A method for monitoring abnormality of business data

Abstract

The present invention relates to the field of data monitoring, and in particular to a method for monitoring abnormal business data. The present invention obtains interaction records between several business terminals and a predetermined intelligent interaction model, segments the interaction records, determines explicit features of several obtained record segments, calculates explicit difference characterization parameters, determines whether to mark corresponding record segments, extracts marked record segments, obtains adjacent record segments of the record segments, determines the semantic collocation degree and sentence structure of each sentence in the adjacent record segments corresponding to the business terminal, identifies abnormal sentence structures, constructs an abnormal label pool for the business terminal, determines whether to pre-process information currently sent by the business terminal based on the abnormal label pool of each business terminal, transmits the information or / and the pre-processed information to a predetermined intelligent interaction model, and timely intervenes in abnormal data sent by the business terminal to improve the accuracy of information feedback from the intelligent interaction model.

Description

Technical Field

[0001] The present invention relates to the field of data monitoring, and in particular to a method for monitoring abnormality of business data. Background Art

[0002] With the development of artificial intelligence technology, a large number of fields have introduced intelligent employees implanted with natural language large models. They can receive business demand consultation texts from different business terminals and provide effective responses to the business demand consultation texts, effectively improving business efficiency. Therefore, when there are many business terminals connected, it is necessary to monitor the relevant business data of intelligent employees for timely intervention and improved system stability.

[0003] Chinese patent publication number: CN116796908A, discloses a digital employee automated business process processing system based on an artificial intelligence cloud platform, including a data collection unit, a cloud database, a business process efficiency evaluation unit, a core process optimization processing unit, an optimization effect verification processing unit, a re-optimization repair processing unit and a display terminal. This invention clarifies the procurement efficiency of the enterprise, and based on this, adopts the method of hierarchical division, collection and integration, and two-way matching, thereby completing the re-matching operation between the enterprise and the supplier, realizing the optimization analysis of the core body of the enterprise's procurement business process, and laying the foundation for improving the work efficiency and accuracy of the enterprise; and through verification analysis, optimization and remedial processing, effectively realizing the digital transformation of the enterprise's business process and the efficient optimization of the process, improving the production efficiency of the enterprise, and reducing the cost and risk of the enterprise.

[0004] Chinese patent publication number: CN115602165A, discloses a digital employee intelligent system based on a financial system, which converts the digital employee's understanding of the customer's consulting intention into a voice topic labeling problem. Specifically, multiple spectrograms are extracted from the voice signal, and the multiple spectrograms are encoded and decoded using a deep neural network model to obtain recognition results for representing the predetermined intention topic label. In this way, on the premise of being able to more accurately understand the user's consultation extraction, it is possible to respond to customers more reasonably and adaptively to improve the user's voice consultation experience.

[0005] However, there are still the following problems in the prior art:

[0006] In actual situations, it is not considered that different business terminals may cause semantic ambiguity or abnormality in the text sent due to different usage habits of operators. The relevant data of business terminals and intelligent employees equipped with large models are not monitored and intervened in a timely manner, which may easily lead to abnormal answers and poor accuracy of intelligent employees equipped with large models. Summary of the invention

[0007] To this end, the present invention provides a method for monitoring business data anomalies, which is used to solve the problem that the prior art does not monitor and intervene in the relevant data of business terminals and intelligent employees equipped with large models in a targeted manner, which easily leads to abnormal answers and poor accuracy of the answers made by intelligent employees equipped with large models.

[0008] To achieve the above object, the present invention provides a method for monitoring abnormality of business data, which comprises:

[0009] Obtaining interaction records between a number of business terminals and a predetermined intelligent interaction model;

[0010] Segmenting the interaction record, determining dominant features of the obtained plurality of record segments, and calculating dominant difference characterization parameters of the record segments based on the dominant features to determine whether to mark the corresponding record segments;

[0011] Only the marked record segments are extracted, and the service terminal to which the record segments belong is traced to obtain the adjacent record segments of the record segments;

[0012] Determine the semantic collocation degree and sentence structure of each sentence in the adjacent record segment corresponding to the service terminal to identify the abnormal sentence structure and build an abnormal label pool for the service terminal;

[0013] Determining whether to pre-process the information currently sent by the service terminal based on the abnormal tag pool of each service terminal, including calling the data in the abnormal tag pool to determine the abnormal sentence, generating a plurality of related sentences based on the abnormal sentence, and screening the related sentences based on the semantic association between the related sentences and the corresponding adjacent record segments of the abnormal sentence;

[0014] transmitting the information and / or the preprocessed information to a predetermined intelligent interaction model;

[0015] The explicit features include the occurrence probability of the sentence structure of the sentence in the record segment in the standard corpus and the occurrence probability of the key words in the record segment in the corresponding technical field.

[0016] Furthermore, the process of determining the technical field to which the obtained plurality of recording segments belong includes:

[0017] Obtain technical field corpus databases corresponding to several technical fields;

[0018] Determine the average occurrence probability of the keywords in each recording segment in the corpus database of each technical field;

[0019] Determine the technical field corpus database corresponding to the highest average occurrence probability;

[0020] The technical field corresponding to the technical field corpus database is determined as the belonging technical field.

[0021] Furthermore, the process of calculating the dominant difference characterization parameter of the recorded segment includes:

[0022] Determine the ratio of the occurrence probability of the keyword in the record segment in the corresponding technical field to the preset field occurrence probability threshold as the first dominant difference influencing factor;

[0023] The ratio of the occurrence probability of the sentence structure in the standard corpus to the occurrence probability of the preset standard is determined as the second dominant difference influencing factor;

[0024] The weighted sum of the first dominant difference influencing factor and the second dominant difference influencing factor is determined as the dominant difference characterization parameter.

[0025] Furthermore, the determination of whether to mark the corresponding record segment comprises:

[0026] If the dominant difference characterization parameter is greater than or equal to the reference dominant difference characterization parameter threshold, then the corresponding record segment is not marked;

[0027] If the dominant difference characterization parameter is less than the reference dominant difference characterization parameter threshold, the corresponding record segment is marked.

[0028] Furthermore, the process of identifying abnormal sentence structures includes:

[0029] Determine the sentence structure of each sentence in the record segment and the corresponding adjacent record segment and the semantic collocation between the sentences;

[0030] If there is a sentence that satisfies the structural variation condition, determining that the sentence structure corresponding to the sentence is a variation sentence structure;

[0031] The structural variation condition includes that the occurrence probability of the sentence structure corresponding to the sentence is less than the preset standard occurrence probability and the semantic matching degree between the sentence and the remaining sentences is less than a predetermined semantic matching degree threshold.

[0032] Furthermore, the process of constructing the abnormal label pool for the service terminal includes:

[0033] The abnormal sentence structure is stored in the abnormal tag pool of the business terminal;

[0034] Among them, the business terminal corresponds to the abnormal label pool one by one.

[0035] Further, the determining whether to pre-process the information currently sent by the service terminal is based on the abnormal label pool of each service terminal, wherein:

[0036] If there is data in the abnormal tag pool of the service terminal, it is determined that the information currently sent by the service terminal needs to be preprocessed;

[0037] If there is no data in the abnormal tag pool of the service terminal, it is determined that there is no need to pre-process the information currently sent by the service terminal.

[0038] Furthermore, the abnormal label pool of the service terminal needs to be updated every predetermined period.

[0039] Furthermore, the process of determining the variant sentence includes:

[0040] Determine the sentence structure of the sentence in the information currently sent by the service terminal and compare it with the abnormal sentence structure stored in the abnormal tag pool;

[0041] If the sentence structure corresponding to the sentence is the same as the mutated sentence structure stored in the mutated label pool, the sentence is determined to be a mutated sentence.

[0042] Furthermore, the process of screening the related sentences includes:

[0043] Determine the semantic relevance between the associated sentence and the corresponding adjacent record segment;

[0044] Filter out the related sentences corresponding to the maximum semantic correlation;

[0045] The associated sentences are generated after the variant sentences are enhanced, and the enhancement process includes reorganization and keyword replacement.

[0046] Compared with the prior art, the present invention obtains interaction records between several business terminals and a predetermined intelligent interaction model, segments the interaction records, determines explicit features of several obtained record fragments, calculates explicit difference characterization parameters of the record fragments, determines whether to mark the corresponding record fragments, extracts the marked record fragments, obtains the corresponding adjacent record fragments of the record fragments, determines the semantic collocation and sentence structure of each sentence in the adjacent record fragments corresponding to the business terminal, identifies abnormal sentence structures, constructs an abnormal label pool for the business terminal, determines whether to pre-process the information currently sent by the business terminal based on the abnormal label pool of each business terminal, transmits the information or / and the pre-processed information to the predetermined intelligent interaction model, monitors the predetermined intelligent interaction model, intervenes in the abnormal data sent by the business terminal in a timely manner, and improves the accuracy of the feedback information of the intelligent interaction model.

[0047] In particular, by calculating the explicit difference characterization parameters of the record fragments to determine whether to mark the corresponding record fragments, a data basis is provided for constructing an anomaly label pool. In actual situations, different business terminals are controlled by different operators. Due to differences in habits, the information sent by some business terminals may normally be abnormal. For example, the information sent contains dialects, commonly used words in non-corresponding technical fields, and special terms. Therefore, the present invention considers calculating explicit difference characterization parameters based on explicit features to characterize the difference between the information sent by the business terminal and the data in the standard corpus, reflecting the variability of the information sent, and providing data support for the subsequent construction of an anomaly label pool and adaptive pre-processing of the information sent by the business terminal, thereby monitoring the predetermined intelligent interaction model, timely intervening in the abnormal data sent by the business terminal, and improving the accuracy of the feedback information of the intelligent interaction model.

[0048] In particular, the present invention identifies abnormal sentence structures and constructs an abnormal label pool for the business terminal. When constructing the abnormal label pool, the record fragments are first marked by calculating explicit difference characterization parameters to improve analysis efficiency. Subsequently, traceability is performed based on the record fragments, and adjacent record fragments are incorporated to identify abnormal sentence structures, thereby improving the data representation of the identified abnormal sentence structures, providing effective data support for subsequent pre-processing of information based on the abnormal label pool, and then timely intervening in abnormal data sent by the business terminal, thereby improving the accuracy of feedback information from the intelligent interaction model.

[0049] In particular, based on the anomaly label pool of each of the service terminals, it is determined whether to pre-process the information currently sent by the service terminal, and the information or / and the pre-processed information is transmitted to the predetermined intelligent interaction model. In actual situations, for service terminals for which there is no data in the anomaly label pool, the information sent by them usually does not induce abnormal feedback from the intelligent interaction model, so computing power can be saved, and the information sent by such service terminals does not need to be pre-processed. However, for service terminals for which there is data in the anomaly label pool, the information sent by them may easily induce abnormal feedback from the intelligent interaction model, and then the information is adaptively pre-processed, and the abnormal data sent by the service terminals is timely intervened to improve the accuracy of the feedback information of the intelligent interaction model.

[0050] In particular, the anomaly label pool of the business terminal needs to be updated at predetermined intervals. In actual situations, abnormal data may increase at any time. If the anomaly label pool is not updated on time, it may cause abnormalities in the monitoring of the information sent by the business terminal. Based on this, the anomaly label pool is periodically updated to improve the accuracy of monitoring the information sent by the business terminal. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic diagram of the method steps of the abnormal monitoring method for business data according to an embodiment of the invention;

[0052] Figure 2 A logic block diagram of determining whether to mark a corresponding record segment according to an embodiment of the invention;

[0053] Figure 3 A logic block diagram of identifying abnormal sentence structures according to an embodiment of the invention;

[0054] Figure 4 A logic block diagram for determining whether to pre-process information sent by a service terminal according to an embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0056] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0057] See also Figure 1-Figure 4 As shown, Figure 1 This is a schematic diagram of the method steps of the abnormal monitoring method for business data according to an embodiment of the invention. Figure 2 A logic block diagram of determining whether to mark a corresponding record segment according to an embodiment of the invention, Figure 3 This is a logic block diagram of identifying abnormal sentence structures according to an embodiment of the invention. Figure 4 A method for monitoring abnormal business data of the present invention includes:

[0058] Step S1, obtaining interaction records between a plurality of service terminals and a predetermined intelligent interaction model. It should be noted that a person skilled in the art needs to obtain authorization from the service terminal before obtaining the interaction records;

[0059] Step S2, segmenting the interaction record, determining dominant features of the obtained plurality of record segments, and calculating dominant difference characterization parameters of the record segments based on the dominant features to determine whether to mark the corresponding record segments;

[0060] Step S3, extracting only the marked record segments, tracing the service terminal to which the record segments belong, and correspondingly obtaining the adjacent record segments of the record segments;

[0061] Step S4, determining the semantic collocation degree and sentence structure of each sentence in the adjacent record segment corresponding to the service terminal, so as to identify the abnormal sentence structure and construct an abnormal label pool for the service terminal;

[0062] Step S5: determining whether to pre-process the information currently sent by the service terminal based on the abnormal label pool of each service terminal,

[0063] The method includes calling data in a mutation tag pool to determine a mutation sentence, generating a plurality of related sentences based on the mutation sentence, and screening the related sentences based on the semantic relevance between the related sentences and the corresponding adjacent record segments of the mutation sentence;

[0064] Step S6, transmitting the information and / or the preprocessed information to a predetermined intelligent interaction model;

[0065] The explicit features include the occurrence probability of the sentence structure of the sentence in the record segment in the standard corpus and the occurrence probability of the key words in the record segment in the corresponding technical field.

[0066] Specifically, the business terminal is a computer or mobile terminal operated by an operator.

[0067] Specifically, the predetermined intelligent interaction model is a natural language big model, which can process the interactive texts sent by each business terminal and generate feedback text information for the interactive texts. In implementation, there is no limitation on the form of the natural language big model. Technical personnel in this field can adopt the existing open source natural language big model or the natural language big model built by themselves, which will not be repeated here.

[0068] It is understandable that, in implementation, the interaction record includes text information sent by the service terminal and text information fed back by the natural language model.

[0069] Specifically, when segmenting the interaction record, it should be avoided as much as possible to avoid sentence breaks at the segmentation nodes. For example, avoid segmenting the same sentence into two different record segments. The number of words in each record segment may be different. Those skilled in the art can choose the number of words contained in the record segment by themselves, which will not be elaborated here.

[0070] Specifically, there is no restriction on the method of tracing the service terminal to which the record fragment belongs. For example, it can be a reverse query method, a method based on information DNA, and a diffusion model based on game theory. As long as the traceability is successful, those skilled in the art can make their own choices. In implementation, the process of tracing by using the reverse query method includes:

[0071] Identify the record segments that need to be traced;

[0072] Obtain all metadata related to the record segment, including the source, creator, modification history, access history, etc. of the data;

[0073] According to the processing and conversion logic of the record fragment, a reverse query model is constructed to convert the record fragment from a current state back to its previous state. It is understandable that the record fragment may involve a change in data type during storage or transmission;

[0074] Using a reverse query model, starting from the target data, gradually tracing back to the previous state of the data until the source of the record fragment is reached, and determining that the source is the business terminal to which the record fragment belongs;

[0075] Among them, during the backtracking process, it is necessary to continuously verify and proofread the accuracy of the data to ensure the correctness of the reverse query results.

[0076] Specifically, the process of determining the technical field to which the obtained plurality of recording segments belong includes:

[0077] Obtain technical field corpus databases corresponding to several technical fields;

[0078] Determine the average occurrence probability of the keywords in each recording segment in the corpus database of each technical field;

[0079] Determine the technical field corpus database corresponding to the highest average occurrence probability;

[0080] The technical field corresponding to the technical field corpus database is determined as the belonging technical field.

[0081] Specifically, the technical field corpus database stores text information belonging to the corresponding technical field. There is no limitation on the method of determining the technical field corpus database. For example, an existing technical field corpus database may be directly adopted, or it may be a technical field corpus database constructed by technical personnel in this field. For example, text may be extracted from a paper database in the corresponding technical field as corpus, and technical processing such as corpus cleaning, corpus duplication checking and desensitization may be performed before storage to construct the corresponding technical field corpus database.

[0082] It is understandable that the corpus in the technical field corpus database is presented in the form of paragraphs, and the probability of a keyword appearing in a single technical field corpus database is the probability of the keyword appearing in the paragraph of the corresponding technical field corpus database.

[0083] Specifically, there is no restriction on the method of determining the keywords in the record segment. For example, any keyword extraction algorithm or model in the prior art can be used to extract the keywords. For example, the rake-nltk algorithm can use the NLTK (Natural Language Toolkit) library to process text data to extract keywords from the text data. Of course, other methods can also be used as long as the keywords can be extracted correctly. I will not go into details.

[0084] Specifically, the process of calculating the explicit difference characterization parameters of the recorded segments includes:

[0085] Determine the ratio of the occurrence probability of the keyword in the record segment in the corresponding technical field to the preset field occurrence probability threshold as the first dominant difference influencing factor;

[0086] The ratio of the occurrence probability of the sentence structure in the standard corpus to the occurrence probability of the preset standard is determined as the second dominant difference influencing factor;

[0087] The weighted sum of the first dominant difference influencing factor and the second dominant difference influencing factor is determined as the dominant difference characterization parameter.

[0088] Specifically, the preset domain occurrence probability threshold is preset, and the average occurrence probability of each keyword in the technical field corpus database is counted, and the average occurrence probability is determined as the preset domain occurrence probability threshold.

[0089] Specifically, the preset standard occurrence probability is obtained by pre-calculation, and a number of probabilities of a number of sentence structures in a standard corpus are obtained in advance, and the average probability is determined as the preset standard occurrence probability.

[0090] Specifically, the sum of the weight coefficients of the first dominant difference influencing factor and the second dominant difference influencing factor is 1. In implementation, the weight coefficient of the first dominant difference influencing factor is 0.4, and the weight coefficient of the second dominant difference influencing factor is 0.6.

[0091] Specifically, it is determined whether to mark the corresponding record segment, wherein:

[0092] If the dominant difference characterization parameter is greater than or equal to the reference dominant difference characterization parameter threshold, then the corresponding record segment is not marked;

[0093] If the dominant difference characterization parameter is less than the reference dominant difference characterization parameter threshold, the corresponding record segment is marked.

[0094] The baseline dominant difference characterization parameter threshold was selected within [0.65, 0.85].

[0095] Specifically, by calculating the explicit difference characterization parameters of the record fragments to determine whether to mark the corresponding record fragments, a data basis is provided for constructing an anomaly label pool. In actual situations, different business terminals are controlled by different operators. Due to differences in habits, the information sent by some business terminals may normally be abnormal. For example, the information sent contains dialects, commonly used words in non-corresponding technical fields, and special terms. Therefore, the present invention considers calculating explicit difference characterization parameters based on explicit features to characterize the difference between the information sent by the business terminal and the data in the standard corpus, reflecting the variability of the information sent, and providing data support for the subsequent construction of an anomaly label pool and adaptive pre-processing of the information sent by the business terminal, thereby monitoring the predetermined intelligent interaction model, intervening in the abnormal data sent by the business terminal in a timely manner, and improving the accuracy of the feedback information of the intelligent interaction model.

[0096] Specifically, the process of identifying variant sentence structures includes:

[0097] Determine the sentence structure of each sentence in the record segment and the corresponding adjacent record segment and the semantic collocation between the sentences;

[0098] If there is a sentence that satisfies the structural variation condition, determining that the sentence structure corresponding to the sentence is a variation sentence structure;

[0099] The structural variation condition includes that the occurrence probability of the sentence structure corresponding to the sentence is less than the preset standard occurrence probability and the semantic matching degree between the sentence and the remaining sentences is less than a predetermined semantic matching degree threshold.

[0100] It is understandable that sentence structure refers to the various components that make up a sentence and their relationship to each other. It is understandable that sentence structure usually follows the order of "subject, predicate, and object", but in actual language use, this order may change due to emphasis, rhetoric, etc.

[0101] Specifically, there is no limitation on the method for recognizing sentence structure, and any one of the existing specific structure extraction tools can be used. For example, the Language Technology Platform toolkit can be used to extract subject, predicate, object and other sentence structures. This toolkit is suitable for syntactic analysis in natural language processing. Of course, other forms can also be used, which will not be repeated here.

[0102] Specifically, there is no limitation on the method for determining the semantic collocation between sentences. For example, the cosine similarity method can be used to determine the semantic collocation. First, the sentences are vectorized, and the cosine angle between the two sentence vectors is calculated as the semantic collocation between the sentences. If the directions of the two vectors are closer, the cosine similarity is closer to 1, indicating that the semantics are more similar. Of course, other methods can also be used. Technical personnel in this field can make a choice based on actual conditions, which will not be elaborated here.

[0103] It is understandable that the purpose of setting the semantic collocation threshold is to characterize that the semantics of two sentences are more compatible. If the semantic collocation is determined by a pre-similarity method, those skilled in the art can select the semantic collocation threshold within [0.75, 0.95].

[0104] Specifically, the process of building a mutation label pool for the service terminal includes:

[0105] The abnormal sentence structure is stored in the abnormal tag pool of the business terminal;

[0106] Among them, the business terminal corresponds to the abnormal label pool one by one.

[0107] Specifically, the mutation tag pool can be a database in any form, and the database is dedicated to the corresponding business terminal and is used to store the corresponding mutation sentence structure.

[0108] Specifically, the present invention identifies the abnormal sentence structure and constructs an abnormal label pool for the business terminal. When constructing the abnormal label pool, the record fragments are first marked by calculating the explicit difference characterization parameters to improve the analysis efficiency. Subsequently, the source is traced based on the record fragments, and the adjacent record fragments are incorporated to identify the abnormal sentence structure, thereby improving the data representation of the identified abnormal sentence structure, providing effective data support for the subsequent pre-processing of information based on the abnormal label pool, and then timely intervening in the abnormal data sent by the business terminal, thereby improving the accuracy of the feedback information of the intelligent interaction model.

[0109] Specifically, it is determined whether to pre-process the information currently sent by the service terminal based on the abnormal label pool of each service terminal, wherein:

[0110] If there is data in the abnormal tag pool of the service terminal, it is determined that the information currently sent by the service terminal needs to be preprocessed;

[0111] If there is no data in the abnormal tag pool of the service terminal, it is determined that there is no need to pre-process the information currently sent by the service terminal.

[0112] It is understandable that if there is no data in the abnormal tag pool of the business terminal, it means that the semantic abnormality tendency of the text information sent by the business terminal is low, and the text information can be directly transmitted to the predetermined intelligent interaction model to obtain feedback text, which will not be repeated here.

[0113] Specifically, based on the anomaly label pool of each business terminal, it is determined whether to preprocess the information currently sent by the business terminal, and the information or / and the preprocessed information is transmitted to the predetermined intelligent interaction model. In actual situations, for business terminals in which there is no data in the anomaly label pool, the information sent by them usually does not induce abnormal feedback from the intelligent interaction model, so computing power can be saved, and the information sent by such business terminals does not need to be preprocessed. However, for business terminals in which there is data in the anomaly label pool, the information sent by them may easily induce abnormal feedback from the intelligent interaction model, and then the information is adaptively preprocessed, and the abnormal data sent by the business terminals is timely intervened to improve the accuracy of the feedback information of the intelligent interaction model.

[0114] Specifically, the abnormal label pool of the service terminal needs to be updated every predetermined period.

[0115] Specifically, the anomaly label pool of the business terminal needs to be updated at predetermined intervals. In actual situations, abnormal data may increase at any time. If the anomaly label pool is not updated on time, it may cause abnormalities in the monitoring of the information sent by the business terminal. Based on this, the anomaly label pool is periodically updated to improve the accuracy of monitoring the information sent by the business terminal.

[0116] Specifically, in implementation, the predetermined period is one month. Of course, those skilled in the art can determine the period according to actual conditions, which will not be elaborated here.

[0117] Specifically, the process of determining variant sentences includes:

[0118] Determine the sentence structure of the sentence in the information currently sent by the service terminal and compare it with the abnormal sentence structure stored in the abnormal tag pool;

[0119] If the sentence structure corresponding to the sentence is the same as the mutated sentence structure stored in the mutated label pool, the sentence is determined to be a mutated sentence.

[0120] Specifically, comparison is required only when there are variant sentences stored in the variant tag pool, which will not be described in detail here.

[0121] Specifically, the process of screening related sentences includes:

[0122] Determine the semantic relevance between the associated sentence and the corresponding adjacent record segment;

[0123] Filter out the related sentences corresponding to the maximum semantic correlation;

[0124] The associated sentences are generated after the variant sentences are enhanced, and the enhancement process includes reorganization and keyword replacement.

[0125] It is understandable that abnormal sentences may have abnormal word order, may be conventional words in non-corresponding technical fields, or may contain dialects. Based on this, when determining related sentences, consider reorganizing the related sentences and replacing keywords.

[0126] Specifically, there is no limitation on the specific method of reorganizing related sentences, as long as the effect of sentence reorganization and comprehension can be achieved. For example, the EDA text data enhancement method can be used, including random insertion, random exchange, random deletion and other methods to reorganize sentences. These technologies can reorganize and enhance sentences while maintaining semantic consistency. Of course, other methods can also be used. Those skilled in the art can choose according to actual conditions, which will not be elaborated here.

[0127] Specifically, there is no limitation on the specific method of replacing keywords in related sentences, as long as the keyword replacement effect can be achieved. For example, the extracted keywords can be replaced with synonyms, which can be achieved through a synonym dictionary or a method based on the word2vec model. The word2vec model can learn the semantic relationship between words, so it can generate synonym replacements that are more suitable for specific fields. Those skilled in the art can make a choice based on the actual situation, which will not be elaborated here.

[0128] It is understandable that when performing synonym replacement, the number of replacements needs to be controlled to avoid semantic changes caused by excessive replacement. For example, it can be set to replace only a certain number of words in a sentence.

[0129] Specifically, there is no limitation on the method for determining the semantic relevance. For example, a method based on word vectors and neural networks can be used. This method converts the text into a vector representation, and then evaluates the semantic relevance by calculating the similarity between several vectors. The similarity can be calculated using cosine similarity. Of course, other methods can also be used as long as the purpose of determining the semantic relevance can be achieved. Those skilled in the art can make a choice based on the actual situation, and will not go into details here.

[0130] A method for monitoring abnormal business data of the present invention, if implemented in the form of a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention, and the aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0131] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0132] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for monitoring abnormality of business data, characterized in that: include: Obtaining interaction records between a number of business terminals and a predetermined intelligent interaction model; Segmenting the interaction record, determining dominant features of the obtained plurality of record segments, and calculating dominant difference characterization parameters of the record segments based on the dominant features to determine whether to mark the corresponding record segments; Only the marked record segments are extracted, and the service terminal to which the record segments belong is traced to obtain the adjacent record segments of the record segments; Determine the semantic collocation degree and sentence structure of each sentence in the adjacent record segment corresponding to the service terminal to identify the abnormal sentence structure and build an abnormal label pool for the service terminal; The process of identifying variant sentence structures includes: Determine the sentence structure of each sentence in the record segment and the corresponding adjacent record segment and the semantic collocation between the sentences; If there is a sentence that satisfies the structural variation condition, determining that the sentence structure corresponding to the sentence is a variation sentence structure; The structural variation condition includes that the occurrence probability of the sentence structure corresponding to the sentence is less than the preset standard occurrence probability and the semantic collocation degree between the sentence and the remaining sentences is less than a predetermined semantic collocation degree threshold; Determining whether to pre-process the information currently sent by the service terminal based on the abnormal tag pool of each service terminal, including calling the data in the abnormal tag pool to determine the abnormal sentence, generating a plurality of related sentences based on the abnormal sentence, and screening the related sentences based on the semantic association between the related sentences and the corresponding adjacent record segments of the abnormal sentence; The process of screening the related sentences includes: Determine the semantic relevance between the associated sentence and the corresponding adjacent record segment; Filter out the related sentences corresponding to the maximum semantic correlation; The related sentences are generated by enhancing the variant sentences, wherein the enhancement includes reorganization and keyword replacement; transmitting the information and / or the preprocessed information to a predetermined intelligent interaction model; The explicit features include the occurrence probability of the sentence structure of the sentence in the record segment in the standard corpus and the occurrence probability of the key words in the record segment in the corresponding technical field.

2. The abnormality monitoring method for business data according to claim 1, characterized in that: The determination process of the technical field includes: Obtain technical field corpus databases corresponding to several technical fields; Determine the average occurrence probability of the keywords in each recording segment in the corpus database of each technical field; Determine the technical field corpus database corresponding to the highest average occurrence probability; The technical field corresponding to the technical field corpus database is determined as the belonging technical field.

3. The abnormal monitoring method for business data according to claim 2 is characterized in that: The process of calculating the dominant difference characterization parameter of the recorded segment includes: Determine the ratio of the occurrence probability of the keyword in the record segment in the corresponding technical field to the preset field occurrence probability threshold as the first dominant difference influencing factor; Determine the ratio of the occurrence probability of the sentence structure in the standard corpus to the occurrence probability of the preset standard as the second dominant difference influencing factor; The weighted sum of the first dominant difference influencing factor and the second dominant difference influencing factor is determined as the dominant difference characterization parameter.

4. The abnormality monitoring method for business data according to claim 3 is characterized in that: The step of determining whether to mark the corresponding recording segment comprises: If the dominant difference characterization parameter is greater than or equal to the reference dominant difference characterization parameter threshold, then the corresponding record segment is not marked; If the dominant difference characterization parameter is less than the reference dominant difference characterization parameter threshold, the corresponding record segment is marked.

5. The abnormality monitoring method for business data according to claim 1, characterized in that: The process of constructing the abnormal label pool for the service terminal includes: The abnormal sentence structure is stored in the abnormal tag pool of the business terminal; Among them, the business terminal corresponds to the abnormal label pool one by one.

6. The abnormality monitoring method for business data according to claim 1, characterized in that: The determining whether to pre-process the information currently sent by the service terminal based on the abnormal label pool of each service terminal, wherein: If there is data in the abnormal tag pool of the service terminal, it is determined that the information currently sent by the service terminal needs to be preprocessed; If there is no data in the abnormal tag pool of the service terminal, it is determined that there is no need to pre-process the information currently sent by the service terminal.

7. The abnormality monitoring method for business data according to claim 1, characterized in that: The abnormal label pool of the service terminal needs to be updated every predetermined period.

8. The abnormality monitoring method for business data according to claim 1, characterized in that: The process of determining the variant sentence includes: Determine the sentence structure of the sentence in the information currently sent by the service terminal and compare it with the abnormal sentence structure stored in the abnormal tag pool; If the sentence structure corresponding to the sentence is the same as the mutated sentence structure stored in the mutated label pool, the sentence is determined to be a mutated sentence.

Images