Enterprise new media operation strategy generation and dynamic correction method based on large model

Abstract

The invention relates to the technical field of commercial operation, and discloses an enterprise new media operation strategy generation and dynamic correction method based on a large model, and the method comprises the steps: obtaining an industry label, enterprise user data, industry hotspot data, enterprise content data in a specified time period, and enterprise competitive product data of a new media enterprise; determining the adjusted hotspot keyword set, the popular keyword set and the key weight of each popular keyword in the adjusted popular keyword set of the new media enterprise; determining follow-up content data of the new media enterprise; and platform suggestion data and operation decision data of each platform of each historical content in the follow-up content data are determined, and operation strategy production and dynamic correction of the new media enterprise are realized. According to the method, the blindness and the hysteresis of the strategy can be avoided, the whole-process intelligence of the operation strategy from data mining to landing execution is realized, the market suitability, the accuracy and the iteration efficiency of the strategy are improved, and the content transmission ability and the core efficiency of user conversion are enhanced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of business operation, in particular to an enterprise new media operation strategy generation and dynamic correction method based on a large model. BACKGROUND

[0002] Early new media operation is in the traffic bonus period, and strategy formulation completely depends on the industry experience of operation personnel, has no data support, and only subjective judgment is used to create content and select a release channel, so the strategy is highly blind. With the development of new media platforms, the industry has entered the initial application stage of data, and enterprises have begun to count the reading and interaction data of single-platform content, and adjust content creation according to basic data, but the data dimension is single, industry hot spots and competitor data are not integrated, and the depth of user demand is also lacking. In the later stage, with the popularization of big data technology, some enterprises try to extract hot keywords and analyze competitor hits, but there are problems such as static strategy, poor market adaptability, and high trial and error cost, which are difficult to adapt to the rapidly changing market environment of new media, and there is an urgent need for intelligent whole-process operation strategy generation and optimization methods.

[0003] Therefore, the present application provides an enterprise new media operation strategy generation and dynamic correction method based on a large model. SUMMARY

[0004] The present application provides an enterprise new media operation strategy generation and dynamic correction method based on a large model, which determines the hot keyword set, the hit keyword set, and the key weight of each hit keyword in the adjusted hit keyword set of the new media enterprise by obtaining the industry label, enterprise user data, industry hot spot data, enterprise content data in a specified time period, and enterprise competitor data of the new media enterprise, determines the follow-up content data, platform suggestion data of each historical content on each platform in the follow-up content data, and operation decision data, and realizes the operation strategy production and dynamic correction of the new media enterprise. It can avoid strategy blindness, hysteresis and high trial and error cost, realize the whole-process intelligentization of operation strategy from data mining to landing execution, improve the market adaptability, accuracy and iteration efficiency of the strategy, and strengthen the core efficiency of content dissemination and user conversion.

[0005] The present application provides an enterprise new media operation strategy generation and dynamic correction method based on a large model, which includes: S1: obtaining the industry label, enterprise user data, and industry hot spot data of the new media enterprise, and obtaining the enterprise content data in a specified time period and enterprise competitor data of the new media enterprise; S2: determining the hot keyword set, hit keyword set, and key weight of each hit keyword in the adjusted hit keyword set of the new media enterprise based on the industry hot spot data and enterprise competitor data of the new media enterprise. S3: determining follow-up content data of the new media enterprise based on the adjusted hot keyword set, the best-selling keyword set, the key weight of all best-selling keywords in the adjusted best-selling keyword set, and the enterprise content data; S4: analyzing the follow-up content data of the new media enterprise, determining platform suggestion data of each historical content on each platform in the follow-up content data, and determining operation decision data of each historical content in the follow-up content data, to realize production of operation strategy of the new media enterprise and dynamic correction.

[0006] Preferably, the method for generating and dynamically correcting operation strategy of new media enterprise based on large model comprises obtaining industry label of new media enterprise, enterprise user data and industry hot spot data, including: Obtaining industry label of new media enterprise and enterprise user data, wherein the enterprise user data comprises user data of multiple users, and the user data comprises user label, user basic vector, user behavior vector and consumption value label, and the consumption value label comprises high value, medium value, low value, zero value and negative value; Obtaining industry hot spot data of new media enterprise based on industry label of new media enterprise.

[0007] Preferably, the method for generating and dynamically correcting operation strategy of new media enterprise based on large model comprises obtaining enterprise content data and enterprise competitor data of new media enterprise in a specified time period, including: Obtaining enterprise content data and enterprise competitor data of new media enterprise in a specified time period, wherein the enterprise content data comprises content performance data of multiple historical contents, and the content performance data comprises content duration, content form label, content performance vector of multiple platforms, content conversion vector, content keyword set and platform feedback data, the platform feedback data comprises feedback content of multiple user numbers, and the enterprise competitor data comprises competitor level label of multiple competitor enterprises and best-selling content data of multiple best-selling contents, the best-selling content data comprises publishing time, best-selling duration, content form label, best-selling performance vector of multiple platforms, best-selling conversion vector and best-selling keyword sub-set, and the content form label comprises image-text, video and live broadcast.

[0008] Preferably, the method for generating and dynamically correcting operation strategy of new media enterprise based on large model comprises determining hot keyword set and best-selling keyword set of new media enterprise based on industry hot spot data and enterprise competitor data of new media enterprise, including: Performing semantic analysis on industry hot spot data of new media enterprise to determine hot keyword set; Input the top-selling content performance vectors and top-selling conversion vectors of all platforms from all top-selling content data of all competitors into the content evaluation model to determine the value assessment value of each top-selling content of each competitor in the competitor data. Based on the subset of trending keywords in the trending content data of all competitor companies in the competitor data, determine the trending keyword set of the new media company and the first trending content set of each trending keyword in the trending keyword set, and count the occurrence frequency of each trending keyword in the trending keyword set. The first trending content set includes the trending content data of multiple trending content from multiple competitor companies.

[0009] The preferred method for generating and dynamically correcting enterprise new media operation strategies based on a large model determines the adjusted set of trending keywords, the set of viral keywords, and the key weight of each viral keyword in the adjusted set of viral keywords for new media enterprises, including: Based on the publication time of the content data corresponding to all subsets of popular keywords in the first popular content set, which includes the popular keyword in the popular keyword set, the publication time range of each popular keyword in the popular keyword set is determined. Based on the release time range and frequency of each popular keyword in the popular keyword set, the specified time period is divided to determine multiple specified time sub-periods for each popular keyword in the popular keyword set and the second popular content set for each specified time sub-period. Semantic normalization is performed on all hot keywords in the hot keyword set and all popular keywords in the popular keyword set to determine the adjusted hot keyword set and popular keyword set. Multiple specified time sub-periods and the second popular content set for each popular keyword in the adjusted popular keyword set are also determined. Based on the adjusted set of trending keywords for new media companies, the competitor level tags of all competitors in the company's competitor data, the adjusted set of popular keywords, all specified time sub-periods of all popular keywords in the adjusted set of popular keywords, and the set of second popular content in each specified time sub-period, the key weight of each popular keyword in the adjusted set of popular keywords for new media companies is calculated.

[0010] The preferred method for generating and dynamically correcting enterprise new media operation strategies based on a large model, uses adjusted sets of trending keywords, sets of viral keywords, key weights of all viral keywords in the adjusted set of viral keywords, and enterprise content data to determine the follow-up content data for new media enterprises, including: The adjusted set of hot keywords, the set of popular keywords, the key weights of all popular keywords in the set of popular keywords, and the set of content keywords in the content performance data of all historical content in the enterprise content data are input into the content recognition model to determine the content tags of each historical content in the enterprise content data. The content tags include follow-up and pause. Input the content performance vectors and content conversion vectors of all platforms from the content performance data of all historical content in the enterprise content data into the content evaluation model to determine the value assessment value of each piece of historical content in the enterprise content data; The value tag of historical content corresponding to each value assessment value exceeding the preset threshold is identified as "follow-up," and the value tag of historical content corresponding to each value assessment value not exceeding the preset threshold is identified as "pause." Based on the content performance data of all historical content tagged as "follow-up" or "value tag as follow-up", the follow-up content data of new media enterprises is determined. The follow-up content data includes the content performance data of multiple historical content.

[0011] The preferred method for generating and dynamically correcting enterprise new media operation strategies based on a large model analyzes the follow-up content data of new media enterprises, determines the platform suggestion data for each platform for each historical content in the follow-up content data, and realizes the production and dynamic correction of the operation strategy of new media enterprises, including: Semantic recognition is performed on the feedback content of each user ID in the platform feedback data of each platform in the content performance data of each historical content in the follow-up content data, and the feedback content tags of each user ID in the platform feedback data of each platform in the content performance data of each historical content are determined. The feedback content tags include product experience and content suggestions. Extract all feedback content from all user IDs whose feedback content tag is "content suggestion" from the platform feedback data of each platform in the content performance data of each historical content in the follow-up content data, and determine the content suggestion data of each platform for each historical content in the follow-up content data; Cluster analysis is performed on all feedback content of all user IDs in the content suggestion data of each platform for each historical content in the follow-up content data to determine multiple suggestion cluster data for each platform for each historical content in the follow-up content data, as well as suggestion labels for each suggestion cluster data. The suggestion cluster data includes suggestion fitting vectors and all user IDs. Based on the consumption value tags of all users in the enterprise user data and the suggestion clustering data of all suggestion tags of each historical content on each platform in the follow-up content data, calculate the initial tag clustering value, the iterative tag clustering value of multiple iterations, and the actual tag clustering value of each suggestion tag of each historical content on each platform in the follow-up content data, and calculate the adopted tag of each suggestion tag of each historical content on each platform in the follow-up content data. Based on the adopted tags of all suggested tags for each platform of each historical content in the follow-up content data, and the suggested fitting vectors in the suggested clustering data, the platform suggested data for each platform of each historical content in the follow-up content data is determined.

[0012] The preferred method for generating and dynamically correcting enterprise new media operation strategies based on a large model determines the operational decision data for each historical piece of content in the follow-up content data, thereby enabling the production and dynamic correction of new media enterprise operation strategies, including: The content duration, content format tags, content performance vectors, content conversion vectors, and platform suggestion data of each historical content in the follow-up content data are input into the content decision model to generate real-time content data and operational decision data for each historical content in the follow-up content data. The operational decision data includes at least the decision duration, release time, decision format tags, predicted performance vectors, and predicted conversion vectors for multiple platforms. Collect real-time performance data for each piece of content, and dynamically correct the content based on the real-time performance data and operational decision data. The real-time performance data includes real-time performance vectors, real-time conversion vectors, and real-time feedback data from multiple platforms.

[0013] The beneficial effects of this invention compared to existing technologies are as follows: By acquiring industry tags, enterprise user data, industry hot topic data, enterprise content data within a specified time period, and competitor data of new media enterprises, this invention determines the set of hot keywords, the set of viral keywords, and the key weight of each viral keyword in the adjusted set of viral keywords for new media enterprises. It also determines the follow-up content data, platform suggestion data for each historical content on each platform, and operational decision data for new media enterprises, enabling the production and dynamic correction of operational strategies. This avoids the blindness, lag, and high trial-and-error costs of strategies, achieving intelligent operation strategies from data mining to implementation, improving the market adaptability, accuracy, and iteration efficiency of strategies, and strengthening the core effectiveness of content dissemination and user conversion.

[0014] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the enterprise new media operation strategy generation and dynamic correction method based on a large model in an embodiment of the present invention. Detailed Implementation

[0017] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0018] Example 1: This invention provides a method for generating and dynamically correcting enterprise new media operation strategies based on a large model, referencing... Figure 1 ,include: S1: Obtain industry tags, enterprise user data, and industry hot topic data of new media companies; obtain enterprise content data and competitor data of new media companies within a specified time period. S2: Based on industry hot data and competitor data of new media companies, determine the set of hot keywords and the set of popular keywords for new media companies, and determine the adjusted set of hot keywords and the set of popular keywords for new media companies, as well as the key weight of each popular keyword in the adjusted set of popular keywords. S3: Based on the adjusted set of trending keywords, the set of popular keywords, the key weights of all popular keywords in the adjusted set of popular keywords, and enterprise content data, determine the follow-up content data for new media enterprises; S4: Analyze the follow-up content data of new media companies, determine the platform recommendation data for each platform of each historical content in the follow-up content data, and determine the operational decision data for each historical content in the follow-up content data, so as to realize the production and dynamic correction of the operational strategy of new media companies.

[0019] In this embodiment, industry tags, enterprise user data, and industry trending data of new media enterprises are acquired, along with enterprise content data and competitor data within a specified time period. Industry tags are classification identifiers that identify the industry sector to which the enterprise belongs, determined based on the enterprise's main business, and used to anchor the industry boundaries of operational strategies. Enterprise user data comprises comprehensive information on user profiles, behaviors, and value feedback reached by the enterprise, collected from the backend of the new media platform and user survey channels. Industry trending data represents the hot topics and events in the enterprise's industry, obtained through public opinion monitoring tools and industry information platforms. The specified time period is set by the enterprise based on its operational goals. Enterprise content data includes all historical content published by the enterprise within that period and its performance and conversion data, extracted from the enterprise's new media account backend. Competitor data comprises the operational content, hit product strategies, and competitor ranking data of competitors in the same industry during the same period, collected through competitor account monitoring and third-party data platforms.

[0020] In this embodiment, the hot keyword set is a collection of words reflecting the core popularity of the industry, extracted through semantic analysis of industry hot data. The best-selling keyword set is a collection of high-frequency core words extracted from best-selling content in competitor data. After semantic normalization to remove redundancy and merge synonyms, the adjusted best-selling keyword set is obtained. The key weight is a quantified importance value for each adjusted best-selling keyword, combining competitor level, keyword frequency, time popularity characteristics, and relevance to industry hot topics. The higher the weight, the stronger the driving force for creating a best-selling product.

[0021] In this embodiment, the adjusted keyword system and key weights clarify the core direction of popular market content. Based on this, historical content in the enterprise's content data is intelligently matched with this system, and its value is evaluated by combining the performance and conversion data of historical content. The follow-up content data is a collection of high-quality historical content that has been screened, aligns with market trends or the logic of viral hits, and has good operational performance, covering all dimensions of information such as basic content attributes and performance conversion data.

[0022] In this embodiment, platform suggestion data is generated by extracting user feedback on content suggestions from various platforms, performing cluster analysis, and quantifying the value of suggestions by combining user consumption value tags. This results in customized optimization suggestions for each historical content across different platforms, aligning with user needs and platform characteristics. Operational decision data is generated by inputting the basic data, performance data, and platform suggestion data of the followed content into a large-scale content decision model. This generates refined execution parameters including optimized content version, release time, decision-making method, and predicted performance. Dynamic correction refers to collecting real-time performance data after the followed content is implemented, comparing it with the predicted indicators of operational decisions, identifying deviations, and making targeted adjustments and optimizations to the content and decisions.

[0023] The beneficial effects of the above technologies are as follows: By acquiring industry tags, enterprise user data, industry hot topic data, enterprise content data within a specified time period, and competitor data of new media enterprises, the technologies can determine the set of hot keywords, the set of viral keywords, and the key weight of each viral keyword in the adjusted set of viral keywords. Furthermore, the technologies can determine the follow-up content data, platform suggestions for each historical content on each platform, and operational decision-making data of new media enterprises, enabling the production and dynamic correction of operational strategies. This avoids the blindness, lag, and high trial-and-error costs of strategies, achieving intelligent operation strategies from data mining to implementation, improving the market adaptability, accuracy, and iteration efficiency of strategies, and strengthening the core effectiveness of content dissemination and user conversion.

[0024] Example 2: Based on Example 1, a method for generating and dynamically correcting enterprise new media operation strategies based on a large model is proposed, which acquires industry tags, enterprise user data, and industry hot topic data for new media enterprises, including: Acquire industry tags and enterprise user data of new media companies. Enterprise user data includes user data of multiple users, and user data includes user tags, user base vectors, user behavior vectors, and consumer value tags. Consumer value tags include high value, medium value, low value, zero value, and negative value. Based on the industry tags of new media companies, we obtain industry hot topics data for new media companies.

[0025] In this embodiment, the industry label is a core classification identifier that identifies the industry field to which the enterprise belongs. It is determined based on the enterprise's main business scope, industry classification standards, or administrative filing information. It is used to clarify the industry boundaries for operational data collection and strategy formulation. The industry label can be beauty and skin care, clothing and footwear, food and beverage, 3C digital products, home appliances, maternal and infant products, education and training, etc.

[0026] In this embodiment, enterprise user data is a collection of relevant information for all users served or reached by the new media enterprise, including independent user data for multiple users. User tags are classification identifiers for users, determined through user registration information and interaction behavior analysis. User tags include: New Customers: registered within 7 days, or first purchase within 30 days; Growing Customers: completed their first order, made 1-2 repeat purchases, or registered 30-90 days ago; Mature Customers: made more than 3 repeat purchases, or registered for more than 90 days and are active; Declining Customers: inactive beyond the average repurchase cycle; Churned Customers: inactive for a long time and unresponsive to reactivation. User base vectors are feature vectors that quantify basic user attributes such as age, gender, and region, obtained through digital processing of basic user information. User behavior vectors are feature vectors that quantify user behaviors such as browsing, liking, forwarding, and purchasing on new media platforms, generated through statistical analysis of platform backend behavior data. Consumer value tags are used to classify and determine the value that users bring to a business. High value refers to users with high-frequency consumption and high average order value; medium value refers to users with stable consumption; low value refers to users with low-frequency and small-amount consumption; zero value refers to users with no consumption behavior but with interaction; and negative value refers to users with malicious complaints and negative dissemination. These tags are determined through user consumption records, interaction behavior, and negative impact assessment.

[0027] In this embodiment, industry hot topic data for new media companies is obtained based on their industry tags. These tags clearly define the industry sector to which the company belongs, and using this as a filtering criterion avoids collecting hot topic information irrelevant to the company. Industry hot topic data refers to recently widely discussed events, topics, trends, policies, and other related information within the company's industry. This data is collected through channels such as public opinion monitoring tools, industry information platforms, social media trending lists, and vertical media within the industry. The collection process focuses on content highly relevant to the industry tags to ensure the relevance of the hot topic data.

[0028] The beneficial effects of the above technologies are: acquiring industry tags, enterprise user data, and industry hot topic data of new media companies can build a comprehensive and accurate data foundation, providing precise user needs and market trend support for strategy generation.

[0029] Example 3: Based on Example 1, a method for generating and dynamically correcting enterprise new media operation strategies based on a large model is proposed, which acquires enterprise content data and competitor data for a specified time period, including: Acquire enterprise content data and competitor data for new media companies within a specified time period. Enterprise content data includes performance data for multiple historical content items, including content duration, content format tags, content performance vectors across multiple platforms, content conversion vectors, content keyword sets, and platform feedback data, including feedback content from multiple user IDs. Competitor data includes competitor level tags for multiple competitor companies and viral content data for multiple viral content items, including publication time, viral duration, content format tags, viral performance vectors across multiple platforms, viral conversion vectors, and a subset of viral keywords. Content format tags include text / images, videos, and live streaming.

[0030] In this embodiment, enterprise content data and competitor data are acquired from new media enterprises within a specified time period. The specified time period is set by the new media enterprise based on its operational goals, such as monthly, quarterly, or marketing campaign cycles, to focus on the effective data range and avoid interference from irrelevant data. Enterprise content data is a collection of relevant data corresponding to all historical content published by the enterprise within this period, including independent content performance data for multiple historical content items. Content duration is the playback or reading duration of a single historical piece of content; for video, it's the playback duration, and for text and image content, it's the estimated reading duration, obtained directly from the content publishing platform's backend. Content format tags are used to distinguish the presentation method of content, including text and image, video, and live streaming, determined based on the actual production and publishing format of the content. Content performance vectors across multiple platforms are feature vectors that quantify the content's performance on each publishing platform, covering metrics such as exposure, readership, likes, shares, and comments, generated through statistical analysis of data from each platform's backend. Content conversion vectors are feature vectors that quantify how content guides users to complete goals such as consumption, following, and registration, calculated based on conversion link data. The content keyword set is the core vocabulary extracted from the title and body tags of historical content, reflecting the core theme of the content. Platform feedback data is a collection of user feedback on historical content, containing feedback content corresponding to multiple user IDs. The user ID uniquely identifies the user providing feedback. Feedback content includes evaluations, suggestions, and questions, collected from various platform message areas, comment sections, customer service systems, and other channels. Competitor data is a collection of relevant data from multiple competing companies in the same industry as the new media company. Competitor level tags are classifications of competitors' market share, influence, and operational strength; for example, Level 1 corresponds to top competitors, Level 2 to mid-tier competitors, and Level 3 to bottom competitors. This is determined through market research using industry reports and third-party data platforms. Data on multiple viral content pieces is detailed data corresponding to outstanding content from competing companies. The release time is the specific moment the viral content was released, the viral duration is the duration the content maintained high popularity, and the content format tags are consistent with the company's content format tags. The viral performance vector across multiple platforms quantifies the characteristics of viral content's exposure and interaction on various platforms. The viral conversion vector quantifies the characteristics of viral content guiding user conversion. The viral keyword subset is the core vocabulary extracted from viral content. This data is collected through monitoring competitor accounts, third-party data tools, and industry analysis platforms.

[0031] The beneficial effects of the above technologies are: acquiring corporate content data and competitor data of new media companies within a specified time period, and comprehensively integrating the company's own content and multi-dimensional data of multiple competitor companies.

[0032] Example 4: Building upon Example 3, the method for generating and dynamically correcting enterprise new media operation strategies based on a large model determines the set of trending keywords and popular keywords for new media enterprises based on industry hot data and competitor data, including: Semantic analysis was conducted on industry hot topics data of new media companies to determine a set of hot keywords; Input the top-selling content performance vectors and top-selling conversion vectors of all platforms from all top-selling content data of all competitors into the content evaluation model to determine the value assessment value of each top-selling content of each competitor in the competitor data. Based on the subset of trending keywords in the trending content data of all competitor companies in the competitor data, determine the trending keyword set of the new media company and the first trending content set of each trending keyword in the trending keyword set, and count the occurrence frequency of each trending keyword in the trending keyword set. The first trending content set includes the trending content data of multiple trending content from multiple competitor companies.

[0033] In this embodiment, semantic analysis is performed on industry hot topic data of new media enterprises to determine a set of hot keywords. Industry hot topic data consists of information such as popular events and topics within the relevant industry, obtained in advance based on the enterprise's industry tags. Semantic analysis involves in-depth parsing of the text content of the hot topic data, extracting core semantic elements, identifying high-frequency related words, and removing irrelevant and redundant information. The set of hot keywords reflects the current core trends in the industry, with each word highly relevant to industry hot topics and possessing dissemination potential.

[0034] In this embodiment, the performance and conversion vectors of all popular content from all competing companies across all platforms are input into the content evaluation model to determine the value assessment value of each popular content from each competing company. The popular content data in the competitive data already includes the performance and conversion vectors of multiple competitors. The content evaluation model is an analytical model used to comprehensively evaluate content value, capable of integrating core features of performance and conversion dimensions for quantitative calculation. The value assessment value is a quantitative result of the comprehensive value of popular content; a higher value indicates better market performance and conversion effect.

[0035] In this embodiment, the content evaluation model first receives the platform-wide performance vectors and conversion vectors of all popular content from all competing companies in the competitor data. It then performs feature analysis on the two types of vectors and assigns appropriate weights to each feature in the performance and conversion vectors based on the value orientation of new media operations. The conversion dimension features have higher weights than the basic performance dimension features. Subsequently, the weighted features are comprehensively quantified and calculated, and the quantification results from each platform are integrated into an overall score. Finally, the value evaluation value of each popular content from each competing company is output. The higher the score, the stronger the comprehensive operational value of the popular content.

[0036] In this embodiment, firstly, subsets of trending keywords from all competitor companies' trending content are collected. These subsets are then integrated and deduplicated, removing words irrelevant to the company's industry tags, forming a set of trending keywords for the new media company. The first set of trending content is a collection of trending content data from all competitor trending content associated with each trending keyword, i.e., detailed data on all trending content containing that keyword. The frequency of occurrence is counted by counting the occurrence frequency of each trending keyword in all competitor trending keyword subsets; a higher frequency indicates a stronger trending-driving ability of that keyword.

[0037] The beneficial effects of the above technologies are as follows: Based on industry hot data and competitor data of new media companies, the set of hot keywords and popular keywords of new media companies can be determined, which can realize the accurate extraction of hot keywords, the quantification of the value of popular competitors, and the visualization of keyword popularity.

[0038] Example 5: Building upon Example 4, this method for generating and dynamically correcting enterprise new media operation strategies based on a large-scale model determines the adjusted set of trending keywords, the set of viral keywords, and the key weight of each viral keyword within the adjusted set of viral keywords for new media enterprises, including: Based on the publication time of the content data corresponding to all subsets of popular keywords in the first popular content set, which includes the popular keyword in the popular keyword set, the publication time range of each popular keyword in the popular keyword set is determined. Based on the release time range and frequency of each popular keyword in the popular keyword set, the specified time period is divided to determine multiple specified time sub-periods for each popular keyword in the popular keyword set and the second popular content set for each specified time sub-period. Semantic normalization is performed on all hot keywords in the hot keyword set and all popular keywords in the popular keyword set to determine the adjusted hot keyword set and popular keyword set. Multiple specified time sub-periods and the second popular content set for each popular keyword in the adjusted popular keyword set are also determined. Based on the adjusted set of trending keywords for new media companies, the competitor level tags of all competitors in the company's competitor data, the adjusted set of popular keywords, all specified time sub-periods of all popular keywords in the adjusted set of popular keywords, and the set of second popular content in each specified time sub-period, the key weight of each popular keyword in the adjusted set of popular keywords for new media companies is calculated.

[0039] In this embodiment, the first set of trending content consists of all competitor trending content data that contains the trending keyword and is associated with it in the early stages. The publication time is the specific publication moment of each trending content, which is directly extracted from the trending content data. All publication times corresponding to each trending keyword are analyzed, and the earliest and latest publication times are used as boundaries to form the publication time range of the keyword. This range reflects the duration of the keyword's popularity in the market.

[0040] In this embodiment, the specified time period is a pre-defined enterprise operation analysis cycle. The publication time interval reflects the sustained range of keyword popularity, and the frequency of occurrence reflects the keyword's popularity. Combining the length of the publication time interval and the distribution of the frequency of occurrence, the specified time period is divided into multiple consecutive specified time sub-periods. The duration of each sub-period can be flexibly set according to the fluctuation of keyword popularity and the length of the specified time period. The second set of trending content is the data set of all competitor trending content containing the trending keyword within each specified time sub-period, obtained by categorizing the first set of trending content by sub-period.

[0041] In this embodiment, semantic normalization involves semantic parsing of two types of keywords, merging synonyms and near-synonyms, correcting keywords with inconsistent expressions, and eliminating semantically redundant or irrelevant words to ensure the uniqueness and accuracy of the keywords. The adjusted set of hot keywords and the set of trending keywords are core vocabulary sets after semantic optimization, avoiding keyword redundancy or ambiguity. For the adjusted trending keywords, their specified time sub-periods and the second set of trending content directly use the information corresponding to them before semantic normalization, because semantic normalization only optimizes the keyword expression and does not change their associated time and content data.

[0042] In this embodiment, based on the adjusted set of trending keywords for new media companies, the competitor level tags of all competitors in the company's competitor data, the adjusted set of popular keywords, all specified time periods of all popular keywords in the adjusted set of popular keywords, and the set of second popular content in each specified time period, the key weight of each popular keyword in the adjusted set of popular keywords for new media companies is calculated. The calculation formula is expressed as follows: ; ; ; ; in, This indicates the key weight of the b-th keyword in the adjusted set of top-selling keywords. This represents the trend and popularity weight of the b-th keyword in the adjusted set of trending keywords. This represents the ranking evaluation weight of the b-th keyword in the adjusted set of top-selling keywords. These represent the trend hotspot weight adjustment factor and the grade assessment weight adjustment factor, respectively. This represents the adjusted set of trending keywords. This represents the b-th popular keyword in the adjusted set of popular keywords. This represents the competitor rating label value of the c-th competitor in the competitor data. This represents the number of trending content items in the second set of trending content items for the b-th trending keyword in the a-th specified time sub-period of the adjusted trending keyword set. This represents the subset of trending keywords for the d-th trending content from the c-th competitor in the competitor data, where bN4 represents the number of specified time periods for the b-th trending keyword in the adjusted set of trending keywords. The second indicator function represents the b-th popular keyword in the adjusted set of popular keywords, based on the adjusted set of trending keywords. This represents the d-th most popular content from the c-th competitor in the competitor data. This represents the set of second trending content related to the b-th trending keyword in the a-th specified time sub-period within the adjusted set of trending keywords. This represents the value assessment of the dth most popular content from the cth competitor in the competitor data.

[0043] The beneficial effects of the above technologies are: determining the adjusted set of hot keywords, the set of trending keywords, and the key weight of each trending keyword in the adjusted set of trending keywords for new media companies, realizing the time-sequential and precise weight allocation of keywords, and providing core guidance for subsequent content adjustments and operational strategies that are more in line with market dynamics and the competitive landscape.

[0044] Example 6: Building upon Example 5, the method for generating and dynamically correcting enterprise new media operation strategies based on a large model determines the follow-up content data for new media enterprises based on the adjusted set of trending keywords, the set of viral keywords, the key weights of all viral keywords in the adjusted set of viral keywords, and enterprise content data. This includes: The adjusted set of hot keywords, the set of popular keywords, the key weights of all popular keywords in the set of popular keywords, and the set of content keywords in the content performance data of all historical content in the enterprise content data are input into the content recognition model to determine the content tags of each historical content in the enterprise content data. The content tags include follow-up and pause. Input the content performance vectors and content conversion vectors of all platforms from the content performance data of all historical content in the enterprise content data into the content evaluation model to determine the value assessment value of each piece of historical content in the enterprise content data; The value tag of historical content corresponding to each value assessment value exceeding the preset threshold is identified as "follow-up," and the value tag of historical content corresponding to each value assessment value not exceeding the preset threshold is identified as "pause." Based on the content performance data of all historical content tagged as "follow-up" or "value tag as follow-up", the follow-up content data of new media enterprises is determined. The follow-up content data includes the content performance data of multiple historical content.

[0045] In this embodiment, the adjusted set of trending keywords and the set of best-selling keywords are core vocabulary sets optimized through semantic normalization in the early stages. Key weights reflect the importance of each best-selling keyword. The set of content keywords for enterprise content is core thematic vocabulary extracted from historical content. These data collectively constitute the core basis for content matching. The content recognition model is an intelligent model used to analyze the fit between content and keywords. It can comprehensively consider the semantic relevance of keywords and key weights to determine whether historical content aligns with market trends and competitor best-selling logic. Content tags include two categories: follow-up and pause. The follow-up tag indicates that the content conforms to trending trends and best-selling directions, making it worthwhile to continue optimizing and promoting. The pause tag indicates that the content is out of touch with market trends and investment should be stopped.

[0046] In this embodiment, the content recognition model first receives the adjusted set of hot keywords, the set of popular keywords, the key weights of all popular keywords, and the set of content keywords for the company's historical content. It first performs semantic matching between the set of content keywords for the company's historical content and the two sets of keywords to calculate semantic similarity. Then, it combines the key weights of popular keywords to perform weighted correction on the similarity. The higher the weight of a popular keyword, the greater its matching result has on the overall fit. Subsequently, it compares the weighted overall fit with a preset threshold. If the fit is higher than the threshold, the historical content is assigned a follow-up label; if the fit is lower than the threshold, it is assigned a pause label. Finally, it outputs the content label corresponding to each piece of historical content of the company.

[0047] In this embodiment, core data on the performance and conversion of historical content are first extracted, then input into the model for quantitative analysis, ultimately yielding an evaluation result reflecting the actual value of the content. The content evaluation model is an analytical model capable of integrating performance and conversion dimension data for comprehensive quantification. By analyzing and calculating the core features of the two types of vectors, it outputs a value assessment value for each piece of historical content. The value assessment value is a quantitative result of the comprehensive operational value of historical content; a higher value indicates stronger market performance and conversion capabilities.

[0048] In this embodiment, the content evaluation model first receives the platform content performance vectors and content conversion vectors of all historical content in the enterprise's content data. Following the feature analysis logic of competitor's hit content value evaluation, it reallocates appropriate weights to each feature based on the enterprise's own operational goals and industry operational benchmarks, aligning with the enterprise's own conversion needs and platform operational characteristics. Then, it performs quantitative calculations on the weighted features of each platform, integrates the quantitative results of each platform to obtain the overall quantitative score of the historical content, and finally outputs the value evaluation value of each piece of historical content of the enterprise. The higher the score, the stronger the actual operational value and development potential of the historical content.

[0049] In this embodiment, the preset threshold is a value judgment standard set by new media companies based on industry benchmarks for operational goals and their own resources, used to distinguish between high-value and low-value content. Value tags include two categories: "follow-up" and "pause." When the value assessment value of historical content exceeds the preset threshold, it indicates that its operational value meets the standard, and a "follow-up" tag is assigned; when the value assessment value does not reach the preset threshold, it indicates that its operational value is insufficient, and a "pause" tag is assigned.

[0050] In this embodiment, the content tag "follow-up" means the content aligns with market trends and popular trends, while the value tag "follow-up" means the content possesses good practical operational value. Only historical content that meets either of these two conditions has the value for continuous optimization and promotion. Follow-up content data is a collection of content performance data from these high-quality historical contents, providing a detailed content foundation for subsequently developing specific operational strategies.

[0051] The beneficial effects of the above technologies are as follows: Based on the adjusted set of hot keywords, the set of popular keywords, the key weights of all popular keywords in the adjusted set of popular keywords, and enterprise content data, the follow-up content data of new media enterprises can be determined. This can avoid inefficient or out-of-touch content investment, accurately target high-quality content that is both market-adaptable and of practical value, improve the efficiency of operational resource utilization, and strengthen the core effectiveness of content dissemination and conversion.

[0052] Example 7: Building upon Example 6, this method for generating and dynamically correcting enterprise new media operation strategies based on a large-scale model analyzes the follow-up content data of new media enterprises, determines the platform suggestion data for each platform for each historical content in the follow-up content data, and realizes the production and dynamic correction of new media enterprise operation strategies, including: Semantic recognition is performed on the feedback content of each user ID in the platform feedback data of each platform in the content performance data of each historical content in the follow-up content data, and the feedback content tags of each user ID in the platform feedback data of each platform in the content performance data of each historical content are determined. The feedback content tags include product experience and content suggestions. Extract all feedback content from all user IDs whose feedback content tag is "content suggestion" from the platform feedback data of each platform in the content performance data of each historical content in the follow-up content data, and determine the content suggestion data of each platform for each historical content in the follow-up content data; Cluster analysis is performed on all feedback content of all user IDs in the content suggestion data of each platform for each historical content in the follow-up content data to determine multiple suggestion cluster data for each platform for each historical content in the follow-up content data, as well as suggestion labels for each suggestion cluster data. The suggestion cluster data includes suggestion fitting vectors and all user IDs. Based on the consumption value tags of all users in the enterprise user data and the suggestion clustering data of all suggestion tags of each historical content on each platform in the follow-up content data, calculate the initial tag clustering value, the iterative tag clustering value of multiple iterations, and the actual tag clustering value of each suggestion tag of each historical content on each platform in the follow-up content data, and calculate the adopted tag of each suggestion tag of each historical content on each platform in the follow-up content data. Based on the adopted tags of all suggested tags for each platform of each historical content in the follow-up content data, and the suggested fitting vectors in the suggested clustering data, the platform suggested data for each platform of each historical content in the follow-up content data is determined.

[0053] In this embodiment, each historical content in the follow-up content data is first identified, and the platform feedback data corresponding to each platform in its content performance data is extracted. Then, the independent feedback content corresponding to each user ID is separated from the platform feedback data of each platform. Subsequently, semantic recognition is performed on the feedback content corresponding to each user ID. By analyzing the core semantics and expressive intent of the feedback content, the core type of feedback request is distinguished. Finally, the feedback content of each user ID is matched with the corresponding feedback content tag. The tags are divided into two categories: product experience and content suggestion. Product experience is the feedback from users about their experience using the company's products or services. Content suggestion is the optimization and improvement suggestions put forward by users for the new media content itself. Content request is the new demands and expectations expressed by users for the new media content. In this way, the tag determination of all user ID feedback content in the platform feedback data of each platform in the content performance data of each historical content is completed.

[0054] In this embodiment, the follow-up content data is a collection of high-quality historical content data tagged as "follow-up" selected in the early stages. The platform feedback data consists of user feedback information on historical content from various platforms. Feedback content tags are classification identifiers for user feedback types. Content suggestions are feedback types pointing to content optimization. User IDs are unique identifiers of the users who provided feedback. First, the content performance data of each historical content in the follow-up content is located. Then, the platform feedback data corresponding to each platform is extracted. From the platform feedback data, the portion of feedback content tagged as "content suggestion" is selected. All feedback content corresponding to all user IDs under this portion is extracted. After integrating and sorting these feedback contents, content suggestion data for each historical content on each platform is formed.

[0055] In this embodiment, cluster analysis is a data analysis method that groups user feedback content with similar semantics and consistent demands into the same category. The suggested cluster data is a collection of suggested feedback of the same category formed after clustering. The suggested tag is a general identifier of the core optimization demand for each category of suggested cluster data. The suggested fitting vector is a feature vector that quantifies the core characteristics and demand tendency of each category of suggestions. All user IDs are the identification information of all users who submitted suggestions of that category. First, content suggestion data for each historical content is obtained on each platform. Then, cluster analysis is performed on all feedback content of all user IDs in the data. Feedback with similar semantics and optimization demands is grouped into multiple independent categories. Corresponding suggested cluster data is generated for each category. Simultaneously, each category of suggested cluster data is assigned a suggested tag that reflects the core demand, and the suggested fitting vector that quantifies the core characteristics and all user IDs corresponding to that category of suggestions are included in the suggested cluster data.

[0056] In this embodiment, based on the consumption value tags of all users in the enterprise user data and the suggestion clustering data of all suggestion tags for each platform of each historical content in the follow-up content data, the initial tag clustering value, the iterative tag clustering value after multiple iterations, and the actual tag clustering value of each suggestion tag for each platform of each historical content in the follow-up content data are calculated. Furthermore, the adopted tag for each suggestion tag for each platform of each historical content in the follow-up content data is calculated. The calculation formula is expressed as follows: ; ; ; ; ; ; ; ; ; ; in, This represents the user ID of the nth user in the enterprise user data. This represents the consumption value tag of the nth user in the enterprise user data. This represents the user ID of the m-th and p-th users in the suggestion clustering data of the j-th platform and the k-th suggestion tag of the i-th historical content in the enterprise content data. , This represents the suggestion fit vector in the suggestion clustering data of the i-th historical content from the j-th platform and the k-th suggestion tag, and the q-th suggestion tag. This represents the user ID of the q-th user in the suggestion clustering data of the j-th platform and the k-th suggestion tag of the i-th historical content in the enterprise content data. The first indicator function represents the user ID of the nth user in the enterprise user data and the user ID of the mth user in the suggestion clustering data of the jth platform and kth suggestion tag of the ith historical content in the enterprise content data. Let ijkN1 represent the value tag mapping value of the nth user in the enterprise user data, and let ijkN1 represent the number of user IDs in the suggestion clustering data of the jth platform and kth suggestion tag of the i-th historical content in the enterprise content data. This represents the frequency of occurrence of the user ID of the m-th user in the clustered data of the suggestion tag of the j-th platform for the i-th historical content in the enterprise content data. This represents the conflict value of the suggestion fit vector in the suggestion clustering data of the i-th historical content from the j-th platform and the k-th suggestion tag from the q-th suggestion tag. This represents the frequency sub-weight of the user ID of the m-th user in the cluster data of the suggestion tag of the j-th platform for the i-th historical content in the enterprise content data. This represents the frequency weight of the user ID of the m-th user in the cluster data of the suggestion tag of the j-th platform for the i-th historical content in the enterprise content data. This represents the number of user IDs in the cluster data of the suggestion tags of the i-th historical content on the j-th platform in the enterprise content data. Let represent the initial tag clustering value of the k-th suggested tag of the i-th historical content on the j-th platform in the enterprise content data, and let ijN3 represent the number of suggested clustering data of the i-th historical content on the j-th platform in the enterprise content data. Let represent the iterative tag clustering values ​​of the j-th platform and the k-th suggested tag of the i-th historical content in the enterprise content data, respectively, at the t-th and t-1-th iterations. This represents the initial tag clustering value of the j-th suggested tag of the i-th historical content in the enterprise content data, on the j-th platform. This represents the actual tag clustering value of the k-th suggested tag on the j-th platform for the i-th historical content in the enterprise content data. This represents the adopted tag for the k-th suggested tag of the j-th platform for the i-th historical content in the enterprise's content data.

[0057] This represents the set of user IDs for the suggestion clustering data of the i-th historical content from the j-th platform with the q-th suggestion tag in the enterprise content data.

[0058] In this embodiment, the adoption tag determines whether the optimization suggestion corresponding to the suggestion tag is adopted, which is the core screening basis for content optimization. The suggestion fitting vector quantifies the core characteristics, optimization needs and tendencies of the adopted suggestions. The platform suggestion data is a set of specific and implementable content optimization suggestions for each historical content on each platform, which is in line with the core needs of the platform's users. First, the adoption tags corresponding to all suggestion tags of each historical content in the follow-up content on each platform are extracted, and the suggestion tags that are adopted are filtered out. Then, the suggestion fitting vectors in the suggestion clustering data corresponding to these adopted suggestion tags are extracted. Combining the core optimization needs and characteristics reflected by the suggestion fitting vectors, specific optimization suggestions that meet the needs of the platform's users are integrated and sorted out. These suggestions are integrated to form the platform suggestion data for each historical content on each platform.

[0059] The beneficial effects of the above technologies are as follows: analyzing the follow-up content data of new media enterprises, determining the platform suggestion data for each platform for each historical content in the follow-up content data, realizing the production and dynamic correction of the operation strategy of new media enterprises, realizing the value stratification and precise decomposition of user feedback, making platform suggestions fit the core needs of users on each platform and the preferences of high-value users of enterprises, improving the pertinence and implementation of content optimization, and strengthening the dynamic adaptability of operation strategies.

[0060] Example 8: Building upon Example 7, the method for generating and dynamically correcting enterprise new media operation strategies based on a large model determines the operational decision data for each historical content in the follow-up content data, thereby realizing the production and dynamic correction of new media enterprise operation strategies, including: The content duration, content format tags, content performance vectors, content conversion vectors, and platform suggestion data of each historical content in the follow-up content data are input into the content decision model to generate real-time content data and operational decision data for each historical content in the follow-up content data. The operational decision data includes at least the decision duration, release time, decision format tags, predicted performance vectors, and predicted conversion vectors for multiple platforms. Collect real-time performance data for each piece of content, and dynamically correct the content based on the real-time performance data and operational decision data. The real-time performance data includes real-time performance vectors, real-time conversion vectors, and real-time feedback data from multiple platforms.

[0061] In this embodiment, basic attribute data, multi-platform performance conversion data, and pre-established platform suggestion data for each historical content are first extracted from the follow-up content data. This multi-dimensional data is then integrated and input into the content decision model. The model performs comprehensive analysis and optimization deduction on the data, ultimately generating real-time content data adapted to each platform. Simultaneously, it outputs operational decision data containing multiple core execution parameters, providing specific and refined execution basis for the release and operation of real-time content, connecting content optimization with actual operational execution. The content decision model is an intelligent analysis model used to integrate multi-dimensional content data and generate optimized content and operational decisions. Real-time content data is the optimized version of content data that can be directly released after model analysis and optimization. Operational decision data consists of specific decision parameters guiding the release and operation of real-time content. The decision duration is the optimized content duration suggested by the model; the release time is the suggested release time given by the model based on platform algorithms and user activity patterns; the decision format label is the content format suggestion optimized by the model according to platform characteristics; the predicted performance vector is the quantitative value of the real-time content's predicted operational performance on each platform; and the predicted conversion vector is the quantitative value of the real-time content's predicted user conversion effect.

[0062] In this embodiment, the content decision model first receives the content duration, content format tags, content performance vectors across all platforms, content conversion vectors, and platform suggestion data for each historical content in the follow-up content data. It preprocesses the multi-dimensional input data, unifying feature dimensions and quantification standards, and eliminating invalid and redundant data. Then, it integrates the content performance vectors and conversion vectors from each platform to analyze the operational shortcomings of historical content on different platforms. Combining this with the platform suggestion data regarding the specific optimization requirements of each platform, it adapts the original content duration of the historical content to generate a decision duration that fits the platform's characteristics. Simultaneously, based on platform content format preferences and optimization suggestions, it optimizes the original content format tags to obtain decision format tags, integrating the aforementioned optimization dimensions to form... The system adapts to real-time content data published on various platforms. Then, based on the algorithm rules and user activity patterns of each platform, combined with historical content performance data, it matches the optimal publication time for each platform to the real-time content data. Subsequently, relying on the training data of a large model and industry operational patterns, combined with the optimization characteristics of real-time content data and the operational attributes of each platform, it quantitatively predicts the exposure and interaction performance of real-time content on each platform, generating predictive performance vectors. It also quantitatively predicts user consumption and engagement conversion effects, generating predictive conversion vectors. Finally, it integrates decision duration, publication time, decision form tags, multi-platform predictive performance vectors, and predictive conversion vectors into operational decision data, simultaneously outputting both real-time content data and operational decision data.

[0063] In this embodiment, the real-time content data generated by the content decision model is first published on various platforms according to the parameter requirements of the operational decision data. After publication, real-time performance data of each platform is continuously collected. Then, a comprehensive comparative analysis is conducted between the actual quantitative indicators in the real-time performance data and the predicted indicators such as the predicted performance vector and predicted conversion vector in the operational decision data to identify the deviation between the actual performance and the predicted results and the reasons for the deviation. Finally, based on the deviation analysis results, the real-time content data is optimized at the content level, or the parameters such as the publication time and decision duration in the operational decision data are adjusted to achieve dynamic correction of content and operational strategies, so that the operational strategies always conform to the actual platform operation effect and the immediate needs of users. Real-time content data refers to the content data to be published after optimization by the content decision model. Real-time performance data refers to the actual operational effect data collected after the real-time content is published on various platforms. Real-time performance vector is a feature vector that quantifies the actual exposure, readership, and interaction of real-time content on various platforms. Real-time conversion vector is a feature vector that quantifies the actual conversion effect of real-time content on various platforms, such as guiding users to consume, pay attention, and register. Real-time feedback data refers to the immediate feedback information such as evaluations, suggestions, and questions from users on various platforms regarding the published real-time content. Dynamic correction refers to the targeted adjustment and optimization of real-time content data or operational decision data based on the comparison results between the actual operational performance of real-time content and the predicted indicators in the operational decision data.

[0064] The beneficial effects of the above technologies are as follows: identifying and tracking operational decision data for each historical content in the content data, enabling new media companies to produce and dynamically correct operational strategies, avoiding lag in operational decisions and disconnect between actual operational results, ensuring that decisions align with platform characteristics and user demands, achieving precise iterative optimization of strategies, improving the scientific nature, implementability, and dynamic adaptability of operational decisions, and strengthening the effectiveness of content operations and market responsiveness.

[0065] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for generating and dynamically correcting enterprise new media operation strategies based on a large model, characterized in that, include: S1: Obtain industry tags, enterprise user data, and industry hot topic data of new media companies; obtain enterprise content data and competitor data of new media companies within a specified time period. S2: Based on industry hot data and competitor data of new media companies, determine the set of hot keywords and the set of popular keywords for new media companies, and determine the adjusted set of hot keywords and the set of popular keywords for new media companies, as well as the key weight of each popular keyword in the adjusted set of popular keywords. S3: Based on the adjusted set of trending keywords, the set of popular keywords, the key weights of all popular keywords in the adjusted set of popular keywords, and enterprise content data, determine the follow-up content data for new media enterprises; S4: Analyze the follow-up content data of new media companies, determine the platform recommendation data for each platform of each historical content in the follow-up content data, and determine the operational decision data for each historical content in the follow-up content data, so as to realize the production and dynamic correction of the operational strategy of new media companies.

2. The method for generating and dynamically correcting enterprise new media operation strategies based on a large model according to claim 1, characterized in that, Obtain industry tags, enterprise user data, and industry trend data for new media companies, including: Acquire industry tags and enterprise user data of new media companies. Enterprise user data includes user data of multiple users, and user data includes user tags, user base vectors, user behavior vectors, and consumer value tags. Consumer value tags include high value, medium value, low value, zero value, and negative value. Based on the industry tags of new media companies, we obtain industry hot topics data for new media companies.

3. The method for generating and dynamically correcting enterprise new media operation strategies based on a large model according to claim 1, characterized in that, Obtain content data and competitor data of new media companies within a specified time period, including: Acquire enterprise content data and competitor data for new media companies within a specified time period. Enterprise content data includes performance data for multiple historical content items, including content duration, content format tags, content performance vectors across multiple platforms, content conversion vectors, content keyword sets, and platform feedback data, including feedback content from multiple user IDs. Competitor data includes competitor level tags for multiple competitor companies and viral content data for multiple viral content items, including publication time, viral duration, content format tags, viral performance vectors across multiple platforms, viral conversion vectors, and a subset of viral keywords. Content format tags include text / images, videos, and live streaming.

4. The method for generating and dynamically correcting enterprise new media operation strategies based on a large model according to claim 3, characterized in that, Based on industry trend data and competitor data of new media companies, we determine the set of trending keywords and the set of viral keywords for new media companies, including: Semantic analysis was conducted on industry hot topics data of new media companies to determine a set of hot keywords; Input the top-selling content performance vectors and top-selling conversion vectors of all platforms from all top-selling content data of all competitors into the content evaluation model to determine the value assessment value of each top-selling content of each competitor in the competitor data. Based on the subset of trending keywords in the trending content data of all competitor companies in the competitor data, determine the trending keyword set of the new media company and the first trending content set of each trending keyword in the trending keyword set, and count the occurrence frequency of each trending keyword in the trending keyword set. The first trending content set includes the trending content data of multiple trending content from multiple competitor companies.

5. The method for generating and dynamically correcting enterprise new media operation strategies based on a large model according to claim 4, characterized in that, Determine the adjusted set of trending keywords, the set of viral keywords, and the key weight of each viral keyword within the adjusted set of viral keywords for new media companies, including: Based on the publication time of the content data corresponding to all subsets of popular keywords in the first popular content set, which includes the popular keyword in the popular keyword set, the publication time range of each popular keyword in the popular keyword set is determined. Based on the release time range and frequency of each popular keyword in the popular keyword set, the specified time period is divided to determine multiple specified time sub-periods for each popular keyword in the popular keyword set and the second popular content set for each specified time sub-period. Semantic normalization is performed on all hot keywords in the hot keyword set and all popular keywords in the popular keyword set to determine the adjusted hot keyword set and popular keyword set. Multiple specified time sub-periods and the second popular content set for each popular keyword in the adjusted popular keyword set are also determined. Based on the adjusted set of trending keywords for new media companies, the competitor level tags of all competitors in the company's competitor data, the adjusted set of popular keywords, all specified time sub-periods of all popular keywords in the adjusted set of popular keywords, and the set of second popular content in each specified time sub-period, the key weight of each popular keyword in the adjusted set of popular keywords for new media companies is calculated.

6. The method for generating and dynamically correcting enterprise new media operation strategies based on a large model according to claim 5, characterized in that, Based on the adjusted set of trending keywords, the set of popular keywords, the key weights of all popular keywords in the adjusted set of popular keywords, and enterprise content data, the follow-up content data for new media enterprises is determined, including: The adjusted set of hot keywords, the set of popular keywords, the key weights of all popular keywords in the set of popular keywords, and the set of content keywords in the content performance data of all historical content in the enterprise content data are input into the content recognition model to determine the content tags of each historical content in the enterprise content data. The content tags include follow-up and pause. Input the content performance vectors and content conversion vectors of all platforms from the content performance data of all historical content in the enterprise content data into the content evaluation model to determine the value assessment value of each piece of historical content in the enterprise content data; The value tag of historical content corresponding to each value assessment value exceeding the preset threshold is identified as "follow-up," and the value tag of historical content corresponding to each value assessment value not exceeding the preset threshold is identified as "pause." Based on the content performance data of all historical content tagged as "follow-up" or "value tag as follow-up", the follow-up content data of new media enterprises is determined. The follow-up content data includes the content performance data of multiple historical content.

7. The method for generating and dynamically correcting enterprise new media operation strategies based on a large model according to claim 6, characterized in that, Analyze the follow-up content data of new media companies to determine the platform recommendation data for each historical piece of content on each platform, enabling the production and dynamic adjustment of new media companies' operational strategies, including: Semantic recognition is performed on the feedback content of each user ID in the platform feedback data of each platform in the content performance data of each historical content in the follow-up content data, and the feedback content tags of each user ID in the platform feedback data of each platform in the content performance data of each historical content are determined. The feedback content tags include product experience and content suggestions. Extract all feedback content from all user IDs whose feedback content tag is "content suggestion" from the platform feedback data of each platform in the content performance data of each historical content in the follow-up content data, and determine the content suggestion data of each platform for each historical content in the follow-up content data; Cluster analysis is performed on all feedback content of all user IDs in the content suggestion data of each platform for each historical content in the follow-up content data to determine multiple suggestion cluster data for each platform for each historical content in the follow-up content data, as well as suggestion labels for each suggestion cluster data. The suggestion cluster data includes suggestion fitting vectors and all user IDs. Based on the consumption value tags of all users in the enterprise user data and the suggestion clustering data of all suggestion tags of each historical content on each platform in the follow-up content data, calculate the initial tag clustering value, the iterative tag clustering value of multiple iterations, and the actual tag clustering value of each suggestion tag of each historical content on each platform in the follow-up content data, and calculate the adopted tag of each suggestion tag of each historical content on each platform in the follow-up content data. Based on the adopted tags of all suggested tags for each platform of each historical content in the follow-up content data, and the suggested fitting vectors in the suggested clustering data, the platform suggested data for each platform of each historical content in the follow-up content data is determined.

8. The method for generating and dynamically correcting enterprise new media operation strategies based on a large model according to claim 7, characterized in that, Identify and track operational decision-making data for each historical piece of content within the tracking data to enable new media companies to produce and dynamically adjust their operational strategies, including: The content duration, content format tags, content performance vectors, content conversion vectors, and platform suggestion data of each historical content in the follow-up content data are input into the content decision model to generate real-time content data and operational decision data for each historical content in the follow-up content data. The operational decision data includes at least the decision duration, release time, decision format tags, predicted performance vectors, and predicted conversion vectors for multiple platforms. Collect real-time performance data for each piece of content, and dynamically correct the content based on the real-time performance data and operational decision data. The real-time performance data includes real-time performance vectors, real-time conversion vectors, and real-time feedback data from multiple platforms.

Images