Search

CN-121996824-A - Dynamic optimization method of intelligent content generation workshop platform

CN121996824ACN 121996824 ACN121996824 ACN 121996824ACN-121996824-A

Abstract

The invention relates to a dynamic optimization method of an intelligent content generation workshop platform, and belongs to the technical field of artificial intelligence and content generation. The method comprises the steps of obtaining user demands, constructing demand feature vectors, matching with functional feature matrixes of a pre-training resource cluster, forming target generation resource combinations, distributing adaptive computing power and material resource packages, starting a multi-form content collaborative generation mechanism based on the resource combinations, constructing a content multi-dimensional degree, generating initial content products, constructing a user evaluation scene operation industry standard three-dimensional feedback acquisition system, obtaining feedback data, extracting generation effect evaluation features, calculating, adjusting and generating system core configuration parameters based on feature weights, updating to a dynamic generation framework, and forming an iterated content generation system. Accurate matching of the generated resources and the requirements is realized, the generation flow is dynamically optimized through multidimensional feedback, and the content generation quality and the platform self-adaptive capacity are improved.

Inventors

  • JIN CHONGYING
  • LIU SHUHUA
  • Heng jing

Assignees

  • 上海数熙科技有限公司

Dates

Publication Date
20260508
Application Date
20260115

Claims (12)

  1. 1. The dynamic optimization method of the intelligent content generation workshop platform is characterized by comprising the following steps of: The method comprises the steps of S1, obtaining content generation requirements input by a user, and constructing a requirement feature vector, performing fitness calculation through the requirement feature vector and a functional feature matrix of a pre-training resource cluster to form a target generation resource combination, simultaneously matching material resource packages of corresponding scenes, distributing adaptation calculation force, and performing accurate matching of generation resources and requirements; S2, starting a multi-form content collaborative generation mechanism based on the generated resource combination, inputting the demand feature vector and the material resource package into a generation flow, and constructing the multi-dimension of the content; S3, constructing a user evaluation-scene operation-industry standard three-dimensional feedback acquisition system, acquiring multi-dimensional feedback data corresponding to the initial content product, performing standardized processing operation on the feedback data, extracting generation effect evaluation characteristics, and constructing a three-dimensional feedback driven generation effect evaluation characteristic set; And S4, based on the generating effect evaluation feature set, calculating each feature weight through feature weight calculation logic, adjusting the core configuration parameters of the generating system, updating the optimized generating resource configuration parameters and the collaborative operation strategy to the dynamic generating framework of the platform, and forming an iterated content generating system.
  2. 2. The method according to claim 1, wherein the constructing the demand feature vector is performed by: The method comprises the steps of executing semantic analysis on a demand text input by a user, screening and extracting demand core elements through word segmentation and core elements, calling a preset scene feature library, matching the demand core elements with scene information in the library to obtain scene associated elements of corresponding application scenes, respectively implementing quantitative coding on the demand core elements and the scene associated elements, converting text elements into vectors through word embedding, carrying out normalization processing on numerical value elements, distributing weights based on element importance, and carrying out weighted fusion on the coded element vectors to generate demand feature vectors.
  3. 3. The method according to claim 1, wherein the forming the target generation resource combination is performed by: the method comprises the steps of constructing a functional feature matrix of a pre-training resource cluster, wherein the functional feature matrix comprises the types, the scene adaptation ranges, the quality output capacity and the power calculation demand characteristics of all pre-training algorithm logics in the cluster, acquiring the core weight duty ratio of demand feature vectors based on content subjects in core demand dimensions and application scenes, calculating the adaptation degree of all the pre-training algorithm logic feature vectors in the demand feature vectors and the functional feature matrix, screening the pre-training algorithm logics with the adaptation degree higher than a preset threshold, and carrying out collaborative compatibility check on the screened pre-training algorithm logics to generate a conflict-free target generation resource combination.
  4. 4. The method according to claim 1, wherein the performing the accurate matching of the generated resource and the demand is as follows: Based on the content theme, quality threshold and application scene of the core demand dimension, performing adaptation consistency verification on the target generated resource combination formed by screening, correcting the resource combination parameter, correlating the application scene, output form and audience attribute of the core demand dimension, screening a material resource package containing a scene exclusive element, a format adaptation and a style adaptation from a preset material database, combining the real-time calculation power consumption, material processing complexity and content generation timeliness requirement of the target generated resource combination, dynamically distributing adaptation calculation power and reserving elastic redundancy, performing cross adaptation verification on the target generated resource combination, the material resource package and the adaptation calculation power, and generating a resource matching result meeting the operation stability requirement through a resource collaborative simulation test.
  5. 5. The method of claim 1, wherein the polymorphic content collaborative generation mechanism comprises text generation logic, visual construction logic, logic verification logic and preset multidimensional content integration rules, and wherein each logic module performs collaborative linkage through a data interaction interface to perform related operations of polymorphic content generation and integration.
  6. 6. The method according to claim 1, wherein the multi-dimensional construction of the content is performed by: The method comprises the steps of analyzing theme guide, style standard and scene adaptation requirements of content construction based on demand feature vectors, classifying and disassembling material resource packages according to text basic materials and visual basic materials, starting text generation logic and visual construction logic to operate in parallel, carrying out structural construction of a core document according to the theme guide and style standard by the text generation logic, generating matched visual elements by combining the visual construction logic with the scene adaptation requirements and document core information, and synchronously carrying out core information sharing of the two types of generation logic through a data interaction interface.
  7. 7. The method according to claim 1, wherein the forming of the initial content product is performed by: Invoking a preset multidimensional content integration rule, integrating the text and the visual elements based on the application scene format requirement and the style standard, and executing logic conflict check and compliance check on the integrated content to generate an initial content product containing the text, the visual elements and the scene adaptation format.
  8. 8. The method according to claim 1, wherein the construction of the user evaluation-scene operation-industry standard three-dimensional feedback acquisition system comprises the following specific steps: The method comprises the steps of configuring a user evaluation acquisition link, setting a platform interaction interface triggering rule, triggering an acquisition program after the user submits the evaluation, acquiring satisfaction degree scores and text suggestions, deploying a scene operation program in a content delivery channel, presetting a timing acquisition period of transmission data and interaction data, simultaneously constructing an industry standard acquisition link, accessing an industry compliance checking interface and a quality evaluation system, unifying three types of link data field formats, and establishing a data convergence channel.
  9. 9. The method according to claim 1, wherein the extracting generates an effect evaluation feature, specifically comprising: The method comprises the steps of acquiring multidimensional feedback data acquired by a three-dimensional feedback acquisition system, performing abnormal elimination and standardization processing to generate a standardized data set, performing feature primary extraction on the standardized data set by adopting a feature extraction algorithm to obtain an original feature set, screening the original feature set to generate an associated feature subset based on three feedback feature dimensions of preset requirement matching degree, quality standard reaching rate and scene adaptation deviation, and performing normalization integration on the feature subset to output generated effect evaluation features.
  10. 10. The method according to claim 1, wherein the calculating the feature weights by the feature weight calculating logic is: The method comprises the steps of obtaining a three-dimensional feedback driven generation effect evaluation feature set, carrying out weight assignment on each feature by adopting a hierarchical analysis method to generate initial weights, constructing an optimized objective function based on the initial weights, taking the opposite number of the sum of products of each feature and the corresponding initial weights as a core expression of the objective function, and carrying out iteration solving on the minimum value of the objective function through a gradient descent algorithm to output dynamic optimal weights of each feature.
  11. 11. The method according to claim 1, wherein the adjusting and generating system core configuration parameters comprises the following specific steps: The method comprises the steps of obtaining dynamic optimal weights of all features and generating effect evaluation features, adjusting configuration dimensions corresponding to features with highest weights, optimizing semantic extraction dictionary and feature vector construction parameters based on the required matching degree dimensions, updating matching threshold values, adjusting iteration steps and loss function parameters of a pre-training algorithm based on quality standard rate dimensions, supplementing special training data, updating material classification labels and matching algorithm based on scene adaptation deviation dimensions, and adjusting and generating scene adaptation parameters in a cooperative mode.
  12. 12. The method according to claim 1, wherein the forming the iterative content generation system comprises the following specific steps: Summarizing the core configuration parameters of the adjusted generation system, executing collaborative compatibility verification to generate an optimized parameter set, importing the optimized parameter set into a platform dynamic generation framework based on a preset format to update and cover the optimized parameter set, generating parameter update report record adjustment dimensionality, change value and corresponding feedback characteristic basis, and deploying the updated dynamic generation framework as an iterated content generation system to a platform content generation flow.

Description

Dynamic optimization method of intelligent content generation workshop platform Technical Field The invention belongs to the technical field of artificial intelligence and content generation, and particularly relates to a dynamic optimization method of an intelligent content generation workshop platform. Background In the prior art, an intelligent content generation platform generally adopts a preset and linear flow to respond to the user demand. The platform firstly analyzes a text instruction input by a user, then invokes one or more preset content generation models to perform authoring, and finally outputs single text or visual content. The resource allocation in the whole process is usually static, namely, computing resources are allocated in advance according to model types, and dynamic adjustment according to real-time demand complexity and scene characteristics is lacked. After the content is generated, the effect evaluation is mostly dependent on subjective scores submitted by users or simple click rate and other single indexes, and the feedback data are scattered and single in dimension and are difficult to systematically guide the optimization and evolution of the generation flow. The more prominent problem is that each link in the generating process is relatively isolated. There is a lack of effective synergy and feedback closed loop between demand analysis, resource matching, content construction and effect assessment. This makes it difficult for the platform to precisely understand the composite content requirements (e.g., such as text, graphics and style-specific typesetting), and it is impossible to achieve deep collaborative generation and integration of polymorphic content elements such as text, vision, etc. Meanwhile, due to the lack of a multidimensional feedback system integrating subjective evaluation of users, actual operation data of content and industry objective standards, optimization adjustment of a generated model often has blindness, and accurate iteration cannot be performed aiming at suitability deviation or quality short plates of specific application scenes. Algorithms and materials in a resource library are also always in a static state, and a matching strategy cannot be dynamically updated along with feedback, so that the self-adaption capability of a platform is insufficient when the platform faces an emerging scene or specialized requirements, and the quality, adaption degree and efficiency of a generated product are bottleneck. Thus, there is a need for a dynamic optimization method that breaks the dead office. The method can realize the systematic self-iterative full-link closed loop from demand understanding, resource intelligent matching and polymorphic content collaborative generation to multi-source feedback driving. The method has the core aims that the content generation platform can flexibly schedule resources like a workshop and continuously learn and evolve based on multidimensional feedback like an organism, so that the accuracy, the fusion and the quality reliability of content generation and the self-adaptive capacity of the platform to complex and variable demands are remarkably improved. Disclosure of Invention In order to solve the problems in the prior art, the invention provides a dynamic optimization method of an intelligent content generation workshop platform, and the aim of the invention can be realized by the following technical scheme: a dynamic optimization method of an intelligent content generation workshop platform comprises the following steps: The method comprises the steps of S1, obtaining content generation requirements input by a user, and constructing a requirement feature vector, performing fitness calculation through the requirement feature vector and a functional feature matrix of a pre-training resource cluster to form a target generation resource combination, simultaneously matching material resource packages of corresponding scenes, distributing adaptation calculation force, and performing accurate matching of generation resources and requirements; S2, starting a multi-form content collaborative generation mechanism based on the generated resource combination, inputting the demand feature vector and the material resource package into a generation flow, and constructing the multi-dimension of the content; S3, constructing a user evaluation-scene operation-industry standard three-dimensional feedback acquisition system, acquiring multi-dimensional feedback data corresponding to the initial content product, performing standardized processing operation on the feedback data, extracting generation effect evaluation characteristics, and constructing a three-dimensional feedback driven generation effect evaluation characteristic set; And S4, based on the generating effect evaluation feature set, calculating each feature weight through feature weight calculation logic, adjusting the core configuration parameters of the generating system, u