Enterprise sales closing rate prediction method, medium and system

Abstract

The application provides an enterprise sales transaction rate prediction method, medium and system, relying on a super large enterprise library to obtain an enterprise full set formed by integrating an initial enterprise set and relationship enterprise information of each enterprise, and greatly widening the enterprise coverage of market analysis; then constructing a multi-source and multi-modal enterprise feature set according to the enterprise related information of each enterprise in the enterprise full set, improving the distinguishability of enterprise characteristics, and laying a solid foundation for accurate prediction; then taking the customers who have transacted as positive samples, and randomly sampling non-transacted customers from the enterprise full set as negative samples, and combining to form a training sample set conforming to the market distribution, and improving the fitting ability and generalization ability of the model under the real market scene; finally, taking the enterprise feature set as an input vector to iteratively train the training sample set, obtaining a high-precision, high-fitting and strong-generalization enterprise sales transaction rate prediction model, and systematically improving the sales conversion efficiency.

Description

Technical Field

[0001] This invention relates to the field of deep learning technology, and in particular to methods, media, and systems for predicting enterprise sales conversion rates. Background Technology

[0002] Traditional B2B sales rely heavily on human experience for lead gathering and prioritization. This approach suffers from technical challenges such as scarce sample resources, rigid decision-making rules, difficulties in project cold starts, and insufficient system scalability. Furthermore, current mainstream lead scoring systems often rely on limited feature fields or static labels, making it difficult to depict the complex relationship networks between entities in a massive market environment. They are also susceptible to training bias due to extreme imbalances between positive and negative samples, leading to insufficient accuracy in identifying high-value leads.

[0003] Therefore, there is an urgent need for a method to predict enterprise sales conversion rates that can overcome the aforementioned technical problems, in order to improve the conversion efficiency of B2B business. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a method, medium and system for predicting enterprise sales conversion rate, so as to achieve automatic, efficient and accurate prediction of enterprise sales conversion rate and improve enterprise sales conversion rate.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: Methods for predicting enterprise sales conversion rates include: Step 1: Obtain the initial set of enterprises and the related enterprise information of each enterprise entity, and integrate them to obtain the complete set of enterprises; Step 2: Based on the enterprise-related information of each enterprise entity in the enterprise set, analyze and extract the enterprise feature set corresponding to each enterprise entity; the enterprise feature set includes classification features, numerical range features, and label features; Step 3: Using the non-transactional customer enterprises obtained by stratified random sampling from the entire set of enterprises as negative samples, and the set of completed customer enterprises as positive samples, combine the negative and positive samples to obtain the training sample set; Step 4: Using the set of enterprise features corresponding to each enterprise entity in the enterprise set as the input vector, iteratively train the training sample set to obtain the enterprise sales conversion rate prediction model.

[0006] Optionally, step 1 includes: The collected enterprise information is filtered according to preset screening criteria to obtain an initial set of enterprises; wherein, the preset screening criteria include operating status and a threshold for the number of insured persons. Obtain the relational enterprise information of each enterprise entity in the initial enterprise set; the relational enterprises include at least one of the following: competitor relationship, branch or subsidiary relationship, supplier relationship, and customer relationship. Based on the initial set of enterprises and the obtained information on related enterprises, a complete set of enterprises is obtained.

[0007] Optionally, the related enterprise information may be the name or identifier of the related enterprise; The process of integrating the initial set of enterprises and the acquired information on related enterprises to obtain a complete set of enterprises includes: After parsing the acquired related enterprise information into corresponding standard enterprise identifiers, it is integrated into the initial enterprise set; The integrated set is deduplicated to obtain the complete set of enterprises.

[0008] Optionally, step 2 includes: Obtain relevant enterprise information from all enterprise entities in the enterprise database; the enterprise information includes business registration information, financial information, and online text information. The business registration information of each enterprise is analyzed, the basic classification attribute information corresponding to each enterprise is extracted, and the extracted basic classification attribute information is standardized and categorized to obtain the classification characteristics corresponding to each enterprise. The business registration and financial information of each enterprise entity is analyzed, the corresponding numerical attribute information of each enterprise entity is extracted, and the extracted numerical attribute information is segmented and discretized according to a preset interval to obtain the numerical range characteristics of each enterprise entity. Analyze the online text information of each enterprise entity, extract the business keywords corresponding to each enterprise entity, and obtain the tag features corresponding to each enterprise entity; By combining the classification features, numerical range features, and label features corresponding to each enterprise entity, a set of enterprise features corresponding to each enterprise entity is obtained.

[0009] Optionally, step 3 includes: Obtain the set of completed customer companies corresponding to each enterprise entity in the enterprise database and mark them as positive samples; The entire set of enterprises, excluding the group of enterprises that have already completed transactions, is statistically analyzed in a stratified manner according to preset dimensions; the preset dimensions include at least two of the following: size, industry category, and region. Based on the results of the hierarchical statistics, random sampling is performed in the statistical results of each layer according to the preset layer weights and / or preset sampling ratios to obtain the set of non-transactional customer enterprises corresponding to each enterprise entity, and these are marked as negative samples. Combine the positive and negative samples corresponding to each enterprise entity to obtain the training sample set.

[0010] Optionally, step 4 includes: The enterprise feature set corresponding to each enterprise entity in the enterprise set is encoded and converted into a model input vector; Based on the model input vector and the training sample set, the deep learning model is iteratively trained to obtain the enterprise sales conversion rate prediction model.

[0011] Optionally, it also includes: Step 5: Input the set of enterprise characteristics of each enterprise in the set of enterprises to be evaluated into the enterprise sales conversion rate prediction model, and output the conversion rate score of each enterprise in the set of enterprises to be evaluated. Optionally, it also includes: Step 6: Divide historically completed customer companies into a training set and a validation set, and add a preset number of non-completed customer companies to the validation set to form a validation sample. Use the enterprise sales conversion rate prediction model to score the validation sample, output the corresponding evaluation results, and verify the predictive performance of the enterprise sales conversion rate prediction model based on the evaluation results.

[0012] Another technical solution provided by this invention is: A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, enables the implementation of the above-described method for predicting enterprise sales conversion rates.

[0013] Another technical solution provided by this invention is: An enterprise sales conversion rate prediction system includes the aforementioned computer-readable storage medium and a processor; the processor is connected to the computer-readable storage medium; the processor calls and executes a computer program on the computer-readable storage medium.

[0014] The beneficial effects of this invention are as follows: The enterprise sales conversion rate prediction method, medium, and system provided by this invention first rely on a large-scale enterprise database to obtain a complete set of enterprises formed by integrating the initial set of enterprises and the information of related enterprises of each enterprise, which greatly broadens the enterprise coverage of market analysis; then, based on the enterprise-related information of each enterprise entity in the complete set of enterprises, a multi-source, multi-modal enterprise feature set is constructed to improve the distinguishability of enterprise characteristics and lay a solid foundation for accurate prediction; subsequently, using customers who have made a purchase as positive samples and non-purchase customers randomly sampled from the complete set of enterprises as negative samples, a training sample set that fits the market distribution is formed, which improves the model's fitting ability and generalization ability in real market scenarios; finally, the enterprise feature set is used as the input vector to iteratively train the training sample set to obtain a high-precision, high-fitting, and strong-generalization enterprise sales conversion rate prediction model, thereby systematically improving sales conversion efficiency. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the basic process of a method for predicting enterprise sales conversion rate provided in an embodiment of the present invention. Detailed Implementation

[0016] To explain in detail the technical principles, specific implementable solutions, possible application scenarios, and achievable objectives and effects of the present invention, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. It is understood that the embodiments described herein and the embodiments shown in the accompanying drawings are only used to more clearly illustrate the technical solutions of the present invention, and are therefore only examples intended to explain the present invention, and should not be construed as limiting the present invention. It should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0017] Please see Figure 1 This invention provides a method for predicting enterprise sales conversion rates, which helps improve the conversion rate of B2B sales leads.

[0018] like Figure 1 As shown in the figure, the enterprise sales conversion rate prediction method provided by this embodiment of the invention includes 6 main steps: Step 1: Obtain the initial set of enterprises and the related enterprise information of each enterprise entity, and integrate them to obtain the complete set of enterprises.

[0019] Specifically, step 1 includes the following sub-steps: S11. Obtain enterprise entity information from official and authoritative enterprise information data sources.

[0020] In some specific implementations, enterprise entity information is obtained from enterprise information data sources through network data collection and data aggregation.

[0021] In some specific implementations, the enterprise information data sources include, but are not limited to: enterprise information service platforms, business registration databases, publicly available social security data, and other enterprise information databases.

[0022] Here, the enterprise's main information is obtained from authoritative and reliable enterprise information data sources, which can ensure the authenticity, accuracy and timeliness of the basic data. This provides a high-quality data foundation for subsequent enterprise full set construction, feature extraction, model training and sales lead screening, avoiding deviations in solution implementation and decision-making errors caused by data distortion from the source.

[0023] S12. Filter the collected enterprise information according to the preset screening conditions to obtain the initial enterprise set.

[0024] The preset screening criteria include operating status (operating / existing) and a threshold for the number of insured persons; In some specific implementations, the preset screening conditions are as follows: Condition 1. The business status is "active" or "existing"; Condition 2. The number of employees enrolled in the company's social insurance program is not less than the preset threshold (e.g., 5 or more). Condition 3. The enterprise's main information is complete.

[0025] The initial set of firms formed after screening satisfies all three conditions mentioned above.

[0026] In some implementations, the initial set of enterprises can be in the millions, for example, containing approximately 5 million enterprises.

[0027] Here, by filtering the collected enterprise information, invalid and inactive enterprises can be efficiently eliminated, and effective enterprise entities with operational vitality and basic scale can be accurately selected, thereby improving the quality of the subsequent enterprise set and the efficiency and accuracy of model training and clue mining.

[0028] S13. Obtain the relational enterprise information of each enterprise entity in the initial enterprise set.

[0029] The related enterprises include, but are not limited to: competitor relationships, branch or subsidiary relationships, supplier relationships, and customer relationships.

[0030] In some specific implementations, the related enterprise information can be obtained through the aforementioned enterprise information data source or relational database query.

[0031] S14. Based on the initial set of enterprises and the obtained information on related enterprises, a complete set of enterprises is obtained.

[0032] In some specific implementations, the related enterprise information obtained in S13 above is the name or identifier of the related enterprise.

[0033] In some specific embodiments, the integration specifically includes: After the acquired related enterprise information is uniformly parsed into corresponding standard enterprise identifiers, it is integrated into the initial enterprise set; and the integrated set is deduplicated to form an expanded enterprise set.

[0034] In some specific implementations, the relationship extension achieved through S13 and S14 above can be executed in one round or multiple rounds to gradually expand the size of the enterprise set.

[0035] In some specific implementations, a relationship network diagram corresponding to each enterprise entity will be constructed based on the related enterprise information of each enterprise entity, and a graph neural network will be used to enhance the representation. This not only intuitively presents the relationship topology and ecological structure between enterprises in the enterprise whole set, but also helps the model to discover the hidden relationship value and group characteristics of enterprises, providing structured relationship data support for the model to discover hidden related transaction patterns.

[0036] For example, the specific implementation process of step 1 above includes: An initial enterprise set of approximately 5 million "operating" enterprises with at least 5 employees enrolled in social insurance was formed by screening from the enterprise information database. For each enterprise in the initial enterprise set, the names or identifiers of related enterprises (including competitors, branches or subsidiaries, suppliers, and customers) were queried from the enterprise information database. The obtained related enterprise names or identifiers were parsed into standardized enterprise identifiers through enterprise name resolution or enterprise information matching services. The parsed standardized enterprise identifiers were added to the initial enterprise set, and after deduplication, a complete enterprise set of approximately 12 million+ enterprises was formed.

[0037] Step 2: Based on the enterprise-related information of each enterprise entity in the enterprise set, analyze and extract the enterprise feature set corresponding to each enterprise entity.

[0038] The enterprise feature set includes classification features, numerical range features, and label features, which are used for subsequent model training and inference.

[0039] Specifically, step 2 includes the following sub-steps: S21. Obtain relevant enterprise information for each enterprise entity in the enterprise central database.

[0040] The enterprise-related information includes, but is not limited to: business registration information, financial information, and online text information. In particular, the business registration information within the business registration information.

[0041] In some specific implementations, the channels for obtaining the enterprise-related information include, but are not limited to: business registration information, enterprise official website information, search engine indexed web pages, and enterprise-related web pages in news information or industry platforms.

[0042] In some specific implementations, enterprise-related information of each enterprise entity is obtained through network data collection and data integration.

[0043] S22, Classification feature extraction.

[0044] Specifically, by analyzing the business registration information of each enterprise, especially the business registration information, the basic classification attribute information corresponding to each enterprise is extracted; the extracted basic classification attribute information is standardized and categorized to obtain the classification characteristics corresponding to each enterprise.

[0045] In some specific implementations, the basic classification attribute information includes, but is not limited to, category information such as enterprise type, industry, registered region, and enterprise size.

[0046] In some specific implementations, the standardized category coding specifically involves: first, standardizing the extracted basic classification attribute information, and then converting it into classification features usable by the model through category coding.

[0047] S23, Numerical range feature extraction.

[0048] Specifically, by analyzing the business registration information and publicly available financial information of each enterprise, especially the business registration information, the numerical attribute information of each enterprise is extracted; the extracted numerical attribute information is then segmented and discretized according to a preset interval to obtain the numerical range characteristics of each enterprise.

[0049] In some specific implementations, the enterprise's numerical attribute information includes, but is not limited to, numerical information such as the number of employees, registered capital, number of years since the enterprise was established, and operating revenue.

[0050] In some specific implementations, the preset interval refers to setting at least two intervals for each piece of information, thereby forming a discretized numerical range feature.

[0051] Here, by segmenting and discretizing the extracted enterprise numerical attribute information according to a preset interval, the difference in numerical scale can be reduced.

[0052] For example, the following four ranges are preset for “number of employees”: 0–10, 10–50, 50–200 and more than 200.

[0053] S24, Label Feature Extraction.

[0054] Specifically, by analyzing the online text information of each enterprise entity, the corresponding business keywords of each enterprise entity are extracted, and the corresponding tag features of each enterprise entity are obtained.

[0055] The business keywords refer to keywords or main tags that reflect the company's business direction.

[0056] In some specific implementations, the network text information includes, but is not limited to: corporate website pages, webpage content (e.g., encyclopedias, news, recruitment), and search result summaries.

[0057] In some specific implementations, the business keywords are extracted using methods such as keyword extraction algorithms, topic modeling methods, or topic extraction methods based on pre-trained language models.

[0058] In some specific implementations, pre-trained language models (such as the BERT family) can be combined to perform phrase mining and topic clustering on network text information, construct enterprise semantic vectors, and obtain enhanced label features.

[0059] S25. Combine the classification features, numerical range features, and label features corresponding to each enterprise entity to obtain the enterprise feature set corresponding to each enterprise entity.

[0060] In some specific implementations, the set of enterprise features is stored in an enterprise feature database for use in subsequent model training and inference.

[0061] For example, the specific implementation process of step 2 above includes: Taking a specific company as an example, firstly, based on its company name or logo, the company's business registration and financial information are queried in the enterprise information database. Basic classification attributes, including company type, industry, registered location, and establishment date, as well as numerical attributes, such as number of employees, registered capital, years of establishment, and operating revenue, are extracted. Secondly, the text content of the company's official website or public web pages, such as company introduction pages, product pages, or news pages, is collected. Simultaneously, the company name is searched through search engines to obtain search result summaries and related web page links, and the main text content of these web pages is extracted. The collected text content and extracted web page content are then cleaned, such as removing HTML tags, script code, and irrelevant characters, and then segmented or parsed to obtain structured web page text data. Finally, the basic classification attributes, numerical attributes, and web page text data are analyzed and feature extracted to obtain three types of company features: classification features, numerical range features, and tag features, forming the company's feature set.

[0062] Step 3: Construct the training sample set.

[0063] Specifically, step 3 includes the following sub-steps: S31. Obtain the set of completed customer enterprises corresponding to each enterprise entity in the enterprise set and mark them as positive samples.

[0064] In some specific implementations, the corresponding customers of each enterprise entity can be obtained from the customer relationship management system, sales transaction records, or historical order data. Preferably, the query is performed based on the standard enterprise identifier of each enterprise entity. If it is an enterprise name, it is converted into a standard enterprise identifier through an enterprise name resolution service before the query is performed.

[0065] S32. After removing the set of enterprises with completed transactions, perform stratified random sampling to obtain the set of enterprises with uncompleted transactions corresponding to each enterprise entity, and mark them as negative samples.

[0066] Specifically, firstly, after marking the set of companies with completed transactions as positive samples, these companies are removed from the overall set of companies, resulting in a pool of candidate companies with uncompleted transactions. Then, the pool of candidate companies with uncompleted transactions is statistically analyzed hierarchically according to preset dimensions. These preset dimensions include, but are not limited to, company size, industry category, and region. Preferably, at least two of these dimensions are selected to statistically analyze the distribution of the entire set of companies after removing the set of companies with completed transactions within the corresponding dimensions. Next, a certain number of samples are randomly selected from each hierarchical statistical result, and all selected samples are marked as negative samples. Preferably, in each hierarchical statistical result, random sampling is performed according to preset hierarchical weights and / or preset sampling ratios to obtain the set of candidate companies with uncompleted transactions corresponding to each company entity. That is, each company entity randomly selects a certain number of uncompleted companies from each hierarchical statistical result to form the set of candidate companies with uncompleted transactions corresponding to that company entity. Finally, the positive and negative samples associated with each company entity are combined to obtain a training sample set.

[0067] In some specific implementations, the number of negative samples for each business entity is several times its number of positive samples (e.g., 2-20 times) to alleviate the sample imbalance problem. For example, when a business entity has 100 completed customer companies, approximately 1,000 non-complete customer companies can be randomly selected according to industry and geographical distribution as its negative samples, thereby forming a sample set for training.

[0068] In some specific implementations, when combining and obtaining the training sample set, positive samples and negative samples are labeled as "positive class" and "negative class" respectively to clarify the sample category.

[0069] In particular, this embodiment uses a stratified random sampling method for negative samples, so that the distribution of negative samples in dimensions such as size, industry and region is similar to the overall market structure of the entire set of enterprises, thereby improving the model's generalization ability in the real market environment.

[0070] Step 4: Using the set of enterprise features corresponding to each enterprise entity in the enterprise set as the input vector, iteratively train the training sample set to learn the relationship between enterprise features and transaction probability, and obtain the enterprise sales transaction rate prediction model.

[0071] Specifically, step 4 includes the following sub-steps: S41. Encode the set of enterprise features corresponding to each enterprise entity in the enterprise set and convert it into a model input vector.

[0072] In some specific implementations, one-hot encoding is performed on the classification features in the enterprise feature set; multi-hot encoding is performed on the label features; the numerical range features are discretized; and the encoding results corresponding to the classification features, numerical range features, and label features are concatenated to convert them into a model input vector.

[0073] Here, by uniformly converting the multimodal unstructured enterprise feature set into a standardized input vector that can be recognized by the deep learning model, it is possible to effectively adapt to the model training requirements and improve the effectiveness and accuracy of feature representation.

[0074] S42. Based on the model input vector and the training sample set, the deep learning model is iteratively trained to obtain the enterprise sales conversion rate prediction model.

[0075] In some specific implementations, the deep learning model can be a deep learning binary classification model or a ranking model, such as DNN (Deep Neural Network), Wide&Deep (Width & Depth Hybrid Network Model), DeepFM (Deep Factorization Model), or Transformer (Self-Attention Transformation Model).

[0076] In some specific implementations, to alleviate the imbalance between positive and negative samples in the training sample set, positive samples can be assigned higher weights and negative samples lower weights. Furthermore, the weights of the negative samples can be related to the ratio of negative samples to positive samples or a stratified sampling strategy. This improves the model's ability to fit the characteristics of the transacting companies and gives the model's output prediction scores better discriminative power and ranking reliability.

[0077] Optionally, during model training, a weighted loss function is used for optimization based on the weight allocation results of positive and negative samples in the training sample set. The weighted loss function includes, but is not limited to, the class weights of the cross-entropy loss and the focal loss function.

[0078] S5: Input the set of enterprise characteristics of each enterprise in the set of enterprises to be evaluated into the enterprise sales conversion rate prediction model, and output the conversion rate score of each enterprise in the set of enterprises to be evaluated.

[0079] The conversion rate score is a predicted probability value for the enterprise to be evaluated to belong to the category of customers who have already made a purchase.

[0080] After the model is trained, it can be used to predict and score the set of enterprise characteristics of the enterprise to be evaluated as input.

[0081] Specifically, step 5 includes the following sub-steps: S51. Obtain the set of enterprise characteristics for each enterprise entity in the set of enterprises to be evaluated.

[0082] The set of companies to be evaluated serves as the evaluation pool for the model, encompassing all business entities whose "probability of becoming a customer of the target company" needs to be predicted.

[0083] In some specific implementations, the set of enterprises to be evaluated can be obtained by using the method described in step 1 above; alternatively, a custom entity of enterprises to be evaluated can be defined to form the set of enterprises to be evaluated.

[0084] In some specific implementations, the enterprise characteristic set of each enterprise entity in the set of enterprises to be evaluated is obtained by using the method described in step 2 above, which will not be elaborated further.

[0085] S52. Input the set of enterprise characteristics of the enterprise set to be evaluated into the enterprise sales conversion rate prediction model, and the model outputs the conversion rate score of each enterprise in the enterprise set to be evaluated.

[0086] The conversion rate score represents the probability or relevance score of each enterprise entity to be evaluated becoming a potential customer.

[0087] In some specific implementations, the transaction rate scores output by the model are sorted according to the transaction rate scores of each enterprise being evaluated. Preferably, the enterprises are sorted in descending order of transaction rate to form a list of enterprise scores.

[0088] In some specific implementations, the method also includes: based on the conversion rate scores of the set of companies to be evaluated output by the model, marking the top-scoring companies to generate a TOP-N preferred customer list, so that users can intuitively identify potential preferred customer companies.

[0089] In some specific implementations, the method also includes: performing Platt / equal binning calibration based on the conversion rate score output by the model; and performing revenue-weighted ranking based on expected average order value and sales cycle.

[0090] For example, a business sales conversion rate prediction model is used to score approximately 12 million businesses in the set of businesses to be evaluated; based on the scoring results, the top-scoring businesses are selected to generate a priority follow-up list.

[0091] In some preferred embodiments of this example, step 6 is also included to verify and update the enterprise sales conversion rate prediction model trained in step 4, so as to evaluate and update the predictive performance of the model.

[0092] Step 6: Verify and update the enterprise sales conversion rate prediction model.

[0093] Specifically, step 6 includes the following sub-steps: S51. Randomly divide historically traded customer companies into training and validation sets.

[0094] S52. Add a preset number of non-transactional customer companies to the verification set to form a verification sample.

[0095] Here, the training set is used for model training; the validation set, together with a subset of companies that did not complete a transaction, constitutes the validation sample of the training set.

[0096] S53. Use the enterprise sales conversion rate prediction model to score the verification sample and output the corresponding evaluation results.

[0097] Specifically, the model input is the set of enterprise characteristics of each enterprise in the verification sample, and the model outputs the transaction rate score of each enterprise in the verification sample.

[0098] S54. Verify the predictive performance of the enterprise sales conversion rate prediction model based on the conversion rate score of the verification sample.

[0099] In some specific implementations, the verification includes: analyzing the proportion of actual transactions within different score ranges in the transaction rate score results to assess the model's discriminative ability.

[0100] In some specific implementations, the verification also includes: manually labeling and re-entering high-uncertainty samples for continuous optimization.

[0101] In some specific implementations, in practical applications, newly added customers who have already completed transactions can be continuously added to the training sample, and the enterprise sales conversion rate prediction model can be retrained periodically to continuously update the model and optimize the prediction effect.

[0102] For example, out of 200 historical customers who made transactions, 100 were selected for training (set A), and 100 were combined with the failed samples (companies that did not make transactions) to form set B. The model's scoring of set B showed that about 1 / 3 of the samples scored above 90 points, and the transaction rate increased from the average of 10% to about 60%.

[0103] Another embodiment of the present invention further extends the above embodiments by providing a computer-readable storage medium.

[0104] This embodiment provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program can implement the enterprise sales conversion rate prediction method described in the above embodiment. The specific steps of the method will not be repeated here; please refer to the description in the above embodiment for details.

[0105] Those skilled in the art will understand that all or part of the processes in the above technical solutions can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the methods described above. After being executed by a processor, the program can also achieve the beneficial effects of the corresponding methods.

[0106] The storage medium can be a disk, optical disc, read-only memory (ROM), or random access memory (RAM), etc.

[0107] Another embodiment of the present invention further extends the previous embodiment by providing an enterprise sales conversion rate prediction system.

[0108] The enterprise sales conversion rate prediction system provided in this embodiment includes the computer-readable storage medium and processor described in the previous embodiment; the processor is electrically connected to the computer-readable storage medium; the processor calls and executes the computer program on the computer-readable storage medium, which can realize the enterprise sales conversion rate prediction method described in the first embodiment above. The specific steps of the method will not be repeated here; please refer to the description in the first embodiment above for details.

[0109] In summary, the enterprise sales conversion rate prediction method, medium, and system provided by this invention rely on a massive enterprise database to construct a complete set of enterprises and build a multi-source, multi-modal enterprise feature system. This significantly broadens the enterprise coverage of market analysis and effectively improves the distinguishability of enterprise characteristics. Simultaneously, based on the complete enterprise database, a hierarchical negative sampling combined with sample weighting optimization strategy is implemented to specifically alleviate the problem of positive and negative sample imbalance, significantly improving the model's fitting ability and generalization performance. Furthermore, the conversion rate score output by the model is applied to sales lead priority ranking, greatly improving sales work efficiency and lead conversion efficiency. This invention can achieve automated, efficient, and accurate prediction of enterprise sales conversion rates, thereby effectively improving enterprise sales conversion rates. Market validation shows that in typical customer scenarios, this invention can achieve an average increase of approximately 100% in enterprise sales conversion rates, with a minimum increase of approximately 30% and a maximum increase of approximately 300%.

[0110] Although preferred embodiments of the invention have been described, those skilled in the art, upon understanding the basic inventive concept, can make other changes and modifications to the embodiments. Therefore, the appended claims are intended to cover the preferred embodiments and all equivalent modifications falling within the scope of protection defined by the claims and their equivalents. Clearly, those skilled in the art can make various alterations and variations to the invention without departing from its spirit and scope. If such modifications and variations fall within the scope of protection defined by the claims and their equivalents, the invention also intends to include them.

[0111] It should be noted that any reference signs placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware.

[0112] In this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "multiple" means two or more, unless otherwise explicitly specified. The term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone.

[0113] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0114] In this specification, the use of terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refers to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of the present invention. The illustrative expressions of the above terms in this specification should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0115] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for predicting enterprise sales conversion rates, characterized in that, include: Step 1: Obtain the initial set of enterprises and the related enterprise information of each enterprise entity, and integrate them to obtain the complete set of enterprises; Step 2: Based on the enterprise-related information of each enterprise entity in the enterprise set, analyze and extract the enterprise feature set corresponding to each enterprise entity; The enterprise feature set includes classification features, numerical range features, and label features; Step 3: Using the non-transactional customer enterprises obtained by stratified random sampling from the entire set of enterprises as negative samples, and the set of completed customer enterprises as positive samples, combine the negative and positive samples to obtain the training sample set; Step 4: Using the set of enterprise features corresponding to each enterprise entity in the enterprise set as the input vector, iteratively train the training sample set to obtain the enterprise sales conversion rate prediction model.

2. The enterprise sales conversion rate prediction method as described in claim 1, characterized in that, Step 1 includes: The collected enterprise information is filtered according to preset screening criteria to obtain an initial set of enterprises; wherein, the preset screening criteria include operating status and a threshold for the number of insured persons. Obtain the relational enterprise information of each enterprise entity in the initial enterprise set; the relational enterprises include at least one of the following: competitor relationship, branch or subsidiary relationship, supplier relationship, and customer relationship. Based on the initial set of enterprises and the obtained information on related enterprises, a complete set of enterprises is obtained.

3. The enterprise sales conversion rate prediction method as described in claim 2, characterized in that, The related company information refers to the names or identifiers of the related companies. The process of integrating the initial set of enterprises and the acquired information on related enterprises to obtain a complete set of enterprises includes: After parsing the acquired related enterprise information into corresponding standard enterprise identifiers, it is integrated into the initial enterprise set; The integrated set is deduplicated to obtain the complete set of enterprises.

4. The enterprise sales conversion rate prediction method as described in claim 1, characterized in that, Step 2 includes: Obtain relevant enterprise information from all enterprise entities in the enterprise database; the enterprise information includes business registration information, financial information, and online text information. The business registration information of each enterprise is analyzed, the basic classification attribute information corresponding to each enterprise is extracted, and the extracted basic classification attribute information is standardized and categorized to obtain the classification characteristics corresponding to each enterprise. The business registration and financial information of each enterprise entity is analyzed, the corresponding numerical attribute information of each enterprise entity is extracted, and the extracted numerical attribute information is segmented and discretized according to a preset interval to obtain the numerical range characteristics of each enterprise entity. Analyze the online text information of each enterprise entity, extract the business keywords corresponding to each enterprise entity, and obtain the tag features corresponding to each enterprise entity; By combining the classification features, numerical range features, and label features corresponding to each enterprise entity, a set of enterprise features corresponding to each enterprise entity is obtained.

5. The enterprise sales conversion rate prediction method as described in claim 1, characterized in that, Step 3 includes: Obtain the set of completed customer companies corresponding to each enterprise entity in the enterprise database and mark them as positive samples; The entire set of enterprises, excluding the group of enterprises that have already completed transactions, is statistically analyzed in a stratified manner according to preset dimensions; the preset dimensions include at least two of the following: size, industry category, and region. Based on the results of the hierarchical statistics, random sampling is performed in the statistical results of each layer according to the preset layer weights and / or preset sampling ratios to obtain the set of non-transactional customer enterprises corresponding to each enterprise entity, and these are marked as negative samples. Combine the positive and negative samples corresponding to each enterprise entity to obtain the training sample set.

6. The enterprise sales conversion rate prediction method as described in claim 1, characterized in that, Step 4 includes: The enterprise feature set corresponding to each enterprise entity in the enterprise set is encoded and converted into a model input vector; Based on the model input vector and the training sample set, the deep learning model is iteratively trained to obtain the enterprise sales conversion rate prediction model.

7. The enterprise sales conversion rate prediction method as described in claim 1, characterized in that, Also includes: Step 5: Input the set of enterprise characteristics of each enterprise in the set of enterprises to be evaluated into the enterprise sales conversion rate prediction model, and output the conversion rate score of each enterprise in the set of enterprises to be evaluated.

8. The enterprise sales conversion rate prediction method as described in claim 1, characterized in that, Also includes: Step 6: Divide historically completed customer companies into a training set and a validation set, and add a preset number of non-completed customer companies to the validation set to form a validation sample. Use the enterprise sales conversion rate prediction model to score the validation sample, output the corresponding evaluation results, and verify the predictive performance of the enterprise sales conversion rate prediction model based on the evaluation results.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it can implement the enterprise sales conversion rate prediction method according to any one of claims 1 to 8.

10. An enterprise sales conversion rate prediction system, characterized in that, The device includes the computer-readable storage medium as described in claim 9 and a processor; the processor is connected to the computer-readable storage medium; the processor invokes and executes a computer program on the computer-readable storage medium.

Images