Opportunity signal correlation analysis method and device based on multi-source heterogeneous data

Abstract

The application discloses a business opportunity and mark information association analysis method and device based on multi-source heterogeneous data, the method generates an original data set containing business opportunity data and mark information data by collecting internal business data and external public data; standardization processing is carried out on the business opportunity data and the mark information data, and a standardized business opportunity data set and a standardized mark information data set are obtained; through entity recognition and disambiguation processing, unified identification results of customer, project, region and industry entities are obtained; based on the unified identification results, a candidate mark information set is recalled for a target business opportunity, and an association feature vector is constructed; a rule and model fusion strategy is used to calculate an association confidence, the candidate mark information is processed in a hierarchical manner, and an association evidence chain is output; manual confirmation is completed according to the hierarchical results, the association calculation process is optimized in combination with the confirmation results and subsequent business results, and finally, business output is completed. The application solves the problems of low association accuracy and recall rate and high labor cost in the prior art.

Description

Technical Field

[0001] This invention relates to the fields of data processing and artificial intelligence technology, specifically to a method and apparatus for analyzing business opportunity information based on multi-source heterogeneous data. Background Technology

[0002] With the deep implementation of enterprise digital transformation and government / enterprise marketing and bidding management, enterprises have generally built multiple business information systems, including opportunity management, customer management, bidding management, contract management, and market intelligence analysis. Opportunity data in the government / enterprise market mainly comes from internal operational processes such as customer visits, lead conversion, and pre-sales support, while bidding information data comes from external public channels such as government procurement websites, public bidding platforms, industry portals, and third-party data platforms. Accurately linking internal opportunities with external bidding information can quickly locate corresponding projects, improve opportunity follow-up efficiency, support opportunity listing, premium bid operation, competitor winning notifications, and full-process project management, and has become a core necessity for government / enterprise marketing and bidding.

[0003] Currently, the association between business opportunities and bidding information mainly relies on three methods: manual retrieval and comparison, single keyword string matching, and a small number of fixed rule judgments. However, business opportunities and bidding information are typical multi-source heterogeneous data, with significant differences in data structure, field naming, text description, time granularity, monetary units, and subject naming. Existing technologies generally suffer from problems such as scattered and unintegrated data, inconsistent standards, and a single matching dimension that cannot utilize multi-dimensional features, making it difficult to effectively identify and disambiguate customers and project entities. At the same time, there is a lack of a correlation confidence level processing mechanism, which can only output binary correlation results, and it is a one-time static matching that does not support dynamic recalculation triggered by new data and business feedback optimization. Ultimately, this results in low correlation accuracy and recall, high manual costs, and an inability to adapt to the precise and automated operation requirements of actual business.

[0004] Therefore, there is an urgent need for a business opportunity information association analysis method based on multi-source heterogeneous data to solve the problems of low association accuracy and recall rate and high labor costs in existing technologies. Summary of the Invention

[0005] To address this issue, the present invention provides a method and apparatus for business opportunity and bidding information association analysis based on multi-source heterogeneous data. This method solves the problems of low association accuracy and recall, and high labor costs when internal business opportunities and external bidding information of government and enterprises are multi-source heterogeneous data, such as inconsistent data standards, single matching methods, difficulty in disambiguating entities, lack of association classification and dynamic optimization mechanisms.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for correlation analysis of business opportunity bidding information based on multi-source heterogeneous data, characterized in that it includes:

[0007] By collecting internal business data and external publicly available data, a raw dataset containing business opportunity data and bidding information data is generated; the internal business data includes business opportunity information and customer master data; the external publicly available data includes bidding announcements and winning bid announcements.

[0008] By standardizing the business opportunity data and the bidding information data in the original dataset, a standardized business opportunity dataset and a standardized bidding information dataset are obtained.

[0009] By performing entity recognition and disambiguation processing on the standardized business opportunity dataset and the standardized bidding information dataset, unified identification results for customer entities, project entities, regional entities, and industry entities are obtained.

[0010] Based on the unified identifier result, a set of candidate bids is recalled from the standardized bid dataset according to the target business opportunity;

[0011] Based on the target business opportunity and the candidate bidding information set, subject matching features, project text features, time features, amount features, regional features, industry business features and process data features are extracted to construct the association feature vector between the target business opportunity and each candidate bidding information.

[0012] Based on the associated feature vector, the association confidence between the target business opportunity and each of the candidate bids is calculated through a rule and model fusion strategy.

[0013] Based on the association confidence level, the candidate tags are subjected to association result classification processing to obtain classification results, and the association evidence chain of the classification results is output.

[0014] Based on the classification results, corresponding processing is performed and a manual confirmation result is generated; based on the manual confirmation result and subsequent business results, the calculation process of the association confidence is optimized, and the business output is completed.

[0015] As a preferred solution for the business opportunity and bidding information association analysis method based on multi-source heterogeneous data, the standardization process for the business opportunity data and bidding information data in the original dataset includes: field mapping, text cleaning, time normalization, amount normalization, region normalization, and industry tag normalization.

[0016] As a preferred embodiment of the business opportunity and bidding information association analysis method based on multi-source heterogeneous data, the entity identification and disambiguation processing in the process of performing entity identification and disambiguation processing on the standardized business opportunity dataset and the standardized bidding information dataset includes:

[0017] Extract customer and project entities from opportunity name, customer name, project description, announcement title, and announcement body;

[0018] Based on a customer alias dictionary, unified social credit code, similarity between abbreviation and full name, historical name mapping, and hierarchical unit relationship, customer entities are merged;

[0019] Based on project topic phrase extraction, keyword merging, and semantic similarity analysis, project names are merged.

[0020] Standardized mapping is performed on regional and industry entities to obtain unified entity identifiers;

[0021] In the process of recalling the candidate bidding information set from the standardized bidding information dataset based on the target business opportunity, a two-stage mechanism is adopted: coarse recall is performed based on customer subject, region and time window to obtain a coarse candidate bidding information set; fine recall is performed from the coarse candidate bidding information set based on project keywords, industry tags, amount range and semantic similarity to obtain the candidate bidding information set.

[0022] As a preferred method for analyzing the correlation of business opportunity bids based on multi-source heterogeneous data, the formula for calculating the correlation confidence level is as follows:

[0023] FinalScore=α×RuleScore+β×ModelScore

[0024] α+β=1

[0025] RuleScore=W1×F1+W2×F2+…+W7×F7

[0026] In the formula, FinalScore is the association confidence score; RuleScore is the rule score calculated based on the rule; ModelScore is the association probability score output by the machine learning model; α and β are weights; F1 to F7 are the corresponding subject matching features, project text features, time features, amount features, regional features, industry business features and process data features, respectively; W1 to W7 are the corresponding feature weights.

[0027] As a preferred solution for the business opportunity information association analysis method based on multi-source heterogeneous data, parameters are configured according to province, industry or business type; the configured parameters include: time window, amount tolerance range, automatic association threshold T1, manual confirmation threshold T2, industry dictionary, stop word dictionary and feature weight;

[0028] The classification results are as follows:

[0029] When the association confidence level is greater than or equal to the automatic association threshold T1, an automatic association result is established.

[0030] When the association confidence level is less than the automatic association threshold T1 and greater than or equal to the manual confirmation threshold T2, a manual confirmation task is generated.

[0031] When the association confidence level is less than the manual confirmation threshold T2, a filtering result is generated;

[0032] The associated evidence chain includes: customer matching basis, project similarity basis, time matching basis, amount matching basis, regional matching basis, and industry consistency basis.

[0033] As a preferred solution for the business opportunity information association analysis method based on multi-source heterogeneous data, the feedback optimization process in the calculation of the association confidence includes: updating the customer alias dictionary, updating the project topic dictionary, updating the feature weights, updating the model parameters, and updating the scoring threshold.

[0034] When a set event is triggered, the association relationship is recalculated; the set events include: adding business opportunities, changing business opportunity fields, adding new tender information, adding winning bid announcements, adding bidding records, and adding contract records.

[0035] This invention also provides a business opportunity bidding information correlation analysis device based on multi-source heterogeneous data, employing the above-mentioned business opportunity bidding information correlation analysis method based on multi-source heterogeneous data, including:

[0036] The data acquisition module is used to generate a raw dataset containing business opportunity data and bidding information data by collecting internal business data and external public data; the internal business data includes business opportunity information and customer master data; the external public data includes bidding announcements and winning bid announcements.

[0037] The standardization processing module is used to obtain a standardized business opportunity dataset and a standardized bidding information dataset by standardizing the business opportunity data and the bidding information data in the original dataset.

[0038] The entity recognition and disambiguation module is used to perform entity recognition and disambiguation processing on the standardized business opportunity dataset and the standardized bidding information dataset to obtain unified identification results for customer entities, project entities, regional entities and industry entities;

[0039] The candidate bidding information recall module is used to recall a set of candidate bidding information from the standardized bidding information dataset based on the unified identifier result and the target business opportunity.

[0040] The association feature construction module is used to extract subject matching features, project text features, time features, amount features, regional features, industry business features and process data features based on the target business opportunity and the candidate bidding information set, and construct the association feature vector between the target business opportunity and each candidate bidding information.

[0041] The association scoring module is used to calculate the association confidence between the target business opportunity and each of the candidate keywords based on the association feature vector and through a rule and model fusion strategy.

[0042] The result classification processing module is used to classify the candidate tags based on the association confidence level, obtain the classification result, and output the corresponding association evidence chain.

[0043] The dynamic update and business output module is used to process the corresponding results according to the classification results and generate manual confirmation results; based on the manual confirmation results and subsequent business results, the calculation process of the association confidence is optimized, and the business output is completed.

[0044] As a preferred embodiment of the business opportunity and bidding information association analysis device based on multi-source heterogeneous data, the standardization processing module includes the following steps in the process of standardizing the business opportunity data and bidding information data in the original dataset: field mapping, text cleaning, time normalization, amount normalization, region normalization, and industry tag normalization.

[0045] As a preferred embodiment of a business opportunity and bidding information association analysis device based on multi-source heterogeneous data, the entity recognition and disambiguation module, in the process of performing entity recognition and disambiguation processing on the standardized business opportunity dataset and the standardized bidding information dataset, includes the following:

[0046] Extract customer and project entities from opportunity name, customer name, project description, announcement title, and announcement body;

[0047] Based on a customer alias dictionary, unified social credit code, similarity between abbreviation and full name, historical name mapping, and hierarchical unit relationship, customer entities are merged;

[0048] Based on project topic phrase extraction, keyword merging, and semantic similarity analysis, project names are merged.

[0049] Standardized mapping is performed on regional and industry entities to obtain unified entity identifiers;

[0050] In the candidate bidding information recall module, a two-stage mechanism is adopted in the process of recalling the candidate bidding information set from the standardized bidding information dataset according to the target business opportunity: coarse recall is performed based on customer subject, region and time window to obtain coarse candidate bidding information set; fine recall is performed from the coarse candidate bidding information set based on project keywords, industry tags, amount range and semantic similarity to obtain the candidate bidding information set.

[0051] As a preferred embodiment of the business opportunity information correlation analysis device based on multi-source heterogeneous data, the formula for calculating the correlation confidence level in the correlation scoring module is as follows:

[0052] FinalScore=α×RuleScore+β×ModelScore

[0053] α+β=1

[0054] RuleScore=W1×F1+W2×F2+…+W7×F7

[0055] In the formula, FinalScore is the association confidence score; RuleScore is the rule score calculated based on the rule; ModelScore is the association probability score output by the machine learning model; α and β are weights; F1 to F7 are the corresponding subject matching features, project text features, time features, amount features, regional features, industry business features and process data features, respectively; W1 to W7 are the corresponding feature weights.

[0056] As a preferred solution for a business opportunity information association analysis device based on multi-source heterogeneous data, parameters are configured according to province, industry or business type; the parameters include: time window, amount tolerance range, automatic association threshold T1, manual confirmation threshold T2, industry dictionary, stop word dictionary and feature weight.

[0057] In the result classification processing module, the classification result is:

[0058] When the association confidence level is greater than or equal to the automatic association threshold T1, an automatic association result is established.

[0059] When the association confidence level is less than the automatic association threshold T1 and greater than or equal to the manual confirmation threshold T2, a manual confirmation task is generated.

[0060] When the association confidence level is less than the manual confirmation threshold T2, a filtering result is generated;

[0061] The associated evidence chain includes: customer matching basis, project similarity basis, time matching basis, amount matching basis, regional matching basis, and industry consistency basis.

[0062] As a preferred solution for a business opportunity information association analysis device based on multi-source heterogeneous data, the dynamic update and business output module includes the following feedback optimization content during the process of calculating the association confidence: updating the customer alias dictionary, updating the project topic dictionary, updating the feature weights, updating the model parameters, and updating the scoring threshold.

[0063] When a set event is triggered, the association relationship is recalculated; the set events include: adding business opportunities, changing business opportunity fields, adding new tender information, adding winning bid announcements, adding bidding records, and adding contract records.

[0064] The present invention has the following advantages:

[0065] First, it can uniformly collect and standardize the modeling of multi-source heterogeneous data from internal business systems and external public platforms, and complete the identification and disambiguation of customer, project, regional and industry entities, effectively solving the association obstacles caused by data dispersion, inconsistent standards and inconsistent subject descriptions.

[0066] Second, a scoring method that combines multi-dimensional association features with rules and models is adopted to replace manual retrieval and single keyword matching. This fully utilizes multiple information such as subject, text, time, amount, and region to reduce association bias and lower the cost of manual comparison.

[0067] Third, set up a graded processing of associated confidence levels and an associated evidence chain output. Perform automatic association, manual confirmation, or filtering operations according to the confidence level to adapt to the actual business process. At the same time, clearly present the basis for association and improve the interpretability and ease of use of the solution.

[0068] Fourth, construct a feedback learning and dynamic update closed loop, continuously optimize the dictionary, weights, model parameters and thresholds based on manual confirmation and business results, and automatically trigger recalculation when data is updated, so that the association scheme can continuously adapt to changes in business.

[0069] Fifth, it supports parameterized configuration by province, industry, and business type, allowing for flexible adjustment of parameters such as time window, amount tolerance, and scoring threshold to adapt to the usage needs of different business scenarios and improve the applicability of the solution. Attached Figure Description

[0070] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0071] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0072] Figure 1 This is a flowchart illustrating the business opportunity information association analysis method based on multi-source heterogeneous data provided in Embodiment 1 of the present invention;

[0073] Figure 2 This is a schematic diagram illustrating the specific implementation process of the business opportunity bidding information association analysis method based on multi-source heterogeneous data provided in Embodiment 1 of the present invention;

[0074] Figure 3 This is a schematic diagram of the technical roadmap architecture of the business opportunity bidding information association analysis method based on multi-source heterogeneous data provided in Embodiment 1 of the present invention;

[0075] Figure 4 This is a schematic diagram of the association scoring logic flow in the business opportunity bidding information association analysis method based on multi-source heterogeneous data provided in Embodiment 1 of the present invention;

[0076] Figure 5 This is a schematic diagram of the feedback learning and dynamic update closed loop in the business opportunity bidding information association analysis method based on multi-source heterogeneous data provided in Embodiment 1 of the present invention;

[0077] Figure 6 This is a schematic diagram of the architecture of the business opportunity information association analysis device based on multi-source heterogeneous data provided in Embodiment 2 of the present invention. Detailed Implementation

[0078] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. 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.

[0079] Example 1

[0080] See Figure 1 , Figure 2 and Figure 3 Embodiment 1 of the present invention provides a method for business opportunity bidding information correlation analysis based on multi-source heterogeneous data, including the following steps:

[0081] T1. By collecting internal business data and external publicly available data, a raw dataset containing business opportunity data and bidding information data is generated; the internal business data includes business opportunity information and customer master data; the external publicly available data includes bidding announcements and winning bid announcements.

[0082] T2. By standardizing the business opportunity data and the bidding information data in the original dataset, a standardized business opportunity dataset and a standardized bidding information dataset are obtained.

[0083] T3. By performing entity recognition and disambiguation processing on the standardized business opportunity dataset and the standardized bidding information dataset, a unified identification result for customer entities, project entities, regional entities, and industry entities is obtained;

[0084] T4. Based on the unified identifier result, recall the candidate bid set from the standardized bid set according to the target business opportunity;

[0085] T5. Based on the target business opportunity and the candidate bidding information set, extract subject matching features, project text features, time features, amount features, regional features, industry business features and process data features to construct the association feature vector between the target business opportunity and each candidate bidding information.

[0086] T6. Based on the associated feature vector, calculate the association confidence between the target business opportunity and each of the candidate keywords through a rule and model fusion strategy;

[0087] T7. Based on the association confidence level, perform association result classification processing on the candidate tags to obtain classification results, and output the association evidence chain of the classification results;

[0088] T8. Process the classification results accordingly and generate manual confirmation results; based on the manual confirmation results and subsequent business results, optimize the calculation process of the association confidence level and complete the business output.

[0089] In this embodiment, in step T1, an original dataset containing business opportunity data and bidding information data is generated by collecting internal business data and external public data.

[0090] Specifically, internal business data must include business opportunity information and customer master data, and can be supplemented with bidding records, visit records, contract data, order data, and product / solution tag data according to business needs; external public data must include bidding announcements and winning bid announcements, and can be supplemented with procurement plans, business entity information, industry news data, national policies, and enterprise information.

[0091] The data collection process can be achieved through interface integration, file import, message subscription, web page data scraping, etc., to complete the unified access and temporary storage of multi-source data, forming a raw dataset covering the entire lifecycle of business opportunities and the entire process of bidding information.

[0092] In this embodiment, in step T2, standardized business opportunity datasets and standardized bidding information datasets are obtained by standardizing the business opportunity data and bidding information data in the original dataset.

[0093] Specifically, addressing the format and standard differences of multi-source heterogeneous data, the process involves unifying and standardizing the raw data, which includes six core standardization actions:

[0094] First, field mapping maps fields with the same meaning but different names from different source systems to a pre-defined standard field system.

[0095] Second, text cleaning removes special symbols, web page tags, fixed template language, and invalid spaces from the original text, while retaining core business text information.

[0096] Third, time normalization converts time data in different formats into a standard time format to eliminate differences in time granularity.

[0097] Fourth, the amount is standardized by converting amount data in different units such as yuan, ten thousand yuan, and one hundred million yuan into a standard unit to eliminate differences in amount measurement.

[0098] Fifth, regional normalization maps the descriptions of administrative regions such as provinces, cities, districts, and counties to the nationally unified administrative division code;

[0099] Sixth, industry label normalization maps industry classification descriptions from different sources to a unified industry label system.

[0100] The final output consists of standardized business opportunity datasets and standardized bidding information datasets with completely unified format, standards, and definitions.

[0101] In this embodiment, in step T3, entity recognition and disambiguation processing is performed on the standardized business opportunity dataset and the standardized bidding information dataset to obtain unified identification results for customer entities, project entities, regional entities and industry entities.

[0102] Specifically, firstly, core information about customer and project entities is extracted from business opportunity names, customer names, project descriptions, announcement titles, and announcement body texts. Secondly, for customer entities, different descriptions of the same customer are merged based on a customer alias dictionary, unified social credit code, similarity between abbreviation and full name texts, historical name mapping relationships, and hierarchical unit affiliation relationships. For project entities, different descriptions of the same project are merged based on project theme phrase extraction, core keyword merging, and text semantic similarity analysis. Finally, standard coding mapping is completed for regional and industry entities, a unique unified identifier is assigned to each of the four types of entities, and the unified identifier result for all entities is output.

[0103] In this embodiment, in step T4, based on the unified identifier result, a set of candidate bids is recalled from the standardized bid dataset according to the target business opportunity.

[0104] Specifically, a two-stage recall mechanism is adopted in the process of recalling the candidate bidding information set: The first stage is the coarse recall stage, which completes the initial screening based on the customer subject, regional entity, and preset time window in the unified identification results, and quickly filters out bidding information data that do not match the subject, region, and time completely; The second stage is the fine recall stage, which completes further screening based on the coarse recall results, using project keyword matching, industry tag matching, amount range matching, and text semantic similarity matching, and finally forms a candidate bidding information set that has potential relevance to the target business opportunity.

[0105] In this embodiment, in step T5, based on the target business opportunity and the candidate bid set, a correlation feature vector between the target business opportunity and each candidate bid is constructed.

[0106] Specifically, the constructed associated feature vector contains at least seven core feature categories:

[0107] First, the subject matching features cover the matching degree of the customer's full name, the matching degree of the customer's abbreviation, the matching degree of the customer's alias, and the matching status of the unified social credit code;

[0108] Second, the project text features, covering the text similarity between the business opportunity name and the announcement title, the semantic similarity between the business opportunity description and the announcement body, and the coverage of project keywords;

[0109] Thirdly, time characteristics, covering the time interval between business opportunity creation and announcement release, and the time interval between business opportunity entry and bid opening;

[0110] Fourthly, the monetary characteristics, which cover the difference between the projected amount of the business opportunity and the announced budget amount, and the matching of the monetary tolerance range;

[0111] Fifth, regional characteristics, covering the consistency between the region where the business opportunity is located and the region where the announcement is located;

[0112] Sixth, industry business characteristics, covering the degree of consistency in industry labels, product direction, and solution labels;

[0113] Seventh, process data characteristics, covering whether there are corresponding bidding records, winning bid results, contract results, or electronic forms as supporting evidence.

[0114] The final result is a unified, quantifiable, and calculable association feature vector, which is then input into the association scoring process.

[0115] In this embodiment, in step T6, based on the associated feature vector, the association confidence between the target business opportunity and each of the candidate keywords is calculated using a rule and model fusion strategy.

[0116] Specifically, through a rule and model fusion strategy, with "dual score calculation - fusion weighted output" as the core architecture, the interpretability of the rules and the generalization ability of the machine learning model are taken into account. Finally, the association confidence value in the 0-1 range is output as the core judgment basis for subsequent classification processing.

[0117] like Figure 4 As shown, the specific calculation process is as follows:

[0118] N1, Dual-Path Independent Score Calculation: The rule score (RuleScore) and model score (ModelScore) are calculated independently in parallel through the rule engine path and the machine learning model path, respectively. The two calculation paths do not interfere with each other, ensuring the independence and accuracy of the scores.

[0119] N2. Rule Score Calculation: The rule score is obtained by weighted summation of multi-dimensional related features. The calculation formula is as follows:

[0120] RuleScore=W1×F1+W2×F2+…+W7×F7

[0121] In the formula, RuleScore is the rule score calculated based on the rule; F1 to F7 are the corresponding subject matching features, project text features, time features, amount features, regional features, industry business features and process data features, respectively; W1 to W7 are the corresponding feature weights.

[0122] N3. Model Score Calculation: The model score is directly output by the pre-trained machine learning model. It is the correlation probability score between the target business opportunity and the corresponding candidate bid, and the value range is 0-1.

[0123] N4. Final Score Based on Fusion Weighting: The final association confidence score is calculated using the fusion formula, which is:

[0124] FinalScore=α×RuleScore+β×ModelScore

[0125] In the formula, FinalScore is the association confidence score; α and β are both weights.

[0126] In this embodiment, in step T7, based on the association confidence level, the candidate tags are subjected to association result classification processing to obtain classification results, and the association evidence chain of the classification results is output.

[0127] Specifically, based on the correlation confidence level output in the previous step, differentiated business processing is completed to adapt to the actual business operation process, while the correlation basis is output to improve the interpretability of the solution.

[0128] The hierarchical processing rule is as follows: when the association confidence level is greater than or equal to the automatic association threshold T1, an automatic association result is established;

[0129] When the association confidence level is less than the automatic association threshold T1 and greater than or equal to the manual confirmation threshold T2, a manual confirmation task is generated.

[0130] When the association confidence level is less than the manual confirmation threshold T2, a filtering result is generated;

[0131] The preferred value for the automatic association threshold T1 is 0.85, and the preferred value for the manual confirmation threshold T2 is 0.60.

[0132] In this embodiment, for each graded result, a corresponding chain of evidence is output simultaneously. The chain of evidence includes: customer matching basis, project similarity basis, time matching basis, amount matching basis, region matching basis, and industry consistency basis, clearly presenting the core supporting information for the association determination. The graded results are simultaneously pushed to the business implementation and feedback optimization stages.

[0133] In this embodiment, in step T8, corresponding processing is performed according to the classification result and a manual confirmation result is generated; based on the manual confirmation result and subsequent business results, the calculation process of the association confidence is optimized, and the business output is completed.

[0134] Specifically, corresponding business processes are performed on the tiered results: first, the business opportunities and bidding information are bound and archived for the automatically associated results; second, manual confirmation results are generated after the manual confirmation tasks are followed up and processed; and third, the filtered results are archived and retained.

[0135] In this embodiment, a complete feedback learning and dynamic update closed loop is also established, such as... Figure 5 As shown, the process is as follows:

[0136] M1, Multi-source result input: With manual confirmation results, winning bid / contract results, and bidding / order results as the three core input sources, it collects real correlation judgment data from the business end as the basis for the whole process optimization;

[0137] M2. Unified sample storage: All the above input data are imported into the associated sample library, and the sample classification and management are completed in the library to form positive samples, negative samples and historical high confidence samples, providing complete data support for subsequent parameter optimization;

[0138] M3, Multi-dimensional Parameter Iterative Update: Based on the labeled sample data of the associated sample library, it synchronously drives four core optimization units to complete parameter iteration: First, the dictionary mapping update unit updates the basic mapping library such as the customer alias dictionary and project theme dictionary; second, the feature weight update unit adjusts the feature weights W1-W7 in the rule score calculation; third, the model parameter update unit optimizes the core parameters of the machine learning model and improves the model's generalization ability; fourth, the threshold rule update unit adjusts the scoring thresholds T1 and T2 for automatic association and manual confirmation.

[0139] M4. Recalculation Trigger for Related Results: The recalculation trigger unit automatically triggers the recalculation of the correlation results between business opportunities and tender information based on all optimized parameters, so as to achieve the rapid implementation of optimization effects. At the same time, the recalculation trigger unit will automatically start incremental recalculation when key business events such as the addition of new business opportunities, changes in business opportunity fields, the entry of new tender information into the database, the addition of new winning bid announcements, the addition of new bidding records, and the addition of new contract records occur, to ensure the real-time performance and accuracy of the correlation results.

[0140] M5, Closed-loop continuous iteration: The recalculated correlation results will be combined with new business feedback (manual confirmation, winning bids / contracts, bidding / order results, etc.) and re-integrated into the correlation sample library, forming an infinitely iterative optimization closed loop, continuously improving the effectiveness of correlation analysis;

[0141] Finally, the confirmed correlation results will be output to business scenarios such as business opportunity listing, premium operation, competitor winning bid reminders, project full-process management, and business report statistics, completing the full-process business output.

[0142] In this embodiment, parameters are configured based on province, industry, or business type; the parameters include: time window, amount tolerance range, automatic association threshold T1, manual confirmation threshold T2, industry dictionary, stop word dictionary, and feature weight.

[0143] In one possible implementation, the input to the machine learning model is a standardized multidimensional correlation feature vector corresponding to the pairing of the target business opportunity with a single candidate bid. The input to the machine learning model is an 18-dimensional fixed-length numerical vector, corresponding to the sub-feature decomposition of 7 major feature categories, as specifically defined in Table 1:

[0144]

[0145] Table 1 Inputs to the machine learning model

[0146] All input features are min-max normalized and uniformly mapped to the [0,1] interval to eliminate the dimensional differences between different features; missing feature values ​​are filled with 0 values ​​to ensure the dimensional consistency and integrity of all input vectors; the preprocessed feature vectors are directly input into the model without additional transformation operations.

[0147] The output of the machine learning model is as follows:

[0148] Model direct output: After completing forward inference, the model directly outputs a continuous numerical value ModelScore in the interval [0,1], which is the correlation probability score between the target business opportunity and the corresponding candidate target information.

[0149] Physical meaning of the output value: The magnitude of the ModelScore is positively correlated with the degree of correlation between the business opportunity and the bidding information; the closer the value is to 1, the higher the probability that the two are related to the same project; the closer the value is to 0, the higher the probability that the two are not related.

[0150] Technical application of the output: This output value is directly used as one of the core inputs for the association confidence fusion calculation. It is weighted and fused with the rule score RuleScore using the formula FinalScore=α×RuleScore+β×ModelScore to obtain the final association confidence, which is used in the hierarchical processing flow.

[0151] The training methods for machine learning models are as follows:

[0152] 1. Construction of the training sample set:

[0153] (1) Explicit definition of positive and negative samples:

[0154] This model is a binary classification model, with sample labels being binary labels. The rules for determining positive and negative samples are completely reproducible.

[0155] Positive Samples (Label Value = 1): Opportunity-bid information matching samples that meet any of the following conditions are marked as positive samples: ① Matching samples where the opportunity and the bid information are from the same project, as confirmed manually by business personnel; ② Matching samples where the bidding records, winning bid announcements, contract information and bid information corresponding to the opportunity are completely matched; ③ Matching samples where the association confidence score FinalScore is greater than or equal to the automatic association threshold T1, and there are no objections after business closed-loop verification.

[0156] Negative samples (label value = 0): Opportunity-bid pairing samples that meet any of the following conditions are marked as negative samples: ① Pairing samples that have been manually confirmed by business personnel and are not related to bids; ② Pairing samples with a correlation confidence score FinalScore < manual confirmation threshold T2 and no business data to support the correlation; ③ Random pairing samples with completely mismatched customer subjects, regions, and industries.

[0157] (2) Sample set partitioning and balancing:

[0158] Sample set partitioning: The labeled full sample set is randomly divided into training set, validation set and test set in a ratio of 7:2:1. The training set is used for model parameter iteration, the validation set is used for hyperparameter tuning and early stopping control, and the test set is used for model generalization ability verification.

[0159] Sample balancing: To address the issue of uneven distribution of positive and negative samples in government and enterprise scenarios, the SMOTE oversampling algorithm is used to expand the minority class samples in the training set, while the TomekLinks undersampling algorithm is used to clean the majority class samples, ensuring that the ratio of positive to negative samples in the training set is controlled between 1:1.2 and 1:1.5, thus avoiding prediction bias in the model.

[0160] Cold start initial samples: In the early stages of the project, when there are no manually labeled samples, high-confidence association samples output by the rule engine are used as initial positive samples, and random paired samples with no matching customers / regions / industries are used as initial negative samples to complete the initial model training. Subsequent iterations and optimizations are made through manually labeled samples.

[0161] The machine learning model employs the LightGBM model based on the Gradient Boosting Decision Tree (GBDT) framework. Its core utilizes a histogram algorithm and a depth-constrained leaf-wise growth strategy, balancing training efficiency and generalization ability. The model's core task is binary classification, with a binary cross-entropy loss function as the objective function and a sigmoid activation function in the output layer, mapping the model output to the [0,1] interval. The core hyperparameters of the machine learning model are shown in Table 2.

[0162]

[0163] Table 2 Hyperparameter Design of Machine Learning Models

[0164] The complete training process for a machine learning model is as follows:

[0165] Step 1: Sample preprocessing. Perform min-max normalization and missing value imputation on all features of the training set, validation set, and test set to ensure that all input vectors have consistent dimensions and uniform format.

[0166] Step 2: Model initialization. Initialize the LightGBM model according to the preset hyperparameters mentioned above. For cold start scenarios, random initialization is used, while for iterative optimization scenarios, the optimal model parameters from the previous version are used as the initialization basis.

[0167] Step 3: Iterative training. Input the preprocessed training set into the initial model and use the gradient descent algorithm to iteratively optimize the model parameters. After each iteration, calculate the model's AUC value and accuracy on the validation set.

[0168] Step 4: Early stopping control. Set the early stopping rounds to 50. When the AUC value on the validation set does not improve for 50 consecutive rounds, terminate the training early, save the current optimal model parameters, and avoid model overfitting.

[0169] Step 5: Model performance verification. The trained model is tested on an independent test set for inference verification. The core evaluation metrics include AUC, precision, recall, and F1 score. When the test set AUC is ≥0.92 and the F1 score is ≥0.85, the model is deemed to meet the business accuracy requirements and can be used for online inference. If the requirements are not met, the process is to return to adjusting hyperparameters, expanding the labeled sample set, and re-executing the training process.

[0170] The incremental training and dynamic optimization methods for machine learning models are as follows:

[0171] Incremental sample update: Collect newly added manual confirmation results and business closed-loop data at fixed intervals (monthly / quarterly), build an incremental labeled sample set, and import it into the associated sample library for unified management;

[0172] Incremental training trigger conditions: Incremental training is automatically triggered when the number of incremental samples reaches more than 10% of the original training set, or when the accuracy of online inference of the model drops by more than 5% for 15 consecutive days.

[0173] Incremental training execution: Based on the current optimal model parameters running online, the above complete training process is executed using an incremental sample set combined with 30% of the historical core sample set to complete the iterative update of model parameters; after the updated model is verified to meet the standards on the test set, it replaces the online inference model to achieve continuous optimization of model performance.

[0174] The application scenarios of this invention are as follows:

[0175] In the context of business opportunity follow-up management in government and enterprise marketing, this invention can realize the automated association and matching of internal business opportunities and external listings, reduce the workload of manual retrieval and comparison, improve the efficiency of business opportunity follow-up, and support core business links such as business opportunity listing and follow-up priority classification.

[0176] In the context of enterprise bidding business full-process management, this invention can quickly locate the bidding and winning bid announcement information corresponding to business opportunities, provide data support for the entire process of bidding project establishment, bid preparation, bid opening follow-up, and winning bid archiving, and improve the full-link management system of bidding business.

[0177] In market intelligence and competitor dynamic tracking scenarios, this invention can promptly obtain the bidding status of competitors for corresponding projects through the correlation results of business opportunities and bidding information, providing complete data basis for enterprise market strategy adjustment and competitor layout analysis.

[0178] In the scenario of managing business opportunities across multiple regions and industries in a group enterprise, this invention can support parameterized configuration by province, industry, and business type to adapt to the business needs of different branches and achieve unified management and standardized correlation analysis of business opportunities and bidding information data across the entire group.

[0179] In the context of enterprise digital operation analysis and decision support, this invention can connect the entire chain of data, including business opportunities, tender information, bidding, and contracts, to form a complete business data closed loop, providing stable data support for enterprise operation report statistics, business performance analysis, and marketing resource allocation.

[0180] In the context of enterprise customer lifecycle management, this invention can integrate all business opportunities and bidding information of the same customer by identifying and merging customer entities, improve the customer profile system, and provide data support for daily customer maintenance, potential demand mining, and business scope expansion.

[0181] It should be noted that the method of this disclosure embodiment can be executed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method of this disclosure embodiment, and the multiple devices will interact with each other to complete the method described.

[0182] It should be noted that the above description describes some embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0183] Example 2

[0184] See Figure 6 Embodiment 2 of the present invention also provides a business opportunity bidding information association analysis device based on multi-source heterogeneous data, including:

[0185] The data acquisition module 001 is used to generate a raw dataset containing business opportunity data and bidding information data by collecting internal business data and external public data; the internal business data includes business opportunity information and customer master data; the external public data includes bidding announcements and winning bid announcements.

[0186] The standardization processing module 002 is used to obtain a standardized business opportunity dataset and a standardized bidding information dataset by standardizing the business opportunity data and the bidding information data in the original dataset.

[0187] The entity recognition and disambiguation module 003 is used to perform entity recognition and disambiguation processing on the standardized business opportunity dataset and the standardized bidding information dataset to obtain unified identification results for customer entities, project entities, regional entities and industry entities;

[0188] The candidate bid recall module 004 is used to recall a set of candidate bids from the standardized bid dataset based on the unified identifier result and the target business opportunity.

[0189] The association feature construction module 005 is used to extract subject matching features, project text features, time features, amount features, regional features, industry business features and process data features based on the target business opportunity and the candidate bidding information set, and construct the association feature vector between the target business opportunity and each candidate bidding information.

[0190] The association scoring module 006 is used to calculate the association confidence between the target business opportunity and each of the candidate keywords based on the association feature vector and through a rule and model fusion strategy.

[0191] The result classification processing module 007 is used to classify the candidate tags based on the association confidence level, obtain the classification result, and output the corresponding association evidence chain.

[0192] The dynamic update and business output module 008 is used to perform corresponding processing based on the hierarchical results and generate manual confirmation results; based on the manual confirmation results and subsequent business results, the calculation process of the correlation confidence is optimized, and the business output is completed.

[0193] In this embodiment, the internal business data in the data acquisition module 001 includes business opportunity information and customer master data; the external public data includes bidding announcements and winning bid announcements.

[0194] In the standardization processing module 002, during the standardization processing of the business opportunity data and the bidding information data in the original dataset, the standardization processing includes: field mapping, text cleaning, time normalization, amount normalization, region normalization, and industry tag normalization.

[0195] In this embodiment, the entity recognition and disambiguation module 003, during the entity recognition and disambiguation processing of the standardized business opportunity dataset and the standardized tender information dataset, includes the following:

[0196] Extract customer and project entities from opportunity name, customer name, project description, announcement title, and announcement body;

[0197] Based on a customer alias dictionary, unified social credit code, similarity between abbreviation and full name, historical name mapping, and hierarchical unit relationship, customer entities are merged;

[0198] Based on project topic phrase extraction, keyword merging, and semantic similarity analysis, project names are merged.

[0199] Standardized mapping is performed on regional and industry entities to obtain unified entity identifiers;

[0200] In the candidate bidding information recall module 004, a two-stage mechanism is adopted in the process of recalling the candidate bidding information set from the standardized bidding information dataset according to the target business opportunity: coarse recall is performed based on customer subject, region and time window to obtain coarse candidate bidding information set; fine recall is performed from the coarse candidate bidding information set based on project keywords, industry tags, amount range and semantic similarity to obtain the candidate bidding information set.

[0201] In this embodiment, the associated feature construction module 005 includes the following associated feature vectors: subject matching features, project text features, time features, amount features, regional features, industry business features, and process data features.

[0202] In the association scoring module 006, the formula for calculating the association confidence level is as follows:

[0203] FinalScore=α×RuleScore+β×ModelScore

[0204] α+β=1

[0205] RuleScore=W1×F1+W2×F2+…+W7×F7

[0206] In the formula, FinalScore is the association confidence score; RuleScore is the rule score calculated based on the rule; ModelScore is the association probability score output by the machine learning model; F1 to F7 are the corresponding subject matching features, project text features, time features, amount features, regional features, industry business features, and process data features, respectively; W1 to W7 are the corresponding feature weights.

[0207] In this embodiment, parameters are configured based on province, industry, or business type; the parameters include: time window, amount tolerance range, automatic association threshold T1, manual confirmation threshold T2, industry dictionary, stop word dictionary, and feature weights.

[0208] In the result classification processing module 007, the classification result is:

[0209] When the association confidence level is greater than or equal to the automatic association threshold T1, an automatic association result is established.

[0210] When the association confidence level is less than the automatic association threshold T1 and greater than or equal to the manual confirmation threshold T2, a manual confirmation task is generated.

[0211] When the association confidence level is less than the manual confirmation threshold T2, a filtering result is generated;

[0212] The associated evidence chain includes: customer matching basis, project similarity basis, time matching basis, amount matching basis, region matching basis, and industry consistency basis;

[0213] In this embodiment, in the dynamic update and business output module 008, during the feedback optimization process of calculating the association confidence, the feedback optimization content includes: updating the customer alias dictionary, updating the project topic dictionary, updating the feature weights, updating the model parameters, and updating the scoring threshold.

[0214] When a set event is triggered, the association relationship is recalculated; the set events include: adding business opportunities, changing business opportunity fields, adding new tender information, adding winning bid announcements, adding bidding records, and adding contract records.

[0215] It should be noted that the information interaction and execution process between the modules of the above-mentioned device are based on the same concept as the method embodiment in Embodiment 1 of this application, and the resulting technical effects are the same as those in the method embodiment of this application. For details, please refer to the description in the method embodiment shown above in this application, and it will not be repeated here.

[0216] Example 3

[0217] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium storing program code for a business opportunity and bidding information association analysis method based on multi-source heterogeneous data. The program code includes instructions for executing the business opportunity and bidding information association analysis method based on multi-source heterogeneous data of Embodiment 1 or any possible implementation thereof.

[0218] Computer-readable storage media can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0219] Example 4

[0220] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0221] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor can call the program instructions to execute the business opportunity information association analysis method based on multi-source heterogeneous data in Embodiment 1 or any possible implementation thereof.

[0222] Specifically, a processor can be implemented in hardware or software. When implemented in hardware, the processor can be a logic circuit, an integrated circuit, etc. When implemented in software, the processor can be a general-purpose processor that reads software code stored in memory. This memory can be integrated into the processor or located outside the processor and exist independently.

[0223] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable system. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0224] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing systems. They can be centralized on a single computing system or distributed across a network of multiple computing systems. Optionally, they can be implemented using program code executable by a computing system, thereby storing them in a storage system for execution by the computing system. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0225] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for analyzing the correlation of business opportunity bids based on multi-source heterogeneous data, characterized in that, include: By collecting internal business data and external publicly available data, a raw dataset containing business opportunity data and bidding information data is generated; the internal business data includes business opportunity information and customer master data; the external publicly available data includes bidding announcements and winning bid announcements. By standardizing the business opportunity data and the bidding information data in the original dataset, a standardized business opportunity dataset and a standardized bidding information dataset are obtained. By performing entity recognition and disambiguation processing on the standardized business opportunity dataset and the standardized bidding information dataset, unified identification results for customer entities, project entities, regional entities, and industry entities are obtained. Based on the unified identifier result, a set of candidate bids is recalled from the standardized bid dataset according to the target business opportunity; Based on the target business opportunity and the candidate bidding information set, subject matching features, project text features, time features, amount features, regional features, industry business features and process data features are extracted to construct the association feature vector between the target business opportunity and each candidate bidding information. Based on the associated feature vector, the association confidence between the target business opportunity and each of the candidate bids is calculated through a rule and model fusion strategy. Based on the association confidence level, the candidate tags are subjected to association result classification processing to obtain classification results, and the association evidence chain of the classification results is output. Based on the classification results, corresponding processing is performed and a manual confirmation result is generated; Based on the manual confirmation results and subsequent business results, the calculation process of the association confidence level is optimized through feedback, and the business output is completed.

2. The business opportunity bidding information correlation analysis method based on multi-source heterogeneous data according to claim 1, characterized in that, In the process of standardizing the business opportunity data and the bidding information data in the original dataset, the standardization process includes: field mapping, text cleaning, time normalization, amount normalization, region normalization, and industry tag normalization.

3. The business opportunity bidding information correlation analysis method based on multi-source heterogeneous data according to claim 2, characterized in that, In the process of performing entity recognition and disambiguation processing on the standardized business opportunity dataset and the standardized tender information dataset, the entity recognition and disambiguation processing includes: Extract customer and project entities from opportunity name, customer name, project description, announcement title, and announcement body; Based on a customer alias dictionary, unified social credit code, similarity between abbreviation and full name, historical name mapping, and hierarchical unit relationship, customer entities are merged; Based on project topic phrase extraction, keyword merging, and semantic similarity analysis, project names are merged. Standardized mapping is performed on regional and industry entities to obtain unified entity identifiers; In the process of recalling the candidate bidding information set from the standardized bidding information dataset based on the target business opportunity, a two-stage mechanism is adopted: coarse recall is performed based on customer subject, region and time window to obtain a coarse candidate bidding information set; fine recall is performed from the coarse candidate bidding information set based on project keywords, industry tags, amount range and semantic similarity to obtain the candidate bidding information set.

4. The business opportunity information correlation analysis method based on multi-source heterogeneous data according to claim 3, characterized in that, The formula for calculating the association confidence level is as follows: FinalScore=α×RuleScore+β×ModelScore α+β=1 RuleScore=W1×F1+W2×F2+…+W7×F7 In the formula, FinalScore is the association confidence score; RuleScore is the rule score calculated based on the rule; ModelScore is the association probability score output by the machine learning model; α and β are weights; F1 to F7 are the corresponding subject matching features, project text features, time features, amount features, regional features, industry business features and process data features, respectively. W1 to W7 are the corresponding feature weights.

5. The business opportunity bidding information correlation analysis method based on multi-source heterogeneous data according to claim 4, characterized in that, Configure parameters based on province, industry, or business type; the configured parameters include: time window, amount tolerance range, automatic association threshold T1, manual confirmation threshold T2, industry dictionary, stop word dictionary, and feature weights; The classification results are as follows: When the association confidence level is greater than or equal to the automatic association threshold T1, an automatic association result is established. When the association confidence level is less than the automatic association threshold T1 and greater than or equal to the manual confirmation threshold T2, a manual confirmation task is generated. When the association confidence level is less than the manual confirmation threshold T2, a filtering result is generated; The associated evidence chain includes: customer matching basis, project similarity basis, time matching basis, amount matching basis, regional matching basis, and industry consistency basis.

6. The business opportunity bidding information correlation analysis method based on multi-source heterogeneous data according to claim 5, characterized in that, During the feedback optimization process of calculating the association confidence, the feedback optimization content includes: updating the customer alias dictionary, updating the project topic dictionary, updating the feature weights, updating the model parameters, and updating the scoring threshold; When a set event is triggered, the association relationship is recalculated; the set events include: adding business opportunities, changing business opportunity fields, adding new tender information, adding winning bid announcements, adding bidding records, and adding contract records.

7. A business opportunity bidding information correlation analysis device based on multi-source heterogeneous data, employing the business opportunity bidding information correlation analysis method based on multi-source heterogeneous data as described in any one of claims 1-6, characterized in that, include: The data acquisition module is used to generate a raw dataset containing business opportunity data and bidding information data by collecting internal business data and external public data; the internal business data includes business opportunity information and customer master data; the external public data includes bidding announcements and winning bid announcements. The standardization processing module is used to obtain a standardized business opportunity dataset and a standardized bidding information dataset by standardizing the business opportunity data and the bidding information data in the original dataset. The entity recognition and disambiguation module is used to perform entity recognition and disambiguation processing on the standardized business opportunity dataset and the standardized bidding information dataset to obtain unified identification results for customer entities, project entities, regional entities and industry entities; The candidate bidding information recall module is used to recall a set of candidate bidding information from the standardized bidding information dataset based on the unified identifier result and the target business opportunity. The association feature construction module is used to extract subject matching features, project text features, time features, amount features, regional features, industry business features and process data features based on the target business opportunity and the candidate bidding information set, and construct the association feature vector between the target business opportunity and each candidate bidding information. The association scoring module is used to calculate the association confidence between the target business opportunity and each of the candidate keywords based on the association feature vector and through a rule and model fusion strategy. The result classification processing module is used to classify the candidate tags based on the association confidence level, obtain the classification result, and output the corresponding association evidence chain. The dynamic update and business output module is used to perform corresponding processing based on the classification results and generate manual confirmation results. Based on the manual confirmation results and subsequent business results, the calculation process of the association confidence level is optimized through feedback, and the business output is completed.

8. The business opportunity information correlation analysis device based on multi-source heterogeneous data according to claim 7, characterized in that, In the standardization processing module, during the standardization process of the business opportunity data and the bidding information data in the original dataset, the standardization processing includes: field mapping, text cleaning, time normalization, amount normalization, region normalization, and industry tag normalization.

9. The business opportunity information correlation analysis device based on multi-source heterogeneous data according to claim 8, characterized in that, In the entity recognition and disambiguation module, during the entity recognition and disambiguation processing of the standardized business opportunity dataset and the standardized tender information dataset, the entity recognition and disambiguation processing includes: Extract customer and project entities from opportunity name, customer name, project description, announcement title, and announcement body; Based on a customer alias dictionary, unified social credit code, similarity between abbreviation and full name, historical name mapping, and hierarchical unit relationship, customer entities are merged; Based on project topic phrase extraction, keyword merging, and semantic similarity analysis, project names are merged. Standardized mapping is performed on regional and industry entities to obtain unified entity identifiers; In the candidate bidding information recall module, a two-stage mechanism is adopted in the process of recalling the candidate bidding information set from the standardized bidding information dataset according to the target business opportunity: coarse recall is performed based on customer subject, region and time window to obtain coarse candidate bidding information set; fine recall is performed from the coarse candidate bidding information set based on project keywords, industry tags, amount range and semantic similarity to obtain the candidate bidding information set.

10. The business opportunity information correlation analysis device based on multi-source heterogeneous data according to claim 9, characterized in that, In the association scoring module, the formula for calculating the association confidence level is: FinalScore=α×RuleScore+β×ModelScore α+β=1 RuleScore=W1×F1+W2×F2+…+W7×F7 In the formula, FinalScore is the association confidence score; RuleScore is the rule score calculated based on the rule; ModelScore is the association probability score output by the machine learning model; α and β are weights; F1 to F7 are the corresponding subject matching features, project text features, time features, amount features, regional features, industry business features and process data features, respectively. W1 to W7 are the corresponding feature weights; Configure parameters based on province, industry, or business type; the configured parameters include: time window, amount tolerance range, automatic association threshold T1, manual confirmation threshold T2, industry dictionary, stop word dictionary, and feature weights; In the result classification processing module, the classification result is: When the association confidence level is greater than or equal to the automatic association threshold T1, an automatic association result is established. When the association confidence level is less than the automatic association threshold T1 and greater than or equal to the manual confirmation threshold T2, a manual confirmation task is generated. When the association confidence level is less than the manual confirmation threshold T2, a filtering result is generated; The associated evidence chain includes: customer matching basis, project similarity basis, time matching basis, amount matching basis, region matching basis, and industry consistency basis; In the dynamic update and business output module, during the feedback optimization process of calculating the association confidence, the feedback optimization content includes: updating the customer alias dictionary, updating the project topic dictionary, updating the feature weights, updating the model parameters, and updating the scoring threshold. When a set event is triggered, the association relationship is recalculated; the set events include: adding business opportunities, changing business opportunity fields, adding new tender information, adding winning bid announcements, adding bidding records, and adding contract records.

Images