CN118052632A - Method and device for determining risk assessment model and computer readable storage medium - Google Patents
Method and device for determining risk assessment model and computer readable storage medium Download PDFInfo
- Publication number
- CN118052632A CN118052632A CN202410131104.XA CN202410131104A CN118052632A CN 118052632 A CN118052632 A CN 118052632A CN 202410131104 A CN202410131104 A CN 202410131104A CN 118052632 A CN118052632 A CN 118052632A
- Authority
- CN
- China
- Prior art keywords
- data
- target
- model
- transaction
- features
- 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
- 238000000034 method Methods 0.000 title claims abstract description 134
- 238000012502 risk assessment Methods 0.000 title claims abstract description 119
- 238000003860 storage Methods 0.000 title claims abstract description 26
- 238000012549 training Methods 0.000 claims abstract description 155
- 238000011156 evaluation Methods 0.000 claims abstract description 96
- 238000000605 extraction Methods 0.000 claims abstract description 73
- 238000012545 processing Methods 0.000 claims abstract description 71
- 238000013210 evaluation model Methods 0.000 claims abstract description 22
- 239000013598 vector Substances 0.000 claims description 24
- 230000002159 abnormal effect Effects 0.000 claims description 9
- 230000011218 segmentation Effects 0.000 claims description 6
- 238000004140 cleaning Methods 0.000 claims description 5
- 230000008030 elimination Effects 0.000 claims description 5
- 238000003379 elimination reaction Methods 0.000 claims description 5
- 238000002790 cross-validation Methods 0.000 claims description 4
- 238000001514 detection method Methods 0.000 claims description 4
- 238000013473 artificial intelligence Methods 0.000 abstract description 26
- 238000003058 natural language processing Methods 0.000 description 19
- 230000008569 process Effects 0.000 description 17
- 238000004590 computer program Methods 0.000 description 12
- 230000006399 behavior Effects 0.000 description 10
- 238000010586 diagram Methods 0.000 description 10
- 238000005516 engineering process Methods 0.000 description 8
- 238000004458 analytical method Methods 0.000 description 7
- 230000005540 biological transmission Effects 0.000 description 7
- 238000004422 calculation algorithm Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 7
- 241000282414 Homo sapiens Species 0.000 description 5
- 230000008859 change Effects 0.000 description 5
- 230000003068 static effect Effects 0.000 description 5
- 230000008451 emotion Effects 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 238000003672 processing method Methods 0.000 description 4
- 238000007637 random forest analysis Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000010276 construction Methods 0.000 description 3
- 238000007405 data analysis Methods 0.000 description 3
- 238000013480 data collection Methods 0.000 description 3
- 238000003066 decision tree Methods 0.000 description 3
- 238000012854 evaluation process Methods 0.000 description 3
- 238000007619 statistical method Methods 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- RWSOTUBLDIXVET-UHFFFAOYSA-N Dihydrogen sulfide Chemical compound S RWSOTUBLDIXVET-UHFFFAOYSA-N 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 238000010219 correlation analysis Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 238000013519 translation Methods 0.000 description 2
- 241000282412 Homo Species 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 230000001010 compromised effect Effects 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000002996 emotional effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 239000003607 modifier Substances 0.000 description 1
- 230000007935 neutral effect Effects 0.000 description 1
- 238000013450 outlier detection Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
- 238000011426 transformation method Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The application provides a method and a device for determining a risk assessment model and a computer readable storage medium, wherein the method comprises the following steps: acquiring target transaction data and target text data, wherein the target transaction data is structured data related to target transaction, and the target text data is unstructured data related to target transaction; respectively carrying out feature extraction processing on target transaction data and target text data to obtain target data features; performing repeated self-learning training on the target data characteristics by adopting an intelligent large language model to obtain a plurality of initial training models; and evaluating each initial training model to obtain a corresponding target evaluation score, and taking the initial training model with the highest target evaluation score as a risk evaluation model. The method carries out risk assessment on the clients through an artificial intelligence training model and uses more data sources to provide a more comprehensive client risk profile, so that the accuracy of the assessment is improved.
Description
Technical Field
The present application relates to the field of determination of risk assessment models, and in particular, to a method for determining a risk assessment model, a device for determining a risk assessment model, and a computer-readable storage medium.
Background
With the development of financial science and technology, conventional user risk assessment methods are often based on fixed rules and models, and are difficult to adapt to complex and variable financial environments.
The conventional customer risk assessment method generally comprises the following steps: acquiring client information, and collecting basic information of clients, such as age, occupation, income, credit history and the like; evaluating credit risk, wherein the credit risk of the client is evaluated by using information such as credit score, credit report, bank transaction record and the like of the client; auditing the financial report, auditing the financial statement and financial status of the customer to determine their repayment capacity and repayment willingness; performing manual evaluation, and evaluating the risk degree of the client through methods such as manual auditing, telephone access and the like; and implementing management measures, and taking proper risk management measures according to the risk assessment result of the client so as to reduce the risk.
The following disadvantages may exist with the conventional risk assessment method described above: data limiting, traditional customer risk assessment methods are generally based on limited data, such as credit history, revenue, assets, etc., and lack a comprehensive understanding of the customer; static evaluation, the traditional client risk evaluation method is biased to static evaluation, and can not reflect the change of clients in time; the flexibility is lacking, and the traditional client risk assessment method lacks flexibility and cannot adapt to different client groups and service demands; evaluation bias, which may exist in conventional customer risk evaluation methods, causes different customers to be erroneously evaluated to the same risk level; the time efficiency is low, and the conventional customer risk assessment method is usually manual in assessment process, so that a long time is required, which is unfavorable for quick decision-making.
From the above, the existing client risk assessment method has incomplete data source, low automation and intelligence, and low client risk assessment efficiency and inaccuracy due to the dependence on manual data processing.
Disclosure of Invention
The application aims to provide a method for determining a risk assessment model, a device for determining the risk assessment model and a computer readable storage medium, so as to at least solve the problems of low automation and low intelligence of the existing client risk assessment method, and low and inaccurate client risk assessment efficiency.
In order to achieve the above object, according to an aspect of the present application, there is provided a method of determining a risk assessment model, comprising: acquiring target transaction data and target text data, wherein the target transaction data is structured data related to target transaction, and the target text data is unstructured data related to the target transaction; respectively carrying out feature extraction processing on the target transaction data and the target text data to obtain target data features; performing repeated self-learning training on the target data characteristics by adopting an intelligent large language model to obtain a plurality of initial training models; and evaluating each initial training model to obtain a corresponding target evaluation score, and taking the initial training model with the highest target evaluation score as a risk evaluation model.
Optionally, acquiring the target transaction data and the target text data includes: acquiring initial transaction data and initial text data, wherein the initial transaction data is the target transaction data before being processed by adopting a first processing mode, and the initial text data is the target text data before being processed by adopting a second processing mode; processing the initial transaction data by adopting the first processing mode to obtain the target transaction data, wherein the first processing mode comprises at least one of the following steps: data cleaning, data duplication removal, missing value processing and abnormal value detection; processing the initial text data by adopting the second processing mode to obtain the target text data, wherein the second processing mode comprises at least one of the following steps: keyword extraction, word segmentation processing, non-keyword removal and stem extraction.
Optionally, performing feature extraction processing on the target transaction data and the target text data respectively to obtain target data features, including: performing feature extraction on the target transaction data by adopting a first feature extraction model to obtain basic features of the transaction data, and determining advanced features of the transaction data according to the basic features of the transaction data, wherein the basic features of the transaction data at least comprise one of the following: transaction frequency, transaction amount, transaction time, and transaction type, the transaction data high-level characteristics including at least one of: transaction volatility, transaction time, and abnormal transaction data; and carrying out feature extraction on the target text data by adopting a second feature extraction model to obtain advanced features of the text data, wherein the second feature extraction model at least comprises one of the following steps: a word bag model and a word embedding model; all the transaction data high-level features and all the text data high-level features are determined to be data features, and the target data features are determined from the data features.
Optionally, determining the target data feature from the data features includes: calculating the correlation between each data feature and the risk degree of the client to obtain a plurality of feature correlations, and determining the data feature corresponding to the feature correlations which are larger than or equal to a preset correlation as the target data feature; or recursively deleting the data features with importance degrees smaller than or equal to a preset importance degree by adopting a recursion feature elimination model so as to determine the target data features from the data features.
Optionally, performing feature extraction on the target text data by using a second feature extraction model to obtain advanced features of the text data, including: converting the target text data into vector data by adopting a word bag model, wherein each element in the vector data respectively represents the frequency of the corresponding word in the target text data; according to the vector data, determining characteristic parameters of each word, wherein the characteristic parameters at least comprise the frequency of the word and the rarity of the word, and the frequency of the word represents the frequency of the word in the target text data; and extracting the characteristics of the target text data according to the characteristic parameters of the words to obtain the advanced characteristics of the text data.
Optionally, performing repeated self-learning training on the target data features by using an intelligent large language model to obtain a plurality of initial training models, including: obtaining a client risk score report, wherein the client risk score report is obtained through training according to multiple sets of training data, and each set of training data in the multiple sets of training data comprises data obtained in a historical time period: a historical training model and a client risk score obtained by adopting the corresponding historical training model; determining a GPT model as the intelligent large language model, and initializing the GPT model to obtain an initialized GPT model; and repeating self-learning training on the target data features according to the client risk score report by adopting the initialized GPT model to obtain a plurality of initial training models.
Optionally, evaluating each initial training model to obtain a corresponding target evaluation score, including: determining an evaluation index, the evaluation index comprising at least one of: accuracy, recall and F1 score, wherein the F1 score represents a harmonic average value of the accuracy and the recall; evaluating each initial training model according to the evaluation index to obtain a first evaluation score; evaluating the generalization capability of each initial training model by adopting a cross-validation method to obtain a second evaluation score; and determining the target evaluation score corresponding to each initial training model according to the first evaluation score and the second evaluation score.
Optionally, after determining the initial training model with the highest target evaluation score as a risk evaluation model, the method further includes: and under the condition that new transaction data and/or new text data are acquired, adjusting the risk assessment model according to the new transaction data and/or the new text data to obtain an adjusted risk assessment model.
According to another aspect of the present application, there is provided a determining apparatus of a risk assessment model, including: the system comprises an acquisition unit, a storage unit and a processing unit, wherein the acquisition unit is used for acquiring target transaction data and target text data, the target transaction data is structured data related to target transaction, and the target text data is unstructured data related to the target transaction; the extraction unit is used for respectively carrying out feature extraction processing on the target transaction data and the target text data to obtain target data features; the training unit is used for carrying out repeated self-learning training on the target data characteristics by adopting an intelligent large language model to obtain a plurality of initial training models; and the determining unit is used for evaluating each initial training model to obtain a corresponding target evaluation score, and taking the initial training model with the highest target evaluation score as a risk evaluation model.
According to another aspect of the present application, there is provided a computer readable storage medium, where the computer readable storage medium includes a stored program, and when the program runs, the device in which the computer readable storage medium is controlled to execute any one of the methods for determining a risk assessment model.
By applying the technical scheme of the application, the method for determining the risk assessment model firstly acquires target transaction data and target text data, wherein the target transaction data is structured data related to target transaction, and the target text data is unstructured data related to target transaction; then, respectively carrying out feature extraction processing on the target transaction data and the target text data to obtain target data features; then, repeating self-learning training on the target data characteristics by adopting an intelligent large language model to obtain a plurality of initial training models; and finally, evaluating each initial training model to obtain a corresponding target evaluation score, and taking the initial training model with the highest target evaluation score as a risk evaluation model. According to the method, the risk assessment is carried out on the clients through the artificial intelligence training model, more data sources are used for providing more comprehensive client risk profiles, the assessment accuracy is improved, and the problems that the existing client risk assessment method is low in automation and intelligence, and the client risk assessment efficiency is low and inaccurate are solved.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the application. In the drawings:
Fig. 1 is a block diagram showing a hardware configuration of a mobile terminal for performing a determination method of a risk assessment model according to an embodiment of the present application;
FIG. 2 is a flow chart of a method for determining a risk assessment model according to an embodiment of the present application;
FIG. 3 is a flow chart of another method for determining a risk assessment model according to an embodiment of the present application;
fig. 4 shows a block diagram of a determination apparatus of a risk assessment model according to an embodiment of the present application.
Wherein the above figures include the following reference numerals:
102. A processor; 104. a memory; 106. a transmission device; 108. and an input/output device.
Detailed Description
It should be noted that, without conflict, the embodiments of the present application and features of the embodiments may be combined with each other. The application will be described in detail below with reference to the drawings in connection with embodiments.
In order that those skilled in the art will better understand the present application, a technical solution in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present application, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present application without making any inventive effort, shall fall within the scope of the present application.
It should be noted that the terms "first," "second," and the like in the description and the claims of the present application and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate in order to describe the embodiments of the application herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
For convenience of description, the following will describe some terms or terminology involved in the embodiments of the present application:
Natural language processing: natural Language Processing, NLP for short, is a very broad field in the field of artificial intelligence, aimed at simulating the ability of humans to understand and generate languages. It includes a number of techniques such as speech recognition, machine translation, emotion analysis, etc.
GPT (GENERATIVE PRE-trained Transformer): is a natural language generation model developed by OpenAI that has been trained on a large number of texts to obtain a pre-linguistic representation. The method is a general language generation model and can perform various language tasks such as text generation, question and answer, translation and the like. GPT is a specific NLP model that focuses on generating language.
Customer risk assessment: customer risk assessment is a method of assessing the credit risk of a customer. This typically includes collecting and evaluating financial and credit histories of the customer, and predicting the future financial condition of the customer to determine whether the customer has repayment capabilities for funds provided by the lending institution. This is important to loan institutions such as banks, credit companies, etc. because they need to ensure that their funds are not compromised by loans that are not repayment by customers.
Artificial intelligence: artificial intelligence (ARTIFICIAL INTELLIGENCE, AI) refers to a scientific technology that utilizes computer programs to simulate human intelligence and thinking so that the machine has certain intelligence and autonomy. The method comprises the technologies of machine learning, natural language processing, image recognition and the like, and can simulate the complex intelligent behaviors of human beings such as understanding, processing, reasoning, decision making and the like of information.
As described in the background, the risk assessment method in the prior art has the following disadvantages: data limiting, traditional customer risk assessment methods are generally based on limited data, such as credit history, revenue, assets, etc., and lack a comprehensive understanding of the customer; static evaluation, the traditional client risk evaluation method is biased to static evaluation, and can not reflect the change of clients in time; the flexibility is lacking, and the traditional client risk assessment method lacks flexibility and cannot adapt to different client groups and service demands; evaluation bias, which may exist in conventional customer risk evaluation methods, causes different customers to be erroneously evaluated to the same risk level; the time efficiency is low, and the conventional customer risk assessment method is usually manual in assessment process, so that a long time is required, which is unfavorable for quick decision-making.
In order to solve the problems of low customer risk assessment efficiency and inaccuracy caused by low automation and intellectualization of the existing customer risk assessment method, the embodiment of the application provides a method for determining a risk assessment model, a device for determining the risk assessment model and a computer-readable storage medium.
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
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 mobile terminal as an example, fig. 1 is a block diagram of a hardware structure of the mobile terminal according to a method for determining a risk assessment model according to an embodiment of the present application. 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 application software and a module, such as a computer program corresponding to a method for determining a risk assessment model in an embodiment of the present invention, and the processor 102 executes the computer program stored in the memory 104, thereby performing various functional applications and data processing, that is, implementing 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 a 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 the present embodiment, a method for determining a risk assessment model to be run on a mobile terminal, a computer terminal, or a similar computing device is provided, it should be noted that the steps illustrated in the flowchart of the drawings may be performed in a computer system such as a set of computer executable instructions, and that although a logical order is illustrated in the flowchart, in some cases, the steps illustrated or described may be performed in an order different from that herein.
Fig. 2 is a flowchart of a method of determining a risk assessment model according to an embodiment of the present application. As shown in fig. 2, the method comprises the steps of:
Step S201, obtaining target transaction data and target text data, wherein the target transaction data is structured data related to target transaction, and the target text data is unstructured data related to the target transaction;
Specifically, the target transaction information generally includes transaction information of the customer, credit card information, financial information, and the like. The target text data typically includes a question library, a knowledge library, a risk library, a report library, and the like.
In general, the customer risk assessment method based on the prior art has the following disadvantages: conventional methods typically require manual collection and analysis of data, which can require a significant amount of time, be prone to error, and limit the scope of evaluation, depending on manual data processing. The data sources are limited and conventional methods typically use only limited data sources, such as credit reports, which may not provide a comprehensive view of the customer's financial behavior and risk assessment. Lack of automation, traditional methods are often based on rules and heuristics that are affected by personal bias and interpretation, resulting in inconsistent and possibly inaccurate risk assessment. Lack of intelligence, inability to learn and evolve, and therefore, a balance of efficiency and accuracy is not achieved.
The method for acquiring the target transaction data and the target text data comprises the following steps:
Step S2011, acquiring initial transaction data and initial text data, wherein the initial transaction data is the target transaction data before being processed by adopting a first processing mode, and the initial text data is the target text data before being processed by adopting a second processing mode;
Step S2012, processing the initial transaction data by using the first processing method to obtain the target transaction data, where the first processing method includes at least one of the following: data cleaning, data duplication removal, missing value processing and abnormal value detection;
Step S2013, processing the initial text data by using the second processing method to obtain the target text data, where the second processing method includes at least one of the following: keyword extraction, word segmentation processing, non-keyword removal and stem extraction.
In particular, artificial intelligence models can provide a more complete view of customer financial behavior and risk status using a wide range of data sources, including transactional data, demographic data, and other unstructured data sources. Conventional approaches typically use only limited sources of data, such as credit reports, which may not provide a comprehensive view of the customer's financial behavior and risk profile.
In addition, the first processing mode generally includes: data cleansing, collecting and cleansing relevant transaction data of each customer, such as account balance, transaction frequency, amount and type; data deduplication, removing duplicate records in a dataset; missing value processing, namely selecting filling, deleting or estimating by using a statistical method for missing data; outlier detection, detecting and processing outliers by using a statistical method or a machine learning method; and (3) preprocessing the text, and performing word segmentation, word removal stopping, stem extraction and other operations on the NLP task.
Step S202, respectively carrying out feature extraction processing on the target transaction data and the target text data to obtain target data features;
in particular, the selection and construction of features from the transaction data and the text data that are relevant to the risk level of the customer is a key step, as these features will directly affect the performance of the risk assessment model.
The method comprises the following steps of:
Step S301, performing feature extraction on the target transaction data by adopting a first feature extraction model to obtain basic features of the transaction data, and determining advanced features of the transaction data according to the basic features of the transaction data, wherein the basic features of the transaction data at least comprise one of the following: transaction frequency, transaction amount, transaction time, and transaction type, the transaction data high-level characteristics including at least one of: transaction volatility, transaction time, and abnormal transaction data;
Additionally, based on the target transaction data, the following basic features may be selected: transaction frequency, the number of transactions of a customer in a certain time; transaction amounts, such as average transaction amount, maximum transaction amount, minimum transaction amount, etc.; transaction time, such as the time period during which the transaction occurs (night, working time, etc.); transaction types such as transfer, presentation, payment, etc.
Based on the basic features, the following advanced features may be further constructed: transaction volatility, calculating the volatility of transaction amount by using standard deviation or other statistical methods; transaction habits such as whether customers often transact at night, whether customers often transact on weekends, etc.
Abnormal transaction: based on the historical transaction data, transactions that do not conform to conventional transaction patterns are identified.
Step S302, performing feature extraction on the target text data by using a second feature extraction model to obtain advanced features of the text data, wherein the second feature extraction model at least comprises one of the following steps: a word bag model and a word embedding model;
The method comprises the following steps of:
Step S3021, converting the target text data into vector data by using a bag-of-word model, wherein each element in the vector data represents the frequency of occurrence of the corresponding word in the target text data;
Step S3022, determining, according to the vector data, a characteristic parameter of each of the words, where the characteristic parameter includes at least a frequency of the word and a rarity of the word, and the frequency of the word indicates a frequency of occurrence of the word in the target text data;
Step S3023, performing feature extraction on the target text data according to the feature parameters of the words, to obtain the advanced features of the text data.
Specifically, GPT is used for vectorization and feature extraction of text data. Thus, the risk assessment process can be trained by artificial intelligence and more data sources are used to solve the defect that manual work cannot be done, so that a more comprehensive customer risk profile is provided. Traditional customer risk assessment methods lack flexibility and cannot accommodate different customer groups and business requirements.
In addition, the feature extraction of the target text data can also be performed by adopting the following method: a Bag of Words model (BOW) converts text into a vector in which each element represents the frequency of occurrence of a word in the text; TF-IDF (Term Frequency-Inverse Document Frequency), which takes into account the Frequency of words in the text and the rarity of the whole dataset, assigns a weight to each word; word embedding (Word Embeddings) using pre-trained models such as Word2Vec, gloVe, etc., to transform words into fixed-size vectors that can capture semantic information of the words; dependency syntactic analysis, namely identifying the relation between words in sentences, such as subjects, objects and the like; named Entity Recognition (NER) identifies specific entities in text, such as name, place, date, etc.; part-of-speech tagging (Part-of-SPEECH TAGGING, POS) that assigns a Part-of-speech tag, such as nouns, verbs, adjectives, etc., to each word in the text; semantic role labels (Semantic Role Labeling, SRL), identify verbs and their associated arguments and modifiers in sentences, and assign them specific semantic roles; topic model (Topic Modeling), which uses algorithms such as LDA (LATENT DIRICHLET Allocation) to extract topics from a large number of documents.
The feature extraction of the target text data can be achieved through the following aspects:
1. keyword extraction, extracting keywords from transaction descriptions, such as "foreign", "large", "cash", etc., which may be related to certain risk behaviors. The characteristics are as follows: a binary feature is created for each keyword, indicating whether the keyword appears in the transaction description.
2. The transaction description length, the length of the transaction description is calculated, and too long or too short descriptions may be associated with certain risk behaviors. The characteristics are as follows: word number or character number of transaction description.
3. Named Entity Recognition (NER) extracts specific entities, such as company names, places, etc., from the transaction description. The characteristics are as follows: a binary feature is created for each common entity indicating whether the entity is present in the transaction description.
4. Emotion analysis, analyzing emotion, such as positive, negative or neutral, of a transaction description. The characteristics are as follows: emotional score of the transaction description.
5. Complexity of the transaction description, the complexity of the transaction description is calculated using the lexical richness or other method, wherein the features: lexical richness score of the transaction description.
6. Frequent pattern mining: frequently occurring phrases or patterns are extracted from the transaction description using pattern mining techniques, such as Apriori algorithms. The characteristics are as follows: a binary feature is created for each frequent pattern, indicating whether the pattern appears in the transaction description.
7. Topic modeling: topics are extracted from the transaction description using algorithms such as LDA. The characteristics are as follows: the main topic or topic distribution of the transaction description.
8. Word vector: the transaction description is converted to a vector using a pre-trained Word vector model, such as Word2 Vec. The characteristics are as follows: word vector average of transaction descriptions or vectors obtained by other aggregation methods.
Step S303, determining all the transaction data high-level features and all the text data high-level features as data features, and determining the target data features from the data features.
In particular, this may improve the accuracy of the risk assessment.
Wherein determining the target data feature from the data features comprises: calculating the correlation between each data feature and the risk degree of the client to obtain a plurality of feature correlations, and determining the data feature corresponding to the feature correlations which are larger than or equal to a preset correlation as the target data feature; or recursively deleting the data features with importance degrees smaller than or equal to a preset importance degree by adopting a recursion feature elimination model so as to determine the target data features from the data features.
In particular, this allows to obtain the features most relevant to the degree of risk of the customer. The artificial intelligence model can learn and identify patterns and relationships from historical data that are not visible to an analyst, thereby enabling more accurate and consistent risk assessment. Conventional customer risk assessment methods may have assessment bias, resulting in different customers being incorrectly assessed as the same risk level.
In addition, the target data feature may be determined from the data features, and the following method may be adopted:
1. Correlation analysis: and calculating the correlation of each feature and the customer risk degree.
2. Recursive Feature Elimination (RFE): the model is used to evaluate the importance of features and recursively remove the least important features.
3. Model-based feature selection: such as using models such as decision trees or random forests, and the like, are selected according to the importance of the feature.
In addition, for certain features, transformations may be required to be better used by the model, with specific transformation methods including: normalization/normalization converts the numerical features to the same scale; encoding, converting the classification features into a numerical form, such as using one-hot encoding.
Considering interactions between different features, such as the product of transaction frequency and transaction amount, may have a stronger correlation with the customer risk level.
Over time, customers' transaction habits and risk patterns may change. Thus, the importance of the feature is periodically reevaluated and adjusted.
Step S203, repeating self-learning training on the target data features by adopting an intelligent large language model to obtain a plurality of initial training models;
Specifically, the intelligent large language model is generally a GPT model and an NLP technology, and the artificial intelligence model enables the data collection, analysis and decision making process to be automated, reduces human errors and improves the evaluation speed and expandability. However, the conventional evaluation process of the client risk evaluation method is usually manual, so that a long time is required, which limits the evaluation range and is not beneficial to quick decision. The method of training a customer risk assessment model based on GPT and NLP artificial intelligence models may provide a more efficient, accurate, and scalable method of customer risk assessment than traditional methods.
The method comprises the following steps of:
Step S2031, obtaining a client risk score report, where the client risk score report is obtained by training according to multiple sets of training data, and each set of training data in the multiple sets of training data includes obtained in a historical time period: historical training models and customer risk scores obtained by adopting the corresponding historical training models;
Step S2032, determining a GPT model as the intelligent large language model, and initializing the GPT model to obtain an initialized GPT model;
step S2033, repeating self-learning training on the target data features according to the client risk score report by using the initialized GPT model, so as to obtain a plurality of initial training models.
In particular, the artificial intelligence model can learn and identify patterns and relationships from historical data that are not visible to an analyst, thereby enabling more accurate and consistent risk assessment. Conventional customer risk assessment methods may have assessment bias, resulting in different customers being incorrectly assessed as the same risk level.
In addition, machine learning models are trained using already constructed features and target variables (risk scores for customers). The model is based on various algorithms, such as decision trees, random forests, gradient lifting and other algorithms, and GPT and NLP artificial intelligence models. The method also comprises the following steps:
1. dividing data: dividing data into a training set, a verification set and a test set;
2. Selecting a model: selecting GPT, other NLP models or a traditional machine learning model;
3. Model initialization: setting initial values for model parameters;
4. Training a model: model training is performed using training data and an optimization algorithm.
Step S204, each initial training model is evaluated to obtain a corresponding target evaluation score, and the initial training model with the highest target evaluation score is used as a risk evaluation model.
In particular, this may result in the most accurate risk assessment model.
The method comprises the following steps of:
step S2041, determining an evaluation index, where the evaluation index includes at least one of: accuracy, recall rate, F1 score, the F1 score characterizes the reconciliation average of accuracy and recall rate;
Step S2042, evaluating each initial training model according to the evaluation index to obtain a first evaluation score;
step S2043, evaluating the generalization capability of each initial training model by adopting a cross-validation method to obtain a second evaluation score;
step S2044, according to the first evaluation score and the second evaluation score, determining the target evaluation score corresponding to each initial training model.
In particular, this may result in the most accurate risk assessment model.
Wherein after determining the initial training model with the highest target evaluation score as the risk evaluation model, the method further comprises: and under the condition that new transaction data and/or new text data are acquired, adjusting the risk assessment model according to the new transaction data and/or the new text data to obtain an adjusted risk assessment model.
Specifically, the risk assessment model can be adjusted in real time, so that the accuracy of the risk assessment model can be guaranteed in a long period of time. If necessary, the model is finely tuned, and a fine tuning answer is given to the negative result which does not accord with the expectation, so that the artificial intelligence automatically learns and evolves.
The specific adjustment steps comprise: optimizing the super-parameters, and searching the optimal super-parameters by using methods such as grid search, random search and the like; early stop, to prevent overfitting, stopping training when performance on the validation set is no longer improving; model integration, the performance of the model is improved using an integration method such as Bagging, boosting.
And repeating the data training circularly, and gradually accumulating the effective models for outputting the models.
Model deployment: the trained model is deployed to a production environment where new customers can be scored and their risk levels predicted.
The whole process in the steps realizes training of bank transaction data and text information by using GPT and NLP artificial intelligent models, so that the risk degree of a customer can be more accurately identified, and the accuracy of risk assessment is improved.
In some embodiments, models are trained with existing problem libraries, knowledge bases, and transaction data to identify customer risk, which is a task of multi-source data fusion. The following is an example flow illustrating how training is performed in conjunction with these three data:
1. The data is pre-processed and the data is pre-processed,
1.1 For a question bank (which may include common questions of customers at banks or financial institutions, such as "how to apply for loans; the problem is then converted into a vector using NLP techniques, such as word embedding.
1.2 For a knowledge base (possibly including policies, regulations or other related knowledge of a bank), first extracting key information in a document, such as interest rate, credit, risk, loan, etc.; the document is then converted into a vector using NLP techniques, such as TF-IDF or word embedding.
1.3, Aiming at transaction data, firstly, carrying out text preprocessing on transaction description, such as word segmentation, stop word removal and the like; extracting basic characteristics of the transaction, such as transaction amount, transaction time and the like; finally, the transaction description is converted into a vector using NLP techniques, such as word embedding.
2. Feature engineering, firstly combining vector representations of a question library, a knowledge base, transaction data and the like to construct mixed features; features most relevant to customer risk are then selected using feature selection techniques such as correlation analysis or model-based feature selection.
3. Model training, first training a classification model, such as a logistic regression, random forest, or deep learning model, using tag data (known customer risk tags); the performance of the model is then evaluated using cross-validation.
4. Model evaluation and micro, firstly, evaluating the performance of the model by using a verification set; then, according to the evaluation result, adjusting parameters or structures of the model; finally, repeating the steps until the performance of the model reaches a satisfactory level.
5. Risk identification: and predicting new customer data by using the trained model, and identifying high-risk customers.
6. Continuously updating: over time, the customer's behavior and risk patterns may change. New data is collected periodically and the model retrained to ensure its accuracy and timeliness.
In summary, training in combination with the problem library, knowledge base and transaction data is a complex task requiring multiple iterations and optimizations. However, by this method, the behavior and risk of the customer can be more fully understood, so that more accurate predictions can be made.
The following are some simulated simplified test data that illustrate how training is performed in conjunction with the problem library, knowledge library, and transaction data:
The contents of the problem library are shown in Table 1:
Table 1, problem library table
| Customer ID | Problem content |
| 001 | How do loans apply? |
| 002 | How did my credit card lost? |
| 003 | How do my account balance be queried? |
| 004 | Do i pay out the loan in advance? |
The knowledge base is shown in table 2:
Table 2, knowledge base table
/>
The contents of the transaction data are shown in table 3:
Table 3, transaction data sheet
For training customer risk tags as shown in Table 4
Table 4, customer risk tag table
| Customer ID | Risk label |
| 001 | Low and low |
| 002 | In (a) |
| 003 | Low and low |
| 004 | High height |
Based on the above data, we can perform the following analysis:
customer 001: the customer inquires about how to apply for the loan and makes a transaction for the loan application fee, which is related to the document A in the knowledge base. Thus, the customer may be a new loan customer with a lower risk.
Customer 002: the customer's credit card is lost and pays a annual credit card fee, which is associated with document B in the knowledge base. Thus, the customer may be at some risk and require further attention.
Customer 003: the customer inquires about how to query the account balance and proceeds with the ATM withdrawal, in relation to document C in the knowledge base. The customer's behavior is more regular and the risk is lower.
Client 004: the customer inquires whether the loan can be repayment in advance and proceeds with repayment in advance of the large loan, which is related to the document D in the knowledge base. This may indicate that the customer's financial condition has changed and that the risk is high.
From the above analysis we can see how to combine the problem library, knowledge base and transaction data to assess the risk of the customer.
The above-described embodiments of methods of training a customer risk assessment model based on GPT and NLP artificial intelligence models may provide a more efficient, accurate, and scalable method of customer risk assessment.
According to the method for determining the risk assessment model, target transaction data and target text data are firstly obtained, wherein the target transaction data are structured data related to target transaction, and the target text data are unstructured data related to target transaction; then, respectively carrying out feature extraction processing on the target transaction data and the target text data to obtain target data features; then, repeating self-learning training on the target data characteristics by adopting an intelligent large language model to obtain a plurality of initial training models; and finally, evaluating each initial training model to obtain a corresponding target evaluation score, and taking the initial training model with the highest target evaluation score as a risk evaluation model. According to the method, the risk assessment is carried out on the clients through the artificial intelligence training model, more data sources are used for providing more comprehensive client risk profiles, the assessment accuracy is improved, and the problems that the existing client risk assessment method is low in automation and intelligence, and the client risk assessment efficiency is low and inaccurate are solved.
In order to enable those skilled in the art to more clearly understand the technical solution of the present application, the implementation process of the method for determining the risk assessment model of the present application will be described in detail below with reference to specific embodiments.
The embodiment relates to a specific method for determining a risk assessment model, as shown in fig. 3, including: transaction data (including structured data and unstructured data, such as account balance, transaction frequency, transaction amount, transaction type, loan data, credit card data, financial data, etc.) is first collected; and then, carrying out data training, namely firstly cleaning the data, carrying out feature extraction on the data, selecting and constructing corresponding feature extraction models according to risk related features, data dimension transformation, capturing modes and relations of the data, specifically carrying out feature extraction on transaction data and text information data respectively from different dimensions (such as aspects of transaction volatility, transaction habit and the like), carrying out feature extraction by adopting different models, selecting a model with the most accurate feature for extracting the data from certain data as a model for extracting the data later, carrying out feature extraction on the transaction data and text information respectively in different modes, carrying out extraction on text information by adopting models to convert texts into vectors, and determining specific features according to transaction description lengths, emotion analysis and the like. And inputting the characteristics and the defined risk scores into a plurality of GPT machine models, repeating self-learning training by combining a decision tree, a random forest, gradient lifting and other multi-medium algorithms, putting the trained models into a model library, evaluating a plurality of data trained by the models and the models, taking the machine model with the best indexes such as accuracy as a final model, fine-tuning the model when necessary, and carrying out transaction risk evaluation on the clients by adopting the final model. Under the condition that new data are input into the model subsequently, the model in the model library and the newly input data are trained together, so that self-learning training of the model can be realized, the obtained risk assessment model is deployed and output, risk assessment is carried out on the client based on the model and the client transaction data, and finally a client risk report is output, so that a related system uses the report to carry out subsequent operation.
It should be noted that the steps illustrated in the flowcharts of the figures may be performed in a computer system such as a set of computer executable instructions, and that although a logical order is illustrated in the flowcharts, in some cases the steps illustrated or described may be performed in an order other than that illustrated herein.
The embodiment of the application also provides a determining device for the risk assessment model, and the determining device for the risk assessment model can be used for executing the determining method for the risk assessment model. The device is used for realizing the above embodiments and preferred embodiments, and is not described in detail. 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.
The following describes a determination device of a risk assessment model provided by an embodiment of the present application.
Fig. 4 is a schematic diagram of a determining apparatus of a risk assessment model according to an embodiment of the present application. As shown in fig. 4, the apparatus includes an acquisition unit 10, an extraction unit 20, a training unit 30, and a determination unit 40, where the acquisition unit 10 is configured to acquire target transaction data and target text data, the target transaction data being structured data related to a target transaction, and the target text data being unstructured data related to the target transaction; the extracting unit 20 is configured to perform feature extraction processing on the target transaction data and the target text data, so as to obtain target data features; the training unit 30 is configured to perform repeated self-learning training on the target data features by using an intelligent large language model to obtain a plurality of initial training models; the determining unit 40 is configured to evaluate each of the initial training models to obtain a corresponding target evaluation score, and take the initial training model with the highest target evaluation score as a risk evaluation model.
The determining device of the risk assessment model comprises an acquisition unit, an extraction unit, a training unit and a determining unit, wherein the acquisition unit is used for acquiring target transaction data and target text data, the target transaction data is structured data related to target transaction, and the target text data is unstructured data related to target transaction; the extraction unit is used for respectively carrying out feature extraction processing on the target transaction data and the target text data to obtain target data features; the training unit is used for carrying out repeated self-learning training on the target data characteristics by adopting the intelligent large language model to obtain a plurality of initial training models; the determining unit is used for evaluating each initial training model to obtain a corresponding target evaluation score, and taking the initial training model with the highest target evaluation score as a risk evaluation model. According to the method, the risk assessment is carried out on the clients through the artificial intelligence training model, more data sources are used for providing more comprehensive client risk profiles, the assessment accuracy is improved, and the problems that the existing client risk assessment method is low in automation and intelligence, and the client risk assessment efficiency is low and inaccurate are solved.
In some optional examples, the acquiring unit includes a first acquiring module, a first processing module and a second processing module, where the first acquiring module is configured to acquire initial transaction data and initial text data, where the initial transaction data is the target transaction data before being processed by the first processing mode, and the initial text data is the target text data before being processed by the second processing mode; the first processing module is configured to process the initial transaction data by using the first processing manner to obtain the target transaction data, where the first processing manner includes at least one of: data cleaning, data duplication removal, missing value processing and abnormal value detection; the second processing module is configured to process the initial text data by using the second processing manner to obtain the target text data, where the second processing manner includes at least one of the following: keyword extraction, word segmentation processing, non-keyword removal and stem extraction. The artificial intelligence model can provide a more complete view of customer financial behavior and risk status using a wide range of data sources, including transactional data, demographic data, and other unstructured data sources.
In some optional examples, the extracting unit includes a first extracting module, a second extracting module, and a first determining module, where the first extracting module is configured to perform feature extraction on the target transaction data by using a first feature extracting model to obtain basic features of the transaction data, and determine advanced features of the transaction data according to the basic features of the transaction data, where the basic features of the transaction data include at least one of the following: transaction frequency, transaction amount, transaction time, and transaction type, the transaction data high-level characteristics including at least one of: transaction volatility, transaction time, and abnormal transaction data; the second extraction module is configured to perform feature extraction on the target text data by using a second feature extraction model to obtain advanced features of the text data, where the second feature extraction model at least includes one of the following: a word bag model and a word embedding model; the first determining module is used for determining all the transaction data advanced features and all the text data advanced features as data features, and determining the target data features from the data features. This may improve the accuracy of risk assessment.
In some optional examples, the first determining module includes a first determining sub-module and a second determining sub-module, where the first determining sub-module is configured to calculate a correlation between each of the data features and a risk level of the client, obtain a plurality of feature correlations, and determine, as the target data feature, the data feature corresponding to the feature correlation that is greater than or equal to a preset correlation; the second determining submodule is used for recursively adopting a recursion feature elimination model and deleting the data features with importance degrees smaller than or equal to a preset importance degree so as to determine the target data features from the data features. This allows to obtain the features most relevant to the degree of risk of the customer.
In some optional examples, the second extraction module includes a conversion sub-module, a third determination sub-module, and an extraction sub-module, where the conversion sub-module is configured to convert the target text data into vector data by using a bag-of-word model, where each element in the vector data represents a frequency of occurrence of a corresponding word in the target text data; the third determining submodule is used for determining characteristic parameters of each word according to the vector data, wherein the characteristic parameters at least comprise the frequency of the word and the rarity of the word, and the frequency of the word represents the frequency of the word in the target text data; and the extraction sub-module is used for extracting the characteristics of the target text data according to the characteristic parameters of the words to obtain the advanced characteristics of the text data. Vectorization and feature extraction of text data is performed using GPT. Thus, the risk assessment process can be trained by artificial intelligence and more data sources are used to solve the defect that manual work cannot be done, so that a more comprehensive customer risk profile is provided.
In this embodiment, the training unit includes a second obtaining module, an initializing module, and a training module, where the second obtaining module is configured to obtain a customer risk score report, where the customer risk score report is obtained by training according to multiple sets of training data, and each set of training data in the multiple sets of training data includes obtained in a historical time period: historical training models and customer risk scores obtained by adopting the corresponding historical training models; the initialization module is used for determining the GPT model as the intelligent large language model, initializing the GPT model and obtaining an initialized GPT model; and the training module is used for carrying out repeated self-learning training on the target data characteristics according to the client risk score report by adopting the initialized GPT model to obtain a plurality of initial training models. The artificial intelligence model can learn and identify patterns and relationships from historical data that are not visible to an analyst, thereby enabling more accurate and consistent risk assessment.
Alternatively, the determining unit includes a second determining module, a first evaluating module, a second evaluating module, and a third determining module, where the second determining module is configured to determine an evaluation index, and the evaluation index includes at least one of the following: accuracy, recall rate, F1 score, the F1 score characterizes the reconciliation average of accuracy and recall rate; the first evaluation module is used for evaluating each initial training model according to the evaluation indexes to obtain a first evaluation score; the second evaluation module is used for evaluating the generalization capability of each initial training model by adopting a cross verification method to obtain a second evaluation score; and the third determining module is used for determining the target evaluation score corresponding to each initial training model according to the first evaluation score and the second evaluation score. This allows the most accurate risk assessment model.
As an alternative, the apparatus further includes an adjusting module, where the adjusting module is configured to, after determining the initial training model with the highest target evaluation score as a risk evaluation model, adjust the risk evaluation model according to the new transaction data and/or the new text data when new transaction data and/or new text data are acquired, and obtain an adjusted risk evaluation model. Therefore, the risk assessment model can be adjusted in real time, so that the accuracy of the risk assessment model can be ensured in a long time.
The determining device of the risk assessment model comprises a processor and a memory, wherein the acquiring unit and the like are stored in the memory as program units, and the processor executes the program units stored in the memory to realize corresponding functions. The modules are all located in the same processor; or the above modules may be located in different processors in any combination.
The processor includes a kernel, and the kernel fetches the corresponding program unit from the memory. The kernel can be provided with one or more than one kernel, and the problems of low customer risk assessment efficiency and inaccuracy caused by low automation and intellectualization of the existing customer risk assessment method are solved by adjusting kernel parameters.
The memory may include volatile memory, random Access Memory (RAM), and/or nonvolatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM), among other forms in computer readable media, the memory including at least one memory chip.
The embodiment of the invention provides a computer readable storage medium, which comprises a stored program, wherein the program is controlled to run so as to control equipment where the computer readable storage medium is located to execute the method for determining the risk assessment model.
Specifically, the method for determining the risk assessment model includes:
Step S201, obtaining target transaction data and target text data, wherein the target transaction data is structured data related to target transaction, and the target text data is unstructured data related to the target transaction;
Specifically, the target transaction information generally includes transaction information of the customer, credit card information, financial information, and the like. The target text data typically includes a question library, a knowledge library, a risk library, a report library, and the like.
Step S202, respectively carrying out feature extraction processing on the target transaction data and the target text data to obtain target data features;
in particular, the selection and construction of features from the transaction data and the text data that are relevant to the risk level of the customer is a key step, as these features will directly affect the performance of the risk assessment model.
Step S203, repeating self-learning training on the target data features by adopting an intelligent large language model to obtain a plurality of initial training models;
Specifically, the intelligent large language model is generally a GPT model and an NLP technology, and the artificial intelligence model enables the data collection, analysis and decision making process to be automated, reduces human errors and improves the evaluation speed and expandability. However, the conventional evaluation process of the client risk evaluation method is usually manual, so that a long time is required, which limits the evaluation range and is not beneficial to quick decision. The method of training a customer risk assessment model based on GPT and NLP artificial intelligence models may provide a more efficient, accurate, and scalable method of customer risk assessment than traditional methods.
Step S204, each initial training model is evaluated to obtain a corresponding target evaluation score, and the initial training model with the highest target evaluation score is used as a risk evaluation model.
In particular, the most accurate risk assessment model can be obtained.
The embodiment of the invention provides a processor, which is used for running a program, wherein the program runs to execute the method for determining the risk assessment model.
Specifically, the method for determining the risk assessment model includes:
Step S201, obtaining target transaction data and target text data, wherein the target transaction data is structured data related to target transaction, and the target text data is unstructured data related to the target transaction;
Specifically, the target transaction information generally includes transaction information of the customer, credit card information, financial information, and the like. The target text data typically includes a question library, a knowledge library, a risk library, a report library, and the like.
Step S202, respectively carrying out feature extraction processing on the target transaction data and the target text data to obtain target data features;
in particular, the selection and construction of features from the transaction data and the text data that are relevant to the risk level of the customer is a key step, as these features will directly affect the performance of the risk assessment model.
Step S203, repeating self-learning training on the target data features by adopting an intelligent large language model to obtain a plurality of initial training models;
Specifically, the intelligent large language model is generally a GPT model and an NLP technology, and the artificial intelligence model enables the data collection, analysis and decision making process to be automated, reduces human errors and improves the evaluation speed and expandability. However, the conventional evaluation process of the client risk evaluation method is usually manual, so that a long time is required, which limits the evaluation range and is not beneficial to quick decision. The method of training a customer risk assessment model based on GPT and NLP artificial intelligence models may provide a more efficient, accurate, and scalable method of customer risk assessment than traditional methods.
Step S204, each initial training model is evaluated to obtain a corresponding target evaluation score, and the initial training model with the highest target evaluation score is used as a risk evaluation model.
In particular, the most accurate risk assessment model can be obtained.
The embodiment of the invention provides equipment, which comprises a processor, a memory and a program stored in the memory and capable of running on the processor, wherein the processor realizes at least the following steps when executing the program:
Step S201, obtaining target transaction data and target text data, wherein the target transaction data is structured data related to target transaction, and the target text data is unstructured data related to the target transaction;
Step S202, respectively carrying out feature extraction processing on the target transaction data and the target text data to obtain target data features;
Step S203, repeating self-learning training on the target data features by adopting an intelligent large language model to obtain a plurality of initial training models;
Step S204, each initial training model is evaluated to obtain a corresponding target evaluation score, and the initial training model with the highest target evaluation score is used as a risk evaluation model.
The device herein may be a server, PC, PAD, cell phone, etc.
The application also provides a computer program product adapted to perform, when executed on a data processing device, a program initialized with at least the following method steps:
Step S201, obtaining target transaction data and target text data, wherein the target transaction data is structured data related to target transaction, and the target text data is unstructured data related to the target transaction;
Step S202, respectively carrying out feature extraction processing on the target transaction data and the target text data to obtain target data features;
Step S203, repeating self-learning training on the target data features by adopting an intelligent large language model to obtain a plurality of initial training models;
Step S204, each initial training model is evaluated to obtain a corresponding target evaluation score, and the initial training model with the highest target evaluation score is used as a risk evaluation model.
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.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In one typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include volatile memory in a computer-readable medium, random Access Memory (RAM) and/or nonvolatile memory, etc., such as Read Only Memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.
Computer readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by a computing device. Computer-readable media, as defined herein, does not include transitory computer-readable media (transmission media), such as modulated data signals and carrier waves.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises an element.
From the above description, it can be seen that the above embodiments of the present application achieve the following technical effects:
1) Firstly, acquiring target transaction data and target text data, wherein the target transaction data is structured data related to target transaction, and the target text data is unstructured data related to target transaction; then, respectively carrying out feature extraction processing on the target transaction data and the target text data to obtain target data features; then, repeating self-learning training on the target data characteristics by adopting an intelligent large language model to obtain a plurality of initial training models; and finally, evaluating each initial training model to obtain a corresponding target evaluation score, and taking the initial training model with the highest target evaluation score as a risk evaluation model. According to the method, the risk assessment is carried out on the clients through the artificial intelligence training model, more data sources are used for providing more comprehensive client risk profiles, the assessment accuracy is improved, and the problems that the existing client risk assessment method is low in automation and intelligence, and the client risk assessment efficiency is low and inaccurate are solved.
2) The determining device of the risk assessment model comprises an acquisition unit, an extraction unit, a training unit and a determining unit, wherein the acquisition unit is used for acquiring target transaction data and target text data, the target transaction data is structured data related to target transaction, and the target text data is unstructured data related to target transaction; the extraction unit is used for respectively carrying out feature extraction processing on the target transaction data and the target text data to obtain target data features; the training unit is used for carrying out repeated self-learning training on the target data characteristics by adopting the intelligent large language model to obtain a plurality of initial training models; the determining unit is used for evaluating each initial training model to obtain a corresponding target evaluation score, and taking the initial training model with the highest target evaluation score as a risk evaluation model. According to the method, the risk assessment is carried out on the clients through the artificial intelligence training model, more data sources are used for providing more comprehensive client risk profiles, the assessment accuracy is improved, and the problems that the existing client risk assessment method is low in automation and intelligence, and the client risk assessment efficiency is low and inaccurate are solved.
The above description is only of the preferred embodiments of the present application and is not intended to limit the present application, but various modifications and variations can be made to the present application by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.
Claims (10)
1. A method for determining a risk assessment model, comprising:
acquiring target transaction data and target text data, wherein the target transaction data is structured data related to target transaction, and the target text data is unstructured data related to the target transaction;
respectively carrying out feature extraction processing on the target transaction data and the target text data to obtain target data features;
Repeating self-learning training on the target data characteristics by adopting an intelligent large language model to obtain a plurality of initial training models;
and evaluating each initial training model to obtain a corresponding target evaluation score, and taking the initial training model with the highest target evaluation score as a risk evaluation model.
2. The determination method according to claim 1, wherein acquiring the target transaction data and the target text data includes:
Acquiring initial transaction data and initial text data, wherein the initial transaction data is the target transaction data before being processed by adopting a first processing mode, and the initial text data is the target text data before being processed by adopting a second processing mode;
Processing the initial transaction data by adopting the first processing mode to obtain the target transaction data, wherein the first processing mode comprises at least one of the following steps: data cleaning, data duplication removal, missing value processing and abnormal value detection;
processing the initial text data by adopting the second processing mode to obtain the target text data, wherein the second processing mode comprises at least one of the following steps: keyword extraction, word segmentation processing, non-keyword removal and stem extraction.
3. The method according to claim 1, wherein the feature extraction processing is performed on the target transaction data and the target text data, respectively, to obtain target data features, including:
Performing feature extraction on the target transaction data by adopting a first feature extraction model to obtain basic features of the transaction data, and determining advanced features of the transaction data according to the basic features of the transaction data, wherein the basic features of the transaction data at least comprise one of the following: transaction frequency, transaction amount, transaction time, and transaction type, the transaction data high-level characteristics including at least one of: transaction volatility, transaction time, and abnormal transaction data;
and carrying out feature extraction on the target text data by adopting a second feature extraction model to obtain advanced features of the text data, wherein the second feature extraction model at least comprises one of the following steps: a word bag model and a word embedding model;
All the transaction data high-level features and all the text data high-level features are determined to be data features, and the target data features are determined from the data features.
4. A method of determining according to claim 3, wherein determining the target data feature from the data features comprises:
Calculating the correlation between each data feature and the risk degree of the client to obtain a plurality of feature correlations, and determining the data feature corresponding to the feature correlations which are larger than or equal to a preset correlation as the target data feature;
Or alternatively
And recursively deleting the data features with importance degrees smaller than or equal to a preset importance degree by adopting a recursion feature elimination model so as to determine the target data features from the data features.
5. A method of determining as claimed in claim 3, wherein the feature extraction of the target text data using the second feature extraction model to obtain advanced features of the text data comprises:
Converting the target text data into vector data by adopting a word bag model, wherein each element in the vector data respectively represents the frequency of the corresponding word in the target text data;
according to the vector data, determining characteristic parameters of each word, wherein the characteristic parameters at least comprise the frequency of the word and the rarity of the word, and the frequency of the word represents the frequency of the word in the target text data;
and extracting the characteristics of the target text data according to the characteristic parameters of the words to obtain the advanced characteristics of the text data.
6. The method of determining according to claim 1, wherein repeating the self-learning training of the target data feature using the intelligent large language model to obtain a plurality of initial training models, comprises:
Obtaining a client risk score report, wherein the client risk score report is obtained through training according to multiple sets of training data, and each set of training data in the multiple sets of training data comprises data obtained in a historical time period: a historical training model and a client risk score obtained by adopting the corresponding historical training model;
Determining a GPT model as the intelligent large language model, and initializing the GPT model to obtain an initialized GPT model;
And repeating self-learning training on the target data features according to the client risk score report by adopting the initialized GPT model to obtain a plurality of initial training models.
7. The method of determining of claim 1, wherein evaluating each of the initial training models to obtain a corresponding target evaluation score comprises:
Determining an evaluation index, the evaluation index comprising at least one of: accuracy, recall and F1 score, wherein the F1 score represents a harmonic average value of the accuracy and the recall;
evaluating each initial training model according to the evaluation index to obtain a first evaluation score;
Evaluating the generalization capability of each initial training model by adopting a cross-validation method to obtain a second evaluation score;
And determining the target evaluation score corresponding to each initial training model according to the first evaluation score and the second evaluation score.
8. The determination method according to claim 1, wherein after determining the initial training model with the highest target evaluation score as a risk evaluation model, the method further comprises:
And under the condition that new transaction data and/or new text data are acquired, adjusting the risk assessment model according to the new transaction data and/or the new text data to obtain an adjusted risk assessment model.
9. A risk assessment model determining apparatus, comprising:
the system comprises an acquisition unit, a storage unit and a processing unit, wherein the acquisition unit is used for acquiring target transaction data and target text data, the target transaction data is structured data related to target transaction, and the target text data is unstructured data related to the target transaction;
the extraction unit is used for respectively carrying out feature extraction processing on the target transaction data and the target text data to obtain target data features;
the training unit is used for carrying out repeated self-learning training on the target data characteristics by adopting an intelligent large language model to obtain a plurality of initial training models;
And the determining unit is used for evaluating each initial training model to obtain a corresponding target evaluation score, and taking the initial training model with the highest target evaluation score as a risk evaluation model.
10. A computer-readable storage medium, characterized in that the computer-readable storage medium comprises a stored program, wherein the program, when run, controls a device in which the computer-readable storage medium is located to perform the method of determining a risk assessment model according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410131104.XA CN118052632A (en) | 2024-01-30 | 2024-01-30 | Method and device for determining risk assessment model and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410131104.XA CN118052632A (en) | 2024-01-30 | 2024-01-30 | Method and device for determining risk assessment model and computer readable storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118052632A true CN118052632A (en) | 2024-05-17 |
Family
ID=91044034
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410131104.XA Pending CN118052632A (en) | 2024-01-30 | 2024-01-30 | Method and device for determining risk assessment model and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118052632A (en) |
-
2024
- 2024-01-30 CN CN202410131104.XA patent/CN118052632A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110377911B (en) | Method and device for identifying intention under dialog framework | |
| US20210272040A1 (en) | Systems and methods for language and speech processing with artificial intelligence | |
| JP2021504789A (en) | ESG-based corporate evaluation execution device and its operation method | |
| WO2018184518A1 (en) | Microblog data processing method and device, computer device and storage medium | |
| Akerkar et al. | Intelligent techniques for data science | |
| CN117271767B (en) | Operation and maintenance knowledge base establishing method based on multiple intelligent agents | |
| Fu et al. | A sentiment-aware trading volume prediction model for P2P market using LSTM | |
| Garbero et al. | Harnessing the power of machine learning analytics to understand food systems dynamics across development projects | |
| Jebaseel et al. | M-learning sentiment analysis with data mining techniques | |
| CN115063035A (en) | Customer evaluation method, system, equipment and storage medium based on neural network | |
| CN113392920B (en) | Method, apparatus, device, medium, and program product for generating cheating prediction model | |
| Azzam et al. | Similarity Software Requirement Specification (SRS) Elicitation Based on the Requirement Statement Using Text Mining on the MNC Play Inventory Management Application | |
| CN117668205A (en) | Smart logistics customer service processing method, system, equipment and storage medium | |
| CN116610592B (en) | Customizable software test evaluation method and system based on natural language processing technology | |
| CN115794898B (en) | Financial information recommendation method and device, electronic equipment and storage medium | |
| WO2022271431A1 (en) | System and method that rank businesses in environmental, social and governance (esg) | |
| CN118052632A (en) | Method and device for determining risk assessment model and computer readable storage medium | |
| Jishtu et al. | Prediction of the stock market based on machine learning and sentiment analysis | |
| Ali et al. | Identifying and Profiling User Interest over time using Social Data | |
| Al Ridhawi | Stock Market Prediction Through Sentiment Analysis of Social-Media and Financial Stock Data Using Machine Learning | |
| Lokeswari et al. | Comparative study of classification algorithms in sentiment analysis | |
| CN113064887B (en) | Data management method, device, equipment and storage medium | |
| CN117540004B (en) | Industrial domain intelligent question-answering method and system based on knowledge graph and user behavior | |
| Zhang et al. | A Semantic Search Framework for Similar Audit Issue Recommendation in Financial Industry | |
| Schlaubitz | Natural Language Processing in finance: analysis of sentiment and complexity of news and earnings reports of swiss SMEs and their relevance for stock returns |
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 |