Complaint type prediction method and device, computer device and storage medium

Abstract

The application relates to a complaint type prediction method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: obtaining complaint information of a target user; performing word segmentation processing on the complaint information to obtain segmented complaint text; inputting the segmented complaint text into a pre-trained complaint rate prediction model to obtain a feature vector of the complaint information, and obtaining a complaint rate of the complaint information according to the feature vector of the complaint information; weighting the segmented complaint text based on the complaint rate of the complaint information to obtain weighted complaint text; inputting the weighted complaint text into a pre-trained complaint type prediction model to obtain a feature vector of the complaint type; and obtaining a complaint type of the target user according to the feature vector of the complaint type. According to the method, the accuracy of user complaint type prediction can be improved based on weighting of the complaint text.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus, computer device, storage medium, and computer program product for predicting complaint types. Background Technology

[0002] With the development of internet technology, there is a need to convert users' voice and text in real time to obtain users' emotional information, and combine it with users' historical emotional information to predict users' types.

[0003] In existing methods, keywords are manually extracted from users' voice text based on business experience. This method can only extract keywords from users' voice text, providing simple reference for business personnel, but it cannot predict the user's type. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for predicting complaint types that can improve the accuracy of predicting user complaint types, in order to address the aforementioned technical problems.

[0005] Firstly, this application provides a method for predicting complaint types, the method comprising:

[0006] Obtain complaint information from target users;

[0007] The complaint information is segmented into words to obtain the segmented complaint text;

[0008] The segmented complaint text is input into a pre-trained complaint rate prediction model to obtain the feature vector of the complaint information, and the complaint rate of the complaint information is obtained based on the feature vector of the complaint information.

[0009] Based on the complaint rate of the aforementioned complaint information, the segmented complaint text is weighted to obtain a weighted complaint text.

[0010] The weighted complaint text is input into a pre-trained complaint type prediction model to obtain the feature vector of the complaint type.

[0011] The complaint type of the target user is obtained based on the feature vector of the complaint type.

[0012] In one embodiment, the method for training the complaint rate prediction model includes:

[0013] Obtain a complaint sample training set; the complaint sample training set includes multiple complaint samples and labeled complaint frequencies;

[0014] Obtain complaint samples from the complaint sample training set;

[0015] The complaint sample is segmented into words to obtain the segmented complaint sample.

[0016] The segmented complaint samples are input into the complaint rate prediction model to be trained to obtain the feature vector of the complaint samples. Based on the feature vector of the complaint samples, the predicted complaint frequency of the complaint samples is obtained.

[0017] Based on the difference between the predicted complaint frequency of the complaint samples and the labeled complaint frequency, the parameters of the complaint rate prediction model are adjusted to obtain a trained complaint rate prediction model.

[0018] In one embodiment, obtaining the complaint sample training set includes:

[0019] Obtain an initial sample training set consisting of historical complaint information; the initial complaint sample training set includes multiple complaint samples and labeled complaint frequencies;

[0020] Obtain related text that is semantically similar to the complaint sample;

[0021] The complaint frequency of the labeled complaint samples that are semantically similar to the associated text is taken as the complaint frequency of the labeled associated text.

[0022] Based on the associated text and the frequency of complaints annotated, the complaint sample training set is expanded to obtain the complaint sample training set.

[0023] In one embodiment, the step of weighting the segmented complaint text based on the complaint rate of the complaint information to obtain a weighted complaint text includes:

[0024] The complaint weight is determined based on the complaint rate of the complaint information; the value of the complaint rate is positively correlated with the complaint weight.

[0025] The segmented complaint text is weighted based on the complaint weights to obtain the weighted complaint text.

[0026] In one embodiment, determining the complaint weight based on the complaint rate of the complaint information includes:

[0027] Based on the complaint rate of the complaint information, the complaint severity level corresponding to the complaint rate is determined; the complaint severity level includes at least two types; the higher the complaint rate value, the higher the complaint severity level.

[0028] The complaint weight is determined based on the complaint severity level, wherein the higher the complaint severity level, the greater the complaint weight.

[0029] In one embodiment, the step of segmenting the complaint information to obtain the segmented complaint text includes:

[0030] Based on a word segmentation corpus in the field of financial debt collection, the complaint information is segmented to obtain the segmented complaint text.

[0031] Secondly, this application provides a device for predicting complaint types, the device comprising:

[0032] The acquisition module is used to acquire complaint information from target users;

[0033] The preprocessing module is used to perform word segmentation on the complaint information to obtain the segmented complaint text;

[0034] The complaint rate prediction module is used to input the segmented complaint text into a pre-trained complaint rate prediction model to obtain the feature vector of the complaint information, and to obtain the complaint rate of the complaint information based on the feature vector of the complaint information.

[0035] The weighting module is used to weight the segmented complaint text based on the complaint rate of the complaint information to obtain the weighted complaint text.

[0036] The processing module is used to input the weighted complaint text into a pre-trained complaint type prediction model to obtain the feature vector of the complaint type;

[0037] The complaint type prediction module is used to obtain the complaint type of the target user based on the feature vector of the complaint type.

[0038] Thirdly, this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the following method steps:

[0039] Obtain complaint information from target users;

[0040] The complaint information is segmented into words to obtain the segmented complaint text;

[0041] The segmented complaint text is input into a pre-trained complaint rate prediction model to obtain the feature vector of the complaint information, and the complaint rate of the complaint information is obtained based on the feature vector of the complaint information.

[0042] Based on the complaint rate of the aforementioned complaint information, the segmented complaint text is weighted to obtain a weighted complaint text.

[0043] The weighted complaint text is input into a pre-trained complaint type prediction model to obtain the feature vector of the complaint type.

[0044] The complaint type of the target user is obtained based on the feature vector of the complaint type.

[0045] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the following method steps:

[0046] Obtain complaint information from target users;

[0047] The complaint information is segmented into words to obtain the segmented complaint text;

[0048] The segmented complaint text is input into a pre-trained complaint rate prediction model to obtain the feature vector of the complaint information, and the complaint rate of the complaint information is obtained based on the feature vector of the complaint information.

[0049] Based on the complaint rate of the aforementioned complaint information, the segmented complaint text is weighted to obtain a weighted complaint text.

[0050] The weighted complaint text is input into a pre-trained complaint type prediction model to obtain the feature vector of the complaint type.

[0051] The complaint type of the target user is obtained based on the feature vector of the complaint type.

[0052] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the following method steps:

[0053] Obtain complaint information from target users;

[0054] The complaint information is segmented into words to obtain the segmented complaint text;

[0055] The segmented complaint text is input into a pre-trained complaint rate prediction model to obtain the feature vector of the complaint information, and the complaint rate of the complaint information is obtained based on the feature vector of the complaint information.

[0056] Based on the complaint rate of the aforementioned complaint information, the segmented complaint text is weighted to obtain a weighted complaint text.

[0057] The weighted complaint text is input into a pre-trained complaint type prediction model to obtain the feature vector of the complaint type.

[0058] The complaint type of the target user is obtained based on the feature vector of the complaint type.

[0059] The aforementioned method, apparatus, computer equipment, storage medium, and computer program products for predicting complaint types acquire complaint information from target users, segment the complaint information into words, and input the segmented complaint text into a complaint rate prediction model to obtain the complaint rate of the complaint information. This provides real-time information on user complaint rates, offering accurate reference for business personnel. Furthermore, the complaint text is weighted based on the complaint rate, with different weights assigned according to the complaint rate. The weighted complaint text is then input into a pre-trained complaint type prediction model to obtain the user's complaint type. This method, based on weighting the complaint text, can improve the accuracy of predicting user complaint types. Attached Figure Description

[0060] Figure 1 This is an application environment diagram of a complaint type prediction method in one embodiment;

[0061] Figure 2 This is a flowchart illustrating a method for predicting complaint types in one embodiment;

[0062] Figure 3 This is a flowchart illustrating a method for training a complaint rate prediction model in one embodiment.

[0063] Figure 4 This is a flowchart illustrating a method for obtaining a training set of complaint samples in one embodiment;

[0064] Figure 5 This is a flowchart illustrating a method for predicting complaint types in another embodiment;

[0065] Figure 6 This is a structural block diagram of a complaint type prediction device in one embodiment;

[0066] Figure 7 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0068] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0069] With the development of internet technology, there is a need for real-time conversion of users' voice and text to obtain their emotional information. This information, combined with historical emotional data, can then be used to predict user type. For example, in the financial debt collection field, when collecting payments from late borrowers, the lack of information on the type of complaint may lead to unnecessary complaints and damage the company's image.

[0070] In existing methods, keywords are manually extracted from users' voice text based on business experience. However, this method is not accurate enough in predicting user types, and business personnel cannot communicate with users according to their types, which leads to user complaints.

[0071] In view of this, the complaint type prediction method provided in the embodiments of this application can be applied to, for example, Figure 1 In the application environment shown, the business terminal 102 communicates with the server 104 via a network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated onto the server 104, or it can be located in the cloud or on another network server.

[0072] Server 104 obtains complaint information from the target user from business terminal 102; Server 104 performs word segmentation on the complaint information to obtain segmented complaint text; Server 104 inputs the segmented complaint text into a pre-trained complaint rate prediction model to obtain the feature vector of the complaint information, and obtains the complaint rate of the complaint information based on the feature vector; Server 104 weights the segmented complaint text based on the complaint rate of the complaint information to obtain weighted complaint text; Server 104 inputs the weighted complaint text into a pre-trained complaint type prediction model to obtain the feature vector of the complaint type; Server 104 obtains the complaint type of the target user based on the feature vector of the complaint type.

[0073] The business terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle systems. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0074] In one embodiment, such as Figure 2 As shown, a method for predicting complaint types is provided, which can be applied to... Figure 1 Taking the server in the example, the following steps are included:

[0075] S202, Obtain complaint information from the target user.

[0076] The target user can be a natural person or an organization. Specifically, when a user is talking to a business person, the user is marked as a target user. By obtaining the target user's complaint information, the behavior or intention of the target user can be identified.

[0077] This can be achieved by obtaining complaint information from target users when they make complaint calls, thereby gaining insight into their behavior or intentions. Complaint information can be voice information or emotional information. Specifically, voice-to-text conversion tools can be used to convert voice information into text-based complaint information. Alternatively, deep learning-based sentiment analysis models can be used to obtain target users' complaint information.

[0078] S204, perform word segmentation on the complaint information to obtain the segmented complaint text.

[0079] Word segmentation refers to separating complaint information according to certain keyword extraction rules. For example, user complaint information can be segmented using the shortest path word segmentation algorithm, dictionary-based matching algorithm, or text segmentation model. During the word segmentation process, the part-of-speech tag of each word is also determined.

[0080] Specifically, after obtaining the segmented complaint text, the bag-of-words model can be used to process the complaint information. This processing includes calculating the frequency of occurrence of each segmented complaint text and calculating high-frequency complaint texts based on the complaint frequency. The bag-of-words model is a simplified model used in natural language processing and information retrieval. It is a text vectorization model that generally does not consider grammar or word order, but rather the frequency of occurrence of all words. For example, the segmented words are placed in a bag, and each word is independent.

[0081] The frequency of occurrence of each segmented complaint text refers to the ratio between the number of times that complaint text appears in the complaint text corpus and the total number of complaint texts. A higher frequency indicates a stronger correlation between the complaint text and the user complaint. For example, if the complaint text corpus includes terms like principal, interest rate, and loan / repayment, and the frequency of "loan / repayment" is higher than the other three types, then the correlation between that complaint text (loan / repayment) and the user complaint is stronger. The correlation of user complaints is determined by linking the user's complaint identifier, complaint information, and the segmented complaint text. If the user's complaint identifier is "yes," indicating a complaint has been filed, then the user's complaint information and the segmented complaint text are added to the complaint text corpus.

[0082] The part-of-speech tag (POS) of each segmented complaint text refers to the part-of-speech category to which the complaint text belongs. This can be achieved using a bag-of-words model or other text segmentation extraction models. For example, when segmenting complaint information, the POS category of the complaint text can be extracted. The POS category includes content words such as nouns, verbs, and adjectives, and function words such as adverbs, prepositions, and onomatopoeia.

[0083] Specifically, during the word segmentation process of complaint information, the complaint text with the target part of speech can be removed to obtain the segmented complaint text.

[0084] S206. Input the segmented complaint text into the pre-trained complaint rate prediction model to obtain the feature vector of the complaint information, and obtain the complaint rate of the complaint information based on the feature vector of the complaint information.

[0085] Among them, the complaint rate prediction model is a model that predicts the complaint rate of complaint information. Specifically, after receiving the user's complaint information, it is necessary to predict the probability of the user's complaint based on the specific content of the complaint information.

[0086] Among them, the feature vector refers to the vector representation of each word in the complaint information. The feature vector can be used to compare with the feature vectors of other complaint texts or complaint texts in the complaint text corpus to obtain the similarity between complaint texts after different word segmentation.

[0087] The complaint rate prediction model can be built based on a logistic regression model, which is a generalized linear regression model applicable to data mining, automated disease diagnosis, and economic forecasting. Through logistic regression analysis, the weights of independent variables can be obtained, thus identifying key factors. Predictions are then made based on the independent variables of these key factors and their weights.

[0088] Specifically, the segmented complaint model can be input into a pre-trained complaint rate prediction model based on logistic regression to obtain the feature vector of the complaint information. By comparing the feature vector of the complaint information with the feature vectors of other complaint information in the complaint text corpus, the frequency of occurrence of the feature vector of the complaint information can be calculated. That is, the complaint rate of the complaint information is obtained from the feature vector of the complaint information.

[0089] Specifically, by comparing the feature vectors of complaint information with the feature vectors of other complaint information in the complaint text corpus, feature vectors with similarity within a threshold range can be removed from multiple feature vectors to obtain the removed feature vectors and the corresponding complaint information. Then, based on the feature vectors of the complaint information, the removed feature vectors and the corresponding complaint information, the frequency of occurrence of the feature vectors of the complaint information can be calculated.

[0090] S208. Based on the complaint rate of the complaint information, the segmented complaint text is weighted to obtain the weighted complaint text.

[0091] The complaint rate can be the frequency of occurrence of a complaint; the higher the frequency, the greater the importance of the complaint.

[0092] Specifically, the segmented complaint text is weighted based on the importance of the complaint information.

[0093] One way to weight the segmented complaint text is to grade the weighting according to the importance of the complaint information. For example, according to the complaint rate from high to low, the weighting can be divided into three levels: the first level represents the complaint rate of the complaint information is greater than 70%, the second level represents the complaint rate of the complaint information is greater than 30% and less than or equal to 70%, and the third level represents the complaint rate of the complaint information is less than or equal to 30%.

[0094] S210, input the weighted complaint text into the pre-trained complaint type prediction model to obtain the feature vector of the complaint type.

[0095] The complaint type can be the result of user complaint profile analysis. Specifically, the complaint profile analysis result can include the user's complaint tendency level, the user's current situation analysis result, and the user's demand analysis result. For example, the user's expected repayment time and the user's complaint tendency level are both high.

[0096] The complaint type prediction model can be a topic model. Topic modeling (Latent Dirichlet Allocation, LDA) is a model that automatically encodes a text corpus into a set of meaningful categories. LDA topic models can automatically encode and classify several documents into a certain mathematical topic, with each topic corresponding to multiple feature words, which greatly reduces the human burden and error. By analyzing these topics, key information can be extracted from a large number of articles, allowing for further analysis of the connections between them. Specific applications include semantic analysis, document classification / clustering, article summarization, community mining, and more.

[0097] Specifically, the topic model can be trained based on numerical data such as the number of collection calls and the number of overdue days, or it can be trained based on a word segmentation corpus to obtain feature vectors for complaint types.

[0098] S214, Obtain the complaint type of the target user based on the feature vector of the complaint type.

[0099] The feature vector of the complaint type includes the vectorized complaint profile analysis results. Specifically, the vectorized complaint profile analysis results can include user complaint tendency level, user current status analysis results, user demand analysis results, etc.

[0100] Specifically, the feature vectors of other complaint types in the word segmentation corpus can be compared with the feature vectors of the predicted complaint type to obtain the similarity of the feature vectors. If the similarity of the feature vectors is less than a preset threshold, the two feature vectors are considered to match, and other complaint types in the word segmentation corpus are output as the complaint types of the target user.

[0101] In the above-mentioned method for predicting complaint types, the complaint information of the target user is obtained, the complaint information is segmented into words, and the segmented complaint text is input into the complaint rate prediction model to obtain the complaint rate of the complaint information. The user's complaint rate is obtained in real time, providing accurate reference for business personnel. Then, the complaint text is weighted based on the complaint rate. The complaint text is weighted according to different complaint rates, and the weighted complaint text is input into a pre-trained complaint type prediction model to obtain the user's complaint type. This method, based on weighting the complaint text, can improve the accuracy of predicting the user's complaint type.

[0102] In one embodiment, such as Figure 3 As shown, the method for training the complaint rate prediction model includes:

[0103] S302, Obtain the complaint sample training set; the complaint sample training set includes multiple complaint samples and labeled complaint frequencies.

[0104] The sample training set is the training sample set for the complaint rate prediction model. Existing labeled data can be used as the sample training set, or the training sample set can be manually labeled in advance to obtain multiple complaint samples and labeled complaint frequencies.

[0105] The complaint samples can be user voice information or emotional information, etc. The labeled complaint frequency is the frequency of complaints corresponding to the complaint samples. This frequency can be automatically calculated based on the occurrence frequency of the target tag in the existing labeled data, using the number of occurrences of each segmented sample. Specifically, the complaint frequency of a segmented sample can be obtained by dividing the number of individual segmented samples by the total number of segmented samples.

[0106] Specifically, the sample training set can take into account voice information and emotional information in the financial industry, such as principal, interest rate, loan repayment, upcoming repayment, and on-time repayment.

[0107] S304, Obtain complaint samples from the complaint sample training set.

[0108] S306, perform word segmentation on the complaint sample to obtain the segmented complaint sample.

[0109] The process of segmenting complaint samples involves separating the complaint information according to certain keyword extraction rules. For example, the shortest path segmentation algorithm, dictionary-based matching algorithm, or text segmentation model can be used to segment the complaint information.

[0110] S308: Input the segmented complaint samples into the complaint rate prediction model to be trained to obtain the feature vector of the complaint samples. Based on the feature vector of the complaint samples, the predicted complaint frequency of the complaint samples is obtained.

[0111] The feature vector refers to the vector representation of each segmented complaint sample. The vector representation of each segmented complaint sample is input into the trained complaint rate prediction model, and the complaint frequency of the complaint sample is output, that is, the predicted complaint frequency.

[0112] S310. Based on the difference between the predicted complaint frequency and the labeled complaint frequency of the complaint samples, adjust the parameters of the complaint rate prediction model to obtain the trained complaint rate prediction model.

[0113] The process involves comparing the difference between the predicted and labeled complaint frequencies of the complaint samples, adjusting the parameters of the prediction model based on the magnitude of the difference, and repeating the model training steps. If the difference between the predicted and labeled complaint frequencies of the complaint samples is less than a threshold, the model is considered to have reached the iteration termination condition, and a well-trained complaint rate prediction model is obtained.

[0114] Specifically, the loss function of the algorithm can be used to compare the difference between the predicted frequency of the complaint samples and the labeled complaint frequency, and the parameter size of the prediction model can be adjusted according to the magnitude of the difference.

[0115] Specifically, the difference between the predicted frequency of a complaint sample and the labeled complaint frequency can be the similarity distance between feature vectors. If the similarity distance between the feature vector corresponding to the predicted frequency of the complaint sample and the feature vector corresponding to the labeled complaint frequency is less than a preset threshold, the model is considered to have reached the iteration termination condition, and a trained complaint rate prediction model is obtained. If the similarity distance between the feature vector corresponding to the predicted frequency of the complaint sample and the feature vector corresponding to the labeled complaint frequency is greater than or equal to the preset threshold, the weights corresponding to the parameters of the prediction model are adjusted, the model is updated in reverse, and the prediction process for complaint samples is repeated until the model training is completed, resulting in a trained complaint rate prediction model.

[0116] In this embodiment, by training the complaint rate prediction model and taking into account the voice and emotional information of the financial industry, it is possible to accurately obtain the user's current situation and demands, and accurately predict the user's complaint rate.

[0117] In one embodiment, such as Figure 4As shown, the method for obtaining the complaint sample training set includes:

[0118] S402, Obtain an initial sample training set consisting of historical complaint information; the initial complaint sample training set includes multiple complaint samples and labeled complaint frequencies.

[0119] The initial sample training set, which consists of historical complaint information, can be the first training sample set for the complaint rate prediction model. The initial sample training set can be updated, including the complaint samples and the labeled complaint frequencies.

[0120] Specifically, updating the initial sample training set can include updating the semantically related associated texts in the initial sample training set, and updating the Chinese sentiment dictionary related to the sentiment information in the initial sample training set.

[0121] The Chinese sentiment dictionary can be constructed using a bag-of-words model. The bag-of-words model obtains sentiment information from the user's historical complaint sample training set, and the Chinese sentiment dictionary is updated based on the associated sentiment information in the complaint sample training set.

[0122] S404, Obtain related texts that are semantically similar to the complaint sample.

[0123] One approach is to use a semantic association model to obtain semantically similar related texts from complaint samples. A semantic association model is a model that uses a specific algorithm to find similar words in complaint samples.

[0124] The semantic association model can be a word2vec model. Word2vec is a group of related models used to generate word vectors. These models are shallow, two-layer neural networks used for training to reconstruct linguistic word text. The network represents words and needs to guess the input words in adjacent positions. Under the bag-of-words assumption in word2vec, the order of words is unimportant. After training, the word2vec model can be used to map each word to a vector, which can be used to represent the relationship between words. This vector is the hidden layer of the neural network.

[0125] Specifically, the word2vec model is used to obtain related texts that are semantically similar to the complaint samples.

[0126] S406, the complaint frequency of the labeled complaint samples that are semantically similar to the associated text is used as the complaint frequency of the labeled associated text.

[0127] S408. Based on the associated text and the frequency of complaints marked, expand the complaint sample training set to obtain the complaint sample training set.

[0128] In this embodiment, the complaint sample training set is expanded by using the frequency of complaints based on associated text and annotations. This enriches the complaint sample training set and facilitates the prediction of user complaint rates in subsequent processes.

[0129] In one embodiment, the segmented complaint text is weighted based on the complaint rate of the complaint information to obtain a weighted complaint text, including: determining the complaint weight based on the complaint rate of the complaint information; the value of the complaint rate is positively correlated with the complaint weight; and weighting the segmented complaint text based on the complaint weight to obtain the weighted complaint text.

[0130] The complaint rate can be the frequency of occurrence of a complaint; the higher the frequency, the greater the importance of the complaint.

[0131] Specifically, complaint weights are determined based on the importance of the complaint information. The higher the importance of the complaint information, the higher the complaint rate, and correspondingly, the higher the weight of the complaint rate.

[0132] One way to weight the segmented complaint text is to grade the weighting according to the importance of the complaint information. For example, according to the complaint rate from high to low, the weighting can be divided into three levels: the first level represents the complaint rate of the complaint information is greater than 70%, the second level represents the complaint rate of the complaint information is greater than 30% and less than or equal to 70%, and the third level represents the complaint rate of the complaint information is less than or equal to 30%.

[0133] In this embodiment, the complaint texts are weighted according to the complaint rate, and the weighted complaint texts are input into a pre-trained complaint type prediction model, which ensures the stability of the model while improving its performance.

[0134] In one embodiment, determining the complaint weight based on the complaint rate of the complaint information includes: determining the complaint severity level corresponding to the complaint rate based on the complaint rate of the complaint information; the complaint severity level includes at least two types; the higher the complaint rate, the higher the complaint severity level; and determining the complaint weight based on the complaint severity level, wherein the higher the complaint severity level, the greater the complaint weight.

[0135] Among them, different complaint rate intervals can be divided according to the complaint rate of the complaint information from high to low. The complaint rate intervals correspond to different complaint severity levels, and the complaint level includes at least two.

[0136] Specifically, the complaint rate information can be divided into at least two ranges from high to low, for example, a complaint rate greater than 70% and a complaint rate less than or equal to 70%. The complaint rate range with a complaint rate greater than 70% corresponds to the first level of complaint severity, and the complaint rate range with a complaint rate less than or equal to 70% corresponds to the second level of complaint severity, with the first level of complaint severity being higher than the second level of complaint severity.

[0137] It should be noted that complaint information can be divided into more complaint rate intervals from high to low, such as three or more. The more intervals there are, the more accurate the complaint type prediction model can be, but the less stable the model becomes. Dividing complaint information into three intervals from high to low can ensure the accuracy of the complaint type prediction model while maintaining appropriate stability.

[0138] Complaint information typically contains a wide variety of complaint terms. Directly inputting these terms into a complaint type prediction model would result in an overly complex model, hindering deployment and usability, and leading to poor stability. A solution is to divide the complaint information into multiple complaint rate intervals, from high to low, with each interval representing a different complaint level. This dimensionality reduction, moving from high-dimensional input information with numerous complaint terms to low-dimensional input information corresponding to fewer complaint levels, improves the model's stability.

[0139] This can be achieved by obtaining complaint information from target users when they make complaint calls, thereby gaining insight into their behavior or intentions. Complaint information can be voice information or emotional information. Specifically, voice-to-text conversion tools can be used to convert voice information into text-based complaint information. Alternatively, deep learning-based sentiment analysis models can be used to obtain target users' complaint information.

[0140] In this embodiment, the complaint severity level corresponding to the complaint rate is determined based on the complaint information, and the complaint weight is determined based on the complaint severity level. The complaint text is weighted according to the complaint rate, and the weighted complaint text is input into the pre-trained complaint type prediction model. This reduces the dimensionality of the model input, ensuring the accuracy of the complaint type prediction model while maintaining appropriate stability.

[0141] In one embodiment, the complaint information is segmented into words to obtain the segmented complaint text, including: segmenting the complaint information into words based on a segmentation corpus in the field of financial debt collection to obtain the segmented complaint text.

[0142] In the financial debt collection business, when customers fail to repay on time, collection efforts are initiated. During the collection process, some unnecessary complaints may arise, which can seriously damage the company's image. Therefore, it is necessary to construct a corpus of domain-specific terms related to customer complaint information.

[0143] Existing word segmentation corpora are generally built from vocabulary in the People's Daily, which cannot accurately segment complaint information. A word segmentation corpus for the financial collection field can be built based on customer voice information and historical collection information from the financial collection field.

[0144] Specifically, the word segmentation corpus in the field of financial debt collection includes terms such as principal, interest rate, and loan repayment, which are more suitable for word segmentation processing of complaint information.

[0145] In this embodiment, a word segmentation corpus in the field of financial debt collection is used to segment complaint information to obtain segmented complaint text, which can accurately segment words and facilitate the construction of a complaint word segmentation corpus.

[0146] In one embodiment, such as Figure 5 A method for predicting complaint types is provided:

[0147] Part 1, Model Building:

[0148] Obtain a training set of complaint samples; the training set of complaint samples includes multiple complaint samples and labeled complaint frequencies;

[0149] S502, Obtain complaint samples from the complaint sample training set.

[0150] The process involves obtaining an initial sample training set composed of historical complaint information; the initial complaint sample training set includes multiple complaint samples and labeled complaint frequencies; obtaining related texts that are semantically similar to the complaint samples; using the labeled complaint frequencies of the complaint samples that are semantically similar to the related texts as the labeled complaint frequencies of the related texts; and expanding the complaint sample training set based on the related texts and labeled complaint frequencies to obtain the complaint sample training set.

[0151] S504, perform word segmentation on the complaint sample to obtain the segmented complaint sample.

[0152] S506. Input the segmented complaint samples into the complaint rate prediction model to be trained to obtain the feature vector of the complaint samples. Based on the feature vector of the complaint samples, the predicted complaint frequency of the complaint samples is obtained.

[0153] S508. Based on the difference between the predicted complaint frequency and the labeled complaint frequency of the complaint samples, the parameters of the complaint rate prediction model are adjusted to obtain a trained complaint rate prediction model.

[0154] Part Two, Model Application:

[0155] S510: Obtain complaint information from target users.

[0156] S512, based on the word segmentation corpus in the field of financial debt collection, the complaint information is segmented to obtain the segmented complaint text.

[0157] In the financial debt collection business, when customers fail to repay on time, collection efforts are initiated. During the collection process, some unnecessary complaints may arise, which can seriously damage the company's image. Therefore, it is necessary to construct a corpus of domain-specific terms related to customer complaint information.

[0158] Existing word segmentation corpora are generally built from vocabulary in the People's Daily, which cannot accurately segment complaint information. A word segmentation corpus for the financial collection field can be built based on customer voice information and historical collection information from the financial collection field.

[0159] Specifically, the word segmentation corpus in the field of financial debt collection includes terms such as principal, interest rate, and loan repayment, which are more suitable for word segmentation processing of complaint information.

[0160] S514: Input the segmented complaint text into the pre-trained complaint rate prediction model to obtain the feature vector of the complaint information, and obtain the complaint rate of the complaint information based on the feature vector of the complaint information.

[0161] S516. Based on the complaint rate of the complaint information, the complaint weight is determined; the value of the complaint rate is positively correlated with the complaint weight.

[0162] Specifically, the complaint severity level is determined based on the complaint rate of the complaint information; there are at least two complaint severity levels; the higher the complaint rate, the higher the complaint severity level; the complaint weight is determined based on the complaint severity level, where the higher the complaint severity level, the greater the complaint weight.

[0163] S518, the segmented complaint text is weighted based on the complaint weight to obtain the weighted complaint text.

[0164] S520: Input the weighted complaint text into the pre-trained complaint type prediction model to obtain the feature vector of the complaint type.

[0165] Before the weighted complaint text is input into the pre-trained complaint type prediction model, meaningful complaint text can be retained. Meaningful text refers to filtering the weighted complaint text. Specifically, the weighted complaint text may include dialectal text, unclear translations, and heavily colloquial text. By retaining the meaningful parts of the weighted complaint text, the user's intent can be preserved as much as possible.

[0166] In this approach, the effectiveness of the topic model is improved by assigning weights to keywords in the meaningful parts of the complaint text based on word frequency and business understanding, while retaining the meaningful parts of the complaint text.

[0167] S522, obtain the complaint type of the target user based on the feature vector of the complaint type.

[0168] By mining voice and text data related to debt collection, and identifying customers with a high tendency to complain in advance based on the complaint types of target users, sales personnel can provide more comprehensive explanations and communication services during business contacts, reducing unnecessary complaints. Simultaneously, by analyzing the complaint rate and type of target users, sales personnel can gain a more complete understanding of the customer's current situation and needs before collection, enabling them to quickly reach a consensus with the customer during collection, improving collection efficiency and reducing unnecessary complaints.

[0169] In this embodiment, complaint information from target users is acquired, segmented into words, and then input into a complaint rate prediction model to obtain the complaint rate. This provides real-time information on user complaint rates, offering accurate reference for business personnel. Furthermore, the complaint text is weighted based on the complaint rate, with different weights assigned according to the complaint rate. The weighted complaint text is then input into a pre-trained complaint type prediction model to determine the user's complaint type. This method, by weighting the complaint text, improves the accuracy of complaint type prediction. This method constructs a collection complaint corpus through part-of-speech tagging, word frequency analysis, and manual intervention (selecting words more relevant to collection work), which can be used for subsequent precise word segmentation. Weighting the complaint text according to the complaint rate and inputting the weighted complaint text into the pre-trained complaint type prediction model reduces the dimensionality of the model input, ensuring both accuracy and stability of the complaint type prediction model.

[0170] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0171] Based on the same inventive concept, this application also provides a complaint type prediction device for implementing the complaint type prediction method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more complaint type prediction device embodiments provided below can be found in the limitations of the complaint type prediction method described above, and will not be repeated here.

[0172] In one embodiment, such as Figure 6 As shown, a complaint type prediction device is provided, including: an acquisition module 602, a preprocessing module 604, a complaint rate prediction module 606, a weighting module 608, a processing module 610, and a complaint type prediction module 612, wherein:

[0173] Module 602 is used to obtain complaint information from target users;

[0174] Preprocessing module 604 is used to perform word segmentation on the complaint information to obtain the segmented complaint text;

[0175] The complaint rate prediction module 606 is used to input the segmented complaint text into a pre-trained complaint rate prediction model to obtain the feature vector of the complaint information, and to obtain the complaint rate of the complaint information based on the feature vector of the complaint information.

[0176] The weighting module 608 is used to weight the segmented complaint text based on the complaint rate of the complaint information to obtain the weighted complaint text.

[0177] Processing module 610 is used to input the weighted complaint text into a pre-trained complaint type prediction model to obtain a feature vector of the complaint type;

[0178] The complaint type prediction module 612 is used to obtain the complaint type of the target user based on the feature vector of the complaint type.

[0179] In one embodiment, the complaint type prediction device further includes a training module for acquiring a complaint sample training set; the complaint sample training set includes multiple complaint samples and labeled complaint frequencies; acquiring complaint samples from the complaint sample training set; performing word segmentation on the complaint samples to obtain segmented complaint samples; inputting the segmented complaint samples into a complaint rate prediction model to be trained to obtain feature vectors of the complaint samples; obtaining the predicted complaint frequency of the complaint samples based on the feature vectors of the complaint samples; and adjusting the parameters of the complaint rate prediction model based on the difference between the predicted complaint frequency and the labeled complaint frequency to obtain a trained complaint rate prediction model.

[0180] In one embodiment, the training module is further configured to obtain an initial sample training set consisting of historical complaint information; the initial complaint sample training set includes multiple complaint samples and labeled complaint frequencies; obtain related texts that are semantically similar to the complaint samples; use the labeled complaint frequencies of the complaint samples that are semantically similar to the related texts as the labeled complaint frequencies of the related texts; expand the complaint sample training set according to the related texts and labeled complaint frequencies to obtain the complaint sample training set.

[0181] In one embodiment, the weighting module 608 is further configured to determine the complaint weight based on the complaint rate of the complaint information; the value of the complaint rate is positively correlated with the complaint weight; the segmented complaint text is weighted based on the complaint weight to obtain the weighted complaint text.

[0182] In one embodiment, the weighting module 608 is further configured to determine the complaint severity level corresponding to the complaint rate based on the complaint rate of the complaint information; the complaint severity level includes at least two types; the higher the complaint rate, the higher the complaint severity level; and determine the complaint weight based on the complaint severity level, wherein the higher the complaint severity level, the greater the complaint weight.

[0183] In one embodiment, the preprocessing module 604 is further configured to perform word segmentation on the complaint information based on a word segmentation corpus in the field of financial debt collection, and obtain the segmented complaint text.

[0184] Each module in the aforementioned complaint type prediction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0185] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 7As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores complaint information data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When executed by the processor, the computer program implements a method for predicting complaint types.

[0186] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0187] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the following method steps:

[0188] Obtain complaint information from target users;

[0189] The complaint information is segmented into words to obtain the segmented complaint text.

[0190] The segmented complaint text is input into a pre-trained complaint rate prediction model to obtain the feature vector of the complaint information, and the complaint rate of the complaint information is obtained based on the feature vector of the complaint information.

[0191] Based on the complaint rate of the complaint information, the segmented complaint text is weighted to obtain the weighted complaint text.

[0192] The weighted complaint text is input into a pre-trained complaint type prediction model to obtain the feature vector of the complaint type.

[0193] The complaint type of the target user is obtained based on the feature vector of the complaint type.

[0194] In one embodiment, when the processor executes a computer program, it is used to acquire a complaint sample training set; the complaint sample training set includes multiple complaint samples and labeled complaint frequencies; acquire complaint samples from the complaint sample training set; perform word segmentation on the complaint samples to obtain segmented complaint samples; input the segmented complaint samples into a complaint rate prediction model to be trained to obtain feature vectors of the complaint samples; obtain the predicted complaint frequency of the complaint samples based on the feature vectors of the complaint samples; adjust the parameters of the complaint rate prediction model based on the difference between the predicted complaint frequency and the labeled complaint frequency of the complaint samples to obtain a trained complaint rate prediction model.

[0195] In one embodiment, when the processor executes a computer program, it is used to obtain an initial sample training set consisting of historical complaint information; the initial complaint sample training set includes multiple complaint samples and labeled complaint frequencies; obtain related texts that are semantically similar to the complaint samples; use the labeled complaint frequencies of the complaint samples that are semantically similar to the related texts as the labeled complaint frequencies of the related texts; expand the complaint sample training set according to the related texts and labeled complaint frequencies to obtain the complaint sample training set.

[0196] In one embodiment, when the processor executes a computer program, it determines the complaint weight based on the complaint rate of the complaint information; the value of the complaint rate is positively correlated with the complaint weight; and the segmented complaint text is weighted based on the complaint weight to obtain the weighted complaint text.

[0197] In one embodiment, when the processor executes a computer program, it is used to determine the complaint severity level corresponding to the complaint rate based on the complaint rate of the complaint information; the complaint severity level includes at least two types; the higher the complaint rate, the higher the complaint severity level; and to determine the complaint weight based on the complaint severity level, wherein the higher the complaint severity level, the greater the complaint weight.

[0198] In one embodiment, when the processor executes a computer program, it performs word segmentation on the complaint information based on a word segmentation corpus in the field of financial debt collection, and obtains the segmented complaint text.

[0199] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the following method steps:

[0200] Obtain complaint information from target users;

[0201] The complaint information is segmented into words to obtain the segmented complaint text.

[0202] The segmented complaint text is input into a pre-trained complaint rate prediction model to obtain the feature vector of the complaint information, and the complaint rate of the complaint information is obtained based on the feature vector of the complaint information.

[0203] Based on the complaint rate of the complaint information, the segmented complaint text is weighted to obtain the weighted complaint text.

[0204] The weighted complaint text is input into a pre-trained complaint type prediction model to obtain the feature vector of the complaint type.

[0205] The complaint type of the target user is obtained based on the feature vector of the complaint type.

[0206] In one embodiment, when the computer program is executed by the processor, it is used to acquire a complaint sample training set; the complaint sample training set includes multiple complaint samples and labeled complaint frequencies; acquire complaint samples from the complaint sample training set; perform word segmentation on the complaint samples to obtain segmented complaint samples; input the segmented complaint samples into a complaint rate prediction model to be trained to obtain the feature vector of the complaint samples; obtain the predicted complaint frequency of the complaint samples based on the feature vector of the complaint samples; adjust the parameters of the complaint rate prediction model based on the difference between the predicted complaint frequency and the labeled complaint frequency of the complaint samples to obtain a trained complaint rate prediction model.

[0207] In one embodiment, when the computer program is executed by the processor, it is used to obtain an initial sample training set consisting of historical complaint information; the initial complaint sample training set includes multiple complaint samples and labeled complaint frequencies; obtain related texts that are semantically similar to the complaint samples; use the labeled complaint frequencies of the complaint samples that are semantically similar to the related texts as the labeled complaint frequencies of the related texts; expand the complaint sample training set according to the related texts and labeled complaint frequencies to obtain the complaint sample training set.

[0208] In one embodiment, when the computer program is executed by the processor, it is used to determine the complaint weight based on the complaint rate of the complaint information; the value of the complaint rate is positively correlated with the complaint weight; the segmented complaint text is weighted based on the complaint weight to obtain the weighted complaint text.

[0209] In one embodiment, when the computer program is executed by the processor, it is used to determine the complaint severity level corresponding to the complaint rate based on the complaint rate of the complaint information; the complaint severity level includes at least two types; the higher the complaint rate, the higher the complaint severity level; and to determine the complaint weight based on the complaint severity level, wherein the higher the complaint severity level, the greater the complaint weight.

[0210] In one embodiment, when a computer program is executed by a processor, it is used to segment complaint information into words based on a segmentation corpus in the field of financial debt collection, thereby obtaining segmented complaint text.

[0211] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the following method steps:

[0212] Obtain complaint information from target users;

[0213] The complaint information is segmented into words to obtain the segmented complaint text.

[0214] The segmented complaint text is input into a pre-trained complaint rate prediction model to obtain the feature vector of the complaint information, and the complaint rate of the complaint information is obtained based on the feature vector of the complaint information.

[0215] Based on the complaint rate of the complaint information, the segmented complaint text is weighted to obtain the weighted complaint text.

[0216] The weighted complaint text is input into a pre-trained complaint type prediction model to obtain the feature vector of the complaint type.

[0217] The complaint type of the target user is obtained based on the feature vector of the complaint type.

[0218] In one embodiment, when the computer program is executed by the processor, it is used to acquire a complaint sample training set; the complaint sample training set includes multiple complaint samples and labeled complaint frequencies; acquire complaint samples from the complaint sample training set; perform word segmentation on the complaint samples to obtain segmented complaint samples; input the segmented complaint samples into a complaint rate prediction model to be trained to obtain the feature vector of the complaint samples; obtain the predicted complaint frequency of the complaint samples based on the feature vector of the complaint samples; adjust the parameters of the complaint rate prediction model based on the difference between the predicted complaint frequency and the labeled complaint frequency of the complaint samples to obtain a trained complaint rate prediction model.

[0219] In one embodiment, when the computer program is executed by the processor, it is used to obtain an initial sample training set consisting of historical complaint information; the initial complaint sample training set includes multiple complaint samples and labeled complaint frequencies; obtain related texts that are semantically similar to the complaint samples; use the labeled complaint frequencies of the complaint samples that are semantically similar to the related texts as the labeled complaint frequencies of the related texts; expand the complaint sample training set according to the related texts and labeled complaint frequencies to obtain the complaint sample training set.

[0220] In one embodiment, when the computer program is executed by the processor, it is used to determine the complaint weight based on the complaint rate of the complaint information; the value of the complaint rate is positively correlated with the complaint weight; the segmented complaint text is weighted based on the complaint weight to obtain the weighted complaint text.

[0221] In one embodiment, when the computer program is executed by the processor, it is used to determine the complaint severity level corresponding to the complaint rate based on the complaint rate of the complaint information; the complaint severity level includes at least two types; the higher the complaint rate, the higher the complaint severity level; and to determine the complaint weight based on the complaint severity level, wherein the higher the complaint severity level, the greater the complaint weight.

[0222] In one embodiment, when a computer program is executed by a processor, it is used to segment complaint information into words based on a segmentation corpus in the field of financial debt collection, thereby obtaining segmented complaint text.

[0223] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0224] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0225] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method of predicting a complaint type, characterized by, The method comprises: obtaining complaint information of a target user; performing word segmentation processing on the complaint information to obtain segmented complaint text; inputting the segmented complaint text into a pre-trained complaint rate prediction model to obtain a feature vector of the complaint information, and calculating a complaint rate of the complaint information by comparing the feature vector of the complaint information with feature vectors of other complaint information in a complaint text corpus; determining a complaint weight according to the complaint rate of the complaint information; the value of the complaint rate is positively correlated with the complaint weight; weighting the segmented complaint text based on the complaint weight to obtain weighted complaint text; inputting the weighted complaint text into a pre-trained complaint type prediction model to obtain a feature vector of a complaint type; obtaining a complaint type of the target user according to the feature vector of the complaint type.

2. The method of claim 1, wherein, The method for training the complaint rate prediction model comprises: obtaining a complaint sample training set; the complaint sample training set comprises a plurality of complaint samples and labeled complaint frequencies; obtaining a complaint sample from the complaint sample training set; performing word segmentation processing on the complaint sample to obtain segmented complaint sample text; inputting the segmented complaint sample text into a to-be-trained complaint rate prediction model to obtain a feature vector of the complaint sample, and obtaining a predicted complaint frequency of the complaint sample according to the feature vector of the complaint sample; adjusting parameters of the complaint rate prediction model according to differences between the predicted complaint frequency of the complaint sample and the labeled complaint frequency to obtain a trained complaint rate prediction model.

3. The method of claim 2, wherein, The method for obtaining the complaint sample training set comprises: obtaining an initial sample training set composed of historical complaint information; the initial sample training set comprises a plurality of complaint samples and labeled complaint frequencies; obtaining associated text semantically similar to the complaint sample; taking the labeled complaint frequency of the complaint sample semantically similar to the associated text as the labeled complaint frequency of the associated text; expanding the initial sample training set according to the associated text and the labeled complaint frequency to obtain a complaint sample training set.

4. The method of claim 1, wherein, Determining a complaint weight according to a complaint rate of complaint information comprises: determining a complaint degree level corresponding to the complaint rate according to the complaint rate of the complaint information; the complaint degree level comprises at least two levels; the greater the value of the complaint rate, the higher the complaint degree level; determining a complaint weight according to the complaint degree level, wherein the higher the complaint degree level, the greater the complaint weight.

5. The method of claim 1, wherein, The method for performing word segmentation processing on the complaint information to obtain segmented complaint text comprises: performing word segmentation processing on the complaint information according to a word segmentation corpus in the field of financial collection to obtain segmented complaint text.

6. A complaint type prediction device characterized by comprising: The device comprises: an obtaining module configured to obtain complaint information of a target user; a preprocessing module configured to perform word segmentation processing on the complaint information to obtain segmented complaint text; and The complaint rate prediction module is configured to input the segmented complaint text into a pre-trained complaint rate prediction model to obtain a feature vector of the complaint information, and to calculate a complaint rate of the complaint information by comparing the feature vector of the complaint information with feature vectors of other complaint information in a complaint text corpus. The weighting module is configured to determine a complaint weight according to the complaint rate of the complaint information, wherein the value of the complaint rate is positively correlated with the complaint weight. The complaint text is weighted based on the complaint weight to obtain a weighted complaint text. The processing module is configured to input the weighted complaint text into a pre-trained complaint type prediction model to obtain a feature vector of a complaint type. The complaint type prediction module is configured to obtain the complaint type of the target user according to the feature vector of the complaint type.

7. The apparatus of claim 6, wherein, The device further comprises a training module configured to: obtain a complaint sample training set, wherein the complaint sample training set comprises a plurality of complaint samples and labeled complaint frequencies; obtain a complaint sample from the complaint sample training set; segment the complaint sample to obtain a segmented complaint sample; input the segmented complaint sample into a to-be-trained complaint rate prediction model to obtain a feature vector of the complaint sample, and obtain a predicted complaint frequency of the complaint sample according to the feature vector of the complaint sample; adjust parameters of the complaint rate prediction model according to a difference between the predicted complaint frequency of the complaint sample and the labeled complaint frequency to obtain a trained complaint rate prediction model.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to implement the steps of the method in any one of claims 1 to 5.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method in any one of claims 1 to 5.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method in any one of claims 1 to 5.

Images