Text auditing quality inspection method, device and equipment and storage medium

Abstract

The invention discloses a quality inspection method and device for text auditing, equipment and a storage medium, and belongs to the field of application software. The method comprises the steps of obtaining an auditing sample, wherein the auditing sample comprises text content and an auditing result obtained by auditing the text content; matching a reference sample corresponding to the auditing sample from a historical sample library; the historical sample library is used for storing historical auditing samples, and the historical auditing samples comprise historical text contents and historical auditing results of the historical text contents; the reference sample is the historical auditing sample of which the similarity between the historical text content and the text content meets a condition; and determining the auditing sample as a sample to be subjected to quality inspection according to the historical auditing result of the reference sample. The method can improve the quality inspection accuracy of text auditing.

Description

Technical Field

[0001] The present application relates to the field of application software, and in particular to a quality inspection method, device, equipment and storage medium for text review. Background Art

[0002] User-generated content (UGC) refers to various forms of content created by users rather than professional content creators, such as text, images, videos, comments, etc. It is commonly found on platforms such as social media, forums, and blogs.

[0003] To ensure the security and compliance of platform content, user-generated content must be reviewed. In related technologies, manual review is often employed. This involves manually browsing platform content to obtain review results. A random sample of these results is then subjected to quality inspection, generating a quality inspection report. Any errors found during manual review are then corrected.

[0004] However, the random sampling method for quality inspection is inefficient and has poor accuracy. Summary of the Invention

[0005] This application provides a method, device, equipment, and storage medium for quality inspection of text review, which can improve the accuracy of quality inspection of text review. The technical solution is as follows:

[0006] According to one aspect of the present application, a quality inspection method for text review is provided, the method comprising:

[0007] Obtaining an audit sample, the audit sample including text content and an audit result obtained by auditing the text content;

[0008] Matching a reference sample corresponding to the audit sample from a historical sample library; the historical sample library is used to store historical audit samples, the historical audit samples including historical text content and historical audit results of the historical text content; the reference sample is the historical audit sample whose historical text content meets the similarity condition with the text content;

[0009] According to the historical audit results of the reference sample, the audit sample is determined as a sample to be quality inspected.

[0010] According to another aspect of the present application, a quality inspection device for text review is provided, the device comprising:

[0011] An acquisition module is used to acquire an audit sample, wherein the audit sample includes text content and an audit result obtained by auditing the text content;

[0012] a matching module for matching a reference sample corresponding to the audit sample from a historical sample library; the historical sample library is used to store historical audit samples, the historical audit samples including historical text content and historical audit results of the historical text content; the reference sample is the historical audit sample whose historical text content meets the similarity condition with the text content;

[0013] A determination module is used to determine the audit sample as a sample to be quality inspected based on the historical audit result of the reference sample.

[0014] According to another aspect of the present application, a computer device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the quality inspection method for text review as described above.

[0015] According to another aspect of the present application, a computer-readable storage medium is provided, in which at least one instruction, at least one program, a code set or an instruction set is stored. The at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the quality inspection method for text review as described above.

[0016] According to another aspect of the present application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the quality inspection method for text review provided in various optional implementations of the above aspects.

[0017] The beneficial effects of the technical solution provided by this application include at least:

[0018] After reviewing a document, you can search for similar historical documents from historical review records. By analyzing the review results of the current document with reference to these historical documents, you can determine whether the document requires quality inspection. By referencing historical review samples to screen for quality inspection, you can filter out audit samples whose current review results don't match those of historical review results and use them as samples for quality inspection confirmation. Compared to random sampling quality inspection, this method can accurately screen audit samples with questionable results, conduct precise quality inspections on questionable audit samples, improve quality inspection efficiency, and ensure the accuracy of audit results. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0020] Figure 1 is a schematic diagram of the structure of a computer system provided by an exemplary embodiment of the present application;

[0021] Figure 2 This is a flow chart of a quality inspection method for text review provided by an exemplary embodiment of the present application;

[0022] Figure 3 This is a flow chart of a quality inspection method for text review provided by an exemplary embodiment of the present application;

[0023] Figure 4 This is a flow chart of a quality inspection method for text review provided by an exemplary embodiment of the present application;

[0024] Figure 5 This is a flow chart of a quality inspection method for text review provided by an exemplary embodiment of the present application;

[0025] Figure 6 is a schematic diagram of a quality inspection method for text review provided by an exemplary embodiment of the present application;

[0026] Figure 7 is a schematic diagram of a quality inspection method for text review provided by an exemplary embodiment of the present application;

[0027] Figure 8 This is a structural diagram of a quality inspection device for text review provided by an exemplary embodiment of the present application;

[0028] Figure 9 It is a structural diagram of a computer device provided by an exemplary embodiment of the present application.

[0029] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application. DETAILED DESCRIPTION

[0030] In order to make the objectives, technical solutions and advantages of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0031] Figure 1 A schematic structural diagram of a computer system provided by an exemplary embodiment of the present application is shown. The computer system may include a terminal device 101 and a server 103.

[0032] For example, the quality inspection method for text review shown in the embodiment of the present application can be applied to a terminal device, wherein the terminal device 101 runs an application 102 that supports user-generated content. The terminal device may include a mobile phone, a tablet computer, a notebook computer, a laptop computer, a desktop computer, an all-in-one computer, an Internet of Things device, an intelligent robot workstation, a television, a set-top box, smart glasses, a smart watch, a digital camera, an MP4 player, an MP5 player, a learning machine, a point-reading machine, an electronic paper book, an electronic dictionary, an in-vehicle device, a virtual reality (VR) player, or an augmented reality (AR) player, etc.

[0033] Exemplarily, the quality inspection method for text review provided in this application can be executed by a client in a terminal device. The client is a client of an application that supports user-generated content. Any application with social attributes, comment functions, or information publishing functions supports user-generated content. Therefore, this application is not limited to the type of application. For example, the application may include at least one of the following: a social application, an audio application, a video application, a shopping application, a live broadcast application, an information application, a browser, a local service application, a travel application, a financial application, a game application, and a communication application.

[0034] The terminal device 101 includes a first memory and a first processor. The first memory stores a quality inspection program for text review; the quality inspection program for text review is called and executed by the first processor to implement the quality inspection method for text review provided in this application. The first memory may include, but is not limited to, the following: random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), and electrically erasable programmable read-only memory (EEPROM).

[0035] The first processor may be composed of one or more integrated circuit chips. Alternatively, the first processor may be a general-purpose processor, such as a central processing unit (CPU) or a network processor (NP). Alternatively, the first processor may implement the text review quality inspection method provided in this application by running a program or code.

[0036] In an optional embodiment, the terminal device 101 and the server 103 may be connected to each other via a wired or wireless network.

[0037] The server 103 is used to provide background services for the client of the terminal device 101. Optionally, the server 103 performs the primary computing work and the terminal device 101 performs the secondary computing work; or, the server 103 performs the secondary computing work and the terminal device 101 performs the primary computing work; or, the server 103 and the terminal device 101 use a distributed computing architecture to perform collaborative computing.

[0038] The server 103 may be a single server, a server cluster consisting of multiple servers, or a cloud computing service center.

[0039] Optionally, server 103 includes a second memory and a second processor. A text review quality inspection program is stored in the second memory; the text review quality inspection program is invoked by the second processor to implement the text review quality inspection method provided in this application. Optionally, the second memory may include, but is not limited to, RAM, ROM, PROM, EPROM, and EEPROM. Optionally, the second processor may be a general-purpose processor, such as a CPU or NP.

[0040] Figure 2 This is a flow chart of a quality inspection method for text review provided by an exemplary embodiment of the present application. This method can be used for Figure 1 The terminal device or server shown in FIG.

[0041] Step 210: Obtain an audit sample, where the audit sample includes text content and an audit result obtained by auditing the text content.

[0042] Exemplarily, the method is executed by a computer device, which may be a terminal device or a server.

[0043] Exemplarily, the audit sample is obtained after an initial audit, which can be a manual audit or an automatic audit using an audit algorithm. The audit sample includes: the text content being audited, and the audit results obtained by auditing the text content.

[0044] The text content may be user-generated content (UGC) within the application. The text content may include at least one of text, tables, and emoticons. Emoticons are images or animated images that can be mixed with text content. For example, emoticons may include emojis. For example, the text content may include user-posted comments, posts, articles, and messages.

[0045] User-generated content (UGC) is user-generated, open, and uncontrollable, and may contain illegal, low-quality, or false information. Therefore, it is necessary to review UGC to ensure that the UGC posted on the application complies with regulations.

[0046] In an optional embodiment, the audit is used to determine whether the text content violates a rule, and the audit result may include either a violation or a non-violation. That is, the audit sample may include the text content and a violation label; or the audit sample may include the text content and a non-violation label.

[0047] In other optional embodiments, the review can also be used to perform multi-dimensional annotation of text content. For example, the review results may include at least one of the following: no violation, false information, low-quality information, advertising, and screen-spamming information.

[0048] For example, the method provided in this embodiment can be used to screen samples for quality inspection for each reviewed text content. For example, if there are 100 reviewed text contents, the method provided in this embodiment will obtain 100 review samples corresponding to the 100 text contents, and determine whether each review sample requires quality inspection in turn. Full screening can ensure that no review samples with potential problems are missed, improve the review accuracy, and ensure the quality of user-generated content in the application.

[0049] Step 220: Match the reference sample corresponding to the audit sample from the historical sample library; the historical sample library is used to store historical audit samples, and the historical audit samples include historical text content and historical audit results of the historical text content; the reference sample is a historical audit sample whose historical text content meets the similarity conditions with the text content.

[0050] The historical sample library stores historical audit samples that have been audited. For example, the historical audit samples may be audit samples that have been audited and do not require quality inspection, and / or audit samples that have been audited and have undergone quality inspection. The historical audit samples in the historical sample library cover all types of audit results. For example, if the audit results include violations and non-violations, the historical sample library will contain both historical audit samples with violations and historical audit samples with non-violations.

[0051] For example, a large number of historical audit samples are stored in the historical sample library, which can provide rich historical audit experience. By referring to the historical audit results of similar historical text content in the historical sample library, the audit results of the current audit sample can be simply reviewed. If the audit result of the current audit sample is consistent with the historical audit result, the audit result of the audit sample is accurate and no quality inspection review is required; if the audit result of the current audit sample is inconsistent with the historical audit result, the audit result of the audit sample may be wrong and a quality inspection review is required; if the historical audit results of the historical audit sample corresponding to the current audit sample are confusing (including multiple audit results, and the number of samples of various audit results is similar), the situation of the audit sample is more complicated and a quality inspection review is required.

[0052] Historical audit samples include historical text content and historical audit results. Historical text content refers to text content that has been audited previously; historical audit results refer to the audit results obtained by auditing historical text content. It should be noted that historical audit results are the correct audit results corresponding to the finalized historical text content. That is, historical audit results can be audit results that have undergone quality inspection and review, or audit results that do not require quality inspection and review.

[0053] Optionally, if the audit sample is determined not to require quality inspection through the method provided in this embodiment, the audit sample can be placed in the historical sample library as a historical audit sample. Alternatively, if the audit sample requires quality inspection through the method provided in this embodiment and passes the quality inspection review, the reviewed audit sample can be placed in the historical sample library as a historical audit sample. In other words, the historical audit samples in the historical sample library can be updated in real time to ensure that the historical audit samples in the historical sample library comply with the latest audit rules.

[0054] Exemplarily, the computer device may match similar historical text content from a historical sample library based on the text content in the audit sample, so as to determine whether the audit result of the text content is accurate by referring to the historical audit results of the historical text content.

[0055] The similarity condition may refer to the similarity being greater than a similarity threshold. For example, the similarity between the text content and the historical text content may be calculated, and historical audit samples with similarity greater than the similarity threshold may be selected as reference samples. The similarity condition may also refer to the top n most similar samples, where n is a positive integer. For example, the top n historical text contents most similar to the text content may be selected from a historical sample library. Similarity may be measured using semantic similarity, character similarity, word frequency similarity, etc.

[0056] Alternatively, a matching model can be used to match historical text content similar to the text content from the historical sample library. The matching model can be a neural network model trained to output historical text content similar to the text content from the historical sample library. Alternatively, a large language model can be used to match historical text content similar to the text content from the historical sample library. The historical review sample corresponding to the historical text content is then determined as the reference sample.

[0057] The reference sample is a historical audit sample whose historical text content is similar to the text content of the audit sample. The number of reference samples may be at least one. For example, when there are multiple reference samples, the historical audit results of the multiple reference samples may be comprehensively considered to determine whether the current audit sample requires quality inspection.

[0058] Step 230: Based on the historical audit results of the reference sample, the audit sample is determined as the sample to be quality inspected.

[0059] Whether the audit sample needs to be quality inspected can be determined based on the historical audit results of the reference sample; for example, when there are multiple reference samples and the historical audit results of the multiple reference samples are inconsistent or the historical audit results are scattered, the audit sample needs to be quality inspected.

[0060] You can also determine whether an audit sample requires quality inspection based on the historical audit results of the reference sample and the audit results of the audit sample. That is, based on the historical audit results of the reference sample and the audit results of the audit sample, the audit sample is identified as a sample for quality inspection. If the historical audit results are inconsistent with the audit results, the audit sample requires quality inspection.

[0061] Samples awaiting quality inspection are audit samples that require quality inspection and re-verification. These samples undergo a second review of the audit results during the quality inspection process. If the audit result is incorrect, the audit result will be revised and confirmed as the final audit result for the text content. If the audit result is correct, the audit result will be confirmed as the final audit result for the text content.

[0062] For example, if it is determined based on the reference sample that the audit sample does not need to undergo quality inspection, the quality inspection process is skipped and the audit result is directly determined as the final audit result corresponding to the text content.

[0063] Exemplarily, the text content and its corresponding final audit result may be determined as a historical audit sample, and the historical audit sample may be stored in a historical sample library.

[0064] In summary, the method provided in this embodiment can, after auditing a text, query historical texts similar to the text from historical audit records, analyze the audit results of the current text with reference to the audit results of the historical text, and then determine whether the text needs to be quality inspected. By referring to historical audit samples to screen samples to be quality inspected, audit samples whose current audit results do not match the historical audit results can be screened out and used as samples to be quality inspected for quality inspection confirmation. Compared with random sampling quality inspection, this method can accurately screen audit samples with questionable audit results, conduct precise quality inspection on questionable audit samples, improve quality inspection efficiency, and ensure the accuracy of audit results.

[0065] By way of example, an embodiment of determining samples to be inspected based on historical audit results is provided.

[0066] Figure 3 This is a flow chart of a quality inspection method for text review provided by an exemplary embodiment of the present application. This method can be used for Figure 1 The terminal device or server shown. Figure 2 In the illustrated embodiment, step 230 includes step 231 and step 232 .

[0067] Step 210: Obtain an audit sample, where the audit sample includes text content and an audit result obtained by auditing the text content.

[0068] Exemplarily, the audit result includes one of violation and no violation.

[0069] Step 220: Match the reference sample corresponding to the audit sample from the historical sample library; the historical sample library is used to store historical audit samples, and the historical audit samples include historical text content and historical audit results of the historical text content; the reference sample is a historical audit sample whose historical text content meets the similarity conditions with the text content.

[0070] Exemplarily, the number of reference samples is at least two; and the historical audit results include one of violation and non-violation.

[0071] Step 231: Determine a reference audit result based on historical audit results of at least two reference samples; the reference audit result includes one of violation, no violation, and uncertainty.

[0072] For example, when the vast majority of historical audit results are violations, the reference audit result is a violation; when the vast majority of historical audit results are no violations, the reference audit result is no violation; when the number of violations and no violations in the historical audit results is not much different, the reference audit result is uncertain.

[0073] In an optional embodiment, the difference between the number of violation samples and the number of non-violation samples in at least two reference samples is calculated; the number of violation samples is the number of reference samples with historical audit results indicating violations, and the number of non-violation samples is the number of reference samples with historical audit results indicating non-violation.

[0074] When the difference is greater than a first threshold, the reference audit result is determined to be a violation; the first threshold is a positive number.

[0075] When the difference is less than a second threshold, the reference audit result is determined to be non-violation; the second threshold is a negative number.

[0076] When the difference is greater than the second threshold and less than the first threshold, the reference audit result is determined to be uncertain.

[0077] In another optional embodiment. Calculate the violation sample score of the violation sample in at least two reference samples, and the violation sample score is equal to the ratio of the number of reference samples with historical audit results of violation to the total number of reference samples; calculate the non-violation sample score of the non-violation sample in at least two reference samples, and the non-violation sample score is equal to the ratio of the number of reference samples with historical audit results of non-violation to the total number of reference samples. When the violation sample score is much larger than the non-violation sample score, the reference audit result is violation; when the non-violation sample score is much larger than the violation sample score, the reference audit result is non-violation; when the violation sample score is approximately equal to the non-violation sample score, the reference audit result is uncertain. Among them, the judgment thresholds of much larger than and approximately equal to can be set arbitrarily. For example, when the difference between the two scores is more than 0.8, it is considered to be much larger than; when the difference between the two scores is within 0.8, it is considered to be approximately equal.

[0078] In another optional embodiment, when the proportion of samples with historical audit results indicating violations is higher than a proportion threshold, the reference audit result is determined to be a violation; when the proportion of samples with historical audit results indicating no violations is higher than a proportion threshold, the reference audit result is determined to be no violation; when the proportion of samples with historical audit results indicating violations, and the proportion of samples with historical audit results indicating no violations do not reach the proportion threshold, the reference audit result is determined to be uncertain.

[0079] Step 232: According to the reference audit result, the audit sample is determined as a sample to be inspected.

[0080] In an optional embodiment, if the reference audit result is inconclusive, the audit sample is determined as a sample for quality inspection. When the reference audit result is inconclusive, it indicates that there is no consensus on the historical audit of similar text content. In order to ensure the accuracy of the audit, such text content needs to be quality inspected and reviewed.

[0081] And / or, if the reference audit result is a violation and the audit result is not a violation, the audit sample will be determined as a sample for quality inspection. If the reference audit result is a violation and the audit result is not a violation, it indicates that the audit result may be incorrect and needs to be quality inspected.

[0082] And / or, if the reference audit result is no violation and the audit result is a violation, the audit sample will be determined as a sample for quality inspection. If the reference audit result is no violation, but the audit result is a violation, it also indicates that the audit result may be incorrect and needs to be quality inspected.

[0083] For example, if the reference audit result is no violation and the audit result is no violation, it is determined that the audit sample does not require quality inspection, and the audit result is determined as the final audit result of the text content. If the reference audit result is a violation and the audit result is a violation, it is determined that the audit sample does not require quality inspection, and the audit result is determined as the final audit result of the text content.

[0084] For example, the following formula is used to screen samples for quality inspection.

[0085]

[0086] Among them, Label history For reference to the audit results; num label＝1 is the number of reference samples with a historical audit result of 1; num label＝0 is the number of reference samples whose historical audit results are 0; k is the total number of reference samples; 1 indicates that the audit result is a violation, and 0 indicates that the audit result is not a violation.

[0087] Then, when Label history If it is uncertain, it means that the audit result of the current audit sample is unstable, and it will be determined as a sample to be inspected and pushed to the quality inspection process for review;

[0088] When Label history When it is 1 or 0, the audit results in the audit sample are compared with the Label history Compare, if the audit result is the same as Label history If there is any inconsistency, it will be identified as a sample for quality inspection and pushed to the quality inspection process for review.

[0089] In summary, the method provided in this embodiment can determine the reference audit result based on the historical audit results of multiple reference samples when there are multiple reference samples, and determine whether the audit sample needs quality inspection based on the reference audit result. When the reference audit result is consistent with the audit result, there is no need to directly conduct quality inspection on the audit result; when the reference audit result is inconsistent with the audit result, the audit result needs to be quality inspected. In addition, when various audit results are mixed in the reference audit result, it means that there is no unified audit opinion on similar text content, and this type of text content needs to be quality inspected and reviewed. Compared with random sampling quality inspection, this method can accurately screen audit samples with questionable audit results, conduct precise quality inspection on questionable audit samples, improve quality inspection efficiency, and ensure the accuracy of audit results.

[0090] The method provided in this embodiment uses the historical audit results of similar samples that have been historically audited to vote to generate a reference audit result, and performs a conflict analysis between the reference audit result and the audit result of the audit sample to determine whether the audit sample needs to be quality inspected, thereby improving the efficiency of quality inspection and significantly increasing the number of error corrections in quality inspection.

[0091] An embodiment of matching a reference sample is given.

[0092] Figure 4 This is a flow chart of a quality inspection method for text review provided by an exemplary embodiment of the present application. This method can be used for Figure 1 The terminal device or server shown. Figure 2 or Figure 3 In the illustrated embodiment, step 220 includes step 221 and step 222 .

[0093] Step 210: Obtain an audit sample, where the audit sample includes text content and an audit result obtained by auditing the text content.

[0094] For example, due to the diversity of user-generated content, text content may include text in multiple languages, a mixture of text and emoticons, and a large number of symbols. To accurately match historical text content similar to the text content from the historical sample library, the method provided in this embodiment uses multiple similarity matching algorithms to calculate the similarity between the text content and historical text content, ensuring that historical text content similar to the text content can be accurately matched.

[0095] Step 221: Calculate the similarity between the historical text content and the text content of at least two historical review samples respectively.

[0096] Exemplarily, the historical sample library stores at least two historical audit samples.

[0097] In an optional embodiment, a semantic matching algorithm is called to calculate the semantic similarity between the historical text content and the text content of at least two historical audit samples, and determine the top a historical audit samples with the highest semantic similarity; a is a positive integer.

[0098] Then, the TF IDF (term frequency inverse document frequency) algorithm is called to calculate the term frequency similarity between the historical text content and the text content of the first a historical review samples, and determine the first b historical review samples with the highest term frequency similarity; b is a positive integer, b is less than a, and b is greater than n.

[0099] Then call the BM25 (Best Match 25) algorithm to calculate the similarity between the historical text content and the text content of the first b historical review samples.

[0100] That is, multiple similarity matching algorithms are called in sequence to perform similarity matching on the historical text content in the historical sample library. For example, the first similarity matching algorithm is used to filter out the top a historical audit samples with the highest similarity from the historical sample library; then the second similarity matching algorithm is used to filter out the top b historical audit samples with the highest similarity from the top a historical audit samples; then the third similarity matching algorithm is used to filter out the top n historical audit samples with the highest similarity from the top b historical audit samples to obtain the reference sample.

[0101] Among them, the semantic matching algorithm encodes the text content into a high-dimensional semantic vector through a deep language model (such as BERT (Bidirectional Encoder Representations from Transformers, bidirectional Transformer encoder), Sentence-BERT (sentence-BERT)), and then calculates the similarity between the vectors (such as cosine similarity). For example, a pre-trained model (such as BERT) is used to encode the text content and historical text content respectively to obtain a semantic vector of fixed dimension. The cosine similarity between the text content vector and the content vector of each historical text is calculated (the value range is [-1,1], the closer to 1, the more similar). Output the top a historical text contents with the highest similarity.

[0102] The TF-IDF (Term Frequency-Inverse Document Frequency) algorithm constructs text vectors by statistically analyzing the importance of words in a text (high-frequency words that are rare in other historical texts are given a higher weight). These vectors are then compared using cosine similarity. For example, a vocabulary is constructed for all historical text content, and the TF-IDF weight for each word is calculated: TF (Term Frequency): how often a word appears in all text content (including the current text and historical text content); IDF (Inverse Document Frequency): log(total number of historical text content / number of historical text content containing the word), with the weight of common words being reduced. The current text and historical text content are each represented as a TF-IDF vector (dimension = vocabulary size). The cosine similarity between the text content vector and each historical text content vector is calculated. The top b historical text content with the highest similarity is output.

[0103] The BM25 (Best Match 25) algorithm is an improved version of TF-IDF. It introduces text content length normalization to prevent overmatching of long texts due to high word frequency. For example, the IDF value of each word in all historical text content is calculated (the same as TF-IDF). The BM25 score is calculated for the text content and historical text content: for each word in the text content, its weighted word frequency in the historical text content is calculated: the word frequency (TF) is normalized by the length of the text content (to prevent the dominance of long texts). Hyperparameters are also introduced to adjust word frequency saturation and text length penalties. The weighted scores of all words are accumulated (BM25 scores). The historical text content is sorted in descending order by the BM25 score. The top n historical text content with the highest scores are output.

[0104] Step 222: Determine the top n historical audit samples with the highest similarity as reference samples, where n is a positive integer.

[0105] Step 230: Based on the historical audit results of the reference sample, the audit sample is determined as the sample to be quality inspected.

[0106] In summary, the method provided in this embodiment employs a hybrid similarity matching algorithm to match historical text content similar to the text content from a historical sample library. Because text content is user-generated, and user-generated content is diverse, using a single similarity matching algorithm may not accurately extract the features of user-generated content, nor accurately identify similar historical text content. Therefore, the method provided in this embodiment employs a semantic matching algorithm, a TF IDF algorithm, and a BM25 algorithm to perform similarity matching on text content, improving the accuracy of text content matching. The audit results of the current text content are then analyzed with reference to the audit results of historical text content to determine whether the text requires quality inspection. By referencing historical audit samples to screen for samples to be inspected, audit samples whose current audit results do not match those of historical audit results can be selected as samples for quality inspection confirmation. Compared to random sampling quality inspection, this method can accurately screen audit samples with questionable audit results, conduct precise quality inspections on questionable audit samples, improve quality inspection efficiency, and ensure the accuracy of audit results.

[0107] An embodiment of using a large language model to assist quality inspection is given.

[0108] Figure 5 This is a flow chart of a quality inspection method for text review provided by an exemplary embodiment of the present application. This method can be used for Figure 1 The terminal device or server shown. Figure 2 、 Figure 3 or Figure 4 The illustrated embodiment further includes steps 241 and 242 after step 230 .

[0109] Step 210: Obtain an audit sample, where the audit sample includes text content and an audit result obtained by auditing the text content.

[0110] Step 220: Match the reference sample corresponding to the audit sample from the historical sample library; the historical sample library is used to store historical audit samples, and the historical audit samples include historical text content and historical audit results of the historical text content; the reference sample is a historical audit sample whose historical text content meets the similarity conditions with the text content.

[0111] Step 230: Based on the historical audit results of the reference sample, the audit sample is determined as the sample to be quality inspected.

[0112] Step 241: Call the large language model, analyze the correctness of the audit results based on the audit samples and reference samples, and obtain the reasoning process text and analysis results.

[0113] For example, when the audit results of an audit sample are inaccurate and require quality inspection review, the large language model can be invoked to analyze the audit sample. The audit sample and reference sample are input into the large language model, which then uses the reference sample to analyze whether the audit results in the audit sample are correct. The large language model can be a model with deep thinking capabilities, and can output a textual description of the reasoning process along with the analysis results. The reasoning process text and analysis results output by the large language model can then be provided to quality inspectors, who can then refer to the reference sample, the reasoning process text, and the analysis results to review the audit results of the audit sample, thereby improving quality inspection efficiency.

[0114] Step 242: Determine the reasoning process text and analysis results as quality inspection auxiliary information for the sample to be quality inspected.

[0115] For example, the audit sample and retrieved reference samples are fed into an LLM (Large Language Model) for further analysis. The LLM also provides analysis results and reasoning text for the current audit results, providing quality inspectors with additional information to assist with quality inspections. LLM is a natural language processing model based on artificial intelligence technology that can understand and generate human language and is widely used in tasks such as text generation, translation, question answering, and dialogue systems.

[0116] In summary, the method provided in this embodiment, after obtaining a sample to be inspected, can also use a large language model to analyze the sample to obtain analysis results and inference process text. The analysis results are used to indicate whether the audit results of the sample to be inspected are correct. This provides quality inspectors with more streamlined quality inspection reference information, improving their quality inspection efficiency.

[0117] By way of example, an embodiment is provided of using the quality inspection method for text review provided in an embodiment of the present application to perform quality inspection on a manually reviewed UGC text.

[0118] Audio players incorporate content security review algorithms to audit user-generated content, preventing harmful content from spreading on the platform and damaging the user experience. However, these algorithms typically lack 100% accuracy, requiring human reviewers to conduct manual evaluations. While manual evaluations typically achieve over 99% accuracy, they can still misidentify legitimate content or overlook illegal content. Therefore, a quality control process is necessary to monitor the quality of human review.

[0119] like Figure 6As shown, the application generates a manual review task based on user-generated UGC content. The reviewer then performs a manual review 301 on the manual review task to obtain a manual review tag 302. Subsequently, a mixed text search 303 is performed on the manual review task and the manual review tag. Historical tasks similar to the manual review task are retrieved from historical review records 304, and a conflict analysis is performed on the manual review tag 302 based on the historical tags of the historical tasks. Manual review tasks whose manual review tags 302 conflict with historical tags are identified as tasks awaiting quality inspection, and / or manual review tasks with mixed historical tags are identified as tasks awaiting quality inspection.

[0120] Optionally, a large language model (LLM) is invoked to analyze the quality inspection task with reference to historical tasks, generating an LLM audit prompt 305. This LLM audit prompt 305 may include an analysis result of whether the manual review label 302 obtained by the LLM analysis is correct, as well as the LLM's analytical reasoning process. The LLM audit prompt 305 is sent to the quality inspector as auxiliary quality inspection information for quality inspection 306. The quality inspector can refer to the LLM audit prompt 305, historical audit results, and other information to perform a quality inspection on the manual review label 302, correct any erroneous manual review label 302, and generate a quality inspection report.

[0121] Due to the particularity of UGC text: 1) there are typos, misspellings and Emoji symbols. 2) UGC text length varies greatly. Using only a single search algorithm cannot achieve the best search results. Therefore, Figure 7 As shown, a method combining semantic vector similarity 401 and a string matching algorithm (eg, TF-IDF algorithm 402, BM25 algorithm 403) is used to retrieve similar reference samples from historical audit records based on the audit samples, thereby improving the retrieval effect.

[0122] In summary, the method provided in this embodiment employs a hybrid retrieval technique for UGC text, mitigating the problem of reduced retrieval accuracy caused by the presence of typos, misspellings, emojis, and large variations in text length. LLM is used to analyze quality inspection tasks using historical reference tasks, generating LLM audit prompts and providing more user-friendly audit content to quality inspectors, thereby improving their work efficiency.

[0123] It should be noted that before collecting the user's relevant data and during the process of collecting the user's relevant data, this application can display a prompt interface, pop-up window or output voice prompt information. The prompt interface, pop-up window or voice prompt information is used to remind the user that its relevant data is currently being collected, so that this application only starts to execute the relevant steps of obtaining the user's relevant data after obtaining the user's confirmation operation on the prompt interface or pop-up window. Otherwise (that is, when the user's confirmation operation on the prompt interface or pop-up window is not obtained), the relevant steps of obtaining the user's relevant data are terminated, that is, the user's relevant data is not obtained. In other words, all user data collected by this application are collected with the user's consent and authorization, and the collection, use and processing of relevant user data need to comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0124] It should be noted that the order of the method steps provided in the embodiments of the present application can be appropriately adjusted, and the steps can be increased or decreased according to the circumstances. Figures 2 to 5 The steps in the illustrated embodiments may also be arbitrarily combined to obtain new embodiments. Any person skilled in the art can easily come up with variations within the technical scope disclosed in this application, and all such variations should be included within the scope of protection of this application, so these variations will not be described in detail.

[0125] Figure 8 This is a schematic diagram of a quality inspection device for text review provided by an exemplary embodiment of the present application. The device includes:

[0126] An acquisition module 1001 is configured to acquire an audit sample, wherein the audit sample includes text content and an audit result obtained by auditing the text content;

[0127] Matching module 1002, for matching a reference sample corresponding to the audit sample from a historical sample library; the historical sample library is used to store historical audit samples, the historical audit samples including historical text content and historical audit results of the historical text content; the reference sample is the historical audit sample whose historical text content meets the similarity condition with the text content;

[0128] The determination module 1003 is configured to determine the audit sample as a sample to be quality inspected based on the historical audit result of the reference sample.

[0129] In an optional embodiment, the number of the reference samples is at least two; the audit result includes one of a violation and a non-violation; and the historical audit result includes one of a violation and a non-violation;

[0130] The determination module 1003 is configured to determine a reference audit result based on the historical audit results of the at least two reference samples; the reference audit result includes one of violation, no violation, and uncertainty;

[0131] The determining module 1003 is configured to determine the audit sample as the sample to be quality inspected based on the reference audit result.

[0132] In an optional embodiment, the determination module 1003 is configured to calculate a difference between the number of violation samples and the number of non-violation samples in the at least two reference samples; the number of violation samples is the number of reference samples whose historical audit results indicate violations, and the number of non-violation samples is the number of reference samples whose historical audit results indicate non-violation;

[0133] The determining module 1003 is configured to determine the reference audit result as a violation if the difference is greater than a first threshold value; the first threshold value is a positive number;

[0134] The determining module 1003 is configured to determine the reference audit result as non-violation if the difference is less than a second threshold value; the second threshold value is a negative number;

[0135] The determination module 1003 is configured to determine the reference audit result as uncertain when the difference is greater than the second threshold and smaller than the first threshold.

[0136] In an optional embodiment, the determination module 1003 is configured to determine the audit sample as the sample to be quality inspected when the reference audit result is uncertain.

[0137] In an optional embodiment, the determining module 1003 is configured to determine the audit sample as the sample to be inspected if the reference audit result is a violation and the audit result is no violation;

[0138] And / or, when the reference audit result is no violation and the audit result is a violation, the audit sample is determined as the sample to be inspected.

[0139] In an optional embodiment, the historical sample library stores at least two historical audit samples;

[0140] The matching module 1002 is configured to calculate the similarity between the historical text contents of the at least two historical review samples and the text content;

[0141] The matching module 1002 is configured to determine the first n historical audit samples with the highest similarity as the reference samples, where n is a positive integer.

[0142] In an optional embodiment, the matching module 1002 is configured to call a semantic matching algorithm to respectively calculate the semantic similarity between the historical text content of the at least two historical audit samples and the text content, and determine the top a historical audit samples with the highest semantic similarity; wherein a is a positive integer;

[0143] The matching module 1002 is configured to call a term frequency inverse document frequency (TF IDF) algorithm to calculate the term frequency similarity between the historical text content of the first a historical review samples and the text content, and determine the first b historical review samples with the highest term frequency similarity; b is a positive integer, which is less than a and greater than n;

[0144] The matching module 1002 is used to call the best matching BM25 algorithm to calculate the similarity between the historical text content of the first b historical review samples and the text content.

[0145] In an optional embodiment, the device further includes:

[0146] An analysis module 1004 is configured to call a large language model, analyze the correctness of the audit result based on the audit sample and the reference sample, and obtain a reasoning process text and an analysis result;

[0147] The analysis module 1004 is configured to determine the reasoning process text and the analysis result as auxiliary quality inspection information of the sample to be quality inspected.

[0148] It should be noted that the quality inspection device for text review provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the quality inspection device for text review provided in the above embodiment and the quality inspection method embodiment for text review belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0149] An embodiment of the present application further provides a computer device comprising: a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set, or instruction set, and the at least one instruction, at least one program, code set, or instruction set is loaded and executed by the processor to implement the quality inspection methods for text review provided in the above-mentioned method embodiments. The computer device can be implemented as a terminal device.

[0150] For example, Figure 9 It is a structural diagram of a computer device provided by an exemplary embodiment of the present application.

[0151] Typically, the computer device 1700 includes a processor 1701 and a memory 1702 .

[0152] The processor 1701 may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. The processor 1701 may be implemented in at least one hardware form of a DSP (Digital Signal Processing), an FPGA (Field-Programmable Gate Array), or a PLA (Programmable Logic Array). The processor 1701 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 1701 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 1701 may also include an AI (Artificial Intelligence) processor, which is used to handle computing operations related to machine learning.

[0153] Memory 1702 may include one or more computer-readable storage media, which may be non-transitory. Memory 1702 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices and flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in memory 1702 is used to store at least one instruction, which is used to be executed by processor 1701 to implement the quality inspection method for text review provided in the method embodiment of the present application.

[0154] In some embodiments, computer device 1700 may optionally include a peripheral device interface 1703 and at least one peripheral device. Processor 1701, memory 1702, and peripheral device interface 1703 may be connected via a bus or signal lines. Each peripheral device may be connected to peripheral device interface 1703 via a bus, signal lines, or circuit boards. Specifically, the peripheral device may include at least one of a radio frequency circuit 1704, a display screen 1705, a camera assembly 1706, an audio circuit 1707, and a power supply 1708.

[0155] The peripheral device interface 1703 can be used to connect at least one I / O (Input / Output)-related peripheral device to the processor 1701 and the memory 1702. In some embodiments, the processor 1701, the memory 1702, and the peripheral device interface 1703 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 1701, the memory 1702, and the peripheral device interface 1703 can be implemented on separate chips or circuit boards, which is not limited in this embodiment of the present application.

[0156] RF circuit 1704 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. RF circuit 1704 communicates with communication networks and other communication devices via electromagnetic signals. RF circuit 1704 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals into electrical signals. RF circuit 1704 may optionally include an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, and the like. RF circuit 1704 may communicate with other computer devices via at least one wireless communication protocol. Such wireless communication protocols include, but are not limited to, the World Wide Web, metropolitan area networks, intranets, various generations of mobile communication networks (2G, 3G, 4G, and 5G), wireless local area networks, and / or WiFi (Wireless Fidelity) networks. In some embodiments, RF circuit 1704 may also include circuitry related to Near Field Communication (NFC), although this application does not limit this.

[0157] The display screen 1705 is used to display a UI (User Interface). The UI may include graphics, text, icons, videos, and any combination thereof. When the display screen 1705 is a touch screen display, the display screen 1705 also has the ability to collect touch signals on the surface or above the surface of the display screen 1705. The touch signal can be input as a control signal to the processor 1701 for processing. At this time, the display screen 1705 can also be used to provide virtual buttons and / or virtual keyboards, also known as soft buttons and / or soft keyboards. In some embodiments, there can be one display screen 1705, which is set on the front panel of the computer device 1700; in other embodiments, there can be at least two display screens 1705, which are respectively set on different surfaces of the computer device 1700 or in a folding design; in still other embodiments, the display screen 1705 can be a flexible display screen, which is set on the curved surface or folding surface of the computer device 1700. Even more, the display screen 1705 can be set as a non-rectangular irregular shape, that is, a special-shaped screen. The display screen 1705 can be made of materials such as LCD (Liquid Crystal Display) and OLED (Organic Light-Emitting Diode).

[0158] The camera assembly 1706 is used to capture images or videos. Optionally, the camera assembly 1706 includes a front camera and a rear camera. Typically, the front camera is set on the front panel of the computer device 1700, and the rear camera is set on the back of the computer device. In some embodiments, there are at least two rear cameras, which are any one of a main camera, a depth of field camera, a wide-angle camera, and a telephoto camera, so as to realize the fusion of the main camera and the depth of field camera to realize the background blur function, the fusion of the main camera and the wide-angle camera to realize panoramic shooting and VR (Virtual Reality) shooting function or other fusion shooting functions. In some embodiments, the camera assembly 1706 may also include a flash. The flash can be a single-color temperature flash or a dual-color temperature flash. A dual-color temperature flash refers to a combination of a warm light flash and a cold light flash, which can be used for light compensation under different color temperatures.

[0159] The audio circuit 1707 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, and convert the sound waves into electrical signals that are input into the processor 1701 for processing, or input into the radio frequency circuit 1704 to achieve voice communication. For the purpose of stereo sound collection or noise reduction, there may be multiple microphones, each located in different parts of the computer device 1700. The microphone may also be an array microphone or an omnidirectional collection microphone. The speaker is used to convert electrical signals from the processor 1701 or the radio frequency circuit 1704 into sound waves. The speaker may be a traditional thin film speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can not only convert electrical signals into sound waves audible to humans, but also convert electrical signals into sound waves inaudible to humans for purposes such as distance measurement. In some embodiments, the audio circuit 1707 may also include a headphone jack.

[0160] Power supply 1708 is used to power various components in computer device 1700. Power supply 1708 can be AC ​​power, DC power, a disposable battery, or a rechargeable battery. When power supply 1708 includes a rechargeable battery, the rechargeable battery can be wired or wirelessly rechargeable. A wired rechargeable battery is charged via a wired line, while a wireless rechargeable battery is charged via a wireless coil. The rechargeable battery can also support fast charging technology.

[0161] In some embodiments, the computer device 1700 further includes one or more sensors 1709 , including but not limited to an acceleration sensor 1710 , a gyroscope sensor 1711 , a pressure sensor 1712 , an optical sensor 1713 , and a proximity sensor 1714 .

[0162] The accelerometer 1710 can detect the magnitude of acceleration along the three coordinate axes of the coordinate system established by the computer device 1700. For example, the accelerometer 1710 can be used to detect the components of gravity acceleration along the three coordinate axes. The processor 1701 can control the touch screen display 1705 to display the user interface in a landscape or portrait view based on the gravity acceleration signal collected by the accelerometer 1710. The accelerometer 1710 can also be used to collect game or user motion data.

[0163] The gyroscope sensor 1711 can detect the orientation and rotation angle of the computer device 1700. It can also work with the accelerometer 1710 to collect 3D motions of the user on the computer device 1700. Based on the data collected by the gyroscope sensor 1711, the processor 1701 can implement the following functions: motion sensing (e.g., changing the UI based on the user's tilt), image stabilization during shooting, game control, and inertial navigation.

[0164] The pressure sensor 1712 can be installed on the side frame of the computer device 1700 and / or below the touch screen display 1705. When the pressure sensor 1712 is installed on the side frame of the computer device 1700, it can detect the user's grip signal of the computer device 1700, and the processor 1701 can perform left and right hand recognition or shortcut operations based on the grip signal collected by the pressure sensor 1712. When the pressure sensor 1712 is installed below the touch screen display 1705, the processor 1701 controls the operational controls on the UI interface based on the user's pressure operation on the touch screen display 1705. The operational controls include at least one of a button control, a scroll bar control, an icon control, and a menu control.

[0165] Optical sensor 1713 is used to detect ambient light intensity. In one embodiment, processor 1701 can control the display brightness of touchscreen display 1705 based on the ambient light intensity detected by optical sensor 1713. Specifically, when the ambient light intensity is high, the display brightness of touchscreen display 1705 is increased; when the ambient light intensity is low, the display brightness of touchscreen display 1705 is decreased. In another embodiment, processor 1701 can also dynamically adjust the shooting parameters of camera assembly 1706 based on the ambient light intensity detected by optical sensor 1713.

[0166] Proximity sensor 1714, also known as a distance sensor, is typically located on the front panel of computer device 1700. Proximity sensor 1714 is used to detect the distance between the user and the front of computer device 1700. In one embodiment, when proximity sensor 1714 detects that the distance between the user and the front of computer device 1700 is gradually decreasing, processor 1701 controls touchscreen display 1705 to switch from the screen-on state to the screen-off state. When proximity sensor 1714 detects that the distance between the user and the front of computer device 1700 is gradually increasing, processor 1701 controls touchscreen display 1705 to switch from the screen-off state to the screen-on state.

[0167] Those skilled in the art will understand that Figure 9 The structure shown in the figure does not constitute a limitation on the computer device 1700, and the computer device 1700 may include more or fewer components than shown in the figure, or combine some components, or adopt a different component arrangement.

[0168] A computer-readable storage medium is also provided in an embodiment of the present application, which stores at least one instruction, at least one program, code set or instruction set. When the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor of a computer device, the quality inspection method for text review provided by the above-mentioned method embodiments is implemented.

[0169] The present application also provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the quality inspection method for text review provided in each of the above method embodiments.

[0170] Those skilled in the art will understand that all or part of the steps of implementing the above embodiments may be accomplished by hardware, or may be accomplished by a program instructing the relevant hardware, and the program may be stored in a computer-readable storage medium, and the above-mentioned readable storage medium may be a read-only memory, a disk or an optical disk, etc.

[0171] The above description is merely an optional embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent switches, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A quality inspection method for text review, characterized in that: The method comprises: Obtaining an audit sample, the audit sample including text content and an audit result obtained by auditing the text content; Matching a reference sample corresponding to the audit sample from a historical sample library; the historical sample library is used to store historical audit samples, the historical audit samples including historical text content and historical audit results of the historical text content; the reference sample is the historical audit sample whose historical text content meets the similarity condition with the text content; According to the historical audit results of the reference sample, the audit sample is determined as a sample to be quality inspected.

2. The method according to claim 1, characterized in that The number of the reference samples is at least two; the audit result includes one of a violation and no violation, and the historical audit result includes one of a violation and no violation; Determining the audit sample as a sample to be inspected based on the historical audit result of the reference sample includes: Determining a reference audit result based on historical audit results of the at least two reference samples; the reference audit result includes one of violation, no violation, and uncertainty; According to the reference audit result, the audit sample is determined as the sample to be quality inspected.

3. The method according to claim 2, characterized in that Determining the reference audit result based on the historical audit results of the at least two reference samples includes: Calculating the difference between the number of violation samples and the number of non-violation samples in the at least two reference samples; the number of violation samples is the number of reference samples whose historical audit results indicate violations, and the number of non-violation samples is the number of reference samples whose historical audit results indicate non-violation; If the difference is greater than a first threshold, determining the reference audit result as a violation; the first threshold is a positive number; If the difference is less than a second threshold, determining the reference audit result as no violation; the second threshold is a negative number; When the difference is greater than the second threshold and smaller than the first threshold, the reference audit result is determined to be uncertain.

4. The method according to claim 2, characterized in that Determining the audit sample as the sample to be inspected based on the reference audit result includes: In the case that the reference audit result is uncertain, the audit sample is determined as the sample to be quality inspected.

5. The method according to claim 2, characterized in that Determining the audit sample as the sample to be inspected based on the reference audit result includes: If the reference audit result is a violation and the audit result is no violation, the audit sample is determined as the sample to be inspected; And / or, when the reference audit result is no violation and the audit result is a violation, the audit sample is determined as the sample to be inspected.

6. The method according to any one of claims 1 to 5, characterized in that: The historical sample library stores at least two historical audit samples; The matching of the reference sample corresponding to the audit sample from the historical sample library includes: respectively calculating similarities between the historical text contents of the at least two historical review samples and the text content; The first n historical audit samples with the highest similarity are determined as the reference samples, where n is a positive integer.

7. The method according to claim 6, characterized in that The respectively calculating the similarity between the historical text contents of the at least two historical review samples and the text content includes: Invoking a semantic matching algorithm to calculate the semantic similarity between the historical text content of the at least two historical audit samples and the text content, and determining the top a historical audit samples with the highest semantic similarity; where a is a positive integer; Calling the TF-IDF algorithm to calculate the word frequency similarity between the historical text content of the first a historical audit samples and the text content, and determining the first b historical audit samples with the highest word frequency similarity; b is a positive integer, b is less than a, and b is greater than n; The best matching BM25 algorithm is called to respectively calculate the similarity between the historical text content of the first b historical review samples and the text content.

8. The method according to any one of claims 1 to 5, characterized in that: The method further comprises: Calling a large language model, analyzing the correctness of the audit result based on the audit sample and the reference sample, and obtaining a reasoning process text and an analysis result; The reasoning process text and the analysis result are determined as quality inspection auxiliary information of the sample to be quality inspected.

9. A quality inspection device for text review, characterized in that: The device comprises: An acquisition module is used to acquire an audit sample, wherein the audit sample includes text content and an audit result obtained by auditing the text content; a matching module for matching a reference sample corresponding to the audit sample from a historical sample library; the historical sample library is used to store historical audit samples, the historical audit samples including historical text content and historical audit results of the historical text content; the reference sample is the historical audit sample whose historical text content meets the similarity condition with the text content; A determination module is used to determine the audit sample as a sample to be quality inspected based on the historical audit result of the reference sample.

10. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores at least one program, and the at least one program is loaded and executed by the processor to implement the quality inspection method for text review as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that The readable storage medium stores at least one program, and the at least one program is loaded and executed by the processor to implement the quality inspection method for text review according to any one of claims 1 to 8.

12. A computer program product, characterized in that The computer program product includes computer instructions, which are stored in a computer-readable storage medium. The processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the quality inspection method for text review as described in any one of claims 1 to 8.

Images