Custom cataloguing template generation method and system

Abstract

The invention discloses a user-defined cataloguing template generation method and system, and the method comprises the steps: obtaining a program feature description, obtaining a business demand description, and carrying out the program feature description and semantic coding of the program feature description through employing a natural language processing and analysis technology based on deep learning, thereby obtaining a user-defined cataloguing template. Therefore, the definition coding template is intelligently generated according to the dynamic memory interaction representation between the program characteristic description semantic features and the business requirement description semantic features. Through the mode, the most suitable cataloguing template can be automatically generated according to the specific demand and demand feedback of each user, and personalized services are provided. And user feedback can be considered in real time, and the cataloguing template is automatically updated according to latest data and changed contents, so that the cataloguing template is ensured to always fit an actual service scene.

Description

Technical Field

[0001] The present application relates to the field of intelligent generation, and more specifically, to a method and system for generating a custom cataloging template. Background Art

[0002] In the media and entertainment industry, program content is becoming increasingly rich and diverse. From traditional TV broadcasts to emerging online video platforms, the speed of content production and consumption is accelerating. At the same time, enterprises and organizations are increasingly in need of efficient management of massive program resources. An effective cataloging system can help users quickly locate the content they need and improve user experience. It can also help enterprises better analyze and utilize their own content libraries to create more business value.

[0003] However, traditional cataloging relies on a predefined classification system. Although it ensures consistency, it has poor adaptability to emerging content forms (such as short videos and live broadcasts), takes a long time to adjust rules, and is difficult to respond quickly to content diversity. In addition, traditional cataloging systems tend to use unified templates, lack flexibility, and cannot meet the personalized needs of users. This standardized approach ignores users' specific business scenarios and feedback, resulting in a gap between the generated cataloging templates and actual needs. At the same time, the traditional system is slow to update and cannot reflect changes in program content in a timely manner, limiting its applicability and responsiveness in a dynamic environment.

[0004] Therefore, a customized cataloging template generation scheme is desired, which can provide a more intelligent, flexible and efficient cataloging solution to better serve the needs of modern media management and content applications. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a method and system for generating a custom cataloging template, which obtains the program feature description and the business demand description, and uses natural language processing and analysis technology based on deep learning to perform semantic coding of the program feature description and the program feature description, so as to intelligently generate a definition coding template based on the dynamic memory interaction representation between the semantic features of the program feature description and the semantic features of the business demand description. In this way, the most suitable cataloging template can be automatically generated according to the specific needs and demand feedback of each user, and personalized services can be provided. It can also consider user feedback in real time, and automatically update the cataloging template according to the latest data and changed content to ensure that it always fits the actual business scenario.

[0006] According to one aspect of the present application, a method for generating a custom catalog template is provided, which includes: Get the program feature description; Obtain business requirement description; Performing semantic coding on the program feature description to obtain a semantic feature of the program feature description; Semantically encoding the business requirement description to obtain semantic features of the business requirement description; Performing dynamic memory interactive attention fusion on the semantic features describing the program characteristics and the semantic features describing the business requirements to obtain fine-grained semantic interactive features of program characteristics and business requirements; Based on the program characteristics-business demand fine-grained semantic interaction features, a custom encoding template is obtained.

[0007] According to another aspect of the present application, a wind-solar power generation energy storage management system is provided, comprising: A program feature description acquisition module, used to acquire a program feature description; Business requirement description acquisition module, used to obtain business requirement description; A program feature description semantic coding module, used for semantically coding the program feature description to obtain a program feature description semantic feature; A business requirement description semantic coding module, used to semantically code the business requirement description to obtain semantic features of the business requirement description; A dynamic memory interactive attention fusion module is used to perform dynamic memory interactive attention fusion on the semantic features describing the program characteristics and the semantic features describing the business requirements to obtain fine-grained semantic interactive features of program characteristics and business requirements; The custom coding template generation module is used to obtain a custom coding template based on the fine-grained semantic interaction features of the program characteristics and business requirements.

[0008] Compared with the prior art, the present application provides a method and system for generating a custom cataloging template, which obtains a program feature description and a business demand description, and uses natural language processing and analysis technology based on deep learning to perform semantic coding of the program feature description and the program feature description, so as to intelligently generate a definition coding template based on the dynamic memory interaction representation between the semantic features of the program feature description and the semantic features of the business demand description. In this way, the most suitable cataloging template can be automatically generated according to the specific needs and demand feedback of each user, and personalized services can be provided. It can also consider user feedback in real time, and automatically update the cataloging template according to the latest data and changed content to ensure that it always fits the actual business scenario. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 A flowchart of a method for generating a custom cataloging template according to an embodiment of the present application; Figure 2 A data flow diagram of a method for generating a custom cataloging template according to an embodiment of the present application; Figure 3 It is a flowchart of sub-step S5 of the wind-solar power generation energy storage management method according to an embodiment of the present application; Figure 4 A block diagram of a system for generating a custom catalog template according to an embodiment of the present application. DETAILED DESCRIPTION

[0011] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0012] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0013] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0014] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0015] Next, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all embodiments of the present application. It should be understood that the present application is not limited by the exemplary embodiments described herein.

[0016] In the technical solution of the present application, a method for generating a custom catalog template is proposed. Figure 1 FIG. is a flowchart of a method for generating a custom catalog template according to an embodiment of the present application. Figure 2 FIG. is a schematic diagram of data flow of a method for generating a custom catalog template according to an embodiment of the present application. As Figure 1 and Figure 2 shown, the method for generating a custom catalog template according to an embodiment of the present application includes the steps of: S1, obtaining a program feature description; S2, obtaining a service requirement description; S3, performing semantic encoding on the program feature description to obtain a program feature description semantic feature; S4, performing semantic encoding on the service requirement description to obtain a service requirement description semantic feature; S5, performing dynamic memory interaction attention fusion on the program feature description semantic feature and the service requirement description semantic feature to obtain a program feature-service requirement fine-grained semantic interaction feature; S6, obtaining a custom encoding template based on the program feature-service requirement fine-grained semantic interaction feature.

[0017] Specifically, in S1, a program feature description is obtained. Obtaining a program feature description involves collecting raw data about the program from multiple channels. These data sources are extensive, including metadata provided by the program production party (such as title, introduction, director, actor information, etc.), user comments and feedback (audience evaluations of the program, discussion topics, preferences, etc.), social media analysis (scraping content such as posts and tags on social networks related to the program to understand the public's interest points and social responses), and automatic video content analysis (using computer vision technology to automatically identify elements such as scenes, objects, and human expressions in the video). By integrating information from different channels, the characteristics of the program content and the user's feelings towards it can be comprehensively captured.

[0018] After obtaining the above data, necessary text preprocessing work is carried out to ensure the effectiveness of subsequent semantic encoding. This includes removing non-text information such as irrelevant characters and HTML tags; splitting continuous text into meaningful lexical units (words) for further analysis; removing common words that are not very helpful for semantic understanding (such as "de", "shi", etc.); and normalizing different forms of words into basic forms, for example, changing "running" to "run". Such preprocessing steps help to improve the text quality and make it more suitable for in-depth semantic parsing.

[0019] Next, extract the key features that can represent the characteristics of the program from the preprocessed text content. This step can be achieved through a variety of methods, such as using the LDA (Latent Dirichlet Allocation) algorithm to find the implicit topic distribution in the text and summarize the main content direction of the program; determine the key words that best reflect the characteristics of the program based on algorithms such as TF-IDF (Term Frequency-InverseDocument Frequency) and TextRank; evaluate the emotional tendency of the text to determine whether the program tends to express positive, negative or neutral emotions; identify and classify proper nouns in the text (such as names of people, places, and organizations), which is particularly important for certain types of programs. These feature extraction methods can effectively extract core information from large amounts of text data.

[0020] In particular, S2, obtains the business demand description. Obtaining the business demand description not only involves capturing the needs directly expressed by users, but also requires an in-depth understanding of the meaning behind these needs and their relevance to program content. The implementation of this process begins with collecting information about business needs through multiple channels. This includes directly asking users about their specific expectations for program cataloging, such as what information they want to highlight and what the classification criteria are; analyzing previous cataloging records and user usage patterns to identify common demand types and preference trends; referring to industry-recognized cataloging specifications and success cases to ensure that the generated templates meet professional requirements and have wide applicability; and considering various rules and guidelines formulated within the company and any special requirements that may affect the cataloging method.

[0021] In particular, the S3 performs semantic encoding on the program feature description to obtain a program feature description semantic feature. That is, in the technical solution of the present application, the program feature description is subjected to word segmentation processing and then input into a program feature description semantic encoder containing a word embedding layer to obtain a program feature description semantic encoding vector as the program feature description semantic feature. Considering that the program feature description contains important information about program content, form, style, etc., which reflects the specific and unique semantic information of each program, in order to capture and extract the key semantic information of the program feature description to accurately understand the essential characteristics of the program content, in the technical solution of the present application, the program feature description is semantically encoded to obtain a program feature description semantic encoding vector. In particular, in a specific embodiment of the present application, the program feature description may first be subjected to word segmentation processing to divide a complete program feature description information into meaningful vocabulary units (words), and then input into a program feature description semantic encoder containing a word embedding layer to further capture and extract the context information implied in each word and the semantic association between each word, thereby obtaining a program feature description semantic encoding vector containing rich semantic information about the program feature description.

[0022] In particular, the S4 semantically encodes the business requirement description to obtain the semantic features of the business requirement description. That is, in the technical solution of the present application, the business requirement description is segmented and then input into a business requirement description semantic encoder containing a word embedding layer to obtain a business requirement description semantic encoding vector as the semantic features of the business requirement description. Here, considering that the business requirement description elaborates on the user's specific expectations for the program, including the type of program, scene and other elements, these descriptions contain key business requirement semantic information. Based on this, in the technical solution of the present application, the business requirement description is semantically encoded to deeply understand the key information of the business requirements, capture deeper semantic relationships, and obtain a business requirement description semantic encoding vector. In particular, in a specific embodiment of the present application, after the continuous business requirement description text is segmented into meaningful vocabulary units, a business requirement description semantic encoder containing a word embedding layer is used to perform semantic understanding on it, and the contextual semantic relationship between each business requirement word is considered to better understand the semantic meaning of the complex business requirement description, and provide rich and complete data support for the subsequent generation of more accurate and personalized definition coding templates.

[0023] In particular, in S5, the semantic features of the program feature description and the semantic features of the business demand description are dynamically memorized and interactively fused to obtain fine-grained semantic interaction features of program features and business demands. Considering that the semantic features of the program feature description provide a deep understanding of the program content. The semantic features of the business demand description provide users with clear business operation guidelines. That is, the program feature description provides information on "what it is", while the business demand description clarifies "how to use it", and the two together determine the specific form and content of the cataloging template. Therefore, in order to combine the information of these two features to generate a cataloging template that better meets user needs, in the technical solution of the present application, the semantic features of the program feature description and the semantic features of the business demand description are dynamically memorized and interactively fused to fuse the information of program features and business demands at a more detailed level, capture the subtle association between the two, and obtain fine-grained semantic interaction features of program features and business demands. In particular, the dynamic memory interactive attention fusion mechanism introduces external knowledge (such as domain-specific knowledge graphs, industry standards, etc.) to help the system better understand the meaning behind the text, such as identifying implicit professional terms or cultural references, so as to make more informed decisions. In a specific example of this application, Figure 3 As shown, the S5 includes: S51, performing fine-grained semantic interaction on the semantic features describing the program characteristics and the semantic features describing the business requirements to obtain fine-grained semantic features of program characteristics-business requirements; S52, guided by the fine-grained semantic features of program characteristics-business requirements, performing interactive modulation on the semantic features describing the program characteristics and the semantic features describing the business requirements based on external knowledge to obtain the fine-grained semantic interaction features of program characteristics-business requirements.

[0024] Specifically, the S51 performs fine-grained semantic interaction on the semantic features of the program feature description and the semantic features of the business demand description to obtain the fine-grained features of the program feature-business demand semantics. That is, in the technical solution of the present application, the program feature description semantic coding vector is multiplied by the transposed vector of the business demand description semantic coding vector to obtain the program feature-business demand semantic fine-grained feature interaction matrix as the program feature-business demand semantic fine-grained features. By inputting the program feature description semantic features and the business demand description semantic features into a fine-grained feature interaction network to capture the fine-grained interaction information between the program feature description semantic coding vector and the business demand description semantic coding vector, the construction of a fine-grained feature interaction matrix allows the model to understand the correlation between different features at a micro level, and this correlation is crucial for subsequent information fusion. More specifically, the program feature description semantic features and the business demand description semantic features are subjected to fine-grained semantic interaction using the following fine-grained semantic interaction formula to obtain the program feature-business demand semantic fine-grained features; wherein, the fine-grained semantic interaction formula is: ;in, A semantic coding vector describing the program characteristics, Describe the semantic encoding vector for the business requirement, for The transposed vector of is matrix multiplication, It is the program feature-business requirement semantic fine-grained feature interaction matrix.

[0025] Specifically, the S52, guided by the program feature-business demand semantic fine-grained features, performs interactive modulation based on external knowledge on the program feature description semantic features and the business demand description semantic features to obtain the program feature-business demand fine-grained semantic interactive features. That is, in the technical solution of the present application, the specific steps of performing interactive modulation based on external knowledge on the program feature description semantic features and the business demand description semantic features include: first, performing external knowledge attention optimization on the program feature-business demand semantic fine-grained feature interaction matrix to obtain the program feature-business demand semantic external knowledge optimized fine-grained feature interaction matrix. By inputting the program feature-business demand semantic fine-grained feature interaction matrix into the attention unit based on external knowledge, the fine-grained feature interaction matrix is ​​further optimized by using external knowledge. This process not only enhances the model's ability to understand program features-business demands, but also enables the model to capture the potential connection between program features-business demands in a larger range. External knowledge can come from multiple channels, such as the knowledge of domain experts, historical data, etc., which are effectively integrated into the model through the attention mechanism, thereby improving the generalization ability and accuracy of the model. More specifically, the program characteristics-business demand semantic fine-grained feature interaction matrix is ​​optimized with external knowledge attention using the following external knowledge attention optimization formula to obtain a program characteristics-business demand semantic external knowledge optimized fine-grained feature interaction matrix; wherein the external knowledge attention optimization formula is: ;in, and They represent the learnable memory parameter matrix, for The transposed matrix of represents the normalization function, It is the program features-business requirements semantics external knowledge optimization fine-grained feature interaction matrix.

[0026] Next, based on the program feature-business requirement semantic external knowledge optimization fine-grained feature interaction matrix, the program feature description semantic coding vector and the business requirement description semantic coding vector are respectively subjected to external knowledge embedding modulation to obtain the optimized program feature description semantic coding vector and the optimized business requirement description semantic coding vector. Specifically, the program feature description semantic coding vector is linearly transformed to obtain a program feature description query feature vector and a program feature description value feature vector, and the program feature-business requirement semantic external knowledge optimization fine-grained feature interaction matrix is ​​used as a key matrix, and the program feature description query feature vector, the program feature description value feature vector and the key matrix are subjected to converter-based feature modulation to obtain the optimized program feature description semantic coding vector; and the business requirement description semantic coding vector is linearly transformed to obtain a business requirement description query feature vector and a business requirement description value feature vector, and the program feature-business requirement semantic external knowledge optimization fine-grained feature interaction matrix is ​​used as a key matrix, and the business requirement description query feature vector, the business requirement description value feature vector and the key matrix are subjected to converter-based feature modulation to obtain the optimized business requirement description semantic coding vector. That is, the information exchange and integration between the internal and external knowledge of the feature vector is realized through the self-attention mechanism, so as to ensure that the program feature description semantic coding vector can benefit from external knowledge to improve the quality of its feature expression. Similarly, the business demand description semantic coding vector also undergoes the same processing flow, that is, linear transformation generates query and value feature vectors, and acts together with the fine-grained feature interaction matrix optimized by external knowledge in the fine-grained modulation module based on the Transformer structure. In this way, the business demand description semantic coding vector can also benefit from external knowledge, thereby improving the quality of its own feature expression. In particular, when the fine-grained modulation module based on the Transformer structure is used to perform feature fine-grained modulation based on external knowledge on the program feature description semantic coding vector and the business demand description semantic coding vector, it uses the program feature-business demand semantic external knowledge optimized fine-grained feature interaction matrix as a common key matrix space, so that the optimized program feature description semantic coding vector and the optimized business demand description semantic coding vector have relatively good alignment characteristics in the high-dimensional feature space. More specifically, based on the program feature-business requirement semantic external knowledge optimization fine-grained feature interaction matrix, the program feature description semantic coding vector and the business requirement description semantic coding vector are respectively subjected to external knowledge embedding modulation using the following external knowledge embedding modulation formula to obtain an optimized program feature description semantic coding vector and an optimized business requirement description semantic coding vector; wherein the external knowledge embedding modulation formula is: ;in, and are the program feature description query embedding matrix and the program feature description first bias vector, respectively. is the feature vector describing the query, and are the program feature description value embedding matrix and the program feature description second bias vector respectively, is the feature vector describing the program characteristics, for The transposed matrix of yes Length, yes function, is the optimized program feature description semantic encoding vector, and are the business requirement description query embedding matrix and the first bias vector of the business requirement description, respectively. is the business requirement description query feature vector, and are the business requirement description value embedding matrix and the business requirement description second bias vector respectively. is the business requirement description value feature vector, yes Length, It is the optimized semantic encoding vector of business requirement description.

[0027] Subsequently, the semantic interaction between the optimized program feature description semantic coding vector and the optimized business requirement description semantic coding vector is calculated to obtain the program feature-business requirement fine-grained semantic interaction coding vector as the program feature-business requirement fine-grained semantic interaction feature. Here, the final program feature-business requirement fine-grained semantic interaction coding vector is generated by calculating the position-weighted sum of the optimized program feature description semantic coding vector and the optimized business requirement description semantic coding vector. That is, the elements of the corresponding positions of the optimized program feature description semantic coding vector and the optimized business requirement description semantic coding vector are weighted and summed by attention weights to generate the program feature-business requirement fine-grained semantic interaction coding vector to better reflect the interactive relationship between the two. More specifically, the semantic interaction between the optimized program feature description semantic coding vector and the optimized business requirement description semantic coding vector is calculated using the following semantic interaction formula to obtain the program feature-business requirement fine-grained semantic interaction coding vector; wherein, the semantic interaction formula is: ;in, is the optimized program feature description semantic coding vector, is the optimized semantic encoding vector of the business requirement description, It is the program feature - business requirement fine-grained semantic interaction encoding vector.

[0028] In particular, the S6 obtains a custom coding template based on the fine-grained semantic interaction features of the program features and business requirements. That is, in the technical solution of the present application, after adding a prompt word to the tail of the fine-grained semantic interaction coding vector of the program features and business requirements, it is input into the cataloging template generator based on the large language model to obtain the custom coding template. That is, after adding a prompt word to the tail of the fine-grained semantic interaction coding vector of the program features and business requirements, it is input into the cataloging template generator based on the large language model to obtain the custom coding template, wherein the prompt is "Task: Generate a cataloging template based on program features, business requirements and semantic association information of the two". That is, the fine-grained semantic interaction features of program features and business requirements obtained by dynamic memory interaction of the semantic features describing the program features and the semantic features describing the business requirements are generated and processed, so as to intelligently generate a defined coding template. In this way, the most suitable cataloging template can be automatically generated according to the specific needs and feedback of each user, and personalized services can be provided. It can also consider user feedback in real time, and automatically update the cataloging template according to the latest data and changed content to ensure that it always fits the actual business scenario. It should be understood that the prompt words not only tell the model what to do, but also implicitly convey important information about program characteristics and business needs, helping the model to better understand the background and purpose of the input data. The catalog template generator based on the large language model can ensure that the generated catalog template is closely centered around program characteristics and business needs, avoiding irrelevant or off-topic content. In particular, in a specific embodiment of the present application, the large language model can be GPT-4.

[0029] In a specific example, suppose a large news organization wants to create a very detailed catalog template for its report content, including information such as the timestamp, geographic location, people involved, event type, etc. of each report. Such a template can help quickly retrieve news materials within a specific time period, making it easier for editors to produce retrospective features or emergency response breaking news. Here are the detailed steps of the process: First, data about news reports are collected from multiple channels, including manuscripts submitted by journalists (the original manuscripts contain titles, texts, picture captions, etc.), internal communication records of the editorial department (such as meeting minutes, email exchanges, etc., which may contain discussions on the focus and direction of the report), social media monitoring (by capturing posts and hashtags on social networks related to the event to understand the public's interests and social reactions), and automatic analysis of video content (using computer vision technology to automatically identify scenes, objects, human expressions and other elements in the video, which is especially important for television news). At the same time, directly ask the editorial team about their specific expectations for the cataloging of news materials, such as what information they want to highlight and what the classification standards are. In addition, refer to the cataloging specifications recognized in the industry and the various rules formulated within the news organization to ensure that the generated templates meet both professional requirements and have wide applicability.

[0030] Next, text preprocessing is performed to ensure that all collected data is cleaned, segmented, stop words are removed, and stemmed or morphologically restored to ensure the effectiveness of subsequent analysis. This stage of work also includes the standardization of timestamps, converting time information in different formats into a standard format that can be recognized by the system; the annotation of geographic locations, using geographic information system (GIS) tools to determine the specific coordinates of the reporting location; and the identification of people and event types, using named entity recognition (NER) technology to mark key people and event categories. Such preprocessing steps help improve the quality of text, making it more suitable for deep semantic analysis, thereby providing a solid foundation for subsequent feature extraction and semantic encoding.

[0031] After preprocessing, algorithms such as LDA (Latent Dirichlet Allocation), TF-IDF (TermFrequency-Inverse Document Frequency), and TextRank are used to extract key features that can represent the characteristics of news reports. These methods can effectively summarize the direction and characteristics of the main content of the report from a large amount of text data. At the same time, sentiment analysis is used to evaluate the emotional tendency of the text to understand whether the reader's attitude towards certain events is positive or negative; named entity recognition (NER) is used to identify and classify proper nouns in the text, such as names of people, places, organizations, etc.; and intent recognition technology is used to identify the user's actual intention, such as distinguishing different types of information input such as query requests, command instructions, or suggestion feedback. These feature extraction methods can effectively extract core information from a large amount of text data, laying a solid foundation for further semantic encoding.

[0032] In order for computers to understand and process business requirement descriptions, they also need to be converted into structured digital representations, namely, semantic encoding vectors of business requirement descriptions. This step is usually performed using a deep learning model, as follows: with the help of a pre-trained language model (such as BERT), each word is mapped to a vector representation in a high-dimensional space. These vectors not only contain the meaning of the words themselves, but also retain the semantic relationship between words; use RNN (recurrent neural network), LSTM (long short-term memory network) or Transformer architecture encoders to capture contextual information within and across sentences, thereby generating richer and more accurate semantic representations. This encoding method can not only understand the meaning of individual words, but also parse their usage and relationships in specific contexts, providing a solid foundation for subsequent dynamic memory interactive attention fusion.

[0033] After completing all the above preparations, the most important step is to conduct fine-grained semantic interaction between the semantic features describing the characteristics of news reports and the semantic features describing business needs, and to perform interactive modulation based on external knowledge. This is to combine the specific characteristics of the news content with the user's clear business operation guidelines to generate a cataloging template that fits the news itself and meets the user's specific needs. In this process, the system will introduce external knowledge (such as domain-specific knowledge graphs, industry standards, etc.) to help better understand the meaning behind the text, such as identifying implicit professional terms or cultural references, so as to make more informed decisions. In this way, it can be ensured that the generated cataloging template is not only closely centered around the characteristics of the news, but also fully considers the user's business needs to avoid irrelevant or off-topic content.

[0034] Finally, after all the above preparations are completed, add a prompt word to the end of the fine-grained semantic interaction encoding vector of news report characteristics and business requirements: "Task: Generate a cataloging template based on news report characteristics, business requirements, and the semantic association information between the two", and then input it into the cataloging template generator based on a large language model (such as GPT-4). This prompt not only tells the model what to do, but also implicitly conveys important information about the characteristics of news reports and business requirements, helping the model to better understand the background and purpose of the input data. Through this mechanism, the cataloging template generator can intelligently generate a personalized cataloging template that conforms to the characteristics of news content and meets specific business needs based on the information provided.

[0035] After the above steps, the catalog template generator may output the following structured catalog template: [Report Name] [Automatically generated] [Report Type] News Report

Release time

Geographic location

Event Type

Main content

[0036] - Mayor’s speech: Mayor Zhang San elaborated on the city’s development plan for the next five years.

[0037] - Citizen interaction: Some citizens asked questions and received responses on the spot.

[0038]

Key words

[0039] In order to support more complex retrieval requirements, the cataloging template can be further refined, including but not limited to the following: the time of report release accurate to the minute level, the geographical location extended to streets, buildings and even GPS coordinates, the people involved with detailed information such as not only their names but also their positions and affiliated institutions, the types of events subdivided according to internationally accepted news classification standards (such as ICFJ classification), and multimedia links to photos, videos or other forms of multimedia resources mentioned in the related reports. Such detailed cataloging templates can not only meet the needs of daily editing work, but also provide strong support for future data analysis and research.

[0040] Through such a comprehensive process, the news organization successfully created a personalized cataloging template for each report that meets business needs. This template not only closely focuses on the characteristics of the news and avoids irrelevant or off-topic content, but also fully considers the actual needs of users and improves the quality of content management and user experience. This approach not only improves the flexibility and responsiveness of the cataloging system, but also provides users with a more intelligent and personalized service experience, while also enhancing the competitiveness of enterprises in the rapidly changing content market.

[0041] Preferably, after adding a prompt word to the tail of the program feature-business requirement fine-grained semantic interaction coding vector, the vector is input into a catalog template generator based on a large language model to obtain a custom coding template, including: The distance between each pair of feature values ​​of the program feature-business requirement fine-grained semantic interaction encoding vector is calculated, for example L 2 distance, and take the square root of the distance to obtain the program feature-business demand fine-grained semantic interaction encoding distance representation matrix, that is, ;in, and is the eigenvalue of the program feature-business requirement fine-grained semantic interaction encoding vector, The distance between the feature values ​​of the fine-grained semantic interaction coding vector representing the program feature-business requirement, is the program feature-business requirement fine-grained semantic interaction encoding vector, is the program feature-business requirement fine-grained semantic interaction encoding distance representation matrix; The program feature-business requirement fine-grained semantic interaction coding autocorrelation matrix of the program feature-business requirement fine-grained semantic interaction coding vector as a row vector is obtained, that is, ; The program characteristics-business requirements fine-grained semantic interaction coding vector is matrix-multiplied with the program characteristics-business requirements fine-grained semantic interaction coding distance representation matrix to obtain the program characteristics-business requirements fine-grained semantic interaction coding first-level mapping vector, that is, ; The program characteristics-business requirements fine-grained semantic interaction coding first-level mapping vector is matrix-multiplied with the program characteristics-business requirements fine-grained semantic interaction coding distance representation matrix and the program characteristics-business requirements fine-grained semantic interaction coding self-association matrix to obtain the program characteristics-business requirements fine-grained semantic interaction coding multi-level mapping vector ; The program characteristics-business requirements fine-grained semantic interaction coding multi-level mapping vector is interpolated with the program characteristics-business requirements fine-grained semantic interaction coding associated eigenvector (if the eigenvalue is insufficient, interpolation or zero padding) composed of the eigenvalue of the program characteristics-business requirements fine-grained semantic interaction coding auto-association matrix to obtain an optimized program characteristics-business requirements fine-grained semantic interaction coding vector; and After adding the prompt word to the tail of the optimized program feature-business requirement fine-grained semantic interaction coding vector, it is input into a catalog template generator based on a large language model to obtain a custom coding template.

[0042] Taking into account that the program feature description semantic coding vector and the business requirement description semantic coding vector represent the coded semantic features of the program feature description and the business requirement description respectively, when performing fine-grained feature interaction based on external knowledge modulation, insufficient correspondence in external knowledge modulation of the coded semantic features caused by different source semantics will lead to the lack of fine-grained interaction feature instance determination of the program feature-business requirement fine-grained semantic interaction coding vector, thereby affecting the semantic quality of the custom coding template obtained after adding the prompt words and inputting it into the catalog template generator based on the large language model.

[0043] Therefore, through the linear target mapping representation based on the similarity distance representation matrix of the program characteristics-business needs fine-grained semantic interaction coding vector, the self-correlation of the complete similarity instantiation of the program characteristics-business needs fine-grained semantic interaction coding vector is represented by a secondary target mapping representation based on a multi-level distribution hierarchy, and the association mismatch negative impact factor is compensated by the association fusion kernel bias to improve the degree of instance determination of the eigenvalue of the program characteristics-business needs fine-grained semantic interaction coding vector under similarity constraints, that is, the significance of the eigenvalue as an instance for generative regression judgment, and improve the semantic quality of the custom coding template obtained by adding a prompt word to the tail of the program characteristics-business needs fine-grained semantic interaction coding vector and inputting it into the cataloging template generator based on the large language model.

[0044] In summary, the method for generating a custom cataloging template according to the embodiment of the present application is explained, which obtains the program feature description and the business demand description, and uses natural language processing and analysis technology based on deep learning to perform semantic coding of the program feature description and the program feature description, so as to intelligently generate a definition coding template based on the dynamic memory interaction representation between the semantic features of the program feature description and the semantic features of the business demand description. In this way, the most suitable cataloging template can be automatically generated according to the specific needs and demand feedback of each user, and personalized services can be provided. It can also consider user feedback in real time, and automatically update the cataloging template according to the latest data and changed content to ensure that it always fits the actual business scenario.

[0045] Furthermore, a custom cataloging template generation system is also provided.

[0046] Figure 4 FIG. 1 is a block diagram of a system for generating a custom catalog template according to an embodiment of the present application. Figure 4 As shown, according to the embodiment of the present application, the custom cataloging template generation system 300 includes: a program feature description acquisition module 310, which is used to obtain the program feature description; a business requirement description acquisition module 320, which is used to obtain the business requirement description; a program feature description semantic encoding module 330, which is used to semantically encode the program feature description to obtain the program feature description semantic features; a business requirement description semantic encoding module 340, which is used to semantically encode the business requirement description to obtain the business requirement description semantic features; a dynamic memory interactive attention fusion module 350, which is used to perform dynamic memory interactive attention fusion on the program feature description semantic features and the business requirement description semantic features to obtain program feature-business requirement fine-grained semantic interaction features; a custom coding template generation module 360, which is used to obtain a custom coding template based on the program feature-business requirement fine-grained semantic interaction features.

[0047] As described above, the custom catalog template generation system 300 according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with a custom catalog template generation algorithm. In a possible implementation, the custom catalog template generation system 300 according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the custom catalog template generation system 300 can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the custom catalog template generation system 300 can also be one of the many hardware modules of the wireless terminal.

[0048] Alternatively, in another example, the custom catalog template generation system 300 and the wireless terminal may be separate devices, and the custom catalog template generation system 300 may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0049] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for generating a custom catalog template, characterized in that: include: Get the program feature description; Obtain business requirement description; Performing semantic coding on the program feature description to obtain a semantic feature of the program feature description; Semantically encoding the business requirement description to obtain semantic features of the business requirement description; Performing dynamic memory interactive attention fusion on the semantic features of the program feature description and the semantic features of the business requirement description to obtain fine-grained semantic interactive features of program features-business requirements, including: performing fine-grained semantic interaction on the semantic features of the program feature description and the semantic features of the business requirement description to obtain fine-grained semantic features of program features-business requirements; Guided by the program feature-business requirement semantic fine-grained features, the program feature description semantic features and the business requirement description semantic features are interactively modulated based on external knowledge to obtain the program feature-business requirement fine-grained semantic interactive features; Based on the program characteristics-business demand fine-grained semantic interaction features, a custom encoding template is obtained.

2. The method for generating a custom catalog template according to claim 1, characterized in that: The program feature description is semantically encoded to obtain a program feature description semantic feature, including: the program feature description is word segmented and then input into a program feature description semantic encoder including a word embedding layer to obtain a program feature description semantic encoding vector as the program feature description semantic feature.

3. The method for generating a custom catalog template according to claim 2, characterized in that: The business requirement description is semantically encoded to obtain a semantic feature of the business requirement description, including: the business requirement description is word segmented and then input into a business requirement description semantic encoder including a word embedding layer to obtain a business requirement description semantic encoding vector as the semantic feature of the business requirement description.

4. The method for generating a custom catalog template according to claim 3, characterized in that: Fine-grained semantic interaction is performed on the semantic features describing the program features and the semantic features describing the business requirements to obtain fine-grained semantic features of program features-business requirements, including: multiplying the semantic coding vector of the program features description with the transposed vector of the semantic coding vector of the business requirements description to obtain a fine-grained feature interaction matrix of program features-business requirements semantics as the fine-grained semantic features of program features-business requirements.

5. The method for generating a custom catalog template according to claim 4, characterized in that: Guided by the program feature-business requirement semantic fine-grained features, the program feature description semantic features and the business requirement description semantic features are interactively modulated based on external knowledge to obtain the program feature-business requirement fine-grained semantic interactive features, including: Performing external knowledge attention optimization on the program characteristics-business demand semantic fine-grained feature interaction matrix to obtain a program characteristics-business demand semantic external knowledge optimized fine-grained feature interaction matrix; Based on the program feature-business requirement semantic external knowledge, the fine-grained feature interaction matrix is ​​optimized, and the program feature description semantic coding vector and the business requirement description semantic coding vector are respectively subjected to external knowledge embedding modulation to obtain an optimized program feature description semantic coding vector and an optimized business requirement description semantic coding vector; The semantic interaction between the optimized program feature description semantic coding vector and the optimized business requirement description semantic coding vector is calculated to obtain a program feature-business requirement fine-grained semantic interaction coding vector as the program feature-business requirement fine-grained semantic interaction feature.

6. The method for generating a custom catalog template according to claim 5, characterized in that: Based on the program feature-business requirement semantic external knowledge, the fine-grained feature interaction matrix is ​​optimized, and external knowledge embedding modulation is performed on the program feature description semantic coding vector and the business requirement description semantic coding vector to obtain an optimized program feature description semantic coding vector and an optimized business requirement description semantic coding vector, including: Performing a linear transformation on the program feature description semantic coding vector to obtain a program feature description query feature vector and a program feature description value feature vector, and using the program feature-business demand semantic external knowledge optimized fine-grained feature interaction matrix as a key matrix, performing converter-based feature modulation on the program feature description query feature vector, the program feature description value feature vector and the key matrix to obtain the optimized program feature description semantic coding vector; The business requirement description semantic coding vector is linearly transformed to obtain a business requirement description query feature vector and a business requirement description value feature vector, and the program characteristics-business requirement semantic external knowledge optimized fine-grained feature interaction matrix is ​​used as a key matrix. The business requirement description query feature vector, the business requirement description value feature vector and the key matrix are subjected to the converter-based feature modulation to obtain the optimized business requirement description semantic coding vector.

7. The method for generating a custom catalog template according to claim 6, characterized in that: Based on the program characteristics-business requirements fine-grained semantic interaction coding vector, a custom coding template is obtained, including: after adding a prompt word to the end of the program characteristics-business requirements fine-grained semantic interaction coding vector, it is input into a cataloging template generator based on a large language model to obtain the custom coding template.

8. The method for generating a custom catalog template according to claim 7, characterized in that: The prompt is "Task: Generate cataloging template based on program characteristics, business requirements and semantic association information between the two." 9. A custom catalog template generation system, characterized in that: include: A program feature description acquisition module, used to acquire a program feature description; Business requirement description acquisition module, used to obtain business requirement description; A program feature description semantic coding module, used for semantically coding the program feature description to obtain a program feature description semantic feature; A business requirement description semantic coding module, used to semantically code the business requirement description to obtain semantic features of the business requirement description; A dynamic memory interactive attention fusion module is used to perform dynamic memory interactive attention fusion on the semantic features describing the program characteristics and the semantic features describing the business requirements to obtain fine-grained semantic interactive features of program characteristics and business requirements; The custom coding template generation module is used to obtain a custom coding template based on the fine-grained semantic interaction features of the program characteristics and business requirements.

Images