Method, device and server for enterprise assessment processing
By using a weighted sum of information entropy, similarity, and difference from text feature vectors, combined with enterprise business scope and patent information, a classification model is trained, which solves the problem of poor reliability of pre-trained models and achieves a highly reliable assessment of enterprise categories.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CCB FINTECH CO LTD
- Filing Date
- 2022-08-31
- Publication Date
- 2026-06-09
Smart Images
Figure CN115471055B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of natural language processing technology, and in particular to a method, apparatus and server for enterprise evaluation processing. Background Technology
[0002] Green enterprises are those whose business activities are conducive to supporting environmental improvement, adapting to or mitigating climate change and conserving and efficiently utilizing resources, and which have funding needs.
[0003] Typically, whether a company is a green enterprise is a key indicator in a bank's risk assessment before granting it a credit loan. Therefore, companies need to be categorized to determine their green enterprise status.
[0004] Currently, pre-trained classification models can be used to classify companies. However, due to the poor reliability of the samples used to train the classification model, the reliability of the resulting model is also low. Consequently, the reliability of determining whether a company is a green company based on this model is poor, leading to inaccurate evaluation results. Summary of the Invention
[0005] This application provides a method, apparatus, and server for enterprise assessment processing to improve the problem of poor reliability in determining whether an enterprise is a green enterprise by a classification model, which leads to inaccurate assessment results for enterprises.
[0006] Firstly, this application provides a method for enterprise evaluation processing, applied to a server. The method includes: the server obtaining an evaluation request, the evaluation request including an enterprise to be evaluated; the server obtaining text feature vectors of the enterprise to be evaluated according to the evaluation request, wherein the text feature vectors of the enterprise to be evaluated are obtained from the enterprise's business scope and / or the enterprise's patent information; the server classifying the text feature vectors of the enterprise to be evaluated using a pre-trained classification model to determine the category of the enterprise to be evaluated, wherein the category of the enterprise is a green enterprise or a non-green enterprise. The classification model is obtained by selecting the top N text feature vectors in value from the set of text feature vectors of each group of enterprises to label the category of the enterprise, and using the top N text feature vectors as input data and the category of the enterprise as output data to train an initial training network. The value of any text feature vector is obtained by weighted summation based on the information entropy of the text feature vector, the similarity between the text feature vector and the cluster center of the set of text feature vectors to which it belongs, and the difference between the text feature vector and other text feature vectors in the set of text feature vectors to which it belongs; the server obtaining an evaluation mode corresponding to the category of the enterprise to be evaluated, and performing evaluation processing on the enterprise to be evaluated according to the evaluation mode.
[0007] In one possible implementation, the value of any text feature vector is calculated using the formula Q(x) = H(x) based on information entropy H(x), similarity I(x), and difference D(x). λ ×I(x) μ ×D(x) (1-λ-μ) The result is given by, where Q(x) is the value, λ and μ are the weights, λ and μ ∈ [0, 1] and λ + μ ≤ 1.
[0008] This ensures that the value of any obtained text feature vector is highly accurate and reliable.
[0009] In one possible implementation, the information entropy H(x) is calculated according to the formula H(x) = -∑ i p(y i |x)logp(y i |x) is obtained, where p(y) i |x) represents the text feature vector x, which belongs to the enterprise category y. i The probability of the enterprise category y i Including green and non-green enterprises, the enterprise category y i It is the output after the text feature vector x is input into the initial classification model. The initial classification model is obtained by inputting the text feature vectors of the categories of labeled enterprises (less than a preset number) into the initial training network for training.
[0010] Understandably, the information entropy obtained in the manner described above has high reliability.
[0011] In one possible implementation, the similarity I(x) is calculated according to the formula... The result is that, where K is the number of clusters formed by clustering the set of text feature vectors of each group of enterprises, and x... (k) The feature vector representing the cluster center of the k-th cluster is x, where x is the text feature vector.
[0012] Understandably, the similarity obtained in the manner described is highly reliable.
[0013] In one possible implementation, the difference D(x) is determined according to the formula... The obtained, where x m Let x be the m-th text feature vector. j Let be the j-th text feature vector (excluding the m-th text feature vector) in the set of text feature vectors of the same group of enterprises, U be the set of text feature vectors of the same group of enterprises, n be the number of text feature vectors in the set of text feature vectors of the same group of enterprises, and s be the number of text feature vectors. m It is the standard deviation of the magnitude of each text feature vector in the set of text feature vectors of the same group of enterprises.
[0014] Understandably, the reliability of the differences obtained in the manner described is high.
[0015] In one possible implementation, the server obtains the text feature vector of the enterprise to be evaluated based on the evaluation request, including: the server determining the Boolean word frequency, part-of-speech ratio, and text length ratio of each keyword in the text describing green and non-green enterprises in the enterprise's business scope and / or patent information, and the ratio of the text length to the text length of the longest word in the enterprise's business scope and / or patent information; the server processes each keyword in the text describing green and non-green enterprises based on the formula CF=WF(w)*[POS(w)+WL(w)] to obtain the text feature vector of the enterprise to be evaluated, where CF is the feature vector of the keywords in the text feature vector of the enterprise to be evaluated, POS(w) is the part-of-speech ratio, and WL(w) is the ratio of the text length of the keywords to the text length of the longest word in the enterprise's business scope and / or patent information.
[0016] Understandably, since the keyword feature vector in the text feature vector of the company to be evaluated is obtained by considering the Boolean word frequency, part-of-speech ratio, and text length ratio of each keyword in the text describing green and non-green enterprises within the company's business scope and / or patent information, it better expresses the characteristics of the text describing keywords for green and non-green enterprises. Therefore, the text feature vector of the company to be evaluated obtained based on the formula CF=WF(w)*[POS(w)+WL(w)] has high reliability.
[0017] In one possible implementation, the text feature vector of the company to be evaluated also includes subject features used to describe the company's patent name.
[0018] This enriches the content of the text feature vectors of the companies to be evaluated, resulting in high reliability.
[0019] In one possible implementation, the topic features are extracted using a topic feature extraction model based on Latent Dirichlet Allocation (LDA), where the expression for the topic feature L is L = [l1, l2, l3, ..., l...]. t ], where t is the dimension of the topic feature, l t Let be the probability of a patent name under the t-th topic.
[0020] This makes the extracted subject features more reliable.
[0021] Secondly, this application also provides an enterprise evaluation processing apparatus applied to a server. The apparatus provided by this application includes: a request acquisition unit for acquiring an evaluation request, the evaluation request including an enterprise to be evaluated; a text acquisition unit for acquiring text feature vectors of the enterprise to be evaluated according to the evaluation request, wherein the text feature vectors of the enterprise to be evaluated are obtained from the enterprise's business scope and / or the enterprise's patent information; and an enterprise classification unit for classifying the text feature vectors of the enterprise to be evaluated using a pre-trained classification model to determine the category of the enterprise to be evaluated, wherein the category of the enterprise is a green enterprise or a non-green enterprise, and the classification model is derived from the text feature vectors of each group of enterprises. From the set of feature vectors, the top N text feature vectors by value are selected to label the enterprise category. The initial training network is trained using the top N text feature vectors as input data and the enterprise category as output data. The value of any text feature vector is obtained by weighted summation based on the information entropy of the text feature vector, the similarity between the text feature vector and the cluster center of its set of text feature vectors, and the difference between the text feature vector and other text feature vectors in its set of text feature vectors. The enterprise evaluation unit is used to obtain the evaluation mode corresponding to the category of the enterprise to be evaluated and to evaluate the enterprise according to the evaluation mode.
[0022] Thirdly, this application also provides a server, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the server performs the method provided in the first aspect of this application.
[0023] Fourthly, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, causes the computer to perform the method provided in the first aspect of this application.
[0024] Fifthly, this application also provides a computer program product, including a computer program that, when run, causes a computer to perform the method provided in the first aspect of this application.
[0025] This application provides a method for evaluating enterprises. The server uses a pre-trained classification model to classify the text feature vectors of the enterprises to be evaluated. The classification model selects the top N text feature vectors by value from the set of text feature vectors of each group of enterprises to label the enterprise category, and uses the top N text feature vectors as input data and the enterprise category as output data to train an initial training network. This results in a highly reliable classification model. Furthermore, the value of any text feature vector is obtained by weighted summation based on the information entropy of the text feature vector, the similarity between the text feature vector and the cluster center of its set, and the difference between the text feature vector and other text feature vectors in its set. This also results in high accuracy and reliability of the value of any text feature vector. This further enhances the reliability of the trained classification model. Consequently, the accuracy of determining the category of the enterprise to be evaluated based on the classification model is also high. Therefore, the reliability of the evaluation results based on the evaluation mode corresponding to the category of the enterprise to be evaluated is also high. Attached Figure Description
[0026] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0027] Figure 1 A flowchart illustrating the training steps of the classification model for the enterprise evaluation processing method provided in this application embodiment;
[0028] Figure 2 A flowchart illustrating the steps of the enterprise evaluation process provided in this application embodiment;
[0029] Figure 3 Functional unit block diagram of the enterprise evaluation processing device provided in the embodiments of this application;
[0030] Figure 4 This is a structural block diagram of the server provided in an embodiment of this application. Detailed Implementation
[0031] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments made by those skilled in the art under the guidance of these embodiments are within the scope of protection of this application.
[0032] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0033] Currently, pre-trained classification models can be used to classify companies. However, due to the poor reliability of the samples used to train the classification model, the reliability of the trained model is low. Consequently, the reliability of determining whether a company is a green company based on the classification model is also poor, leading to inaccurate evaluation results for the companies.
[0034] Based on the aforementioned technical problems, the inventive concept of this application lies in: using the information entropy of text feature vectors, the similarity between a text feature vector and the cluster center of its corresponding set of text feature vectors, and the difference between a text feature vector and other text feature vectors in its corresponding set of text feature vectors, to determine the value of the text feature vectors of each group of enterprises, and selecting the top N text feature vectors in value ranking to train a classification model, thereby accurately determining the category of the enterprise to be evaluated. In this way, the reliability of the subsequent evaluation results based on the enterprise category is also high.
[0035] Explanation of technical terms in this application:
[0036] Information entropy is a fundamental concept in information theory, describing the uncertainty of the occurrence of various possible events from an information source.
[0037] Cluster center: A cluster center is a special sample in cluster analysis. It represents a certain cluster, and other samples determine whether they belong to that cluster by calculating their distance from it.
[0038] Boolean term frequency: If a document contains a keyword, the Boolean term frequency is 1; otherwise, the Boolean term frequency is 0.
[0039] The technical solutions of this application and how they solve the aforementioned technical problems will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.
[0040] This application provides an evaluation processing method for enterprises, applied to a server. The evaluation processing method for enterprises includes two steps: a training step for a classification model and an evaluation step for the enterprise. Figure 1 As shown, the training steps for the classification model include:
[0041] S101: The server obtains a set of text feature vectors of a group of companies, each text feature vector including information describing the company's business scope and / or the company's patent information.
[0042] The set of text feature vectors for a group of companies can be understood as a set of training samples used to train the classification model. The specific details of the company's business scope and patent information are shown in Table 1.
[0043]
[0044]
[0045] Table 1
[0046] S102: The server determines each text feature vector in the set of text feature vectors as a training sample to train the classification model.
[0047] The classification model is used to determine the category of the company to be evaluated, which is either a green company or a non-green company, based on the text feature vector of the company.
[0048] In addition, the value of any text feature vector is obtained by weighted summation based on the information entropy of the text feature vector, the similarity between the text feature vector and the cluster center of the set of text feature vectors to which it belongs, and the difference between any text feature vector and other text feature vectors in the set of text feature vectors to which it belongs.
[0049] Understandably, information entropy represents the amount of information required for a distribution; a higher information entropy indicates greater uncertainty in the system, while a lower information entropy indicates less uncertainty. Therefore, information entropy can be used as one of the indicators for determining value. Furthermore, text feature vectors may contain outliers or anomalies, making them unsuitable as indicators for determining value. In such cases, similarity can be used as one indicator. To further mitigate redundant information among text feature vectors, differences can also be used as an indicator. This approach ensures high accuracy and reliability in determining the value of any given text feature vector.
[0050] For example, the weighted summation method can be as follows: the value of any text feature vector is calculated based on information entropy H(x), similarity I(x), and difference D(x), using the formula Q(x) = H(x). λ ×I(x) μ ×D(x) (1-λ-μ) The result is given by, where Q(x) is the value, λ and μ are the weights, λ and μ ∈ [0, 1] and λ + μ ≤ 1.
[0051] For example, the information entropy H(x) is the value of the server according to the formula H(x) = -∑ i p(y i |x)logp(y i |x) is obtained, where p(y) i |x) represents the text feature vector x, which belongs to the enterprise category y. i The probability of the enterprise category y i Including green and non-green enterprises, the enterprise category y i The output is obtained by inputting the text feature vector x into the initial classification model. The initial classification model is trained by inputting the text feature vectors of the categories of labeled enterprises (less than a preset number) into the initial training network. Understandably, the information entropy obtained in the above way has high reliability.
[0052] For example, the similarity I(x) is calculated according to the formula The result is that, where K is the number of clusters formed by clustering the set of text feature vectors of each group of enterprises, and x... (k) Let x be the feature vector representing the cluster center of the k-th cluster, and x be the text feature vector. It is understandable that the similarity obtained in this manner is highly reliable.
[0053] For example, the difference D(x) is based on the formula The obtained, where x m Let x be the m-th text feature vector. jLet be the j-th text feature vector (excluding the m-th text feature vector) in the set of text feature vectors of the same group of enterprises, U be the set of text feature vectors of the same group of enterprises, n be the number of text feature vectors in the set of text feature vectors of the same group of enterprises, and s be the number of text feature vectors. m Let x be the standard deviation of the magnitude of each text feature vector in the set of text feature vectors of the same group of enterprises. i Let i be the feature vector of the i-th text. This is the average value of the set of text feature vectors.
[0054] Understandably, the reliability of the differences obtained in the manner described is high.
[0055] S103: The server sorts the values of each text feature vector from high to low.
[0056] S104: The server recommends the top N text feature vectors by value to the terminal device, so that the user can label the categories of each of the top N text feature vectors on the terminal device, where N is a positive integer.
[0057] S105: The server uses the input data and output data to train the initial network to be trained, and obtains a classification model. The input data consists of the top N text feature vectors, and the output data consists of the categories of the top N text feature vectors.
[0058] S106: The server obtains the next set of text feature vectors of enterprises and returns the execution server to determine the value of each text feature vector in the set of text feature vectors for training the classification model, until the output data obtained in two consecutive adjacent times meet the preset termination condition, so as to obtain the classification model trained for the last time.
[0059] Based on S101-S106 above, it is known that the classification model selects the top N text feature vectors by value from the set of text feature vectors of each group of enterprises to label the enterprise category, and uses the top N text feature vectors as input data and the enterprise category as output data to train the initial training network. Thus, the trained classification model has high reliability. Furthermore, since the value of any text feature vector is obtained by weighted summation based on the information entropy of the text feature vector, the similarity between the text feature vector and the cluster center of its set, and the difference between the text feature vector and other text feature vectors in its set, the accuracy and reliability of the value of any text feature vector are high. This further enhances the reliability of the trained classification model.
[0060] Below, in conjunction with Figure 2 Explain the steps involved in evaluating a company. For example... Figure 2As shown, the evaluation steps for server companies include:
[0061] S201: The server received an evaluation request, which included the company to be evaluated.
[0062] For example, a terminal device may display a list of companies. In response to selecting a company from the list, the terminal device may determine the company to be evaluated. The terminal device may then send an evaluation request to the server. This evaluation request may be used to request a risk assessment.
[0063] S202: The server obtains the text feature vector of the company to be evaluated based on the evaluation request.
[0064] The text feature vector of the company to be evaluated is obtained from the company's business scope and / or patent information. For example, S202 can be implemented as follows:
[0065] The server determines the Boolean word frequency, part-of-speech ratio, and text length ratio of each keyword in the text describing green and non-green enterprises, based on preset keywords, within the enterprise's business scope and / or patent information, and the ratio of the text length to the longest word in the enterprise's business scope and / or patent information.
[0066] For example, for keyword A, if the company's business scope and / or patent information includes keyword A, then the Boolean frequency of keyword A is 1; otherwise, the Boolean frequency of keyword A is 0. If the part of speech of keyword A is an adjective, and the company's business scope and / or patent information includes adjectives, verbs, nouns, quantifiers, and pronouns, then the part of speech ratio of keyword A is 1 / 5. If the text length of keyword A is 4, and the text length of the longest word in the company's business scope and / or patent information is 12, then the ratio of the text length of keyword A to the text length of the longest word in the company's business scope and / or patent information is 4 / 12.
[0067] Then, the server processes each keyword in the text describing green and non-green enterprises based on the formula CF=WF(w)*[POS(w)+WL(w)], and obtains the text feature vector of the enterprise to be evaluated.
[0068] Wherein, CF is the feature vector of keywords in the text feature vector of the enterprise to be evaluated, POS(w) is the part-of-speech ratio, and WL(w) is the ratio of the text length of the keywords to the text length of the longest word in the enterprise's business scope and / or patent information.
[0069] Understandably, since the keyword feature vector in the text feature vector of the company to be evaluated is obtained by considering the Boolean word frequency, part-of-speech ratio, and text length ratio of each keyword in the text describing green and non-green enterprises within the company's business scope and / or patent information, it better expresses the characteristics of the text describing keywords for green and non-green enterprises. Therefore, the text feature vector of the company to be evaluated obtained based on the formula CF=WF(w)*[POS(w)+WL(w)] has high reliability.
[0070] It should be noted that because the set of text feature vectors of the companies to be evaluated is of a large magnitude, the dimensionality of the set of text feature vectors will be excessively high, and data sparsity will exist. Therefore, the server can use principal component analysis (PCA) for dimensionality reduction. Through linear transformation, the set of text feature vectors of the companies to be evaluated is transformed into a set of linearly independent representations, extracting the principal feature components of the set of text feature vectors. This process maps the set of text feature vectors of the companies to be evaluated from a high-dimensional space to a low-dimensional space, obtaining a meaningful low-dimensional representation of the high-dimensional set of text feature vectors of the companies to be evaluated.
[0071] In another possible implementation, the text feature vector of the company to be evaluated also includes subject features describing the company's patent name. This enriches the content of the text feature vector of the company to be evaluated and improves its reliability.
[0072] Specifically, the topic features are extracted using a topic feature extraction model based on Latent Dirichlet Allocation (LDA), where the expression for the topic feature L is L = [l1, l2, l3, ..., l...]. t ], where t is the dimension of the topic feature, l t Let be the probability of a patent name under the t-th topic.
[0073] It should be noted that the set of text feature vectors of enterprises in S101 above can also be implemented in the manner of S202, and is not limited here.
[0074] S203: The server uses a pre-trained classification model to classify the text feature vectors of the company to be evaluated in order to determine the category of the company to be evaluated.
[0075] It should be noted that the training process for the classification model can refer to the above. Figure 1 This will not be elaborated upon here.
[0076] In addition, the classification model can also output the green rating of the company being evaluated. For example, the green rating can be light green, medium green, or dark green, etc., without limitation.
[0077] S204: The server obtains the assessment mode corresponding to the category of the enterprise to be assessed, and performs assessment processing on the enterprise to be assessed according to the assessment mode.
[0078] For example, the server can use the enterprise category as an indicator for risk assessment when evaluating the enterprise.
[0079] In summary, the enterprise evaluation method provided in this application involves a server using a pre-trained classification model to classify the text feature vectors of the enterprises to be evaluated. The classification model selects the top N text feature vectors by value from the set of text feature vectors of each group of enterprises to label the enterprise category, and uses the top N text feature vectors as input data and the enterprise category as output data to train an initial training network. This results in a highly reliable classification model. Furthermore, the value of any text feature vector is obtained by weighted summation based on the information entropy of the text feature vector, the similarity between the text feature vector and the cluster center of its set, and the difference between the text feature vector and other text feature vectors in its set. This also results in high accuracy and reliability of the value of any text feature vector. This further enhances the reliability of the trained classification model. Consequently, the accuracy of determining the category of the enterprise to be evaluated based on the classification model is also high. Therefore, the reliability of the evaluation results based on the evaluation mode corresponding to the category of the enterprise to be evaluated is also high.
[0080] Please see Figure 3 This application also provides an enterprise evaluation processing device 300, applied to a server. It should be noted that the basic principle and technical effects of the enterprise evaluation processing device 300 provided in this application are the same as those in the above embodiments. For the sake of brevity, any parts not mentioned in the embodiments of this application can be referred to the corresponding content in the above embodiments. The enterprise evaluation processing device 300 provided in this application includes:
[0081] The system comprises the following components: a request acquisition unit for acquiring an evaluation request, which includes the enterprise to be evaluated; a text acquisition unit for acquiring the text feature vector of the enterprise to be evaluated based on the evaluation request, wherein the text feature vector of the enterprise to be evaluated is obtained from the enterprise's business scope and / or patent information; an enterprise classification unit for classifying the text feature vector of the enterprise to be evaluated using a pre-trained classification model to determine the category of the enterprise to be evaluated, wherein the enterprise category is either a green enterprise or a non-green enterprise. The classification model selects the top N text feature vectors by value from the set of text feature vectors of each group of enterprises to label the category of the enterprise, and uses the top N text feature vectors as input data and the enterprise category as output data to train an initial training network. The value of any text feature vector is obtained by weighted summation based on the information entropy of the text feature vector, the similarity between the text feature vector and the cluster center of the set of text feature vectors to which it belongs, and the difference between the text feature vector and other text feature vectors in the set of text feature vectors to which it belongs; and an enterprise evaluation unit for acquiring the evaluation mode corresponding to the category of the enterprise to be evaluated and performing evaluation processing on the enterprise to be evaluated according to the evaluation mode.
[0082] In one possible implementation, the enterprise evaluation processing apparatus 300 provided in this application embodiment further includes: a value calculation unit, used to calculate the value based on information entropy H(x), similarity I(x), and difference D(x) using the formula Q(x) = H(x). λ ×I(x) μ ×D(x) (1-λ-μ) Calculate the value of any text feature vector. Where Q(x) is the value, λ and μ are the weights, λ and μ ∈ [0, 1] and λ + μ ≤ 1.
[0083] In one possible implementation, the value calculation unit is specifically used to calculate the value based on the formula H(x) = -∑ i p(y i |x)logp(y i Calculate the information entropy H(x) using |x). Where p(y) i |x) represents the text feature vector x, which belongs to the enterprise category y. i The probability of the enterprise category y i Including green and non-green enterprises, the enterprise category y i It is the output after the text feature vector x is input into the initial classification model. The initial classification model is obtained by inputting the text feature vectors of the categories of labeled enterprises (less than a preset number) into the initial training network for training.
[0084] In one possible implementation, the value calculation unit is specifically used to calculate according to the formula. Calculate the similarity I(x). Where K is the number of clusters formed by clustering the text feature vectors of each enterprise, and x... (k) The feature vector representing the cluster center of the k-th cluster is x, where x is the text feature vector.
[0085] In one possible implementation, the value calculation unit is specifically used to calculate according to the formula. Calculate the difference D(x). Where x m Let x be the m-th text feature vector. j Let be the j-th text feature vector (excluding the m-th text feature vector) in the set of text feature vectors of the same group of enterprises, U be the set of text feature vectors of the same group of enterprises, n be the number of text feature vectors in the set of text feature vectors of the same group of enterprises, and s be the number of text feature vectors. m It is the standard deviation of the magnitude of each text feature vector in the set of text feature vectors of the same group of enterprises.
[0086] In one possible implementation, the text acquisition unit is specifically used to determine the Boolean word frequency, part-of-speech ratio, and text length ratio of each keyword in the text describing green and non-green enterprises in the enterprise's business scope and / or patent information, as well as the ratio of the text length of the keyword to the text length of the longest word in the enterprise's business scope and / or patent information. The server processes each keyword in the text describing green and non-green enterprises based on the formula CF = WF(w) * [POS(w) + WL(w)] to obtain the text feature vector of the enterprise to be evaluated, where CF is the feature vector of the keyword in the text feature vector of the enterprise to be evaluated, POS(w) is the part-of-speech ratio, and WL(w) is the ratio of the text length of the keyword to the text length of the longest word in the enterprise's business scope and / or patent information.
[0087] In one possible implementation, the text feature vector of the enterprise to be evaluated also includes topic features describing the enterprise's patent names. These topic features are extracted using a topic feature extraction model based on Latent Dirichlet Allocation (LDA), where the expression for the topic feature L is L = [l1, l2, l3, ..., l...]. t ], where t is the dimension of the topic feature, l t Let be the probability of a patent name under the t-th topic.
[0088] Figure 4 This is a block diagram illustrating a server according to an exemplary embodiment. The server may include one or more of the following components: a processing component 402, a memory 404, a power supply component 406, an input / output (I / O) interface 412, and a communication component 416.
[0089] Processing component 402 typically controls the overall operation of device 400. Processing component 402 may include one or more processors 520 to execute instructions to complete all or part of the steps of the method described above. In addition, processing component 402 may include one or more modules to facilitate interaction between processing component 402 and other components.
[0090] Memory 404 is configured to store various types of data to support the operation of device 400. Examples of such data include instructions for any application or method operating on device 400. Memory 404 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), and read-only memory (ROM).
[0091] Power supply component 406 provides power to various components of device 400. Power supply component 406 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to device 400.
[0092] I / O interface 412 provides an interface between processing component 402 and peripheral interface module, which may be a USB interface, keyboard interface, etc.
[0093] Communication component 416 is configured to facilitate wired or wireless communication between device 400 and other devices. Device 400 can access wireless networks based on communication standards. In one exemplary embodiment, communication component 416 also includes a near-field communication (NFC) module to facilitate short-range communication.
[0094] In an exemplary embodiment, the apparatus 400 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the methods described above.
[0095] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 404 including instructions, which can be executed by a processor 420 of device 400 to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc. When the instructions in this non-transitory computer-readable storage medium are executed by the processor of a server, the server is able to perform the above-described method. Figure 1 or Figure 2 The method.
[0096] This application also provides a computer program product, including a computer program, which, when executed by a processor, performs the above-described... Figure 1 and Figure 2 The method.
[0097] Finally, it should be noted that the collection, storage, use, processing, transmission, provision, and disclosure of financial data or user data involved in the technical solutions of this application all comply with the provisions of relevant laws and regulations and do not violate public order and good morals. The above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit them; although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.
Claims
1. A method of enterprise assessment processing, characterized by, Applied to a server, the method includes: The server receives an evaluation request, which includes the enterprise to be evaluated. The server obtains the text feature vector of the enterprise to be evaluated according to the evaluation request, wherein the text feature vector of the enterprise to be evaluated is obtained from the enterprise's business scope and / or the enterprise's patent information; The server uses a pre-trained classification model to classify the text feature vectors of the enterprises to be evaluated, thereby determining the category of the enterprises to be evaluated. The category of the enterprises is either green enterprises or non-green enterprises. The classification model selects the top N text feature vectors by value from the set of text feature vectors of each group of enterprises to label the category of the enterprise. The top N text feature vectors are used as input data and the category of the enterprise is used as output data to train an initial training network. The value of any text feature vector is obtained by weighted summation based on the information entropy of the text feature vector, the similarity between the text feature vector and the cluster center of the set of text feature vectors to which it belongs, and the difference between the text feature vector and other text feature vectors in the set of text feature vectors to which it belongs. The server obtains the evaluation mode corresponding to the category of the enterprise to be evaluated, and performs evaluation processing on the enterprise to be evaluated according to the evaluation mode; wherein the value of any of the text feature vectors is derived from the information entropy , the similarity , the difference , using the equation , where is the value, are the weights, respectively, and ; wherein the information entropy is obtained according to an algorithm wherein, is a probability that the text feature vector x belongs to a category of enterprises , the category of enterprises including green enterprises and non-green enterprises, the category of enterprises is output after inputting the text feature vector x into an initial classification model, and the initial classification model is obtained by inputting text feature vectors of categories of enterprises less than a preset number into the initial training network for training. Wherein, the similarity It is based on the formula The result is that, where K is the number of clusters formed by clustering the set of text feature vectors of each group of enterprises. Representing the The feature vectors of the cluster centers of each cluster, where x is the text feature vector; Among them, the difference It is based on the formula Of the results, Let m be the feature vector of the text. Let U be the set of text feature vectors of the same group of enterprises, excluding the m-th text feature vector, and n be the number of text feature vectors in the set of text feature vectors of the same group of enterprises. It is the standard deviation of the magnitude of each text feature vector in the set of text feature vectors of the same group of enterprises.
2. The method according to claim 1, characterized in that, The server obtains the text feature vector of the enterprise to be evaluated based on the evaluation request, including: The server determines the Boolean word frequency, part-of-speech ratio, and text length ratio of each keyword in the text containing preset keywords describing green and non-green enterprises, in the enterprise's business scope and / or the enterprise's patent information, and the ratio of the text length to the longest word in the enterprise's business scope and / or the enterprise's patent information. The server is based on an algorithm. Each keyword in the text describing green and non-green enterprises is processed to obtain the text feature vector of the enterprise to be evaluated, where CF is the feature vector of the keyword in the text feature vector of the enterprise to be evaluated, POS(w) is the part-of-speech ratio, WL(w) is the ratio of the text length of the keyword to the text length of the longest word in the business scope and / or the patent information of the enterprise, and WF(w) is the Boolean word frequency of the keyword in the business scope and / or the patent information of the enterprise.
3. The method according to claim 1, characterized in that, The text feature vector of the company to be evaluated also includes thematic features used to describe the company's patent names.
4. The method according to claim 3, characterized in that, The topic features are extracted using a topic feature extraction model based on Latent Dirichlet Allocation (LDA), wherein the expression for the topic feature L is: Where t is the dimension of the topic feature, Let be the probability of the patent name under the t-th topic.
5. An evaluation and processing device for an enterprise, characterized in that, Applied to a server, the device includes: The request acquisition unit is used to acquire an evaluation request, wherein the evaluation request includes the enterprise to be evaluated; The text acquisition unit is configured to acquire the text feature vector of the enterprise to be evaluated according to the evaluation request, wherein the text feature vector of the enterprise to be evaluated is acquired from the business scope of the enterprise and / or the patent information of the enterprise; The enterprise classification unit is used to classify the text feature vectors of the enterprises to be evaluated using a pre-trained classification model to determine the category of the enterprises to be evaluated. The category of the enterprises is either green enterprises or non-green enterprises. The classification model selects the top N text feature vectors by value from the set of text feature vectors of each group of enterprises to label the category of the enterprises. The top N text feature vectors are used as input data and the category of the enterprises is used as output data to train an initial training network. The value of any text feature vector is obtained by weighted summation based on the information entropy of the text feature vector, the similarity between the text feature vector and the cluster center of the set of text feature vectors to which it belongs, and the difference between the text feature vector and other text feature vectors in the set of text feature vectors to which it belongs. The enterprise assessment unit is used to obtain the assessment model corresponding to the category of the enterprise to be assessed, and to conduct assessment processing on the enterprise to be assessed according to the assessment model. The value of any of the text feature vectors is based on the information entropy. The similarity The aforementioned differences Using formulas Of the results, For value, They are weights, and ; Among them, the information entropy It is based on the formula Of the results, The text feature vector x belongs to the category of the enterprise. The probability, the type of enterprise Including green enterprises and non-green enterprises, the categories of the enterprises It is the output after inputting the text feature vector x into the initial classification model. The initial classification model is obtained by inputting the text feature vectors of the categories of labeled enterprises (less than a preset number) into the initial training network for training. Wherein, the similarity It is based on the formula The result is that, where K is the number of clusters formed by clustering the set of text feature vectors of each group of enterprises. Representing the The feature vectors of the cluster centers of each cluster, where x is the text feature vector; Among them, the difference It is based on the formula Of the results, Let m be the feature vector of the text. Let U be the set of text feature vectors of the same group of enterprises, excluding the m-th text feature vector, and n be the number of text feature vectors in the set of text feature vectors of the same group of enterprises. It is the standard deviation of the magnitude of each text feature vector in the set of text feature vectors of the same group of enterprises.
6. A server comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it causes the server to perform the method as described in any one of claims 1 to 4.
7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it causes the computer to perform the method as described in any one of claims 1 to 4.