Sensitive word screening and checking method and device, terminal and storage medium

Abstract

The invention discloses a sensitive word screening and troubleshooting method and device, a terminal and a storage medium. The method comprises the steps that sensitive words are collected in multiple channels, the sensitive words are divided according to categories and grades, and a sensitive word library is established; preprocessing the received to-be-checked text, and performing sensitive word matching on the preprocessed to-be-checked text according to a preset sensitive word matching algorithm and a sensitive word library to obtain a matching result; performing preliminary screening on the to-be-checked text according to the matching result to obtain a preliminary screening result; and constructing a comprehensive judgment model, performing comprehensive violation judgment in combination with the preliminary screening result and a preset graph neural algorithm, and obtaining and outputting a judgment result of the to-be-checked text. According to the method, the comprehensive and dynamic sensitive word library is constructed, and the efficient sensitive word recognition algorithm and various logic combination judgment mechanisms are combined, so that accurate, efficient and real-time screening and troubleshooting of texts are realized, and the health and safety of a network environment are effectively maintained.

Description

Technical Field

[0001] The present application relates to the field of Internet information processing technology, and in particular to a sensitive word screening method, device, terminal and storage medium. Background Art

[0002] Sensitive word screening is an important part of the information security field. With the rapid development of Internet technology, online content is becoming increasingly rich, but the spread of bad information is also increasing, posing a serious threat to the network environment and social stability.

[0003] The traditional screening method based on sensitive word configuration still has some shortcomings, such as single recognition algorithm, high false positive rate and false negative rate, etc. Therefore, it is particularly important to develop a network content screening system that can adaptively learn, dynamically update the sensitive word library, and has efficient recognition capabilities.

[0004] Therefore, how to improve the accuracy and efficiency of sensitive word screening has become a key issue that needs to be urgently addressed in this field. Summary of the invention

[0005] The invention relates to an efficient network content screening and investigation method based on the integration of multi-dimensional sensitive word configuration and intelligent logic. The method realizes accurate, efficient and real-time screening and investigation of text by constructing a comprehensive and dynamic sensitive word library, combining an efficient sensitive word recognition algorithm and a variety of logical combination and judgment mechanisms, and effectively maintains the health and security of the network environment.

[0006] In the first aspect, the present application provides a method for screening and checking sensitive words, which adopts the following technical solution: A method for screening sensitive words includes the following steps: Collect sensitive words through multiple channels, classify the sensitive words according to categories and levels, and establish a sensitive word library; Preprocessing the received text to be checked, and performing sensitive word matching on the preprocessed text to be checked according to a preset sensitive word matching algorithm and the sensitive word library to obtain a matching result; Preliminarily screening the text to be checked according to the matching result to obtain a preliminary screening result; A comprehensive judgment model is constructed, and a comprehensive violation judgment is performed by combining the preliminary screening results and a preset graph neural algorithm to obtain and output the judgment result of the text to be checked.

[0007] By adopting the above-mentioned technical means, sensitive words are collected through multiple channels and a sensitive word library is established. The received text to be checked is pre-processed and matched with sensitive words, and then preliminary screening and comprehensive violation judgment are carried out, thereby improving the accuracy and efficiency of sensitive word screening. It can not only process large-scale complex data, but also maintain a high adaptability in various application scenarios.

[0008] Preferably, the preprocessing of the received text to be checked specifically comprises the following steps: Receiving a text to be checked, and preprocessing the text to be checked, wherein the preprocessing includes removing irrelevant information, text cleaning, and word segmentation processing; Among them, the removal of irrelevant information includes removing advertisements, links and HTML tags, and the text to be checked after removing the irrelevant information is cleaned. The text cleaning includes removing set symbols, removing punctuation marks, removing stop words and unified coding. The text to be checked after the text cleaning is segmented to obtain a word segmentation result.

[0009] By adopting the above-mentioned technical means, the text to be screened is preprocessed and converted into a format that is more suitable for screening and screening, and the text is divided into independent vocabulary units, which provides a basis for subsequent sensitive word matching and can also improve the efficiency of subsequent screening and screening.

[0010] Preferably, the method of performing sensitive word matching on the preprocessed text to be checked and obtaining a matching result based on a preset sensitive word matching algorithm and the sensitive word library specifically includes the following steps: A preset sensitive word matching algorithm is adopted, and the sensitive word matching algorithm includes a Trie tree algorithm. The text to be checked after the preprocessing is traversed, and a comparison is performed in the sensitive word library. A quick search is performed to determine whether the text to be checked after the preprocessing exists. When a matching sensitive word is found, the found sensitive word is recorded in the matching result.

[0011] By adopting the above technical means and using the Trie tree algorithm, it is possible to quickly find out whether sensitive words exist in the text to be checked, thereby improving the matching efficiency while ensuring the matching accuracy.

[0012] Preferably, the preliminary screening of the to-be-checked text according to the matching result to obtain a preliminary screening result specifically includes the following steps: Recording the positions, categories, and matching degrees of all sensitive words in the matching results, and assigning different weights to each sensitive word according to the category and the preset category severity; All sensitive words and the corresponding positions, categories, matching degrees and weights are summarized to obtain sensitive word data, and the text to be checked is preliminarily screened in combination with preset violation judgment rules and the sensitive word data to obtain a preliminary screening result.

[0013] By adopting the above-mentioned technical means, recording the sensitive word information in the matching results can help reduce the burden of subsequent processing and improve the efficiency of the entire screening process.

[0014] Preferably, the violation judgment rule at least includes keyword combination judgment, and the keyword combination judgment specifically includes the following steps: Determine whether the sensitive word data includes a preset sensitive word combination. If included, perform a preliminary violation judgment based on the weight and the matching degree of each sensitive word in the sensitive word combination. The preliminary violation judgment includes respectively calculating a weighted average of the weight and the matching degree. When the weighted average of any one of the weight and the matching degree exceeds a preset threshold, summarize the text to be checked into the preliminary screening result.

[0015] By adopting the above-mentioned technical means, preliminary screening of the text to be checked can be carried out according to the violation judgment rules, which can reduce the workload of the subsequent comprehensive judgment model, thereby speeding up the screening speed, helping to cope with a large amount of screening work and ensuring screening efficiency.

[0016] Preferably, the construction of a comprehensive judgment model combines the preliminary screening results and a preset graph neural algorithm to perform a comprehensive violation judgment, specifically including the following steps: Constructing a graph structure of the text to be checked according to a preset graph neural algorithm and the preliminary screening result, and training and extracting features of the graph structure through the graph neural algorithm to obtain graph features of the graph structure; Extracting word features of the sensitive words according to the preliminary screening results, the word features including the position, category and occurrence frequency of the sensitive words; A comprehensive judgment model is constructed, and the word features and the graph features are combined to perform a comprehensive violation judgment on all the sensitive words in the preliminary screening results.

[0017] By adopting the above-mentioned technical means, the graph structure features and the word features of sensitive words in the text that have been preliminarily screened are extracted, so as to capture the potential relationships and features in the text, thereby improving the accuracy and efficiency of judgment and realizing accurate screening and troubleshooting of the text.

[0018] Preferably, the comprehensive violation judgment of all the sensitive words in the preliminary screening results specifically includes the following steps: The graph features and the word features are feature fused to obtain a comprehensive feature vector, which is input into the comprehensive judgment model. The comprehensive judgment model outputs a prediction result based on the comprehensive feature vector. The prediction result is a violation probability value of the preliminary screening result. The prediction result is processed and a judgment result is output.

[0019] By adopting the above-mentioned technical means, a comprehensive judgment model is used to comprehensively consider the information of multiple features to make more comprehensive and accurate judgment predictions on the text. In the process, the judgment ability and accuracy are continuously optimized, so that the system can adapt to changes in the network environment and new types of violations, and improve the accuracy of subsequent judgments.

[0020] In the second aspect, the present application provides a sensitive word screening device, which adopts the following technical solution: A sensitive word screening device includes the following devices: A sensitive word library building module is used to collect sensitive words through multiple channels, classify the sensitive words according to categories and levels, and build a sensitive word library; A sensitive word matching module is used to pre-process the received text to be checked, and perform sensitive word matching on the pre-processed text to be checked according to a preset sensitive word matching algorithm and the sensitive word library to obtain a matching result; A preliminary screening module is used to perform preliminary screening on the text to be checked based on the matching results to obtain preliminary screening results; a comprehensive judgment module is used to construct a comprehensive judgment model, combine the preliminary screening results and the preset graph neural algorithm to perform comprehensive violation judgment, and obtain and output the judgment result of the text to be checked.

[0021] By adopting the above technical means, setting up the collaborative work of sensitive word library building module, sensitive word matching module, preliminary screening module and comprehensive judgment module, effective screening and screening of sensitive words are achieved. The device can efficiently process large-scale complex data, has high accuracy and real-time performance, and is suitable for sensitive word screening and screening needs in various application scenarios.

[0022] In a third aspect, the present application provides a smart terminal, which adopts the following technical solution: A smart terminal includes a memory and a processor, 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 sensitive word screening method as described above.

[0023] In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium, wherein at least one instruction, at least one program, code set or instruction set is stored in the computer-readable storage medium, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the sensitive word screening method as described above.

[0024] In summary, this application has at least the following beneficial effects: 1. This application collects sensitive words through multiple channels and establishes a sensitive word library, pre-processes and matches sensitive words for the received text to be checked, and then conducts preliminary screening and comprehensive violation judgment, thereby improving the accuracy and efficiency of sensitive word screening. It can not only process large-scale complex data, but also maintain high adaptability in various application scenarios.

[0025] 2. This application combines violation judgment rules and graph neural algorithms to achieve accurate screening and screening of texts to be screened, and improves the accuracy and efficiency of judgment by capturing the potential relationships and features in the text. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a flow chart of the sensitive word screening method of this application; Figure 2 It is a structural diagram of the sensitive word screening and investigation device of the present application. DETAILED DESCRIPTION

[0027] The present application provides a sensitive word screening method, device, terminal and storage medium. In order to make the purpose, technical solution and advantages of the present application clearer, the implementation method of the present application will be further described in detail below.

[0028] The following is a further detailed description of an embodiment of a sensitive word screening method of the present application in conjunction with the drawings of the specification.

[0029] A sensitive word screening method of the present application, such as Figure 1 As shown, the following steps are included: S1. Collect sensitive words through multiple channels, classify them according to categories and levels, and establish a sensitive word library.

[0030] In this embodiment, sensitive words can be collected from multiple dimensions, such as from laws and regulations, industry norms, historical cases, user feedback and other channels, using natural language processing technology (NLP) and social media analysis to automatically mine potential sensitive words in hot topics on the Internet.

[0031] Sensitive words are managed hierarchically by category and level to establish a comprehensive sensitive word library. At the same time, network hot spots and changes in laws and regulations are tracked in real time, and the sensitive word library is updated regularly to maintain its timeliness and accuracy.

[0032] There are several sensitive word libraries, all belonging to the sensitive word library management system. The sensitive word library management system supports operations such as adding, deleting, modifying, enabling, disabling, and querying sensitive words.

[0033] Historical records are kept for the referenced sensitive word libraries. It should be noted that directly deleting a previously referenced sensitive word library is not supported to ensure data integrity and stability. Ensure data integrity and traceability.

[0034] At the same time, the sensitive word library supports import and export functions, facilitating sharing and migration between different systems or businesses.

[0035] S2. Preprocess the received text to be checked. According to the preset sensitive word matching algorithm and the sensitive word library, perform sensitive word matching on the preprocessed text to be checked and obtain the matching result. The specific steps are as follows: S21. Receive the text to be checked and preprocess it. The preprocessing includes removing irrelevant information, text cleaning, and word segmentation.

[0036] Among them, removing irrelevant information includes removing advertisements, links, and HTML tags; Perform text cleaning on the text to be checked after removing irrelevant information. Text cleaning includes removing specified symbols, punctuation marks, stop words, and unifying the encoding.

[0037] Specified symbols include characters without specific meanings, and stop words include words such as "de", "le", "zai", etc. In this embodiment, the encoding is unified as UTF-8.

[0038] Perform word segmentation on the text to be checked after text cleaning. Use word segmentation technology to split the text to be checked into independent lexical units to obtain the word segmentation result, which is convenient for subsequent retrieval and checking.

[0039] S22. Adopt the preset sensitive word matching algorithm to traverse the preprocessed text to be checked. The sensitive word matching algorithm includes the Trie tree algorithm.

[0040] Compare in the sensitive word library and quickly check whether there is a match in the preprocessed text to be checked. When a matching sensitive word is found, record the found sensitive word in the matching result.

[0041] S3. Conduct a preliminary screening of the text to be checked according to the matching result to obtain the preliminary screening result. The specific steps are as follows: S31. Record the positions, categories, and matching degrees of all sensitive words in the matching result; This step helps to reduce the burden of subsequent processing and improve the efficiency of the entire screening and checking process.

[0042] S32. Assign different weights to each sensitive word based on the category and the preset category severity.

[0043] The matching degree includes partial matching and complete matching. In this embodiment, the categories of sensitive words include: Class A sensitive, Class B sensitive, Class C sensitive, and Class D sensitive. The severity of the categories is from low to high, namely, mild, medium, and severe.

[0044] In a specific implementation method, the weights of sensitive words in a text to be checked are distributed as follows: Sensitive word category Severity Weight distribution Class A sensitive medium 3 Class A sensitive serious 5 Class B sensitive slight 1 Class B sensitive serious 4 Class C sensitive slight 1 Class C sensitive serious 5 Class D Sensitive medium 3 Class D Sensitive serious 4 The severity and matching degree of sensitive words are the key factors in weight allocation. The higher the severity, the greater the weight. Similarly, the higher the matching degree, the greater the weight.

[0045] All sensitive words and their corresponding positions, categories, matching degrees and weights are summarized to obtain sensitive word data.

[0046] S33. Preliminary screening is performed on the text to be screened in combination with the preset violation judgment rules and sensitive word data to obtain preliminary screening results, which are texts that may violate the regulations.

[0047] The violation judgment rules at least include keyword combination judgment, which specifically includes the following steps: S331. Determine whether the sensitive word data includes a preset sensitive word combination.

[0048] S332: If included, a preliminary violation judgment is made based on the weight and matching degree of each sensitive word in the sensitive word combination. The preliminary violation judgment includes respectively calculating the weighted average of the weight and matching degree.

[0049] S333. When the weighted average of any one of the weight and the matching degree exceeds a preset threshold, the text to be checked is summarized into the preliminary screening result.

[0050] In another specific method that can be implemented, the violation judgment rules also include sensitive word frequency judgment.

[0051] Sensitive word frequency judgment includes counting the occurrence frequencies of various categories of sensitive words in the sensitive word data. When the occurrence frequency of any category of sensitive words exceeds a preset frequency threshold, the text to be screened is summarized into the preliminary screening result.

[0052] In another specific implementable manner, the violation judgment rule also includes semantic similarity judgment.

[0053] The semantic similarity judgment includes using natural language processing technology to calculate the semantic similarity between the sensitive word data and the sensitive word library. When the similarity exceeds a preset threshold, the text to be screened is summarized into the preliminary screening results.

[0054] In another specific implementation, the violation judgment rule also includes evaluating the user's credit in combination with the user's historical behavior data, where the historical behavior data includes records of published content, reporting records, etc. When the user's credit value is lower than a preset credit threshold, the text to be checked is summarized into the preliminary screening result.

[0055] In a specific implementable method, all of the above-mentioned preliminary violation judgment processes are the preliminary screening process, and the sensitive word data are sequentially subjected to the above-mentioned keyword combination judgment, sensitive word frequency judgment and semantic similarity judgment. When the sensitive word data exceeds the preset threshold of any of the above items, the text to be checked is summarized into the preliminary screening result. If none of them exceeds the above three preset thresholds, the user is evaluated for credit. If it is not lower than the preset credit threshold, the possibility of violation of the text to be checked is directly excluded, and no subsequent comprehensive judgment is performed, and a "normal" judgment result is output.

[0056] S4. Construct a comprehensive judgment model, combine the preliminary screening results and the preset graph neural algorithm to make a comprehensive violation judgment, obtain and output the judgment result of the text to be checked, which specifically includes the following steps: S41, constructing a graph structure of the text to be checked based on the preset graph neural algorithm and the preliminary screening results. Specifically, the nodes in the graph represent the words, sentences or paragraphs in the preliminary screening results, and the edges represent the association between words and the similarity of sentences.

[0057] S42. Train and extract features of the graph structure through a graph neural algorithm to obtain graph features of the graph structure.

[0058] In this embodiment, through the training of the graph neural algorithm, the embedded representation of each node, that is, the low-dimensional vector, is learned. These embedded representations can capture the contextual information and potential semantic relationships in the preliminary screening results.

[0059] Based on the embedded representation of nodes, the graph structure features are further extracted for subsequent logical judgment.

[0060] S42. Extract word features of sensitive words based on the preliminary screening results. The word features include the position, category, and frequency of occurrence of the sensitive words.

[0061] S43. Construct a comprehensive judgment model, combine word features and image features, and make a comprehensive violation judgment on all sensitive words in the preliminary screening results, which specifically includes the following steps: S431: Fusing the graph features and word features to obtain a comprehensive feature vector. The comprehensive feature vector includes the position, category, matching degree, weight of all sensitive words in the preliminary screening results, as well as the semantic information and structural information of the text.

[0062] S432. Input the comprehensive feature vector into the comprehensive judgment model. The comprehensive judgment model outputs a prediction result based on the comprehensive feature vector. The prediction result is a violation probability value of the preliminary screening result.

[0063] S44: Process the prediction result and output the judgment result.

[0064] The judgment conclusion is output based on the prediction results of the comprehensive judgment model. When the violation probability value is greater than the preset probability threshold, the text to be checked is judged to be in violation, otherwise it is not in violation.

[0065] If the judgment result is a violation, the violation type, sensitive word information and violation probability value of the text to be checked will be output. If the judgment result is not a violation, the "normal" conclusion will be output.

[0066] In a specific implementable method, the output judgment result also includes confidence, which indicates the reliability of the judgment result. The confidence is calculated based on the weight and matching degree of the sensitive words. A high confidence usually means that the judgment result is more reliable. When the confidence score of a text to be checked is too low, a secondary judgment is required on the text to be checked, which is conducive to improving the accuracy of the judgment.

[0067] At the same time, the sensitive word library is updated and expanded based on the confidence feedback results, and the comprehensive judgment model is retrained and verified to improve its accuracy and reliability and reduce the probability of misjudgment and missed judgment.

[0068] For the illegal text to be checked, all its sensitive words are displayed. When the user clicks on a sensitive word, the sensitive word can be replaced, deleted, marked or blocked.

[0069] Based on the same inventive concept as above, the present application also discloses a sensitive word screening device, the structure of which is as follows: Figure 2 As shown, the device includes the following modules: A sensitive word library building module is used to collect sensitive words through multiple channels, classify the sensitive words according to categories and levels, and build a sensitive word library; A sensitive word matching module is used to pre-process the received text to be checked, and perform sensitive word matching on the pre-processed text to be checked according to a preset sensitive word matching algorithm and the sensitive word library to obtain a matching result; A preliminary screening module is used to perform preliminary screening on the text to be checked based on the matching results to obtain preliminary screening results; a comprehensive judgment module is used to construct a comprehensive judgment model, combine the preliminary screening results and the preset graph neural algorithm to perform comprehensive violation judgment, and obtain and output the judgment result of the text to be checked.

[0070] In a specific implementation scheme, the sensitive word matching module includes the following units: The first sensitive word matching unit is used to receive the text to be checked, and pre-process the text to be checked, wherein the pre-processing includes removing irrelevant information, text cleaning and word segmentation processing; wherein the removal of irrelevant information includes removing advertisements, links and HTML tags, and the text to be checked after the irrelevant information is removed is cleaned, and the text cleaning includes removing special symbols, removing punctuation marks, removing stop words and unified coding, and the text to be checked after the text cleaning is segmented to obtain a word segmentation processing result.

[0071] The second sensitive word matching unit is used to adopt a preset sensitive word matching algorithm, wherein the sensitive word matching algorithm includes a Trie tree algorithm, traverses the text to be checked after the preprocessing, compares it in the sensitive word library, and quickly searches for whether the text to be checked after the preprocessing exists. When a matching sensitive word is found, the found sensitive word is recorded in the matching result.

[0072] In a specific implementation scheme, the preliminary screening module includes the following units: A first preliminary screening unit, configured to record the positions, categories and matching degrees of all sensitive words in the matching results, and assign different weights to each sensitive word according to the category and a preset category severity; The second preliminary screening unit is used to summarize all sensitive words and the corresponding positions, categories, matching degrees and weights to obtain sensitive word data, and to perform preliminary screening on the text to be checked in combination with preset violation judgment rules and the sensitive word data to obtain preliminary screening results, which include text that may violate the regulations.

[0073] In a specific embodiment, the second preliminary screening unit includes the following subunits: The second preliminary screening subunit is used to determine whether the sensitive word data includes a preset sensitive word combination. If so, a preliminary violation judgment is performed based on the weight and the matching degree of each sensitive word in the sensitive word combination. The preliminary violation judgment includes respectively calculating a weighted average of the weight and the matching degree. When the weighted average of any one of the weight and the matching degree exceeds a preset threshold, the text to be checked is summarized into the preliminary screening result.

[0074] In a specific feasible implementation scheme, the comprehensive judgment module includes the following units: A first comprehensive judgment unit is used to construct a graph structure of the text to be checked according to a preset graph neural algorithm and the preliminary screening result, and train and extract features of the graph structure through the graph neural algorithm to obtain graph features of the graph structure; a second comprehensive judgment unit is used to extract word features of the sensitive words according to the preliminary screening result, and the word features include the position, category and occurrence frequency of the sensitive words; The third comprehensive judgment unit is used to construct a comprehensive judgment model, and to perform a comprehensive violation judgment on all the sensitive words in the preliminary screening results by combining the word features and the graph features.

[0075] In a specific implementation scheme, the third comprehensive judgment unit includes the following subunits: The third comprehensive judgment subunit is used to fuse the graph features and the word features to obtain a comprehensive feature vector, and input the comprehensive feature vector into the comprehensive judgment model. The comprehensive judgment model outputs a prediction result based on the comprehensive feature vector, and the prediction result is a violation probability value of the preliminary screening result. The prediction result is processed and a judgment result is output.

[0076] From the above functional introduction, it can be seen that a sensitive word screening and investigation device in this application has built a comprehensive sensitive word investigation and prediction system, and through the collaborative work of the sensitive word library establishment module, the sensitive word matching module, the preliminary screening module and the comprehensive judgment module, it has achieved effective screening and investigation of sensitive words. The device can efficiently process large-scale complex data, has high accuracy and real-time performance, is suitable for the needs of sensitive word screening and investigation in various application scenarios, and promotes the sustained, rapid and healthy development of the economy and society.

[0077] Based on the same inventive concept mentioned above, an embodiment of the present application also discloses a computer-readable storage medium, which stores at least one instruction, at least one program, code set or instruction set. The at least one instruction, at least one program, code set or instruction set can be loaded and executed by a processor to implement the sensitive word screening method provided in the above method embodiment.

[0078] Also based on the same inventive concept mentioned above, an embodiment of the present application also discloses a computer-readable storage medium, which stores at least one instruction, at least one program, code set or instruction set. The at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the sensitive word screening method as described above.

[0079] Those skilled in the art will appreciate that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in the computer-readable storage medium, which may include, for example, various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0080] The above description is only an optional embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for screening and checking sensitive words, characterized in that: The steps include: Collect sensitive words through multiple channels, classify the sensitive words according to categories and levels, and establish a sensitive word library; Preprocessing the received text to be checked, and performing sensitive word matching on the preprocessed text to be checked according to a preset sensitive word matching algorithm and the sensitive word library to obtain a matching result; Preliminarily screening the text to be checked according to the matching result to obtain a preliminary screening result; A comprehensive judgment model is constructed, and a comprehensive violation judgment is performed by combining the preliminary screening results and a preset graph neural algorithm to obtain and output the judgment result of the text to be checked.

2. The sensitive word screening method according to claim 1 is characterized in that: The preprocessing of the received text to be checked specifically includes the following steps: Receiving a text to be checked, and preprocessing the text to be checked, wherein the preprocessing includes removing irrelevant information, text cleaning, and word segmentation processing; Among them, the removal of irrelevant information includes removing advertisements, links and HTML tags, and the text to be checked after removing the irrelevant information is cleaned. The text cleaning includes removing set symbols, removing punctuation marks, removing stop words and unified coding. The text to be checked after the text cleaning is segmented to obtain a word segmentation result.

3. The sensitive word screening method according to claim 1 is characterized in that: According to the preset sensitive word matching algorithm and the sensitive word library, the preprocessed text to be checked is matched with sensitive words and a matching result is obtained, which specifically includes the following steps: A preset sensitive word matching algorithm is adopted, and the sensitive word matching algorithm includes a Trie tree algorithm. The text to be checked after the preprocessing is traversed, and a comparison is performed in the sensitive word library. A quick search is performed to determine whether the text to be checked after the preprocessing exists. When a matching sensitive word is found, the found sensitive word is recorded in the matching result.

4. The sensitive word screening method according to claim 1 is characterized in that: The preliminary screening of the to-be-screened text according to the matching result to obtain a preliminary screening result specifically includes the following steps: Recording the positions, categories, and matching degrees of all sensitive words in the matching results, and assigning different weights to each sensitive word according to the category and the preset category severity; All sensitive words and the corresponding positions, categories, matching degrees and weights are summarized to obtain sensitive word data, and the text to be checked is preliminarily screened in combination with preset violation judgment rules and the sensitive word data to obtain a preliminary screening result.

5. The sensitive word screening method according to claim 4 is characterized in that: The violation judgment rule at least includes keyword combination judgment, and the keyword combination judgment specifically includes the following steps: Determine whether the sensitive word data includes a preset sensitive word combination. If included, perform a preliminary violation judgment based on the weight and the matching degree of each sensitive word in the sensitive word combination. The preliminary violation judgment includes respectively calculating a weighted average of the weight and the matching degree. When the weighted average of any one of the weight and the matching degree exceeds a preset threshold, summarize the text to be checked into the preliminary screening result.

6. The sensitive word screening method according to claim 3 is characterized in that: The construction of the comprehensive judgment model combines the preliminary screening results and the preset graph neural algorithm to make a comprehensive violation judgment, specifically including the following steps: Constructing a graph structure of the text to be checked according to a preset graph neural algorithm and the preliminary screening result, and training and extracting features of the graph structure through the graph neural algorithm to obtain graph features of the graph structure; Extracting word features of the sensitive words according to the preliminary screening results, the word features including the position, category and occurrence frequency of the sensitive words; A comprehensive judgment model is constructed, and the word features and the graph features are combined to perform a comprehensive violation judgment on all the sensitive words in the preliminary screening results.

7. The sensitive word screening method according to claim 6, characterized in that: The comprehensive violation judgment of all the sensitive words in the preliminary screening results specifically includes the following steps: The graph features and the word features are feature fused to obtain a comprehensive feature vector, which is input into the comprehensive judgment model. The comprehensive judgment model outputs a prediction result based on the comprehensive feature vector. The prediction result is a violation probability value of the preliminary screening result. The prediction result is processed and a judgment result is output.

8. A sensitive word screening and checking device, characterized in that: Includes the following devices: A sensitive word library building module is used to collect sensitive words through multiple channels, classify the sensitive words according to categories and levels, and build a sensitive word library; A sensitive word matching module is used to pre-process the received text to be checked, and perform sensitive word matching on the pre-processed text to be checked according to a preset sensitive word matching algorithm and the sensitive word library to obtain a matching result; A preliminary screening module, used to perform preliminary screening on the text to be checked according to the matching result to obtain a preliminary screening result; The comprehensive judgment module is used to construct a comprehensive judgment model, combine the preliminary screening results and the preset graph neural algorithm to make a comprehensive violation judgment, and obtain and output the judgment result of the text to be checked.

9. An intelligent terminal, characterized in that: It includes a memory and a processor, 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 sensitive word screening method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The readable storage medium 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 sensitive word screening method as described in any one of claims 1 to 7.

Images