Business opportunity recommendation method and device, computer equipment and storage medium

Abstract

The invention discloses a business opportunity recommendation method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring internal data and external data of an enterprise to obtain related data; enterprise features, business opportunity features and interaction features are extracted from the related data; inputting the enterprise features, the business opportunity features and the interaction features into a business opportunity recommendation model for business opportunity matching prediction and transaction prediction to obtain a business opportunity matching score and a transaction probability prediction value; and outputting the business opportunity matching score and the transaction probability prediction value. By implementing the method, the business opportunities can be accurately matched, the transaction possibility can be predicted, the recommendation precision and the decision-making efficiency are improved, dynamic adaptability is achieved, continuous updating and comprehensive multi-dimensional feature analysis can be achieved, and the decision-making ability of enterprises in competition is enhanced.

Description

Technical Field

[0001] The present invention relates to computers, and more specifically to a business opportunity recommendation method, apparatus, computer device, and storage medium. Background Art

[0002] The use of a business opportunity recommendation system is mainly reflected in helping enterprises quickly identify and capture potential business opportunities in a complex and changing market environment, so as to improve business growth and decision-making efficiency.

[0003] Most of the existing business opportunity recommendation technologies rely on traditional rule engine-based methods or simple statistical analysis methods, such as using collaborative filtering, decision trees, etc. for recommendation. However, traditional recommendation methods usually rely on single numerical features or basic user behavior data, lacking comprehensive analysis of multi-source and heterogeneous data. This limits the accuracy and precision of business opportunity recommendation and is difficult to meet the requirements of complex business scenarios. In business opportunity transaction prediction of the prior art, simple statistical models or shallow machine learning algorithms are often used, which are difficult to handle complex non-linear relationships and high-dimensional features. Therefore, the prediction results may not be accurate and cannot effectively improve the decision-making ability and business execution ability of enterprises. Most traditional business opportunity recommendation systems have limitations in optimization and usually can only optimize business opportunity matching or transaction prediction alone. This single optimization method ignores multi-objective joint optimization, resulting in unsatisfactory recommendation effects and unable to optimize the accuracy and precision of both business opportunity matching and transaction prediction simultaneously. Data update and model training of traditional recommendation systems usually lag behind and are difficult to reflect the dynamic changes of the business opportunity market in real time. Especially in the face of a rapidly changing market environment, the real-time performance and adaptability of the recommendation system are poor, affecting its effect in practical applications.

[0004] Therefore, it is necessary to design a new method to achieve accurate matching of business opportunities and predict the possibility of transactions, improve recommendation accuracy and decision-making efficiency, have dynamic adaptability, be able to continuously update and comprehensively analyze multi-dimensional features, and enhance the decision-making ability of enterprises in competition. Summary of the Invention

[0005] The purpose of the present invention is to overcome the defects of the prior art and provide a business opportunity recommendation method, apparatus, computer device, and storage medium.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions: A business opportunity recommendation method, including:

[0007] Obtain internal data and external data of an enterprise to obtain relevant data;

[0008] Extract enterprise features, business opportunity features, and interaction features from the relevant data;

[0009] Input the enterprise characteristics, business opportunity characteristics, and interaction characteristics into a business opportunity recommendation model for business opportunity matching prediction and transaction prediction to obtain a business opportunity matching score and a transaction probability prediction value;

[0010] Output the business opportunity matching score and the transaction probability prediction value.

[0011] Its further technical solution is that the extraction of enterprise characteristics, business opportunity characteristics, and interaction characteristics from the relevant data includes:

[0012] Extract data on the scale, industry category, geographical location, historical cooperation records, and financial status of the enterprise from the relevant data to obtain initial enterprise characteristics;

[0013] Extract information on business opportunity category, demand quantity, budget range, expected cooperation period, and market potential from the relevant data to obtain initial business opportunity characteristics;

[0014] Analyze the historical interaction records and feedback evaluations between the enterprise and the business opportunity in the relevant data to obtain initial interaction characteristics;

[0015] Process missing values, detect outliers, normalize numerical features, and perform encoding conversion on categorical features for the initial enterprise characteristics, the initial business opportunity characteristics, and the initial interaction characteristics respectively to obtain enterprise characteristics, business opportunity characteristics, and interaction characteristics.

[0016] Its further technical solution is that the input of the enterprise characteristics, the business opportunity characteristics, and the interaction characteristics into a business opportunity recommendation model for business opportunity matching prediction and transaction prediction to obtain a business opportunity matching score and a transaction probability prediction value includes:

[0017] Perform feature engineering processing on the enterprise characteristics, the business opportunity characteristics, and the interaction characteristics to obtain a processing result;

[0018] Input the processing result into a business opportunity recommendation model for business opportunity matching prediction and transaction prediction to obtain a business opportunity matching score and a transaction probability prediction value.

[0019] Its further technical solution is that the performing feature engineering processing on the enterprise characteristics, the business opportunity characteristics, and the interaction characteristics to obtain a processing result includes:

[0020] Use the feature importance evaluation of XGBoost to normalize the importance of each of the enterprise characteristics, the business opportunity characteristics, and the interaction characteristics, sort each of the enterprise characteristics, the business opportunity characteristics, and the interaction characteristics from largest to smallest according to importance, calculate the cumulative importance, and screen all features whose cumulative importance meets the requirements to obtain a processing result.

[0021] Its further technical solution is: the business opportunity recommendation model is obtained by acquiring historical relevant data, extracting enterprise features, business opportunity features, and interaction features, and using them as a sample set to train a deep learning architecture after feature engineering processing.

[0022] Its further technical solution is: the business opportunity recommendation model is obtained by acquiring historical relevant data, extracting enterprise features, business opportunity features, and interaction features, and using them as a sample set to train a deep learning architecture after feature engineering processing, including:

[0023] Acquire historical relevant data, extract enterprise features, business opportunity features, and interaction features, and perform feature engineering processing to obtain a sample set;

[0024] Construct a deep learning architecture and define a loss function;

[0025] Use the sample set to train the deep learning architecture, combine the loss function with the trained deep learning architecture, and determine the hyperparameters in the loss function through grid search or cross-validation methods to obtain the business opportunity recommendation model.

[0026] Its further technical solution is: the structure of the deep learning architecture includes an input layer, a feature fusion layer, a deep learning network, and an output layer. The deep learning network includes a first fully connected layer, a first batch normalization layer, a first activation function ReLU, a first Dropout layer, a second fully connected layer, a second batch normalization, a second activation function ReLU, a second Dropout layer. The output layer includes two branches and respectively uses the sigmoid activation function.

[0027] The present invention also provides a business opportunity recommendation device, including:

[0028] A data acquisition unit for acquiring internal and external data of an enterprise to obtain relevant data;

[0029] A feature extraction unit for extracting enterprise features, business opportunity features, and interaction features from the relevant data;

[0030] A prediction unit for inputting the enterprise features, the business opportunity features, and the interaction features into a business opportunity recommendation model for business opportunity matching prediction and transaction prediction to obtain a business opportunity matching score and a transaction probability prediction value;

[0031] An output unit for outputting the business opportunity matching score and the transaction probability prediction value.

[0032] The present invention also provides a computer device, which includes a memory and a processor. A computer program is stored on the memory, and when the processor executes the computer program, the above method is implemented.

[0033] The present invention also provides a storage medium storing a computer program, which, when executed by a processor, implements the above method.

[0034] The beneficial effects of the present invention compared with the prior art are as follows: The present invention obtains enterprise internal information and external market data through multiple channels to ensure that the required relevant data is comprehensive, real-time and of high quality; processes the collected data, extracts enterprise characteristics and business opportunity characteristics from it to form input characteristics of a data model; inputs the extracted enterprise and business opportunity characteristics into a business opportunity recommendation model, performs business opportunity matching and transaction probability prediction through machine learning algorithms to obtain a business opportunity matching score and a transaction possibility; provides accurate business opportunity recommendations based on the matching score and transaction probability output by the model, adjusts the prediction strategy according to real-time market changes, and optimizes the decision-making process; continuously optimizes feature analysis and model prediction accuracy by dynamically adapting and periodically updating the model, combining new data with user feedback, to help enterprises maintain a decision-making advantage in competition.

[0035] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0037] Figure 1 It is a schematic diagram of the application scenario of the business opportunity recommendation method provided by the embodiment of the present invention;

[0038] Figure 2 It is a schematic flowchart of the business opportunity recommendation method provided by the embodiment of the present invention;

[0039] Figure 3 It is a schematic sub-flowchart of the business opportunity recommendation method provided by the embodiment of the present invention;

[0040] Figure 4 It is a schematic sub-flowchart of the business opportunity recommendation method provided by the embodiment of the present invention;

[0041] Figure 5 It is a schematic sub-flowchart of the business opportunity recommendation method provided by the embodiment of the present invention;

[0042] Figure 6 It is a schematic block diagram of the business opportunity recommendation device provided by the embodiment of the present invention;

[0043] Figure 7A schematic block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0046] It should also be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0047] It should be further understood that the term "and / or" used in the present description and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0048] See also Figure 1 and Figure 2 , Figure 1 A schematic diagram of an application scenario of the business opportunity recommendation method provided in an embodiment of the present invention. Figure 2 A schematic flow chart of a business opportunity recommendation method provided in an embodiment of the present invention. The business opportunity recommendation method is applied in a server. The server interacts with the terminal to perform data, extracts enterprise features and business opportunity features by collecting and analyzing multi-dimensional data of enterprises and business opportunities, and uses feature engineering to process data to ensure the accuracy and availability of the data. By inputting the processed features into a business opportunity recommendation model based on deep learning, the model can accurately match business opportunities and predict transaction probabilities, thereby improving recommendation accuracy. The model evaluates feature importance through XGBoost, selects the most influential features, and optimizes matching effects. The deep learning architecture supports dynamic adaptation to new data changes, continuously updates the model, and ensures the real-time and accuracy of recommendations. Ultimately, the business opportunity recommendation system enhances the decision-making ability of enterprises in competition, improves decision-making efficiency and the conversion rate of business opportunities.

[0049] Figure 2It is a schematic flowchart of the business opportunity recommendation method provided by an embodiment of the present invention. As Figure 2 shown, the method includes the following steps S110 to S140.

[0050] S110. Obtain internal data and external data of the enterprise to obtain relevant data.

[0051] In this embodiment, the relevant data includes internal data and external data of the enterprise. Specifically, the internal data of the enterprise refers to all relevant information generated during the enterprise's own operation process. Such data is crucial for understanding the current situation, needs, and preferences of the enterprise and is the basis for building a personalized business opportunity recommendation system. Specifically:

[0052] Customer Relationship Management (CRM) system: Extract customer basic information, purchase history, service records, etc. from the CRM system, which helps to understand the customer's business model and needs.

[0053] Enterprise Resource Planning (ERP) system: The ERP system provides in-depth insights into aspects such as the supply chain, inventory level, and production efficiency, and can help the enterprise evaluate its resource allocation situation.

[0054] Data of the sales and marketing departments: Integrate information such as the effects of marketing activities and feedback on advertising placements to help identify potential market demands and development trends.

[0055] Financial statements: Analyze the financial situation of the enterprise, such as indicators such as revenue, profit, and cash flow, which can provide economic background support for business opportunity evaluation.

[0056] Internal transaction records: Collect documents such as purchase orders and sales invoices to reflect the purchase and sales dynamics of the enterprise.

[0057] External data comes from third-party channels outside the enterprise and usually contains broader socio-economic environment information, which is very important for broadening the vision and discovering new business opportunities. The main channels are:

[0058] Publicly available data sources: Utilize official resources such as published statistical data and industry reports, as well as publicly available information on the Internet captured by legal means, such as news reports and social media discussions.

[0059] Third-party data service providers: Subscribe to professional business information services or cooperate with data providers in specific fields to obtain high-quality data products such as market research and competitor intelligence.

[0060] Social platform API interfaces: Connect to the official APIs provided by social networks such as Twitter and LinkedIn to monitor unstructured data such as the trend of public opinion and brand mention rate.

[0061] Partner data sharing: Establish partnerships with other enterprises, exchange non-sensitive business data with each other, and promote common growth.

[0062] Once sufficient internal and external data has been collected, the next step is to merge these scattered data sets and perform the necessary cleaning and transformation operations to facilitate subsequent analysis and modeling:

[0063] Ensure that data fields from different sources are consistent and in a unified format so that they can be effectively utilized on the same platform. Eliminate duplicates, correct error values, fill in missing values, and ensure data quality. Create new derived features based on the original data, such as calculating growth rates and predicting future trends, to enhance the richness of model inputs. Standardize numerical features to ensure that each feature is within a similar scale range and improve the performance of machine learning algorithms.

[0064] In summary, in step S110, by comprehensively collecting internal and external data resources of the enterprise through multiple channels and after a series of preprocessing operations, a high-quality and insightful data set is finally formed, providing a solid foundation for the construction of the business opportunity recommendation system. This process not only enhances the value of the data but also lays a crucial first step towards achieving more intelligent and accurate business opportunity matching and transaction prediction.

[0065] S120. Extract enterprise features, business opportunity features, and interaction features from the relevant data.

[0066] In this embodiment, enterprise features include static attributes such as the scale, industry, geographical location, historical cooperation, and financial status of the enterprise, while business opportunity features contain dynamic information such as the category, demand, budget, expected cooperation period, and market potential of the business opportunity.

[0067] Specifically, the enterprise scale is classified according to the number of employees, annual revenue, or market share, etc., such as small, medium, and large enterprises. The data is sourced from enterprise registration information, annual financial reports, etc. It can be numerically processed (such as using categorical numerical representations) and standardized, or one-hot encoded.

[0068] The industry category refers to the industry in which the enterprise is located, such as manufacturing, finance, healthcare, etc. The data is sourced from enterprise registration information, industry classification data, etc. One-hot encoding or label encoding is used to convert it into a numerical feature.

[0069] The geographical location refers to the geographical location of the enterprise, including provinces and cities, etc.; the data is sourced from enterprise registration information, GIS data, etc.; through hierarchical coding or geocoding techniques, geographical information is converted into numerical features (such as longitude and latitude).

[0070] The historical cooperation record refers to the historical cooperation data of an enterprise, reflecting the number of cooperations, depth, and duration. The data is sourced from the enterprise's internal transaction records, CRM system, etc. By counting the number of cooperations or calculating the average duration of cooperation, it is converted into numerical features.

[0071] The financial condition refers to the financial health of an enterprise, such as profit margin, debt ratio, etc. The data is sourced from financial statements, credit ratings, etc. The financial indicators are standardized, or they are integrated into a financial health index.

[0072] The business opportunity category refers to the specific type of business opportunity, such as equipment procurement, software development, etc. The data is sourced from business opportunity release platforms, internal business opportunity management systems, etc. One-hot encoding or label encoding is used to convert it into numerical features.

[0073] The demand volume refers to the demand or scale of a business opportunity, such as the purchase quantity, service demand, etc. The data is sourced from business opportunity description information, enterprise demand analysis reports, etc. The demand volume is numericalized and standardized.

[0074] The budget range refers to the budget amount set by an enterprise for a business opportunity. The data is sourced from business opportunity release information, enterprise budget plans, etc. The budget is numericalized and standardized, or interval encoding is used to represent different budget intervals.

[0075] The expected cooperation period refers to the expected cooperation time period of a business opportunity, such as several months or years. The data is sourced from business opportunity release information, project management systems, etc. The cooperation period is converted into days or months and standardized.

[0076] The market potential refers to the potential of a business opportunity in the market, usually referring to the growth space or market demand expectation of this business opportunity. The data is sourced from market research reports, industry analysis, etc. Based on market data, a potential score is generated and standardized.

[0077] The historical interaction record between an enterprise and a business opportunity refers to the historical interaction between the enterprise and the business opportunity or similar business opportunities, such as the number of interactions, interaction quality, etc. The data is sourced from the CRM system, enterprise communication records, etc. The number of interactions is counted or the interaction quality score (such as customer satisfaction, conversion rate, etc.) is calculated.

[0078] The feedback evaluation refers to the feedback of an enterprise on previous business opportunities, such as satisfaction score, cooperation evaluation, etc. The data is sourced from enterprise feedback surveys, evaluation systems after the completion of business opportunities, etc. The evaluation score is numericalized and standardized, or the sentiment tendency is extracted through sentiment analysis.

[0079] In one embodiment, please refer to Figure 3 , the above step S120 may include steps S121 to S124.

[0080] S121. Extract data on the scale, industry category, geographical location, historical cooperation records, and financial status of the enterprise from the relevant data to obtain initial enterprise characteristics;

[0081] S122. Extract information on business opportunity categories, demand quantities, budget ranges, expected cooperation cycles, and market potential from the relevant data to obtain initial business opportunity characteristics;

[0082] S123. Analyze the historical interaction records and feedback evaluations between the enterprise and the business opportunity in the relevant data to obtain initial interaction characteristics;

[0083] In this embodiment, the construction process of interaction characteristics aims to improve the deep learning model's understanding of the relationship between the enterprise and the business opportunity through refined feature engineering, thereby enhancing the accuracy of business opportunity recommendation and transaction prediction. Specifically, the construction process of interaction characteristics is as follows:

[0084] The interaction frequency and behavior quality between the enterprise and the business opportunity are key factors. Therefore, first, count the monthly contact frequency between the enterprise and the business opportunity to quantify the enterprise's interaction activity. Interaction behaviors include:

[0085] Online browsing: The enterprise's online browsing behavior of the business opportunity. Such behavior is used as a basic interaction method and is given a basic weight.

[0086] Consultation: Consultation initiated by the enterprise, such as asking questions or requesting more information about the business opportunity, is given a higher weight.

[0087] Request for quotation: The request for quotation initiated by the enterprise belongs to a relatively in-depth business communication and is given an even higher weight.

[0088] Business negotiation: When the enterprise enters the business negotiation stage with the business opportunity, it is given an even higher weight.

[0089] On-site inspection: If the enterprise arranges an on-site inspection of the business opportunity, this is the most in-depth interaction and is given the highest weight.

[0090] To ensure that the historical interaction has greater reference value for the current business opportunity, a time decay factor is introduced for all interaction behaviors, that is, the more recent the interaction, the higher its weight. Through the weighted calculation of interaction frequency, behavior weight, and time decay, a comprehensive enterprise activity score is finally obtained, which can reflect the interaction intensity between the enterprise and the business opportunity within a specific time period.

[0091] The interaction quality of the enterprise not only depends on the quantity of interactions but also needs to focus on the depth and effectiveness of the interactions. Therefore, the following evaluation dimensions are adopted:

[0092] Interaction duration: The duration of each interaction reflects the degree of attention of the enterprise to the business opportunity. The longer the time, the deeper the interaction usually indicates.

[0093] Completeness of information exchange: During the interaction process, whether the requirement documents, technical solutions, quotation plans, etc. involved are completely exchanged affects the depth of cooperation.

[0094] Response timeliness: The feedback speed and timeliness of enterprises to business opportunities are also important indicators for measuring the quality of interaction.

[0095] By calculating the weighted average of these dimensions, a quantitative index of interaction quality is finally obtained. This index not only reflects the frequency of interaction, but also captures the effectiveness and depth of interaction, helping the model understand the cooperation intentions and behavioral characteristics of enterprises.

[0096] To further refine the understanding of enterprise behavior, it is necessary to conduct a quantitative analysis of the cooperation satisfaction of enterprises. Cooperation satisfaction includes both explicit evaluations and implicit feedback:

[0097] Explicit evaluations: Include direct scores in past cooperation, customer satisfaction surveys, text evaluations, etc.

[0098] Implicit feedback: By analyzing the behavior of enterprises such as complaint records, renewal situations, and whether they recommend other customers, the satisfaction of enterprises with business opportunities can be reflected. In addition, it is also very important to conduct sentiment analysis on evaluation texts and extract the sentiment tendency (such as positive or negative sentiment) from them.

[0099] By combining explicit and implicit feedback through weighting, a comprehensive satisfaction index is finally obtained, which can help the model better understand the performance of enterprises in past cooperation and their overall satisfaction with business opportunities.

[0100] The decision-making process of enterprises is crucial for business opportunity matching and transaction prediction. To accurately capture the decision-making behavioral characteristics of enterprises, the following indicators are very key:

[0101] Decision-making cycle: The average time from the first contact with a business opportunity to the final decision can reflect the decision-making speed of enterprises.

[0102] Complexity of the decision-making chain: The number of departments and levels involved in the decision-making process. The more complex the decision-making chain usually means greater decision-making difficulty and longer cycle.

[0103] Price sensitivity: By analyzing the deviation between the historical transaction price of an enterprise and the market average price, the sensitivity of the enterprise to price changes is measured.

[0104] These decision-making behavioral characteristics can help the model identify the decision-making patterns of enterprises in business opportunity selection, thus providing a more accurate basis for the screening and matching of business opportunities.

[0105] An enterprise's preference for different types of business opportunities is often closely related to the characteristics of its industry, so it is necessary to conduct a quantitative analysis of the enterprise's industry preferences:

[0106] Industry distribution entropy: Calculates the entropy of industry distribution of historical interactive business opportunities, reflecting the company's focus or diversity on a certain industry.

[0107] Budget range distribution: Understand the budget preferences of enterprises by counting the budget ranges of historical closed business opportunities.

[0108] Innovation Receptivity: Assessing the company's receptivity to new business opportunity categories can help predict whether the company is willing to accept more innovative business opportunities.

[0109] These industry preference indicators will help the model more accurately grasp the business opportunity needs and selection preferences of enterprises and optimize the business opportunity recommendation process.

[0110] The ultimate goal of business opportunity conversion is to successfully close a deal, so it is crucial to understand the conversion characteristics of business opportunities. The key conversion features extracted include:

[0111] Historical transaction conversion rate: The historical transaction conversion rate of different types of business opportunities can reflect the potential of the business opportunities themselves.

[0112] Average advancement time: The advancement time of business opportunities at each stage can help evaluate the progress of business opportunity conversion.

[0113] Main reasons for unsuccessful cases: Recording the reasons for unsuccessful business opportunities (such as too high price, competitor advantage, etc.) can provide a reference for the evaluation and decision-making of future business opportunities.

[0114] By extracting these conversion features, the model can more accurately predict the likelihood of closing a business opportunity, the duration of promotion, and potential risks.

[0115] The above interaction features will eventually be passed as input data to the deep learning model. Specifically:

[0116] In the business opportunity matching stage, these features will be used to adjust the matching weights, screen business opportunities that meet the enterprise's decision-making model, and predict the potential degree of matching.

[0117] In the deal prediction stage, the model uses these features to evaluate the deal possibility of business opportunities, predict the duration of promotion, and identify potential risk points.

[0118] During the business opportunity matching process, interaction features affect the matching results through multiple dimensions. First, based on the interaction activity and behavior quality scores of enterprises, the weight coefficients for business opportunity matching are dynamically adjusted. For business opportunity types with higher interaction frequencies and good interaction quality, higher weights are assigned during matching to increase the likelihood of recommending such business opportunities. Second, according to the decision-making behavior characteristics of enterprises, such as decision-making cycles and price sensitivities, business opportunities that conform to the enterprise's decision-making mode are screened. For example, for enterprises with longer decision-making cycles, large business opportunities that require a longer investigation period are recommended first; while for enterprises with high price sensitivities, business opportunities with price advantages are emphasized. Third, by analyzing the industry preference indicators of enterprises, such as industry distribution entropy values and budget interval distributions, the potential matching degrees of enterprises for different types of business opportunities are predicted, thus achieving more accurate business opportunity screening.

[0119] In the deal closing prediction stage, the roles of interaction features are mainly reflected in the following aspects: First, based on the conversion features in historical interaction data, such as the deal closing conversion rates and average promotion durations of different types of business opportunities, a deal closing probability prediction model is constructed. This model evaluates the deal closing possibility of the current business opportunity by analyzing interaction feature indicators such as the interaction quality between the enterprise and the business opportunity and satisfaction evaluations. Second, the decision-making link complexity indicators of enterprises are used to predict the promotion cycle of business opportunities, providing a time expectation reference for the sales team. Third, a risk warning mechanism is established in combination with the cause analysis of unclosed cases. For example, when certain interaction patterns that are prone to failure in history are detected, the system will promptly identify the risks and provide warnings. Additionally, by continuously tracking and analyzing changes in the interaction behaviors of enterprises, such as sudden drops in interaction frequencies and slower response timings, the deal closing probability prediction values are dynamically adjusted to achieve real-time optimization of the prediction results.

[0120] By deeply integrating interaction features with enterprise features and business opportunity features, the model can more comprehensively understand the decision-making preferences and behavior patterns of enterprises. For example, when the enterprise scale feature shows a large enterprise and the interaction features reflect a high acceptance of innovative business opportunities, the model will increase the matching weight of innovative business opportunities. Another example is that when the enterprise's financial condition is good and the historical interaction records show a high conversion rate for high-end business opportunities, the model will give priority to recommending business opportunities with a higher budget interval. This feature integration mechanism enables the recommendation system to make decisions based on multi-dimensional feature combinations, thereby providing business opportunity suggestions that are more in line with the actual needs of enterprises.

[0121] Through such refined feature engineering, the deep learning model can more accurately understand the behavior patterns of enterprises and the conversion laws of business opportunities, thereby improving the accuracy of business opportunity recommendation and prediction.

[0122] By constructing a refined and multi-dimensional interaction feature system, the behavioral patterns and enterprise preference information in historical interaction data can be fully utilized to provide a more reliable basis for business opportunity recommendation and deal prediction. This solution not only enhances the depth of feature engineering but also improves the prediction accuracy of the recommendation system, providing strong data support for decision-makers.

[0123] In this embodiment, the interaction features obtained by analyzing the historical interaction records and feedback evaluations between enterprises and business opportunities will serve as important input features for the deep learning model. Specifically, the construction process of the interaction features is as follows:

[0124] First, the interaction activity of the enterprise is quantitatively calculated. Specifically, the monthly contact frequency between the enterprise and the business opportunity is counted, including interaction behaviors such as online browsing, consultation, quotation requests, etc. At the same time, different weight coefficients are assigned to different types of interaction behaviors. For example, the browsing behavior is assigned a basic weight, the consultation behavior is assigned a higher weight, the business negotiation is assigned an even higher weight, and the on-site inspection is assigned the highest weight. In addition, a time decay factor is introduced to make recent interaction behaviors have higher reference value. By performing weighted calculations on the interaction frequency, behavior weights, and time decay factor, the activity score of the enterprise is finally obtained.

[0125] Second, the interaction quality is comprehensively evaluated. Specifically, the duration of each interaction is recorded, the completeness of information exchange in links such as requirement documents, technical solutions, and quotation plans is counted, and the response timeliness of the enterprise to business opportunity information is calculated. The scores of these dimensions are used to obtain a quantitative index of interaction quality through weighted averaging.

[0126] Third, the cooperation satisfaction is quantified in multiple dimensions. Specifically, the direct ratings and text evaluations in historical cooperation are collected as explicit evaluation data, and the complaint records, renewal situations, recommendation behaviors, etc. are analyzed as implicit feedback data, and sentiment analysis is performed on the evaluation text to extract the sentiment tendency. By performing weighted combination on these data, a comprehensive satisfaction index is obtained.

[0127] At the same time, the decision-making behaviors of the enterprise are characterized. Specifically, the average duration from the first contact to the final decision is counted as the decision cycle index, the number of departments and levels involved in the decision-making process is recorded as the decision-making link complexity index, and the deviation between the historical transaction price and the market average price is analyzed to obtain the price sensitivity index. These indicators together constitute the decision-making behavior characteristics of the enterprise.

[0128] In addition, the industry preferences of the enterprise are quantitatively analyzed. Specifically, the entropy value of the industry distribution of historical interaction business opportunities is calculated to represent the concentration in the vertical field, the budget interval distribution of historical transaction business opportunities is counted to reflect the budget tendency, and the acceptance degree of the enterprise for new business opportunity categories is evaluated to measure the innovation acceptance. These dimensions together constitute the preference intensity index of the enterprise.

[0129] Finally, extract the business opportunity conversion features. Specifically, count the historical conversion rates of different types of business opportunities, calculate the average promotion duration of each stage, and record the distribution of the main reasons for unclosed cases, so as to construct a complete conversion feature vector.

[0130] The above interaction features will be used as the input of the deep learning model. In the business opportunity matching stage, they are used to adjust the matching weights, screen business opportunities that conform to the enterprise decision-making mode, and predict the potential matching degree; in the deal prediction stage, they are used to evaluate the deal possibility, predict the business opportunity promotion duration, and predict potential risks. In this way, the interaction features can help the model more accurately understand the business opportunity preferences and behavior patterns of the enterprise, and significantly improve the accuracy of recommendation and prediction.

[0131] In this embodiment, by constructing such a refined interaction feature system and using it as the model input, the behavior patterns and preference information contained in the historical interaction data can be fully utilized, thereby providing a more reliable basis for business opportunity recommendation. This solution not only improves the depth of feature engineering, but also provides important support for improving the prediction accuracy of the recommendation system.

[0132] S124. Process the missing values, detect outliers, normalize the numerical features, and perform encoding conversion on the categorical features for the initial enterprise features, the initial business opportunity features, and the initial interaction features respectively, so as to obtain enterprise features, business opportunity features, and interaction features.

[0133] In this embodiment, step S120 focuses on extracting and constructing two main types of features for the business opportunity recommendation system from the collected relevant data: enterprise features and business opportunity features. This step is a key link in the overall system design because high-quality features directly determine the learning effect and final prediction performance of the model. Specifically, enterprise features reflect the attributes and historical behaviors of the enterprise, while business opportunity features describe the characteristics of the business opportunities themselves.

[0134] Extract important information about the enterprise from the relevant data to construct an initial set of enterprise features. These features include but are not limited to:

[0135] Enterprise scale: Classified as large, medium, or small based on indicators such as the number of employees, annual revenue, or market share.

[0136] Industry category: The industry field to which the enterprise belongs, such as manufacturing, IT, finance, healthcare, etc.

[0137] Geographical location: The geographical area where the enterprise is located, which can be refined to the national, provincial, city, etc. levels.

[0138] Historical cooperation records: The historical cooperation situations of the enterprise with other enterprises or business opportunities, such as the number of cooperations, depth, and duration.

[0139] Financial status: Indicators reflecting the financial health of the enterprise, such as profit margin, debt ratio, cash flow, etc.

[0140] The data sources of these features are extensive, including but not limited to enterprise registration information, annual financial reports, market research data, CRM system data, etc. For categorical variables (such as enterprise size, industry category), one-hot encoding or label encoding is usually used to convert them into numerical representations; while for continuous variables (such as financial scores), standardization processing is required.

[0141] Next, focus on the characteristics of the business opportunity itself and extract initial business opportunity features, which will help the system understand the uniqueness of each business opportunity. Business opportunity features cover:

[0142] Business opportunity category: The specific business area or activity type involved in the business opportunity, such as equipment procurement, software development, market promotion, etc.

[0143] Demand quantity: The demand scale or quantity of the business opportunity, such as the quantity purchased or the demand quantity of the service.

[0144] Budget range: The capital budget reserved by the enterprise for this business opportunity.

[0145] Expected cooperation period: The length of time expected to complete this business opportunity.

[0146] Market potential: The potential development space or growth expectation of the business opportunity in the market.

[0147] The data sources of business opportunity features include business opportunity release platforms, enterprise internal business opportunity management systems, market research reports, etc. Similarly, for categorical features (such as business opportunity category), coding techniques are used to convert them into numerical forms, while for continuous features (such as budget range), standardization processing is carried out.

[0148] In addition to static enterprise features and business opportunity features, interaction features capture the dynamic relationship between the enterprise and the business opportunity. This part of the features comes from the historical interaction records and feedback evaluations between the enterprise and the business opportunity, such as:

[0149] Historical interaction records: Count the number of interactions, calculate the interaction quality score (such as satisfaction), conversion rate, etc.

[0150] Feedback evaluation: The feedback evaluation of the enterprise on previous business opportunities, including satisfaction scores, cooperation evaluations, etc.

[0151] Through sentiment analysis technology, the feedback content in text form can be further mined and converted into quantitative indicators, thus enriching the feature set. Such features help improve the model's predictive ability for the possibility of future cooperation.

[0152] Finally, after obtaining the preliminary enterprise features, business opportunity features, and interaction features, a series of data processing tasks are still required to ensure the quality and applicability of the features:

[0153] Missing value handling: Use mean filling, median filling, or interpolation method to fill in the missing data points, and delete samples with serious missing values if necessary.

[0154] Outlier detection: Use statistical methods such as box plots or Z-scores to identify and handle outliers to ensure the reliability of the data.

[0155] Data normalization: Apply standardization (Z-score standardization) or Min-Max normalization to numerical features to make all features within a similar scale range.

[0156] Categorical encoding: For categorical features, use one-hot encoding or label encoding to convert them into numerical features for easy processing by machine learning algorithms.

[0157] In summary, through in-depth analysis and processing of relevant data, step S120 effectively constructs a comprehensive feature set including enterprise features, business opportunity features, and interaction features, providing a solid foundation for subsequent model training. This process not only improves the quality of the features but also creates conditions for achieving more intelligent and accurate business opportunity matching and transaction prediction.

[0158] S130. Input the enterprise features, the business opportunity features, and the interaction features into the business opportunity recommendation model for business opportunity matching prediction and transaction prediction to obtain a business opportunity matching score and a transaction probability prediction value.

[0159] In this embodiment, the business opportunity matching score: This is the result of a binary classification problem, indicating the degree of match between the business opportunity recommended to a specific enterprise and its needs. The output is a value between 0 and 1, where a value close to 1 means a high degree of match, and a value close to 0 means a low degree of match. The transaction probability prediction value is also the result of a binary classification problem, but this time it focuses on whether the business opportunity can ultimately be successfully converted into an actual transaction. The output is also a probability value between 0 and 1, and the higher this value, the greater the likelihood of the business opportunity being successfully transacted.

[0160] Specifically, the business opportunity matching score and the deal closing probability prediction value can be defined as the Y value, which is used to represent the target variable in the business opportunity recommendation system and covers two stages: business opportunity matching and deal closing: Business opportunity matching: It represents the degree of match between the recommended business opportunity and the enterprise's needs. Deal closure prediction: It represents the possibility of the recommended business opportunity being finally closed.

[0161] Define the Y value as a two-dimensional vector, including two dimensions: Y 1 (Whether the business opportunity is matched): The value is 0 or 1. 0 means not matched, and 1 means matched. Y 2 (Whether the business opportunity is closed): The value is 0 or 1. 0 means not closed, and 1 means closed. As shown in Table 1.

[0162] Table 1. Y value

[0163]

[0164] In one embodiment, please refer to Figure 4 , the above step S130 may include steps S131 to S132.

[0165] S131. Perform feature engineering processing on the enterprise features, the business opportunity features, and the interaction features to obtain a processing result.

[0166] In this embodiment, the processing result refers to the final form of the enterprise features and business opportunity features after feature engineering processing, and these features have been optimized, transformed, and are ready to be input into the business opportunity recommendation model.

[0167] Specifically, using the feature importance evaluation of XGBoost, normalize the importance of each of the enterprise features, the business opportunity features, and the interaction features, sort each of the enterprise features, the business opportunity features, and the interaction features from largest to smallest according to importance, calculate the cumulative importance, and screen all features whose cumulative importance meets the requirements to obtain the processing result. As shown in Table 2.

[0168] Table 2. Processing result

[0169]

[0170]

[0171] During the feature engineering process, machine learning algorithms such as XGBoost are used to evaluate the importance of each feature. This step aims to determine which features have the greatest contribution to predicting the target variables (i.e., business opportunity matching score and deal probability prediction value). By calculating the importance of each feature and normalizing it, the features can be sorted according to their importance, and then the features with cumulative importance reaching a certain threshold can be selected to remove redundant or irrelevant features, so as to optimize the model performance and reduce complexity.

[0172] All categorical features (such as enterprise size, industry category, etc.) have been converted into numerical representations, and numerical features (such as the number of historical cooperations, financial scores, etc.) have been standardized or normalized to ensure that different features are on the same scale. Missing values in the data have been filled (for example, using the mean, median filling or interpolation method), and outliers have been identified and processed to ensure the quality of the data and the stability of model training. The importance of features is evaluated by methods such as XGBoost, the importance of each feature is normalized, and the features are sorted according to their importance. Features with cumulative importance meeting the requirements are selected to remove redundant and irrelevant features, thereby reducing model complexity and improving prediction performance. If applicable, new features describing the relationship between the enterprise and the business opportunity (such as historical interaction records, feedback evaluations, etc.) are created to capture additional information that may affect business opportunity matching and deal probability. Finally, a feature set is formed, which consists of the above-mentioned processed features and can be directly used for model training. This feature set is the basis for the model to understand the characteristics of enterprises and business opportunities and is crucial for improving the accuracy of business opportunity matching score and deal probability prediction.

[0173] S132. Input the processing result into the business opportunity recommendation model for business opportunity matching prediction and deal prediction to obtain a business opportunity matching score and a deal probability prediction value.

[0174] In this embodiment, the above-mentioned business opportunity recommendation model is obtained by acquiring historical relevant data, extracting enterprise features, business opportunity features, and interaction features, and training a deep learning architecture with the processed data as a sample set.

[0175] In one embodiment, please refer to Figure 5 , the above-mentioned business opportunity recommendation model is obtained by acquiring historical relevant data, extracting enterprise features, business opportunity features, and interaction features, and training a deep learning architecture with the processed data as a sample set, including steps S1321 to S1323.

[0176] S1321. Acquire historical relevant data, extract enterprise features, business opportunity features, and interaction features, and perform feature engineering processing to obtain a sample set.

[0177] In this embodiment, the acquisition of the above-mentioned history-related data, the extraction of enterprise features and business opportunity features, and the feature engineering processing are the same as the processing in steps S110, S120, and S31 above, and will not be elaborated here.

[0178] S1322. Construct a deep learning architecture and define a loss function.

[0179] In this embodiment, the structure of the deep learning architecture includes an input layer, a feature fusion layer, a deep learning network, and an output layer. The deep learning network includes a first fully connected layer, a first batch normalization layer, a first activation function ReLU, a first Dropout layer, a second fully connected layer, a second batch normalization layer, a second activation function ReLU, and a second Dropout layer. The output layer includes two branches and respectively uses the sigmoid activation function.

[0180] Specifically, construct a deep learning architecture based on the following structure:

[0181] Input layer: Receive the enterprise features and business opportunity features processed by feature engineering and input them as numerical features and categorical features respectively.

[0182] Feature fusion layer: Concatenate the enterprise features and business opportunity features to form a unified feature vector.

[0183] Deep learning network:

[0184] First fully connected layer: Process the fused feature vector for preliminary feature extraction.

[0185] First batch normalization layer: Normalize the output of the fully connected layer to improve training stability.

[0186] First activation function ReLU: Introduce non-linearity to improve the model's expressive ability.

[0187] First Dropout layer: Prevent overfitting and improve the model's generalization ability.

[0188] Second fully connected layer, second batch normalization layer, second activation function ReLU, second Dropout layer: Further extract and normalize features.

[0189] Output layer: Divide into two independent branches:

[0190] Business opportunity matching prediction layer: Use the sigmoid activation function to output the business opportunity matching score (0 or 1).

[0191] Transaction prediction layer: Use the sigmoid activation function to output the transaction probability (0 or 1).

[0192] In addition, to achieve multi-objective optimization of business opportunity matching and transaction prediction, a weighted combination multi-objective loss function is designed: Loss = α · business opportunity matching loss + (1 - α) · transaction prediction loss; where α is a hyperparameter that controls the weight distribution between the business opportunity matching loss and the transaction prediction loss, and its value range is [0, 1].

[0193] Both the business opportunity matching loss and the transaction prediction loss use binary cross-entropy loss to measure the difference between the prediction result and the actual label; where where is the actual business opportunity matching score (0 or 1) of the i-th sample, is the business opportunity matching score predicted by the model, and N is the number of samples.

[0194] where is the actual transaction label (0 or 1) of the i-th sample, is the transaction probability predicted by the model.

[0195] S1323. Use the sample set to train the deep learning architecture, adopt the loss function in combination with the trained deep learning architecture, and determine the hyperparameters in the loss function through grid search or cross-validation method to obtain a business opportunity recommendation model.

[0196] Finally, use the sample set prepared in the previous steps to train this deep learning architecture. During the training process, adopt the loss function in combination with the trained deep learning architecture, and determine the hyperparameter α in the loss function through grid search or cross-validation method to ensure that the model can achieve optimal performance in both business opportunity matching and transaction prediction tasks. By adjusting α, the importance of the two objectives can be flexibly balanced, so as to achieve the optimal balance of the two objectives.

[0197] By adjusting the hyperparameter α, the importance of business opportunity matching and transaction prediction can be flexibly balanced:

[0198] Increase α: Make the model focus more on optimizing the accuracy of the business opportunity matching score.

[0199] Decrease α: Make the model focus more on optimizing the accuracy of the transaction probability.

[0200] In practical applications, the value of α is usually tuned through methods such as grid search or cross-validation to find the best weight distribution and achieve the optimal balance of the two objectives.

[0201] For example: Assume α = 0.7. For a sample:

[0202] Actual business opportunity matching score Y1 = 1, predicted score

[0203] Actual transaction label Y 2 = 1, predicted probability

[0204] Then: Business opportunity matching loss = -[1·log(0.8)+(1 - 1)·log(1 - 0.8)] = -log(0.8) ≈ 0.2231; Transaction prediction loss = -[1·log(0.75)+(1 - 1)·log(1 - 0.75)] = -log(0.75) ≈ 0.2877; Total loss = 0.7·0.2231 + 0.3·0.2877 ≈ 0.15617 + 0.08631 = 0.24248

[0205] Through this loss function design, the model can optimize both business opportunity matching and transaction prediction objectives during the training process, improving the performance of the overall recommendation system.

[0206] In this embodiment, to comprehensively evaluate the performance of the model, the following key indicators are used in this embodiment:

[0207] AUC (Area Under Curve): Measures the ability of the model to distinguish between positive and negative samples.

[0208] Accuracy: The proportion of correctly predicted samples to the total number of samples.

[0209] Precision: The proportion of actually positive samples among the predicted positive samples.

[0210] Recall: The proportion of actually positive samples that are correctly predicted as positive samples.

[0211] F1-score: The harmonic mean of precision and recall, comprehensively evaluating the performance of the model.

[0212] Misclassified sample analysis: Deeply analyze the misclassified samples, identify the weaknesses of the model, and provide guidance for subsequent optimization. For example, if the prediction accuracy of business opportunities in a certain industry category is low, it may be necessary to increase the sample size of this category or optimize the feature engineering.

[0213] Re-evaluate feature importance: According to the results of error analysis, re-evaluate the influence of features and adjust the feature selection strategy to ensure that the model can utilize the most valuable features.

[0214] Configure servers for the production environment and install the required software dependencies, such as Python environment, deep learning frameworks (such as TensorFlow or PyTorch), and database drivers, etc.

[0215] Package the model and its dependencies through container technologies such as Docker to ensure the consistency and portability of the deployment environment.

[0216] Design RESTful APIs to provide business opportunity matching scoring and transaction probability prediction services. The APIs include the following main endpoints:

[0217] Accept the characteristic data of enterprises and business opportunities and return the prediction results.

[0218] Optimize the response speed of the APIs to meet the requirements of real-time prediction. Adopt a caching mechanism (such as Redis) to reduce repeated calculations and improve the response efficiency.

[0219] Integrate the prediction module with the existing business system to ensure the seamless connection of data streams and functions. For example, embed the prediction results into the CRM system for the reference of the sales team. Conduct comprehensive functional tests to ensure the correctness and stability of each module, including interface tests, data stream tests, boundary condition tests, etc. Through stress tests, load tests, etc., evaluate the performance of the system under high concurrency to ensure its scalability and stability.

[0220] Deploy monitoring tools (such as Prometheus and Grafana) to monitor the running status and performance metrics of the system in real time (such as API response time, error rate, system load, etc.). Record the system running logs for convenient problem tracking and troubleshooting. Adopt a centralized log management system (such as the ELK stack) to collect, store, and visually analyze the logs.

[0221] Set a regular data collection plan to ensure the timeliness and accuracy of the model input data. For example, perform data updates at midnight every day to ensure that the latest data is incorporated into model training in a timely manner. Continuously monitor the data quality, promptly handle data anomalies, and ensure the integrity of the data warehouse. Use automated tools to generate data quality reports regularly.

[0222] According to the data change situation, conduct model retraining regularly, update the model parameters, and ensure its efficiency. It is recommended to conduct a comprehensive retraining every quarter or half year. Adopt an incremental learning method to use new data to quickly update the model and reduce the retraining time. For example, use online learning algorithms to update the model parameters in real time to adapt to the dynamic changes of the data.

[0223] Establish a user feedback channel to collect users' opinions and suggestions on the recommendation results. For example, user feedback can be obtained through in-app feedback buttons, regular questionnaires, etc. Analyze user feedback, identify the deficiencies of the system, and guide the subsequent optimization direction. Use natural language processing technologies for sentiment analysis and topic extraction to discover potential problems and improve the system.

[0224] S140. Output the business opportunity matching score and the predicted value of the closing probability.

[0225] After completing the above steps, the business opportunity recommendation model is ready to accept new inputs. When new enterprise features and business opportunity features are input into the model, the model will perform business opportunity matching prediction and closing prediction based on the knowledge obtained from previous training, and finally output the business opportunity matching score and the predicted value of the closing probability to provide decision-making support for the enterprise.

[0226] In summary, through the above steps, not only a deep learning model that can simultaneously perform business opportunity matching prediction and closing prediction is constructed, but also it is ensured that the model can adapt to different business needs and focus on optimizing a specific objective by adjusting the hyperparameter α.

[0227] In this embodiment, relevant data of enterprises and business opportunities are collected in real time from the internal database and external data sources through the API interface. Web crawler technology is used to scrape public data, such as enterprise announcements, news and information, etc. Timed tasks are set to ensure the real-time nature and integrity of the data. The collected data is preprocessed and cleaned, dealing with missing values, outliers and performing data normalization. The conversion and standardization of data formats are realized to ensure data consistency. Features are extracted, processed, transformed and selected to generate a feature set suitable for model training. Feature construction and dimensionality reduction processing are carried out to improve the model training efficiency and prediction accuracy. A deep learning model is constructed, and a multi-objective loss function is defined for training and optimization. The accuracy of business opportunity matching and closing probability prediction is improved through multi-task learning. Real-time data input is received, and the trained model is used for business opportunity matching and closing probability prediction. The prediction results are output to support the business opportunity screening and decision-making support of the enterprise. The prediction results are displayed through a visual dashboard to help the enterprise more intuitively understand and analyze the recommendation results. Interactive reports and charts are provided to support the enterprise in deeply mining business opportunity data. Precise matching of business opportunities and prediction of the possibility of closing are realized, the recommendation accuracy and decision-making efficiency are improved, and it has dynamic adaptability, can be continuously updated and comprehensively analyze multi-dimensional features, enhancing the enterprise's decision-making ability in competition.

[0228] Specifically, the system uses a deep learning network to automatically extract potentially valuable features from large-scale multi-source data, greatly improving the accuracy of business opportunity matching and transaction prediction. Compared with traditional methods, the present invention can provide more accurate and timely recommendation results, significantly optimizing the accuracy of business opportunity identification and decision-making. Through a multi-objective optimization strategy, the system can simultaneously optimize business opportunity matching and transaction prediction, comprehensively improving the performance of the system. Compared with the traditional single-objective optimization method, the system can more accurately identify business opportunities that meet the needs of enterprises in business opportunity matching, and make more accurate predictions in transaction prediction, thereby helping enterprises make more scientific decisions. The method of this embodiment has a flexible retraining mechanism, which can be adjusted and optimized in real time according to market changes and feedback information from enterprises. This enables the system to maintain efficient recommendation capabilities in a rapidly changing business environment and always provide accurate business opportunity predictions. By comprehensively considering the multi-dimensional characteristics of enterprises (such as enterprise size, industry category, historical cooperation records, etc.) and the multi-dimensional characteristics of business opportunities (such as demand, budget range, etc.), the system achieves comprehensive business opportunity matching and transaction prediction, avoiding the limitations of traditional methods that rely too much on a single feature. Through accurate business opportunity recommendations and transaction forecasts, the system helps companies identify and seize high-quality business opportunities in a complex and ever-changing business environment, improves decision-making efficiency and quality, and enables companies to gain an advantage in fierce market competition.

[0229] By defining the Y value of the two dimensions, the system considers the two stages of business opportunity matching and transaction prediction at the same time, and combines the deep learning network for multi-objective optimization. Traditional methods usually only focus on one aspect of business opportunity matching or transaction prediction, while the dual-objective optimization strategy of the present invention significantly improves the overall effect of the recommendation system. In feature engineering, by integrating the multi-dimensional features of the enterprise, the multi-dimensional features of the business opportunity and the interactive features between them, and performing strict data preprocessing and standardization. This provides high-quality input data for the deep learning model. Compared with the traditional method, the system can mine more potential feature relationships and further improve the recommendation accuracy. A deep neural network architecture is adopted, including multi-layer fully connected layers, batch normalization layers, Dropout layers, etc., which can enable the network to automatically learn complex nonlinear features. Through a multi-level learning mechanism, the present invention effectively captures the complex correlation between business opportunities and enterprises, thereby significantly improving the prediction performance of the model. A weighted combination of multi-objective loss functions is designed, which can flexibly adjust the weights of business opportunity matching and transaction prediction according to different business scenarios. This design enables the system to achieve a more accurate balance in practical applications and optimize the performance of the model in a multi-objective environment.

[0230] In addition, when the data set is small or computing resources are limited, traditional machine learning methods are still a good choice. For example, algorithms such as decision trees, support vector machines (SVMs), and random forests are relatively lightweight, computationally efficient, and suitable for processing small-scale data. However, these methods may not be able to compete with deep learning methods on large-scale data sets, especially in capturing complex nonlinear relationships and feature interactions.

[0231] In some specific industries or application scenarios, business opportunity recommendation based on rule engines is still an effective solution. By setting clear rules (such as matching based on enterprise size, industry category, and other characteristics), the rule engine can quickly screen business opportunities in some simple scenarios. This method is relatively straightforward to implement, but its limitation is that the definition of rules requires manual intervention and it is difficult to handle more complex situations or dynamically changing data.

[0232] In order to avoid the complexity and long computing time of deep learning methods during training, ensemble learning methods (such as XGBoost, LightGBM, etc.) can be used as an effective alternative. Ensemble methods improve the accuracy and robustness of the overall model by combining the prediction results of multiple weak classifiers. These methods are usually easier to train than deep learning and can provide superior performance in many scenarios.

[0233] In general, although the above alternatives have advantages in some scenarios, deep learning methods are still the best technical choice for business opportunity recommendation tasks that process large-scale and high-dimensional data sets. By combining multi-source data features and multi-objective optimization design, deep learning can significantly improve the accuracy and effectiveness of recommendation systems, especially in complex data and high-demand commercial applications.

[0234] The above-mentioned business opportunity recommendation method obtains internal enterprise information and external market data through multiple channels to ensure that the required relevant data is comprehensive, real-time and high-quality; processes the collected data to extract enterprise characteristics and business opportunity characteristics from them to form the input characteristics of the data model; inputs the extracted enterprise and business opportunity characteristics into the business opportunity recommendation model, and uses machine learning algorithms to match business opportunities and predict transaction probabilities to derive business opportunity matching scores and transaction possibilities; provides accurate business opportunity recommendations based on the matching scores and transaction probabilities output by the model, and adjusts the prediction strategy according to real-time market changes to optimize the decision-making process; through dynamic adaptability and regular model updates, combined with new data and user feedback, continuously optimizes feature analysis and model prediction accuracy to help companies maintain decision-making advantages in competition.

[0235] Figure 6 FIG. 3 is a schematic block diagram of a business opportunity recommendation device 300 provided in an embodiment of the present invention. Figure 6As shown, corresponding to the above business opportunity recommendation method, the present invention also provides a business opportunity recommendation device 300. The business opportunity recommendation device 300 includes units for executing the above business opportunity recommendation method, and the device can be configured in a server. Specifically, please refer to Figure 6 , the business opportunity recommendation device 300 includes a data acquisition unit 301, a feature extraction unit 302, a prediction unit 303, and an output unit 304.

[0236] The data acquisition unit 301 is used to acquire internal data and external data of an enterprise to obtain relevant data; the feature extraction unit 302 is used to extract enterprise features, business opportunity features, and interaction features from the relevant data; the prediction unit 303 is used to input the enterprise features, the business opportunity features, and the interaction features into a business opportunity recommendation model for business opportunity matching prediction and transaction prediction to obtain a business opportunity matching score and a transaction probability prediction value; the output unit 304 is used to output the business opportunity matching score and the transaction probability prediction value.

[0237] In one embodiment, the feature extraction unit 302 includes:

[0238] A first extraction subunit is used to extract data on the scale, industry category, geographical location, historical cooperation records, and financial status of an enterprise from the relevant data to obtain initial enterprise features; a second extraction subunit is used to extract information on business opportunity categories, demand quantities, budget ranges, expected cooperation cycles, and market potentials from the relevant data to obtain initial business opportunity features; an analysis subunit is used to analyze the historical interaction records and feedback evaluations between the enterprise and the business opportunity in the relevant data to obtain initial interaction features; a preprocessing subunit is used to process missing values, detect outliers, normalize numerical features, and perform encoding conversion on categorical features for the initial enterprise features, the initial business opportunity features, and the initial interaction features respectively to obtain enterprise features, business opportunity features, and interaction features.

[0239] In one embodiment, the prediction unit 303 includes:

[0240] A feature engineering subunit is used to perform feature engineering processing on the enterprise features, the business opportunity features, and the interaction features to obtain a processing result; a matching prediction subunit is used to input the processing result into a business opportunity recommendation model for business opportunity matching prediction and transaction prediction to obtain a business opportunity matching score and a transaction probability prediction value.

[0241] In one embodiment, the feature engineering subunit is configured to normalize the importance degrees of each of the enterprise features, the business opportunity features, and the interaction features by using the feature importance evaluation of XGBoost, sort each of the enterprise features, the business opportunity features, and the interaction features in descending order of importance, calculate the cumulative importance, and filter all the features whose cumulative importance meets the requirements to obtain a processing result.

[0242] In one embodiment, the business opportunity recommendation device 300 further includes a training unit, configured to:

[0243] Obtain historical relevant data, extract enterprise features, business opportunity features, and interaction features, perform feature engineering processing to obtain a sample set; construct a deep learning architecture and define a loss function; train the deep learning architecture by using the sample set, combine the loss function with the trained deep learning architecture, and determine hyperparameters in the loss function by using a grid search or cross-validation method to obtain a business opportunity recommendation model.

[0244] It should be noted that those skilled in the art can clearly understand that the specific implementation processes of the above business opportunity recommendation device 300 and each unit can refer to the corresponding descriptions in the foregoing method embodiments. For the sake of convenience and brevity of description, they will not be elaborated herein.

[0245] The above business opportunity recommendation device 300 can be implemented in the form of a computer program, and this computer program can run on a computer device as shown in Figure 7 shown.

[0246] Please refer to Figure 7 , Figure 7 which is a schematic block diagram of a computer device provided by an embodiment of the present application. The computer device 500 can be a server. Among them, the server can be an independent server or a server cluster composed of multiple servers.

[0247] Referring to Figure 7 , the computer device 500 includes a processor 502, a memory, and a network interface 505 connected through a system bus 501. Among them, the memory can include a non-volatile storage medium 503 and an internal memory 504.

[0248] The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions. When the program instructions are executed, the processor 502 can be caused to execute a business opportunity recommendation method.

[0249] The processor 502 is configured to provide computing and control capabilities to support the operation of the entire computer device 500.

[0250] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can be caused to execute a business opportunity recommendation method.

[0251] The network interface 505 is used for network communication with other devices. Those skilled in the art can understand that Figure 7 the structure shown in is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device 500 to which the solution of this application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0252] Among them, the processor 502 is used to run the computer program 5032 stored in the memory to implement the following steps:

[0253] Obtain internal enterprise data and external data to obtain relevant data; extract enterprise characteristics, business opportunity characteristics, and interaction characteristics from the relevant data; input the enterprise characteristics, the business opportunity characteristics, and the interaction characteristics into a business opportunity recommendation model for business opportunity matching prediction and transaction prediction to obtain a business opportunity matching score and a transaction probability prediction value; output the business opportunity matching score and the transaction probability prediction value.

[0254] In one embodiment, when the processor 502 implements the step of extracting enterprise characteristics, business opportunity characteristics, and interaction characteristics from the relevant data, the following steps are specifically implemented:

[0255] Extract data on the scale, industry category, geographical location, historical cooperation records, and financial status of the enterprise from the relevant data to obtain initial enterprise characteristics; extract information on business opportunity categories, demand quantities, budget ranges, expected cooperation cycles, and market potentials from the relevant data to obtain initial business opportunity characteristics; analyze the historical interaction records and feedback evaluations between the enterprise and the business opportunity in the relevant data to obtain initial interaction characteristics; respectively process missing values, detect outliers, normalize numerical features for the initial enterprise characteristics, the initial business opportunity characteristics, and the initial interaction characteristics, and perform encoding conversion on categorical features to obtain enterprise characteristics, business opportunity characteristics, and interaction characteristics.

[0256] In one embodiment, when the processor 502 implements the step of inputting the enterprise characteristics, the business opportunity characteristics, and the interaction characteristics into a business opportunity recommendation model for business opportunity matching prediction and transaction prediction to obtain a business opportunity matching score and a transaction probability prediction value, the following steps are specifically implemented:

[0257] Perform feature engineering on the enterprise features, the business opportunity features, and the interaction features to obtain a processing result; input the processing result into a business opportunity recommendation model for business opportunity matching prediction and transaction prediction to obtain a business opportunity matching score and a transaction probability prediction value.

[0258] Among them, the business opportunity recommendation model is obtained by acquiring historical relevant data, extracting enterprise features, business opportunity features, and interaction features, performing feature engineering processing, and then using them as a sample set to train a deep learning architecture.

[0259] In one embodiment, when the processor 502 implements the step of performing feature engineering on the enterprise features, the business opportunity features, and the interaction features to obtain a processing result, the following specific steps are implemented:

[0260] Use the feature importance evaluation of XGBoost to normalize the importance of each of the enterprise features, the business opportunity features, and the interaction features, sort each of the enterprise features, the business opportunity features, and the interaction features from largest to smallest according to importance, calculate the cumulative importance, and screen all features whose cumulative importance meets the requirements to obtain a processing result.

[0261] In one embodiment, when the processor 502 implements the step that the business opportunity recommendation model is obtained by acquiring historical relevant data, extracting enterprise features, business opportunity features, and interaction features, performing feature engineering processing, and then using them as a sample set to train a deep learning architecture, the following specific steps are implemented:

[0262] Acquire historical relevant data, extract enterprise features, business opportunity features, and interaction features, perform feature engineering processing to obtain a sample set; construct a deep learning architecture and define a loss function; use the sample set to train the deep learning architecture, combine the loss function with the trained deep learning architecture, and determine the hyperparameters in the loss function through grid search or cross-validation methods to obtain a business opportunity recommendation model.

[0263] Among them, the structure of the deep learning architecture includes an input layer, a feature fusion layer, a deep learning network, and an output layer. The deep learning network includes a first fully connected layer, a first batch normalization layer, a first activation function ReLU, a first Dropout layer, a second fully connected layer, a second batch normalization, a second activation function ReLU, a second Dropout layer. The output layer includes two branches and respectively uses the sigmoid activation function.

[0264] It should be understood that in the embodiments of the present application, the processor 502 may be a central processing unit (CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0265] Those of ordinary skill in the art can understand that all or part of the processes in the methods of implementing the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program includes program instructions, and the computer program can be stored in a storage medium, and the storage medium is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the above method embodiments.

[0266] Therefore, the present invention also provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, and when the computer program is executed by a processor, the processor is caused to execute the following steps:

[0267] Obtain internal data and external data of the enterprise to obtain relevant data; extract enterprise features, business opportunity features, and interaction features from the relevant data; input the enterprise features, the business opportunity features, and the interaction features into a business opportunity recommendation model for business opportunity matching prediction and transaction prediction to obtain a business opportunity matching score and a transaction probability prediction value; output the business opportunity matching score and the transaction probability prediction value.

[0268] In one embodiment, when the processor executes the computer program to implement the step of extracting enterprise features, business opportunity features, and interaction features from the relevant data, the following steps are specifically implemented:

[0269] Extract data on the scale, industry category, geographical location, historical cooperation records, and financial status of the enterprise from the relevant data to obtain initial enterprise characteristics; extract information on business opportunity categories, demand volume, budget range, expected cooperation cycle, and market potential from the relevant data to obtain initial business opportunity characteristics; analyze the historical interaction records and feedback evaluations between the enterprise and the business opportunity in the relevant data to obtain initial interaction characteristics; process missing values, detect outliers, normalize numerical features, and perform encoding conversion on categorical features for the initial enterprise characteristics, the initial business opportunity characteristics, and the initial interaction characteristics respectively to obtain enterprise characteristics, business opportunity characteristics, and interaction characteristics.

[0270] In one embodiment, when the processor executes the computer program to implement the step of inputting the enterprise characteristics, the business opportunity characteristics, and the interaction characteristics into the business opportunity recommendation model for business opportunity matching prediction and transaction prediction to obtain a business opportunity matching score and a transaction probability prediction value, the specific implementation is as follows:

[0271] Perform feature engineering processing on the enterprise characteristics, the business opportunity characteristics, and the interaction characteristics to obtain a processing result; input the processing result into the business opportunity recommendation model for business opportunity matching prediction and transaction prediction to obtain a business opportunity matching score and a transaction probability prediction value.

[0272] Among them, the business opportunity recommendation model is obtained by acquiring historical relevant data, extracting enterprise characteristics, business opportunity characteristics, and interaction characteristics, and performing feature engineering processing and then using them as a sample set to train a deep learning architecture.

[0273] In one embodiment, when the processor executes the computer program to implement the step of performing feature engineering processing on the enterprise characteristics, the business opportunity characteristics, and the interaction characteristics to obtain a processing result, the specific implementation is as follows:

[0274] Use the feature importance evaluation of XGBoost to normalize the importance of each of the enterprise characteristics, the business opportunity characteristics, and the interaction characteristics, sort each of the enterprise characteristics, the business opportunity characteristics, and the interaction characteristics from largest to smallest according to importance, calculate the cumulative importance, and screen all features whose cumulative importance meets the requirements to obtain a processing result.

[0275] In one embodiment, when the processor executes the computer program to implement the step that the business opportunity recommendation model is obtained by acquiring historical relevant data, extracting enterprise characteristics, business opportunity characteristics, and interaction characteristics, performing feature engineering processing, and then using them as a sample set to train a deep learning architecture, the specific implementation is as follows:

[0276] Obtain historical relevant data and extract enterprise features, business opportunity features, and interaction features, and perform feature engineering processing to obtain a sample set; construct a deep learning architecture and define a loss function; use the sample set to train the deep learning architecture, adopt the loss function in combination with the trained deep learning architecture, and determine the hyperparameters in the loss function through grid search or cross-validation methods to obtain a business opportunity recommendation model.

[0277] Among them, the structure of the deep learning architecture includes an input layer, a feature fusion layer, a deep learning network, and an output layer. The deep learning network includes a first fully connected layer, a first batch normalization layer, a first activation function ReLU, a first Dropout layer, a second fully connected layer, a second batch normalization, a second activation function ReLU, a second Dropout layer. The output layer includes two branches and respectively uses the sigmoid activation function.

[0278] The storage medium can be various computer-readable storage media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disc that can store program codes.

[0279] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0280] In several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of each unit is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0281] The steps in the method embodiments of the present invention can be adjusted, combined, and deleted according to actual needs. The units in the device embodiments of the present invention can be combined, divided, and deleted according to actual needs. In addition, the functional units in each embodiment of the present invention can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

[0282] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0283] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A business opportunity recommendation method, characterized in that: include: Obtain internal and external data of the enterprise to obtain relevant data; Extracting enterprise characteristics, business opportunity characteristics, and interaction characteristics from the relevant data; Inputting the enterprise characteristics, the business opportunity characteristics and the interaction characteristics into the business opportunity recommendation model to perform business opportunity matching prediction and transaction prediction, so as to obtain a business opportunity matching score and a transaction probability prediction value; Output the business opportunity matching score and the transaction probability prediction value.

2. The business opportunity recommendation method according to claim 1, characterized in that: The extracting of enterprise features, business opportunity features and interaction features from the relevant data includes: Extracting data on the size, industry category, geographic location, historical cooperation record and financial status of the enterprise from the relevant data to obtain initial enterprise characteristics; Extracting information on business opportunity categories, demand, budget range, expected cooperation period and market potential from the relevant data to obtain initial business opportunity characteristics; Analyze the historical interaction records and feedback evaluations between the enterprise and the business opportunity in the relevant data to obtain initial interaction features; The initial enterprise features, the initial business opportunity features and the initial interaction features are respectively processed for missing values, detected for outliers, and normalized for numerical features, and the categorical features are encoded and converted to obtain enterprise features, business opportunity features and interaction features.

3. The business opportunity recommendation method according to claim 2, characterized in that: The step of inputting the enterprise characteristics, the business opportunity characteristics and the interaction characteristics into the business opportunity recommendation model to perform business opportunity matching prediction and transaction prediction to obtain a business opportunity matching score and a transaction probability prediction value includes: Performing feature engineering processing on the enterprise features, the business opportunity features, and the interaction features to obtain processing results; The processing results are input into the business opportunity recommendation model to perform business opportunity matching prediction and transaction prediction to obtain a business opportunity matching score and a transaction probability prediction value.

4. The business opportunity recommendation method according to claim 3, characterized in that: The performing feature engineering processing on the enterprise features, the business opportunity features, and the interaction features to obtain processing results includes: Using XGBoost’s feature importance evaluation, the importance of each of the enterprise feature, the business opportunity feature, and the interaction feature is normalized, and each of the enterprise feature, the business opportunity feature, and the interaction feature is sorted from large to small according to importance, the cumulative importance is calculated, and all features whose cumulative importance meets the requirements are screened to obtain the processing result.

5. The business opportunity recommendation method according to claim 3, characterized in that: The business opportunity recommendation model is obtained by acquiring historical relevant data and extracting enterprise characteristics, business opportunity characteristics and interaction characteristics, performing feature engineering processing, and then using them as sample sets to train a deep learning architecture.

6. The business opportunity recommendation method according to claim 5, characterized in that: The business opportunity recommendation model is obtained by obtaining historical relevant data and extracting enterprise features, business opportunity features and interaction features, and then performing feature engineering processing as a sample set to train a deep learning architecture, including: Obtain historical relevant data and extract enterprise features, business opportunity features, and interaction features, and perform feature engineering processing to obtain a sample set; Build deep learning architecture and define loss function; The deep learning architecture is trained using the sample set, a loss function is used in combination with the trained deep learning architecture, and hyperparameters in the loss function are determined by a grid search or cross-validation method to obtain a business opportunity recommendation model.

7. The business opportunity recommendation method according to claim 6, characterized in that: The structure of the deep learning architecture includes an input layer, a feature fusion layer, a deep learning network and an output layer. The deep learning network includes a first fully connected layer, a first normalization layer, a first activation function ReLU, a first Dropout layer, a second fully connected layer, a second batch normalization, a second activation function ReLU, and a second Dropout layer. The output layer includes two branches, and an igmoid activation function is used respectively.

8. A business opportunity recommendation device, characterized in that: include: A data acquisition unit is used to acquire internal and external data of the enterprise to obtain relevant data; A feature extraction unit, used to extract enterprise features, business opportunity features and interaction features from the relevant data; A prediction unit, used for inputting the enterprise characteristics, the business opportunity characteristics and the interaction characteristics into a business opportunity recommendation model to perform business opportunity matching prediction and transaction prediction, so as to obtain a business opportunity matching score and a transaction probability prediction value; The output unit is used to output the business opportunity matching score and the transaction probability prediction value.

9. A computer device, characterized in that: The computer device comprises a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images