Quality inspection method, device, medium and electronic equipment for auditing text
Through word segmentation processing and machine learning models to calculate keyword probability and weight, the problem of high cost and low accuracy of manual review text quality inspection is solved, and automated and intelligent quality inspection methods are realized, which improves the quality inspection speed and accuracy. .
Patent Information
- Application Number
- CN202110077289.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-01-20
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2041-01-20
AI Technical Summary
The prior art has problems of high cost and low accuracy in the quality inspection of manual review texts, and lacks adaptability.
By obtaining the review basis and conclusions, performing word segmentation processing to obtain keywords, determining the mapping relationship between keywords and conclusions, calculating keyword probability and weights, performing probability prediction to determine quality inspection conclusions, and using machine learning models to achieve automation and intelligence Chemical quality inspection.
It improves the speed and efficiency of quality inspection, ensures the accuracy and adaptability of quality inspection conclusions, and significantly reduces costs.
Smart Images

Figure CN113011156B_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to a quality inspection method for reviewing texts, a quality inspection device for reviewing texts, a computer-readable medium, and an electronic device. Background Art
[0002] Quality inspection of manual review conclusions can ensure the accuracy of the conclusions.
[0003] However, quality inspection through manual sampling will further consume a lot of manpower and time, and cannot avoid human negligence; and quality inspection by checking the keywords or sensitive words in the manual review conclusions lacks adaptability.
[0004] In view of this, there is an urgent need in this field to develop a new quality inspection method and device for reviewing texts.
[0005] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the technical background of the present application, and therefore may include information that does not constitute prior art known to ordinary technicians in the field. Summary of the invention
[0006] The purpose of the present disclosure is to provide a quality inspection method for audit texts, a quality inspection device for audit texts, a computer-readable medium, and an electronic device, thereby overcoming the technical problems of high quality inspection costs and low accuracy, at least to a certain extent.
[0007] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.
[0008] According to one aspect of the embodiments of the present disclosure, a quality inspection method for an audit text is provided, the method comprising: obtaining an audit basis and an audit conclusion, and performing word segmentation processing on the audit basis to obtain audit keywords;
[0009] Determining an audit mapping relationship between the audit keyword and the audit conclusion, and determining at least two audit probabilities according to the audit mapping relationship;
[0010] Performing probability calculation on the at least two audit probabilities to obtain a keyword probability corresponding to the audit keyword;
[0011] Determine the audit weight of the audit keyword according to the audit keyword and the audit basis;
[0012] A probability prediction calculation is performed on the keyword probability and the audit weight to obtain a conclusion probability, and the conclusion probability is used to determine a quality inspection conclusion in the audit conclusion.
[0013] According to one aspect of an embodiment of the present disclosure, a quality inspection device for reviewing texts is provided, the device comprising: a word segmentation processing module configured to obtain review basis and review conclusion, and perform word segmentation processing on the review basis to obtain review keywords;
[0014] A mapping relationship module, configured to determine an audit mapping relationship between the audit keyword and the audit conclusion, and determine at least two audit probabilities according to the audit mapping relationship;
[0015] A probability calculation module, configured to perform probability calculation on the at least two review probabilities to obtain a keyword probability corresponding to the review keyword;
[0016] A weight determination module, configured to determine the audit weight of the audit keyword according to the audit keyword and the audit basis;
[0017] The conclusion probability module is configured to perform probability prediction calculation on the keyword probability and the audit weight to obtain the conclusion probability, and use the conclusion probability to determine the quality inspection conclusion in the audit conclusion.
[0018] In some embodiments of the present disclosure, based on the above technical solution, the weight determination module includes: a position determination submodule configured to determine the text position of the audit keyword in the audit basis;
[0019] The position weight submodule is configured to determine the review weight of the review keyword according to the text position.
[0020] In some embodiments of the present disclosure, based on the above technical solution, the position determination submodule module includes: a weight mapping unit configured to determine a weight mapping relationship between the text position and the audit weight of the audit keyword;
[0021] A mapping determination unit is configured to determine the audit weight of the audit keyword according to the weight mapping relationship and the text position.
[0022] In some embodiments of the present disclosure, based on the above technical solution, the position determination submodule includes: a candidate weight unit configured to determine the position weight of the audit keyword according to the text position;
[0023] The weight calculation unit is configured to obtain the influence weight corresponding to the audit keyword, and perform weight calculation on the position weight and the influence weight to obtain the audit weight.
[0024] In some embodiments of the present disclosure, based on the above technical solution, the probability calculation module includes: a model calculation submodule, configured to use a machine learning model to perform probability calculation on the at least two review probabilities to obtain a keyword probability corresponding to the review keyword.
[0025] In some embodiments of the present disclosure, based on the above technical solution, the probability calculation module includes: a matrix calculation submodule, configured to perform probability calculation on the at least two audit probabilities to obtain a probability matrix;
[0026] The probability search submodule is configured to determine the keyword probability corresponding to the audit keyword in the probability matrix.
[0027] In some embodiments of the present disclosure, based on the above technical solution, the mapping relationship module includes: a separate probability submodule configured to determine the conclusion probability corresponding to the audit conclusion and determine the text probability corresponding to the audit keyword;
[0028] A quantity determination submodule, configured to determine the number of conclusions of the audit keyword in the audit conclusion according to the audit mapping relationship;
[0029] The audit probability submodule is configured to determine the relevance probability of the audit keyword in the audit conclusion according to the number of conclusions, and determine the conclusion probability, the text probability and the relevance probability as at least two audit probabilities.
[0030] In some embodiments of the present disclosure, based on the above technical solution, the conclusion probability module includes: a probability number submodule, configured to determine that the conclusion probability is at least two, and the at least two conclusion probabilities are obtained by at least two audit bases and at least two audit conclusions;
[0031] The probability comparison submodule is configured to compare the at least two conclusion probabilities to obtain a target conclusion probability, and use the target conclusion probability to determine a quality inspection conclusion from the at least two audit conclusions.
[0032] In some embodiments of the present disclosure, based on the above technical solution, the quality inspection device for reviewing texts further includes: a conclusion quality inspection module configured to obtain a manual review conclusion corresponding to the review basis, and compare the manual review conclusion with the quality inspection conclusion to determine a quality inspection result;
[0033] The quality inspection result module is configured to update the keyword probability according to the quality inspection result.
[0034] In some embodiments of the present disclosure, based on the above technical solution, the quality inspection result module includes: a conclusion review submodule configured to review the quality inspection conclusion to obtain a review result if the quality inspection result is that the manual review conclusion is different from the quality inspection conclusion;
[0035] The probability updating submodule is configured to update the keyword probability if the review result is that the quality inspection conclusion is wrong.
[0036] In some embodiments of the present disclosure, based on the above technical solution, the word segmentation processing module includes: a quality inspection word segmentation submodule, configured to perform word segmentation processing on the audit basis to obtain quality inspection word segmentation;
[0037] The feature extraction submodule is configured to perform feature extraction processing on the quality inspection segmented words to obtain audit keywords.
[0038] In some embodiments of the present disclosure, based on the above technical solution, the word segmentation processing module includes: a data cleaning submodule configured to clean the audit basis to obtain a target audit basis;
[0039] The target basis submodule is configured to perform word segmentation processing on the target review basis to obtain review keywords.
[0040] According to one aspect of an embodiment of the present disclosure, a computer-readable medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the quality inspection method of the audit text in the above technical solution is implemented.
[0041] According to one aspect of an embodiment of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute a quality inspection method for an audit text as in the above technical solution by executing the executable instructions.
[0042] In the technical solution provided by the embodiments of the present disclosure, on the one hand, the keyword probability and audit weight of the audit keyword are determined by utilizing the audit mapping relationship between the audit keyword and the audit conclusion, thereby realizing the quantification of the correlation between the audit conclusion and the audit basis, and providing the possibility for the quantitative application of the audit basis and the audit conclusion in the text quality inspection scenario; on the other hand, the quantitative keyword probability and audit weight are subjected to probability prediction calculation to obtain the conclusion probability, and the conclusion probability is further used to determine the quality inspection conclusion. This is an automated and intelligent audit text quality inspection method, which improves the quality inspection speed and efficiency, ensures the accuracy and adaptability of the quality inspection conclusion, and greatly reduces the quality inspection cost of the audit text.
[0043] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] The drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present disclosure, and together with the specification, are used to explain the principles of the present disclosure. Obviously, the drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:
[0045] Figure 1 The schematic diagram schematically shows an exemplary system architecture to which the technical solution of the present disclosure is applied;
[0046] Figure 2 A flowchart schematically shows a method for quality inspection of a review text in some embodiments of the present disclosure;
[0047] Figure 3 A flowchart schematically shows a method for obtaining audit keywords in some embodiments of the present disclosure;
[0048] Figure 4 Schematically shows another flow chart of steps for obtaining audit keywords in some embodiments of the present disclosure;
[0049] Figure 5 A flowchart schematically illustrates the steps of a method for determining at least two audit probabilities in some embodiments of the present disclosure;
[0050] Figure 6 A flowchart schematically shows the steps of a method for determining keyword probability in some embodiments of the present disclosure;
[0051] Figure 7 A flowchart schematically shows the steps of a method for determining an audit weight in some embodiments of the present disclosure;
[0052] Figure 8 A flowchart schematically shows the steps of a first method for further determining the audit weight in some embodiments of the present disclosure;
[0053] Fig. 9 A flowchart schematically shows the steps of a second method for further determining the audit weight in some embodiments of the present disclosure;
[0054] Fig.10 A flowchart schematically shows the steps of a method for determining a quality inspection conclusion in some embodiments of the present disclosure;
[0055] Fig.11A flowchart schematically shows the steps of a method for updating keyword probability in some embodiments of the present disclosure;
[0056] Fig.12 A flowchart schematically shows the steps of a method for further updating keyword probability in some embodiments of the present disclosure;
[0057] Fig.13 The flowchart schematically shows the steps of the quality inspection method of the audit text in the application scenario of some embodiments of the present disclosure;
[0058] Fig.14 The structure block diagram of a quality inspection device for reviewing texts in some embodiments of the present disclosure is schematically shown;
[0059] Fig.15 The structure diagram of a computer system suitable for implementing the electronic device of the embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION
[0060] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be more comprehensive and complete and will fully convey the concept of the example embodiments to those skilled in the art.
[0061] In addition, the described features, structures or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring the various aspects of the present disclosure.
[0062] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.
[0063] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.
[0064] In the relevant technologies in this field, manually ensuring the quality of quality inspection of manual review conclusions will consume a lot of manpower and time, and human negligence may also occur; and quality inspection through manual sampling, although the quality of quality inspection can be guaranteed within a certain range, there are also cases of incomplete quality inspection of manual review conclusions and human negligence in the quality inspection process; quality inspection of manual review conclusions using keyword or sensitive word inspection can only check and quality inspect known or specific keywords or sensitive words, and lacks adaptability.
[0065] Based on the problems existing in the above solutions, the present disclosure provides an artificial intelligence-based quality inspection method for audit texts, a quality inspection device for audit texts, a computer-readable medium, and an electronic device.
[0066] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making.
[0067] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.
[0068] Among them, machine learning (ML) is a multi-disciplinary interdisciplinary subject, involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.
[0069] The method of using machine learning technology in artificial intelligence to conduct quality inspection on audit texts can quantify the correlation between audit conclusions and audit basis, and provide the possibility for the quantitative application of audit basis and audit conclusions in text quality inspection scenarios. In addition, it also improves the speed and efficiency of quality inspection, ensures the accuracy and adaptability of quality inspection conclusions, and greatly reduces the cost of quality inspection of audit texts.
[0070] Figure 1 A schematic diagram of an exemplary system architecture applying the technical solution of the present disclosure is shown.
[0071] like Figure 1 As shown, the system architecture 100 may include a terminal 110 , a network 120 , and a server 130 . The terminal 110 and the server 130 are connected via the network 120 .
[0072] The terminal 110 may be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited thereto. The network 120 may be a communication medium of various connection types that can provide a communication link between the terminal 110 and the server 130, such as a wired communication link, a wireless communication link, or an optical fiber cable, etc., which is not limited in this application. The server 130 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms.
[0073] Specifically, the server 130 obtains the audit basis and the audit conclusion, and performs word segmentation processing on the audit basis to obtain the audit keywords. Then, the audit mapping relationship between the audit keywords and the audit conclusion is determined, and at least two audit probabilities are determined based on the audit mapping relationship, so as to perform probability calculation on the at least two audit probabilities to obtain the keyword probability corresponding to the audit keyword. Further, the audit weight of the audit keyword is determined based on the audit keyword and the audit basis, and the keyword probability and the audit weight are probability predicted and calculated to obtain the conclusion probability, so as to determine the quality inspection conclusion in the audit conclusion using the conclusion probability.
[0074] It is worth noting that, in the quality inspection method or device for auditing texts disclosed in this application, multiple servers can be combined into a blockchain, and the servers are nodes on the blockchain.
[0075] In addition, the quality inspection method of the audit text in the embodiment of the present disclosure can be applied to the terminal or the server, and the present disclosure does not make any special limitation on this. The embodiment of the present disclosure mainly uses the application of the quality inspection method of the audit text to the server 130 as an example.
[0076] The quality inspection method for audit texts, the quality inspection device for audit texts, the computer-readable medium, and the electronic device provided by the present disclosure are described in detail below in conjunction with specific implementation methods.
[0077] Figure 2 The flowchart schematically shows the steps of the quality inspection method of the audit text in some embodiments of the present disclosure, such as Figure 2 As shown, the quality inspection method of the audit text can mainly include the following steps:
[0078] Step S210: Obtain the audit basis and audit conclusion, and perform word segmentation on the audit basis to obtain audit keywords.
[0079] Step S220. Determine the audit mapping relationship between the audit keywords and the audit conclusions, and determine at least two audit probabilities based on the audit mapping relationship.
[0080] Step S230. Perform probability calculation on at least two audit probabilities to obtain a keyword probability corresponding to the audit keyword.
[0081] Step S240. Determine the audit weight of the audit keyword based on the audit keyword and the audit basis.
[0082] Step S250. Perform probability prediction calculation on the keyword probability and the audit weight to obtain the conclusion probability, and use the conclusion probability to determine the quality inspection conclusion in the audit conclusion.
[0083] In the exemplary embodiments of the present disclosure, on the one hand, the audit mapping relationship between audit keywords and audit conclusions is used to determine the keyword probability and audit weight of the audit keywords, thereby realizing the quantification of the correlation between the audit conclusion and the audit basis, and providing the possibility for the quantitative application of the audit basis and audit conclusion in the text quality inspection scenario; on the other hand, the quantitative keyword probability and audit weight are subjected to probability prediction calculation to obtain the conclusion probability, and the conclusion probability is further used to determine the quality inspection conclusion. This is an automated and intelligent audit text quality inspection method, which improves the quality inspection speed and efficiency, ensures the accuracy and adaptability of the quality inspection conclusion, and greatly reduces the quality inspection cost of the audit text.
[0084] The following is a detailed description of each step of the quality inspection method for the audit text.
[0085] In step S210, the audit basis and the audit conclusion are obtained, and the audit basis is segmented to obtain audit keywords.
[0086] In the exemplary embodiment of the present disclosure, the quality inspection of the audit basis and the audit conclusion is the quality inspection, which is the operation of inspecting the manual audit results. The audit basis is the basis of the quality inspection, and the audit conclusion is the conclusion that can be obtained from the audit basis.
[0087] The audit basis and audit conclusion refer to the valid audit basis and audit conclusion, that is, the one obtained by excluding invalid audit records such as rejection or retrial. Since invalid audit records such as rejection and retrial are marked with corresponding identification information, the rejected or invalid audit records can be identified by the identification information to eliminate the audit records and obtain the valid audit basis and audit conclusion. At this time, the audit conclusion may include conclusions such as pass or fail.
[0088] After obtaining effective audit basis and audit conclusions, the audit basis can be further segmented to obtain audit keywords.
[0089] Figure 3 and Figure 4 The following flowcharts show two methods for obtaining audit keywords. Figure 3 The flowchart shows the steps of the method of performing word segmentation and feature extraction on the audit basis to obtain audit keywords. Figure 4 A flow chart showing the steps of a method for cleaning and segmenting audit evidence to obtain audit keywords.
[0090] It is worth mentioning that Figure 4 The way to segment the audit evidence after cleaning may include: Figure 3 The word segmentation processing method and feature extraction processing method are shown.
[0091] In an alternative embodiment, Figure 3 A flowchart of a method for obtaining audit keywords is shown. Figure 3 As shown, the method includes at least the following steps: In step S310, the audit basis is segmented to obtain quality inspection segmentation.
[0092] The word segmentation of the audit basis can be achieved by using Jieba word segmentation. Jieba word segmentation supports three word segmentation modes, namely full mode, precise mode and search engine mode. Among them, the full mode is to scan all the words that can be formed into words in the sentence, which is very fast; the precise mode attempts to split the sentence most accurately, which is suitable for text segmentation; the search engine mode is based on the precise mode, and the long words are segmented again, which can improve the recall rate and is suitable for search engine segmentation. In addition, Jieba word segmentation also supports traditional Chinese word segmentation and custom dictionaries.
[0093] Although the Jieba word segmentation has a certain ability to recognize new words, the new words added by yourself and the corresponding weights can ensure a higher accuracy. Therefore, the custom dictionary of the Jieba word segmentation is used for word segmentation.
[0094] The custom dictionary consists of three parts: words, word frequency, and part of speech. The three parts are separated by spaces and the order cannot be reversed. Among them, part of speech and word frequency can be omitted. For example, the custom dictionary can include "queried 100v", "list person 100n", "blacklist 100n", etc.
[0095] Specifically, the part-of-speech comparison table is shown in Table 1:
[0096] Table 1
[0097] Part of Speech Part of Speech a adjective n name v verb … …
[0098] When creating a custom dictionary, in addition to using parts of speech and word frequency, you can also add information such as the business attributes of the words, the position of the words, and the previous and next positions of the words, so that the custom dictionary you create contains more word dimensions and ensures the accuracy of word segmentation processing.
[0099] Furthermore, the words in the custom dictionary may be derived from business terms, i.e., corresponding audit guidance manuals and industry terminology descriptions, etc., or may be obtained by recycling new audit keywords obtained during feature extraction in step S320, or may be obtained by other means. This exemplary embodiment does not make any special limitations on this.
[0100] In addition, the word segmentation processing methods based on rules and statistics can be used to obtain quality inspection word segmentation.
[0101] Among them, rule-based word segmentation is to segment words according to the matching method by pre-building a dictionary. The dictionary can be a polynomial N-gram (Chinese language model) dictionary, and according to the word segmentation strategy, the character string contained in the original text is matched with the pre-built polynomial N-gram dictionary to obtain all possible segmentation results for each character, and then the shortest path method based on the polynomial N-gram dictionary is used to calculate the final quality inspection word segmentation.
[0102] Statistical word segmentation is to use a classifier constructed by training annotated corpus to perform word segmentation. The classifier can be trained and constructed using machine learning or deep learning algorithms. Such algorithms can use hidden Markov models (HMM), conditional random field algorithms (CRF), and deep learning.
[0103] In addition, it is also possible to directly call a variety of different word segmentation tools as word segmentation models, and call different types of word segmentation tools to perform word segmentation processing on the original text to obtain quality inspection word segmentation. The quality inspection word segmentation can also be called a rough segmentation corpus. Call different types of word segmentation tools to perform preliminary word segmentation processing on the original text to obtain multiple initial word segmentations corresponding to different word segmentation tools, and merge the multiple initial word segmentations into an initial word segmentation set. At this time, the initial word segmentation set contains a lot of initial word segmentation data, and the number of initial word segmentations can be reduced by voting for each initial word segmentation. Among them, the voting process can be obtained according to the statistics of the word segmentation tools. For example, for a certain initial word segmentation, three word segmentation tools can all obtain the initial word segmentation from the original text, and the initial word segmentation is used as a quality inspection word segmentation. If the word segmentation results of the three word segmentation tools for the initial word segmentation are inconsistent, the initial word segmentation is directly discarded; if the word segmentation results of two word segmentation tools for the initial word segmentation are consistent, and the word segmentation results of another word segmentation tool are inconsistent, it can be determined that the initial word segmentation is a quality inspection word segmentation.
[0104] When using multiple word segmentation tools for word segmentation, the number of initial word segmentations in the initial word segmentation set can be initially reduced through voting to ensure the effectiveness of quality inspection word segmentation.
[0105] Among them, the word segmentation tool can use open source Chinese word segmentation tools, such as Jieba Chinese word segmentation, Hanlp word segmenter, Harbin Institute of Technology Language Technology Platform (Language Technology Platform, referred to as LTP), a set of Chinese lexical analysis toolkits (THU Lexial Analyzer for Chinese, referred to as THULAC) developed and launched by the Natural Language Processing and Social Humanities Computing Laboratory of Tsinghua University, Stanford word segmenter, natural language processing and information retrieval sharing platform NLPIR, etc. These word segmentation tools have their own word segmentation characteristics. For example, the multiple word segmentation tools called can be LTP, THULAV and NLPIR three word segmentation tools to perform word segmentation processing on the original text.
[0106] The word segmentation module of LTP is trained and decoded based on the CRF model. It models the target sequence based on the observed sequence, and the data source used is the data in the People's Daily from January to June 1998. The word segmentation interface is initialized by obtaining the file path, and the word segmentation interface is called to segment the original text to obtain the quality inspection word segmentation.
[0107] The THULAC toolkit comes with model training raw corpus, but authorization is required. The Chinese word segmentation and part-of-speech tagging functions of the THULAC toolkit are powerful and accurate. It can configure interface parameters and call word segmentation statements to segment the original text to obtain quality inspection word segmentation.
[0108] The NLPIR tool is a full-chain analysis tool that can segment the original text. In the specific segmentation process, it is necessary to import a pre-built dictionary and call the dictionary to perform the initial segmentation to obtain the segmentation result. Furthermore, probabilistic statistical methods and simple rules are used to eliminate ambiguous words, and word frequency information is used to identify unregistered words. After eliminating ambiguity and identifying unregistered words, quality inspection segmentation is obtained.
[0109] In step S320, feature extraction is performed on the quality inspection segmented words to obtain audit keywords.
[0110] After obtaining the quality inspection segmentation words, the quality inspection segmentation words can be further processed for feature extraction to obtain the audit keywords.
[0111] Feature extraction starts from an initial measured data set and then constructs informative and non-redundant derived values, namely eigenvalues.
[0112] Eigenvalues can help to continue the learning process and the inductive step, and in some cases can make it easier for people to make better interpretations of the data. Feature extraction is a dimensionality reduction step, where the initial data set is reduced to more manageable groups (features) for learning, while maintaining the accuracy and completeness of the description of the original data set.
[0113] Specifically, the feature extraction processing method can be implemented by using TF-IDF (term frequency-inverse document frequency) technology.
[0114] TF-IDF is a common weighting technique used in information retrieval and text mining. TF-IDF is a statistical method used to assess the importance of a word to a document set or a document in a corpus. The importance of a word increases in direct proportion to the number of times the word appears in the document, but decreases in inverse proportion to the frequency of the word in the corpus.
[0115] TF-IDF refers to an algorithm for extracting audit keywords from quality inspection segmented words. Specifically, the product of the word frequency corresponding to each quality inspection segmented word in the audit basis, that is, the inverse text frequency, is used as the TF-IDF value of the quality inspection segmented word, and then the audit keywords are extracted based on the product of the TF-IDF value of each quality inspection segmented word and the part-of-speech weight.
[0116] The order of the part-of-speech weights from high to low may be verb>noun>adjective>auxiliary word>pronoun, etc. There may also be other ways of setting the part-of-speech weights, which are not particularly limited in this exemplary embodiment.
[0117] Since the TF-IDF value is proportional to the frequency of the quality inspection word in the audit basis, and inversely proportional to the frequency of the quality inspection word in the entire corpus, audit keywords can be extracted from the audit basis based on this feature of TF-IDF. The method of extracting audit keywords using TF-IDF is simple and fast, which improves the efficiency of extracting audit keywords.
[0118] In this exemplary embodiment, audit keywords are obtained through word segmentation processing and feature extraction processing, which ensures the accuracy of audit keyword extraction and improves the efficiency of audit keyword extraction.
[0119] In an alternative embodiment, Figure 4 A flow chart showing the steps of another method for obtaining audit keywords is shown in FIG4 . The method includes at least the following steps: In step S410 , the audit basis is cleaned to obtain the target audit basis.
[0120] Cleansing is the process of re-examining and verifying the audit basis with the aim of removing duplicate audit basis, correcting errors in the audit basis, and providing data consistency for the audit basis.
[0121] Specifically, the cleaning of the audit basis includes removing incomplete data, removing data format errors, removing abnormal or noisy data, and unifying specifications.
[0122] Among them, incomplete data may include missing or incomplete data, that is, the code of the audit basis is not in the Chinese code range or the text length of the audit basis does not meet the first preset length; data format errors include the display of the audit basis showing garbled characters, or the format of the identity card or other identification information contained in the audit basis does not meet the predetermined format, or there are too many spaces, etc.; abnormal or noisy data includes the text length of the audit basis exceeding the second preset length, or the vocabulary in the audit basis belongs to the preset uncommon vocabulary, etc.
[0123] The first preset length and the second preset length may both be preset according to actual conditions, and this exemplary embodiment does not impose any special limitation on this.
[0124] The cleaning method for incomplete data, data with incorrect data format, and abnormal or noisy data is removal.
[0125] Unified specifications can mean merging synonyms or similar words, and unifying the format of text such as dates, times or regions.
[0126] After cleaning the audit basis, the target audit basis can be obtained.
[0127] In step S420, the target audit basis is segmented to obtain audit keywords.
[0128] After obtaining the target audit basis, you can Figure 3 The word segmentation processing method shown is used to obtain the audit keywords, and the audit keywords can also be obtained by performing the word segmentation processing method of step S310 on the target audit basis. This exemplary embodiment does not make any special limitation on this.
[0129] In this exemplary embodiment, audit keywords are obtained through cleaning and word segmentation processing, which ensures the data validity and format uniformity of the audit keywords and provides data support for the subsequent determination of quality inspection conclusions.
[0130] In step S220, an audit mapping relationship between audit keywords and audit conclusions is determined, and at least two audit probabilities are determined according to the audit mapping relationship.
[0131] In an exemplary embodiment of the present disclosure, after obtaining the audit keywords, the audit mapping relationship between the audit keywords and the audit conclusions may be determined.
[0132] Furthermore, before determining the audit mapping relationship, the audit conclusions may be cleaned. The audit conclusions may be cleaned by removing the audit conclusions that are not within the preset audit conclusion range, or other cleaning methods may be set according to actual needs, which is not specifically limited in this exemplary embodiment.
[0133] Furthermore, the audit mapping relationship between the audit keywords and the audit conclusions after the cleaning process is determined.
[0134] The audit mapping relationship may reflect the inclusion relationship between the audit conclusion and the audit keyword. For example, audit conclusion 1 includes audit keyword 1, audit conclusion 1 includes audit keyword 2, and audit conclusion 2 includes audit keyword 2.
[0135] After the audit mapping relationship is determined, at least two audit probabilities between the audit keywords and the audit conclusions may be determined based on the audit mapping relationship.
[0136] Figure 5 A flow chart showing the steps of a method for determining at least two audit probabilities, such as Figure 5 As shown, the method includes at least the following steps: In step S510, the conclusion probability corresponding to the audit conclusion is determined, and the text probability corresponding to the audit keyword is determined.
[0137] The conclusion probability may be the probability of occurrence of the audit conclusion P(A). The conclusion probability is the prior probability (or marginal probability) of the audit conclusion. The reason why P(A) is called "prior" is that it does not consider any audit keyword factors.
[0138] The text probability is the probability of the review keyword appearing P(B). Correspondingly, P(B) is the prior probability of the review keyword.
[0139] In step S520, the number of audit keywords in the audit conclusion is determined according to the audit mapping relationship.
[0140] For example, the number of audit keywords in the audit conclusion determined according to the audit mapping relationship is shown in Table 2:
[0141] Table 2
[0142]
[0143]
[0144] Among them, the number of conclusions for audit keyword 1 in audit conclusion 1 is number 1, the number of conclusions for audit keyword 1 in audit conclusion 2 is number 2, the number of conclusions for audit keyword 1 in audit conclusion 3 is number 3,..., the number of conclusions for audit keyword 2 in audit conclusion 1 is number 4, the number of conclusions for audit keyword 2 in audit conclusion 2 is number 5, the number of conclusions for audit keyword 2 in audit conclusion 3 is number 6,...
[0145] In step S530, the relevance probability of the audit keyword in the audit conclusion is determined according to the number of conclusions, and the conclusion probability, the text probability and the relevance probability are determined to be at least two audit probabilities.
[0146] The relevant probability may be the probability P(B|A) of the audit keyword being included in the audit conclusion. P(B|A) is the conditional probability of the audit keyword appearing when the audit conclusion is known to appear.
[0147] When the conclusion probability P(A), text probability P(B) and correlation probability P(B|A) are obtained respectively, the conclusion probability P(A), text probability P(B) and correlation probability P(B|A) can be determined as at least two audit probabilities.
[0148] In this exemplary embodiment, at least two audit probabilities corresponding to the audit keywords and the audit conclusions are determined. The determination method is simple and accurate, and is closely related to the application scenario of the audit conclusion review, providing a mathematical theoretical basis for the subsequent determination of the quality inspection conclusion.
[0149] In step S230, probability calculation is performed on at least two audit probabilities to obtain a keyword probability corresponding to the audit keyword.
[0150] In an exemplary embodiment of the present disclosure, after obtaining at least two review probabilities, probability calculation may be performed on the at least two review probabilities to obtain a keyword probability.
[0151] Specifically, the probability of the keyword is obtained by performing probability calculation on at least two review probabilities according to formula (1):
[0152]
[0153] Among them, A and B are random events, event A is the event of the audit conclusion, event B is the event of the audit keyword, and they are not zero. P(A|B) refers to the probability of the audit conclusion appearing when the audit keyword appears.
[0154] Since the conclusion probability P(A), text probability P(B) and correlation probability P(B|A) have been obtained, the probability of the audit conclusion corresponding to the audit keyword can be calculated as P(A|B) according to formula (1).
[0155] It is worth mentioning that the parallel computing capability of cloud computing can also be used to quickly improve the probability calculation speed of keyword probabilities, so as to further improve the quality inspection capability and speed of the audit text.
[0156] Among them, cloud computing refers to the delivery and use model of IT infrastructure, which means obtaining required resources through the network in an on-demand and easily scalable manner; in a broad sense, cloud computing refers to the delivery and use model of services, which means obtaining required services through the network in an on-demand and easily scalable manner. This service can be related to IT and software, the Internet, or other services. Cloud computing is the product of the development and integration of traditional computer and network technologies such as grid computing, distributed computing, parallel computing, utility computing, network storage technologies, virtualization, and load balancing.
[0157] With the development of the Internet, real-time data streams, and the diversification of connected devices, as well as the demand for search services, social networks, mobile commerce, and open collaboration, cloud computing has developed rapidly. Different from the previous parallel distributed computing, the emergence of cloud computing will promote revolutionary changes in the entire Internet model and enterprise management model from a conceptual perspective.
[0158] Furthermore, the parallel capability of cloud computing can be used not only for probability calculations here, but also for other processes that can be implemented using cloud computing, such as weight calculations and probability prediction calculations involved in this disclosure.
[0159] In addition, the keyword probability can also be calculated through machine learning models.
[0160] In an optional embodiment, a machine learning model is used to perform probability calculation on at least two review probabilities to obtain a keyword probability corresponding to the review keyword.
[0161] The machine learning model is trained using the Naive Bayes algorithm. The training of the machine learning model is actually the process of calculating the keyword probability using the corresponding relationship of the audit keywords in the known audit basis.
[0162] Bayes' theorem is a theorem in probability theory that describes the probability of an event occurring given some known conditions. For example, if it is known that cancer is related to life expectancy, Bayes' theorem can be used to accurately calculate the probability of a person getting cancer by knowing their age.
[0163] In machine learning, the Naive Bayes algorithm is a series of probabilistic classifiers based on the Bayesian theorem under the assumption of strong (naive) independence between features. Using the machine learning model trained by the Naive Bayes algorithm to perform probability calculations on at least two audit probabilities can transform the quality inspection problem of audit results into a classification judgment problem.
[0164] Training machine learning models using the Naive Bayes algorithm mainly includes two stages: the preparation stage and the classifier training stage.
[0165] Among them, the task of the preparation stage is to make necessary preparations for the naive Bayes classification. The main work is to determine the feature attributes according to the specific situation, and to divide each feature attribute appropriately, and then manually classify a part of the items to be classified to form a training sample set. The input of this stage is all the data to be classified, and the output is the feature attributes and training samples. This stage is the only stage in the entire naive Bayes classification that needs to be completed manually. The quality of this stage has an important impact on the entire process. The quality of the subsequent classifier is largely determined by the feature attributes, feature attribute division and training sample quality.
[0166] The task of the classifier training phase is to generate a classifier. The main work is to calculate the frequency of occurrence of each category in the training samples and the conditional probability estimate of each feature attribute for each category, and record the results. The input of this phase is the feature attribute and the training samples, and the output is the classifier. This phase is a mechanical phase and can be automatically calculated according to formula (1).
[0167] After the machine learning model is trained, the pre-trained machine learning model can be used to calculate the keyword probability corresponding to the audit keyword.
[0168] In addition, the machine learning model is not limited to training using the Naive Bayes algorithm, but can also be trained using other supervised or unsupervised classification algorithms, and multiple algorithms can also be combined to achieve the training purpose. This exemplary embodiment does not make any special limitations on this.
[0169] In this exemplary embodiment, the keyword probability is calculated using a machine learning model, thereby achieving automatic calculation and intelligent prediction of the keyword probability.
[0170] The use of machine learning models to calculate keyword probabilities can be applied in the training phase and actual use phase of the machine learning model. However, since the keyword probabilities calculated in the training phase can also be recorded, the corresponding keyword probabilities can also be determined in the actual use phase by querying the recorded keyword probabilities.
[0171] In an alternative embodiment, Figure 6 A flow chart showing the steps of a method for determining keyword probability is shown as follows: Figure 6 As shown, the method includes at least the following steps: In step S610, probability calculation is performed on at least two audit probabilities to obtain a probability matrix.
[0172] The probability matrix obtained by probability calculation of at least two audit probabilities can be realized in three steps. First, the keyword probability corresponding to an audit keyword is obtained by using formula (1) or a pre-trained machine learning model. Then, the keyword probabilities corresponding to other audit keywords are calculated in this way to obtain other keyword probabilities. Finally, all the calculated keyword probabilities form a probability matrix. This probability matrix is a Bayesian probability matrix.
[0173] Specifically, the probability matrix can refer to Table 3:
[0174] Table 3
[0175] word Audit conclusion Probability Word 1 Conclusion 1 P11 Word 1 Conclusion 2 P12 Word 2 Conclusion 1 P21 Word 2 Conclusion 2 P22 Word 3 Conclusion 1 P31 Word 3 Conclusion 2 P32 … … …
[0176] Among them, P11, P12, P21, P22, P31 and P32, ..., are the keyword probabilities corresponding to the respective audit keywords. Therefore, a one-dimensional probability matrix can be formed by P11, P12, P21, P22, P31 and P32, ...
[0177] The probability matrix can be executed regularly to ensure the accuracy and adaptability of the probability of keywords in subsequent queries.
[0178] In step S620, the keyword probability corresponding to the audit keyword is determined in the probability matrix.
[0179] After the probability matrix is obtained in the actual application stage, the corresponding keyword probability can be queried from the probability matrix through the audit keyword to be queried.
[0180] In this exemplary embodiment, the keyword probability is determined by querying, which realizes the maximum application of the keyword probability in the training stage, saves the time of determining the keyword probability, and improves the efficiency of determining the keyword probability.
[0181] In step S240, the audit weight of the audit keyword is determined according to the audit keyword and the audit basis.
[0182] In an exemplary embodiment of the present disclosure, the audit weight of the audit keyword may also be determined by the audit keyword and the audit basis.
[0183] In an alternative embodiment, Figure 7 A flowchart showing the steps of a method for determining audit weights is shown in FIG. Figure 7 As shown, the method includes at least the following steps: In step S710, determine the text position of the audit keyword in the audit basis.
[0184] The text position of the audit keyword in the audit basis can be determined by the word number of the audit keyword in the audit basis. For example, when the audit keyword is C and the audit basis is AABC, the text position is 4. In addition, there may be other methods for determining the text position, which are not specifically limited in this exemplary embodiment.
[0185] The text position of audit keywords in the audit basis can be found in Table 4:
[0186] Table 4
[0187] word Audit Basis Location Word 1 Based on 1 Position 1 Word 2 Based on 1 Position 2 Word 3 Based on 1 Location 3 … … …
[0188] In step S720, the review weight of the review keyword is determined according to the text position.
[0189] In an alternative embodiment, Figure 8A flowchart showing the steps of a first method for further determining the audit weight is shown, such as Figure 8 As shown, the method at least includes the following steps: In step S810, a weight mapping relationship between a text position and an audit weight of an audit keyword is determined.
[0190] The weight mapping relationship may represent a proportional relationship between the text position and the audit weight, or an inverse relationship between the text position and the audit weight, and this exemplary embodiment does not specifically limit this.
[0191] The positive relationship means that the later the text is, the greater the review weight is; the negative relationship means that the earlier the text is, the greater the review weight is. Moreover, the greater the review weight is, the greater the impact on the final result is.
[0192] The weight mapping relationship may be queried through a proportional relationship of a direct or inverse relationship or calculated through a proportional relationship, or may be calculated through a functional relationship, and the functional relationship may not be a proportional relationship, and this exemplary embodiment does not make any special limitation on this.
[0193] In step S820, the audit weight of the audit keyword is determined according to the weight mapping relationship and the text position.
[0194] When the weight mapping relationship and text position are known, the review weight of the review keyword can be queried or calculated through the weight mapping relationship.
[0195] It is worth noting that the audit weights of all audit keywords in the audit basis can be determined, or the audit weights of only some audit keywords in the audit basis can be determined to meet the influence of the audit keywords in different application scenarios.
[0196] In this exemplary embodiment, the audit weight can be determined by determining the weight mapping relationship and the text position. The determination method is simple and accurate and can meet the requirements for the audit weight in different situations.
[0197] In an alternative embodiment, Fig. 9 A flowchart showing the steps of a second method for further determining the audit weight is shown in FIG. Fig. 9 As shown, the method at least includes the following steps: In step S910, the position weight of the audit keyword is determined according to the text position.
[0198] The method of using text position to determine position weight is similar to Figure 8 The determination method is the same as shown, that is, first determine the weight mapping relationship between the text position and the position weight of the audit keyword, and then determine the position weight of the audit keyword based on the weight mapping relationship and the text position. Therefore, it will not be repeated here.
[0199] The weight mapping relationship between the text position and the position weight of the audit keyword may be the same as or different from the weight mapping relationship between the text position and the audit weight of the audit keyword, and this exemplary embodiment does not specifically limit this.
[0200] In step S920, the influence weight corresponding to the audit keyword is obtained, and the position weight and the influence weight are weighted to obtain the audit weight.
[0201] The impact weight may be determined based on the business value of the audit keyword, may be manually set, or may be calculated or queried based on the relationship between the audit keyword and the quantitative standard of business value, and this exemplary embodiment does not specifically limit this.
[0202] After the influence weight is obtained, the position weight and the influence weight can be weighted to obtain the audit weight. The weight calculation method can be to multiply the position weight and the influence weight, or to calculate according to the functional relationship between the position weight and the influence weight, which is not specifically limited in this exemplary embodiment.
[0203] Therefore, the factors determining the audit weight are not limited to the position weight, but may also include the position weight, or other weights that are important in actual situations. When other weights also affect the audit weight, the corresponding audit weight can also be obtained by calculation or other methods.
[0204] In this exemplary embodiment, the audit weight is calculated by position weight and influence weight. The calculation method is simple and accurate, which also enriches the method of determining the audit weight and further enriches the application scenarios of the quality inspection method of the audit text.
[0205] In step S250, a probability prediction calculation is performed on the keyword probability and the audit weight to obtain the conclusion probability, and the conclusion probability is used to determine the quality inspection conclusion in the audit conclusion.
[0206] In an exemplary embodiment of the present disclosure, after the keyword probability and the review weight are calculated, a probability prediction calculation may be further performed on the keyword probability and the review weight to obtain a conclusion probability.
[0207] For example, when there is an audit keyword in the audit basis, that is, there is a keyword probability and an audit weight, the probability prediction calculation method can be to multiply a keyword probability and an audit weight to obtain the conclusion probability; when there are at least two audit keywords in the audit basis, there can also be at least two keyword probabilities and at least two audit weights. The probability prediction calculation method can be to respectively multiply at least two keyword probabilities and at least two audit weights to obtain the conclusion probability.
[0208] Specifically, the method of multiplying at least two keyword probabilities and at least two review weights can refer to formula (2):
[0209] P1=P1*W1*P2*W2*P3*W3 (2)
[0210] Among them, P1, P2 and P3 represent at least two keyword probabilities, and W1, W2 and W3 represent at least two review weights.
[0211] If there is only one audit conclusion, after the conclusion probability is calculated, the conclusion probability can be compared with the corresponding probability threshold, and when the conclusion probability meets the requirement of the probability threshold, the audit conclusion is determined to be a quality inspection conclusion. Alternatively, this audit conclusion can be directly determined as a quality inspection conclusion, and this exemplary embodiment does not specifically limit this.
[0212] When there are two or more audit conclusions, the two calculated conclusion probabilities can also be compared to determine the quality inspection conclusion.
[0213] In an alternative embodiment, Fig.10 A flow chart showing the steps of the method for determining the quality inspection conclusion is shown in FIG. Fig.10 As shown, the method includes at least the following steps: in step S1010, the conclusion probability is determined to be at least two, and the at least two conclusion probabilities are determined by at least two audit conclusions.
[0214] When the same audit basis and at least two corresponding audit conclusions are obtained in step S210, or when at least two audit bases and at least two corresponding audit conclusions are obtained, at least two conclusion probabilities can be calculated according to steps S210 to S250.
[0215] Among them, when at least two conclusion probabilities are determined by the same audit basis and at least two audit conclusions, there is an audit mapping relationship between the audit keywords obtained by word segmentation processing of the audit basis and the at least two audit conclusions, and at least two keyword probabilities can be determined, and at least two audit weights can also be determined based on the audit keywords and the audit basis to obtain at least two conclusion probabilities; when at least two conclusion probabilities are determined by at least two audit bases and the corresponding at least two audit conclusions, the audit keywords determined based on the at least two audit bases are different, and the processing of separate audit keywords and audit conclusions is the same as steps S210-S250, therefore, at least two conclusion probabilities can be calculated separately.
[0216] For example, the corresponding relationship between audit keywords and audit weights is shown in Table 5:
[0217] Table 5
[0218] word Weight Word 1 W1 Word 2 W2 Word 3 W3
[0219] Moreover, word 1, word 2 and word 3 correspond to two audit conclusions at the same time. Furthermore, by multiplying the three keyword probabilities P11, P21 and P31 of audit conclusion 1 and the three audit weights W1, W2 and W3, we can get the conclusion probability P1=P11*W1*P21*W2*P31*W3; by multiplying the three keyword probabilities P12, P22 and P32 of audit conclusion 2 and the three audit weights W1, W2 and W3, we can get the conclusion probability P2=P12*W1*P22*W2*P32*W3. Among them, the three keyword probabilities are obtained according to step S230, which will not be repeated here.
[0220] In step S1020, at least two conclusion probabilities are compared to obtain a target conclusion probability, and the target conclusion probability is used to determine a quality inspection conclusion from at least two audit conclusions.
[0221] After obtaining at least two conclusion probabilities, the at least two conclusion probabilities may be compared to determine a target conclusion probability from the at least two conclusion probabilities.
[0222] For example, when there are two conclusion probabilities P1 and P2, the sizes of P1 and P2 can be compared, and the larger conclusion probability can be determined from P1 and P2 as the target conclusion probability.
[0223] Furthermore, the audit conclusion corresponding to the target conclusion probability is determined to be the quality inspection conclusion. That is, when P2 is large, the audit conclusion 2 is determined to be the quality inspection conclusion.
[0224] In this exemplary embodiment, a method for determining the quality inspection probability in the case of at least two conclusion probabilities is provided, and the calculation and determination method is efficient, thereby reducing the cost of determining the quality inspection conclusion.
[0225] After the quality inspection conclusion is determined, it can be used to conduct quality inspection on the manual review conclusion, so as to continuously update the keyword probability according to the quality inspection results and ensure the accuracy and adaptability of the quality inspection conclusion in the long term.
[0226] In an alternative embodiment, Fig.11 A flowchart showing the steps of the method for updating keyword probability is shown as follows: Fig.11 As shown, the method includes at least the following steps: In step S1110, a manual review conclusion corresponding to the review basis is obtained, and the manual review conclusion is compared with the quality inspection conclusion to determine the quality inspection result.
[0227] The manual review conclusion may be a review conclusion determined manually based on the review basis. The comparison between the manual review conclusion and the quality inspection conclusion may be to compare whether the conclusion texts are the same, or to compare whether the semantics of the manual review conclusion and the quality inspection conclusion are the same, so as to determine whether the quality inspection result is consistent or inconsistent with the manual review conclusion.
[0228] In step S1120, the keyword probability is updated according to the quality inspection result.
[0229] In an alternative embodiment, Fig.12 A flowchart of the steps of further updating the keyword probability is shown, such as Fig.12 As shown, the method includes at least the following steps: In step S1210, if the quality inspection result is that the manual review conclusion is different from the quality inspection conclusion, the quality inspection conclusion is reviewed to obtain a review result.
[0230] When the quality inspection result is that the text of the manual review conclusion is different from that of the quality inspection conclusion, or the semantics of the manual review conclusion is different from that of the quality inspection conclusion, it is determined that the manual review conclusion is different from the quality inspection conclusion.
[0231] Furthermore, the quality inspection conclusion can be reviewed manually or by other means to determine whether the quality inspection conclusion is correct.
[0232] In step S1220, if the review result is that the quality inspection conclusion is wrong, the keyword probability is updated.
[0233] When the review result is that the quality inspection conclusion is wrong, it indicates that the determination of the quality inspection conclusion in step S210-step S250 is wrong. Therefore, the machine learning model for calculating the keyword probability can be re-optimized to update the keyword probability using the optimized machine learning model to ensure the accuracy of the quality inspection probability determination.
[0234] In this exemplary embodiment, after the quality inspection conclusion is obtained, in order to ensure the accuracy of the quality inspection conclusion, the keyword probability may be updated regularly to ensure the quality inspection accuracy and adaptability of the audit text.
[0235] In addition, the determined quality inspection conclusions are not only suitable for quality inspection of manual review conclusions, but can also be used as review conclusions in application scenarios, providing an automated and intelligent way to determine review conclusions, thereby improving the timeliness of review conclusion determination in such application scenarios and reducing the cost of review conclusion determination.
[0236] The quality inspection method for the audit text provided in the embodiment of the present disclosure is described in detail below in conjunction with a specific application scenario.
[0237] Fig.13 The flowchart of the steps of the quality inspection method of the audit text in the application scenario is shown as follows: Fig.13As shown, in step S1301, data is collected.
[0238] That is, to obtain the audit conclusion and audit basis. The audit basis and audit conclusion refer to the valid audit basis and audit conclusion, that is, to exclude invalid audit records such as rejection or retrial. Since invalid audit records such as rejection and retrial are marked with corresponding identification information, the rejected or invalid audit records can be identified through the identification information to eliminate the audit records and obtain the valid audit basis and audit conclusion.
[0239] In step S1302, word segmentation processing is performed.
[0240] After obtaining effective audit basis and audit conclusions, the audit basis can be further segmented to obtain audit keywords.
[0241] The word segmentation processing of the audit basis can be realized by using Jieba word segmentation. Although Jieba word segmentation has a certain ability to recognize new words, the new words added by oneself and the corresponding weights can ensure a higher accuracy rate. Therefore, the custom dictionary of Jieba word segmentation is used for word segmentation processing.
[0242] In addition, the word segmentation processing methods based on rules and statistics can be used to obtain quality inspection word segmentation, or a variety of different word segmentation tools can be called as word segmentation models to perform word segmentation processing to obtain quality inspection word segmentation. This exemplary embodiment does not make any special limitations on this.
[0243] In step S1303, data is cleaned.
[0244] Cleansing is the process of re-examining and verifying the audit basis with the aim of removing duplicate audit basis, correcting errors in the audit basis, and providing data consistency for the audit basis.
[0245] Specifically, the cleaning process of the audit basis includes removing incomplete data, removing data format errors, removing abnormal or noise data, and unifying specifications. Among them, incomplete data may include missing or incomplete data, that is, the code of the audit basis is not in the Chinese code interval or the text length of the audit basis does not meet the first preset length; data format errors include garbled characters in the display of the audit basis, or the format of the ID card or other identification information contained in the audit basis does not meet the predetermined format, or there are too many spaces, etc.; abnormal or noise data includes the text length of the audit basis exceeds the second preset length, or the vocabulary in the audit basis belongs to the preset uncommon vocabulary, etc.
[0246] The cleaning method for incomplete data, data with incorrect data format, and abnormal or noisy data is removal.
[0247] Unified specifications can mean merging synonyms or similar words, and unifying the format of text such as dates, times or regions.
[0248] After cleaning the audit basis, the target audit basis can be obtained.
[0249] Therefore, the quality inspection segmentation at this time can be the quality inspection segmentation corresponding to the target audit basis, that is, the quality inspection segmentation obtained in step S1302 can be cleaned by cleaning the audit basis, so as to use the cleaned quality inspection segmentation for feature extraction.
[0250] In step S1304, feature extraction is performed.
[0251] The feature extraction processing method can be implemented by using TF-IDF technology. Specifically, the product of the word frequency corresponding to each quality inspection word in the audit basis, that is, the inverse text frequency, is used as the TF-IDF value of the quality inspection word, and then the audit keywords are extracted according to the product of the TF-IDF value of each quality inspection word and the part-of-speech weight.
[0252] The order of the part-of-speech weights from high to low may be verb>noun>adjective>auxiliary word>pronoun, etc. There may also be other ways of setting the part-of-speech weights, which are not particularly limited in this exemplary embodiment.
[0253] In step S1305, training is performed.
[0254] Specifically, after obtaining the audit keywords, the audit mapping relationship between the audit keywords and the audit conclusions can be determined.
[0255] After the audit mapping relationship is determined, at least two audit probabilities between the audit keywords and the audit conclusions may be determined based on the audit mapping relationship.
[0256] Specifically, the conclusion probability corresponding to the audit conclusion is determined, and the text probability corresponding to the audit keyword is determined. The conclusion probability may be the occurrence probability of the audit conclusion, and the text probability is the occurrence probability of the audit keyword.
[0257] Furthermore, the number of audit keywords in the audit conclusion is determined according to the audit mapping relationship, so as to determine the relevant probability of the audit keywords in the audit conclusion according to the number of conclusions. The relevant probability can be the probability of the audit keyword being included in the audit conclusion, or the conditional probability of the audit keyword appearing when the audit conclusion appears.
[0258] After the conclusion probability, text probability and related probability are determined as at least two review probabilities, the naive Bayes algorithm can be used to train the machine learning model to perform probability calculation on the at least two review probabilities to obtain the keyword probability.
[0259] The machine learning model is trained using the Naive Bayes algorithm. The training of the machine learning model is actually the process of calculating the keyword probability using the corresponding relationship of the audit keywords in the known audit basis.
[0260] In step S1306, data is collected.
[0261] Specifically, obtain the audit basis and audit conclusion currently submitted for quality inspection.
[0262] In step S1307, word segmentation processing is performed.
[0263] That is, the review basis for quality inspection is subjected to the same word segmentation processing as step S1302, which will not be repeated here.
[0264] In step S1308, data is cleaned.
[0265] That is, the audit basis for quality inspection is subjected to the same data cleaning process as step S1303, which will not be repeated here.
[0266] In step S1309, a conclusion is predicted.
[0267] After obtaining the keyword probability in step S1305, a corresponding probability matrix can also be generated according to the keyword probability. Therefore, the probability matrix obtained in step S1305 can be used to query the audit probability corresponding to the audit keyword obtained in step S1308, and the corresponding conclusion probability can be calculated according to the audit weight corresponding to the audit keyword.
[0268] Among them, the audit weight is to first determine the text position of the audit keyword in the audit basis, and then determine the audit weight of the audit keyword based on the text position. Specifically, it can be to first determine the weight mapping relationship between the text position and the audit weight of the audit keyword, and then determine the audit weight of the audit keyword based on the weight mapping relationship and the text position; it can also be to determine the position weight of the audit keyword based on the text position, and then obtain the impact weight corresponding to the audit keyword, and perform weight calculation on the position weight and impact weight to obtain the audit weight.
[0269] After obtaining the conclusion probability, the quality inspection conclusion can be determined using the conclusion probability.
[0270] If there is only one audit conclusion, after calculating the conclusion probability, the conclusion probability can be compared with the corresponding probability threshold, and when the conclusion probability meets the requirements of the probability threshold, the audit conclusion is determined to be a quality inspection conclusion.
[0271] When there are two or more audit conclusions, the two calculated conclusion probabilities can also be compared to determine the quality inspection conclusion.
[0272] In step S1310, manual confirmation is performed.
[0273] After the quality inspection conclusion is determined, it can be used to conduct quality inspection on the manual review conclusion, so as to continuously update the keyword probability according to the quality inspection results and ensure the accuracy and adaptability of the quality inspection conclusion in the long term.
[0274] When the quality inspection result is that the text of the manual review conclusion is different from that of the quality inspection conclusion, or the semantics of the manual review conclusion is different from that of the quality inspection conclusion, it is determined that the manual review conclusion is different from the quality inspection conclusion.
[0275] Furthermore, the quality inspection conclusion can be reviewed manually or by other means to determine whether the quality inspection conclusion is correct.
[0276] When the review result is that the quality inspection conclusion is wrong, it indicates that the determination of the quality inspection conclusion is wrong. Therefore, the machine learning model for calculating the keyword probability can be re-optimized to update the keyword probability using the optimized machine learning model to ensure the accuracy of the quality inspection probability determination.
[0277] use Fig.13 The quality inspection method for the audit text shown takes about 100 milliseconds to perform word quality inspection on the existing audit results, while a single manual quality inspection takes about 2 minutes. Therefore, the quality inspection method for the audit text disclosed in the present invention is 1000 times faster than the manual quality inspection method.
[0278] In addition, judging from the manual review records, the success rate of the quality inspection method of the audit text disclosed in the present invention can be maintained at at least about 83%, which is an extremely high success rate.
[0279] Based on the above application scenarios, it can be seen that the quality inspection method for the audit text provided by the embodiment of the present disclosure, on the one hand, utilizes the audit mapping relationship between the audit keywords and the audit conclusions to determine the keyword probability and audit weight of the audit keywords, thereby realizing the quantification of the correlation between the audit conclusions and the audit basis, and providing the possibility for the quantitative application of the audit basis and the audit conclusion in the text quality inspection scenario; on the other hand, the quantitative keyword probability and audit weight are subjected to probability prediction calculation to obtain the conclusion probability, and the conclusion probability is further used to determine the quality inspection conclusion. This is an automated and intelligent audit text quality inspection method, which improves the quality inspection speed and efficiency, ensures the accuracy and adaptability of the quality inspection conclusion, and greatly reduces the quality inspection cost of the audit text.
[0280] It should be noted that although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.
[0281] The following describes an apparatus embodiment of the present disclosure, which can be used to execute the quality inspection method of the audit text in the above-mentioned embodiment of the present disclosure. For details not disclosed in the apparatus embodiment of the present disclosure, please refer to the embodiment of the quality inspection method of the audit text in the above-mentioned embodiment of the present disclosure.
[0282] Fig.14 The structure block diagram of a quality inspection device for reviewing texts in some embodiments of the present disclosure is schematically shown. Fig.14 As shown, the quality inspection device 1400 for reviewing texts may mainly include: a word segmentation processing module 1410 , a mapping relationship module 1420 , a probability calculation module 1430 , a weight determination module 1440 and a conclusion probability module 1450 .
[0283] The word segmentation processing module 1410 is configured to obtain the audit basis and the audit conclusion, and perform word segmentation processing on the audit basis to obtain the audit keywords; the mapping relationship module 1420 is configured to determine the audit mapping relationship between the audit keywords and the audit conclusion, and determine at least two audit probabilities based on the audit mapping relationship; the probability calculation module 1430 is configured to perform probability calculation on at least two audit probabilities to obtain the keyword probability corresponding to the audit keyword; the weight determination module 1440 is configured to determine the audit weight of the audit keyword based on the audit keyword and the audit basis; the conclusion probability module 1450 is configured to perform probability prediction calculation on the keyword probability and the audit weight to obtain the conclusion probability, and use the conclusion probability to determine the quality inspection conclusion in the audit conclusion.
[0284] In some embodiments of the present disclosure, the weight determination module includes: a position determination submodule configured to determine a text position of the audit keyword in the audit basis;
[0285] The position weight submodule is configured to determine the review weight of the review keyword according to the text position.
[0286] In some embodiments of the present disclosure, the position determination submodule module includes: a weight mapping unit configured to determine a weight mapping relationship between a text position and an audit weight of an audit keyword;
[0287] The mapping determination unit is configured to determine the review weight of the review keyword according to the weight mapping relationship and the text position.
[0288] In some embodiments of the present disclosure, the position determination submodule includes: a candidate weight unit configured to determine the position weight of the audit keyword according to the text position;
[0289] The weight calculation unit is configured to obtain the influence weight corresponding to the audit keyword, and perform weight calculation on the position weight and the influence weight to obtain the audit weight.
[0290] In some embodiments of the present disclosure, the probability calculation module includes: a model calculation submodule, configured to use a machine learning model to perform probability calculation on at least two review probabilities to obtain a keyword probability corresponding to the review keyword.
[0291] In some embodiments of the present disclosure, the probability calculation module includes: a matrix calculation submodule configured to perform probability calculation on at least two audit probabilities to obtain a probability matrix;
[0292] The probability search submodule is configured to determine the keyword probability corresponding to the audit keyword in the probability matrix.
[0293] In some embodiments of the present disclosure, the mapping relationship module includes: a separate probability submodule configured to determine a conclusion probability corresponding to the audit conclusion and determine a text probability corresponding to the audit keyword;
[0294] A quantity determination submodule is configured to determine the number of conclusions of the audit keyword in the audit conclusion according to the audit mapping relationship;
[0295] The audit probability submodule is configured to determine the relevance probability of the audit keyword in the audit conclusion according to the number of conclusions, and determine the conclusion probability, the text probability and the relevance probability as at least two audit probabilities.
[0296] In some embodiments of the present disclosure, the conclusion probability module includes: a probability number submodule configured to determine that the conclusion probability is at least two, the at least two conclusion probabilities being obtained from at least two audit bases and at least two audit conclusions;
[0297] The probability comparison submodule is configured to compare at least two conclusion probabilities to obtain a target conclusion probability, and use the target conclusion probability to determine a quality inspection conclusion from at least two audit conclusions.
[0298] In some embodiments of the present disclosure, the quality inspection device for reviewing texts further includes: a conclusion quality inspection module configured to obtain a manual review conclusion corresponding to the review basis, and compare the manual review conclusion with the quality inspection conclusion to determine a quality inspection result;
[0299] The quality inspection result module is configured to update the keyword probability according to the quality inspection result.
[0300] In some embodiments of the present disclosure, the quality inspection result module includes: a conclusion review submodule configured to review the quality inspection conclusion to obtain a review result if the quality inspection result is that the manual review conclusion is different from the quality inspection conclusion;
[0301] The probability update submodule is configured to update the keyword probability if the review result is that the quality inspection conclusion is wrong.
[0302] In some embodiments of the present disclosure, the word segmentation processing module includes: a quality inspection word segmentation submodule configured to perform word segmentation processing on the audit basis to obtain quality inspection word segments;
[0303] The feature extraction submodule is configured to perform feature extraction processing on the quality inspection segmentation words to obtain the audit keywords.
[0304] In some embodiments of the present disclosure, the word segmentation processing module includes: a data cleaning submodule configured to perform cleaning processing on the audit basis to obtain the target audit basis;
[0305] The target basis submodule is configured to perform word segmentation processing on the target review basis to obtain review keywords.
[0306] The specific details of the quality inspection device for reviewing texts provided in each embodiment of the present disclosure have been described in detail in the corresponding method embodiments, and therefore will not be repeated here.
[0307] Fig.15 A schematic diagram of the structure of a computer system suitable for implementing an electronic device of an embodiment of the present disclosure is shown.
[0308] It should be noted that Fig.15 The computer system 1500 of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.
[0309] like Fig.15 As shown, the computer system 1500 includes a central processing unit (CPU) 1501, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 1502 or the program loaded from the storage part 1508 to the random access memory (RAM) 1503. In the RAM 1503, various programs and data required for system operation are also stored. The CPU 1501, the ROM 1502, and the RAM 1503 are connected to each other through a bus 1504. The input / output (I / O) interface 1505 is also connected to the bus 1504.
[0310] The following components are connected to the I / O interface 1505: an input section 1506 including a keyboard, a mouse, etc.; an output section 1507 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1508 including a hard disk, etc.; and a communication section 1509 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section 1509 performs communication processing via a network such as the Internet. A drive 1510 is also connected to the I / O interface 1505 as needed. A removable medium 1511, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1510 as needed so that a computer program read therefrom is installed into the storage section 1508 as needed.
[0311] In particular, according to an embodiment of the present disclosure, the process described in each method flow chart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains a program code for executing the method shown in the flow chart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication part 1509, and / or installed from a removable medium 1511. When the computer program is executed by a central processing unit (CPU) 1501, various functions defined in the system of the present application are executed.
[0312] It should be noted that the computer-readable medium shown in the embodiment of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by an instruction execution system, device or device or used in combination with it. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a computer-readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate, or transmit programs for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.
[0313] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0314] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.
[0315] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the implementation of the present disclosure.
[0316] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any modifications, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure.
[0317] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.
Claims
1. A quality inspection method for audit text, characterized in that: The method comprises: Obtaining the audit basis and audit conclusion, and performing word segmentation processing on the audit basis to obtain audit keywords; Determine an audit mapping relationship between the audit keyword and the audit conclusion, and determine at least two audit probabilities based on the audit mapping relationship; the at least two audit probabilities include a conclusion probability corresponding to the audit conclusion, a text probability corresponding to the audit keyword, and a relevant probability of the audit keyword in the audit conclusion, the relevant probability is determined based on the number of conclusions of the audit keyword in the audit conclusion, and the number of conclusions is determined based on the audit mapping relationship; Probabilistically calculating the at least two audit probabilities to obtain a keyword probability corresponding to the audit keyword; wherein the keyword probability is determined based on the ratio between the product of the conclusion probability and the relevant probability and the text probability after calculating the product of the conclusion probability and the relevant probability; Determine the audit weight of the audit keyword based on the audit keyword and the basis for quality inspection; A probability prediction calculation is performed on the keyword probability and the audit weight to obtain the conclusion probability corresponding to the audit basis to be quality inspected, and the conclusion probability corresponding to the audit basis to be quality inspected is used to determine the quality inspection conclusion corresponding to the audit basis to be quality inspected in the audit conclusion.
2. The quality inspection method for audit text according to claim 1, characterized in that: The step of determining the audit weight of the audit keyword according to the audit keyword and the audit basis to be inspected includes: Determine the text position of the audit keyword in the audit basis to be inspected; The audit weight of the audit keyword is determined according to the text position.
3. The quality inspection method for audit text according to claim 2, characterized in that: The step of determining the audit weight of the audit keyword according to the text position includes: Determining a weight mapping relationship between the text position and the audit weight of the audit keyword; The audit weight of the audit keyword is determined according to the weight mapping relationship and the text position.
4. The quality inspection method for audit text according to claim 2, characterized in that: The step of determining the audit weight of the audit keyword according to the text position includes: Determining the position weight of the audit keyword according to the text position; The influence weight corresponding to the audit keyword is obtained, and the position weight and the influence weight are weighted to obtain the audit weight.
5. The quality inspection method for audit text according to claim 1, characterized in that: The performing probability calculation on the at least two audit probabilities to obtain a keyword probability corresponding to the audit keyword includes: The at least two audit probabilities are probability calculated using a machine learning model to obtain a keyword probability corresponding to the audit keyword.
6. The quality inspection method for audit text according to claim 1, characterized in that: The performing probability calculation on the at least two audit probabilities to obtain a keyword probability corresponding to the audit keyword includes: Performing probability calculation on the at least two audit probabilities to obtain a probability matrix; The keyword probability corresponding to the audit keyword is determined in the probability matrix.
7. The quality inspection method for audit text according to claim 1, characterized in that: The determining at least two audit probabilities according to the audit mapping relationship includes: Determine a conclusion probability corresponding to the audit conclusion, and determine a text probability corresponding to the audit keyword; Determine the number of conclusions of the audit keyword in the audit conclusion according to the audit mapping relationship; The relevant probability of the audit keyword in the audit conclusion is determined based on the number of conclusions.
8. The quality inspection method for audit text according to claim 1, characterized in that: The using the conclusion probability corresponding to the to-be-inspected-audit basis to determine the quality inspection conclusion corresponding to the to-be-inspected-audit basis in the audit conclusion includes: Determine that the conclusion probabilities corresponding to the quality inspection and audit basis are at least two, and the at least two conclusion probabilities are obtained from at least two audit conclusions; The at least two conclusion probabilities are compared to obtain a target conclusion probability, and the target conclusion probability is used to determine a quality inspection conclusion from among the at least two audit conclusions.
9. The quality inspection method for audit text according to claim 1, characterized in that: The method further comprises: Obtaining a manual review conclusion corresponding to the basis for quality inspection and review, and comparing the manual review conclusion with the quality inspection conclusion to determine a quality inspection result; The keyword probability is updated according to the quality inspection result.
10. The quality inspection method for audit text according to claim 9, characterized in that: The updating of the keyword probability according to the quality inspection result includes: If the quality inspection result is that the manual review conclusion is different from the quality inspection conclusion, review the quality inspection conclusion to obtain a review result; If the review result is that the quality inspection conclusion is wrong, the keyword probability is updated.
11. The quality inspection method for audit text according to claim 1, characterized in that: The word segmentation processing of the audit basis to obtain audit keywords includes: Performing word segmentation processing on the review basis to obtain quality inspection word segmentation; Perform feature extraction processing on the quality inspection segmented words to obtain audit keywords.
12. The quality inspection method for audit text according to claim 1, characterized in that: The word segmentation processing of the audit basis to obtain audit keywords includes: Cleaning the audit basis to obtain the target audit basis; The target audit basis is segmented to obtain audit keywords.
13. A quality inspection device for reviewing texts, characterized in that: The device comprises: A word segmentation processing module is configured to obtain audit basis and audit conclusion, and perform word segmentation processing on the audit basis to obtain audit keywords; a mapping relationship module, configured to determine an audit mapping relationship between the audit keyword and the audit conclusion, and determine at least two audit probabilities according to the audit mapping relationship; the at least two audit probabilities include a conclusion probability corresponding to the audit conclusion, a text probability corresponding to the audit keyword, and a relevant probability of the audit keyword in the audit conclusion, the relevant probability is determined according to the number of conclusions of the audit keyword in the audit conclusion, and the number of conclusions is determined according to the audit mapping relationship; A probability calculation module is configured to perform probability calculation on the at least two audit probabilities to obtain a keyword probability corresponding to the audit keyword; wherein the keyword probability is determined according to a ratio between the product of the conclusion probability and the relevant probability and the text probability after calculating the product of the conclusion probability and the relevant probability; A weight determination module, configured to determine the audit weight of the audit keyword according to the audit keyword and the audit basis to be inspected; The conclusion probability module is configured to perform probability prediction calculation on the keyword probability and the audit weight to obtain the conclusion probability corresponding to the audit basis to be quality inspected, and use the conclusion probability corresponding to the audit basis to be quality inspected to determine the quality inspection conclusion corresponding to the audit basis to be quality inspected in the audit conclusion.
14. A computer readable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the quality inspection method for the audit text described in any one of claims 1 to 12 is implemented.
15. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; Wherein, the processor is configured to execute the quality inspection method of the audit text according to any one of claims 1 to 12 by executing the executable instructions.
16. A computer program product, characterized in that It comprises a computer program, which, when executed by a processor, implements the quality inspection method for the audit text described in any one of claims 1 to 12.
