Creative activity planning and copy management system and method

Abstract

The application relates to the technical field of planning copy management, and discloses a creative activity planning copy management method and system. First, the user uploads the to-be-audited creative activity planning copy, and the text cleaning tool is used to remove redundant spaces and special characters, so as to ensure data purity. Then, the cleaned copy is scanned character by character by using a sensitive word Trie tree, and potential sensitive words and context information thereof are identified. Based on the information, the semantic embedding coding and converter model are further used to intelligently analyze the potential sensitive words and evaluate whether the potential sensitive words are in violation. Finally, a detailed copy audit report is automatically generated to help the user quickly understand the specific problems of the copy and make corresponding adjustments. In this way, the audit efficiency and accuracy can be improved, and the problems of false positives, false negatives and insufficient processing of complex formats in the traditional method can be solved.

Description

Technical Field

[0001] This application relates to the field of planning document management technology, and more specifically, to a creative event planning document management system and method. Background Technology

[0002] In the planning and management of modern creative events, the review and management of copywriting plays a crucial role. With the development of the internet and the accelerated speed of information dissemination, businesses, organizations, and individuals increasingly rely on online platforms for publicity and promotion. This makes the quality and compliance of creative event planning copywriting a key factor. However, traditional copywriting review methods mainly rely on manual review, which is not only inefficient but also prone to human error, leading to the failure to promptly identify sensitive or inappropriate content.

[0003] In existing technologies, copywriting review often relies on simple keyword matching for initial screening. While this method can identify obvious sensitive words to some extent, its ability to understand context is limited, making it prone to false positives or false negatives. For example, some words may be perfectly appropriate in a specific context but could be considered sensitive in others. Furthermore, manual review is not only time-consuming and labor-intensive but also requires reviewers to possess high levels of expertise and sensitivity, posing a significant challenge for large-scale copywriting review.

[0004] Therefore, we look forward to an optimized method for managing creative event planning documents. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a creative event planning document management system and method. This system preprocesses the document using a text cleaning tool, detects potential sensitive words using a sensitive word Trie tree, and performs intelligent analysis based on context to determine compliance, finally generating an audit report. This improves audit efficiency and accuracy, solving the problems of false positives, missed positives, and insufficient ability to handle complex formats in traditional methods.

[0006] According to one aspect of this application, a method for managing creative event planning documents is provided, comprising: receiving creative event planning documents uploaded by users for review; preprocessing the creative event planning documents for review using a text cleaning tool to obtain cleaned creative event planning documents; detecting potential sensitive words in the cleaned creative event planning documents based on a sensitive word Trie tree to obtain a set of potential sensitive word detection results, wherein the potential sensitive word detection results include potential sensitive words, sensitive word positions, and sensitive word context information; and performing contextual intelligent processing on the potential sensitive words based on the sensitive word context information in the potential sensitive word detection results. The analysis yields semantic compliance analysis results for potential sensitive words, which indicate whether potential sensitive words violate regulations. Based on these results, a document review report is generated. Specifically, based on a sensitive word Trie tree, potential sensitive words in the cleaned creative event planning document are detected to obtain a set of potential sensitive word detection results, including: acquiring a sensitive word library; constructing the sensitive word library into a sensitive word Trie tree; and scanning each character in the cleaned creative event planning document character by character based on the sensitive word Trie tree to obtain the set of potential sensitive word detection results.

[0007] In the above-mentioned creative event planning document management method, the creative event planning document to be reviewed is preprocessed using a text cleaning tool to obtain a cleaned creative event planning document, including: removing extra spaces and special characters from the creative event planning document to be reviewed.

[0008] In the above-mentioned creative event planning document management method, based on the sensitive word Trie tree, each character in the cleaned creative event planning document is scanned character by character to obtain the set of potential sensitive word detection results. This includes: for each character in the cleaned creative event planning document, searching along the path starting from the root node of the sensitive word Trie tree; if the character has a corresponding child node in the sensitive word Trie tree, continuing to search for the next character, and if an end node is reached, a potential sensitive word is matched; if the character does not have a corresponding child node in the sensitive word Trie tree, backtracking to the next starting character in the cleaned creative event planning document.

[0009] In the aforementioned creative activity planning copy management method, based on the sensitive word context information in the potential sensitive word detection results, the potential sensitive words are subjected to contextual intelligent analysis to obtain the potential sensitive word semantic compliance analysis results. This includes: performing semantic embedding encoding on the potential sensitive words to obtain a potential sensitive word semantic embedding encoding vector; performing semantic encoding on the sensitive word context information based on a converter model to obtain a sensitive word context semantic information encoding vector; calculating the semantic alignment between the potential sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector as the potential sensitive word scene semantic fit; and generating the potential sensitive word semantic compliance analysis results based on the comparison between the potential sensitive word scene semantic fit and a preset threshold.

[0010] In the above-mentioned creative activity planning copy management method, calculating the semantic alignment between the potential sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector as the potential sensitive word scene semantic fit degree includes: concatenating the potential sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector to obtain the potential sensitive word-sensitive word context semantic joint encoding vector; and inputting the potential sensitive word-sensitive word context semantic joint encoding vector into the semantic alignment degree measurement network to obtain the semantic alignment degree.

[0011] In the above-mentioned creative activity planning copywriting management method, the potential sensitive word-sensitive word context semantic joint encoding vector is input into the semantic alignment measurement network to obtain the semantic alignment, including: ;in, This represents the joint encoding vector of the potential sensitive word and its context. Weight vector for semantic alignment measurement For the joint semantic linear connection weight matrix, For joint semantic bias vector, Indicates semantic alignment.

[0012] In the above-mentioned creative activity planning copy management method, calculating the semantic alignment degree between the potential sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector as the potential sensitive word scene semantic fit degree includes: performing semantic space probabilistic geometric pre-alignment on the potential sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector to obtain aligned potential sensitive word semantic embedding encoding vector and aligned sensitive word context semantic information encoding vector; concatenating the aligned potential sensitive word semantic embedding encoding vector and the aligned sensitive word context semantic information encoding vector to obtain a potential sensitive word-sensitive word context semantic joint encoding vector; and inputting the potential sensitive word-sensitive word context semantic joint encoding vector into a semantic alignment degree measurement network to obtain the semantic alignment degree.

[0013] In the aforementioned creative activity planning copy management method, semantic space probabilistic geometric pre-alignment is performed on the potential sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector to obtain aligned potential sensitive word semantic embedding encoding vector and aligned sensitive word context semantic information encoding vector. This includes: transforming the potential sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector, and then performing probabilistic transformation based on the sigmoid function to obtain potential sensitive word semantic embedding encoding probability vector and sensitive word context semantic information encoding probability vector respectively; in response to the absolute value of the difference between the feature values ​​corresponding to the potential sensitive word semantic embedding encoding probability vector and the sensitive word context semantic information encoding probability vector being greater than a distance threshold, obtaining alignment count and non-alignment count; dividing the alignment count and the non-alignment count by the length of the potential sensitive word semantic embedding encoding probability vector respectively to obtain alignment probability count and non-alignment probability count; and optimizing the potential sensitive word semantic embedding encoding probability vector and the sensitive word context semantic information encoding probability vector through cross-space alignment based on the alignment probability count and the non-alignment probability count to obtain aligned potential sensitive word semantic embedding encoding vector and aligned sensitive word context semantic information encoding vector.

[0014] According to another aspect of this application, a creative event planning document management system is provided, comprising: a planning document receiving module for receiving creative event planning documents uploaded by users for review; a planning document preprocessing module for preprocessing the creative event planning documents for review using a text cleaning tool to obtain cleaned creative event planning documents; a potential sensitive word detection module for detecting potential sensitive words in the cleaned creative event planning documents based on a sensitive word Trie tree to obtain a set of potential sensitive word detection results, wherein the potential sensitive word detection results include potential sensitive words, sensitive word positions, and sensitive word context information; a potential sensitive word semantic compliance analysis module for performing contextual intelligent analysis on the potential sensitive words based on the sensitive word context information in the potential sensitive word detection results to obtain potential sensitive word semantic compliance analysis results, wherein the potential sensitive word semantic compliance analysis results are used to indicate whether the potential sensitive words violate regulations; and a document review report generation module for generating a document review report based on the potential sensitive word semantic compliance analysis results.

[0015] Compared with existing technologies, the creative event planning document management system and method provided in this application aim to solve the problems of low efficiency, poor accuracy, and insufficient ability to handle complex formats in existing document review technologies. It first receives creative event planning documents uploaded by users for review and removes redundant spaces and special characters using text cleaning tools to ensure data purity. Next, it uses a sensitive word Trie tree to scan the cleaned document character by character to identify potential sensitive words and their contextual information. Based on this information, it further employs semantic embedding encoding and converter models to intelligently analyze potential sensitive words and assess whether they violate regulations. Finally, it automatically generates a detailed document review report to help users quickly understand the specific problems in the document and make corresponding adjustments. This improves review efficiency and accuracy, solving the problems of false positives, false negatives, and insufficient ability to handle complex formats in traditional methods. Attached Figure Description

[0016] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0017] Figure 1 The illustration shows a flowchart of a creative event planning document management method according to an embodiment of this application.

[0018] Figure 2 The illustration shows a flowchart of step S3 in the creative event planning document management method according to an embodiment of this application.

[0019] Figure 3 The illustration shows a flowchart of step S33 in the creative event planning document management method according to an embodiment of this application.

[0020] Figure 4 The illustration shows a flowchart of step S4 in the creative event planning document management method according to an embodiment of this application.

[0021] Figure 5 The illustration shows a flowchart of step S43 in the creative event planning document management method according to an embodiment of this application.

[0022] Figure 6 The illustration shows a structural diagram of a creative event planning and copywriting management system according to an embodiment of this application. Detailed Implementation

[0023] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0024] Based on this, this application provides a method for managing creative event planning documents. Figure 1 The illustration shows a flowchart of a creative event planning document management method according to an embodiment of this application. Figure 1 As shown, the creative event planning document management method includes: S1, receiving creative event planning documents uploaded by users for review; S2, preprocessing the creative event planning documents for review using a text cleaning tool to obtain cleaned creative event planning documents; S3, detecting potential sensitive words in the cleaned creative event planning documents based on a sensitive word Trie tree to obtain a set of potential sensitive word detection results, the potential sensitive word detection results including potential sensitive words, sensitive word positions, and sensitive word context information; S4, performing contextual intelligent analysis on the potential sensitive words based on the sensitive word context information in the potential sensitive word detection results to obtain potential sensitive word semantic compliance analysis results, the potential sensitive word semantic compliance analysis results being used to indicate whether the potential sensitive words violate regulations; S5, generating a document review report based on the potential sensitive word semantic compliance analysis results.

[0025] Specifically, in step S1, users upload their creative event planning documents through a specially designed web interface or mobile application. On this interface, users first need to log in to their account to ensure that only authorized users can access and submit content. After logging in, users will see a clean and intuitive upload interface that supports not only plain text files (such as Word documents, PDFs, etc.) but also multimedia files (such as images and videos). Users can choose to drag and drop files to a designated area or click the "Select File" button to select a file from their local computer. Once a file is selected and the upload is confirmed, a preliminary data verification process will automatically begin in the background, checking whether the file format meets the requirements and assessing whether the file size is within the allowed range. If everything is normal, a progress bar will be displayed to provide real-time feedback on the upload progress; if any problems arise, a prompt message will immediately pop up to inform the user and provide solutions or suggestions.

[0026] This design considers both user convenience and stability and security. First, by offering multiple file upload methods (drag and drop and file selection), it caters to different user habits and enhances the user experience. Second, requiring users to log in and verify permissions not only protects user privacy but also effectively prevents unauthorized access and potential security threats. Furthermore, real-time data verification during the upload process helps users promptly identify and correct errors, avoiding wasted time and resources. For example, if a user attempts to upload a file exceeding the maximum allowed size, they will be immediately prompted to compress the file or select another file, rather than waiting until the entire upload process is complete to receive an error message.

[0027] Furthermore, upon successful file upload, a unique identifier (ID) is generated for each file, and this ID is associated with the user's account for subsequent management and tracking. Metadata such as upload time and file type is also recorded, which is extremely useful for future auditing and analysis. Considering the specific needs of large enterprises, batch upload functionality is also supported, allowing users to upload multiple files at once. This significantly improves work efficiency, especially suitable for companies or organizations that frequently review large volumes of documents.

[0028] Specifically, in step S2, considering that user-uploaded text often contains various redundant information and formatting issues, such as extra spaces, unnecessary line breaks, and non-standard symbols, these elements not only affect the readability of the text but also interfere with subsequent data processing steps. Therefore, it is necessary to use a text cleaning tool to preprocess the creative event planning text to be reviewed to obtain a cleaned creative event planning text. Preprocessing with a text cleaning tool can remove this redundant information, ensuring the consistency and simplicity of the text structure. For example, multiple consecutive spaces are reduced to a single space, and non-standard punctuation marks are standardized, which can significantly improve the quality and parsability of the text. Sensitive word detection is a crucial step in text review, aiming to identify potential violations. However, if the text contains a large number of irrelevant characters or formatting issues, it may cause false positives or false negatives in the sensitive word matching algorithm. For example, certain special characters may be similar to words in the sensitive word library, leading to incorrect matching results. Removing these interfering factors through a text cleaning tool can make sensitive word detection more accurate, reduce the possibility of misjudgment, and ensure the reliability of the final review report. When reviewing copy, it's necessary to combine Natural Language Processing (NLP) technology for contextual intelligence analysis to assess whether potential sensitive words actually violate regulations. This analysis relies on high-quality input data; any poorly formatted copy or copy containing irrelevant characters will reduce the accuracy of semantic analysis. Text cleaning tools standardize the copy format, ensuring that each word can be understood and analyzed in the correct context. For example, removing redundant line breaks and spaces can help the converter model more accurately capture the logical relationships between sentences, improving the overall effectiveness of semantic analysis.

[0029] In one example, a text cleaning tool is used to preprocess the creative event planning document to be reviewed, resulting in a cleaned version. This includes removing redundant spaces and special characters from the document. It's understandable that a user-uploaded Word document may contain multiple consecutive spaces, unnecessary line breaks, or non-standard symbols (such as misuse of punctuation). While these elements don't directly affect the reading experience, they can lead to misjudgments or increased computational complexity during automated processing. To achieve this, text cleaning tools typically employ a series of predefined rules and algorithms. First, it scans the entire document line by line, identifying all redundant spaces and special characters. For example, multiple consecutive spaces are reduced to a single space; unnecessary line breaks are judged based on context to determine whether to retain or delete them. Furthermore, the tool filters out uncommon symbols or character sets, such as certain specially encoded punctuation marks or non-printable characters. These operations not only simplify the document's structure but also reduce interference in subsequent processing steps.

[0030] Specifically, in step S3, considering that sensitive word detection is a crucial step in copywriting review, it must be efficient and accurate. While traditional keyword matching methods are simple and direct, they often fall short when dealing with large-scale copywriting and complex contexts. Therefore, in this application, a sensitive word Trie tree is used to detect potential sensitive words in the cleaned creative event planning copywriting to obtain a set of potential sensitive word detection results. Compared to traditional keyword matching, the sensitive word Trie tree-based detection method has several significant advantages: First, the Trie tree structure significantly improves matching speed. Since each sensitive word is decomposed into multiple characters and stored in a tree structure, character-by-character matching only requires searching along the path, without needing to compare all words in the entire vocabulary. This linear search process greatly reduces computation, enabling sensitive word detection for a large number of copywriting documents to be completed in a short time. The advantages of the Trie tree are particularly evident when processing long copywriting documents or batch reviewing multiple documents. Second, the Trie tree supports multi-pattern matching. In practical applications, sensitive word libraries typically contain a large number of words that may cover different topics and fields. Traditional single-pattern matching methods struggle to handle such a large number of words simultaneously, while Trie trees can detect multiple sensitive words in a single scan. Furthermore, Trie trees can flexibly adapt to constantly updated sensitive word databases. As the social environment changes and laws and regulations adjust, sensitive word databases need to be updated regularly to reflect the latest censorship standards. By constructing a Trie tree structure, new sensitive words can be easily added or outdated words can be deleted without major modifications to the entire matching algorithm. This provides strong assurance for the system's scalability and adaptability.

[0031] Another crucial aspect is the extraction of contextual information. In sensitive word detection, simply identifying potential sensitive words is insufficient to determine whether they are truly in violation; a comprehensive analysis combining their contextual information is also necessary. For example, the word "horror" might be a legitimate description in some situations (such as the movie title *Triangle*), while in other contexts it might be considered sensitive content. By recording the specific location of each potential sensitive word and the fragments consisting of the characters preceding and following it, rich background information can be provided for subsequent intelligent contextual analysis, helping to more accurately assess the actual meaning and impact of sensitive words.

[0032] In one example, such as Figure 2As shown, step S3, based on the sensitive word Trie tree, detects potential sensitive words in the cleaned creative event planning document to obtain a set of potential sensitive word detection results, including: S31, obtaining a sensitive word library; S32, constructing the sensitive word library into the sensitive word Trie tree; S33, based on the sensitive word Trie tree, scanning each character in the cleaned creative event planning document character by character to obtain the set of potential sensitive word detection results.

[0033] Understandably, the first step is to obtain a pre-built sensitive word database. This database typically contains a large list of words that may cover multiple areas such as politics, religion, pornography, and violence—all content that requires strict review in different application scenarios. For example, certain political terms or inappropriate remarks may be prohibited in an internal corporate event; while promotional materials for educational institutions may focus on the presence of inappropriate language or content.

[0034] Once the sensitive word database is ready, it is converted into a Trie tree structure. A Trie tree is an efficient string matching data structure, particularly suitable for prefix search and multi-pattern matching tasks. In this process, each sensitive word is broken down into its constituent characters and inserted into the Trie tree according to certain rules. For example, if the sensitive word database contains words like "horror" or "violence," these words will be progressively inserted into different branches of the Trie tree. The resulting Trie tree structure supports fast character-by-character matching operations, greatly improving the speed and efficiency of sensitive word detection.

[0035] Next, the cleaned text will be scanned character by character. For each character, starting from the root node of the Trie tree, the search will proceed along the path to find a corresponding child node. If a matching child node is found for the current character in the Trie tree, the search will continue to the next character. This process is similar to looking up a word in a dictionary: starting from the first letter, the search proceeds step by step until a complete word is found or an unmatched character is encountered.

[0036] That is, in one example, such as Figure 3As shown in the figure, in step S33, based on the sensitive word Trie tree, each character in the cleaned creative activity planning copywriting is scanned character by character to obtain a set of potential sensitive word detection results, including: S331, for each character in the cleaned creative activity planning copywriting, start searching along the path from the root node of the sensitive word Trie tree; S332, if there is a corresponding child node for the character in the sensitive word Trie tree, continue to search for the next character and if an end node is reached, a potential sensitive word is matched; S333, if there is no corresponding child node for the character in the sensitive word Trie tree, backtrack to the next starting character in the cleaned creative activity planning copywriting.

[0037] In a specific example, there is a passage in the cleaned copywriting: "This event aims to promote positive energy and oppose terrorism." It will start from the first character "本" and search for the corresponding node of this character in the Trie tree. If no matching node is found, the character is directly skipped and the next character is continued to be searched. When the character "恐" is found, it is found that there is a corresponding node for this character in the Trie tree, so the next character "怖" is continued to be searched. At this time, a complete sensitive word "恐怖" has been found, and the position of this sensitive word (such as the 10th character) and its context information (such as the fragment composed of several characters before and after) are recorded. This character-by-character scanning method can not only efficiently identify potential sensitive words, but also accurately locate their specific positions in the copywriting, facilitating subsequent further analysis and processing. If no matching child node can be found at a certain character, it will backtrack to the next starting character in the cleaned copywriting and start the matching process again. For example, in the above example, if no matching node is found for the characters after "本", the character "本" will be skipped and a new round of matching will start from the character "次". Through this method, the entire copywriting can be traversed to ensure that no potential sensitive words are missed.

[0038] Specifically, in step S4, while simple sensitive word matching can quickly identify potential problems, it cannot accurately determine the actual meaning and impact of words in a specific context. Modern copywriting review requires a higher level of intelligence, capable of making more accurate judgments based on understanding and parsing the context. Therefore, in this application, based on the sensitive word context information in the potential sensitive word detection results, the potential sensitive words are subjected to contextual intelligent analysis to obtain a potential sensitive word semantic compliance analysis result, which is used to indicate whether the potential sensitive word violates regulations. Specifically, traditional keyword matching methods rely solely on the words themselves, which is prone to false positives or false negatives. For example, the word "horror" may be a legitimate description in some situations (such as the movie title "Triangle"), while it may be considered sensitive content in other scenarios. By converting potential sensitive words into semantic embedding encoding vectors and combining them with semantic encoding vectors of their context information, the meaning of words in specific contexts can be better understood, reducing the possibility of false positives.

[0039] In one example, such as Figure 4 As shown, step S4, based on the sensitive word context information in the potential sensitive word detection result, performs contextual intelligent analysis on the potential sensitive words to obtain the potential sensitive word semantic compliance analysis result, including: S41, performing semantic embedding encoding on the potential sensitive words to obtain the potential sensitive word semantic embedding encoding vector; S42, performing semantic encoding on the sensitive word context information based on a converter model to obtain the sensitive word context semantic information encoding vector; S43, calculating the semantic alignment between the potential sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector as the potential sensitive word scene semantic fit; S44, generating the potential sensitive word semantic compliance analysis result based on the comparison between the potential sensitive word scene semantic fit and a preset threshold.

[0040] As you can understand, the first step is to perform semantic embedding encoding on potential sensitive words. Semantic embedding is a technique that maps words to a continuous vector space, capturing the semantic relationships between words. For example, the detected potential sensitive word "terror" is transformed into a vector of a specific dimension; this vector is called the potential sensitive word semantic embedding encoding vector. Each dimension represents the features of the word in different semantic dimensions, ensuring that words with similar meanings have a high degree of similarity in the vector space.

[0041] Simultaneously, it is necessary to semantically encode the contextual information of potential sensitive words based on a converter model. Converter models (such as BERT) are advanced natural language processing techniques capable of generating high-quality semantic representations based on an understanding of sentence structure and semantics. They extract the contextual information surrounding potential sensitive words (such as segments composed of preceding and following characters) and transform it into a semantic encoding vector—the sensitive word contextual semantic information encoding vector. This vector contains specific semantic information from the context, helping to more accurately understand the actual meaning of potential sensitive words.

[0042] Next, we will calculate the semantic alignment between the latent sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector. In one example, such as Figure 5 As shown, step S43, calculating the semantic alignment between the potential sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector as the potential sensitive word scene semantic fit, includes: S431, concatenating the potential sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector to obtain the potential sensitive word-sensitive word context semantic joint encoding vector; S432, inputting the potential sensitive word-sensitive word context semantic joint encoding vector into the semantic alignment measurement network to obtain the semantic alignment.

[0043] In one example, the latent sensitive word-sensitive word context semantic joint encoding vector is input into a semantic alignment metric network to obtain the semantic alignment, including: ;in, This represents the joint encoding vector of the potential sensitive word and its context. Weight vector for semantic alignment measurement For the joint semantic linear connection weight matrix, For joint semantic bias vector, Indicates semantic alignment.

[0044] In other words, the latent sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector are first concatenated to generate a new joint encoding vector, called the latent sensitive word-sensitive word context semantic joint encoding vector. This latent sensitive word-sensitive word context semantic joint encoding vector not only contains the semantic information of the latent sensitive word itself, but also integrates the specific semantic background of its context, providing rich input data for subsequent semantic alignment calculation.

[0045] Then, the joint semantic encoding vector of the potential sensitive word and its context is input into the semantic alignment measurement network. The semantic alignment measurement network is a pre-trained neural network model specifically designed to evaluate the similarity or consistency between two semantic vectors. This network can calculate the semantic alignment between a potential sensitive word and its context. Specifically, the network contains a semantic alignment measurement weight vector, a joint semantic linear connection weight matrix, and a joint semantic bias vector; these parameters collectively determine the final output semantic alignment value. As those skilled in the art will understand, the specific parameters of the semantic alignment measurement weight vector, the joint semantic linear connection weight matrix, and the joint semantic bias vector are obtained through training; that is, through dataset preparation, parameter initialization, forward propagation, loss function, and backpropagation, the semantic alignment measurement weight vector, the joint semantic linear connection weight matrix, and the joint semantic bias vector are progressively optimized and adjusted. This method not only improves semantic understanding and analysis capabilities but also ensures the accuracy and reliability of document review, providing strong support for modern enterprise document management and review work.

[0046] Based on the above calculations, the semantic alignment value is further compared with a preset threshold to generate semantic compliance analysis results for potential sensitive words. Here, the setting of the preset threshold needs to consider business needs and risk preferences. Different application scenarios may have different tolerances for false positives (identifying compliant content as non-compliant) and false negatives (identifying non-compliant content as compliant). Combined with dynamic adjustment and feedback mechanisms for continuous optimization, this method not only improves accuracy and reliability but also ensures its flexibility in responding to different application scenarios and needs.

[0047] In a specific example, the text is detected to contain the potentially sensitive word "terrorism," with the context information "This activity aims to promote positive energy and oppose terrorism." First, "terrorism" is converted into a semantic embedding encoding vector for the potentially sensitive word, and the context information "This activity aims to promote positive energy and oppose terrorism" is converted into another semantic encoding vector for the sensitive word's context. Next, these two vectors are concatenated to generate a joint encoding vector for the potentially sensitive word and its context. Finally, this joint encoding vector is input into a semantic alignment measurement network to obtain a specific semantic alignment value, such as 0.85. Based on the above calculation result, the semantic alignment value is further compared with a preset threshold to generate a semantic compliance analysis result for the potentially sensitive word. If the semantic alignment value is higher than the preset threshold (in this specific example, the preset threshold is set to 0.8), the potentially sensitive word is considered compliant with regulations in the current context and is not considered non-compliant content; conversely, if the semantic alignment value is lower than the preset threshold, the potentially sensitive word may pose a violation risk in the current context and requires further manual review or corresponding rectification measures.

[0048] In a preferred example, calculating the semantic alignment between the latent sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector as the latent sensitive word scene semantic fit includes: performing semantic space probabilistic geometric pre-alignment on the latent sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector to obtain aligned latent sensitive word semantic embedding encoding vector and aligned sensitive word context semantic information encoding vector; concatenating the aligned latent sensitive word semantic embedding encoding vector and the aligned sensitive word context semantic information encoding vector to obtain a latent sensitive word-sensitive word context semantic joint encoding vector; and inputting the latent sensitive word-sensitive word context semantic joint encoding vector into a semantic alignment metric network to obtain the semantic alignment.

[0049] Specifically, semantic space probabilistic geometric pre-alignment is performed on the latent sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector to obtain aligned latent sensitive word semantic embedding encoding vector and aligned sensitive word context semantic information encoding vector. This includes: transforming the latent sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector, and then performing probabilistic transformation based on the sigmoid function to obtain latent sensitive word semantic embedding encoding probability vector and sensitive word context semantic information encoding probability vector respectively; in response to the absolute value of the difference between the feature values ​​corresponding to the latent sensitive word semantic embedding encoding probability vector and the sensitive word context semantic information encoding probability vector being greater than a distance threshold, obtaining alignment count and non-alignment count; dividing the alignment count and the non-alignment count by the length of the latent sensitive word semantic embedding encoding probability vector respectively to obtain alignment probability count and non-alignment probability count; and optimizing the latent sensitive word semantic embedding encoding probability vector and the sensitive word context semantic information encoding probability vector through cross-space alignment based on the alignment probability count and the non-alignment probability count to obtain aligned latent sensitive word semantic embedding encoding vector and aligned sensitive word context semantic information encoding vector.

[0050] Specifically, before directly concatenating the latent sensitive word semantic embedding encoding vector with the sensitive word context semantic information encoding vector to obtain the latent sensitive word-sensitive word context semantic joint encoding vector, in order to improve the intuitive encoding consistency between semantic embedding encoding and semantic encoding based on the converter model, and to improve the accuracy of semantic alignment calculation, a semantic vector intuitive alignment mechanism is introduced to improve the encoding interpretability consistency of the alignment result through the semantic space probabilistic geometric intuition.

[0051] Specifically, firstly, the latent sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector are converted to the same length, for example, by interpolation of the short vectors. Then, probabilistic transformation based on the sigmoid function is performed to obtain the latent sensitive word semantic embedding encoding probability vector and the sensitive word context semantic information encoding probability vector, for example, represented as... and .

[0052] Then, in response to the absolute value of the difference between the feature values ​​corresponding to the latent sensitive word semantic embedding encoding probability vector and the sensitive word context semantic information encoding probability vector being greater than the distance threshold, ... Obtain alignment count Unaligned count ,Right now: ;in, Represents a counting function. and These represent the first and second edges of the semantic embedding encoding probability vector of the potential sensitive words, respectively. The and the first Feature values ​​at each position, and The first and second lines respectively represent the first and third lines of the encoding probability vector of the context semantic information of the sensitive words. The and the first The feature values ​​at each position.

[0053] Then count according to the alignment. and the unaligned count Divide each probability by the length of the latent sensitive word semantic embedding encoding probability vector to obtain the alignment probability count. Unaligned probability count .

[0054] Finally, the semantic embedding and encoding probability vector of the potential sensitive words is performed using a cross-space alignment method. and the probability vector encoded from the contextual semantic information of the sensitive words Optimization: ;in, This indicates subtraction by position. This indicates addition by position. This indicates dot product by position. Representing vectors sum vector The inner product, and For scaling hyperparameters, This represents the aligned semantic embedding encoding vector of potential sensitive words. This represents the encoded vector of the contextual semantic information of sensitive words after alignment.

[0055] In this way, the semantic space probabilistic geometric pre-alignment of the potential sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector based on intuitive semantic similarity can be performed without using a spatial mapping weight matrix that requires determining a large number of parameters. This improves the weight convergence of the joint semantic linear connection weight matrix of the subsequent semantic alignment measurement network and improves the accuracy of semantic alignment calculation.

[0056] Specifically, in step S5, a text review report is generated based on the semantic compliance analysis results of potential sensitive words. First, the potential sensitive words and their contextual fragments are marked in the original text according to their location. For example, if the text contains the potential sensitive word "terrorism," it will be highlighted before and after the word, allowing users to quickly locate the problem. Simultaneously, the report will list the specific location of each potential sensitive word (e.g., paragraph number, line number) and its contextual information to help users better understand its actual meaning and impact. Next, each potential sensitive word is categorized and labeled based on the semantic compliance analysis results. If a potential sensitive word is deemed compliant, the report will specify its explanation and the reasons for compliance; conversely, if a potential sensitive word is deemed non-compliant, the report will provide specific reasons for the violation and corresponding improvement suggestions. For example, the word "terrorism" in "This activity aims to promote positive energy and oppose terrorism" may be marked as compliant content, with its context explaining that it refers to "terrorism" rather than other sensitive meanings.

[0057] This approach not only significantly improves the transparency and credibility of document review but also provides users with practical improvement suggestions. First, by marking potential sensitive words and their contextual information in the original document, users can quickly pinpoint the problem, saving considerable time. Second, by categorizing and labeling each potential sensitive word, users can clearly understand the actual meaning and impact of each term, avoiding unnecessary misunderstandings and misjudgments. Another important aspect is flexibility and scalability. Different application scenarios and business needs may have different requirements for the content and format of the review report. For example, in some scenarios, users may be more concerned with the specific location and contextual information of potential sensitive words; while in other scenarios, users may be more concerned with the overall compliance assessment results. By flexibly configuring report templates and content, diverse business needs can be met, providing more personalized services.

[0058] This application also provides a creative event planning document management system for implementing the above-described creative event planning document management method. Figure 6 The illustration shows a structural diagram of a creative event planning and copywriting management system according to an embodiment of this application. Figure 6As shown, the creative event planning document management system 600 includes: a planning document receiving module 610, used to receive creative event planning documents uploaded by users for review; a planning document preprocessing module 620, used to preprocess the creative event planning documents for review using a text cleaning tool to obtain cleaned creative event planning documents; a potential sensitive word detection module 630, used to detect potential sensitive words in the cleaned creative event planning documents based on a sensitive word Trie tree to obtain a set of potential sensitive word detection results, the potential sensitive word detection results including potential sensitive words, sensitive word positions, and sensitive word context information; a potential sensitive word semantic compliance analysis module 640, used to perform contextual intelligent analysis on the potential sensitive words based on the sensitive word context information in the potential sensitive word detection results to obtain potential sensitive word semantic compliance analysis results, the potential sensitive word semantic compliance analysis results used to indicate whether the potential sensitive words are in violation; and a document review report generation module 650, used to generate a document review report based on the potential sensitive word semantic compliance analysis results.

[0059] Those skilled in the art will understand that the specific operations of each module in the aforementioned creative event planning and copywriting management system have been referenced above. Figures 1 to 5 The description of creative event planning copywriting management methods is detailed here, and therefore, its repeated description will be omitted.

[0060] This application also provides a computer-readable storage medium storing computer program code. When the computer program code is run on a computer, the computer executes the above-described related method steps to implement the creative activity planning document management method provided in the above embodiments.

[0061] This application also provides a computer program product that, when run on a computer, causes the computer to perform the aforementioned steps to achieve the creative event planning document management method provided in the above embodiments.

[0062] In this application, the system, computer-readable storage medium, or computer program product provided in the embodiments are all used to execute the corresponding methods provided above. Therefore, the beneficial effects that can be achieved can be referred to the beneficial effects in the corresponding methods provided above, and will not be repeated here.

[0063] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments.

[0064] The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous. The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A creative event planning script management method, characterized by, The method comprises the following steps: receiving a creative activity plan script uploaded by a user for review; preprocessing the creative activity plan script to be reviewed by using a text cleaning tool to obtain a cleaned creative activity plan script; detecting potential sensitive words in the cleaned creative activity plan script based on a sensitive word Trie tree to obtain a set of potential sensitive word detection results, wherein the potential sensitive word detection results include a potential sensitive word, a sensitive word position, and sensitive word context information; performing context intelligent analysis on the potential sensitive word based on the sensitive word context information in the potential sensitive word detection results to obtain a potential sensitive word semantic compliance analysis result, wherein the potential sensitive word semantic compliance analysis result is used to indicate whether the potential sensitive word is in violation; generating a script review report based on the potential sensitive word semantic compliance analysis result; wherein, based on the sensitive word Trie tree, detecting potential sensitive words in the cleaned creative activity plan script to obtain a set of potential sensitive word detection results comprises: obtaining a sensitive word library; constructing the sensitive word library into the sensitive word Trie tree; based on the sensitive word Trie tree, performing character-by-character scanning on each character in the cleaned creative activity plan script to obtain the set of potential sensitive word detection results; wherein, based on the sensitive word context information in the potential sensitive word detection results, performing context intelligent analysis on the potential sensitive word to obtain a potential sensitive word semantic compliance analysis result comprises: performing semantic embedding coding on the potential sensitive word to obtain a potential sensitive word semantic embedding coding vector; performing semantic coding based on a converter model on the sensitive word context information to obtain a sensitive word context semantic information coding vector; calculating a semantic alignment degree between the potential sensitive word semantic embedding coding vector and the sensitive word context semantic information coding vector as a potential sensitive word scene semantic adaptation degree; based on a comparison between the potential sensitive word scene semantic adaptation degree and a preset threshold, generating the potential sensitive word semantic compliance analysis result; wherein, in the process of calculating the semantic alignment degree between the potential sensitive word semantic embedding coding vector and the sensitive word context semantic information coding vector as the potential sensitive word scene semantic adaptation degree, performing semantic space probability geometric pre-alignment on the potential sensitive word semantic embedding coding vector and the sensitive word context semantic information coding vector to obtain an aligned potential sensitive word semantic embedding coding vector and an aligned sensitive word context semantic information coding vector, comprising: after converting the potential sensitive word semantic embedding coding vector and the sensitive word context semantic information coding vector, performing probability-based processing based on a sigmoid function respectively to obtain a potential sensitive word semantic embedding coding probability vector and a sensitive word context semantic information coding probability vector; in response to the absolute values of the differences between the feature values corresponding to the potential sensitive word semantic embedding coding probability vector and the sensitive word context semantic information coding probability vector being greater than a distance threshold respectively, obtaining an alignment count and a non-alignment count; divide the aligned count and the unaligned count by a length of the potential sensitive word semantic embedding encoding probability vector to obtain an aligned probability count and an unaligned probability count, respectively; based on the aligned probability count and the unaligned probability count, performing optimization of the potential sensitive word semantic embedding encoding probability vector and the sensitive word context semantic information encoding probability vector by a cross-space alignment manner to obtain an aligned potential sensitive word semantic embedding encoding vector and an aligned sensitive word context semantic information encoding vector.

2. The creative event planning script management method of claim 1, wherein, performing preprocessing on the to-be-audited creative activity planning script by using a text cleaning tool to obtain a cleaned creative activity planning script, including: removing redundant spaces and special characters in the to-be-audited creative activity planning script.

3. The creative event planning script management method of claim 2, wherein, based on the sensitive word Trie tree, performing character-by-character scanning on each character in the cleaned creative activity planning script to obtain a set of potential sensitive word detection results, including: starting from a root node of the sensitive word Trie tree, performing path searching for each character in the cleaned creative activity planning script; if the character has a corresponding child node in the sensitive word Trie tree, continuing to search for a next character, and if an end node is reached, a potential sensitive word is matched; if the character does not have a corresponding child node in the sensitive word Trie tree, backtracking to a next starting character in the cleaned creative activity planning script.

4. The creative event planning script management method of claim 2, wherein, calculating a semantic alignment degree between the potential sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector as a potential sensitive word scene semantic adaptation degree, including: concatenating the potential sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector to obtain a potential sensitive word-sensitiveword context semantic joint encoding vector; inputting the potential sensitive word-sensitiveword context semantic joint encoding vector into a semantic alignment degree measurement network to obtain the semantic alignment degree.

5. The creative event planning script management method of claim 4, wherein, inputting the potential sensitive word-sensitive word context semantic joint encoding vector into a semantic alignment degree measurement network to obtain the semantic alignment degree, comprising: ; wherein, represents the potential sensitive word-sensitive word context semantic joint encoding vector, is a semantic alignment degree measurement weight vector, is a joint semantic linear connection weight matrix, is a joint semantic bias vector, represents the semantic alignment degree.

6. The creative event planning script management method of claim 2, wherein, calculating a semantic alignment degree between the potential sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector as a potential sensitive word scene semantic adaptation degree, including: performing semantic space probability geometric pre-alignment on the potential sensitive word semantic embedding encoding vector and the sensitive word context semantic information encoding vector to obtain an aligned potential sensitive word semantic embedding encoding vector and an aligned sensitive word context semantic information encoding vector; concatenating the aligned potential sensitive word semantic embedding encoding vector and the aligned sensitive word context semantic information encoding vector to obtain a potential sensitive word-sensitiveword context semantic joint encoding vector; inputting the potential sensitive word-sensitiveword context semantic joint encoding vector into a semantic alignment degree measurement network to obtain the semantic alignment degree.

7. A creative event planning script management system for carrying out the creative event planning script management method according to any one of claims 1 to 6, characterized by, including: a planning script receiving module, configured to receive a to-be-audited creative activity planning script uploaded by a user; a planning script preprocessing module, configured to perform preprocessing on the to-be-audited creative activity planning script by using a text cleaning tool to obtain a cleaned creative activity planning script; The potential sensitive word detection module is configured to detect potential sensitive words in the cleaned creative activity planning script based on a sensitive word Trie tree to obtain a set of potential sensitive word detection results, the potential sensitive word detection results including a potential sensitive word, a sensitive word position, and sensitive word context information; The potential sensitive word semantic compliance analysis module is configured to perform context intelligent analysis on the potential sensitive word based on the sensitive word context information in the potential sensitive word detection result to obtain a potential sensitive word semantic compliance analysis result, the potential sensitive word semantic compliance analysis result being used to indicate whether the potential sensitive word violates a rule. The script review report generation module is configured to generate a script review report based on the potential sensitive word semantic compliance analysis result.

Images