Intelligent sensitive word detection and word recommendation method and system, medium and equipment

Abstract

The invention discloses an intelligent sensitive word detection and word recommendation method and system, a medium and equipment, and relates to the technical field of word detection and recommendation. The method comprises the following steps: obtaining product information of a target product, and determining a propaganda material based on the product information; formula information of the target product is extracted from the product information, and a first recommendation word set is determined based on the formula information and the propaganda materials; detecting and processing sensitive words in the first recommended word set to obtain a second recommended word set; and performing association expansion on the second recommendation word set, and performing sensitive word verification on the expanded recommendation word set to obtain a target recommendation word set. By implementing the technical scheme provided by the invention, the accuracy and correlation of word recommendation of the product can be improved, so that the advertising effect and the market performance of the product are enhanced.

Description

Technical Field

[0001] The present application relates to the technical field of word detection and recommendation, and specifically to an intelligent sensitive word detection and word recommendation method, system, medium and device. Background Art

[0002] In today's era of rapid information development, marketing strategies and methods are becoming increasingly diverse. Especially in product promotion and advertising, the choice of language plays a decisive role in attracting consumers and conveying product information. In this process, choosing the right words can not only accurately convey the characteristics and advantages of the product, but also avoid potential sensitivity issues, thereby protecting the corporate image and avoiding legal risks.

[0003] In the prior art, word recommendation systems are often used to assist in text generation, such as advertising, product descriptions, etc. These systems generally generate recommended words based on basic product information and a preset keyword library, and also include a sensitive word detection function to screen out words that may cause legal or moral risks. However, the prior art usually uses a simple deletion method when processing sensitive words, which may limit the accuracy and relevance of word recommendations, resulting in the final recommended word set failing to meet marketing needs, thereby affecting the effectiveness of advertising and the market performance of products. Summary of the invention

[0004] The present application provides an intelligent sensitive word detection and word recommendation method, system, medium and device, which can improve the accuracy and relevance of product word recommendations, thereby enhancing the effectiveness of advertising and the market performance of products.

[0005] In a first aspect, the present application provides an intelligent sensitive word detection and word recommendation method, the method comprising: Acquire product information of a target product, and determine promotional materials based on the product information; extracting the recipe information of the target product from the product information, and determining a first set of recommended words based on the recipe information and the promotional material; Detecting and processing sensitive words in the first recommended word set to obtain a second recommended word set; The second recommended word set is associated and expanded, and the expanded recommended word set is checked for sensitive words to obtain a target recommended word set.

[0006] By adopting the above technical scheme, appropriate promotional materials related to the product can be obtained through accurate matching of target product information, and formula information can be extracted from product information to determine the composition, nutritional ingredients, efficacy and other characteristics of the product. The first batch of recommended word sets are determined based on the formula information and promotional materials, and preliminary personalized customization is performed. By detecting and effectively removing risky words in the first recommended word set, a safe and compliant second recommended word set is obtained, and finally expanded by mining the word association network, the recommended word set can have a wider coverage and richer information, and risk verification is performed on the expanded words, so that the final output target word set not only ensures safety, but also has good relevance and accuracy, and can accurately convey product information, thereby enhancing the effect of advertising and the market performance of the product.

[0007] Optionally, determining the promotional material based on the product information includes: Matching the product category of the target product from a preset product category library; Based on the product category, extracting a first promotional material associated with the product category from a preset promotional material library; Performing semantic analysis on the first promotional material to extract keywords and themes; Generate second promotional materials based on the keywords and themes combined with the feature description in the product information; The first promotional material and the second promotional material are merged to obtain the promotional material of the target product.

[0008] By adopting the above technical solution, by matching product categories, extracting related first promotional materials, performing semantic analysis to extract keywords and themes, generating second promotional materials, and integrating the first and second promotional materials, customized promotional materials for target products are finally obtained. This technical means can use product categories for targeted extraction, perform semantic analysis to enhance the first material, and then autonomously generate the second material for supplementation, so that the final promotional materials can fully combine the category characteristics and key information of the product, have strong relevance and pertinence, and provide high-quality input materials for subsequent word recommendations.

[0009] Optionally, determining the first recommended word set based on the recipe information and the promotional material includes: Establishing multi-dimensional formula evaluation indicators, including nutritional indicators, functional indicators and market indicators; Based on the parameter values ​​in the formula information, calculating various indexes corresponding to the formula evaluation indicators of the target product; According to the various indices, corresponding descriptive words are extracted from a preset index mapping vocabulary; Performing sentiment analysis and key information extraction on the promotional material, and calculating semantic relevance of the descriptive words based on the sentiment analysis results and key information; A number of target description words with semantic relevance higher than a relevance threshold are selected, and the target description words are combined and transformed according to product formula characteristics, and the combined and transformed word set is used as the first recommended word set.

[0010] By adopting the above technical solution, we first establish a formula evaluation index that considers nutrition, function and market multi-dimensional angles, then calculate the index value of each evaluation index based on the formula information, and then extract the corresponding descriptive words from the preset mapping vocabulary according to the index value. At the same time, we analyze the promotional materials to extract key information, calculate the semantic relevance between the descriptive words and the key information, and finally select target descriptive words with high relevance for combination and deformation to generate a first recommended word set that is highly correlated and meets the characteristics of the product formula. Comprehensive formula analysis and evaluation can be carried out, combined with customized extraction and processing of product key information, so that the first recommended word set can accurately reflect the formula characteristics of the product, laying the foundation for subsequent precise word recommendations.

[0011] Optionally, detecting and processing the sensitive words in the first recommended word set to obtain a second recommended word set includes: Performing word segmentation processing on each word in the first recommended word set to obtain a word segmentation result; Matching the word segmentation result with a preset multi-level sensitive word library to obtain a first sensitive word list; Performing context semantic analysis on each word in the first sensitive word list, and filtering the first sensitive word list based on the context semantic analysis result to obtain a second sensitive word list; The second sensitive word list is graded, and corresponding sensitive word processing is performed according to a preset processing strategy to obtain a second recommended word set.

[0012] By adopting the above technical solution, through a series of technical means such as word segmentation, matching with multi-level sensitive word libraries, contextual semantic analysis and filtering, classification by sensitivity level and corresponding processing, various sensitive words in the first recommended word set can be effectively identified, and different processing strategies can be adopted according to the specific semantic situation and sensitivity of the words to accurately remove risky words and obtain the second recommended word set that meets security and compliance requirements. The words can be processed more finely to avoid the loss of effective information caused by simple direct deletion, thereby improving the quality of the recommended word set.

[0013] Optionally, the step of classifying the second sensitive word list and performing corresponding sensitive word processing according to a preset processing strategy to obtain a second recommended word set includes: Classify each sensitive word in the second sensitive word list into high-level sensitive words, medium-level sensitive words and low-level sensitive words; For the highly sensitive words, a deletion strategy is adopted; For the intermediate sensitive words, a synonym replacement processing strategy is adopted; For the low-level sensitive words, a processing strategy of reconstructing the context to generate new words is adopted; The word set obtained by processing the sensitive words of each level with the corresponding processing strategy is used as the second recommended word set.

[0014] By adopting the above technical scheme, firstly, words are classified into three levels: high, medium and low according to their sensitivity, and then different strategies are adopted for processing each level, wherein high-level sensitive words are deleted, medium-level sensitive words are replaced with synonyms, and low-level sensitive words are reconstructed by using context to generate new words. Finally, the processed words of each level are combined as the second recommended word set. This hierarchical processing can adopt different strategies according to the actual sensitivity of the words. Deleting high-level words can effectively avoid legal risks, and replacing and reconstructing medium and low-level words can achieve information preservation, avoiding data loss caused by excessive deletion, thereby improving the quality and richness of the second recommended word set.

[0015] Optionally, the step of performing association expansion on the second recommended word set and performing sensitive word verification on the expanded recommended word set to obtain a target recommended word set includes: Constructing a word association network, wherein the word association network includes a synonym layer, a near-sense layer, a hypernym layer, and a hyponym layer; For each word in the second recommended word set, mining is performed in the word association network to obtain an associated word list, wherein the associated word list includes an association degree of each word; Filtering out words with a correlation degree higher than a correlation degree threshold from the associated word list as candidate expansion words; Performing semantic adaptation scoring on each of the candidate expansion words, filtering out words with scores below a scoring threshold, and obtaining an initial expansion word set; A sensitive word check is performed on each word in the initial expanded word set to obtain a target recommended word set.

[0016] By adopting the above technical scheme, by constructing a word network with multi-level associations such as synonyms and near-synonyms, word mining is performed and the correlation is calculated based on the second recommended word set, and words with high correlation are selected for semantic matching scoring and verification, and finally an expanded target recommended word set is obtained, thereby realizing intelligent expansion of the second recommended word set, and making full use of the association relationship between words for expansion, and combining semantic matching with risk verification mechanism, while expanding the word set, the security of the expanded word set and its correlation with the original word set are guaranteed, thereby improving the richness and coverage of the recommended word set.

[0017] Optionally, the method further includes: Establish a multimodal recommendation system that integrates text analysis module, image analysis module and audio analysis module; Inputting the target recommendation word set into the text analysis module to generate corresponding visual element suggestions and audio element suggestions; Based on the visual element suggestions, filter product display images from the image library; Based on the audio element suggestion, select audio material from an audio material library; The target recommendation word set, the product display pictures and the audio material are multimodally fused to generate a comprehensive promotional plan including text, images and audio.

[0018] By adopting the above technical solution and establishing a text, image, and audio integrated analysis system, the recommended word set is input into the text module to generate visual and audio element suggestions, product images are screened from the gallery based on visual suggestions, and materials are selected from the audio library based on audio suggestions. Finally, multimodal fusion is performed to generate a comprehensive promotional plan of text, images, and audio, thereby realizing the cross-modal application and presentation of the recommended word set. The communication advantages of different modes can be fully utilized for combination, making product promotion more three-dimensional, rich, and vivid, and improving the overall communication effect of the promotional plan.

[0019] In a second aspect of the present application, an intelligent sensitive word detection and word recommendation system is provided, the system comprising: A material determination module, used to obtain product information of a target product and determine promotional materials based on the product information; An initial word set recommendation module, configured to extract the recipe information of the target product from the product information, and determine a first recommended word set based on the recipe information and the promotional material; A sensitive word detection module, used to detect and process the sensitive words in the first recommended word set to obtain a second recommended word set; The target word set recommendation module is used to perform association expansion on the second recommended word set, and perform sensitive word verification on the expanded recommended word set to obtain a target recommended word set.

[0020] In a third aspect of the present application, a computer storage medium is provided, wherein the computer storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded by a processor and executing the above method steps.

[0021] In a fourth aspect of the present application, an electronic device is provided, comprising: a processor and a memory; wherein the memory stores a computer program, and the computer program is suitable for being loaded by the processor and executing the above-mentioned method steps.

[0022] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: This application obtains appropriate promotional materials related to the product through precise matching of target product information. Formula information extracted from product information can determine the composition, nutritional ingredients, efficacy and other characteristics of the product. The first batch of recommended word sets is then determined based on the formula information and promotional materials, and preliminary personalized customization is performed. By detecting and effectively removing risky words in the first recommended word set, a safe and compliant second recommended word set is obtained. Finally, the recommended word set is expanded by mining the word association network, which can make the coverage of the recommended word set wider and the amount of information richer. Risk verification is then performed on the expanded words, so that the final output target word set not only ensures safety, but also has good relevance and accuracy, and can accurately convey product information, thereby enhancing the effectiveness of advertising and the market performance of the product. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flow chart of an intelligent sensitive word detection and word recommendation method provided in an embodiment of the present application; Figure 2 This is a module diagram of an intelligent sensitive word detection and word recommendation system provided in an embodiment of the present application; Figure 3 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application.

[0024] Description of reference numerals: 300, electronic device; 301, processor; 302, communication bus; 303, user interface; 304, network interface; 305, memory. DETAILED DESCRIPTION

[0025] In order to enable technicians in this field to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments.

[0026] In the description of the embodiments of the present application, words such as "for example" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "for example" or "for example" is intended to present related concepts in a specific way.

[0027] In the description of the embodiments of the present application, the meaning of the term "multiple" refers to two or more. For example, multiple systems refer to two or more systems, and multiple screen terminals refer to two or more screen terminals. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. The terms "include", "comprise", "have" and their variations all mean "including but not limited to", unless otherwise specifically emphasized.

[0028] The following will provide a clear and complete description of the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments.

[0029] Please refer to Figure 1 , a flowchart of an intelligent sensitive word detection and word recommendation method is proposed. The method can be implemented by a computer program, a single-chip microcomputer, or run on an intelligent sensitive word detection and word recommendation system. The computer program can be integrated into a computer device or run as an independent tool application. Specifically, the method includes steps 10 to 40, and the steps are as follows: Step 10: Obtain product information of the target product and determine promotional materials based on the product information.

[0030] In the embodiment of the present application, the target product refers to a specific product that needs to be intelligently detected and recommended, and may be a commodity or service that is about to be launched or is currently being promoted.

[0031] Product information in the embodiment of the present application refers to a comprehensive data set related to the target product, which may include but is not limited to the product category, formula information, and feature description.

[0032] In the embodiment of the present application, promotional materials refer to textual content used for product promotion and marketing.

[0033] Specifically, first, it is necessary to obtain the product information of the target product. The product information usually contains various attributes, characteristics and related data of the target product, such as product name, category, formula ingredients, functional characteristics, etc. The way to obtain product information can be through database query, user input or automatic extraction from the product management system. Once the complete product information is obtained, the system can fully understand the characteristics of the target product and provide an accurate basis for the subsequent generation of promotional materials. The system will match the product category of the target product from the preset product category library, which can ensure that the generated promotional materials are highly relevant to the product type. Then, based on the matched product category, the first promotional material associated with the product category is extracted from the preset promotional material library. However, only general promotional materials are not enough, and personalized processing is also required for the uniqueness of the target product. Therefore, the system will perform semantic analysis on the extracted first promotional material, extract the keywords and themes therein, so that the system can understand the core content and structure of the general promotional material. Combined with the feature description extracted from the product information, the second promotional material is generated based on the keywords and themes obtained from the previous analysis. This process actually integrates the unique characteristics of the product into the general slogan, thereby creating promotional content that meets industry standards and highlights the personality of the product. Finally, the first promotional material and the second promotional material are intelligently integrated to obtain the promotional material of the final target product.

[0034] Based on the above embodiment, as an optional embodiment, the step of obtaining product information of the target product and determining the promotional material based on the product information may further include the following steps: Step 101: Match the product category of the target product from a preset product category library.

[0035] Specifically, the preset product category library is a manually designed and classified data structure that contains various possible product types and their feature descriptions. The existence of this library enables the system to quickly and accurately identify and classify new target products, thereby improving the efficiency and accuracy of the entire recommendation process.

[0036] First, the system extracts the key features of the target product, which may include information such as product name, main functions, and applicable population. Then, text analysis techniques such as keyword extraction and semantic analysis are used to process these features and generate a feature vector. This feature vector is then compared with each category in the preset product category library. This comparison process may use algorithms such as cosine similarity and Euclidean distance to calculate the similarity or distance between the feature vector and each category, and select the product category with the highest similarity or distance as the target product.

[0037] To improve the accuracy of matching, the system can also use machine learning models, such as support vector machines (SVM) or classification algorithms such as random forests. These models can better handle complex classification situations by learning a large amount of classified product data. In addition, the hierarchical relationship of products will be considered. For example, a product may belong to both the "skin care products" and "sunscreen" categories, where "sunscreen" is a subcategory of "skin care products". By establishing a hierarchical structure of categories, the system can more accurately locate the specific category of the product.

[0038] Step 102: Based on the product category, extract the first promotional material associated with the product category from a preset promotional material library.

[0039] Specifically, the system will use the previously matched product categories as search keywords to perform preliminary screening in the preset promotional material library. This screening process not only considers the exact match, but also uses semantic similarity algorithms to identify related categories. For example, if the target product belongs to the "sunscreen" category, the system will not only extract materials directly related to "sunscreen", but may also consider materials in related categories such as "sun protection" and "skin care products". The screened materials are sorted by relevance. This sorting process may consider multiple factors, such as the frequency of use of the material, the last update time, user feedback ratings, etc. The system may use machine learning algorithms, such as PageRank, to combine these factors and derive the final relevance score, which can ensure that the first promotional material extracted is not only relevant, but also the best and most applicable content at the moment. In order to further improve the accuracy of extraction, the system may also consider the specific attributes and target market of the product. For example, if the target product is a sunscreen specifically for sensitive skin, the system will give priority to promotional materials containing keywords such as "sensitive skin" and "gentle formula". This fine-grained matching can be achieved by using advanced natural language processing technologies such as named entity recognition (NER) and relationship extraction. During the extraction process, the system also pays attention to maintaining the diversity of materials. This means not only extracting direct product descriptions, but also including product advantages, usage methods, applicable scenarios, and other aspects. To achieve this, the system may use topic modeling algorithms, such as LDA, to ensure that the extracted materials cover multiple topic dimensions.

[0040] Step 103: Perform semantic analysis on the first promotional material to extract keywords and themes.

[0041] Specifically, the system uses natural language processing (NLP) techniques, such as the term frequency-inverse document frequency (TF-IDF) algorithm, word vector models (such as Word2Vec or GloVe), etc., to analyze the first promotional material and extract keywords. At the same time, the system also uses topic model algorithms, such as latent Dirichlet allocation (LDA) or non-negative matrix factorization (NMF), to identify the main topics in the first promotional material. The purpose of this step is to understand the core content and structure of the first promotional material and lay the foundation for subsequent personalized processing.

[0042] Step 104: Generate second promotional material based on the keywords and themes combined with the feature description in the product information.

[0043] Specifically, the second promotional material is generated by combining the feature description extracted from the product information, as well as the keywords and themes obtained based on the previous analysis. This process uses natural language generation (NLG) technology, such as template-based generation methods or more advanced neural network generation models. The system will incorporate the unique features of the product (such as special functions, innovative formulas, target user groups, etc.) into the generated content. For example, if the target product is a sunscreen for sensitive skin, the system will emphasize its mild formula, high protection index and other characteristics. The purpose of this is to create personalized promotional content that can highlight the unique selling points of the target product. In the process of generating the second promotional material, the system will also consider the diversity and innovation of the language. For example, techniques such as synonym replacement and sentence reorganization will be used to avoid being too similar to the first promotional material. At the same time, attention will be paid to maintaining the coherence and logic of the generated content, which can be achieved by using attention mechanisms or rule-based post-processing.

[0044] Step 105: The first promotional material and the second promotional material are integrated to obtain the promotional material of the target product.

[0045] Specifically, the first promotional material is intelligently integrated with the second promotional material to obtain the final promotional material for the target product. This integration process is not a simple splicing, but a complex content reorganization and optimization process, using text summarization technology to extract the core content of the two materials, and then using natural language generation technology to create a new, coherent text. In this process, the system will pay attention to maintaining the integrity of key information while optimizing the overall structure and fluency.

[0046] Step 20: Extract the recipe information of the target product from the product information, and determine the first set of recommended words based on the recipe information and the promotional materials.

[0047] In the embodiment of the present application, the formula information refers to the specific composition and ingredient data of the target product, including but not limited to the various raw materials or ingredients of the product and their proportions, the specific content or percentage of each ingredient, and physical or chemical property parameters related to the product.

[0048] In the embodiment of the present application, the first recommended word set refers to a group of recommended words or phrases initially generated based on the recipe information and promotional materials.

[0049] Specifically, to extract formula information from product information, the system uses natural language processing (NLP) technologies, such as named entity recognition (NER) and relation extraction, to identify and extract the formula-related parts of the product information, including various raw materials or ingredients and their proportions, the specific content or percentage of each ingredient, the functional description or characteristic description of the ingredients, etc. For example, for a skin care product, the system may extract ingredients such as water, glycerin, various plant extracts, preservatives, and their specific content percentages. At the same time, the system also uses rule matching or machine learning models to identify the functions and characteristics of these ingredients, such as moisturizing, anti-oxidation, whitening, etc. Then, the extracted formula information is deeply analyzed, such as using professional formula analysis models, such as ingredient efficacy evaluation models or formula optimization algorithms. These models can evaluate the overall effect of the formula, identify key ingredients and their synergies, and predict possible product performance. The formula analysis results are then intelligently integrated with the previously generated promotional materials. This process uses text matching algorithms and semantic similarity calculations to find the correlation between formula features and promotional focus. Based on these analysis and matching results, the system begins to generate the first set of recommended words. This process involves a variety of technologies, including keyword extraction, phrase generation, synonym expansion, and attribute-effect mapping. During the generation process, the system also considers the diversity and balance of words, ensuring that both technical and emotional words are included, while maintaining a balance between professional terms and popular expressions.

[0050] Based on the above embodiment, as an optional embodiment, the step of determining the first recommended word set based on the recipe information and the promotional material may further include the following steps: Step 201: Establish multi-dimensional formula evaluation indicators, which include nutritional indicators, functional indicators and market indicators.

[0051] Specifically, we first establish a multi-dimensional formula evaluation index system, which includes three categories of indicators: nutritional indicators, functional indicators and market indicators. Nutritional indicators may include but are not limited to protein content, fat content, vitamin content, etc.; functional indicators include but are not limited to moisturizing effect, antioxidant capacity, sun protection index, etc.; market indicators include but are not limited to cost-effectiveness, market demand, competitive advantages, etc. The system will set a weight for each indicator to reflect its importance in the overall evaluation. These weights can also be adjusted dynamically based on factors such as product category and target market.

[0052] Step 202: Based on the parameter values ​​in the recipe information, calculate various indexes corresponding to the target product in each recipe evaluation index.

[0053] Specifically, the original parameter values ​​in the formula information are standardized to ensure that parameters of different units and scales can be compared. The system then uses a preset algorithm or model to calculate the score for each indicator. For example, for the protein content in the nutritional indicator, the system may set an ideal range based on the product category, and then calculate the score based on the degree of fit between the actual content and the ideal range. For functional indicators, the system may use a specialized efficacy evaluation model, such as a moisturizing effect evaluation model or an antioxidant capacity evaluation model, to calculate the corresponding index. For market indicators, the calculation can be combined with market research data and competitive product analysis results.

[0054] Step 203: According to each index, corresponding description words are extracted from a preset index mapping vocabulary.

[0055] Specifically, according to the various indices calculated previously, the corresponding descriptive words are extracted from the preset index mapping word library. The index mapping word library is a data structure that contains the mapping relationship between various index ranges and corresponding descriptive words. For example, for the protein content index, there may be descriptive words such as "high protein", "rich in protein", and "high-quality protein source"; for the moisturizing effect index, there may be descriptive words such as "deep moisturizing", "long-term moisture lock", and "hydrating and nourishing". The system will select the most appropriate descriptive words based on the specific values ​​of each index. This process also involves fuzzy matching algorithms to handle situations where the index value falls on the boundary of the interval.

[0056] Step 204: Perform sentiment analysis and key information extraction on the promotional material, and calculate semantic relevance of the description words based on the sentiment analysis results and key information.

[0057] Specifically, sentiment analysis and key information extraction are performed on the promotional materials. Sentiment analysis uses natural language processing (NLP) technology, such as sentiment lexicon or deep learning models, to identify the sentiment tendency and intensity in the promotional materials. This may include the judgment of positive / negative sentiment, as well as the identification of specific sentiment categories (such as excitement, trust, expectation, etc.). At the same time, the system uses keyword extraction algorithms, such as TF-IDF or TextRank, to identify the core concepts and key information in the promotional materials. Based on the sentiment analysis results and the extracted key information, the semantic relevance of the descriptive words previously selected from the index mapping vocabulary is calculated. This process uses semantic similarity algorithms, such as cosine similarity or Word2Vec models, to calculate the degree of relevance between each descriptive word and the sentiment tendency and key information of the promotional materials. For example, if the sentiment analysis results of the promotional materials show that the product is positioned as "high-end luxury", then descriptive words related to "affordable" may receive a lower relevance score, even if they are appropriate in the index mapping.

[0058] Step 205: Select several target description words whose semantic relevance is higher than the relevance threshold, and combine and transform the target description words according to the product formula characteristics, and use the combined and transformed word set as the first recommended word set.

[0059] Specifically, the first step is to set a semantic relevance threshold. This threshold is set based on a large amount of historical data analysis and expert experience, with the goal of screening out descriptive words that are highly relevant to product characteristics and promotional positioning. The system will compare the semantic relevance score of each descriptive word with the set threshold. Only words with a score higher than the threshold will be selected as target descriptive words, ensuring that the selected words have high relevance and accuracy. Then, according to the specific formula characteristics of the product, the selected target descriptive words are combined and transformed. For word combination: phrase structure analysis and dependency syntax analysis are used to identify possible word combination methods. For example, "high protein" and "low fat" are combined into "high protein and low fat formula". For word transformation: word inflection algorithms and synonym replacement techniques are used to appropriately transform the target descriptive words. For example, "antioxidant" is transformed into "strong antioxidant capacity" or "excellent antioxidant effect". In the process of word combination and transformation, the system also considers grammatical correctness, semantic coherence and market acceptance. It uses a pre-trained language model to evaluate whether the generated word combination is natural and fluent, and to ensure that no ambiguous or misleading expressions are generated. At the same time, the system will refer to industry norms and laws and regulations to avoid generating words that involve exaggerated propaganda or illegal descriptions. Finally, the combined and transformed word set is used as the first recommended word set.

[0060] Step 30: Detect and process sensitive words in the first recommended word set to obtain a second recommended word set.

[0061] In the embodiments of the present application, sensitive words refer to words that need to be used with caution or avoided during product publicity and promotion, including but not limited to: illegal and unlawful words, inappropriate efficacy claims, words that may cause controversy or misunderstanding, and words that are banned or restricted in the industry.

[0062] The second recommended word set in the embodiment of the present application refers to the recommended word set after sensitive word detection and processing.

[0063] Specifically, each word in the first recommended word set is segmented to obtain the segmentation result. This step uses the segmentation algorithm in natural language processing (NLP) technology, such as the dictionary-based maximum forward matching method or the statistical conditional random field (CRF) method, to split compound words into the smallest semantic units. The segmentation result is matched with the preset multi-level sensitive word library to obtain the first sensitive word list. The hierarchical sensitive word library is a continuously updated data structure containing various types of sensitive words. Since simple string matching may lead to misjudgment, the system will perform contextual semantic analysis on each word in the first sensitive word list, filter the first sensitive word list based on the contextual semantic analysis results, and obtain a more accurate second sensitive word list. The second sensitive word list is then graded into high-level sensitive words, medium-level sensitive words and low-level sensitive words. This division is based on factors such as the sensitivity of the words, potential risks and usage restrictions. Finally, sensitive words of different levels are processed accordingly according to the preset processing strategy to obtain the second recommended word set. This new word set not only retains the compliant and appropriate descriptive words in the first recommended word set, but also includes modified or replaced safe words. It is more compliant and safe as a whole, significantly reduces the risk of using inappropriate words in product promotion, and improves the compliance and safety of promotional content.

[0064] Based on the above embodiment, as an optional embodiment, the step of detecting and processing sensitive words in the first recommended word set to obtain the second recommended word set may further include the following steps: Step 301: Perform word segmentation processing on each word in the first recommended word set to obtain a word segmentation result.

[0065] Specifically, the system performs word segmentation on each word in the first set of recommended words. The system uses word segmentation algorithms in natural language processing (NLP) technology, such as the dictionary-based maximum forward matching method or the statistical conditional random field (CRF) method, to split compound words into the smallest semantic units. For example, for a recommended word such as "high-efficiency antioxidant formula", the system may decompose it into several basic semantic units such as "high-efficiency", "antioxidant", and "formula". The purpose of this step is to break down complex words or phrases into smaller units to facilitate subsequent sensitive word matching.

[0066] Step 302: Match the word segmentation result with a preset multi-level sensitive word library to obtain a first sensitive word list.

[0067] Specifically, the segmentation results are matched with a preset multi-level sensitive word library, which contains various types of sensitive words, such as illegal and illegal words, inappropriate efficacy claims, and words that may cause controversy or misunderstanding. The multi-level structure of the sensitive word library reflects the severity and usage restrictions of different sensitive words. The matching process uses efficient string matching algorithms, such as the KMP algorithm, to quickly identify potential sensitive words. These algorithms can complete sensitive word matching of large-scale texts in linear time, ensuring processing efficiency. In the matching process, the system not only considers complete matching, but also approximate matching and variant matching to deal with possible circumvention methods, such as homophone replacement or spelling deformation. After the matching is completed, the system obtains the first sensitive word list, which contains all potential sensitive words found in the segmentation results. Each sensitive word is marked with its position in the original text, the sensitive word category to which it belongs, and the severity.

[0068] Step 303: Perform context semantic analysis on each word in the first sensitive word list, and filter the first sensitive word list based on the context semantic analysis result to obtain a second sensitive word list.

[0069] Specifically, contextual semantic analysis is performed on each word in the first sensitive word list, such as a word vector model, which can capture the semantic relationship and contextual dependency between words, so as to more accurately understand the actual intention of the words. When performing contextual semantic analysis, the system not only considers the target word itself, but also the surrounding words and the structure of the entire sentence. For example, for the potential sensitive word "high", the system will analyze whether it is used to describe the objective characteristics of the product (such as "high protein content") or used for exaggerated publicity (such as "highest effect"). The system will also consider industry-specific language habits and expressions. For example, in certain specific industries, certain words may have special, non-literal meanings. Through the preset filtering rules, the second sensitive word list is finally obtained. This new list not only contains sensitive words that have been verified by the context, but also adds more detailed semantic information and risk assessment results for each sensitive word.

[0070] Step 304: The second sensitive word list is graded and corresponding sensitive word processing is performed according to a preset processing strategy to obtain a second recommended word set.

[0071] Specifically, the first step is to classify the sensitive words in the second sensitive word list into three levels: high, medium, and low. This classification is based on multiple factors, including but not limited to: the degree of legal risk of the word, the degree of misunderstanding or controversy that may be caused, industry regulatory requirements, and the actual impact of the word in a specific context. The system uses pre-trained machine learning models, such as support vector machines (SVM) or random forests, to automatically perform this classification. The input to the model includes the word itself, its contextual information, historical usage data, and the latest regulations and industry standards information.

[0072] Differentiated processing strategies are adopted for sensitive words of different levels. For highly sensitive words, the system adopts a direct deletion strategy. These words usually involve serious legal risks or obvious false propaganda. Direct deletion is the safest and most effective processing method. While deleting these words, the system retains the relevant context information for possible subsequent manual review or interpretation.

[0073] For medium-level sensitive words, the system adopts a synonym replacement strategy, which can use natural language processing technology, such as word vector models or pre-trained language models (such as BERT), to generate semantically similar but non-sensitive replacement words. The system will consider the applicability of the replacement word in the current context to ensure that the replaced expression retains the original core meaning while avoiding potential risks. For example, replace "best" with "high-quality" or "excellent".

[0074] For low-level sensitive words, the system adopts a more flexible context reconstruction strategy, which involves the reconstruction of the entire phrase or sentence, rather than just the replacement of a single word. The system uses natural language generation (NLG) technology, such as rule-based generation methods or more complex neural network generation models (such as the GPT series), to re-express the original meaning. This method can avoid potential sensitivity by changing the expression while retaining the core information. For example, "no side effects" is reconstructed as "mild formula, suitable for daily use". In the process of these processing, the system also considers the overall semantic coherence and naturalness of expression, using language models to evaluate whether the generated new expression is fluent and natural, and ensure that the overall semantics will not be distorted or incoherent due to local modifications. Finally, the word set obtained through these different processing strategies is used as the second recommended word set. This new word set not only removes high-risk sensitive words, but also contains optimized and reconstructed safe expressions. It is more compliant and safer overall, while still retaining the core intention and appeal of the original expression.

[0075] Step 40: perform association expansion on the second recommended word set, and perform sensitive word verification on the expanded recommended word set to obtain a target recommended word set.

[0076] Specifically, the association expansion of the second set of recommended words can be achieved by using a variety of natural language processing (NLP) technologies, such as using word vector models (such as Word2Vec or GloVe) to find words that are similar to existing recommended words in the semantic space. This method can find words that are semantically related but expressed in different forms, enriching the expression of recommended words. Secondly, the system uses pre-trained language models (such as BERT) to generate new words or phrases that are related to existing recommended words in context. This method can produce more complex and rich expressions, increasing the diversity of recommended words.

[0077] In addition, knowledge graph technology can be used to expand concepts. Through pre-built industry-specific knowledge graphs, the system can find words that are conceptually related to existing recommended words. This method is particularly suitable for discovering cross-domain but related expressions, adding new dimensions to the recommended word set. For example, for the word "moisturizing" for a skin care product, the system may expand to related concepts such as "hydrating", "locking moisture", and "hydrating" through the knowledge graph. After completing the associated expansion, the system will conduct a comprehensive sensitive word check on the expanded recommended word set. This step uses the same sensitive word detection technology as before, including multi-level sensitive word library matching, contextual semantic analysis, etc., which will not be repeated here. After a series of expansion and verification processes, the system finally obtains the target recommended word set. This word set not only includes the original safe words, but also adds verified new words and expressions, which is richer, more diverse and strictly compliant as a whole.

[0078] Based on the above embodiment, as an optional embodiment, the step of performing association expansion on the second recommended word set and performing sensitive word verification on the expanded recommended word set to obtain the target recommended word set may further include the following steps: Step 401: construct a word association network, wherein the word association network includes a synonym layer, a near-sense layer, a hypernym layer, and a hyponym layer.

[0079] Specifically, a word association network is constructed, which is a complex multi-level structure including synonymy, near-meaning, hypernym, and hyponym layers. The system uses a variety of language resources and technologies to build this network, including existing dictionaries (such as WordNet), large-scale text corpora, and pre-trained word vector models (such as Word2Vec or GloVe). The synonymy layer contains words with exactly or almost the same meaning; the near-meaning layer contains words with similar meanings but with subtle differences; the hypernym layer contains broader or abstract concepts; and the hyponym layer contains more specific or specialized concepts. During the construction process, the system also considers industry-specific language habits and terminology. For example, for the cosmetics industry, "moisturizing" may be associated with words such as "hydrating" and "locking water" at the near-meaning layer, while "skin care" may be a hypernym of "moisturizing". The system uses natural language processing (NLP) techniques, such as semantic similarity calculation and hierarchical clustering algorithms, to automatically identify and classify these relationships.

[0080] Step 402: For each word in the second recommended word set, mining is performed in the word association network to obtain an associated word list, wherein the associated word list includes the association degree of each word.

[0081] Specifically, after constructing the word association network, the system performs association mining on each word in the second recommended word set. This process involves performing a breadth-first search or a depth-first search in the constructed network to find all words related to the target word. The system considers different levels in the network. For example, for the word "moisturizing", the system not only finds synonyms and near synonyms, but also considers its hypernyms (such as "skin care") and hyponyms (such as "deep moisturizing" and "long-lasting moisturizing"). The association between each associated word and the target word is calculated during the mining process. This association is based on multiple factors, including hierarchical distance in the network, cosine similarity in the word vector space, and co-occurrence frequency in a large-scale corpus. The system uses machine learning algorithms, such as random forests or neural networks, to combine these factors and calculate the final association score. Finally, a list of associated words is generated, including each associated word and its corresponding association. This list is sorted from high to low by association, providing a clear priority for subsequent word selection.

[0082] Step 403: Filter out words with a relevance higher than a relevance threshold from the relevance word list as candidate expansion words.

[0083] Specifically, words with a correlation higher than a preset correlation threshold are selected from the list of related words as candidate expansion words. This threshold is set based on a large amount of experimental data and expert experience, aiming to retain words with sufficient correlation while excluding associations that are too loose or irrelevant. A simple comparison algorithm can be used to quickly complete this screening, ensuring that only the most relevant words enter the next stage of processing.

[0084] Step 404: Perform semantic adaptation scoring on each candidate expansion word, filter out words with scores below a scoring threshold, and obtain an initial expansion word set.

[0085] Specifically, the semantic adaptation score of each candidate expansion word is evaluated, taking into account multiple factors, including but not limited to: the matching degree between the word and product characteristics: the system analyzes whether each candidate word accurately reflects the core characteristics and advantages of the product; the language preference of the target audience: based on the preset target audience portrait, the acceptance and attractiveness of the word are evaluated; the consistency of brand tone: ensuring that the word is consistent with the overall image and communication style of the brand; the fit of market trends: considering the current market hot spots and consumer concerns, the timeliness of the word is evaluated. Finally, machine learning algorithms, such as support vector machines (SVM) or random forests, are used to integrate these factors and calculate the final semantic adaptation score, which reflects the overall applicability of the candidate word in the current context.

[0086] Step 405: Perform sensitive word verification on each word in the initial expanded word set to obtain a target recommended word set.

[0087] Specifically, each word in the initial expanded word set is checked for sensitive words. This step uses the same sensitive word detection technology as before, including multi-level sensitive word library matching, contextual semantic analysis, etc. However, at this stage, the system will adopt stricter standards to ensure that the final recommended word set fully complies with laws, regulations and industry standards. The system not only checks individual words, but also analyzes potential problems that may arise from words in specific contexts to ensure that even specific combinations of legal words will not cause misleading or inappropriate implications.

[0088] Based on the above embodiment, as an optional embodiment, a method for intelligent sensitive word detection and word recommendation may also include the following process: Specifically, a multimodal recommendation system integrating text analysis module, image analysis module and audio analysis module is established. The core of this system is to realize semantic mapping and conversion between text, image and audio. The text analysis module is responsible for understanding and parsing text content and extracting key concepts and themes; the image analysis module can understand the visual features and semantic content of the image; the audio analysis module focuses on the recognition of sound features and emotional expressions. These three modules are connected and interacted through a common semantic space to realize cross-modal information conversion and recommendation.

[0089] The system inputs the target recommended word set into the text analysis module, which not only analyzes the literal meaning of the words, but also considers the emotional color, style characteristics and implicit semantics of the words. Based on these analysis results, the system generates corresponding visual element suggestions and audio element suggestions. For example, for the word "fresh", the system may recommend the use of bright and clear image elements, as well as brisk and soothing audio elements.

[0090] Based on the visual element suggestions, product display pictures are selected from the pre-established image library. This process involves semantic matching and visual feature analysis of the images. The system will consider factors such as the color, composition, and theme of the image to select those pictures that can best reflect the semantics and emotions of the recommended word set. At the same time, the system will also consider the characteristics of the product and the preferences of the target audience to ensure that the selected pictures not only meet the semantic requirements, but also effectively display the product features and attract target users.

[0091] Similarly, the system selects appropriate audio materials from the audio material library based on audio element suggestions. This may include background music, sound effects or voice content. The selection process will take into account the rhythm, emotional expression, style characteristics, etc. of the audio to ensure that the selected audio elements can enhance the expressiveness of text and images and create a harmonious and unified listening experience.

[0092] Finally, the system performs multimodal fusion of the target recommendation word set, the screened product display pictures and the selected audio materials to generate a comprehensive promotional plan including text, images and audio. This fusion process is not a simple superposition of elements, but an intelligent content integration. The system will consider the semantic consistency, emotional coordination and expressive balance between the various elements to ensure that the final promotional plan can form an organic whole, with each element echoing and reinforcing each other.

[0093] See also Figure 3 , is a module diagram of an intelligent sensitive word detection and word recommendation system provided in an embodiment of the present application, the intelligent sensitive word detection and word recommendation system may include: a material determination module, an initial word set recommendation module, a sensitive word detection module and a target word set recommendation module, wherein: A material determination module, used to obtain product information of a target product and determine promotional materials based on the product information; An initial word set recommendation module, configured to extract the recipe information of the target product from the product information, and determine a first recommended word set based on the recipe information and the promotional material; A sensitive word detection module, used to detect and process the sensitive words in the first recommended word set to obtain a second recommended word set; The target word set recommendation module is used to perform association expansion on the second recommended word set, and perform sensitive word verification on the expanded recommended word set to obtain a target recommended word set.

[0094] Optionally, the material determination module is further used to match the product category of the target product from a preset product category library; based on the product category, extract a first promotional material associated with the product category from a preset promotional material library; perform semantic analysis on the first promotional material to extract keywords and themes; generate a second promotional material based on the keywords and themes in combination with a feature description in the product information; and merge the first promotional material with the second promotional material to obtain the promotional material of the target product.

[0095] Optionally, the initial word set recommendation module is also used to establish multi-dimensional formula evaluation indicators, which include nutritional indicators, functional indicators and market indicators; based on the parameter values ​​in the formula information, calculate the various indexes corresponding to the target product in each of the formula evaluation indicators; according to the various indexes, extract corresponding descriptive words from a preset index mapping vocabulary; perform sentiment analysis and key information extraction on the promotional materials, and based on the sentiment analysis results and key information, calculate the semantic relevance of the descriptive words; select several target description words whose semantic relevance is higher than the relevance threshold, and combine and transform the target description words according to the product formula characteristics, and use the combined and transformed word set as the first recommended word set.

[0096] Optionally, the sensitive word detection module is also used to perform word segmentation processing on each word in the first recommended word set to obtain a word segmentation result; match the word segmentation result with a preset multi-level sensitive word library to obtain a first sensitive word list; perform contextual semantic analysis on each word in the first sensitive word list, and filter the first sensitive word list based on the contextual semantic analysis result to obtain a second sensitive word list; classify the second sensitive word list, and perform corresponding sensitive word processing according to a preset processing strategy to obtain a second recommended word set.

[0097] Optionally, the sensitive word detection module is further used to divide the sensitive words in the second sensitive word list into high-level sensitive words, medium-level sensitive words and low-level sensitive words; for the high-level sensitive words, a deletion processing strategy is adopted; for the medium-level sensitive words, a synonym replacement processing strategy is adopted; for the low-level sensitive words, a context reconstruction processing strategy is adopted to generate new words; and the word sets obtained by the corresponding processing strategies for sensitive words of each level are used as the second recommended word set.

[0098] Optionally, the target word set recommendation module is also used to construct a word association network, which includes a synonym layer, a near-meaning layer, a hypernym layer and a hyponym layer; for each word in the second recommended word set, the word association network is mined to obtain a list of associated words, and the associated word list includes the association degree of each word; from the associated word list, words with an association degree higher than an association degree threshold are screened out as candidate expansion words; semantic adaptation scoring is performed on each of the candidate expansion words, and words with a score lower than the score threshold are screened out to obtain an initial expansion word set; sensitive word verification is performed on each word in the initial expansion word set to obtain a target recommended word set.

[0099] Optionally, the target word set recommendation module is also used to establish a multimodal recommendation system integrating a text analysis module, an image analysis module and an audio analysis module; inputting the target recommended word set into the text analysis module to generate corresponding visual element suggestions and audio element suggestions; based on the visual element suggestions, screening product display pictures from the image material library; based on the audio element suggestions, selecting audio materials from the audio material library; performing multimodal fusion of the target recommended word set, the product display pictures and the audio materials to generate a comprehensive promotional plan including text, images and audio.

[0100] It should be noted that: when the system provided in the above embodiment realizes its functions, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0101] The embodiment of the present application also provides a computer storage medium, which can store multiple instructions. The instructions are suitable for being loaded by a processor and executing an intelligent sensitive word detection and word recommendation method of the above embodiment. The specific execution process can be found in the specific description of the above embodiment, which will not be repeated here.

[0102] Please refer to Figure 3 The application also discloses an electronic device. Figure 3 The electronic device 300 may include: at least one processor 301 , at least one network interface 304 , a user interface 303 , a memory 305 , and at least one communication bus 302 .

[0103] The communication bus 302 is used to realize the connection and communication between these components.

[0104] The user interface 303 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 303 may also include a standard wired interface and a wireless interface.

[0105] The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0106] Among them, the processor 301 may include one or more processing cores. The processor 301 uses various interfaces and lines to connect various parts in the entire server, and executes various functions of the server and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 305, and calling data stored in the memory 305. Optionally, the processor 301 can be implemented in at least one hardware form of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field-Programmable Gate Array, FPGA), and programmable logic array (Programmable Logic Array, PLA). The processor 301 can integrate one or a combination of a central processing unit (Central Processing Unit, CPU), a graphics processing unit (Graphics Processing Unit, GPU) and a modem. Among them, the CPU mainly processes the operating system, user interface and application programs; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 301, and it can be implemented separately through a chip.

[0107] Among them, the memory 305 may include a random access memory (Random Access Memory, RAM) and may also include a read-only memory (Read-Only Memory). Optionally, the memory 305 includes a non-transitory computer-readable storage medium. The memory 305 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 305 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data involved in the above-mentioned various method embodiments, etc. The memory 305 may optionally also be at least one storage device located away from the aforementioned processor 301. Refer to Figure 3, the memory 305 as a computer storage medium may include an operating system, a network communication module, a user interface module, and an application program of an intelligent sensitive word detection and word recommendation method.

[0108] exist Figure 3 In the electronic device 300 shown, the user interface 303 is mainly used to provide an input interface for the user and obtain the data input by the user; and the processor 301 can be used to call the application program storing an intelligent sensitive word detection and word recommendation method in the memory 305, and when executed by one or more processors 301, the electronic device 300 executes one or more of the methods described in the above embodiments. It should be noted that for the aforementioned method embodiments, for the sake of simple description, they are all expressed as a series of action combinations, but those skilled in the art should know that the present application is not limited to the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required for the present application.

[0109] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0110] In the several embodiments provided in this application, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are only schematic, such as the division of units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interfaces, and the indirect coupling or communication connection of devices or units can be electrical or other forms.

[0111] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0112] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0113] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a memory and includes several instructions for a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned memory includes: various media that can store program codes, such as USB flash drives, mobile hard drives, magnetic disks or optical disks.

[0114] The above is only an exemplary embodiment of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the disclosure of the specification and the truth of practice, those skilled in the art will easily think of other embodiments of the present disclosure.

[0115] This application is intended to cover any variation, use or adaptation of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary technical means in the art not described in the present disclosure. The description and examples are to be regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. An intelligent sensitive word detection and word recommendation method, characterized in that: The method comprises: Acquire product information of a target product, and determine promotional materials based on the product information; extracting the recipe information of the target product from the product information, and determining a first set of recommended words based on the recipe information and the promotional material; Detecting and processing sensitive words in the first recommended word set to obtain a second recommended word set; The second recommended word set is associated and expanded, and the expanded recommended word set is checked for sensitive words to obtain a target recommended word set.

2. The intelligent sensitive word detection and word recommendation method according to claim 1, characterized in that: The determining of promotional materials based on the product information includes: Matching the product category of the target product from a preset product category library; Based on the product category, extracting a first promotional material associated with the product category from a preset promotional material library; Performing semantic analysis on the first promotional material to extract keywords and themes; Generate second promotional materials based on the keywords and themes combined with the feature description in the product information; The first promotional material and the second promotional material are merged to obtain the promotional material of the target product.

3. The intelligent sensitive word detection and word recommendation method according to claim 1, characterized in that: The determining of the first recommended word set based on the recipe information and the promotional material includes: Establishing multi-dimensional formula evaluation indicators, including nutritional indicators, functional indicators and market indicators; Based on the parameter values ​​in the formula information, calculating various indexes corresponding to the formula evaluation indicators of the target product; According to the various indices, corresponding descriptive words are extracted from a preset index mapping vocabulary; Performing sentiment analysis and key information extraction on the promotional material, and calculating semantic relevance of the descriptive words based on the sentiment analysis results and key information; A number of target description words with semantic relevance higher than a relevance threshold are selected, and the target description words are combined and transformed according to product formula characteristics, and the combined and transformed word set is used as the first recommended word set.

4. The intelligent sensitive word detection and word recommendation method according to claim 1, characterized in that: The detecting and processing the sensitive words in the first recommended word set to obtain a second recommended word set includes: Performing word segmentation processing on each word in the first recommended word set to obtain a word segmentation result; Matching the word segmentation result with a preset multi-level sensitive word library to obtain a first sensitive word list; Performing context semantic analysis on each word in the first sensitive word list, and filtering the first sensitive word list based on the context semantic analysis result to obtain a second sensitive word list; The second sensitive word list is graded, and corresponding sensitive word processing is performed according to a preset processing strategy to obtain a second recommended word set.

5. The intelligent sensitive word detection and word recommendation method according to claim 4, characterized in that: The step of classifying the second sensitive word list and performing corresponding sensitive word processing according to a preset processing strategy to obtain a second recommended word set includes: Classify each sensitive word in the second sensitive word list into high-level sensitive words, medium-level sensitive words and low-level sensitive words; For the highly sensitive words, a deletion strategy is adopted; For the intermediate sensitive words, a synonym replacement processing strategy is adopted; For the low-level sensitive words, a processing strategy of reconstructing the context to generate new words is adopted; The word set obtained by processing the sensitive words of each level with the corresponding processing strategy is used as the second recommended word set.

6. The intelligent sensitive word detection and word recommendation method according to claim 1, characterized in that: The step of performing association expansion on the second recommended word set and performing sensitive word verification on the expanded recommended word set to obtain a target recommended word set includes: Constructing a word association network, wherein the word association network includes a synonym layer, a near-sense layer, a hypernym layer, and a hyponym layer; For each word in the second recommended word set, mining is performed in the word association network to obtain an associated word list, wherein the associated word list includes an association degree of each word; Filtering out words with a correlation degree higher than a correlation degree threshold from the associated word list as candidate expansion words; Performing semantic adaptation scoring on each of the candidate expansion words, filtering out words with scores below a scoring threshold, and obtaining an initial expansion word set; A sensitive word check is performed on each word in the initial expanded word set to obtain a target recommended word set.

7. The intelligent sensitive word detection and word recommendation method according to claim 1, characterized in that: The method further comprises: Establish a multimodal recommendation system that integrates text analysis module, image analysis module and audio analysis module; Inputting the target recommendation word set into the text analysis module to generate corresponding visual element suggestions and audio element suggestions; Based on the visual element suggestions, filter product display images from the image library; Based on the audio element suggestion, select audio material from an audio material library; The target recommendation word set, the product display pictures and the audio material are multimodally fused to generate a comprehensive promotional plan including text, images and audio.

8. An intelligent sensitive word detection and word recommendation system, characterized in that: The system comprises: A material determination module, used to obtain product information of a target product and determine promotional materials based on the product information; An initial word set recommendation module, configured to extract the recipe information of the target product from the product information, and determine a first recommended word set based on the recipe information and the promotional material; A sensitive word detection module, used to detect and process the sensitive words in the first recommended word set to obtain a second recommended word set; The target word set recommendation module is used to perform association expansion on the second recommended word set, and perform sensitive word verification on the expanded recommended word set to obtain a target recommended word set.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor and executing the method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: It includes a processor, a memory, a user interface and a network interface, the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the method as described in any one of claims 1-7.

Images