boundary entropy, and the preset word segmentation ambiguity processing strategy to perform word segmentation semantic processing on the field with the largest intersection ambiguity, thereby realizing word segmentation processing of business message data, effectively improving word segmentation accuracy, and laying the foundation for subsequent automated tagging processing of business message data.

[0111] Based on the second and / or third embodiments of this application, in the fourth embodiment of this application, the content that is the same as or similar to the above-described embodiments two and / or three can be referred to the above description and will not be repeated hereafter. Based on this, please refer to Figure 4. The historical service message data has a service request identifier, and operation S230 includes operations S410 to S440:

[0112] Operation S410: Based on the service request identifier, call the full-link service information to construct the full-link directed acyclic graph (DAG);

[0113] Operation S420, based on the depth-first search algorithm, locates key message nodes in the full-link DAG and obtains the service information of the historical service message data at the key message nodes;

[0114] It should be noted that the business request identifier refers to the specific request ID used to identify a particular business request in the business system. This request ID is globally unique throughout the entire link path of the business system. In this embodiment, the business request identifier refers to the traceId in distributed tracing technology. The traceID is a unique identifier used to track requests in a distributed system.

[0115] In some embodiments, the business message processing system first needs to construct the full-link directed acyclic graph (DAG) of the business system in real time by calling the global link data of the business system based on the business request identifier.

[0116] Then, based on the business request identifier, the request path is traced back in the entire DAG using a depth-first search algorithm to determine the key message nodes in the request path of the business request, and the message data corresponding to the key message nodes of the historical business message data is obtained. The message data is then parsed to obtain the business information corresponding to the key message nodes of the historical business message data.

[0117] In one implementable manner, operation S420 includes operations C01 to C02:

[0118] Operation C01: If the current message node is abnormal, then according to the depth-first search algorithm and the service request identifier, the reverse tracing of the link is invoked to perform a reverse search from the current message node in the full-link DAG until the key message node is found.

[0119] In some embodiments, starting from the current message node, the process traces backwards to the upstream critical message node. Nodes along the traced path are pushed onto a stack in sequence, forming a call order from top to bottom of the stack.

[0120] In some embodiments, when it is impossible to trace back to the critical message node upstream, the current message node is placed in a delay queue corresponding to the service request identifier. In response to detecting a critical message node with the corresponding service request identifier, a retry operation is triggered in the delay queue. Specific implementations of the delay queue and retry operation will be described below.

[0121] Operation C02 obtains the corresponding message parsing result of the historical service message data at the key message node, and obtains the service information.

[0122] In some embodiments, service information is propagated downwards from key message nodes along the direction of the DAG.

[0123] In some embodiments, if the parsing result of the historical service message data at the current message node is empty, that is, the current message node is abnormal, the service message processing system needs to use a depth-first search algorithm to trace the reverse link according to the service request identifier.

[0124] First, mark the current packet node as visited. Then, recursively visit each unvisited adjacent packet node associated with the current packet node in the full-link DAG to confirm whether there are critical packet nodes. Recursively visiting adjacent packet nodes ensures that all possible link paths are completely traversed.

[0125] If the critical message node is still not found after all adjacent message nodes have been visited, it is necessary to trace back to the upper-level message nodes according to the end-to-end DAG. Each upper-level message node is visited, and if it is not found there, its adjacent message nodes are visited for confirmation. This process is repeated until the critical message node is found. The critical message node mentioned above refers to the message data for which the parsing result for the historical service message data exists, and which contains scenario information. It should be understood that the critical message node corresponding to each historical service message data is different.

[0126] Once the key message nodes are found using the above method, the corresponding message of the historical service message data at the key message nodes is determined, and its message parsing result is obtained. Then, based on the relevant service information corresponding to the message parsing result, the historical service message data is colorized, that is, the relevant service information obtained from the above parsing is used as a tag and labeled to the historical service message data, thereby realizing the service message tagging processing of the historical service message data.

[0127] In this implementation, a depth-first search method and a full-link DAG are used to achieve reverse tracing of the link. By invoking the reverse tracing of the link, it is ensured that for a certain historical service message data, after the message is successfully parsed at any message node on its requested link path, the relevant service information of the historical service message data can be obtained, thereby realizing the automatic synchronization and annotation of service information.

[0128] For example, referring to Figure 5, Figure 5 is a logical flowchart of an exemplary service coloring method for service messages.

[0129] In Figure 5, the business steps involved in the business request include business step A, business step B, business step C, and business step D. The traceId is consistent across different business steps in the call chain. The spanId in the call chain identifies the position of a remote procedure call within the distributed request. Each business step generates a spanId for the request after it arrives at it. The parentSpanId identifies the position of the upstream remote procedure call within the distributed request. When the business request arrives at each business step, the spanId of the upstream business step, which is passed along with the business request, is recorded as the parentSpanId.

[0130] In Figure 5, when business link D, i.e. the current message node, cannot parse the message, the call link is traced backwards using the depth-first search method and the full-link DAG until it is traced back to business link A. The business information in the message parsing result of A is then used to color the message data corresponding to the downstream business link.

[0131] Operation S430: Based on the preset message data vector set, perform initial label prediction on the second service message data to obtain the initial predicted label;

[0132] In operation S440, the second service message data is labeled according to the service information and the initial predicted label to obtain a labeled dataset.

[0133] In some embodiments, the business message processing system performs similarity matching between a preset message data vector set and second business message data. When the similarity reaches a preset threshold, a preliminary predicted label, i.e., an initial predicted label, can be obtained. The preset message data vector set is a message data vector set pre-constructed by relevant personnel based on known tagged message data; the preset threshold refers to the vector similarity threshold information set by relevant personnel based on actual needs or experience.

[0134] Then, the business message processing system uses the aforementioned business information, combined with the initial predicted labels, to further supplement the label information of the second business message data, thereby obtaining an automatically labeled dataset.

[0135] Referring to Table 1, which uses the business scenario of a business request to recharge traffic as an example, Table 1 shows a specific example of a business request and its specific business steps.

[0136] Table 1

[0137] Combining Table 1 and Figure 5, in the business request for traffic recharge, business links A, B, C and D share the same TraceID, for example, TraceID = 12345 in Figure 5, which serves as a globally unique identifier for the entire business request.

[0138] Each business process has a unique spanID in the call chain of the business request, which identifies the position of each business process in the business request. As shown in Figure 5, the spanID corresponding to business process A is aaa, the spanID corresponding to business process B is bbb, the spanID corresponding to business process C is ccc, and the spanID corresponding to business process D is ddd.

[0139] `parentSpanID` can be used to identify the upstream business link of the current business link in the call chain of the business request, to represent the hierarchical relationship between different business links and form a call relationship. As shown in Figure 5, the `parentSpanID` of business link A is null, indicating that it has no upstream business link. The `parentSpanID` of business link B is the `spanID` of business link A, indicating that the upstream business link of business link B is business link A. The cases for business links C and D are similar and will not be elaborated further here.

[0140] The full-link DAG or chain corresponding to the traffic recharge service request is: Business link A (aaa) → Business link B (bbb) → Business link C (ccc) → Business link D (ddd). In this disclosure, a directed acyclic graph (DAG) can be constructed by calling and / or parsing information such as TraceID, spanID, and parentSpanID. This disclosure can call the marked business scenario information in the DAG in reverse direction, that is, trace the business scenario information in reverse.

[0141] When performing reverse tracing, the operation can begin from the current message node. Taking the example of business link D in Table 1 encountering an anomaly and being unable to parse the message, business link D can be placed on the stack. Based on the parentSpanID of business link D, the operation is traced back to its upstream business link C, which is also pushed onto the stack. This reverse tracing operation is repeated sequentially, with business links B and A also being pushed onto the stack. At this point, the stack structure is: stack top - business link A (aaa) → business link B (bbb) → business link C (ccc) → business link D (ddd) – stack bottom. The stack follows the "last-in, first-out" principle. During the reverse tracing process, if a colored business link is found at the top node of the stack (such as business link A), the business information is propagated downwards along the call chain DAG starting from that business link to complete the coloring operation for the downstream business links. Similarly, if a colored business link is found in another business link, such as business link B, business link A does not need to be pushed onto the stack. Instead, the business information can be propagated downwards along the call chain DAG starting from the colored business link B to complete the coloring operation for the downstream business links.

[0142] In distributed request systems, call chain data often arrives out of order, and some business processes may lack business messages, making it impossible to directly parse the business scenario. Therefore, the technical solution disclosed herein addresses this issue by establishing a delayed queue.

[0143] In some embodiments, for a newly received business segment, taking business segment D in Table 1 and Figure 5 as an example, if no business information is found using the stack reverse tracing algorithm described above, business segment D is placed in the first delay queue for subsequent processing. The delay queue is managed by grouping according to TraceID. For example, the first delay queue is used to store business segments with TraceID as the first TraceID. When other business segments with TraceID as the first TraceID are received, and also have no business information, these business segments can also be placed in the same delay queue. When business segment A with business information is detected, and business segment A corresponds to the first TraceID, the retry mechanism of the first delay queue can be triggered. The stack reverse tracing algorithm described above is executed on the business segments in the first delay queue to complete the business coloring process.

[0144] In some embodiments, when a business process is placed into a first delay queue, a preset retry timestamp can be set for each entry in the queue. Then, a scheduled task can periodically attempt to color the business processes in the first delay queue. After each coloring attempt, the attempt count for the corresponding business process message can be incremented. If the attempt count reaches a preset threshold (e.g., 5 times) and coloring fails, the message is discarded, thereby ensuring the controllability of system resources and processing efficiency.

[0145] This embodiment provides a business message processing method that achieves reverse tracing of the call chain through a depth-first search method and a full-link DAG. By reverse tracing the call chain, it ensures that for a certain historical business message data, after normal message parsing at any message node on its request chain path, the relevant business information of the historical business message data can be obtained. This enables automatic synchronization and labeling of business information, effectively solving the problem of not being able to obtain business information when the current message node is abnormal, and laying the foundation for automated labeling of subsequent business message data. Given the inherent uncertainty of out-of-order arrival of call chain data in distributed systems, this algorithm also proposes a delayed processing mechanism. Messages of business links that cannot be parsed temporarily are stored in a delayed processing queue, and the business links in the delayed processing queue are re-attempted for labeling periodically or by listening to business links with the same TraceID. This disclosure significantly improves the labeling accuracy and system robustness in asynchronous and out-of-order environments.

[0146] This application provides a service message processing device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the service message processing method in the above embodiment.

[0147] Referring now to Figure 6, a schematic diagram of a service message processing device suitable for implementing embodiments of this application is shown. The service message processing device in embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. The service message processing device shown in Figure 6 is merely an example and should not impose any limitations on the functionality and scope of use of embodiments of this application.

[0148] As shown in Figure 6, the service message processing device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 1002 or a program loaded from storage device 1003 into random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the service message processing device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus 1005. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the message processing equipment to communicate wirelessly or wiredly with other devices to exchange data. Although a message processing equipment with various systems is shown in the figure, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems may be implemented alternatively.

[0149] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 1009, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0150] The service message processing device provided in this application, employing the service message processing method described in the above embodiments, can solve the technical problem of low accuracy and efficiency in the existing service message processing methods for service message tagging. Compared with the prior art, the beneficial effects of the service message processing device provided in this application are the same as those of the service message processing method provided in the above embodiments, and other technical features of this service message processing device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0151] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0152] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0153] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the service message processing method described in the above embodiments.

[0154] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. Further examples of the computer-readable storage medium in some embodiments may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (Radio Frequency), etc., or any suitable combination thereof.

[0155] The aforementioned computer-readable storage medium may be included in the service message processing device; or it may exist independently and not be assembled into the service message processing device.

[0156] The aforementioned computer-readable storage medium carries one or more programs. When the aforementioned one or more programs are executed by the service message processing device, the service message processing device causes the following: to acquire raw service message data; to perform text vectorization processing on the raw service message data to obtain vectorized service message data; and to perform label prediction on the vectorized service message data based on a pre-trained large model to obtain labeled service message data. The pre-trained large model is trained on a labeled dataset generated by an unsupervised word segmentation algorithm, a pre-trained word embedding model, and a stack reverse tracing algorithm.

[0157] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0158] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0159] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0160] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described service message processing method. This solves the technical problem of low accuracy and efficiency in the existing service message processing methods for service message tagging. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the service message processing method provided in the above embodiments, and will not be repeated here.

[0161] This application provides a computer program product, including a computer program that, when executed by a processor, implements the operation of the business message processing method described above.

[0162] The computer program product provided in this application can solve the technical problem of low accuracy and efficiency in the tagging of business messages in existing business message processing methods. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as the beneficial effects of the business message processing methods provided in the above embodiments, and will not be repeated here.

[0163] The above are only some embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural changes made based on the content of this application’s specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent scope of this application.

Claims

1. A method for processing business messages, wherein, The method includes: Obtain the original business message data; The original business message data is processed into text vectorization to obtain vectorized business message data. Based on a pre-trained large model, label prediction is performed on the vectorized business message data to obtain labeled business message data. The pre-trained large model is trained on a labeled dataset generated by an unsupervised word segmentation algorithm, a pre-trained word embedding model, and a stack reverse tracing algorithm.

2. The method as described in claim 1, wherein, Before the operation of predicting labels on the vectorized service message data based on the pre-trained large model to obtain the labeled service message data, the following steps are included: Based on historical business message data, an labeled dataset is generated using unsupervised word segmentation algorithm, pre-trained word embedding model, and stack reverse tracing algorithm; The labeled dataset is divided into a training set, a validation set, and a test set; Based on the training set, the service message tag prediction capability of the large model is trained to obtain a pre-trained large model; and / or Based on the validation set and test set, the pre-trained large model is evaluated for model performance, and the model is optimized based on the evaluation results to obtain a pre-optimized large model.

3. The method as described in claim 2, wherein, The operation of generating a labeled dataset based on historical business message data using an unsupervised word segmentation algorithm, a pre-trained word embedding model, and a stack reverse tracing algorithm includes: Based on an unsupervised word segmentation algorithm, the historical business message data is segmented and ambiguous to obtain the first business message data.

4. The method of claim 3, wherein, The operation of generating a labeled dataset based on historical business message data using an unsupervised word segmentation algorithm, a pre-trained word embedding model, and a stack reverse tracing algorithm also includes: Based on a pre-trained word embedding model, the first business message data is processed into text vectors to obtain the second business message data.

5. The method of claim 4, wherein, The operation of generating a labeled dataset based on historical business message data using an unsupervised word segmentation algorithm, a pre-trained word embedding model, and a stack reverse tracing algorithm also includes: Based on the stack reverse tracing algorithm and the preset message data vector set, the second service message data is labeled to obtain the labeled dataset.

6. The method of claim 3, wherein, The operation of performing word segmentation and ambiguity processing on the historical business message data based on the unsupervised word segmentation algorithm to obtain the first business message data includes: Based on a preset dictionary segmentation strategy, the historical business message data is initially segmented to obtain several fields with the largest intersection of ambiguities. Based on the unsupervised word segmentation algorithm, the mutual information and boundary entropy of the maximum intersection ambiguity field are calculated to obtain the mutual information value and boundary entropy value of the maximum intersection ambiguity field. Based on a preset word segmentation ambiguity handling strategy, the maximum intersection ambiguity field is segmented according to the mutual information value and the boundary entropy value to obtain the first business message data.

7. The method according to claim 6, wherein, The preset dictionary segmentation strategy is a combination of the forward maximum matching method and the backtracking one-character mechanism.

8. The method according to claim 6, wherein, The calculation of mutual information and boundary entropy of the maximum intersection ambiguity field includes: determining the strings corresponding to the possible segmentation methods of the maximum intersection ambiguity field, and calculating the multi-word mutual information value and left and right boundary entropy values ​​of the strings corresponding to the possible segmentation methods.

9. The method according to claim 8, wherein, The preset word segmentation ambiguity handling strategy performs word segmentation on the maximum intersection ambiguity field based on the mutual information value and the boundary entropy value to obtain the first business message data, including: A comprehensive index is determined to be a sum of the mutual information value and boundary entropy value corresponding to different segmentation methods of the maximum intersection ambiguity field, wherein the comprehensive index is a weighted sum of the mutual information value and boundary entropy value corresponding to the segmentation method; and From the different segmentation methods of the maximum intersection ambiguity field, select the segmentation method with the highest comprehensive index that is greater than the preset index threshold as the final word segmentation method.

10. The method according to claim 9, wherein, The pre-defined word segmentation ambiguity handling strategy, which segments the maximum intersection ambiguity field according to the mutual information value and the boundary entropy value to obtain the first business message data, further includes: When the comprehensive index corresponding to different segmentation methods of the maximum intersection ambiguity field does not exceed the preset index threshold, the serial frequency segmentation method is used to determine the final word segmentation method.

11. The method of claim 3, wherein, The operation of processing the first service message data into text vectors based on the pre-trained word embedding model to obtain the second service message data includes: Based on a pre-trained word embedding model, a pre-defined set of word vectors is loaded; The first service message data is preprocessed and segmented to obtain the segmented first service message data. The segmented first business message data is processed into text vectorization using a pre-trained word embedding model and a preset set of word vectors to obtain the second business message data.

12. The method of claim 3, wherein, The historical service message data has a service request identifier. The operation of labeling the second service message data based on the stack reverse tracing algorithm and a preset message data vector set to obtain a labeled dataset includes: Based on the business request identifier, the full-link business information is invoked to construct a full-link directed acyclic graph (DAG). Based on the depth-first search algorithm, key message nodes are located in the full-link DAG, and the service information of the historical service message data at the key message nodes is obtained. Based on the preset message data vector set, the initial label prediction is performed on the second service message data to obtain the initial predicted label; Based on the business information and the initial predicted label, the second business message data is labeled to obtain a labeled dataset.

13. The method of claim 12, wherein, The key message node refers to a node for which there is a corresponding parsing result for the historical service message data, and the parsing result contains scenario information.

14. The method of claim 12, wherein, The operation of locating key message nodes in the full-link DAG based on the depth-first search algorithm and obtaining the service information of the historical service message data at the key message nodes includes: If the current message node is abnormal, then according to the depth-first search algorithm and the service request identifier, the reverse tracing of the call link is used to perform a reverse search in the full-link DAG from the current message node until the key message node is found. The corresponding message parsing results of the historical service message data at the key message node are obtained to obtain the service information.

15. The method according to claim 14, wherein, If the current message node is abnormal, the operation of performing a reverse search from the current message node in the full-link DAG by calling the depth-first search algorithm and the service request identifier until the key message node is found includes: Starting from the current message node, trace back upstream to the key message node; The nodes along the traced path are pushed onto the stack in sequence, forming a calling order from the top of the stack to the bottom of the stack; The operation of obtaining the corresponding message parsing result of the historical service message data at the key message node to obtain the service information includes: The service information is propagated downwards from the key message node along the direction of the DAG.

16. The method as described in claim 15, characterized in that, In cases where it is impossible to trace back to the upstream critical message node, the current message node is placed in the delay queue corresponding to the service request identifier; In response to the detection of a key message node corresponding to the business request identifier, a retry operation is triggered in the delay queue.

17. A service message processing device, wherein, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the operation of the service message processing method as described in any one of claims 1 to 16 when executed by the processor.

18. A storage medium, wherein, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the operation of the service message processing method as described in any one of claims 1 to 16.

19. A computer program product, wherein, The computer program product includes a computer program that, when executed by a processor, implements the operation of the service message processing method according to any one of claims 1 to 16.

Images