A QC document and media analysis application platform and method

Abstract

The present invention discloses a QC document and media parsing application platform and method thereof, belonging to the document and media content parsing technology field. The platform includes a document parsing module, a video text recognition module, a media voice recognition module, a retrieval module, a relationship diagram drawing module and a rights management module. The document parsing module is used to extract content and perform word segmentation storage on QC documents uploaded by users; the media voice recognition module and the video text recognition module are used to perform voice and text recognition on media videos uploaded by users and associated with QC documents, respectively; the retrieval module is used to retrieve and display document and video content using a dynamic weighted similarity algorithm; the relationship diagram drawing module is used to draw relationship diagrams online; and the rights management module is used to set and manage platform user rights. The present invention effectively improves the efficiency and accuracy of QC document management and information retrieval, realizes rapid information sharing and tracing, and optimizes the quality control process.

Description

Technical Field

[0001] The present invention belongs to the technical field of document and media content analysis, and relates to an application platform and method for analyzing QC documents and their media. Background Art

[0002] In current quality control (QC) document management processes, processing large volumes of QC documents and their associated media files, such as video and audio, is a highly manual task. Traditional processing methods typically require manual review of documents, viewing video content, and recording or analyzing information. This approach has several significant drawbacks and shortcomings:

[0003] Inefficiency: QC documents and their associated media files are large in volume. Manual processing is not only time-consuming but also prone to omissions or misjudgments, resulting in overall low efficiency.

[0004] Isolated information is difficult to integrate: The correlation between text, images, and video content in QC documents is weak, and there is a lack of a unified processing mechanism. Information extraction and sharing are inconvenient, making it difficult to achieve rapid query and comprehensive tracking.

[0005] Difficulty in tracing: During subsequent quality control inspections or problem tracing, traditional document and video management methods cannot effectively support the rapid search for specific content, posing a significant challenge to problem troubleshooting and analysis.

[0006] Especially in quality control work, accurate and timely parsing and searching for relevant documents and video content is crucial. However, existing manual or semi-automated processing methods can no longer meet the current high efficiency and accuracy requirements of the quality control industry. Therefore, a more intelligent and automated QC document and media parsing application platform is urgently needed to improve work efficiency, ensure the comprehensiveness and accuracy of information, and support efficient content retrieval and traceability. Summary of the Invention

[0007] In order to address the deficiencies in the prior art, the present invention provides an application platform and method for parsing QC documents and their media, which can efficiently and accurately parse QC documents and multimedia attachments, and support content-based full-text retrieval and tracing. By automatically and uniformly parsing, storing, and retrieving text, images, and audio content, manual operations are greatly reduced, the efficiency and accuracy of QC document management and information retrieval are effectively improved, the rapid sharing and tracing of information are realized, the quality control process is optimized, and a more intelligent management method is provided for the enterprise's QC work.

[0008] The present invention adopts the following technical solutions.

[0009] The first aspect of the present invention provides a QC document and media parsing application platform, including a document parsing module, a video text recognition module, a media voice recognition module, a retrieval module, a relationship diagram drawing module and a rights management module;

[0010] The document parsing module is used to extract the content of the QC documents uploaded by the user and store them in categories;

[0011] The media voice recognition module and the video text recognition module are used to perform voice and text recognition on the media videos uploaded by the user and associated with the QC documents respectively;

[0012] The retrieval module is used to retrieve and display documents and video content using a dynamic weighted similarity algorithm;

[0013] The relationship graph drawing module is used to draw the relationship graph online in real time based on the graph data structure;

[0014] The rights management module is used for role management, batch maintenance and configuration of user rights.

[0015] Preferably, the document parsing module extracts and parses the text content and image data in the QC document by combining document content extraction technology with OCR recognition technology, and stores them in the database and Elasticsearch. The text content and image data are classified, stored and indexed according to the document structure.

[0016] Preferably, the document parsing module displays the document parsing progress in real time and provides an exception prompt function. When an exception occurs during the parsing process, the exception situation is immediately displayed.

[0017] Preferably, the media speech recognition module uses an audio processing tool to extract audio content from the media video;

[0018] The video text recognition module transcribes the audio content into text content through a speech recognition model, extracts key information according to the document structure, stores it in the database and Elasticsearch, and associates it with the corresponding QC document.

[0019] Preferably, the retrieval module performs full-text retrieval based on the query conditions input by the user through a dynamic weighted similarity algorithm, matches the text content of the corresponding QC document, the text content obtained by transcribing the image data and audio content, and displays the QC documents and media videos containing relevant content.

[0020] Preferably, the algorithm flow of the dynamic weighted similarity algorithm includes:

[0021] The query conditions entered by the user and the text content in the QC document are word-vectorized, and then the dynamic weighted similarity between the query conditions and the QC document is calculated based on the word vectors. The text content of the most matching QC document is obtained based on the dynamic weighted similarity, and the text content obtained by transcribing the image data and audio content associated with the text content is obtained from the database and Elasticsearch.

[0022] Preferably, the calculation formula of the dynamic weighted similarity is:

[0023] ,

[0024] in: For query conditions and text content Dynamic weighted similarity of; For text content Middle field The weight of For fields Dynamic factors; is the normalization factor; For text content The number of fields in ; For query conditions Text content field No. The query word vector that matches the word vector; For fields Middle word vectors; For fields The number of word vectors in .

[0025] Preferably, the dynamic factor The calculation formula is:

[0026] ,

[0027] ,

[0028] ,

[0029] ,

[0030] in, For fields In the text content The global importance of

[0031] For fields Local relevance to the query conditions;

[0032] U i Adjustment factor for user behavior feedback;

[0033] 、 For fields , Field The word vector in the text content word frequency in ;

[0034] Respectively query term vectors and fields Middle word vectors;

[0035] is the cosine function; is the weight adjustment parameter;

[0036] The field for the hth interaction in the user's historical behavior record Importance rating;

[0037] The field for the hth interaction in the user's historical behavior record relevance;

[0038] is the total number of samples in the user's history.

[0039] Preferably, the importance score The calculation formula is:

[0040] ,

[0041] in, For the Fields in Interaction The probability of the corresponding content being clicked;

[0042] For the Fields in Interaction Average dwell time on the corresponding content;

[0043] For the From the field The proportion of actions that lead to download, open, or preview;

[0044] is the weight parameter.

[0045] Preferably, the correlation The calculation formula is: ,

[0046] in, For the The semantic vector of the user query in this interaction;

[0047] For fields In the Semantic vector of the interaction content.

[0048] Preferably, the normalization factor The calculation formula is:

[0049] ,

[0050] in, Indicates the influence coefficient of user feedback on field length normalization.

[0051] Preferably, the relationship graph drawing module draws the relationship graph online in real time based on the graph data structure, including:

[0052] Based on the nodes and edges of the graph data structure, you can draw a relationship diagram online in real time in the browser, freely set the node arrangement, edit nodes and edges through the event mechanism, or adjust the node position by dragging and dropping.

[0053] Preferably, the authority management module is based on a role management mechanism, manages user authority through roles, assigns authority configurations to different roles, and assigns one or more roles to users through batch maintenance, configuring authority according to actual needs.

[0054] Preferably, the platform adopts a B / S architecture and is deployed in multiple operating system environments, and the permissions of multiple operating systems are associated, and can be expanded in different regions or departments according to needs.

[0055] A second aspect of the present invention provides a method for parsing and applying a QC document and its media, comprising:

[0056] Extract content and classify and store QC documents uploaded by users;

[0057] Perform voice and text recognition on media videos uploaded by users and associated with QC documents;

[0058] Use dynamic weighted similarity algorithm to retrieve and display document and video content;

[0059] Draw relationship diagrams online in real time based on graph data structures;

[0060] Through role management, batch maintenance and configuration of user permissions.

[0061] Compared with the prior art, the beneficial effects of the present invention include at least:

[0062] (1) Openness: The analytical application platform of the present invention is based on an open technology architecture and has good scalability and portability. The platform can be deployed in a variety of operating system environments, such as Linux or Windows, supports permission association, and can be flexibly expanded to different regions or departments according to needs. The system design allows for integration with other business systems or data platforms, facilitating subsequent functional expansion and docking with third-party systems.

[0063] (2) Practicality and friendliness: The analytical application platform of the present invention adopts a B / S architecture, has a simple and intuitive interface, is user-friendly, supports simultaneous operation by multiple users, and is highly practical.

[0064] (3) Compatibility and advancement: The analysis application platform of the present invention adopts advanced document analysis and speech recognition technology, which can process a variety of document formats and audio and video formats. The platform has strong compatibility and supports seamless connection with the company's existing content management systems, databases, search engines and other tools, making it convenient to share and call data with other business systems. At the same time, the platform has advanced functions such as audio-to-text and full-text retrieval, which greatly improves the efficiency of information management.

[0065] (4) Efficiency: The parsing application platform of the present invention is capable of efficiently processing large-scale QC documents and multimedia attachments. Through batch uploading and parallel parsing, the platform can quickly extract text, images, and audio content, and display the parsing progress in real time, reducing user waiting time. At the same time, the retrieval module supports precise keyword-based searches, helping users quickly find required documents and related multimedia content. This efficient processing capability significantly improves the speed and accuracy of document management.

[0066] (5) Security: The analytical application platform of the present invention has powerful security management capabilities, supports user permission control and encryption technology, and ensures the security of documents and media content. The platform's access control policy can be customized according to company needs to ensure that only authorized users can access and operate. At the same time, the platform can also set up system modules to support logging and audit processes to ensure transparency and security compliance of data use. It is particularly suitable for quality control scenarios involving sensitive information.

[0067] (6) Dynamic weighted similarity algorithm: Dynamic factors and normalization factors are introduced. The dynamic factor can dynamically adjust the weight based on the importance of the field in the global and local contexts. The normalization factor solves the impact of field length on similarity, so that long fields will not have too high similarity scores due to the large number of words, which can significantly improve the accuracy of search results. In addition, the user behavior feedback adjustment factor is applied to the dynamic factor and the normalization factor, which takes into account the impact of user behavior feedback on weight distribution and the impact of user behavior feedback on the trade-off of field length in similarity calculation.

[0068] Intelligent semantic matching: Retrieval is no longer limited to simple keyword matching, but is able to understand the contextual semantics of the text to achieve more accurate content matching.

[0069] Efficient full-text retrieval: The combination of NLP and Elasticsearch ensures the system's high efficiency and can quickly respond to user queries in large-scale documents and multimedia data.

[0070] Improved user experience: The system not only displays the retrieved documents, but also intuitively displays matching paragraphs and related images, allowing users to quickly locate the required content.

[0071] (7) The relationship diagram drawing module supports users to draw relationship diagrams directly in the browser without downloading additional software, which greatly facilitates use; the module supports automatic definition of layout, and users can also freely set the node arrangement method, which enhances the flexibility of drawing; through the event mechanism, users can easily edit nodes and edges by right-clicking, or adjust the node position by dragging, which improves the user experience.

[0072] (8) The permission management module manages user permissions through roles and assigns permission configurations to different roles, which simplifies the permission management process and improves management efficiency. It has batch maintenance capabilities, allowing users to quickly assign one or more roles to users, facilitating batch permission maintenance and reducing duplication of work and the possibility of errors. Users can flexibly configure permissions based on actual needs to ensure that users can only access the functions they need, thereby enhancing the security and controllability of the platform.

[0073] (9) The document parsing module can accurately locate the location of image data and automatically extract key information, such as the image title. It can also identify the attributes and type of the image based on the image title and associate the document paragraph to which the image belongs. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 Schematic diagram of the QC document and media analysis application platform architecture of the present invention;

[0075] Figure 2This is a schematic diagram of the document parsing and retrieval business process of the present invention;

[0076] Figure 3 This is a schematic diagram of the video text recognition and retrieval business process of the present invention;

[0077] Figure 4 This is the retrieval process of integrating NLP into Elasticsearch of the present invention. DETAILED DESCRIPTION

[0078] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. The embodiments described in this application are only part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, other embodiments obtained by ordinary technicians in this field without making creative efforts are all within the scope of protection of the present invention.

[0079] like Figure 1 As shown, embodiment 1 of the present invention provides an application platform for parsing QC documents and their media, which solves the problems of low efficiency, information isolation, and difficulty in retrieval in existing QC document and multimedia content parsing. Specifically, it includes a document parsing module, a video text recognition module, a media voice recognition module, a retrieval module, a relationship diagram drawing module, and a rights management module.

[0080] The document parsing module is used to extract content and store word segments of the QC documents uploaded by users;

[0081] More preferably, Figure 2 As shown, the document parsing module is primarily used to automatically extract text and image content from user-submitted documents using document content extraction technology combined with OCR recognition technology, and store it in a database and Elasticsearch. Text content and image data are categorized, stored, and indexed according to the document structure to facilitate subsequent retrieval and management. Compared to traditional image recognition and content extraction methods, this platform's document parsing module can also accurately locate image data, automatically extract key information such as image titles, identify image attributes and types based on the image title, and associate the document paragraph to which the image belongs.

[0082] Furthermore, the document parsing module supports multi-file uploads. Users can upload multiple PDF documents or compressed packages containing multiple documents, supporting various document formats (such as PDF and Word). Using OCR (optical character recognition) technology, once the document is uploaded, the system stores it on the server and begins automatic parsing. Using PDF and other parsing tools, the system can extract text paragraphs and images from each document, accurately identify the text within the images, and store them alongside the rest of the document. The parsed text and images are then stored in a database and Elasticsearch for subsequent full-text retrieval and information tracking.

[0083] Furthermore, the document parsing module includes content classification and annotation functions, which can automatically identify document structures such as titles, paragraphs, and images, and store and index them according to different categories, providing more accurate data support for subsequent retrieval. Image data is also stored and indexed according to the document structure.

[0084] The document parsing module also displays the parsing progress in real time and provides an exception notification feature. When an anomaly occurs during parsing, the system immediately displays the anomaly, helping users quickly locate and resolve the issue, ensuring a smooth parsing process. Users can view the real-time progress of document parsing and clearly understand the processing status of each document, avoiding the long wait times associated with large-scale file uploads.

[0085] The media voice recognition module and the video text recognition module are used to perform voice and text recognition on the media videos uploaded by the user and associated with the QC documents respectively;

[0086] More preferably, Figure 3 As shown, the media speech recognition module and the video text recognition module are mainly used to extract audio content from video files through audio and video analysis technology, and perform speech-to-text processing on the audio, and finally store the transcribed text content in Elasticsearch for subsequent full-text retrieval.

[0087] Users can upload multiple multimedia files (audio and video) associated with QC documents. The uploaded media files will be stored in the system and associated with the corresponding QC documents;

[0088] Audio Extraction: Use FFmpeg to extract audio files from videos. It supports parsing and processing of various video formats (such as MP4, AVI, etc.). The extracted audio files will be further identified and processed.

[0089] Audio-to-text: The extracted audio files are transcribed using a speech recognition model (such as ASR, automatic speech recognition). The system converts the spoken content in the audio into structured text. This recognition process supports recognition of multiple languages ​​and industry terms, ensuring the accuracy of the text content.

[0090] Furthermore, the video text recognition module uses machine learning to optimize recognition accuracy when converting audio to text, incorporating scenario-specific language models to enhance the ability to recognize specialized terminology or complex audio. The system extracts key information based on the document structure, automatically identifying and tagging keywords, document titles, authors, groups, and other information. The speech-to-text conversion results are stored in a database and Elasticsearch for subsequent full-text retrieval.

[0091] Data association: Each audio transcription result will be automatically associated with the corresponding QC document, ensuring that users can retrieve relevant multimedia content through QC document information, facilitating subsequent retrieval of media attachments through QC project information.

[0092] The retrieval module is used to retrieve and display documents and video content using a dynamic weighted similarity algorithm;

[0093] Further preferably, the retrieval module provides users with a powerful full-text search function, allowing them to quickly find QC documents and their multimedia attachments by keywords.

[0094] Full-text search and precise display: Users can enter keywords to search, and the system will match the text content, image descriptions, and audio transcripts of QC documents, displaying documents and multimedia attachments containing relevant content. Specifically, by integrating Elasticsearch's natural language processing (NLP) algorithms with keyword-based full-text search technology, the accuracy and intelligence of document retrieval are improved. Based on NLP's text analysis capabilities, the system can semantically understand the content in documents, enabling searches not only based on simple keyword matching but also enabling semantically based intelligent search.

[0095] NLP retrieval in Elasticsearch relies on text embedding and semantic matching technologies. The main process includes the following steps:

[0096] 1) Text preprocessing: Preprocess the text content and generate semantic embeddings using deep learning models (such as BERT, RoBERTa, and other pre-trained models). Preprocessing includes stemming, stop word removal, and synonym expansion. Extract key document information, including the title, author, company, QC team, and paragraphs. Specifically, this includes:

[0097] Perform unified preprocessing on all parsed content, including the following steps:

[0098] Stemming: Remove the inflections of words (such as plurals, tense changes) through the stemming algorithm of NLTK.

[0099] Lemmatization: Use a dictionary library such as WordNet to restore words to their standard forms and unify the expressions.

[0100] Stop word filtering: Delete meaningless stop words (such as "in", "is", etc.) to reduce noise.

[0101] Synonym expansion: With the help of Word2Vec or a pre-trained word vector model, generate a list of synonyms for the query word to expand the query coverage. The formula is as follows:

[0102] ,

[0103] where is the vector representation of the word , is the vector representation of the candidate synonym. Determine the synonym relationship by calculating the cosine similarity. is the threshold, is the cosine function.

[0104] 2) Vectorization processing:

[0105] NLP models (such as BERT, Word2Vec, etc.) convert sentences or paragraphs in the document into numerical vector representations. The text of each document is converted into a vector (embedding) of a fixed dimension by the model to capture the semantic relationships between words.

[0106] Specifically, use the BERT pre-trained language model to generate the semantic vector of the text:

[0107] Sentence vector generation: Input each paragraph into the model to obtain the vector representation of each word ;

[0108] Use the CLS vector or the average vector to generate the embedding representation of the paragraph or the embedding representation of the sentence :

[0109] ,

[0110] Vector storage: The generated vectors are stored in a vector database such as Elasticsearch to provide a basis for subsequent similarity retrieval.

[0111] Achieving multi-dimensional semantic understanding improves the intelligence of text matching and makes search results more semantically relevant.

[0112] 3) Similarity calculation:

[0113] After the semantic vector is generated, it is matched by weighted cosine similarity. The weight is dynamically adjusted according to the importance of different fields, and a dynamic factor C is introduced. i This factor dynamically adjusts the weight based on the importance of the field in the global and local context to improve the personalization and accuracy of search results:

[0114] Query vector generation: After the user enters a query, the same BERT model is used to convert the query text into a query vector .

[0115] Weighted cosine similarity calculation: weights of weighted cosine similarity This is usually determined in the following ways to ensure that certain specific fields (such as titles, keywords, etc.) are given higher priority in similarity calculations.

[0116] Specific weight setting methods include manual setting based on domain knowledge, weight determination based on statistical analysis, and dynamic weight optimization based on machine learning.

[0117] Manual weight setting in the early stage: You can combine domain knowledge and statistical analysis to manually set weights to lay a good foundation for the model. The following is an example of manual setting:

[0118] ,

[0119] Statistical weighting: TF-IDF (Term Frequency-Inverse Document Frequency) analysis can be used to measure the weight of different fields. For example, a title may have a low term frequency but high information density, and the keyword has a high IDF value, so it is assigned a higher weight.

[0120] ,

[0121] Mature Machine Learning: This approach uses machine learning to automatically adjust the weights of different fields, ensuring that the weighted cosine similarity score better reflects users' real needs. Once the system receives user interaction data (such as clicks and likes), it can train field weights using machine learning models (such as LightGBM and XGBoost). The goal is to improve the accuracy of the system's recommendations by optimizing weights.

[0122] The training steps are: collect user query and click data; use field similarity scores as features and click-through rate or user feedback as target variables; optimize field weights through model training to obtain the optimal weight combination .

[0123] The weights determined by the above method can enable the system to more accurately capture the semantic intent of user queries, thereby obtaining better similarity calculation results in multiple content types.

[0124] Furthermore, the present invention preferably uses a dynamic weighted similarity algorithm to assign different weights to different fields (such as title, author, content, etc.) to highlight the matching degree of the query field. The calculation formula is as follows:

[0125] ,

[0126] in: Field weights are adjusted dynamically by manual or machine learning algorithms;

[0127] is the dynamic weight adjustment factor, which is calculated by the following formula:

[0128] ,

[0129] : The global importance of the field in the entire document collection, based on the TF-IDF score of the field content, calculated using the following formula: ,

[0130] in, For fields The TF-IDF score in the document.

[0131] : The local relevance of the field in the current document, based on the field's content density (word frequency) or semantic similarity score, calculated using the following formula:

[0132] ,

[0133] in, and Respectively represent Semantic vectors of query terms and field content terms.

[0134] It is the user behavior feedback adjustment factor, which directly participates in the context dynamic weight adjustment and reflects the importance of user behavior to the field. It is calculated as follows:

[0135] ,

[0136] : Field The importance score of the hth interaction in the user's historical behavior record (based on indicators such as click-through rate, dwell time, conversion rate, etc.). The calculation formula is:

[0137] ,

[0138] in, For the Fields in Interaction The probability of the corresponding content being clicked; For the Fields in Interaction Average dwell time on the corresponding content; For the From the field The proportion of actions that lead to download, open, or preview; is a weight parameter that can be determined through experiments or machine learning methods.

[0139] : Field in the hth interaction in the user's historical behavior record The correlation (which can be calculated by semantic similarity) is calculated as follows: ,

[0140] in, For the The semantic vector of the user query in this interaction; For fields In the Semantic vector of the interaction content. : Cosine similarity of semantic vectors.

[0141] : The total number of samples in the user's history.

[0142] : Weight adjustment parameter ( ), determined through experiments or automatic optimization.

[0143] is a normalization factor used to address the impact of field length on similarity, so that long fields will not have too high similarity scores due to the large number of words; i Indicates the number of word segments of field i, Indicates the influence coefficient of user feedback on field length normalization, Represents the user behavior feedback adjustment factor, which is used to amplify or suppress the impact of field length on similarity calculation.

[0144] The user behavior feedback adjustment factor Ui is applied to the normalization factor to reflect the correction effect of user feedback on field length, so that user behavior feedback not only affects the weight distribution, but also indirectly affects the trade-off of field length in similarity calculation.

[0145] When field normalization or weight optimization is not required, , , we can return to the original formula, where Manually set initial weight for field i.

[0146] The formula is explained as follows:

[0147] Assumptions is the document to be retrieved, including fields, the vector representation of each field is considered when calculating the weighted similarity:

[0148] ,

[0149] in: For text content Middle field The weight of For fields Dynamic factors; For text content The number of fields in ; For query conditions Text content field No. The query word vector that matches the word vector; For fields The number of word vectors in ; For fields Middle word vectors; is the normalization factor.

[0150] Weight optimization: Optimize weights based on user feedback or using supervised learning algorithms (such as gradient descent) to adapt the system to user search needs and improve search accuracy.

[0151] By calculating the similarity score, the retrieval module can provide users with the document content that best matches the query semantics, and obtain the image data associated with the document content and the text content transcribed from the audio content.

[0152] 4) Result display and sorting:

[0153] For QC documents, search results will display paragraphs matching the keywords according to defined document weighting rules, allowing users to quickly locate the information they need. For media videos, the text content transcribed from the audio content associated with the QC document that best matches the keyword or sentence is displayed.

[0154] The system sorts search results based on similarity scores, prioritizing the most similar documents. It also displays matching paragraphs and related images within specific documents, enhancing the user experience.

[0155] The working principle of NLP in ES is as follows Figure 4 Shown, including:

[0156] The user submits a keyword or sentence query Q.

[0157] The query content is processed by the NLP model to generate a query vector .

[0158] The system converts the content in the document into vectors And stored in ES.

[0159] The similarity between the query vector and the document vector is calculated through dynamic weighted similarity.

[0160] The search results are displayed in order of similarity, including matching document paragraphs and image content.

[0161] By integrating NLP algorithms, the document parsing and retrieval module has the following advantages:

[0162] Intelligent semantic matching: Retrieval is no longer limited to simple keyword matching, but is able to understand the contextual semantics of the text to achieve more accurate content matching.

[0163] Efficient full-text retrieval: The combination of NLP and Elasticsearch ensures the system's high efficiency and can quickly respond to user queries in large-scale documents and multimedia data.

[0164] Improved user experience: The system not only displays the retrieved documents, but also intuitively displays matching paragraphs and related images, allowing users to quickly locate the required content.

[0165] The search module supports full-text search based on audio text, allowing users to quickly search for video-related content by entering keywords. The system will accurately locate media text content containing the keyword and display it in conjunction with the video file, allowing users to jump directly to the relevant video clip for viewing, and also display associated QC information.

[0166] Furthermore, the retrieval module also supports complex query conditions, such as filtering functions based on date, document type, associated media type, etc., to help users obtain the required information more accurately.

[0167] The relationship graph drawing module is used to draw the relationship graph online in real time based on the graph data structure;

[0168] Furthermore, the module supports users to draw a relationship graph online in real time, based on a graph data structure consisting of nodes and edges. The data model is usually represented in the form of a JSON object.

[0169] Browser support: Users can draw relationship diagrams directly in the browser without downloading additional software, which greatly facilitates use.

[0170] Smart layout: The module supports automatic layout definition, and users can also freely set the node arrangement, which enhances the flexibility of drawing.

[0171] Convenient operation mechanism: Through the event mechanism, users can easily edit nodes and edges by right-clicking, or adjust the node position by dragging, which improves the user experience.

[0172] The rights management module is used for role management, batch maintenance and configuration of user rights.

[0173] Further preferably, the rights management module provides a role management mechanism:

[0174] User permissions are managed through roles, and permission configurations are assigned to different roles, which simplifies the permission management process and improves management efficiency.

[0175] Batch maintenance capability: Users can quickly assign one or more roles to users, facilitating batch permission maintenance and reducing duplication of work and the possibility of errors.

[0176] Flexible permission configuration: Users can flexibly configure permissions according to actual needs, ensuring that users can only access the functions they need, enhancing the security and controllability of the platform.

[0177] Embodiment 2 of the present invention provides a method for parsing and applying a QC document and its media, including:

[0178] Extract content and classify and store QC documents uploaded by users;

[0179] Perform voice and text recognition on media videos uploaded by users and associated with QC documents;

[0180] Use dynamic weighted similarity algorithm to retrieve and display document and video content;

[0181] Draw relationship diagrams online in real time based on graph data structures;

[0182] Through role management, batch maintenance and configuration of user permissions.

[0183] Embodiment 3 of the present invention provides a terminal, including a processor and a storage medium; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of the method.

[0184] Embodiment 4 of the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the method when executed by a processor.

[0185] Compared with the prior art, the beneficial effects of the present invention include at least:

[0186] (1) Openness: The analytical application platform of the present invention is based on an open technology architecture and has good scalability and portability. The platform can be deployed in a variety of operating system environments, such as Linux or Windows, supports permission association, and can be flexibly expanded to different regions or departments according to needs. The system design allows for integration with other business systems or data platforms, facilitating subsequent functional expansion and docking with third-party systems.

[0187] (2) Practicality and friendliness: The analytical application platform of the present invention adopts a B / S architecture, has a simple and intuitive interface, is user-friendly, supports simultaneous operation by multiple users, and is highly practical.

[0188] (3) Compatibility and advancement: The analysis application platform of the present invention adopts advanced document analysis and speech recognition technology, which can process a variety of document formats and audio and video formats. The platform has strong compatibility and supports seamless connection with the company's existing content management systems, databases, search engines and other tools, making it convenient to share and call data with other business systems. At the same time, the platform has advanced functions such as audio-to-text and full-text retrieval, which greatly improves the efficiency of information management.

[0189] (4) Efficiency: The parsing application platform of the present invention is capable of efficiently processing large-scale QC documents and multimedia attachments. Through batch uploading and parallel parsing, the platform can quickly extract text, images, and audio content, and display the parsing progress in real time, reducing user waiting time. At the same time, the retrieval module supports precise keyword-based searches, helping users quickly find required documents and related multimedia content. This efficient processing capability significantly improves the speed and accuracy of document management.

[0190] (5) Security: The analytical application platform of the present invention has powerful security management capabilities, supporting user permission control and encryption technology to ensure the security of documents and media content. The platform's access control policies can be customized according to company needs, ensuring that only authorized users can access and operate. At the same time, the platform supports logging and auditing processes to ensure transparency and security compliance of data use, making it particularly suitable for quality control scenarios involving sensitive information.

[0191] (6) Dynamic weighted similarity algorithm: Dynamic factors and normalization factors are introduced. The dynamic factor can dynamically adjust the weight based on the importance of the field in the global and local contexts. The normalization factor solves the impact of field length on similarity, so that long fields will not have too high similarity scores due to the large number of words, which can significantly improve the accuracy of search results. In addition, the user behavior feedback adjustment factor is applied to the dynamic factor and the normalization factor, which takes into account the impact of user behavior feedback on weight distribution and the impact of user behavior feedback on the trade-off of field length in similarity calculation.

[0192] Intelligent semantic matching: Retrieval is no longer limited to simple keyword matching, but is able to understand the contextual semantics of the text to achieve more accurate content matching.

[0193] Efficient full-text retrieval: The combination of NLP and Elasticsearch ensures the system's high efficiency and can quickly respond to user queries in large-scale documents and multimedia data.

[0194] Improved user experience: The system not only displays the retrieved documents, but also intuitively displays matching paragraphs and related images, allowing users to quickly locate the required content.

[0195] (7) The relationship diagram drawing module supports users to draw relationship diagrams directly in the browser without downloading additional software, which greatly facilitates use; the module supports automatic definition of layout, and users can also freely set the node arrangement method, which enhances the flexibility of drawing; through the event mechanism, users can easily edit nodes and edges by right-clicking, or adjust the node position by dragging, which improves the user experience.

[0196] (8) The permission management module manages user permissions through roles and assigns permission configurations to different roles, which simplifies the permission management process and improves management efficiency. It has batch maintenance capabilities, allowing users to quickly assign one or more roles to users, facilitating batch permission maintenance and reducing duplication of work and the possibility of errors. Users can flexibly configure permissions based on actual needs to ensure that users can only access the functions they need, thereby enhancing the security and controllability of the platform.

[0197] (9) The document parsing module can accurately locate the location of image data and automatically extract key information, such as the image title. It can also identify the attributes and type of the image based on the image title and associate the document paragraph to which the image belongs.

[0198] The present disclosure may be a system, method and / or computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0199] A computer-readable storage medium can be a tangible device that can hold and store instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punched card or raised structure in a groove on which instructions are stored, and any suitable combination thereof. As used herein, a computer-readable storage medium is not to be construed as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse passing through a fiber optic cable), or an electrical signal transmitted through an electrical wire.

[0200] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions to be stored in the computer-readable storage medium in each computing / processing device.

[0201] The computer program instructions for performing the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, the state information of the computer-readable program instructions is used to personalize an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), so that the electronic circuit can execute the computer-readable program instructions, thereby implementing various aspects of the present disclosure.

[0202] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A QC document and media analysis application platform, including a document analysis module, a video text recognition module, a media voice recognition module, a retrieval module, a relationship diagram drawing module, and a rights management module, characterized by: The document parsing module is used to extract content from QC documents uploaded by users and to perform classified storage. The media voice recognition module and video text recognition module are used to perform voice and text recognition on media videos uploaded by users and associated with QC documents, respectively. The retrieval module is used to retrieve and display document and video content using a dynamic weighted similarity algorithm. The relationship graph drawing module is used to draw a relationship graph online in real time based on a graph data structure. The permission management module is used to manage roles, batch maintain, and configure user permissions. The calculation formula for dynamic weighted similarity is: ， in: For query conditions and text content Dynamic weighted similarity of; For text content Middle field The weight of For fields Dynamic factors; is the normalization factor; For text content The number of fields in ; For query conditions Text content field No. The query word vector that matches the word vector; For fields Middle word vectors; For fields The number of word vectors in ; Dynamic Factor The calculation formula is: ， ， ， ， in, For fields In the text content The global importance of For fields Local relevance to the query conditions; U i Adjustment factor for user behavior feedback; 、 For fields , Field The word vector in the text content word frequency in ; Respectively query term vectors and fields Middle word vectors; is the cosine function; is the weight adjustment parameter; The field for the hth interaction in the user's historical behavior record Importance rating; The field for the hth interaction in the user's historical behavior record relevance; is the total number of samples in the user's history.

2. The QC document and media analysis application platform according to claim 1, characterized in that: The document parsing module extracts and parses the text content and image data in the QC document by combining document content extraction technology with OCR recognition technology, and stores them in the database and Elasticsearch. The text content and image data are classified, stored and indexed according to the document structure.

3. The QC document and media analysis application platform according to claim 1, characterized in that: The document parsing module displays the document parsing progress in real time and provides an exception prompt function. When an exception occurs during the parsing process, the exception situation is immediately displayed.

4. The QC document and media analysis application platform according to claim 1, characterized in that: The media speech recognition module uses an audio processing tool to extract audio content from the media video; The video text recognition module transcribes the audio content into text content through a speech recognition model, extracts key information according to the document structure, stores it in the database and Elasticsearch, and associates it with the corresponding QC document.

5. The QC document and media analysis application platform according to claim 1, characterized in that: The retrieval module performs full-text retrieval based on the query conditions input by the user through a dynamic weighted similarity algorithm, matches the text content of the corresponding QC document, the text content obtained by transcribing the image data and audio content, and displays the QC documents and media videos containing relevant content.

6. A QC document and media analysis application platform according to claim 1 or 5, characterized in that: The algorithm flow of the dynamic weighted similarity algorithm includes: The query conditions entered by the user and the text content in the QC document are word-vectorized, and then the dynamic weighted similarity between the query conditions and the QC document is calculated based on the word vectors. The text content of the most matching QC document is obtained based on the dynamic weighted similarity, and the text content obtained by transcribing the image data and audio content associated with the text content is obtained from the database and Elasticsearch.

7. The QC document and media analysis application platform according to claim 1, characterized in that: Importance Rating The calculation formula is: ， actually, For the Fields in Interaction The probability of the corresponding content being clicked; For the Fields in Interaction Average dwell time on the corresponding content; For the From the field The proportion of actions that lead to download, open, or preview; is the weight parameter.

8. The QC document and media analysis application platform according to claim 1, characterized in that: Correlation The calculation formula is: ， in, For the The semantic vector of the user query in this interaction; For fields In the Semantic vector of the interaction content.

9. The QC document and media analysis application platform according to claim 1, characterized in that: Normalization factor The calculation formula is: ， in, Indicates the influence coefficient of user feedback on field length normalization.

10. The QC document and media analysis application platform according to claim 1, characterized in that: The relationship graph drawing module draws the relationship graph online in real time based on the graph data structure, including: Based on the nodes and edges of the graph data structure, you can draw a relationship diagram online in real time in the browser, freely set the node arrangement, edit nodes and edges through the event mechanism, or adjust the node position by dragging and dropping.

11. The QC document and media analysis application platform according to claim 1, characterized in that: The authority management module is based on the role management mechanism, manages user authority through roles, assigns authority configuration to different roles, and assigns one or more roles to users through batch maintenance, and configures authority according to actual needs.

12. The QC document and media analysis application platform according to claim 1, characterized in that: The platform adopts a B / S architecture and is deployed in multiple operating system environments. The permissions of multiple operating systems are associated and can be expanded in different regions or departments according to needs.

13. A method for parsing and applying QC documents and media thereof, utilizing the platform according to any one of claims 1 to 12, characterized in that: The method comprises: Extract content and classify and store QC documents uploaded by users; Perform voice and text recognition on media videos uploaded by users and associated with QC documents; Use dynamic weighted similarity algorithm to retrieve and display document and video content; Draw relationship diagrams online in real time based on graph data structures; Through role management, batch maintenance and configuration of user permissions.

Images