Project cloud platform data management method and system based on process market

Abstract

The invention relates to the technical field of information management, and discloses a project cloud platform data management method based on a process market, and the method comprises the steps: obtaining process demand information inputted by an enterprise user, and obtaining an enterprise name of the enterprise user; crawling enterprise process information through a web crawler tool, and performing multi-level classification to construct a process market platform; obtaining the business scope and the enterprise scale of the enterprise user according to the enterprise name, and determining the enterprise type of the enterprise user according to the business scope; performing keyword extraction on the process demand information input by the enterprise user to determine a target keyword of the process demand information; and matching a demand flow list in a flow market platform according to the enterprise type, the enterprise scale and the target keyword, and pushing the demand flow list to the enterprise user. According to the method, an intelligent process market platform is constructed, keyword extraction and analysis are carried out on demands of enterprise users, automatic aggregation and accurate matching of process resources are realized, and the process adaptation accuracy is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of information management technology, and in particular to a project cloud platform data management method and system based on a process market. Background Art

[0002] In the traditional project management model, enterprises often face pain points such as repetitive and inefficient process design, severe data silos, and difficulties in cross-platform collaboration. Existing enterprise process management technologies have problems such as scattered process resources, poor adaptability, and data silos. Traditional solutions rely on manual screening or closed process libraries, causing enterprises to spend a lot of time searching for suitable templates in massive Internet resources. In addition, due to the lack of unified classification standards and intelligent matching mechanisms, the process has a low degree of fit with the actual needs of the enterprise. At the same time, key characteristics such as enterprise scale and industry attributes are difficult to effectively integrate during cross-platform collaboration, and data interaction has security risks, which cannot meet the rapid response needs of dynamic business scenarios. For example, when a manufacturing company needs to optimize its production process, it needs to download multiple templates from different platforms and verify them one by one, which is time-consuming and labor-intensive and easily misses better solutions. Summary of the Invention

[0003] The present invention provides a project cloud platform data management method and system based on the process market, builds an intelligent process market platform, performs keyword extraction and analysis on the needs of enterprise users, realizes the automatic aggregation and precise matching of process resources, and significantly improves the accuracy of process adaptation.

[0004] The present invention provides a project cloud platform data management method based on a process market, comprising: Obtaining process requirement information input by an enterprise user, and obtaining the enterprise name of the enterprise user; wherein the process requirement information is text information or graphic information; Crawling enterprise process information through a web crawler tool, and classifying the enterprise process information into multiple levels to build a process market platform; Obtaining the business scope and scale of the enterprise user according to the enterprise name, and determining the enterprise type of the enterprise user according to the business scope; Performing keyword extraction on the process requirement information input by the enterprise user to determine target keywords of the process requirement information; The required process list is matched in the process market platform according to the enterprise type, enterprise scale and target keywords, and the required process list is pushed to the enterprise user.

[0005] Furthermore, the step of crawling enterprise process information by a web crawler tool and performing multi-level classification on the enterprise process information to build a process market platform includes: Crawling a number of enterprise process information using a web crawler tool, and classifying the enterprise process information according to process types; wherein the process types include management processes, business processes, and special field processes, and each process type contains multiple corresponding demand keywords; Classify the enterprise process information of each process type according to the enterprise type, and construct the business scope keywords corresponding to each enterprise type; The enterprise process information of each enterprise type is classified according to the enterprise scale to form a three-level classification table and stored to obtain the process market platform.

[0006] Furthermore, the step of obtaining the business scope and enterprise scale of the enterprise user according to the enterprise name, and determining the enterprise type of the enterprise user according to the business scope includes: Search and extract the business scope and scale of the enterprise user on the enterprise information website according to the enterprise name; Searching the extracted business scope using a business scope keyword corresponding to the enterprise type, and when the corresponding business scope keyword appears in the extracted business scope, determining that the enterprise type corresponding to the business scope keyword is the target enterprise type of the enterprise user; When there are multiple target enterprise types, sending the enterprise user to select one from the multiple target enterprise types as the final target enterprise type; When the target enterprise type is empty, all enterprise types are sent to enable the enterprise user to select one of them as the final target enterprise type, and the business scope keyword corresponding to the enterprise type is updated accordingly.

[0007] Furthermore, the step of extracting keywords from the process requirement information input by the enterprise user to determine target keywords of the process requirement information includes: Extracting text content from the process requirement information, encoding the text content using a text encoding module based on a pre-trained language model and extracting key information therein; Extracting pictures from the process requirement information, using an image encoding module to extract key information from the pictures, and converting the extracted key information into corresponding keyword information in combination with a pre-trained model; Output results of the text encoding module and the image encoding module Perform sorting fusion and output the top-n keyword prediction results; specifically including: The two candidate keyword lists are concatenated and deduplicated. The fused candidate keyword lists are then sorted based on the calculated MR scores, and the top-n results are selected as the final output news keyword prediction result R. The semantic gap between process requirement information and extracted key information is defined as the loss function of the entire model:

[0008] in, The semantic encoding vector representing the text content, represents the encoding vector of the image, Represents a vector concatenation operation, Represents the semantic encoding of the output keyword.

[0009] Furthermore, the step of extracting text content from the process requirement information, encoding the text content using a text encoding module based on a pre-trained language model, and extracting key information therein includes: Extract the text content D in the process requirement information, divide the text content into sentences, and convert the complete text content D into a corresponding sentence set , use the pre-trained language model based on RoBERTa to encode each sentence and obtain the corresponding semantic encoding vector : ; After obtaining the semantic encoding of the sentence, the obtained semantic encoding vector is fused and used as the semantic encoding vector of the news text content. :

[0010] Split the sentence into several words to build a candidate keyword list , to complete the word segmentation, use RoBERTa to semantically encode the words in the candidate word list and obtain the corresponding candidate word encoding vector : ; Semantic encoding vector based on text content and the candidate word encoding vector Perform keyword screening, use the maximum marginal correlation MMR to calculate the relevance of candidate keywords and documents and screen the keyword output results, and define the output keyword table as , for each word in the candidate word list, its corresponding marginal relevance score is defined as:

[0011]

[0012]

[0013] in, Represents the similarity calculation function, represents the weight coefficient, The encoding vector representing the output keyword; Initially, the output keywords are empty, that is, only the similarity between the candidate word semantic vector and the document semantic vector is used as the MR score. The candidate word with the highest MR score is selected and added to the output keyword table. After that, the MR scores of the remaining candidate words are recalculated, and the keyword with the highest MR score is selected and added to the output keyword table. This process is iterated until the predetermined number of keywords are output.

[0014] Furthermore, the step of extracting the image from the process requirement information, extracting key information from the image using an image encoding module, and converting the extracted key information into corresponding keyword information in combination with a pre-trained model includes: Extract the pictures from the process requirement information , use the pre-trained visual encoder CLIP to encode the image, obtain the corresponding image semantic encoding, and map it into a fixed-length encoding vector : ; The FasterRCNN-based target detector is used to identify the subject in the image. When one of the regions g identified in the image is subjected to feature encoding, the identified region is subjected to feature encoding based on semantic representation and position representation; Semantic representation uses the pre-trained visual encoder CLIP to encode the recognition region g and map it into a fixed-length semantic encoding vector : ; The coordinates of the region corresponding to g Perform normalization based on the original image The corresponding length and width data are used to convert the region boundary coordinates to a normalized scale and calculate the information importance weight accordingly. :

[0015] Where width and height are the length and width of the original image respectively. After the mapping is completed, the mapped region coordinates are used as the coordinate features of region g. After completing the semantic coding and position coding, the two codes are concatenated and fused to obtain the feature coding vector for each recognition target. , and input into the decoder to decode the corresponding candidate keywords : ; Similarly, after encoding all the targets detected by the target detector, a list of candidate keywords will be obtained. , the maximum marginal relevance algorithm similar to text encoding is used to screen the obtained keyword list, and the obtained information importance weight is used Calculate MR score:

[0016]

[0017]

[0018] in, Represents the feature vector corresponding to the candidate keyword, express The semantic encoding vector of the corresponding image; In the initial stage, randomly select A candidate keyword in the table is used as the initial seed, and then the MR scores of the remaining candidate words in the table are calculated, and the candidate keyword with the highest score is selected. This process is repeated until the expected number of keywords is obtained.

[0019] Furthermore, the step of matching the required process list in the process market platform according to the enterprise type, enterprise scale and target keywords, and pushing the required process list to the enterprise user includes: Matching the target keyword with the demand keyword to obtain a plurality of initial process information; screening the plurality of initial process information according to the enterprise type to obtain a plurality of screening process information; A plurality of demand processes are extracted from a plurality of screening process information according to the enterprise scale to form a demand process list, and the demand process list is pushed to the enterprise user.

[0020] The present invention also provides a project cloud platform data management system based on a process market, comprising: An acquisition module is used to acquire process requirement information input by an enterprise user and to acquire the enterprise name of the enterprise user; wherein the process requirement information is text information or graphic information; A construction module is used to crawl enterprise process information through a web crawler tool and classify the enterprise process information into multiple levels to build a process market platform; a determination module, configured to obtain the business scope and enterprise scale of the enterprise user according to the enterprise name, and determine the enterprise type of the enterprise user according to the business scope; An extraction module, configured to extract keywords from the process requirement information input by the enterprise user to determine target keywords of the process requirement information; A matching module is used to match the demand process list in the process market platform according to the enterprise type, enterprise scale and target keywords, and push the demand process list to the enterprise user.

[0021] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0022] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0023] The beneficial effects of the present invention are: The present invention realizes the automatic aggregation and precise matching of process resources by constructing an intelligent process market platform. Specifically, it uses web crawler technology to obtain the entire network process information in real time and store it in categories to solve the problem of resource dispersion; based on the company name, it automatically extracts business scope, scale and other characteristics, combines the keyword analysis of process requirements, and builds a multi-dimensional matching model to significantly improve the accuracy of process adaptation; through a dynamic push mechanism, the matching results are actively delivered to corporate users, shortening the screening cycle by more than 50%. In addition, it supports cross-platform data interaction security verification and authority management to ensure the compliance of sensitive corporate information. At the same time, through the process use feedback optimization algorithm, it forms a closed-loop ecology of "demand-matching-optimization", promoting the intelligence and efficiency of enterprise process management. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Schematic diagram of a method flow according to an embodiment of the present invention.

[0025] Figure 2 FIG. 1 is a schematic diagram of the device structure according to an embodiment of the present invention.

[0026] Figure 3 Schematic diagram of the internal structure of a computer device according to an embodiment of the present invention.

[0027] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0028] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0029] like Figure 1 As shown, the present invention provides a project cloud platform data management method based on a process market, comprising the following steps: S1. Obtain process requirement information input by an enterprise user, and obtain the enterprise name of the enterprise user; wherein the process requirement information is text information or graphic information.

[0030] The process requirement information can be a word, a sentence, a paragraph of text or a piece of graphic information. When the process requirement information does not contain image information, only keywords are extracted from the text information, and the keywords are used to match the enterprise process information; when the process requirement information is graphic information, keywords are extracted from the image information and text information respectively, and they are fused to obtain the target keywords, and finally the target keywords are used to match the enterprise process information.

[0031] S2. Crawling enterprise process information through a web crawler tool, and classifying the enterprise process information into multiple levels to build a process market platform.

[0032] Step S2 specifically includes the following steps: S201. Crawling a number of enterprise process information through a web crawler tool, and classifying the enterprise process information according to process types; wherein the process types include management processes, business processes and special field processes, and each process type contains multiple corresponding demand keywords.

[0033] (1) Management process Including strategy and decision-making process, financial management process, human resources management process, administrative management process, and quality management process.

[0034] The demand keywords corresponding to the strategy and decision-making process include strategic planning, market analysis, investment decision-making, risk management, etc. The demand keywords corresponding to the financial management process include budget preparation, cost control, capital management, financial reporting, tax planning, etc. The demand keywords corresponding to the human resources management process include recruitment and appointment, training and development, performance appraisal, salary and benefits, labor relations management, etc. The demand keywords corresponding to the administrative management process include meeting management, document approval, fixed asset management, logistics support, etc. The demand keywords corresponding to the quality management process include ISO certification, quality inspection, problem rectification, customer complaint handling, etc.

[0035] (2) Business Process Including product R&D and production processes, marketing and sales processes, customer service processes, supply chain and logistics processes, and digital transformation related processes.

[0036] The demand keywords corresponding to product R&D and production processes include product design (IPD), production planning, material procurement, supply chain management, quality control, etc. The demand keywords corresponding to marketing and sales processes include market research, brand promotion, lead acquisition (MTL), sales negotiation, order processing (LTC), customer service, etc. The demand keywords corresponding to customer service processes include pre-sales consultation, in-sales support, after-sales maintenance, customer feedback processing (ITR), etc. The demand keywords corresponding to supply chain and logistics processes include supplier management, inventory optimization, warehouse management, logistics distribution, etc. The demand keywords corresponding to digital transformation-related processes include data governance, system integration, process automation (RPA), AI applications, etc.

[0037] (3) Special field processes Including industry-specific processes, compliance and risk control processes, and cross-departmental collaboration processes.

[0038] Key requirements for industry-specific processes include production scheduling, equipment maintenance, and process optimization in the manufacturing industry, and loan approval, anti-money laundering screening, and portfolio management in the financial industry. Key requirements for compliance and risk control processes include audit processes, compliance inspections, risk assessments, and emergency response plans. Key requirements for cross-departmental collaboration processes include cross-departmental project management, resource allocation, and conflict resolution mechanisms.

[0039] S202. Classify the enterprise process information of each process type according to the enterprise type, and construct business scope keywords corresponding to each enterprise type; such as "software development" and "manufacturing", based on which the industry to which it belongs (such as technology or manufacturing) can be determined.

[0040] S203. Classify the enterprise process information of each enterprise type according to enterprise scale (large enterprise, medium-sized enterprise, small enterprise), form a three-level classification table and store it to obtain the process market platform.

[0041] S3. Obtain the business scope and enterprise scale of the enterprise user according to the enterprise name, and determine the enterprise type of the enterprise user according to the business scope.

[0042] Step S3 specifically includes the following steps: S301. Search and extract the business scope and scale of the enterprise user on the enterprise information website according to the enterprise name; for example, search for the business scope and scale of the enterprise user in Enterprise Search or Tianyan Search according to the enterprise name.

[0043] S302: Search the extracted business scope using the business scope keyword corresponding to the enterprise type. When the corresponding business scope keyword appears in the extracted business scope, determine that the enterprise type corresponding to the business scope keyword is the target enterprise type of the enterprise user.

[0044] S303: When there are multiple target enterprise types, the enterprise user is prompted to select one of the multiple target enterprise types as the final target enterprise type. If there are no user-selectable options among the multiple target enterprise types, all enterprise types are displayed for the user to select, and the business scope keywords of the process market platform are updated based on the user's selection.

[0045] S304: When the target enterprise type is empty, all enterprise types are sent to enable the enterprise user to select one of them as the final target enterprise type, and the business scope keyword corresponding to the enterprise type is updated accordingly.

[0046] S4. Perform keyword extraction on the process requirement information input by the enterprise user to determine target keywords of the process requirement information.

[0047] Step S4 specifically includes the following steps: S401: extract text content from the process requirement information, encode the text content using a text encoding module based on a pre-trained language model, and extract key information therein.

[0048] In order to make full use of the relevant key information contained in the text content, the KeyBert method based on the large model is used to extract the keyword information contained therein.

[0049] Extract the text content D in the process requirement information, divide the text content into sentences, and convert the complete text content D into a corresponding sentence set , use the pre-trained language model based on RoBERTa to encode each sentence and obtain the corresponding semantic encoding vector :

[0050] After obtaining the semantic encoding of the sentence, the obtained semantic encoding vector is fused and used as the semantic encoding vector of the news text content. :

[0051] In the standard KeyBert method, in order to better mine the key information that may be contained, the candidate word extraction method based on n-gram is often used, by defining the selection range [range min , range max] to determine the maximum and minimum length of candidate words. However, Chinese is different from languages like English. Compared with languages like English, Chinese has a word structure based on semantics. Directly using the candidate word extraction method based on n-gram may destroy the semantic information contained in the text, thus affecting the prediction results. Therefore, based on the Chinese word segmentation tool to segment the text content, the sentence is split into several words to construct a candidate keyword list. , to complete the word segmentation, use RoBERTa to semantically encode the words in the candidate word list and obtain the corresponding candidate word encoding vector : ; After completing the encoding of candidate keywords, the semantic encoding vector based on the text content and the candidate word encoding vector Perform keyword screening. Considering that the selected keywords should describe the key information contained in the text as non-repetitive as possible, that is, the output keywords should have both relevance and diversity, the maximum marginal relevance (MMR) is used to calculate the relevance between candidate keywords and documents and filter the keyword output results. Define the output keyword table as (Initially, the keyword list is empty.) For each word in the candidate word list, the corresponding marginal relevance score is defined as:

[0052]

[0053]

[0054] in, Represents the similarity calculation function, represents the weight coefficient, Represents the encoded vector of the output keyword.

[0055] In the initial stage, the output keywords are empty, that is, only the similarity between the candidate word semantic vector and the document semantic vector is used as the MR score. The candidate word with the highest MR score is selected and added to the output keyword table. After that, the MR scores of the remaining candidate words are recalculated, and the keyword with the highest MR score is selected and added to the output keyword table. This process is iterated until the predetermined number of keywords are output.

[0056] S402: extracting pictures from the process requirement information, extracting key information from the pictures using an image encoding module, and converting the extracted key information into corresponding keyword information in combination with a pre-trained model.

[0057] Semantic coding methods can directly map image content into high-dimensional semantic vectors and then decode them to obtain corresponding keyword outputs. However, since the entire image is directly encoded, the judgment of key content in the image is easily affected by the overall image information, which may lead to misjudgment. To address this problem, the encoding process is optimized, using an object recognition-based encoding method to filter and encode the main content in the image. The encoding also considers the multi-dimensional feature information of the subject, thereby further improving the ability to mine key information in the image.

[0058] Extract the pictures from the process requirement information , use the pre-trained visual encoder CLIP to encode the image, obtain the corresponding image semantic encoding, and map it into a fixed-length encoding vector :

[0059] The FasterRCNN-based target detector is used to identify the subject in the image. In order to obtain more abundant and effective semantic information from the identified region, the feature encoding of the identified region g is performed from the semantic representation and position representation. The semantic representation uses the pre-trained visual encoder CLIP to encode the identified region g and maps it to a fixed-length semantic encoding vector. :

[0060] Since the identified target subjects have corresponding positional relationships in the image, and this relationship affects both the recognition results and the importance of information, in order to characterize this relationship, the region coordinates corresponding to g are Perform normalization based on the original image The corresponding length and width data are used to convert the region boundary coordinates to a normalized scale and calculate the information importance weight accordingly. :

[0061] Among them, width and height are the length and width of the original image respectively; after the mapping is completed, the mapped area coordinates are used as the coordinate features of area g.

[0062] After completing the semantic coding and position coding, the two codes are concatenated and fused to obtain the feature coding vector for each recognition target. , and input into the decoder to decode the corresponding candidate keywords :

[0063]

[0064] Similarly, after encoding all the targets detected by the target detector, a list of candidate keywords will be obtained. , the maximum marginal relevance algorithm similar to text encoding is used to screen the obtained keyword list, and the obtained information importance weight is used Calculate MR score:

[0065]

[0066]

[0067] in, Represents the feature vector corresponding to the candidate keyword, express The semantic encoding vector of the corresponding image.

[0068] In the initial stage, randomly select A candidate keyword in the table is used as the initial seed, and then the MR scores of the remaining candidate words in the table are calculated, and the candidate keyword with the highest score is selected. This process is repeated until the expected number of keywords is obtained.

[0069] S403: Output results of the text encoding module and the image encoding module Perform sorting fusion and output the top-n keyword prediction results. Specifically including: After the text encoding module and the image encoding module, the candidate keyword lists corresponding to the two modalities are obtained and After inputting the fusion ranking module, the two candidate keyword lists are spliced and deduplicated, and then the fused candidate keyword lists are sorted based on the calculated MR scores, and the Top-n results are selected as the final output news keyword prediction result R.

[0070] In the loss function, the semantic gap between process requirement information and extracted key information is defined as the loss function of the entire model:

[0071] in, The semantic encoding vector representing the text content, represents the encoding vector of the image, Represents a vector concatenation operation, Represents the semantic encoding of the output keyword.

[0072] S5. Match the required process list in the process market platform according to the enterprise type, enterprise scale and target keywords, and push the required process list to the enterprise user.

[0073] Step S5 specifically includes: S501, matching the target keyword with the requirement keyword to obtain a plurality of initial process information; S502: Filter multiple initial process information according to the enterprise type to obtain multiple filtered process information; S503 , extracting multiple demand processes from the multiple screening process information according to the enterprise scale, organizing them to form a demand process list, and pushing the demand process list to the enterprise user.

[0074] The present invention realizes the automatic aggregation and precise matching of process resources by constructing an intelligent process market platform. Specifically, it uses web crawler technology to obtain the entire network process information in real time and store it in categories to solve the problem of resource dispersion; based on the company name, it automatically extracts business scope, scale and other characteristics, combines the keyword analysis of process requirements, and builds a multi-dimensional matching model to significantly improve the accuracy of process adaptation; through a dynamic push mechanism, the matching results are actively delivered to corporate users, shortening the screening cycle by more than 50%. In addition, it supports cross-platform data interaction security verification and authority management to ensure the compliance of sensitive corporate information. At the same time, through the process use feedback optimization algorithm, it forms a closed-loop ecology of "demand-matching-optimization", promoting the intelligence and efficiency of enterprise process management.

[0075] like Figure 2 As shown, the present invention also provides a project cloud platform data management system based on the process market, including: Acquisition module 1 is used to acquire process requirement information input by an enterprise user and acquire the enterprise name of the enterprise user; wherein the process requirement information is text information or graphic information; Construction module 2 is used to crawl enterprise process information through a web crawler tool and classify the enterprise process information into multiple levels to build a process market platform; Determination module 3, used to obtain the business scope and enterprise scale of the enterprise user according to the enterprise name, and determine the enterprise type of the enterprise user according to the business scope; Extraction module 4, used for extracting keywords from the process requirement information input by the enterprise user to determine target keywords of the process requirement information; The matching module 5 is used to match the required process list in the process market platform according to the enterprise type, enterprise scale and target keywords, and push the required process list to the enterprise user.

[0076] In one embodiment, building block 2 includes: A crawling unit, configured to crawl a plurality of enterprise process information using a web crawler tool, and classify the enterprise process information according to process types; wherein the process types include management processes, business processes, and special field processes, and each process type includes a plurality of corresponding demand keywords; A classification unit is used to classify the enterprise process information of each process type according to the enterprise type and construct the business scope keywords corresponding to each enterprise type; The storage unit is used to classify the enterprise process information of each enterprise type according to the enterprise scale, form a three-level classification table and store it to obtain the process market platform.

[0077] In one embodiment, the determination module 3 includes: A query unit, configured to query and extract the business scope and enterprise scale of the enterprise user on the enterprise information website according to the enterprise name; a search unit, configured to search the extracted business scope using a business scope keyword corresponding to the enterprise type, and when the corresponding business scope keyword appears in the extracted business scope, determine that the enterprise type corresponding to the business scope keyword is the target enterprise type of the enterprise user; A first target determination unit is configured to, when there are multiple target enterprise types, send the enterprise user a request to select one from the multiple target enterprise types as a final target enterprise type; The second target determination unit is configured to, when the target enterprise type is empty, send all enterprise types to enable the enterprise user to select one of them as the final target enterprise type, and thereby update the business scope keyword corresponding to the enterprise type.

[0078] In one embodiment, the extraction module 4 includes: A text extraction unit, configured to extract text content from the process requirement information, encode the text content using a text encoding module based on a pre-trained language model, and extract key information therein; An image extraction unit is used to extract images from the process requirement information, extract key information from the images using an image encoding module, and convert the extracted key information into corresponding keyword information in combination with a pre-trained model; A fusion unit is used to combine the output results of the text encoding module and the image encoding module Perform sorting fusion and output the top-n keyword prediction results; specifically including: The two candidate keyword lists are concatenated and deduplicated. The fused candidate keyword lists are then sorted based on the calculated MR scores, and the top-n results are selected as the final output news keyword prediction result R. The semantic gap between process requirement information and extracted key information is defined as the loss function of the entire model:

[0079] in, The semantic encoding vector representing the text content, represents the encoding vector of the image, Represents a vector concatenation operation, Represents the semantic encoding of the output keyword.

[0080] In one embodiment, the text extraction unit includes: Extract the text content D in the process requirement information, divide the text content into sentences, and convert the complete text content D into a corresponding sentence set , use the pre-trained language model based on RoBERTa to encode each sentence and obtain the corresponding semantic encoding vector : ; After obtaining the semantic encoding of the sentence, the obtained semantic encoding vector is fused and used as the semantic encoding vector of the news text content. :

[0081] Split the sentence into several words to build a candidate keyword list , to complete the word segmentation, use RoBERTa to semantically encode the words in the candidate word list and obtain the corresponding candidate word encoding vector : ; Semantic encoding vector based on text content and the candidate word encoding vector Perform keyword screening, use the maximum marginal correlation MMR to calculate the relevance between candidate keywords and documents and screen the keyword output results, and define the output keyword table as , for each word in the candidate word list, its corresponding marginal relevance score is defined as:

[0082]

[0083]

[0084] in, Represents the similarity calculation function, represents the weight coefficient, The encoding vector representing the output keyword; Initially, the output keywords are empty, that is, only the similarity between the candidate word semantic vector and the document semantic vector is used as the MR score. The candidate word with the highest MR score is selected and added to the output keyword table. After that, the MR scores of the remaining candidate words are recalculated, and the keyword with the highest MR score is selected and added to the output keyword table. This process is iterated until the predetermined number of keywords are output.

[0085] In one embodiment, the image extraction unit includes: Extract the pictures from the process requirement information , use the pre-trained visual encoder CLIP to encode the image, obtain the corresponding image semantic encoding, and map it into a fixed-length encoding vector : ; The FasterRCNN-based target detector is used to identify the subject in the image. When one of the regions g identified in the image is subjected to feature encoding, the identified region is subjected to feature encoding based on semantic representation and position representation; Semantic representation uses the pre-trained visual encoder CLIP to encode the recognition region g and map it into a fixed-length semantic encoding vector : ; The coordinates of the region corresponding to g Perform normalization based on the original image The corresponding length and width data are used to convert the region boundary coordinates to a normalized scale and calculate the information importance weight accordingly. :

[0086] Where width and height are the length and width of the original image respectively. After the mapping is completed, the mapped region coordinates are used as the coordinate features of region g. After completing the semantic coding and position coding, the two codes are concatenated and fused to obtain the feature coding vector for each recognition target. , and input into the decoder to decode the corresponding candidate keywords : ; Similarly, after encoding all the targets detected by the target detector, a list of candidate keywords will be obtained. , the maximum marginal relevance algorithm similar to text encoding is used to screen the obtained keyword list, and the obtained information importance weight is used Calculate MR score:

[0087]

[0088]

[0089] in, Represents the feature vector corresponding to the candidate keyword, express The semantic encoding vector of the corresponding image; In the initial stage, randomly select A candidate keyword in the table is used as the initial seed, and then the MR scores of the remaining candidate words in the table are calculated, and the candidate keyword with the highest score is selected. This process is repeated until the expected number of keywords is obtained.

[0090] In one embodiment, the matching module 5 includes: a matching unit, configured to match the target keyword with the required keyword to obtain a plurality of initial process information; a screening unit, configured to screen the plurality of initial process information according to the enterprise type to obtain a plurality of screening process information; The arranging unit is used to extract multiple demand processes from the multiple screening process information according to the enterprise scale, to form a demand process list, and to push the demand process list to the enterprise user.

[0091] The above modules and units are used to execute the corresponding steps in the above process market-based project cloud platform data management method. The specific implementation method is described in the above method embodiment and will not be repeated here.

[0092] like Figure 3 As shown, the present invention also provides a computer device, which can be a server, and its internal structure can be as follows Figure 3As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store all data required for the process of the project cloud platform data management method based on the process market. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the project cloud platform data management method based on the process market is implemented.

[0093] Those skilled in the art will understand that Figure 3 The structure shown in is merely a block diagram of a portion of the structure related to the present application solution and does not constitute a limitation on the computer device to which the present application solution is applied.

[0094] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, any one of the above-mentioned process market-based project cloud platform data management methods is implemented.

[0095] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided in this application and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct RAM bus dynamic RAM (DRDRAM), and RAM bus dynamic RAM (RDRAM).

[0096] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, apparatus, article, or method. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, apparatus, article, or method comprising the element.

[0097] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A project cloud platform data management method based on process market, characterized in that: include: Obtaining process requirement information input by an enterprise user, and obtaining the enterprise name of the enterprise user; wherein the process requirement information is text information or graphic information; Crawling enterprise process information through web crawler tools, and classifying the enterprise process information into multiple levels to build a process market platform; Obtaining the business scope and scale of the enterprise user according to the enterprise name, and determining the enterprise type of the enterprise user according to the business scope; Performing keyword extraction on the process requirement information input by the enterprise user to determine target keywords of the process requirement information; The required process list is matched in the process market platform according to the enterprise type, enterprise scale and target keywords, and the required process list is pushed to the enterprise user.

2. A project cloud platform data management method based on process market according to claim 1, characterized in that: The step of crawling enterprise process information by a web crawler tool and performing multi-level classification on the enterprise process information to build a process market platform includes: Crawling a number of enterprise process information using a web crawler tool, and classifying the enterprise process information according to process types; wherein the process types include management processes, business processes, and special field processes, and each process type contains multiple corresponding demand keywords; Classify the enterprise process information of each process type according to the enterprise type, and construct the business scope keywords corresponding to each enterprise type; The enterprise process information of each enterprise type is classified according to the enterprise scale to form a three-level classification table and stored to obtain the process market platform.

3. A project cloud platform data management method based on process market according to claim 2, characterized in that: The step of obtaining the business scope and enterprise scale of the enterprise user according to the enterprise name, and determining the enterprise type of the enterprise user according to the business scope includes: Search and extract the business scope and scale of the enterprise user on the enterprise information website according to the enterprise name; Searching the extracted business scope using a business scope keyword corresponding to the enterprise type, and when the corresponding business scope keyword appears in the extracted business scope, determining that the enterprise type corresponding to the business scope keyword is the target enterprise type of the enterprise user; When there are multiple target enterprise types, sending the enterprise user to select one from the multiple target enterprise types as the final target enterprise type; When the target enterprise type is empty, all enterprise types are sent to enable the enterprise user to select one of them as the final target enterprise type, and the business scope keyword corresponding to the enterprise type is updated accordingly.

4. A project cloud platform data management method based on process market according to claim 3, characterized in that: The step of extracting keywords from the process requirement information input by the enterprise user to determine target keywords of the process requirement information includes: Extracting text content from the process requirement information, encoding the text content using a text encoding module based on a pre-trained language model and extracting key information therein; Extracting pictures from the process requirement information, using an image encoding module to extract key information from the pictures, and converting the extracted key information into corresponding keyword information in combination with a pre-trained model; Output results of the text encoding module and the image encoding module Perform sorting fusion and output the top-n keyword prediction results; specifically including: The two candidate keyword lists are concatenated and deduplicated. The fused candidate keyword lists are then sorted based on the calculated MR scores, and the top-n results are selected as the final output news keyword prediction result R. The semantic gap between process requirement information and extracted key information is defined as the loss function of the entire model: ； in, The semantic encoding vector representing the text content, represents the encoding vector of the image, Represents a vector concatenation operation, Represents the semantic encoding of the output keyword.

5. A project cloud platform data management method based on process market according to claim 4, characterized in that: The step of extracting text content from the process requirement information, encoding the text content using a text encoding module based on a pre-trained language model, and extracting key information therein includes: Extract the text content D in the process requirement information, divide the text content into sentences, and convert the complete text content D into a corresponding sentence set , use the pre-trained language model based on RoBERTa to encode each sentence and obtain the corresponding semantic encoding vector : ; After obtaining the semantic encoding of the sentence, the obtained semantic encoding vector is fused and used as the semantic encoding vector of the news text content. : ； Split the sentence into several words to build a candidate keyword list , to complete the word segmentation, use RoBERTa to semantically encode the words in the candidate word list and obtain the corresponding candidate word encoding vector : ; Semantic encoding vector based on text content and the candidate word encoding vector Perform keyword screening, use the maximum marginal correlation MMR to calculate the relevance of candidate keywords and documents and screen the keyword output results, and define the output keyword table as , for each word in the candidate word list, its corresponding marginal relevance score is defined as: ； ； ； in, Represents the similarity calculation function, represents the weight coefficient, The encoding vector representing the output keyword; Initially, the output keywords are empty, that is, only the similarity between the candidate word semantic vector and the document semantic vector is used as the MR score. The candidate word with the highest MR score is selected and added to the output keyword table. After that, the MR scores of the remaining candidate words are recalculated, and the keyword with the highest MR score is selected and added to the output keyword table. This process is iterated until the predetermined number of keywords are output.

6. A project cloud platform data management method based on process market according to claim 5, characterized in that: The step of extracting the image from the process requirement information, extracting key information from the image using an image encoding module, and converting the extracted key information into corresponding keyword information in combination with a pre-trained model includes: Extract the pictures from the process requirement information , use the pre-trained visual encoder CLIP to encode the image, obtain the corresponding image semantic encoding, and map it into a fixed-length encoding vector : ; The FasterRCNN-based target detector is used to identify the subject in the image. When one of the regions g identified in the image is subjected to feature encoding, the identified region is subjected to feature encoding based on semantic representation and position representation; Semantic representation uses the pre-trained visual encoder CLIP to encode the recognition region g and map it into a fixed-length semantic encoding vector : ; The coordinates of the region corresponding to g Perform normalization based on the original image The corresponding length and width data are used to convert the region boundary coordinates to a normalized scale and calculate the information importance weight accordingly. : ； Where width and height are the length and width of the original image respectively. After the mapping is completed, the mapped region coordinates are used as the coordinate features of region g. After completing the semantic coding and position coding, the two codes are concatenated and fused to obtain the feature coding vector for each recognition target. , and input into the decoder to decode the corresponding candidate keywords : ; Similarly, after encoding all the targets detected by the target detector, a list of candidate keywords will be obtained. , the maximum marginal relevance algorithm similar to text encoding is used to screen the obtained keyword list, and the obtained information importance weight is used Calculate MR score: ； ； ； in, Represents the feature vector corresponding to the candidate keyword, express The semantic encoding vector of the corresponding image; In the initial stage, randomly select A candidate keyword in the table is used as the initial seed, and then the MR scores of the remaining candidate words in the table are calculated, and the candidate keyword with the highest score is selected. This process is repeated until the expected number of keywords is obtained.

7. A project cloud platform data management method based on process market according to claim 6, characterized in that: The step of matching the required process list in the process market platform according to the enterprise type, enterprise scale and target keywords, and pushing the required process list to the enterprise user includes: Matching the target keyword with the demand keyword to obtain a plurality of initial process information; screening the plurality of initial process information according to the enterprise type to obtain a plurality of screening process information; A plurality of demand processes are extracted from a plurality of screening process information according to the enterprise scale to form a demand process list, and the demand process list is pushed to the enterprise user.

8. A project cloud platform data management system based on process market, characterized in that: include: An acquisition module is used to acquire process requirement information input by an enterprise user and to acquire the enterprise name of the enterprise user; wherein the process requirement information is text information or graphic information; A construction module is used to crawl enterprise process information through a web crawler tool and classify the enterprise process information into multiple levels to build a process market platform; a determination module, configured to obtain the business scope and enterprise scale of the enterprise user according to the enterprise name, and determine the enterprise type of the enterprise user according to the business scope; An extraction module, configured to extract keywords from the process requirement information input by the enterprise user to determine target keywords of the process requirement information; A matching module is used to match the demand process list in the process market platform according to the enterprise type, enterprise scale and target keywords, and push the demand process list to the enterprise user.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images