A data prediction method and system for automobile installment loan overdue rate

Abstract

The present invention relates to the field of data prediction technology, specifically to a data prediction method and system for the overdue rate of auto installment loans, comprising: obtaining a target customer's credit report, converting the credit report into a credit report text based on a pre-trained text recognition model, pre-processing the credit report text to obtain a standard credit report text; and credit classification of the standard credit report text based on a pre-trained credit classification model to obtain a credit type label corresponding to the standard credit report text. The present invention predicts a customer's overdue rate by combining information from multiple different sources. This multi-dimensional information fusion is more comprehensive than a single feature and can more accurately reflect the customer's comprehensive risk. The combination of sentiment analysis and credit classification allows not only the customer's past financial behavior to be seen, but also the customer's current sentiment and attitude to be identified from their voice and text communication.

Description

Technical Field

[0001] The present invention relates to the technical field of data prediction, and in particular to a data prediction method and system for automobile installment loan overdue rate. Background Art

[0002] The delinquency rate of auto installment loans refers to the ratio between the amount of auto installment loans that have been overdue and not repaid on time within a certain period of time and the total amount of all auto installment loans. Specifically, the delinquency rate reflects the proportion of borrowers who fail to repay the loan according to the agreed time and amount. It is an important indicator for measuring loan risks and the credit management level of financial institutions.

[0003] Traditional methods usually rely on a single type of information, mainly based on the customer's credit report data (such as credit score, borrowing history, repayment record, etc.) to assess overdue risk. These methods usually only focus on the customer's historical financial behavior, but ignore other factors that may affect the customer's credit status, such as the customer's emotional attitude, social behavior, and interaction with financial institutions. Traditional methods, such as models based on logistic regression and decision trees, although they can predict risks to a certain extent, these methods usually cannot capture complex nonlinear relationships well, especially in the context of multi-dimensional data fusion, the performance of traditional methods will be limited; and traditional overdue rate prediction methods usually treat all customers uniformly based on the overall risk assessment, such as deciding whether to approve a loan or take collection measures based on a certain credit score threshold. This traditional method does not take into account the differences in each customer's specific financial situation, behavioral patterns, and emotional state. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a data prediction method and system for the overdue rate of automobile installment loans.

[0005] The technical solution adopted to solve the above technical problems is: a data prediction method for the overdue rate of automobile installment loans, comprising: Obtaining a credit report of a target customer, converting the credit report into a credit report text based on a pre-trained text recognition model, and pre-processing the credit report text to obtain a standard credit report text; Based on the pre-trained credit classification model, the standard credit report text is credit classified to obtain the credit type label corresponding to the standard credit report text, that is, to obtain the credit type label of the target customer; Acquire the target customer's interactive speech and the first interactive text, convert the interactive speech into a second interactive text based on a pre-trained speech recognition model, and merge the first interactive text and the second interactive text to obtain a communication text; Performing sentiment classification on the communication text based on a pre-trained sentiment classification model to obtain a sentiment type label corresponding to the communication text, that is, obtaining the sentiment type label of the target customer; The credit type label and the emotion type label of the target customer are label mapped based on a pre-trained label mapping network to obtain the overdue rate label corresponding to the target customer.

[0006] Preferably, the credit report text is preprocessed to obtain a standard credit report text, including: Performing a word segmentation operation on the credit report text by using a word segmentation tool to obtain a keyword sequence; Traversing the keyword sequence, selecting two adjacent keywords in the keyword sequence as a combination keyword, and matching the combination keyword with a preset tag table to obtain a tag corresponding to the combination keyword; Obtaining a label mapping text corresponding to the label, and adding the label mapping text to a position corresponding to the combined keyword in the credit report text; Repeat the above operation until each keyword in the keyword sequence is traversed to obtain the standard credit report text.

[0007] Preferably, the credit classification model includes a semantic representation module, a semantic association module and a semantic classification module. The semantic representation module is used to perform feature extraction on the standard credit report text based on a pre-trained BERT model to obtain a BERT vector corresponding to the standard credit report text. The semantic association module is used to perform a deep convolution operation on the BERT vector based on a deep pyramid convolutional neural network layer to obtain local features in the standard credit report text, and to capture the long-range dependencies of the standard credit report text based on the convolution layer and the pooling layer. The semantic classification module is used to classify the standard credit report text based on the local features and long-range dependencies in the standard credit report text to obtain a credit type label corresponding to the standard credit report text.

[0008] Preferably, the sentiment classification model includes a vector embedding layer, an encoder, a hybrid attention mechanism and a decoder, the vector embedding layer adopts a BER model, the encoder includes a bidirectional gated recurrent unit and a long short-term memory network, the hybrid attention mechanism is used to fuse the encoder output results, and the decoder is used to map the fused encoder output results to the sentiment type.

[0009] Preferably, sentiment classification of the communication text is performed based on a pre-trained sentiment classification model to obtain a sentiment type label corresponding to the communication text, including: Inputting the communication text into the BER model, and converting the communication text into a target word vector through the BER model; Inputting the target word vector into the bidirectional gated recurrent unit and the long short-term memory network respectively, and obtaining the inter-sentence semantic features of the target word vector through the bidirectional gated recurrent unit; The inter-sentence semantic features are converted into inter-sentence semantic feature vectors, the intra-sentence semantic features of the target word vector are obtained through the long short-term memory network, and the intra-sentence semantic features are converted into intra-sentence semantic feature vectors.

[0010] Preferably, sentiment classification of the communication text is performed based on a pre-trained sentiment classification model to obtain a sentiment type label corresponding to the communication text, and further includes: fusing the inter-sentence semantic feature vector and the intra-sentence semantic feature vector through the hybrid attention mechanism to obtain a first fused vector; fusing the inter-sentence semantic feature vector, the intra-sentence semantic feature vector, and the first fusion vector to obtain a second fusion vector; The second fusion vector is input into the decoder, and the second fusion vector is mapped to the emotion type by the decoder to obtain the emotion type label corresponding to the communication text.

[0011] Preferably, the first interactive text and the second interactive text are merged to obtain a communication text, including: Sentence the first interactive text and the second interactive text to obtain a first sentence sequence corresponding to the first interactive text and a second sentence sequence corresponding to the second interactive text; Encoding the first sentence sequence and the second sentence sequence to obtain a first sentence encoding sequence corresponding to the first sentence sequence and a second sentence encoding sequence corresponding to the second sentence sequence; Calculating sentence similarity between the first sentence and the second sentence based on the first sentence encoding sequence and the second sentence encoding sequence; If the sentence similarity is greater than a preset similarity threshold, merging the second sentence into the first sentence; The above operation is repeatedly performed until the first sentence coding sequence and the second sentence coding sequence are traversed to merge the first interactive text and the second interactive text to obtain a communication text.

[0012] Preferably, the label mapping network adopts BP neural network.

[0013] Preferably, label mapping is performed on the credit type label and the emotion type label of the target customer based on a pre-trained label mapping network to obtain the overdue rate label corresponding to the target customer, including: Inputting the credit type label and the emotion type label of the target customer into the label mapping network; Through the forward propagation process of the label mapping network, the overdue rate label corresponding to the target customer is generated, wherein the overdue rate label is the overdue probability.

[0014] The technical solution adopted to solve the above technical problem is: a data prediction system for automobile installment loan overdue rate, which is applicable to the data prediction method for automobile installment loan overdue rate, comprising: An information processing unit, the information processing unit is used to obtain a credit report of a target customer, convert the credit report into a credit report text based on a pre-trained text recognition model, and pre-process the credit report text to obtain a standard credit report text; A credit classification unit, the credit classification unit is used to perform credit classification on the standard credit report text based on a pre-trained credit classification model to obtain a credit type label corresponding to the standard credit report text, that is, to obtain the credit type label of the target customer; A text merging unit, the text merging unit is used to obtain the interactive speech and the first interactive text of the target customer, convert the interactive speech into a second interactive text based on a pre-trained speech recognition model, and merge the first interactive text and the second interactive text to obtain a communication text; A sentiment classification unit, wherein the sentiment classification unit is used to perform sentiment classification on the communication text based on a pre-trained sentiment classification model to obtain a sentiment type label corresponding to the communication text, that is, to obtain a sentiment type label of the target customer; A label mapping unit is used to perform label mapping on the credit type label and the emotion type label of the target customer based on a pre-trained label mapping network to obtain an overdue rate label corresponding to the target customer.

[0015] The beneficial effects of the present invention are as follows: (1) The present invention predicts the customer's delinquency rate by combining information from multiple different sources. This multi-dimensional information fusion is more comprehensive than a single feature and can more accurately reflect the customer's comprehensive risk. Among them, the credit report text provides the customer's credit history and financial behavior, and the interactive voice and text can reflect the customer's emotional attitude and interactive behavior. By combining these features, the customer's credit status and behavior pattern can be better captured; (2) The present invention can independently extract information at different levels through multiple pre-trained models. This deep learning method can better capture complex nonlinear relationships and improve prediction accuracy. In addition, through the combination of sentiment analysis and credit classification, it is possible to see not only the customer's past financial behavior but also the customer's past financial behavior. In addition, the present emotion and attitude of the customer can be identified from his / her voice and text communication. This emotional understanding is very helpful for identifying potential default risks. Moreover, by mapping the labels of combined keywords, the information in the credit report can be accurately matched with the features related to financial risk assessment. (3) The present invention maps the credit type label and the emotion type label to obtain the overdue rate label of the target customer. This process can integrate the financial status and emotional state of the customer and provide each customer with a personalized overdue rate prediction, which helps to accurately identify the customer's overdue probability. Financial institutions can adopt differentiated risk management strategies based on this prediction, such as adjusting loan conditions, increasing guarantee requirements or taking more stringent collection measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic flow chart of the steps of an overall method in an embodiment of the present invention; Figure 2 A schematic diagram of the system architecture of an overall system in an embodiment of the present invention.

[0017] Figure numerals: 1. Information processing unit; 2. Credit classification unit; 3. Text merging unit; 4. Sentiment classification unit; 5. Label mapping unit. DETAILED DESCRIPTION

[0018] Embodiment 1, as Figure 1 As shown, the present invention proposes a data prediction method for automobile installment loan overdue rate, comprising: S1. Obtain the credit report of the target customer, convert the credit report into a credit report text based on a pre-trained text recognition model, and pre-process the credit report text to obtain a standard credit report text; S2. Classify the standard credit report text based on the pre-trained credit classification model to obtain the credit type label corresponding to the standard credit report text, that is, obtain the credit type label of the target customer; S3, obtaining the target customer's interactive voice and the first interactive text, converting the interactive voice into a second interactive text based on a pre-trained speech recognition model, and merging the first interactive text and the second interactive text to obtain a communication text; S4. Perform sentiment classification on the communication text based on the pre-trained sentiment classification model to obtain the sentiment type label corresponding to the communication text, that is, obtain the sentiment type label of the target customer; S5. Based on the pre-trained label mapping network, label mapping is performed on the target customer's credit type label and sentiment type label to obtain the overdue rate label corresponding to the target customer.

[0019] In the present invention, a credit report is a document provided by a credit scoring agency or a financial institution, which contains information such as an individual's credit history, debt record, loan repayment status, etc., and it reflects the customer's credit status; optical character recognition (OCR) is a technology that converts text content in an image into editable text, and common application scenarios include scanning documents, recognizing printed and handwritten text, etc.; a credit type label is a classification label generated based on a customer's credit report, which usually indicates the customer's credit status, such as good, general or bad credit, etc.; interactive voice refers to the voice communication record between the customer and the financial institution, such as telephone consultation, customer service call, etc.; the customer's voice record reflects the customer's attitude and emotion, and may play an important role in predicting the delinquency rate; a speech recognition model (Speech Recognition) is a model that converts speech into text through machine learning technology, which can process various languages ​​and extract information from speech; the first interactive text refers to the written communication record between the customer and the institution, which may occur through e-mail, online chat, etc.; the emotion type label is a classification label for the emotion in the customer's communication text. Common sentiment type labels include "positive", "negative" and "neutral"; the label mapping network is a deep learning network that is usually used to map different types of labels (for example, credit type labels and sentiment type labels) to a specific prediction target (such as overdue rate labels).

[0020] Embodiment 2, a data prediction method for overdue rate of automobile installment loans proposed by the present invention, compared with embodiment 1, this embodiment further includes: preprocessing the credit report text to obtain a standard credit report text, including: A1. Use a word segmentation tool to segment the credit report text to obtain a keyword sequence; A2. Traverse the keyword sequence, select two adjacent keywords in the keyword sequence as a combination keyword, and match the combination keyword with a preset tag table to obtain a tag corresponding to the combination keyword; A3. Obtain the label mapping text corresponding to the label, and add the label mapping text to the position corresponding to the combined keyword in the credit report text; Repeat the above operation until each keyword in the keyword sequence is traversed to obtain the standard credit report text.

[0021] In this embodiment, the word segmentation tool is a basic tool in natural language processing (NLP), which is used to segment text into words or keywords. In Chinese processing, since there is no space to distinguish words, the word segmentation tool uses language models, dictionaries and algorithms (such as hidden Markov models, conditional random fields, etc.) to segment text into independent words or phrases; a combined keyword refers to a combination of two adjacent keywords extracted from a keyword sequence. For example, if the keyword sequence is ["credit", "card", "repayment", "record"], then possible combination keywords are ["credit card", "card repayment", "repayment record"]; the tag table is a preset list or mapping table containing specific tags and their corresponding keywords. Each tag represents a specific category or information point, and these categories or information points can be identified by keywords or combination of keywords. For example, the tag table may contain tags such as "credit score" and "loan history", and each tag corresponds to a set of keywords that can be used to identify the tag; tag mapping text refers to the descriptive text or content corresponding to a specific tag, which describes the information represented by the tag. For example, the tag "credit score" may correspond to the tag mapping text "Your credit score is 700, which is a good level"; position mapping refers to finding the position corresponding to a keyword or combination of keywords in the credit report text.

[0022] In an optional embodiment, the credit classification model includes a semantic representation module, a semantic association module and a semantic classification module. The semantic representation module is used to extract features of the standard credit report text based on a pre-trained BERT model to obtain a BERT vector corresponding to the standard credit report text. The semantic association module is used to perform deep convolution operations on the BERT vector based on a deep pyramid convolutional neural network layer to obtain local features in the standard credit report text, and to capture long-range dependencies of the standard credit report text based on the convolution layer and the pooling layer. The semantic classification module is used to classify the standard credit report text based on the local features and long-range dependencies in the standard credit report text to obtain a credit type label corresponding to the standard credit report text.

[0023] It should be noted that BERT is a pre-trained language model that can capture contextual information in text. It uses a bidirectional Transformer structure and a large amount of corpus for unsupervised training to extract semantic features of the text. The BERT vector is a text vector representation generated by the BERT model, which captures the semantic information of each word and sentence in the text. The Deep Pyramid Convolutional Neural Network (DPCNN) is a convolutional neural network (CNN) architecture designed to process text data. It uses multiple convolutional layers and pooling layers to extract local features of the text in a pyramid structure, and gradually enhances the performance of the model in a hierarchical manner. Long-range dependencies refer to the semantic connection between distant words or phrases in the text. In language, important information may span a long distance, requiring the model to understand the global context.

[0024] In an optional embodiment, the sentiment classification model includes a vector embedding layer, an encoder, a hybrid attention mechanism and a decoder, the vector embedding layer adopts a BER model, the encoder includes a bidirectional gated recurrent unit and a long short-term memory network, the hybrid attention mechanism is used to fuse the encoder output results, and the decoder is used to map the fused encoder output results to the sentiment type.

[0025] It should be noted that the purpose of the vector embedding layer is to convert the words in the text into a dense vector form, so that each word can correspond to a low-dimensional vector representation; BERT is a pre-trained language model based on the Transformer architecture; the encoder is a part of the model that is used to process the sequence of input text and extract features; the bidirectional gated recurrent unit (Bi-GRU) means that when processing a text sequence, the model not only processes the sequence from front to back (i.e., sequential input), but also processes the sequence from back to front (i.e., reverse input), which enables the model to simultaneously utilize the previous and next information in the context to obtain a more comprehensive semantic representation; LSTM is a special RNN structure used to solve the gradient vanishing problem encountered by traditional RNNs when processing long-range dependencies. It controls the flow of information by introducing "memory units" and multiple gating mechanisms (input gate, forget gate, and output gate), so that it can capture long-term dependencies; the hybrid attention mechanism is used to combine different information output by the encoder (such as the output of Bi-GRU and LSTM). It can automatically select the most relevant information by calculating the weights of different parts, thereby improving the accuracy of the model for sentiment classification tasks; the decoder generates sentiment classification prediction results by processing the vector output by the encoder.

[0026] In an optional embodiment, sentiment classification is performed on the communication text based on a pre-trained sentiment classification model to obtain a sentiment type label corresponding to the communication text, including: B1. Input the communication text into the BER model and convert the communication text into the target word vector through the BER model; B2. Input the target word vector into the bidirectional gated recurrent unit and the long short-term memory network respectively, and obtain the inter-sentence semantic features of the target word vector through the bidirectional gated recurrent unit; B3. Convert the inter-sentence semantic features into an inter-sentence semantic feature vector, obtain the intra-sentence semantic features of the target word vector through the long short-term memory network, and convert the intra-sentence semantic features into an intra-sentence semantic feature vector.

[0027] In an optional embodiment, sentiment classification is performed on the communication text based on a pre-trained sentiment classification model to obtain a sentiment type label corresponding to the communication text, and further includes: B4, fusing the inter-sentence semantic feature vector and the intra-sentence semantic feature vector through a hybrid attention mechanism to obtain a first fused vector; B5, fusing the inter-sentence semantic feature vector, the intra-sentence semantic feature vector and the first fusion vector to obtain a second fusion vector; B6. Input the second fusion vector into the decoder, and map the second fusion vector to the emotion type through the decoder to obtain the emotion type label corresponding to the communication text.

[0028] In an optional embodiment, the first interactive text and the second interactive text are merged to obtain a communication text, including: C1. Segment the first interactive text and the second interactive text to obtain a first sentence sequence corresponding to the first interactive text and a second sentence sequence corresponding to the second interactive text; C2, encoding the first sentence sequence and the second sentence sequence to obtain a first sentence encoding sequence corresponding to the first sentence sequence and a second sentence encoding sequence corresponding to the second sentence sequence; C3, calculating the sentence similarity between the first sentence and the second sentence based on the first sentence encoding sequence and the second sentence encoding sequence; C4. If the sentence similarity is greater than the preset similarity threshold, the second sentence is merged into the first sentence; C5. Repeat the above operations until the first sentence coding sequence and the second sentence coding sequence are traversed to merge the first interactive text and the second interactive text to obtain a communication text.

[0029] In an optional embodiment, the label mapping network adopts a BP neural network.

[0030] It should be noted that the label mapping network refers to a neural network that maps input data to corresponding labels. It is usually used in classification tasks. The output of the network is labels of different categories. The goal of the label mapping network is to achieve effective classification or regression of input data by learning the relationship between input features and target labels; BP neural network, also known as backpropagation neural network, is a neural network trained by gradient descent algorithm. It is the most basic multi-layer feedforward neural network, usually composed of input layer, hidden layer and output layer. The BP algorithm updates the weights by backpropagating errors to optimize the performance of the network.

[0031] In an optional embodiment, label mapping is performed on the credit type label and the emotion type label of the target customer based on a pre-trained label mapping network to obtain the overdue rate label corresponding to the target customer, including: D1. Input the target customer's credit type label and sentiment type label into the label mapping network; D2. Generate the overdue rate label corresponding to the target customer through the forward propagation process of the label mapping network, where the overdue rate label is the overdue probability.

[0032] Embodiment three, as Figure 2 As shown, the data prediction system for the overdue rate of automobile installment loans proposed by the present invention is applicable to the data prediction method for the overdue rate of automobile installment loans, and comprises: Information processing unit 1, which is used to obtain the credit report of the target customer, convert the credit report into a credit report text based on a pre-trained text recognition model, and pre-process the credit report text to obtain a standard credit report text; The credit classification unit 2 is used to classify the standard credit report text based on the pre-trained credit classification model to obtain the credit type label corresponding to the standard credit report text, that is, to obtain the credit type label of the target customer; The text merging unit 3 is used to obtain the target customer's interactive speech and the first interactive text, convert the interactive speech into the second interactive text based on the pre-trained speech recognition model, and merge the first interactive text and the second interactive text to obtain the communication text; The emotion classification unit 4 is used to perform emotion classification on the communication text based on the pre-trained emotion classification model to obtain the emotion type label corresponding to the communication text, that is, to obtain the emotion type label of the target customer; The label mapping unit 5 is used to perform label mapping on the credit type label and the emotion type label of the target customer based on the pre-trained label mapping network to obtain the overdue rate label corresponding to the target customer.

[0033] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A data prediction method for automobile installment loan delinquency rate, characterized in that: include: Obtaining a credit report of a target customer, converting the credit report into a credit report text based on a pre-trained text recognition model, and pre-processing the credit report text to obtain a standard credit report text; Based on the pre-trained credit classification model, the standard credit report text is credit classified to obtain the credit type label corresponding to the standard credit report text, that is, to obtain the credit type label of the target customer; Acquire the target customer's interactive speech and the first interactive text, convert the interactive speech into a second interactive text based on a pre-trained speech recognition model, and merge the first interactive text and the second interactive text to obtain a communication text; Performing sentiment classification on the communication text based on a pre-trained sentiment classification model to obtain a sentiment type label corresponding to the communication text, that is, obtaining the sentiment type label of the target customer; The credit type label and the emotion type label of the target customer are label mapped based on a pre-trained label mapping network to obtain the overdue rate label corresponding to the target customer.

2. The data prediction method for automobile installment loan overdue rate according to claim 1 is characterized in that: Preprocessing the credit report text to obtain a standard credit report text, including: Performing a word segmentation operation on the credit report text by using a word segmentation tool to obtain a keyword sequence; Traversing the keyword sequence, selecting two adjacent keywords in the keyword sequence as a combination keyword, and matching the combination keyword with a preset tag table to obtain a tag corresponding to the combination keyword; Obtaining a label mapping text corresponding to the label, and adding the label mapping text to a position corresponding to the combined keyword in the credit report text; Repeat the above operation until each keyword in the keyword sequence is traversed to obtain the standard credit report text.

3. The data prediction method for automobile installment loan overdue rate according to claim 1 is characterized in that: The credit classification model includes a semantic representation module, a semantic association module and a semantic classification module. The semantic representation module is used to extract features of the standard credit report text based on a pre-trained BERT model to obtain a BERT vector corresponding to the standard credit report text. The semantic association module is used to perform a deep convolution operation on the BERT vector based on a deep pyramid convolutional neural network layer to obtain local features in the standard credit report text, and to capture the long-range dependencies of the standard credit report text based on the convolution layer and the pooling layer. The semantic classification module is used to classify the standard credit report text based on the local features and long-range dependencies in the standard credit report text to obtain a credit type label corresponding to the standard credit report text.

4. The data prediction method for automobile installment loan overdue rate according to claim 1 is characterized in that: The sentiment classification model includes a vector embedding layer, an encoder, a hybrid attention mechanism and a decoder. The vector embedding layer adopts a BER model. The encoder includes a bidirectional gated recurrent unit and a long short-term memory network. The hybrid attention mechanism is used to fuse the encoder output results. The decoder is used to map the fused encoder output results to the sentiment type.

5. The data prediction method for automobile installment loan overdue rate according to claim 4 is characterized in that: The communication text is sentimentally classified based on a pre-trained sentiment classification model to obtain a sentiment type label corresponding to the communication text, including: Inputting the communication text into the BER model, and converting the communication text into a target word vector through the BER model; Inputting the target word vector into the bidirectional gated recurrent unit and the long short-term memory network respectively, and obtaining the inter-sentence semantic features of the target word vector through the bidirectional gated recurrent unit; The inter-sentence semantic features are converted into inter-sentence semantic feature vectors, the intra-sentence semantic features of the target word vector are obtained through the long short-term memory network, and the intra-sentence semantic features are converted into intra-sentence semantic feature vectors.

6. The method for predicting the overdue rate of automobile installment loans according to claim 5, characterized in that: The communication text is sentimentally classified based on a pre-trained sentiment classification model to obtain a sentiment type label corresponding to the communication text, and further includes: fusing the inter-sentence semantic feature vector and the intra-sentence semantic feature vector through the hybrid attention mechanism to obtain a first fused vector; fusing the inter-sentence semantic feature vector, the intra-sentence semantic feature vector, and the first fusion vector to obtain a second fusion vector; The second fusion vector is input into the decoder, and the second fusion vector is mapped to the emotion type by the decoder to obtain the emotion type label corresponding to the communication text.

7. The data prediction method for automobile installment loan overdue rate according to claim 1 is characterized in that: The first interactive text and the second interactive text are combined to obtain a communication text, including: Sentence the first interactive text and the second interactive text to obtain a first sentence sequence corresponding to the first interactive text and a second sentence sequence corresponding to the second interactive text; Encoding the first sentence sequence and the second sentence sequence to obtain a first sentence encoding sequence corresponding to the first sentence sequence and a second sentence encoding sequence corresponding to the second sentence sequence; Calculating sentence similarity between the first sentence and the second sentence based on the first sentence encoding sequence and the second sentence encoding sequence; If the sentence similarity is greater than a preset similarity threshold, merging the second sentence into the first sentence; The above operation is repeatedly performed until the first sentence coding sequence and the second sentence coding sequence are traversed to merge the first interactive text and the second interactive text to obtain a communication text.

8. The method for predicting the overdue rate of automobile installment loans according to claim 1, characterized in that: The label mapping network adopts BP neural network.

9. The data prediction method for automobile installment loan overdue rate according to claim 8, characterized in that: Based on a pre-trained label mapping network, label mapping is performed on the credit type label and the emotion type label of the target customer to obtain the overdue rate label corresponding to the target customer, including: Inputting the credit type label and the emotion type label of the target customer into the label mapping network; Through the forward propagation process of the label mapping network, the overdue rate label corresponding to the target customer is generated, wherein the overdue rate label is the overdue probability.

10. A data prediction system for automobile installment loan overdue rate, which is applicable to a data prediction method for automobile installment loan overdue rate according to any one of claims 1 to 9, characterized in that: include: An information processing unit (1), the information processing unit (1) being used to obtain a credit report of a target customer, convert the credit report into a credit report text based on a pre-trained text recognition model, and pre-process the credit report text to obtain a standard credit report text; A credit classification unit (2), the credit classification unit (2) is used to perform credit classification on the standard credit report text based on a pre-trained credit classification model to obtain a credit type label corresponding to the standard credit report text, that is, to obtain the credit type label of the target customer; A text merging unit (3), the text merging unit (3) being used to obtain the target customer's interactive speech and the first interactive text, convert the interactive speech into a second interactive text based on a pre-trained speech recognition model, and merge the first interactive text and the second interactive text to obtain a communication text; A sentiment classification unit (4), the sentiment classification unit (4) is used to perform sentiment classification on the communication text based on a pre-trained sentiment classification model to obtain a sentiment type label corresponding to the communication text, that is, to obtain the sentiment type label of the target customer; A label mapping unit (5), the label mapping unit (5) is used to perform label mapping on the credit type label and the emotion type label of the target customer based on a pre-trained label mapping network, so as to obtain an overdue rate label corresponding to the target customer.

Images