CN117745102A - Enterprise health degree assessment method and device based on convolutional neural network - Google Patents
Enterprise health degree assessment method and device based on convolutional neural network Download PDFInfo
- Publication number
- CN117745102A CN117745102A CN202311812047.3A CN202311812047A CN117745102A CN 117745102 A CN117745102 A CN 117745102A CN 202311812047 A CN202311812047 A CN 202311812047A CN 117745102 A CN117745102 A CN 117745102A
- Authority
- CN
- China
- Prior art keywords
- information
- enterprise
- matrix
- health
- label
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000036541 health Effects 0.000 title claims abstract description 131
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 28
- 239000011159 matrix material Substances 0.000 claims abstract description 167
- 238000012545 processing Methods 0.000 claims abstract description 49
- 238000013210 evaluation model Methods 0.000 claims abstract description 25
- 238000004364 calculation method Methods 0.000 claims abstract description 14
- 238000004590 computer program Methods 0.000 claims description 16
- 230000006870 function Effects 0.000 claims description 15
- 238000002372 labelling Methods 0.000 claims description 14
- 239000013598 vector Substances 0.000 claims description 13
- 238000010606 normalization Methods 0.000 claims description 10
- 238000010276 construction Methods 0.000 claims description 9
- 238000011176 pooling Methods 0.000 claims description 9
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 230000010365 information processing Effects 0.000 claims description 6
- 238000005516 engineering process Methods 0.000 abstract description 4
- 230000000694 effects Effects 0.000 abstract description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 16
- 238000010586 diagram Methods 0.000 description 8
- 238000011156 evaluation Methods 0.000 description 6
- 230000005540 biological transmission Effects 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 230000002159 abnormal effect Effects 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000000737 periodic effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The embodiment of the invention provides an enterprise health assessment method and device based on a convolutional neural network, and relates to the technical field of enterprise health assessment technology. The method comprises the following steps: acquiring first information of an enterprise, and constructing the first information of the enterprise into an information matrix and a label matrix according to a preset rule; carrying out enterprise health score calculation on the information square matrix and the tag matrix through a preset health evaluation model so as to obtain enterprise health scores; and under the condition that the enterprise health score does not meet the preset condition, performing risk feedback processing. The method solves the problem of low enterprise health degree recognition precision, and further achieves the effect of improving enterprise health degree recognition efficiency.
Description
Technical Field
The embodiment of the invention relates to the technical field of enterprise health assessment technology, in particular to an enterprise health assessment method and device based on a convolutional neural network.
Background
Convolutional neural networks have been widely used and developed in the field of financial analysis in recent years. With the continuous development of deep learning technology, CNN starts to be applied to enterprise bankruptcy prediction, and by evaluating the health degree of enterprises, the enterprises can be helped to select proper partners, and investors can be helped to make correct investment decisions.
Convolutional neural networks perform well in the field of image recognition, however, the data adopted for enterprise health assessment are usually numerical data, so research in this field is still in the beginning stage.
A comprehensive bankruptcy database comprising 11827 American marketing companies is constructed by adopting a bankruptcy prediction deep learning model based on text disclosure, text data is used in combination with traditional financial ratio-based and market-based variables, and the prediction precision of the deep learning model is further improved. The method also adopts the convolutional neural network, and the feasibility of the convolutional neural network in the field of enterprise health evaluation is proved again.
The deep learning-based bankruptcy prediction collects 44 financial ratio information of 28000 enterprises and 1120 bankruptcy enterprises as training data, and the method also adopts a convolutional neural network as a model. A total of 88 of two years of financial ratio information, filled with 100 pixels, was fed into the convolutional neural network in a three-dimensional matrix arranged in time series in 7 x 2. The method demonstrates that the time evolution of data can be captured even in the manner of convolutional neural networks.
The above method has the following problems: 1. the data sets are all from the annual newspaper of the enterprise, and the information provided by the annual newspaper of the enterprise usually has certain hysteresis; 2. part of enterprises have lost financial interest rate items, so that the model has low universality; 3. the falsification of financial interest rate data results in inaccurate assessment models.
Disclosure of Invention
The embodiment of the invention provides an enterprise health assessment method and device based on a convolutional neural network, which at least solve the problem of inaccurate assessment model in the related technology.
According to one embodiment of the present invention, there is provided an enterprise health assessment method based on a convolutional neural network, including:
acquiring first information of an enterprise, and constructing the first information of the enterprise into an information matrix and a label matrix according to a preset rule; the first enterprise information at least comprises enterprise bank flow information;
carrying out enterprise health score calculation on the information square matrix and the tag matrix through a preset health evaluation model so as to obtain enterprise health scores;
and under the condition that the enterprise health score does not meet the preset condition, performing risk feedback processing.
In an exemplary embodiment, the constructing the first information of the enterprise into an information square matrix and a tag matrix according to a preset rule includes:
constructing the first information of the enterprise into the information square matrix according to the time sequence;
performing binary labeling processing on the first information of the enterprise to obtain an enterprise information label, and constructing a label matrix based on the enterprise information label;
constructing a three-dimensional information matrix based on the information square matrix and the tag matrix;
normalizing the three-dimensional information matrix to obtain a target three-dimensional matrix;
and the health assessment model calculates enterprise health scores of the target three-dimensional matrix to obtain the enterprise health scores.
In one exemplary embodiment, the health assessment model performing an enterprise health score calculation on the target three-dimensional matrix to obtain the enterprise health score comprises:
the health evaluation model carries out convolution pooling treatment on the target three-dimensional matrix in a convolution pooling layer so as to obtain a first treatment result;
performing one-dimensional expansion processing on the first processing result to obtain a one-dimensional vector;
inputting the one-dimensional vector into a full connection layer of the health evaluation model, so that the full connection layer performs a first operation on the one-dimensional vector;
and carrying out health probability calculation on a first operation result through a first function to obtain the enterprise health score, wherein the health evaluation model comprises the first function.
In an exemplary embodiment, the performing binary labeling processing on the first information of the enterprise to obtain an enterprise information tag, and constructing a tag matrix based on the enterprise information tag includes:
classifying types of the first information of the enterprise to obtain tag types, wherein the number of the tag types is a preset value;
based on the label type, performing binary labeling processing on the first information of the enterprise to obtain an enterprise information label with the label length of the preset value;
carrying out binary conversion processing on the enterprise information label to obtain a pixel brightness value corresponding to the enterprise information label;
and constructing the label matrix based on the pixel brightness value.
In an exemplary embodiment, the constructing the first information of the enterprise into an information square matrix and a tag matrix according to a preset rule includes:
and under the condition that the data volume of the first enterprise information does not meet the data volume condition, constructing the preset element values and the first enterprise information into the information square matrix and the tag matrix according to a preset rule.
In an exemplary embodiment, after the first information of the enterprise is constructed into an information matrix and a tag matrix according to a preset rule, the method further includes:
and under the condition that the data size of the information matrix and/or the label matrix does not meet the size requirement, carrying out data size adjustment processing on the information matrix and/or the label matrix to obtain an adjusted information matrix and/or label matrix.
According to another embodiment of the present invention, there is provided an enterprise health assessment system based on a convolutional neural network, including:
the enterprise information processing module is used for acquiring first enterprise information and constructing the first enterprise information into an information square matrix and a label matrix according to a preset rule; the first enterprise information at least comprises enterprise bank flow information;
the health scoring module is used for calculating enterprise health scores of the information square matrix and the tag matrix through a preset health evaluation model so as to obtain enterprise health scores;
and the risk feedback module is used for carrying out risk feedback processing under the condition that the enterprise health score does not meet the preset condition.
In one exemplary embodiment, the enterprise information processing module includes:
the information square matrix construction unit is used for constructing the first information of the enterprise into the information square matrix according to the time sequence;
the label matrix construction unit is used for performing binary labeling processing on the first information of the enterprise to obtain an enterprise information label, and constructing a label matrix based on the enterprise information label;
the three-dimensional matrix construction unit is used for constructing a three-dimensional information matrix based on the information square matrix and the tag matrix;
the normalization processing unit is used for carrying out normalization processing on the three-dimensional information matrix to obtain a target three-dimensional matrix;
and the scoring calculation unit is used for calculating enterprise health scores of the target three-dimensional matrix by the health evaluation model so as to obtain the enterprise health scores.
According to a further embodiment of the invention, there is also provided a computer readable storage medium having stored therein a computer program, wherein the computer program is arranged to perform the steps of any of the method embodiments described above when run.
According to a further embodiment of the invention, there is also provided an electronic device comprising a memory having stored therein a computer program and a processor arranged to run the computer program to perform the steps of any of the method embodiments described above.
According to the invention, the information square matrix and the label matrix are processed on the enterprise information, and the two models are processed by combining the deep learning neural network model, so that the accurate identification of the enterprise health condition is realized, the problem of low enterprise health degree identification precision can be solved, and the effect of improving the enterprise health degree identification precision and efficiency is achieved.
Drawings
FIG. 1 is a block diagram of a hardware architecture of a mobile terminal of an enterprise health assessment method based on a convolutional neural network according to an embodiment of the present invention;
FIG. 2 is a flow chart of an enterprise health assessment method based on convolutional neural network, in accordance with an embodiment of the present invention;
FIG. 3 is a schematic diagram of a specific embodiment according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a model structure according to an embodiment of the present invention;
FIG. 5 is a block diagram of an enterprise health assessment apparatus based on a convolutional neural network, in accordance with an embodiment of the present invention;
FIG. 6 is a schematic diagram of a system architecture according to an embodiment of the invention;
fig. 7 is an output diagram of layers according to an embodiment of the present invention.
Detailed Description
Embodiments of the present invention will be described in detail below with reference to the accompanying drawings in conjunction with the embodiments.
It should be noted that the terms "first," "second," and the like in the description and the claims of the present invention and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order.
The method embodiments provided in the embodiments of the present application may be performed in a mobile terminal, a computer terminal or similar computing device. Taking the operation on a mobile terminal as an example, fig. 1 is a hardware structure block diagram of a mobile terminal of an enterprise health assessment method based on a convolutional neural network according to an embodiment of the present invention. As shown in fig. 1, a mobile terminal may include one or more (only one is shown in fig. 1) processors 102 (the processor 102 may include, but is not limited to, a microprocessor MCU or a processing device such as a programmable logic device FPGA) and a memory 104 for storing data, wherein the mobile terminal may also include a transmission device 106 for communication functions and an input-output device 108. It will be appreciated by those skilled in the art that the structure shown in fig. 1 is merely illustrative and not limiting of the structure of the mobile terminal described above. For example, the mobile terminal may also include more or fewer components than shown in fig. 1, or have a different configuration than shown in fig. 1.
The memory 104 may be used to store a computer program, for example, a software program of an application software and a module, such as a computer program corresponding to an enterprise health assessment method based on a convolutional neural network in an embodiment of the present invention, and the processor 102 executes the computer program stored in the memory 104 to perform various functional applications and data processing, that is, implement the method described above. Memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include memory remotely located relative to the processor 102, which may be connected to the mobile terminal via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The transmission device 106 is used to receive or transmit data via a network. Specific examples of the network described above may include a wireless network provided by a communication provider of the mobile terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, simply referred to as NIC) that can connect to other network devices through a base station to communicate with the internet. In one example, the transmission device 106 may be a Radio Frequency (RF) module, which is configured to communicate with the internet wirelessly.
In this embodiment, an enterprise health assessment method based on a convolutional neural network is provided, and fig. 2 is a flowchart of an enterprise health assessment method based on a convolutional neural network according to an embodiment of the present invention, as shown in fig. 2, where the flowchart includes the following steps:
step S201, acquiring first information of an enterprise, and constructing the first information of the enterprise into an information square matrix and a label matrix according to a preset rule; the first enterprise information at least comprises enterprise bank flow information;
in this embodiment, relevant information of the enterprise is obtained, so that accurate evaluation of the health degree of the enterprise can be conveniently performed from multiple angles.
The first information of the enterprise comprises (but is not limited to) enterprise bank running information and label information corresponding to each running water, the default bank running information is 100 ten thousand pieces/year, and the size of a winning information square matrix of the enterprise bank running information is 1000 x 1000; the tag information includes at least the following categories: 27 kinds of labels such as privacy, public, overseas, incomplete data, large amount, new enterprises, small and micro enterprises, financial, non-main industry large amount transaction, periodic expenditure, concerned party, non-risk party, stakeholder, same legal person, same stakeholder, present enterprise share, inflow and outflow coincidence, fund borrowing, inward borrowing, outward borrowing, borrowing analysis, financial institution, whether the total amount of the medium mark in the last year is more than fifty percent of registered funds, whether the number of the medium marks in the last year is more than 10, and the like.
It will be readily appreciated that the number and type of labels may take other forms, only as one of which is illustrated herein.
The preset rules may be different according to actual requirements, for example, labeling each piece of flowing information and marking the labeling information.
Specific: each piece of pipeline data is compressed to a range of 0-255 using a normalization method. And secondly, presetting the sorting of label categories, namely 27 categories in total, wherein each label can generate a 27-bit binary code, and the bit is 1 if the label accords with the corresponding category, otherwise, the binary code is 0. And arranging the flow information and the label information into a plurality of square matrixes of 1000 x 1000 according to the time sequence and superposing the square matrixes. Step S202, calculating enterprise health scores of the information square matrix and the tag matrix through a preset health evaluation model to obtain enterprise health scores;
in this embodiment, the matrix formed by the running water information and the running water label is input into the health degree evaluation model, and the health degree of the enterprise is automatically and efficiently calculated through the evaluation model, so that the manual consumption is reduced, and meanwhile, the accuracy of evaluation can be improved because the calculation is performed on the matrix from multiple angles.
The health evaluation model adopts a neural network model with deep learning, such as a CNN model, an RNN model and the like.
Step S203, performing risk feedback processing when the enterprise health score does not meet a preset condition.
In this embodiment, when the health score of the enterprise is low, it is indicated that the enterprise has a high risk, and risk feedback is needed at this time, and an alarm is timely given.
The preset condition may be that the enterprise health score is greater than a certain preset threshold, and the threshold may be set according to a standard of an evaluation mechanism or a unit, or may be set according to an industry standard; similarly, the preset conditions may be set by the evaluation unit itself.
In an optional embodiment, the building the first information of the enterprise into the information square matrix and the tag matrix according to the preset rule includes:
step S2011, constructing the first information of the enterprise into the information square matrix according to a time sequence;
step S2012, performing binary labeling processing on the first information of the enterprise to obtain an enterprise information label, and constructing a label matrix based on the enterprise information label;
step S2013, constructing a three-dimensional information matrix based on the information square matrix and the tag matrix;
step S2014, carrying out normalization processing on the three-dimensional information matrix to obtain a target three-dimensional matrix;
and step S2015, the health assessment model calculates enterprise health scores of the target three-dimensional matrix to obtain the enterprise health scores.
In this embodiment, in addition to sorting the running water information according to the time sequence, the information matrix may be constructed according to other rules, for example, according to the amount of running water, the date of running water, and the like; the labels are marked in the form of binary codes, if the labels accord with a certain category, the category corresponds to the binary code being 1, otherwise, the binary code is 0, so that a plurality of labels can be formed into an N multiplied by M matrix, wherein N is the number of the labels, and M is the number of the categories in the labels; as shown in fig. 3, the three-dimensional information matrix is constructed by superposing a square matrix composed of flow information and a tag matrix composed of tag information into a three-dimensional matrix of 1000×1000×2; the normalization processing is performed on the one hand to reduce the amount of computation, and on the other hand to convert the elements in the tag matrix into values that can be represented by pixels, so that the relevant information can be computed from a pixel point of view.
It should be noted that the execution sequence of step S2011 and step S2012 may be exchanged, that is, step S2012 may be executed first, and then step S2011 may be executed.
In an optional embodiment, the performing binary labeling processing on the first information of the enterprise to obtain an enterprise information tag, and constructing a tag matrix based on the enterprise information tag includes:
step S20121, performing type classification on the first information of the enterprise to obtain label types, wherein the number of the label types is a preset value;
step S20122, based on the label type, performing binary labeling processing on the first information of the enterprise to obtain an enterprise information label with a label length of the preset value;
step S20123, carrying out binary conversion processing on the enterprise information label to obtain a pixel brightness value corresponding to the enterprise information label;
and step S20124, constructing the label matrix based on the pixel brightness value.
In this embodiment, the binary conversion process is performed to facilitate the conversion of binary information into data that can be subjected to other processes, thereby facilitating the processing of related data from other angles.
In the binary labeling process of the labels, as shown in fig. 3, a part of labels are selected as labels in the training process, for example, ten frequently occurring categories are selected and coded, for example, label information of a certain running string is represented by binary codes: 1010101010, a bit of 1 indicates that the pen is flowing according to a certain type of label, and so on; the health assessment model then converts the binary code of the tag to decimal, for example binary code 1010101010 to decimal 682, so that the model can identify the luminance value of the pixel for subsequent processing; the preset value may be a number of tag types, e.g. 27, to facilitate converting the binary code into a value (i.e. a grey value) that can represent the brightness of the pixel.
After obtaining the pixel luminance value, the processing may determine whether the related data is abnormal according to the distribution characteristics of the pixel luminance value (i.e., the position coordinate characteristics, the luminance of the adjacent point, etc.), for example, the distribution of the related luminance is (35/28/70/120/150) in the normal case, but if the actually obtained distribution is (33/30/68/120/255), the data represented by "255" is obviously abnormal, so that the related data is subjected to subsequent processing such as alarm. Namely, the method comprises the following steps: after the performing the binary conversion processing on the enterprise information tag to obtain the pixel brightness value corresponding to the enterprise information tag, the method further includes: determining distribution characteristics of the pixel brightness values based on the pixel brightness values, wherein the distribution characteristics at least comprise position distribution characteristics, pixel brightness value size characteristics and adjacent pixel brightness value distribution characteristics; normalizing the distribution characteristics to obtain target distribution characteristics, for example, marking the brightness value of the pixel points larger than a certain threshold value as 0, or else, marking the brightness value of the pixel points as 1; constructing a distribution feature matrix based on the target distribution feature; and carrying out correlation calculation on the distribution feature matrix, determining whether the distribution feature matrix is abnormal according to a correlation calculation result, and carrying out alarm processing under the condition that the distribution feature matrix is abnormal.
In an alternative embodiment, the calculating the enterprise health score for the target three-dimensional matrix by the health assessment model includes:
step S20151, the health assessment model performs convolution pooling processing on the target three-dimensional matrix in a convolution pooling layer to obtain a first processing result;
step S20152, performing one-dimensional expansion processing on the first processing result to obtain a one-dimensional vector;
step S20153, inputting the one-dimensional vector into a fully connected layer of the health assessment model, so that the fully connected layer performs a first operation on the one-dimensional vector;
in step S20154, a health probability is calculated on the first operation result through a first function, so as to obtain the enterprise health score, where the health assessment model includes the first function.
In this embodiment, as shown in fig. 4, after normalization processing is performed on an information matrix composed of stream information and a tag matrix, two gray images composed of pixel points are respectively obtained, data obtained after convolution and pooling of two gray images are expanded into a one-dimensional column vector (corresponding to the one-dimensional vector), the one-dimensional column vector is input into a full-connection layer, and the probability of enterprise health degree can be obtained after the data after operation of the full-connection layer is subjected to a layer of Softmax function (corresponding to the first function). The Softmax function causes the value of the model output (corresponding to the first operation result described above) to be between 0 and 1.
In an optional embodiment, the building the first information of the enterprise into the information square matrix and the tag matrix according to the preset rule includes:
in step S20121, when the data size of the first information of the enterprise does not meet the data size condition, the preset element value and the first information of the enterprise are constructed as the information square matrix and the tag matrix according to a preset rule.
In this embodiment, it is considered that some enterprises often have missing data, so that these data need to be filled to ensure consistency of data formats.
The preset element value may be 0 or other values.
In an optional embodiment, after the first information of the enterprise is constructed into an information matrix and a label matrix according to a preset rule, the method further includes:
step S20122 performs data size adjustment processing on the information matrix and/or the tag matrix to obtain an adjusted information matrix and/or tag matrix when the data size of the information matrix and/or the tag matrix does not meet the size requirement.
In this embodiment, it is considered that some enterprises often have missing data, and thus, the data needs to be adjusted to ensure consistency of data formats.
For example, the enterprise health assessment model employs a resnet50, wherein the input size of the model is modified to 1000×1000×2 (the original input size is 224×224×3), and the input sizes of stage1, stage2, stage3, stage4 are modified. Considering that part of enterprises are small micro-enterprises, when the flow rate is small, the input data can be filled to the default size by adopting a 0-pixel filling mode. The size of the last full connection layer is changed to 2048 x 2, and the probability value of the health and unhealthy of the enterprise is obtained through a softmax function.
From the description of the above embodiments, it will be clear to a person skilled in the art that the method according to the above embodiments may be implemented by means of software plus the necessary general hardware platform, but of course also by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) comprising instructions for causing a terminal device (which may be a mobile phone, a computer, a server, or a network device, etc.) to perform the method according to the embodiments of the present invention.
The embodiment also provides an enterprise health assessment system based on a convolutional neural network, and the device is used for realizing the embodiment and the preferred implementation mode, and the description is omitted. As used below, the term "module" may be a combination of software and/or hardware that implements a predetermined function. While the means described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware, is also possible and contemplated.
FIG. 5 is a block diagram of an enterprise health assessment system based on convolutional neural network, as shown in FIG. 5, comprising:
the enterprise information processing module 51 is configured to obtain first enterprise information, and construct the first enterprise information into an information square matrix and a tag matrix according to a preset rule; the first enterprise information at least comprises enterprise bank flow information;
the health scoring module 52 is configured to calculate an enterprise health score for the information square matrix and the tag matrix through a preset health evaluation model, so as to obtain an enterprise health score;
and the risk feedback module 53 is configured to perform risk feedback processing when the enterprise health score does not meet a preset condition.
The enterprise information processing module 51 includes:
the information square matrix construction unit is used for constructing the first information of the enterprise into the information square matrix according to the time sequence;
the label matrix construction unit is used for performing binary labeling processing on the first information of the enterprise to obtain an enterprise information label, and constructing a label matrix based on the enterprise information label;
the three-dimensional matrix construction unit is used for constructing a three-dimensional information matrix based on the information square matrix and the tag matrix;
the normalization processing unit is used for carrying out normalization processing on the three-dimensional information matrix to obtain a target three-dimensional matrix;
and the scoring calculation unit is used for calculating enterprise health scores of the target three-dimensional matrix by the health evaluation model so as to obtain the enterprise health scores.
It should be noted that each of the above modules may be implemented by software or hardware, and for the latter, it may be implemented by, but not limited to: the modules are all located in the same processor; alternatively, the above modules may be located in different processors in any combination.
The invention is illustrated below by means of specific examples.
Firstly, as shown in fig. 6, the invention is an end-to-end framework, comprising a connection module, an adaptive input module, a monitoring module and a risk early warning module; the problems of inconsistent data input size, inaccurate enterprise health degree research and judgment, untimely risk early warning and the like are solved, and a good tool for evaluating the enterprise health degree is provided for a business operator.
Wherein:
the connection module is used for accessing a service system and connecting the original data;
the self-adaptive input module is used for adjusting the shape of the model and self-adaptively changing the receiving size of the model according to the different sizes of input data;
the health degree evaluation module is used for evaluating the health degree of the enterprise and outputting health degree probability;
the risk early warning module is used for feeding back health degree early warning information to business personnel in time.
Establishing an enterprise health evaluation model:
step 1: collecting and arranging bank flow information of enterprises; step 2: inputting the running water information and the running water label into a health degree evaluation model; step 3: and calculating the enterprise health degree score.
The content of the step 1 is mainly as follows: collecting and arranging enterprise bank running water information, arranging the running water information into a square matrix according to time sequence, wherein the default bank running water information is 100 ten thousand pieces/year, the square matrix size is 1000 x 1000, and each running water has corresponding label information, and the method comprises the following steps: 27 kinds of labels such as privacy, public, overseas, incomplete data, large amount, new enterprises, small and micro enterprises, financial, non-main industry large amount transaction, periodic expenditure, concerned correlation party, non-risk correlation party, stakeholder, same legal person, same stakeholder, present enterprise stock, inflow and outflow coincidence, fund borrowing, inward borrowing, outward borrowing, borrowing analysis, financial institution, whether the total amount of the medium mark in the last year is more than fifty percent of registered funds, whether the amount of the medium mark in the last year is more than 10 and the like.
As shown in fig. 3, the tag is marked with a binary code, and if the tag meets a certain category, the binary code corresponding to the category is 1, otherwise, the tag is 0. During the preprocessing stage, a portion of the tags will be selected as tags during the training process. Ten frequently occurring categories are selected and coded, for example, label information of a certain flowing water is expressed as binary codes: 1010101010, a bit of 1 indicates that the flowing water meets a certain type of label, and the flowing water is input into the health evaluation model to convert the binary code into decimal, and the binary code is converted into decimal 682, so that the model can identify the brightness value of the pixel point. The square matrix dimension is also 1000 x 1000. And then, the running water information matrix and the label information matrix are superimposed into a 1000 x 2 three-dimensional matrix, the value of each pixel point is normalized, and the value is compressed to be within the range of 0-255, so that the enterprise health evaluation model can normally operate. And superposing the label matrix and the pipeline information matrix into a three-dimensional matrix to be used as the input of the health evaluation model.
The content of the step 2 is as follows: as shown in fig. 4, the enterprise health assessment model is a convolutional neural network model, two gray images with the size of 1000 x 1000 are obtained in step 1, the health assessment model obtains two inputs at the same time, data obtained after convolution and pooling of two gray images are unfolded into one-dimensional column vectors and then input into a full-connection layer, and the probability of the enterprise health can be obtained after the data after operation of the full-connection layer is subjected to a softmax function. The Softmax function causes the value of the model output to be between 0 and 1;
the enterprise health assessment model comprises a convolution layer, a pooling layer, a full connection layer and an output layer.
As shown in fig. 7, the enterprise health assessment model employs a resnet50, wherein the input size of the model is modified to 1000×1000×2 (the original input size is 224×224×3), and the input sizes of stage1, stage2, stage3, stage4 are modified. Considering that part of enterprises are small micro-enterprises, when the flow rate is small, the input data can be filled to the default size by adopting a 0-pixel filling mode. The size of the last full connection layer is changed to 2048 x 2, and the probability value of the health and unhealthy of the enterprise is obtained through a softmax function.
The content of the step 3 is as follows: and giving out the probability of the enterprise health degree according to the model to score.
The model is input as enterprise bank flow information and label information, and the model gives a probability (between 0 and 1) of enterprise health degree through calculation of the evaluation model. The step 3 is completed as follows: the probability is multiplied by 100, namely the score of the health degree of the enterprise, and when the score of a certain enterprise is lower than a set threshold value, the system automatically alarms and feeds back risk information to business personnel.
Embodiments of the present invention also provide a computer readable storage medium having a computer program stored therein, wherein the computer program is arranged to perform the steps of any of the method embodiments described above when run.
In one exemplary embodiment, the computer readable storage medium may include, but is not limited to: a usb disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a removable hard disk, a magnetic disk, or an optical disk, or other various media capable of storing a computer program.
An embodiment of the invention also provides an electronic device comprising a memory having stored therein a computer program and a processor arranged to run the computer program to perform the steps of any of the method embodiments described above.
In an exemplary embodiment, the electronic apparatus may further include a transmission device connected to the processor, and an input/output device connected to the processor.
Specific examples in this embodiment may refer to the examples described in the foregoing embodiments and the exemplary implementation, and this embodiment is not described herein.
It will be appreciated by those skilled in the art that the modules or steps of the invention described above may be implemented in a general purpose computing device, they may be concentrated on a single computing device, or distributed across a network of computing devices, they may be implemented in program code executable by computing devices, so that they may be stored in a storage device for execution by computing devices, and in some cases, the steps shown or described may be performed in a different order than that shown or described herein, or they may be separately fabricated into individual integrated circuit modules, or multiple modules or steps of them may be fabricated into a single integrated circuit module. Thus, the present invention is not limited to any specific combination of hardware and software.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. An enterprise health assessment method based on a convolutional neural network is characterized by comprising the following steps:
acquiring first information of an enterprise, and constructing the first information of the enterprise into an information matrix and a label matrix according to a preset rule; the first enterprise information at least comprises enterprise bank flow information;
carrying out enterprise health score calculation on the information square matrix and the tag matrix through a preset health evaluation model so as to obtain enterprise health scores;
and under the condition that the enterprise health score does not meet the preset condition, performing risk feedback processing.
2. The method of claim 1, wherein the constructing the first information of the enterprise into an information matrix and a tag matrix according to a preset rule comprises:
constructing the first information of the enterprise into the information square matrix according to the time sequence;
performing binary labeling processing on the first information of the enterprise to obtain an enterprise information label, and constructing a label matrix based on the enterprise information label;
constructing a three-dimensional information matrix based on the information square matrix and the tag matrix;
normalizing the three-dimensional information matrix to obtain a target three-dimensional matrix;
and the health assessment model calculates enterprise health scores of the target three-dimensional matrix to obtain the enterprise health scores.
3. The method of claim 2, wherein the health assessment model performing an enterprise health score calculation on the target three-dimensional matrix to obtain the enterprise health score comprises:
the health evaluation model carries out convolution pooling treatment on the target three-dimensional matrix in a convolution pooling layer so as to obtain a first treatment result;
performing one-dimensional expansion processing on the first processing result to obtain a one-dimensional vector;
inputting the one-dimensional vector into a full connection layer of the health evaluation model, so that the full connection layer performs a first operation on the one-dimensional vector;
and carrying out health probability calculation on a first operation result through a first function to obtain the enterprise health score, wherein the health evaluation model comprises the first function.
4. The method of claim 2, wherein binary tagging the first information of the business to obtain a business information tag, and constructing a tag matrix based on the business information tag comprises:
classifying types of the first information of the enterprise to obtain tag types, wherein the number of the tag types is a preset value;
based on the label type, performing binary labeling processing on the first information of the enterprise to obtain an enterprise information label with the label length of the preset value;
carrying out binary conversion processing on the enterprise information label to obtain a pixel brightness value corresponding to the enterprise information label;
and constructing the label matrix based on the pixel brightness value.
5. The method of claim 1, wherein the constructing the first information of the enterprise into an information matrix and a tag matrix according to a preset rule comprises:
and under the condition that the data volume of the first enterprise information does not meet the data volume condition, constructing the preset element values and the first enterprise information into the information square matrix and the tag matrix according to a preset rule.
6. The method of claim 5, wherein after said building the first information of the business into the information matrix and the tag matrix according to the preset rule, the method further comprises:
and under the condition that the data size of the information matrix and/or the label matrix does not meet the size requirement, carrying out data size adjustment processing on the information matrix and/or the label matrix to obtain an adjusted information matrix and/or label matrix.
7. An enterprise health assessment system based on convolutional neural network, comprising:
the enterprise information processing module is used for acquiring first enterprise information and constructing the first enterprise information into an information square matrix and a label matrix according to a preset rule; the first enterprise information at least comprises enterprise bank flow information;
the health scoring module is used for calculating enterprise health scores of the information square matrix and the tag matrix through a preset health evaluation model so as to obtain enterprise health scores;
and the risk feedback module is used for carrying out risk feedback processing under the condition that the enterprise health score does not meet the preset condition.
8. The system of claim 7, wherein the enterprise information processing module comprises:
the information square matrix construction unit is used for constructing the first information of the enterprise into the information square matrix according to the time sequence;
the label matrix construction unit is used for performing binary labeling processing on the first information of the enterprise to obtain an enterprise information label, and constructing a label matrix based on the enterprise information label;
the three-dimensional matrix construction unit is used for constructing a three-dimensional information matrix based on the information square matrix and the tag matrix;
the normalization processing unit is used for carrying out normalization processing on the three-dimensional information matrix to obtain a target three-dimensional matrix;
and the scoring calculation unit is used for calculating enterprise health scores of the target three-dimensional matrix by the health evaluation model so as to obtain the enterprise health scores.
9. A computer readable storage medium, characterized in that the computer readable storage medium has stored therein a computer program, wherein the computer program is arranged to execute the method of any of the claims 1 to 6 when run.
10. An electronic device comprising a memory and a processor, characterized in that the memory has stored therein a computer program, the processor being arranged to run the computer program to perform the method of any of the claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311812047.3A CN117745102A (en) | 2023-12-26 | 2023-12-26 | Enterprise health degree assessment method and device based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311812047.3A CN117745102A (en) | 2023-12-26 | 2023-12-26 | Enterprise health degree assessment method and device based on convolutional neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117745102A true CN117745102A (en) | 2024-03-22 |
Family
ID=90256649
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311812047.3A Pending CN117745102A (en) | 2023-12-26 | 2023-12-26 | Enterprise health degree assessment method and device based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117745102A (en) |
-
2023
- 2023-12-26 CN CN202311812047.3A patent/CN117745102A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112036550B (en) | Client intention identification method and device based on artificial intelligence and computer equipment | |
| CN109165645A (en) | A kind of image processing method, device and relevant device | |
| CN113159147A (en) | Image identification method and device based on neural network and electronic equipment | |
| CN110097451B (en) | Bank business monitoring method and device | |
| CN105528465A (en) | Credit status assessment method and device | |
| CN112785441B (en) | Data processing method, device, terminal equipment and storage medium | |
| US20220292861A1 (en) | Docket Analysis Methods and Systems | |
| CN115302963B (en) | Bar code printing control method, system and medium based on machine vision | |
| WO2019214455A1 (en) | Data sequence prediction method and computing apparatus | |
| CN110705225A (en) | Contract marking method and device | |
| CN113011961B (en) | Method, device, equipment and storage medium for monitoring risk of company-related information | |
| CN113220875A (en) | Internet information classification method and system based on industry label and electronic equipment | |
| CN117036947A (en) | Image recognition-based agricultural risk early warning method, device, equipment and medium | |
| CN117745102A (en) | Enterprise health degree assessment method and device based on convolutional neural network | |
| CN111401154A (en) | Accurate delivery transparentization auxiliary operation device of commodity circulation based on AR | |
| CN115810006A (en) | Reinforcing steel bar counting method and system based on MobileNet V3 improved model | |
| CN112580505B (en) | Method and device for identifying network point switch door state, electronic equipment and storage medium | |
| US11436816B2 (en) | Information processing device, learning device, and storage medium storing learnt model | |
| CN115130473A (en) | Key information extraction method, model training method, related device and electronic equipment | |
| CN113504865A (en) | Work order label adding method, device, equipment and storage medium | |
| CN113901768A (en) | Standard file generation method, device, equipment and storage medium | |
| CN117217904A (en) | Project rating method and device, storage medium and electronic device | |
| CN116738343B (en) | Material data identification method and device for construction industry and electronic equipment | |
| Ratusny et al. | Characterizing customer ordering Behaviors in semiconductor supply chains with convolutional neural networks | |
| CN116109080B (en) | Building integrated management platform based on BIM and AR |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |