Product scheme PPT automatic generation method, system, device and medium

Abstract

The invention provides a product scheme PPT automatic generation method, system and device and a medium, and belongs to the technical field of artificial intelligence. The method comprises the steps that product demand information is collected, an NLP tool is called for entity recognition and relation extraction, and a demand feature vector is generated; based on a MySQL database, retrieval is performed according to the demand feature vector, a structured record set is returned, and an associated document ID is obtained; based on a Milvus vector database, retrieval is carried out according to the demand feature vectors, and an unstructured document fragment set is returned; filtering the unstructured document fragment set according to the document ID, retrieving the unstructured document fragment set by adopting a mixed strategy of a BM25 algorithm and semantic retrieval, comprehensively sequencing retrieval results through a weighted scoring mechanism, and generating a final content fragment set; and calling a PPT template according to user requirements, and filling the final content fragment set into the PPT template to generate a PPT draft file.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence technology, and more specifically relates to a method, system, device and medium for automatically generating product solution PPTs. Background Technology

[0002] In the field of product solution development, the current traditional approach relies entirely on manual writing and formatting: product managers need to retrieve, extract, and integrate information from multiple sources, and then manually draw charts and apply templates. This process is not only inefficient, but the output quality also fluctuates significantly with differences in personnel experience, making it difficult to guarantee the consistency and accuracy of key information such as technical parameters and version numbers, becoming a prominent bottleneck for companies to shorten their time-to-market.

[0003] Currently, while enterprises have accumulated a large amount of knowledge assets, these assets are scattered in the form of PDF technical documents, Word white papers, email records, and heterogeneous ERP / CRM databases, lacking unified indexes and semantic associations. This results in high knowledge retrieval costs and low reuse rates. When market or technological demands change rapidly, it is difficult for humans to complete cross-system and cross-modal information integration in a short period of time, further exacerbating the lag in solution development.

[0004] To alleviate these pain points, general-purpose PPT generation tools such as ChatPPT and Kimi have emerged in the market. However, these tools are mainly geared towards general scenarios, and their underlying large language models lack a deep understanding of the enterprise's private knowledge base. They cannot recognize vertical domain concepts such as model numbers, parameters, and compliance clauses, and the generated content is often vague and may even contain factual errors. At the same time, the preset templates and static rules of general-purpose tools are difficult to adapt to the ever-evolving business logic and brand standards of enterprises, still requiring a lot of manual secondary corrections, and failing to fundamentally solve the contradiction between efficiency and quality.

[0005] Furthermore, the commonly used approach of simply relying on large language models to directly output the entire PPT document carries risks such as formatting deviations, misaligned charts and data, and outdated version information. It also lacks a real-time verification mechanism for structured databases, resulting in insufficient reliability of the output. Therefore, the industry urgently needs an automated PPT generation technology that can deeply integrate enterprise knowledge assets and balance accuracy, efficiency, and scalability. Summary of the Invention

[0006] To address the above problems, the present invention aims to provide a method, system, device, and medium for automatically generating product solution PPTs. Through multi-database integration, semantic parsing, hybrid retrieval, dynamic optimization, and full-process automation technologies, it effectively improves the completeness, accuracy, adaptability, and efficiency of product solution PPT generation.

[0007] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, embodiments of this application provide a method, system, device, and medium for automatically generating product solution PPTs, including: Collect product requirement information, call NLP tools to perform entity recognition and relationship extraction on the product requirement information, and generate requirement feature vectors; Based on a MySQL database storing structured product data, the system performs retrieval according to the requirement feature vector, returns the corresponding set of structured records, and retrieves the associated document ID. Based on the Milvus vector database which stores unstructured product data, the system performs retrieval based on requirement feature vectors and returns a corresponding set of unstructured document fragments. The unstructured document fragment set is filtered based on the document ID, and a hybrid strategy of BM25 algorithm and semantic retrieval is used to retrieve the unstructured document fragment set. The retrieval results are sorted by a weighted scoring mechanism to generate the final content fragment set. Based on user needs, a PPT template is invoked, and the final collection of content snippets is filled into the PPT template to generate a PPT draft file, which is then distributed and displayed.

[0008] In an optional implementation, the step of collecting product demand information and calling NLP tools to perform entity recognition and relationship extraction on the product demand information to generate a demand feature vector includes: Collect product requirement information entered by users in the form of text boxes and drop-down menus using a reactive front-end such as React or Vue.js; Call spaCy or Stanford CoreNLP to perform entity recognition and relation extraction on the input text of product requirement information, and generate requirement feature vectors; The requirement feature vector is sent to the backend service via a RESTful API.

[0009] In an optional implementation, the MySQL database storing structured product data is used to perform a search based on a demand feature vector, returning a corresponding set of structured records and obtaining the associated document ID, including: Based on the MySQL 8.0 relational database, and utilizing SQLAlchemy ORM mapping, SQL queries are executed according to the preset product model, technical parameters, and document version metadata model. The result is a set of structured records R that matches the required feature vector, and the associated document ID is obtained.

[0010] In an optional implementation, the Milvus vector database storing unstructured product data performs a search based on demand feature vectors and returns a corresponding set of unstructured document fragments, including: The demand feature vector is mapped into a high-dimensional vector through a pre-trained embedding model; The HNSW approximate nearest neighbor algorithm is used to retrieve a set V of unstructured document fragments with the top-K similarity to the high-dimensional vector in the Milvus vector database; The unstructured document fragments are PDF or Word document fragments that have been scanned by OCR and intelligently parsed.

[0011] In an optional implementation, the step of filtering the unstructured document fragment set according to the document ID, retrieving the unstructured document fragment set using a hybrid strategy of the BM25 algorithm and semantic retrieval, and generating a final content fragment set by comprehensively ranking the retrieval results through a weighted scoring mechanism includes: Using the document IDs in the structured record set R as the filter key, perform a key-value match on the unstructured document fragment set V to form a candidate subset V0; In the candidate subset V0, the BM25 full-text relevance score and Milvus vector similarity score are calculated in parallel for the text fields of each document fragment. The two scores are then merged according to preset weights and sorted in descending order to generate a comprehensive ranking list. The document fragments of the top N items in the comprehensive sorted list are input into the Elasticsearch engine for secondary verification of synonym expansion and noise filtering. Document fragments below the set relevance threshold are removed, and the final content fragment set C is generated.

[0012] In an optional implementation, the step of calling a PPT template according to user needs, filling the final set of content fragments into the PPT template to generate a PPT draft file, and then distributing and displaying it includes: The PPT scene tags are determined by analyzing user needs, and the corresponding PPT templates are called from the preset template library; Based on the placeholder rules of the PPT template, the structured fields in the final content fragment set C are mapped to title, chart, and table placeholders; The python-pptxSDK is used to automatically generate a PPT draft file P0 containing charts, flowcharts, and technical parameter tables; The PPT draft file P0 is rendered as a binary stream, returned to the front end via a RESTful API interface, and then previewed.

[0013] In an optional implementation, the method further includes: The enterprise knowledge source directory is monitored by the watchdog library. When a new or modified product file is detected, an asynchronous Kafka task is triggered to parse, structure or vectorize the product file and write it to a MySQL database or Milvus database. Real-time feedback from users on search results and PPT draft files is obtained, and a feedback dataset is generated. The Embedding model was tuned based on the feedback dataset using Jupyter Notebook scripts, and the Milvus index was rebuilt.

[0014] Secondly, embodiments of this application also provide a product solution PPT automatic generation system, including: The requirement acquisition module is used to collect product requirement information, call NLP tools to perform entity recognition and relationship extraction on the product requirement information, and generate requirement feature vectors. The structured data retrieval module is used to retrieve data based on the required feature vectors from a MySQL database that stores structured product data, return the corresponding set of structured records, and obtain the associated document IDs. The unstructured data retrieval module is used to retrieve data based on the Milvus vector database, which stores unstructured product data, and return the corresponding set of unstructured document fragments. The hybrid retrieval and result fusion module is used to filter the unstructured document fragment set according to the document ID, retrieve the unstructured document fragment set using a hybrid strategy of BM25 algorithm and semantic retrieval, and generate the final content fragment set by comprehensively ranking the retrieval results through a weighted scoring mechanism. The PPT generation module is used to call PPT templates according to user needs, fill the final collection of content fragments into the PPT template to generate a PPT draft file, and then distribute and display it.

[0015] Thirdly, embodiments of this application also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the product solution PPT automatic generation method described in any of the above descriptions.

[0016] Fourthly, embodiments of this application also provide a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the product solution PPT automatic generation method described in any of the above.

[0017] As can be seen from the above technical solutions, the present invention has the following advantages: The product solution PPT automatic generation method provided in this application uses NLP technology to accurately analyze product requirements and generate feature vectors. It combines MySQL and Milvus databases to achieve efficient retrieval of structured and unstructured data. After optimizing the results through a hybrid strategy of BM25 algorithm and semantic retrieval and weighted scoring, it calls PPT templates to automatically fill in and generate drafts. It also relies on dynamic updates and human feedback mechanisms to continuously optimize the model and index, effectively improving the efficiency, accuracy and relevance of product solution PPT generation, and adapting to the needs of enterprises in multiple scenarios.

[0018] This application achieves a significant improvement in product solution generation efficiency through a deep collaborative mechanism between enterprise knowledge base and large language model, compressing the traditional manual-driven production cycle to an extremely short time; simultaneously activating the value of enterprise knowledge assets and greatly improving knowledge reuse efficiency; effectively controlling content risks through a dynamic compliance verification system and significantly reducing the deviation rate of key information; and continuously optimizing the model's enterprise scenario adaptability by combining human feedback reinforcement learning mechanism, ultimately constructing an intelligent solution production closed loop with self-evolutionary characteristics.

[0019] This application utilizes a MySQL 8.0 database to standardize the storage of structured data such as product models and technical parameters, and combines this with a Milvus vector database to vectorize unstructured documents such as PDFs and Word documents, achieving comprehensive coverage of different types of knowledge within the enterprise. Simultaneously, by leveraging metadata extraction and intelligent document parsing technologies, it transforms scattered Excel spreadsheets and scanned documents into standardized data, solving the problems of chaotic data formats and difficulty in unified access in traditional methods, and providing complete information support for PPT generation.

[0020] This application employs an NLP toolkit to semantically analyze user requirements and extract key entities. A hybrid retrieval strategy combines structured data retrieval from MySQL with vector similarity retrieval from Milvus (HNSW algorithm). The results are then comprehensively ranked using the BM25 algorithm and a weighted scoring mechanism, significantly improving the accuracy of matching requirements with content. Furthermore, document ID-based filtering and Elasticsearch full-text search further reduce irrelevant information interference, ensuring that the generated PPT content closely aligns with user needs, accurately presenting core information such as product features and technical parameters.

[0021] This application relies on a watchdog library to monitor file changes and a scheduled task mechanism to achieve automatic updates to the knowledge base, ensuring that newly added or modified product documents can be promptly structured or vectorized and stored in the database. At the same time, it collects user annotation information through a manual feedback interface, and combines the Jupyter Notebook environment to fine-tune the embedding model and rebuild the Milvus index. This enables the system to continuously adapt to changes in terminology and requirements specific to the enterprise's domain, avoiding content failure due to outdated knowledge or model deviations, and ensuring long-term effectiveness.

[0022] This application automates the entire process, from user input (responsive front-end supports multi-terminal operation) to data retrieval, content integration, and automatic PPT template generation, using technologies such as RESTful API and Python SDK. This reduces repetitive manual tasks like data filtering and layout. In particular, the OCR processing of unstructured documents, the automatic association and fusion of structured and unstructured content, and the automatic layout function of the python-pptx SDK significantly improve the efficiency of product solution PPT generation and reduce reliance on operators' professional skills. Attached Figure Description

[0023] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 A flowchart illustrating the method for automatically generating product solution PPTs for this application.

[0025] Figure 2 A schematic diagram of the structure of the product solution PPT automatic generation system provided for this application.

[0026] Figure 3 A schematic diagram of the structure of the electronic device provided in this application. Detailed Implementation

[0027] The various embodiments of this disclosure will be described more fully in the detailed steps of the automatic generation method for product solution PPTs described below. This disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of this disclosure to the specific embodiments disclosed herein, but rather this disclosure should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of this disclosure.

[0028] In the following, the terms “comprising” or “may include”, which may be used in various embodiments of this disclosure, indicate the presence of the disclosed functions, operations, or elements, and do not limit the addition of one or more functions, operations, or elements. Furthermore, as used in various embodiments of this disclosure, the terms “comprising,” “having,” and their cognates are intended only to indicate a particular feature, number, step, operation, element, component, or combination of the foregoing, and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations of the foregoing, or the possibility of adding one or more combinations of the foregoing.

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Please see Figure 1 The diagram shown is a flowchart of a method for automatically generating a product solution PPT in a specific embodiment. The method includes: S1: Collect product requirement information, call NLP tools to perform entity recognition and relationship extraction on the product requirement information, and generate requirement feature vectors.

[0031] In a specific implementation, a responsive front-end interface developed based on the React or Vue.js framework is first used to provide interactive components such as text boxes and drop-down menus, allowing users to input product requirement information such as target market, product functions, and technical parameters on mobile or desktop devices.

[0032] Then, NLP toolkits such as spaCy or Stanford CoreNLP are called to perform entity recognition (such as extracting product models and technical terms) and semantic relationship extraction (such as the association between functions and parameters) on the input text, generating a structured requirement feature vector.

[0033] Finally, the requirement feature vector is transmitted to the backend service through the RESTful API to realize real-time data interaction between the frontend and the backend, while ensuring cross-platform compatibility of the interface.

[0034] S2: Based on a MySQL database storing structured product data, perform retrieval according to the requirement feature vector, return the corresponding set of structured records, and obtain the associated document ID.

[0035] In this specific implementation, a pre-defined metadata model is constructed based on a MySQL 8.0 relational database. This model includes structured data fields such as product model, technical parameters, and document version, and ORM mapping is implemented using SQLAlchemy to simplify database operations. The pandas library in Python is used to clean (e.g., deduplication and format standardization) structured data such as Excel spreadsheets synchronized from enterprise ERP / CRM systems. Metadata extraction technology is then used to map this data to a pre-defined MySQL table structure, where semi-structured data is stored using MySQL 8.0 JSON type fields.

[0036] After receiving the requirement feature vector, execute an SQL query to match the structured data, return a set of structured records R containing technical parameters, and extract the associated unstructured document IDs from the set.

[0037] In this step, a relational data model is built using the open-source MySQL 8.0 database for structured data (such as product models, technical parameters, document versions, etc.) in the company's product materials. The core task is to transform the company's internal product document information into standardized structured data and automatically import it into the database through metadata extraction technology. For example, for an Excel spreadsheet containing product models, parameter names, and parameter values, the system can use a Python script to parse the spreadsheet content and map it to a preset table structure in the MySQL database.

[0038] Furthermore, this method supports integration with existing enterprise ERP or CRM systems via API interfaces to achieve dynamic synchronization of structured data. To enhance data management flexibility, MySQL 8.0 introduced JSON type fields, allowing the storage of semi-structured data. Technically, Python's pandas library is used for data cleaning, and SQLAlchemy is used to implement ORM (Object-Relational Mapping) operations, ensuring efficient data storage and retrieval.

[0039] S3: Based on the Milvus vector database which stores unstructured product data, it performs retrieval based on the requirement feature vector and returns the corresponding set of unstructured document fragments.

[0040] In a specific implementation, for unstructured data such as PDF technical documents and Word white papers, the plain text content is first extracted using intelligent document parsing technology. Then, for scanned documents, OCR technology is used to recognize the text information in the image. The parsed text is then converted into a high-dimensional vector using Milvus's built-in pre-trained embedding model and stored in the Milvus vector database.

[0041] After receiving the required feature vector, it is mapped to a high-dimensional vector using the same embedding model. The HNSW approximate nearest neighbor algorithm is used to retrieve the top-K unstructured document fragments with similarity to the vector in Milvus, forming a set of unstructured document fragments V. The retrieval process is implemented through database read and write operations using MilvusPythonSDK.

[0042] For example, when a user initiates a query, the query statement is also converted into a vector, and the similarity between vectors is calculated using algorithms such as HNSW (Hierarchical Navigable Small World) to quickly locate the most relevant document fragments. For instance, when a user searches for "design of 5G communication modules for industrial-grade routers," PDF documents containing keywords such as "5G module" and "industrial-grade" can be retrieved from Milvus, and the corresponding paragraphs can be returned. Technically, read and write operations on the vector database are implemented based on Milvus's Python SDK, and a RESTful API interface is built for other modules to call.

[0043] S4: Filter the unstructured document fragment set according to the document ID, use a hybrid strategy of BM25 algorithm and semantic retrieval to retrieve the unstructured document fragment set, and generate the final content fragment set by comprehensively ranking the retrieval results through a weighted scoring mechanism.

[0044] In a specific implementation, the obtained document ID is used as a filtering condition. Key-value matching is performed on the generated set of unstructured document fragments V to filter out the associated candidate subset V0. In the candidate subset V0, the BM25 full-text relevance score and Milvus vector similarity score of each document fragment are calculated in parallel based on keyword matching and semantic matching. The two scores are then merged according to a preset weight (e.g., each accounting for 50%) and sorted in descending order to generate a comprehensive ranking list. The top N document fragments in the list are input into the Elasticsearch engine for secondary verification through synonym expansion (e.g., associating "5G module" with "5G communication module") and noise filtering (removing duplicate or low-relevance content), finally generating the final set of content fragments C that meets the relevance threshold.

[0045] For example, if a user needs to query "parameters of the 5G communication module of an industrial-grade router of model X-500", the system can first retrieve the parameter table of X-500 from MySQL, then locate the PDF document stored in Milvus using the document ID, and finally return the parameter table and the detailed description in the document. Technically, a hybrid strategy of the BM25 algorithm and semantic retrieval is adopted, and the search results are comprehensively ranked through a weighted scoring mechanism. Furthermore, this method supports supplementary full-text search functionality through Elasticsearch, further improving the query efficiency of unstructured data.

[0046] S5: Based on user needs, call the PPT template, fill the final collection of content fragments into the PPT template to generate a PPT draft file, and then distribute and display it.

[0047] In a specific implementation, the scenario information in the user's requirements is first parsed (such as "technical solution report" or "product parameter description"), and the corresponding PPT template in the preset template library is called. Then, according to the placeholder rules of titles, charts, tables, etc. in the template, the structured parameters (such as technical parameter tables) in the final content fragment set C are mapped to table placeholders, and the unstructured descriptions (such as functional descriptions) are mapped to text placeholders.

[0048] Finally, the python-pptxSDK is called to automatically generate a PPT draft file P0 containing a technical parameter table, flowchart, and document summary. After rendering it as a binary stream, it is returned to the front end through the RESTful API interface, allowing users to preview and display it in a responsive interface.

[0049] In one embodiment of the present invention, the method further includes a dynamic knowledge base update and human feedback mechanism: First, the enterprise knowledge source directory is monitored in real time using Python's watchdog library. When a new or modified product file (such as an updated technical document or Excel parameter table) is detected, an asynchronous task is triggered in the Kafka message queue to perform structured parsing (suitable for tabular data) or vectorization (suitable for document data) on the file, and write it to a MySQL database or Milvus database to update the knowledge base. User feedback on search results (such as "relevant" and "irrelevant" labels) and PPT drafts is collected in real time to generate a feedback dataset. Technical personnel use Jupyter Notebook scripts to fine-tune the embedding model based on this dataset, optimizing semantic matching accuracy, and triggering Milvus index reconstruction based on the feedback to improve subsequent search efficiency.

[0050] As can be seen, the core functions of this feature include: 1) Automatic updates: Regularly scanning internal knowledge sources through scheduled tasks, parsing newly added or modified documents, and storing them in the database after structuring or vectorizing them; 2) Manual feedback: Users can annotate the generated solutions or search results (e.g., "relevant" or "irrelevant"), and the feedback data is synchronized to the model training module to optimize the embedding model and retrieval algorithm. For example, if a user finds that a search result does not contain a key document, they can manually mark the document and submit it to the system, triggering the reconstruction of the Milvus index. Technically, Python's watchdog library is used to monitor file changes, and asynchronous task processing is achieved through a Kafka message queue. In addition, the system integrates a Jupyter Notebook environment, allowing technical personnel to fine-tune the embedding model through Notebook scripts to adapt to the terminology and expression habits specific to the enterprise's domain.

[0051] In this embodiment, user needs are accurately analyzed through a responsive front-end and NLP technology. Structured and unstructured data are efficiently managed using MySQL and Milvus databases respectively. A hybrid retrieval strategy (HNSW algorithm, BM25 algorithm) and Elasticsearch secondary validation are combined to improve content matching accuracy. PPTs are automatically generated and displayed on multiple platforms using python-pptxSDK. At the same time, through watchdog monitoring, asynchronous updates with Kafka, and manual feedback to fine-tune the model, the cross-platform adaptability, information integrity, content accuracy, and generation efficiency of the product solution PPT generation are fully guaranteed, effectively reducing labor costs and supporting continuous system optimization and upgrades.

[0052] In one embodiment of the present invention, based on step S4, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0053] S401, Data Preparation: Receive the data set generated by the preceding steps: The structured record set R contains doc_id, product model, technical parameters, etc. for each line.

[0054] The unstructured fragment set V: each row contains doc_id, chunk_id, raw text, and Milvus vector similarity score score_v.

[0055] To ensure subsequent data alignment, first build R into a HashSet with doc_id as the key, denoted as Ω.

[0056] S402, Hard Constraint Filtering: Traverse V, keeping only the rows where doc_id ∈ Ω, to obtain the candidate subset V0.

[0057] This sub-step narrows the search scope from the "entire database" to "documents related to product models," reducing computation by 70–90%.

[0058] S403, BM25 Full Text Rating: Create an inverted index in memory for the raw_text of the candidate subset V0 (using Elasticsearch's BM25 implementation, with parameters k1=1.2 and b=0.75).

[0059] The demand feature vector is converted into a keyword list Q as the query. For each segment in V0, the BM25 score score_b is calculated and linearly normalized to [0,1], denoted as score_b′.

[0060] S404, Weighted Fusion: The overall score for each segment is recalculated using the following formula: score_f = α·score_v + (1-α)·score_b′ In this context, α is set to 0.6 by default.

[0061] After the calculation is completed, sort the results in descending order of score_f to generate a preliminary sorted list L.

[0062] S405, Elasticsearch secondary verification: Take the first N records of L and submit their raw_text values ​​to the Elasticsearch index: First, execute the match_phrase query to verify a phrase-level match with Q; then use the must_not clause to remove segments containing low-value identifiers such as "disclaimer", "table of contents", and "revision history". Finally, the new relevance scores (score_es) are returned by Elasticsearch, and then the results are reordered in descending order by score_es, keeping only the top K results to obtain the refined list L′.

[0063] S406. Assemble the final set of content fragments C: First, perform the following operations on each record in the refined list L′: If doc_id comes from R, then directly extract the corresponding technical parameters; If the data comes from V, then use regular expressions / rules to extract numerical values, chart descriptions, or process nodes.

[0064] Then, the extracted data is uniformly encapsulated into the following JSON structure to generate the final content fragment collection C: {source,doc_id,chunk_id,title,body,params,chart_data,score_f}.

[0065] Finally, the final collection of content fragments C is compressed and written to Redis, and the metadata is written to a MySQL audit table to ensure traceability.

[0066] like Figure 2 As shown, the following are embodiments of the product solution PPT automatic generation system provided in this disclosure. This system and the product solution PPT automatic generation method of the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the product solution PPT automatic generation system, please refer to the embodiments of the product solution PPT automatic generation method.

[0067] A product proposal PPT automatic generation system includes: The requirement gathering module is used to collect product requirement information, call NLP tools to perform entity recognition and relationship extraction on the product requirement information, and generate requirement feature vectors.

[0068] The structured data retrieval module is used to retrieve data based on the required feature vectors from a MySQL database that stores structured product data, return the corresponding set of structured records, and obtain the associated document ID.

[0069] The unstructured data retrieval module is used to retrieve data based on the required feature vectors from the Milvus vector database, which stores unstructured product data, and return the corresponding set of unstructured document fragments.

[0070] The hybrid retrieval and result fusion module is used to filter the set of unstructured document fragments based on the document ID, retrieve the set of unstructured document fragments using a hybrid strategy of BM25 algorithm and semantic retrieval, and generate the final set of content fragments by comprehensively ranking the retrieval results through a weighted scoring mechanism.

[0071] The PPT generation module is used to call PPT templates according to user needs, fill the final collection of content fragments into the PPT template to generate a PPT draft file, and then distribute and display it.

[0072] The product solution PPT automatic generation system provided in this embodiment efficiently integrates structured and unstructured data through MySQL and Milvus databases. It solves the problem of data format disorder by using metadata extraction and document parsing technology to ensure information integrity. It achieves accurate content matching by combining NLP semantic parsing, hybrid retrieval strategies and weighted scoring mechanisms to reduce irrelevant interference. It enhances the long-term adaptability of the system by relying on automatic update mechanism and human feedback to optimize the model. The fully automated processing reduces labor costs and operation threshold, and significantly improves the efficiency and content quality of product solution PPT generation.

[0073] Figure 3 A schematic diagram of the hardware structure of an electronic device for implementing various embodiments of the present invention.

[0074] The product solution PPT automatic generation method provided in this application embodiment can be applied to electronic devices. Those skilled in the art will understand that the electronic device structure involved in the embodiments of this invention does not constitute a limitation on the electronic device. An electronic device may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. In the embodiments of this invention, electronic devices include, but are not limited to, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of this application described and / or claimed herein.

[0075] Electronic devices may include processors, external memory interfaces, internal memory, universal serial bus (USB) interfaces, charging management modules, power management modules, batteries, wireless communication modules, audio modules, speakers, microphones, sensor modules, buttons, cameras, displays, and SIM card interfaces, etc.

[0076] A processor may include one or more processing units, such as: a central processing unit (CPU), an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, memory, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural network processing unit (NPU). Different processing units may be independent devices or integrated into one or more processors.

[0077] The processor can serve as the nerve center and command center of an electronic device. The controller can generate operation control signals based on the instruction opcode and timing signals to control the fetching and execution of instructions.

[0078] The processor may also include memory for storing instructions and data. In some embodiments, the memory in the processor is a cache memory. This memory can store instructions or data that the processor has just used or that are used repeatedly. If the processor needs to use the instruction or data again, it can retrieve it directly from this memory. This avoids repeated accesses, reduces processor latency, and thus improves system efficiency.

[0079] An external storage interface (ESI) can be used to connect external memory cards, such as microSD cards, to expand the storage capacity of electronic devices. The external memory card communicates with the processor through the ESI to perform data storage functions, such as saving music and video files on the external memory card.

[0080] Internal memory can be used to store computer executable program code, which includes instructions. The processor executes various functional applications and data processing of electronic devices by running the instructions stored in internal memory. Internal memory can include a program storage area and a data storage area. Internal memory can include high-speed random access memory, and can also include non-volatile memory, such as at least one disk storage device, flash memory device, universal flash storage (UFS), etc.

[0081] Wireless communication functionality in electronic devices can be achieved through antennas, wireless communication modules, modem processors, and baseband processors.

[0082] Wireless communication modules can provide solutions for wireless communication applications in electronic devices, including wireless local area networks (WLANs) (such as wireless fidelity (Wi-Fi) networks), Bluetooth (BT), global navigation satellite system (GNSS), frequency modulation (FM), near field communication (NFC), and infrared (IR) technologies.

[0083] Electronic devices can implement audio functions through audio modules, speakers, receivers, microphones, headphone jacks, and application processors.

[0084] Electronic devices can achieve shooting functions through ISPs, cameras, video codecs, GPUs, displays, and application processors.

[0085] Electronic devices can achieve display functions through GPUs, displays, and application processors.

[0086] A GPU is a microprocessor for image processing, connected to the display screen and application processor. GPUs are used to perform mathematical and geometric calculations for graphics rendering. A processor may include one or more GPUs, which execute program instructions to generate or modify display information.

[0087] A display screen is used to display images, videos, etc. A display screen includes a display panel.

[0088] The aforementioned electronic device realizes the automatic generation method of product solution PPT in this application. It integrates structured and unstructured data through MySQL and Milvus databases, achieves accurate matching by means of NLP semantic parsing and hybrid retrieval strategies, and optimizes the model by relying on watchdog monitoring and human feedback. The entire process is automated through technologies such as RESTful API and Python SDK, which achieves the beneficial effects of improving the information completeness, content accuracy, system adaptability and generation efficiency of product solution PPT generation, and reducing labor costs and operation threshold.

[0089] The storage medium provided in this application stores a program product capable of automatically generating product solution PPTs.

[0090] Methods for automatically generating product proposal PPTs include: Collect product requirement information, call NLP tools to perform entity recognition and relationship extraction on the product requirement information, and generate requirement feature vectors; Based on a MySQL database storing structured product data, the system performs retrieval according to the requirement feature vector, returns the corresponding set of structured records, and retrieves the associated document ID. Based on the Milvus vector database which stores unstructured product data, the system performs retrieval based on requirement feature vectors and returns a corresponding set of unstructured document fragments. The unstructured document fragment set is filtered based on the document ID, and a hybrid strategy of BM25 algorithm and semantic retrieval is used to retrieve the unstructured document fragment set. The retrieval results are sorted by a weighted scoring mechanism to generate the final content fragment set. Based on user needs, a PPT template is invoked, and the final collection of content snippets is filled into the PPT template to generate a PPT draft file, which is then distributed and displayed.

[0091] In some possible implementations, the product solution PPT automatic generation method of this disclosure can be implemented as a program product, which includes program code that, when the program product is run on a terminal device, causes the terminal device to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure.

[0092] The storage medium disclosed herein may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0093] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A product plan (PPT) automatic generation method, characterized by, Comprise: Collect product demand information, call NLP tool to product demand information carries out entity recognition and relationship extraction, generates demand feature vector; Based on the MySQL database storing product structured data, according to demand feature vector carries out retrieval, returns corresponding structured record set, and obtains associated document ID; Based on the Milvus vector database storing product unstructured data, according to demand feature vector carries out retrieval, returns corresponding unstructured document fragment set; According to the document ID, the unstructured document fragment set is filtered, the BM25 algorithm and the hybrid strategy of semantic retrieval are used to retrieve the unstructured document fragment set, the retrieval result is comprehensively sorted through the weighted scoring mechanism, and the final content fragment set is generated; According to user demand, call PPT template, fill the final content fragment set into PPT template to generate PPT draft file, and issue and show.

2. The product plan (PPT) automatic generation method according to claim 1, characterized in that, The collection product demand information, call NLP tool to product demand information carries out entity recognition and relationship extraction, generates demand feature vector, comprising: Collect product demand information through the React or Vue.js responsive front end in the form of text box, drop-down menu; Call spaCy or Stanford CoreNLP to carry out entity recognition and relationship extraction on the input text of product demand information, and generate demand feature vector; The demand feature vector is sent to the backend service through RESTfulAPI.

3. The product plan (PPT) automatic generation method according to claim 2, characterized in that, The MySQL database storing product structured data is based on, according to demand feature vector carries out retrieval, returns corresponding structured record set, and obtains associated document ID, comprising: Based on MySQL8.0 relational database, using SQLAlchemy ORM mapping, according to the preset product model, technical parameter, document version metadata model, executes SQL query, returns the structured record set R matched with demand feature vector, and obtains associated document ID.

4. The product plan (PPT) automatic generation method according to claim 3, characterized in that, The Milvus vector database storing product unstructured data is based on, according to demand feature vector carries out retrieval, returns corresponding unstructured document fragment set, comprising: The demand feature vector is mapped to a high-dimensional vector through a pre-trained Embedding model; Adopt HNSW approximate nearest neighbor algorithm in Milvus vector database retrieves the unstructured document fragment set V with the similarity Top-K of the high-dimensional vector; Wherein, the unstructured document fragment is the PDF, Word document fragment after OCR scanning and document intelligent analysis.

5. The product plan (PPT) automatic generation method according to claim 4, characterized in that, According to the document ID, the unstructured document fragment set is filtered, the BM25 algorithm and the hybrid strategy of semantic retrieval are used to retrieve the unstructured document fragment set, the retrieval result is comprehensively sorted through the weighted scoring mechanism, and the final content fragment set is generated, comprising: The document ID in the structured record set R is used as the filter key, and the unstructured document fragment set V is executed once key value matching, forming the candidate subset V0; In the candidate subset V0, the BM25 full-text relevance score and the Milvus vector similarity score are calculated in parallel for the text fields of each document segment, the two scores are fused according to the preset weight, and the integrated ranking list is generated in descending order; The top N items of the integrated ranking list are input into the Elasticsearch engine for secondary verification of synonym expansion and noise filtering, and the document segments below the set relevance threshold are removed to generate the final content segment set C.

6. The product plan (PPT) automatic generation method according to claim 5, characterized in that, The PPT template is called according to the user demand, the final content segment set is filled into the PPT template to generate a PPT draft file, and the PPT draft file is issued and displayed, including: Determine the PPT scene label by analyzing the user demand, and call the corresponding PPT template in the preset template library; According to the placeholder rules of the PPT template, map the structured fields in the final content segment set C to the title, chart, and table placeholders; Call the python-pptxSDK to automatically generate a PPT draft file P0 containing charts, flowcharts, and technical parameter tables; Render the PPT draft file P0 into a binary stream, return it to the front end through the RESTful API interface, and preview it.

7. The product plan (PPT) automatic generation method according to claim 6, characterized in that, The method further comprises: Monitor the enterprise knowledge source directory through the watchdog library, trigger a Kafka asynchronous task when detecting that a product file is added or modified, and write the product file into the MySQL database or the Milvus database after analyzing, structuring, or vectorizing the product file; Obtain user feedback information on the search results and the PPT draft file in real time to generate a feedback dataset; Adjust the Embedding model based on the feedback dataset using the Jupyter Notebook script, and rebuild the Milvus index.

8. A product plan (PPT) automatic generation system characterized by comprising: The system adopts the product scheme PPT automatic generation method according to any one of claims 1 to 7; The system comprises: A demand collection module for collecting product demand information, calling NLP tools for entity recognition and relationship extraction on the product demand information, and generating a demand feature vector; A structured data retrieval module for retrieving based on the MySQL database storing product structured data according to the demand feature vector, returning a corresponding structured record set, and obtaining associated document IDs; An unstructured data retrieval module for retrieving based on the Milvus vector database storing product unstructured data according to the demand feature vector, and returning a corresponding unstructured document segment set; A hybrid retrieval and result fusion module for filtering the unstructured document segment set according to the document IDs, using a hybrid strategy of BM25 algorithm and semantic retrieval to retrieve the unstructured document segment set, and generating a final content segment set through a weighted scoring mechanism to comprehensively rank the retrieval results; A PPT generation module for calling a PPT template according to user demand, filling the final content segment set into the PPT template to generate a PPT draft file, and issuing and displaying the PPT draft file.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the steps of the product plan PPT automatic generation method of any one of claims 1 to 7 when executing the program.

10. A storage medium having stored thereon a computer program, characterized in that The computer program implements the steps of the product plan PPT automatic generation method of any one of claims 1 to 7 when executed by the processor.

Images