Collection dispatching method and device based on multi-person cooperation mode, equipment and medium

Abstract

The invention relates to the technical field of artificial intelligence technology and financial science and technology, and discloses a collection dispatching method and device based on a multi-person cooperation mode, equipment and a medium, and the method comprises the steps: obtaining an insurance policy recording, and converting the insurance policy recording into a recording text; utilizing a preset attention model to extract a high-level semantic feature vector of the recording text; constructing a feature vector according to the semantic feature, and performing matching calculation on the feature vector and a preset intention vector table to obtain an insurance policy intention; and obtaining a task saturation level of the communication personnel, generating a cooperative order distribution strategy according to the communication personnel saturation level and the insurance policy intention, and performing insurance policy distribution according to the cooperative order distribution strategy. And the insurance policy service processing efficiency and accuracy are effectively improved.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence technology and financial technology, and in particular to a method, device, equipment and medium for debt collection and dispatching based on a multi-person collaboration mode. Background Art

[0002] As insurance companies develop, the number of existing policyholders increases, and the annual renewal payment becomes more and more important. At present, the common practice in the industry is for agents or agency supervisors to be responsible for special follow-up. Before the end of each month, the list of customers whose policies are about to expire in the next month is exported and assigned to the corresponding agents and supervisors for follow-up.

[0003] The traditional order dispatching rules are mainly based on the agent or corresponding agency that signed the order with the customer. Therefore, some agents or supervisors may have a large number of renewal payment customers who need to follow up every month. Due to the overload of workload, the agents may not be able to contact and inform the customers in advance, resulting in customers forgetting to pay or even canceling the insurance in the end.

[0004] At the same time, considering that the customer's renewal payment period generally lasts from one month before the due date to two months after the due date (i.e. the grace period), different customers need to be reminded at different frequencies during these three months. Therefore, a customer may need to be contacted and reminded multiple times, further increasing the workload of agents or supervisors. Summary of the invention

[0005] The present invention provides an intelligent collection dispatching method, device, computer equipment and medium based on a multi-person collaboration mode to solve the problem that agents are overloaded with workload and cannot contact and inform customers in advance, thereby reducing the problem of excessive workload of agents.

[0006] In the first aspect, a method for collecting and dispatching debts based on a multi-person collaboration mode is provided, including:

[0007] Obtaining the policy recording, and converting the policy recording into a recording text;

[0008] Extracting a high-level semantic feature vector of the recorded text using a preset attention model;

[0009] Matching and calculating the high-level semantic feature vector with a preset intent vector table to obtain the policy intent;

[0010] The task saturation level of the communication personnel is obtained, a collaborative dispatching strategy is generated according to the communication personnel saturation level and the policy intention, and the policy is dispatched according to the collaborative dispatching strategy.

[0011] In a second aspect, a debt collection dispatching device based on a multi-person collaboration mode is provided, comprising:

[0012] The recording acquisition module is used to acquire the policy recording;

[0013] A conversion module, used to convert the policy recording into a recording text;

[0014] An extraction module, used to extract a high-level semantic feature vector of the recorded text using a preset attention model;

[0015] A matching calculation module, used for performing matching calculation on the high-level semantic feature vector and a preset intent vector table to obtain the policy intent;

[0016] A module for obtaining saturation levels is used to obtain the task saturation levels of communication personnel;

[0017] A generation module, used for generating a collaborative dispatching strategy according to the communication personnel saturation level and the insurance policy intention;

[0018] The distribution module is used to distribute insurance policies according to the collaborative distribution strategy.

[0019] In a third aspect, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the above-mentioned method for debt collection and dispatching based on a multi-person collaborative mode are implemented.

[0020] In a fourth aspect, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps of the above-mentioned method for debt collection and dispatching based on a multi-person collaborative mode are implemented.

[0021] In the above-mentioned scheme implemented by the method, device, computer equipment and storage medium for debt collection and dispatching based on a multi-person collaborative mode, obtaining the insurance policy recording can reduce the time of remote audio transmission during analysis, speed up the analysis, and make it easier to accurately analyze the voice of each unit, while avoiding problems such as network instability during analysis and data loss during transmission. By converting the recording into text, the content can be clearly presented on the screen, which is easy to organize and manage, and will facilitate subsequent analysis and processing; the semantic features of the recording text are extracted through the attention model, and the key parts of the text are selectively focused on according to the importance of different parts, so that the model is more effective in processing long texts and can effectively extract individual parts. Humanized language features and expressions, precise customer needs, by matching high-level semantic feature vectors and preset intent vector tables, can more accurately classify policy intent, thereby improving the accuracy of policy intent recognition, and generating dispatch strategies based on policy intent can more accurately grasp the core information and specific requirements of the policy, so as to handle the policy more finely, improve the accuracy of policy processing, and realize policy processing automation, thereby saving a lot of manual review and processing time, reducing labor costs, and processing policies faster and more accurately, thereby shortening policy processing time, improving work efficiency, and understanding customer needs more finely, while being able to quickly respond to customer needs and further improve customer service quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0023] Figure 1 This is a schematic diagram of an application environment of a method for collecting and dispatching debts based on a multi-person collaboration mode in one embodiment of the present invention;

[0024] Figure 2 It is a flow chart of a method for collecting and dispatching work based on a multi-person collaboration mode in one embodiment of the present invention;

[0025] Figure 3 It is a structural schematic diagram of a debt collection and dispatching device based on a multi-person collaboration mode in one embodiment of the present invention;

[0026] Figure 4 yes Figure 1 A schematic flow chart of a specific implementation of step S1;

[0027] Figure 5 is a structural schematic diagram of a computer device in one embodiment of the present invention;

[0028] Figure 6 It is another structural schematic diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0030] The embodiment of the present invention provides a method for collecting and dispatching debts based on a multi-person collaboration mode, which can be applied in the following aspects: Figure 1 In the application environment, the client communicates with the server through the network. The server can obtain the policy recording and convert the policy recording into a recording text; use the preset attention model to extract the high-level semantic feature vector of the recording text; match and calculate the high-level semantic feature vector with the preset intention vector table to obtain the policy intention; obtain the task saturation level of the communication personnel, generate a collaborative dispatching strategy according to the communication personnel saturation level and the policy intention, and distribute the policy according to the collaborative dispatching strategy, and feed the dispatching strategy back to the client. The present invention provides a collection and dispatching device based on a multi-person collaborative mode. For the heavy policy business, the policy recording is first obtained and converted into a recording text, and then a scheme of extracting the high-level semantic feature vector of the recording text by the preset attention model is proposed. Through the result mapping method, a collaborative dispatching strategy is generated according to the communication personnel saturation level, policy intention, and policy cycle level, and the policy is distributed according to the collaborative dispatching strategy, which can effectively reduce the workload of the staff to avoid the situation of missing customers, greatly and efficiently improve the effect of policy processing, and reduce labor costs at the same time. The client may be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, and portable wearable devices. The server may be implemented by an independent server or a server cluster consisting of multiple servers. The present invention is described in detail below through specific embodiments.

[0031] See also Figure 2 As shown, Figure 2 A flowchart of a method for collecting and dispatching debts based on a multi-person collaboration mode provided by an embodiment of the present invention includes the following steps:

[0032] S1. Obtain the insurance policy recording, and convert the insurance policy recording into an audio text.

[0033] In the example of the present invention, the policy recording refers to the call recording of the policy. When the policy is assigned to a follow-up person, the follow-up person needs to have a call with the customer of the policy, and the recording during the call becomes the policy recording; the recording text refers to the policy recording converted into readable text information.

[0034] In detail, the insurance policy recording is converted into readable text using speech recognition technology, and the converted text is processed to remove useless information, punctuation marks, word abbreviations, repeated content, spelling errors, etc. to obtain a clean version of the readable text, and then the clean version of the readable text is separated into some individual words to obtain the initial insurance policy recording text.

[0035] In the embodiment of the present invention, the obtaining of the insurance policy recording and converting the insurance policy recording into an audio recording text includes:

[0036] Obtaining the policy information of a preset existing customer group, screening the policy information according to a preset term range, and obtaining target policy information;

[0037] Allocating the target insurance policy information to preset communication personnel according to preset priorities;

[0038] Collecting the policy recordings of the communication personnel communicating with the existing customer group based on the target policy information;

[0039] The insurance policy recording is converted into text to obtain a recording text.

[0040] In an embodiment of the present invention, the preset time range refers to the date range on which the policies of the existing customer group are about to expire, the target policy information refers to the policies within the preset time range, the screening refers to screening out the policies of the existing customer group that are about to expire, and the preset priority refers to arranging the preset personnel who handle the assigned policies with reference to the rules of the relationship with the former agent (former handler) from near to far, and the people who have a relationship with the former agent from near to far are set in order from the former agent himself, agents who jointly develop business, other agents in the same supervisor / area, supervisors in the same area, and random personnel in the same secondary institution; the communication personnel refers to employees who handle policy-related business, and the text conversion of the recording refers to inputting the recording into a deep learning model or other audio-to-text model to output text information corresponding to the content of the recording.

[0041] Specifically, when the rule engine model is used to determine the correspondence personnel to whom the insurance policy is assigned, the hierarchy of the correspondence personnel to whom the insurance policy is assigned is first constructed based on the relationship from near to far. For example, the people who have close relationships with the former agent from near to far are the former agent himself, agents who jointly develop business, other agents with the same supervisor / area, supervisors in the same area, and random personnel in the same secondary agency. By determining whether personnel at all levels are in office, for example, if the former agent is in office, the policy is still assigned to the former agent; if the former agent resigns, the policy is assigned to the agent who jointly develops business, and so on, the correspondence personnel to whom the insurance policy is assigned is determined.

[0042] In the embodiment of the present invention, by allocating the existing insurance policies to the communication personnel according to the preset priority and classifying and allocating the insurance policies based on the preset rules, it is possible to avoid errors caused by human subjectivity and uncertainty, improve the accuracy and consistency of decision-making, and can be adjusted and optimized according to actual needs, thereby ensuring the controllability and sustainability of the decision-making process.

[0043] In detail, obtaining policy recordings can reduce the time for remote audio transmission during analysis, speed up the analysis, and make it easier to accurately analyze the voice of each unit, while avoiding problems such as network instability during analysis and data loss during transmission. It ensures that policy recordings can be analyzed without being affected by network limitations and time zone differences, reduces traffic costs and fees when transmitting data with cloud service providers, and converts recordings into text so that the content can be clearly presented on the screen, making it easier to organize and manage, and will facilitate subsequent analysis and processing.

[0044] S2. Using a preset attention model to extract a high-level semantic feature vector of the recorded text.

[0045] In an embodiment of the present invention, the attention model is a model that identifies recorded text information and outputs the intention expressed by the recorded text. The semantic feature refers to the semantic information in the recorded text that more accurately expresses the text.

[0046] In an embodiment of the present application, an attention mechanism network is used to assign different weights to each word vector in the semantic feature vector output in the previous stage according to their different degrees of semantic representation.

[0047] In an embodiment of the present invention, the step of extracting a high-level semantic feature vector of the recorded text using a preset attention model includes:

[0048] Performing word segmentation processing on the recorded text according to a preset word segmentation method to obtain text segmentation;

[0049] Performing vector conversion on the text segmentation to obtain text segmentation vectors, and splicing the text segmentation vectors into a text vector;

[0050] Extracting semantic features from the text vector to obtain a semantic feature vector;

[0051] Performing attention distribution calculation on the semantic feature vector according to the preset attention model to obtain a vector weight of each text segmentation vector in the semantic feature vector;

[0052] The text word segmentation vectors are weighted according to the vector weights to obtain a high-level semantic feature vector.

[0053] In an embodiment of the present invention, the word segmentation processing is to divide the recorded text into multiple words, and the text vector refers to the set of multiple word vectors after the text is divided into multiple words and the words are represented by vectors. In an embodiment of the present invention, the attention mechanism refers to a model that assigns attention weights to the word vectors of the input text vector, and the semantic feature refers to the feature extracted after analyzing the recorded text and can more accurately express the semantic information of the text.

[0054] In detail, you first need to load a word segmentation dictionary as a reference for the word segmentation algorithm, perform forward maximum matching based on the dictionary, traverse the text from left to right, and each time find a word with the longest dictionary match, output it as the word segmentation result and delete it from the text. Repeat this operation until there are no characters in the text, and get all the word segmentation results of the recorded text.

[0055] In an embodiment of the present invention, a word embedding model is used to convert the segmented words into vector representations. For example, a word embedding model is usually used for representation (such as Word2Vec, GloVe and other models). The type of word vector is set according to the needs, such as vector dimension, vocabulary size and other parameters. For example, the sentence "I am willing to pay insurance premiums" needs to be represented by a vector. First, based on this text, a vocabulary is constructed, that is, Vocabulary = {1. "I", 2. "willing", 3. "pay", 4. "insurance premium", 5. "he", 6. "unwilling"}; this table contains a total of 6 different words. Using the index number of the vocabulary, the above text can be represented by a 6-dimensional vector (the elements in the vector are the frequencies of the words in the vocabulary appearing in the text), that is, "I am willing to pay insurance premiums" is represented as [1, 1, 1, 1, 0, 0].

[0056] In the embodiment of the present invention, the text vector is subjected to semantic feature extraction by a bidirectional long short-term memory network (LSTM) to obtain a semantic feature vector. The bidirectional long short-term memory network in the embodiment of the present application can be a three-layer bidirectional cyclic long short-term memory network, which is used as a text feature extractor to extract the semantic feature vector of the text vector and obtain the semantic feature vector; first, the input text sequence is converted into a word vector sequence through an embedding layer, and the word vector sequence is input into three bidirectional LSTM layers, and forward calculation and backward calculation are performed respectively. In each bidirectional LSTM layer, the forward calculation and backward calculation will output a hidden state vector. These two vectors are spliced ​​together to obtain the output vector of the current layer; the output vector of the current layer is input into the next bidirectional LSTM layer for further processing; the output vector of the last bidirectional LSTM layer is the semantic feature vector of the text vector, and the output vectors of the three LSTM layers are connected together through a concatenation operation to obtain the final semantic feature vector of the text vector.

[0057] In the embodiment of the present invention, the attention distribution calculation is performed on the semantic feature vector based on the preset attention model to obtain the weight of each word segmentation vector in the semantic feature vector, including:

[0058] According to the attention model, each word segmentation vector in the semantic feature vector is scored with the pre-acquired query vector to obtain the attention score of each word segmentation vector in the semantic feature vector;

[0059] The scored word segmentation vectors are weighted according to the Softmax function to obtain the weight of each word segmentation vector in the semantic feature vector.

[0060] In the example of the present invention, the query vector is a set of dynamic vectors, which are mainly vectors dynamically adjusted according to the word segmentation vectors. The score calculation refers to the process of obtaining the attention score of each word segmentation vector in the text vector by treating the query vector and each word segmentation vector in the text vector as the product of two linear transformations through the attention model.

[0061] The embodiment of the present application uses the attention model to assign different weights to each word vector in the semantic feature vector output in the previous stage according to their different degrees of semantic representation. The overall process is divided into two stages. First, the correlation between the semantic feature information of each word in the text and the text representation is calculated to obtain the attention distribution of the text vector; then, based on the attention distribution, the weight of each word vector in the semantic feature vector corresponding to the text vector is calculated. The calculation formula for this stage is as follows:

[0062]

[0063] x t=tanh(W x H x +b x );

[0064]

[0065] In the formula, is the semantic feature vector corresponding to the text vector obtained in the previous stage, h x is the word vector in the semantic feature vector, W x , b x is the network parameter of the attention mechanism, x t is the total attention distribution of the text vector, α t is the weight of the tth word vector in the semantic feature vector, is the weighted word vector in the semantic feature vector, and finally the high-level semantic feature vector corresponding to the text vector is obtained

[0066]

[0067] In detail, the attention model takes the query vector and each word vector in the text vector as the product of two linear transformations to obtain the attention distribution. The specific formula is as follows:

[0068] score(q,k i )=q*k i ;

[0069] In the formula, score(q,k i ) represents the attention score of the word vector, q represents the query vector, and k i Represents one of the word vectors in the text vector.

[0070] Specifically, the formula for weighting the attention score of the word vector based on the Softmax function is as follows:

[0071]

[0072] In the formula, α i Represents the weight of the i-th word vector in the text vector.

[0073] In the embodiment of the present invention, weighting the corresponding word segmentation vector according to the weight of each word segmentation vector in the text vector to obtain a high-level semantic feature vector includes:

[0074] Taking a weighted average of each word segmentation vector in the text and the corresponding vector weight, to obtain a weighted vector for each word segmentation in the text;

[0075] Performing a pooling operation on each word segmentation weighted vector in the text to obtain each word segmentation pooling vector in the text;

[0076] Each word pooling vector in the text is combined to generate a high-level semantic feature vector.

[0077] In an embodiment of the present invention, the weight of the word segmentation vector represents the numerical value of the importance of the word segmentation in the text, the word segmentation weighted vector refers to the weight of the price word segmentation vector further amplifying the numerical value representing the importance of the word segmentation in the text, and the word segmentation pooling vector refers to the vector representation after taking the maximum value of the word segmentation weighted vector on each dimension.

[0078] Specifically, each word segmentation vector in the text is weighted averaged with the weight of the corresponding word segmentation vector to obtain the formula for each word segmentation weighted vector in the text as follows:

[0079]

[0080] In the formula, v i represents the i-th word segmentation vector, w i represents the weight corresponding to the i-th word segmentation vector, n represents the total number of word segments in the text, and S represents the i-th word segmentation weighted vector in the text.

[0081] In detail, each word segmentation weighted vector is regarded as an independent pooling area, and the maximum value is taken in each pooling area dimension; specifically, the word segmentation weighted vector set is {v1, w1}, {v2, w2}... {vn, wn}, and a maximum value is selected from each set as v_max. After maximum pooling, the original word segmentation vector set is compressed into a fixed-length vector v_max. This new vector v_max can be used as an input feature for subsequent tasks, such as classification, regression, or other natural language processing tasks.

[0082] In the embodiment of the present invention, after the recorded text is segmented, each segmented word can be regarded as corresponding to a word, which is convenient for text processing and analysis. At the same time, the segmented text can be used to train models such as text classification and sentiment analysis, thereby constructing a more accurate and efficient data set. The segmented word vectors in the text are used as input, which can improve the accuracy of the model in various natural language processing tasks, such as text classification, sentiment analysis, and information retrieval, and can reduce the subsequent calculation complexity and improve the training efficiency and prediction speed of the model.

[0083] At the same time, the attention model can select key parts in the word segmentation vector for strengthening, while ignoring other irrelevant parts, so as to better capture the semantic information in the text and improve the accuracy of the model. The weights of different parts in the word segmentation vector can be assigned to adapt to different input lengths without fixing the input length in the model. The use of weighted word segmentation vectors can enhance the role of important word segmentations and improve the accuracy of feature representation.

[0084] In the example of the present invention, when using the attention model to extract the semantic features of the recorded text, it is possible to selectively focus on the key parts of the text according to the importance of different parts, rather than processing all the information together. This makes the model more effective when processing long texts, and can effectively extract personalized language features and expressions, thereby better providing users with customized services and personalized experiences.

[0085] S3. Match and calculate the high-level semantic feature vector with a preset intent vector table to obtain the policy intent.

[0086] In an embodiment of the present invention, a high-level semantic feature vector refers to a vector used to represent text features in natural language processing, a preset intention vector table refers to a vector table with preset features as column labels and results corresponding to the preset features as row labels, the matching calculation refers to calculating the similarity between each feature result corresponding to each preset feature in the intention table and the feature vector obtained in the previous stage one by one, and the policy intention refers to the parsed user's intention.

[0087] In the example of the present invention, the high-level semantic feature vector is matched and calculated with a preset intent vector table to obtain the policy intent, including:

[0088] Extracting preset features in the intention vector table and feature results corresponding to the preset features;

[0089] Extracting features from the feature results one by one to obtain a feature vector of each feature result corresponding to a preset feature;

[0090] Select each feature vector corresponding to each preset feature one by one, and perform similarity calculations on the selected feature vectors and the high-level semantic feature vector one by one to obtain a similarity prediction value between each feature result corresponding to the preset feature in the intent table and the high-level semantic feature vector;

[0091] Each feature result with the highest similarity prediction value corresponding to the high-level semantic feature vector is selected one by one, and the selected feature results are merged into the policy intention.

[0092] In an embodiment of the present invention, the feature extraction refers to repeating the above steps for each feature result corresponding to each preset feature in the intention vector table to convert each feature result into a feature vector identical to the high-level semantic feature vector, the similarity calculation refers to calculating the similarity between each feature result corresponding to each preset feature in the intention vector table and the high-level semantic feature vector obtained in the previous stage one by one, and the similarity prediction value refers to the value obtained after the above similarity calculation.

[0093] In the example of the present invention, for example, four groups of preset features are set in the intent vector table, namely reminder notification, payment willingness, next reminder appointment, and contact information, that is, column labels, among which the feature results corresponding to the preset feature for the reminder notification are divided into clear, unclear, and unclear, that is, labels; the feature results corresponding to the preset feature for the payment willingness are divided into willing to pay, unwilling to pay, and unclear; the feature results corresponding to the preset feature for the next reminder appointment are 7 days, 14 days, and pending; the feature results corresponding to the preset feature for the contact information are telephone, WeChat, face-to-face interview, and pending.

[0094] In the embodiment of the present invention, the Euclidean distance algorithm can be used to calculate the distance between the feature vector of each feature result and the high-level semantic feature vector. Through the above steps, the feature vector of a single feature result is expressed as The formula for calculating the two using the Euclidean distance algorithm is as follows

[0095]

[0096] In the formula, The Euclidean distance between the feature vector representing a feature result and the high-level semantic feature vector, represents the i-th vector value in the high-level semantic feature vector, The i-th vector value in the feature vector representing a feature result.

[0097] For example, the Euclidean distance between the high-level semantic feature vector and the feature result corresponding to the willingness to pay is 0.3, the Euclidean distance between the feature result corresponding to the willingness to pay is 0.4, and the Euclidean distance between the feature result corresponding to the willingness to pay is unclear is 0.7; the threshold is set to 0.6, and the intention below the threshold is considered to be matched, and the intention above the threshold is considered to be unmatched. The smaller the distance, the higher the similarity. That is, there are two situations for willingness to pay and unwillingness for the high-level semantic feature vector that matches the high-level semantic feature vector. Then, the results are mapped through normalization. Assuming that the high-level semantic The similarity between the feature vector and the willingness to pay is 0.9, and the similarity between the high-level semantic feature vector and the unwillingness to pay is 0.7. The higher value is taken as the final result (the feature result similarity value with a value lower than 0.6 is deleted), that is, the willingness to pay expressed in the high-level semantic feature vector is willing to pay. Similarly, the intention of reminder notification, next reminder appointment, and contact information expressed in the high-level semantic feature vector can be obtained; finally, the similarity values ​​of the feature results selected in the preset features are added, and the arithmetic mean is calculated as the confidence score of this insurance policy recording.

[0098] In the embodiment of the present invention, by using deep learning technology, more advanced and abstract semantic feature vectors are obtained, so that the semantic feature vectors have stronger expressibility and discriminability, which can significantly improve the accuracy and performance of the model; and, by predicting the similarity between the semantic feature vectors and the feature results corresponding to the preset features, the credibility of the insurance policy recordings can be better predicted. At the same time, the insurance policy recordings can be preliminarily screened and classified, avoiding the need for manual review of all insurance policy recordings one by one, thereby reducing the workload of manual review and improving work efficiency.

[0099] In an embodiment of the present invention, by matching high-level semantic feature vectors with preset intent vector tables, policy intent can be more accurately classified, thereby improving the accuracy of policy intent recognition. Matching calculations using high-level semantic feature vectors and intent vector tables can reduce the dependence of the intent recognition model on specific vocabulary and language structures, thereby enhancing the robustness of the model. At the same time, by using vector matching methods, these classification results can also be migrated to new data for classification, thereby improving the generalization ability of the model.

[0100] S4. Obtain the task saturation level of the communication personnel, generate a collaborative dispatching strategy according to the communication personnel saturation level and the insurance policy intention, and dispatch the insurance policy according to the collaborative dispatching strategy.

[0101] In an embodiment of the present invention, the communication personnel saturation level refers to a plurality of groups of saturation status levels obtained by comparing the number of insurance policies currently followed up by the insurance policy communication personnel and the number when the average number of insurance policies followed up per capita in the communication personnel's work organization is multiplied by a base number; a table consisting of a plurality of groups of levels.

[0102] In the example of the present invention, the step of obtaining the task saturation level of the communication personnel includes:

[0103] Obtain the number of follow-up policies per capita within a preset period, and perform a change operation on the number of follow-up policies per capita and a preset value to obtain the number of policy changes;

[0104] The result of comparing the number of policy changes with the number of policies currently followed up by the communication personnel is mapped to a pre-acquired saturation level table to obtain the communication personnel saturation level.

[0105] In an embodiment of the present invention, the saturation level table refers to a plurality of groups of saturation status levels obtained by comparing the number of policies currently followed up by the policy communication personnel and the number when the average number of policies followed up per capita in the communication personnel's work organization is multiplied by a base number; the table is composed of a plurality of groups of levels, and the collaborative dispatching strategy refers to freely combining and matching the characteristic results of the policy difficulty level, the policy collection cycle level, and the preset personnel task saturation, and setting an dispatching strategy for each combination.

[0106] In the embodiment of the present invention, the preset features in the saturation level table are set as task saturation, and the feature results corresponding to the preset features are high saturation, medium saturation, and low saturation respectively; the number of insurance policies to be followed up by the communication personnel is set to N, and the average number of insurance policies followed up per capita in the communication personnel's work organization last month is M;

[0107] When N≥M*80%, the current task saturation of the preset personnel is in a high saturation state;

[0108] When M*30%≤N<M*80%, the current task saturation of the preset personnel is in the medium saturation state;

[0109] When N<M*30%, the current task saturation of the preset personnel is in a low saturation state.

[0110] In the example of the present invention, generating a collaborative dispatching strategy according to the communication personnel saturation level and the insurance policy intention, and dispatching the insurance policy according to the collaborative dispatching strategy includes:

[0111] Divide the current policy period according to the preset rules to obtain the policy period result;

[0112] Mapping the policy cycle result to the pre-acquired policy urging cycle grade table to obtain the policy urging cycle grade;

[0113] Mapping the policy intention to a pre-acquired policy difficulty level table to obtain a policy difficulty level;

[0114] The result of the combination of the policy difficulty level, communication personnel saturation level, and policy collection cycle level is mapped to a pre-acquired dispatch strategy table, a collaborative dispatch strategy is generated based on the dispatch strategy table, and the policy is dispatched according to the collaborative dispatch strategy.

[0115] In an embodiment of the present invention, the insurance policy difficulty level table refers to a table consisting of multiple groups of levels, in which a combination of feature results corresponding to preset features in the intent table is matched, and a level is set for each combination. The insurance policy collection cycle level table refers to a table consisting of a period status corresponding to the time from the policy expiration date to the preset number of days. The collaborative dispatching strategy refers to a free combination and matching of the feature results of the insurance policy difficulty level, the insurance policy collection cycle level, and the preset personnel task saturation, and a person to be dispatched is set for each combination.

[0116] In an embodiment of the present invention, the preset features in the collection cycle level table are set as collection cycles, and the feature results corresponding to the preset features are respectively the stable policy period, the due date, the grace period, and the suspension period; the policy expiration date is represented as T; when the current policy date is within [T-30, T-1], it is the stable policy period, when the current policy date is within [T, T+1], it is the due date, when the current policy date is within [T+2, T+60], it is the grace period, and when the current policy date is within [T+61,) it is the suspension period.

[0117] In the embodiment of the present invention, the preset features in the difficulty level table are set as difficulty scores, and the feature results corresponding to the preset features are A for easy collection, B for difficult collection, C for pending follow-up, and D for difficult;

[0118] When the final characteristic result of the reminder notification is clear and the final characteristic result of the willingness to pay is willing, this level is rated as A, which is easy to collect;

[0119] When the final characteristic result of the reminder notification is clear, and the final characteristic result of the willingness to pay is unwilling, this level is rated as D, difficult;

[0120] When the final characteristic result of the reminder notification is clear, and the final characteristic result of the willingness to pay is unclear, this level is rated as C and awaits follow-up;

[0121] When the final characteristic result of the reminder notification is unclear and the final characteristic result of the willingness to pay is unwilling, this level is rated as D, difficult;

[0122] When the final characteristic result of the reminder notification is unclear and the final characteristic result of the willingness to pay is unclear, this level is rated as C and awaits follow-up;

[0123] When the final characteristic result of the reminder notification is unclear and the final characteristic result of the willingness to pay is willing, the level is rated as B, which means it is difficult to collect.

[0124] In the embodiment of the present invention, the preset features in the dispatch strategy table are set as dispatch strategies, and the feature results corresponding to the preset features are respectively transferring to supervisory collection and the original agent continuing to collect;

[0125] When the policy difficulty level is easy to collect, the policy collection cycle level is any of the stable payment period, due date, grace period, and suspension period, and the communication personnel saturation level is high saturation, the dispatch strategy is to transfer to supervisory collection;

[0126] When the policy difficulty level is easy to collect, the policy collection cycle level is any level of due date or grace period, and the communication personnel saturation level is any level of medium saturation or low saturation, the dispatching strategy is to collect for the original agent;

[0127] When the policy difficulty level is easy to collect, the policy collection cycle level is not configured, and the communication personnel saturation level is not configured, the dispatch strategy is the original agent collection;

[0128] When the policy difficulty level is difficult to collect, the policy collection cycle level is any of the stable payment period, due date, and grace period, and the communication personnel saturation level is high saturation, the dispatch strategy is to transfer to supervisory collection;

[0129] When the policy difficulty level is difficult to collect, the policy collection cycle level is out of effect, and the communication personnel saturation level is high, the dispatch strategy is to collect by the original agent;

[0130] When the policy difficulty level is difficult to collect, the policy collection cycle level is any of the stable payment period, due date, grace period, and suspension period, and the communication personnel saturation level is any of the medium saturation and low saturation levels, the dispatch strategy is to collect by the original agent;

[0131] When the policy difficulty level is to be followed up, the policy collection cycle level is not configured, and the communication personnel saturation level is not configured, the dispatch strategy is the original agent collection;

[0132] When the policy difficulty level is difficult, the policy collection cycle level is not configured, and the communication personnel saturation level is not configured, the dispatch strategy is to collect by the original agent.

[0133] In an embodiment of the present invention, when the number of policies is large or the policy customers are critical, an order dispatching strategy is formulated based on the characteristic results corresponding to the difficulty level table and the collection cycle level table. This can quickly respond to customer needs in the shortest time, improve customer service quality, and balance responsibility and efficiency. Employees can use this table to improve their work performance, improve employee performance, and increase benefits for the company. Through unified planning of the level table, manual operations and human decision-making can be reduced, thereby reducing the error rate and improving task processing accuracy.

[0134] In an embodiment of the present invention, generating a dispatch strategy based on policy intent can more accurately grasp the core information and specific requirements of the policy, thereby processing the policy more finely, improving the accuracy of policy processing, and automating policy processing, thereby saving a large amount of manual review and processing time, reducing labor costs, and at the same time, can process policies more quickly and accurately, thereby shortening policy processing time, improving work efficiency, and can also understand customer needs more finely, while being able to quickly respond to customer needs and further improve customer service quality.

[0135] It can be seen that in the above scheme, for the heavy insurance policy business, the insurance policy recording is first obtained and converted into an audio text, and then a preset attention model is proposed to extract the high-level semantic feature vector of the audio text. Through the result mapping method, a collaborative dispatching strategy is generated according to the communication personnel saturation level, policy intention, and policy cycle level, and insurance policies are dispatched according to the collaborative dispatching strategy. This can effectively reduce the workload of staff and avoid the omission of customers, greatly and efficiently improve the effect of policy processing, and reduce labor costs.

[0136] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0137] In one embodiment, a device for collecting and dispatching debts based on a multi-person collaborative mode is provided, and the device for collecting and dispatching debts based on a multi-person collaborative mode corresponds one-to-one to the method for collecting and dispatching debts based on a multi-person collaborative mode in the above embodiment. Figure 3 As shown, the debt collection dispatching device based on the multi-person collaboration mode includes a recording acquisition module 101, a conversion module 102, an extraction module 103, a matching calculation module 104, a saturation level acquisition module 105, a generation module 106 and a dispatching module 107. The functional modules are described in detail as follows:

[0138] The recording acquisition module 101 is used to acquire the insurance policy recording;

[0139] A conversion module 102, used to convert the insurance policy recording into a recording text;

[0140] An extraction module 103, used to extract a high-level semantic feature vector of the recorded text using a preset attention model;

[0141] A matching calculation module 104 is used to perform matching calculation on the high-level semantic feature vector and a preset intention vector table to obtain the insurance policy intention;

[0142] A saturation level acquisition module 105 is used to acquire the task saturation level of the communication personnel;

[0143] A generating module 106, configured to generate a collaborative dispatching strategy according to the communication personnel saturation level and the insurance policy intention;

[0144] The distribution module 107 is used to distribute insurance policies according to the collaborative distribution strategy.

[0145] In one embodiment, the recording module 101 is used to:

[0146] Obtaining the policy information of a preset existing customer group, screening the policy information according to a preset term range, and obtaining target policy information;

[0147] Allocating the target insurance policy information to preset communication personnel according to preset priorities;

[0148] Collect the policy recordings of the communication personnel communicating with the existing customer group based on the target policy information.

[0149] In one embodiment, the recording acquisition module 101 is further used for:

[0150] The insurance policy recording is converted into text to obtain a recording text.

[0151] In one embodiment, the extraction module 103 is used to:

[0152] Performing word segmentation processing on the recorded text according to a preset word segmentation method to obtain text segmentation;

[0153] Performing vector conversion on the text segmentation to obtain text segmentation vectors, and splicing the text segmentation vectors into a text vector;

[0154] Extracting semantic features from the text vector to obtain a semantic feature vector;

[0155] Performing attention distribution calculation on the semantic feature vector according to the preset attention model to obtain a vector weight of each text segmentation vector in the semantic feature vector;

[0156] The text word segmentation vectors are weighted according to the vector weights to obtain a high-level semantic feature vector.

[0157] In one embodiment, the extraction module 103 is further configured to:

[0158] According to the attention model, each word segmentation vector in the semantic feature vector is scored with the pre-acquired query vector to obtain the attention score of each word segmentation vector in the semantic feature vector;

[0159] The scored word segmentation vectors are weighted according to the Softmax function to obtain the weight of each word segmentation vector in the semantic feature vector.

[0160] In one embodiment, the extraction module 103 is further configured to:

[0161] Taking a weighted average of each word segmentation vector in the text and the corresponding vector weight, to obtain a weighted vector for each word segmentation in the text;

[0162] Performing a pooling operation on each word segmentation weighted vector in the text to obtain each word segmentation pooling vector in the text;

[0163] Each word pooling vector in the text is combined to generate a high-level semantic feature vector.

[0164] In one embodiment, the matching calculation module 104 is used to:

[0165] Extracting preset features in the intention vector table and feature results corresponding to the preset features;

[0166] Extracting features from the feature results one by one to obtain a feature vector of each feature result corresponding to a preset feature;

[0167] Select each feature vector corresponding to each preset feature one by one, calculate the similarity between the selected feature vector and the high-level semantic feature vector one by one, and obtain a similarity prediction value between each feature result corresponding to the preset feature in the intent vector table and the high-level semantic feature vector;

[0168] Each feature result with the highest similarity prediction value corresponding to the high-level semantic feature vector is selected one by one, and the selected feature results are merged into the policy intention.

[0169] In one embodiment, the generating module 106 is used to:

[0170] Divide the current policy period according to the preset rules to obtain the policy period result;

[0171] Mapping the policy cycle result to the pre-acquired policy urging cycle grade table to obtain the policy urging cycle grade;

[0172] Mapping the policy intention to a pre-acquired policy difficulty level table to obtain a policy difficulty level;

[0173] The result of the combination of the policy difficulty level, the communication personnel saturation level, and the policy collection cycle level is mapped to a pre-acquired dispatch strategy table, and a collaborative dispatch strategy is generated based on the dispatch strategy table.

[0174] The present invention provides a collection and dispatching device based on a multi-person collaborative mode. For the heavy insurance policy business, the insurance policy recording is first obtained and converted into an audio text, and then a preset attention model is proposed to extract the high-level semantic feature vector of the audio text. Through the result mapping method, a collaborative dispatching strategy is generated according to the communication personnel saturation level, policy intention, and policy cycle level, and the insurance policy is dispatched according to the collaborative dispatching strategy, which can effectively reduce the workload of the staff and avoid the situation where there are customers missing, greatly and efficiently improve the effect of insurance policy processing, and reduce labor costs at the same time.

[0175] For the specific definition of a debt collection and dispatching device based on a multi-person collaborative mode, please refer to the definition of a debt collection and dispatching method based on a multi-person collaborative mode above, which will not be repeated here. Each module in the above-mentioned debt collection and dispatching device based on a multi-person collaborative mode can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0176] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 5 As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client via a network connection. When the computer program is executed by the processor, it realizes a function or step on the service side of a collection and dispatching method based on a multi-person collaborative mode.

[0177] In one embodiment, a computer device is provided. The computer device may be a client, and its internal structure diagram may be as follows: Figure 6 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it realizes a function or step on the client side of a collection and dispatching method based on a multi-person collaborative mode.

[0178] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0179] Obtaining the policy recording, and converting the policy recording into a recording text;

[0180] Extracting a high-level semantic feature vector of the recorded text using a preset attention model;

[0181] Matching and calculating the high-level semantic feature vector with a preset intent vector table to obtain the policy intent;

[0182] The task saturation level of the communication personnel is obtained, a collaborative dispatching strategy is generated according to the communication personnel saturation level and the policy intention, and the policy is dispatched according to the collaborative dispatching strategy.

[0183] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0184] Obtaining the policy recording, and converting the policy recording into a recording text;

[0185] Extracting a high-level semantic feature vector of the recorded text using a preset attention model;

[0186] Matching and calculating the high-level semantic feature vector with a preset intent vector table to obtain the policy intent;

[0187] The task saturation level of the communication personnel is obtained, a collaborative dispatching strategy is generated according to the communication personnel saturation level and the policy intention, and the policy is dispatched according to the collaborative dispatching strategy.

[0188] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can refer to the relevant descriptions on the server side and the client side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0189] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0190] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0191] The above-described embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some of the technical features therein by equivalents; and these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should be included in the protection scope of the present invention. It should be noted that if software tools or components other than those of the company appear in the embodiments of this application, they are only used for example introduction and do not represent actual use.

Claims

1. A collection and dispatching method based on a multi-person collaboration mode, characterized in that: include: Obtaining the policy recording, and converting the policy recording into a recording text; Extracting a high-level semantic feature vector of the recorded text using a preset attention model; Matching and calculating the high-level semantic feature vector with a preset intent vector table to obtain the policy intent; The task saturation level of the communication personnel is obtained, a collaborative dispatching strategy is generated according to the communication personnel saturation level and the policy intention, and the policy is dispatched according to the collaborative dispatching strategy.

2. The method for collecting and dispatching debts based on a multi-person collaboration mode as claimed in claim 1, characterized in that: The obtaining of the policy recording and converting the policy recording into a recording text comprises: Obtaining the policy information of a preset existing customer group, screening the policy information according to a preset term range, and obtaining target policy information; Allocating the target insurance policy information to preset communication personnel according to preset priorities; Collecting the policy recordings of the communication personnel communicating with the existing customer group based on the target policy information; The insurance policy recording is converted into text to obtain a recording text.

3. The method for collecting and dispatching debts based on a multi-person collaboration mode as claimed in claim 1, characterized in that: The extracting of the high-level semantic feature vector of the recorded text by using a preset attention model includes: Performing word segmentation processing on the recorded text according to a preset word segmentation method to obtain text segmentation; Performing vector conversion on the text segmentation to obtain text segmentation vectors, and splicing the text segmentation vectors into a text vector; Extracting semantic features from the text vector to obtain a semantic feature vector; Performing attention distribution calculation on the semantic feature vector according to the preset attention model to obtain a vector weight of each text segmentation vector in the semantic feature vector; The text word segmentation vectors are weighted according to the vector weights to obtain a high-level semantic feature vector.

4. The method for collecting and dispatching debts based on a multi-person collaboration mode as claimed in claim 3, characterized in that: The performing attention distribution calculation on the semantic feature vector based on the preset attention model to obtain the weight of each word segmentation vector in the semantic feature vector includes: According to the attention model, each word segmentation vector in the semantic feature vector is scored with the pre-acquired query vector to obtain the attention score of each word segmentation vector in the semantic feature vector; The scored word segmentation vectors are weighted according to the Softmax function to obtain the weight of each word segmentation vector in the semantic feature vector.

5. The method for collecting and dispatching debts based on a multi-person collaboration mode as claimed in claim 3, characterized in that: The step of weighting the corresponding word segmentation vector according to the weight of each word segmentation vector in the text vector to obtain a high-level semantic feature vector includes: Performing a weighted average of each word segmentation vector in the text and the corresponding vector weight to obtain a weighted vector for each word segmentation in the text; Performing a pooling operation on each word segmentation weighted vector in the text to obtain each word segmentation pooling vector in the text; Each word pooling vector in the text is combined to generate a high-level semantic feature vector.

6. The method for collecting and dispatching debts based on a multi-person collaboration mode as claimed in claim 1, characterized in that: The step of matching and calculating the high-level semantic feature vector with a preset intent vector table to obtain the policy intent includes: Extracting preset features in the intention vector table and feature results corresponding to the preset features; Extracting features from the feature results one by one to obtain a feature vector of each feature result corresponding to a preset feature; Select each feature vector corresponding to each preset feature one by one, calculate the similarity between the selected feature vector and the high-level semantic feature vector one by one, and obtain a similarity prediction value between each feature result corresponding to the preset feature in the intent vector table and the high-level semantic feature vector; Each feature result with the highest similarity prediction value corresponding to the high-level semantic feature vector is selected one by one, and the selected feature results are merged into the policy intention.

7. The method for collecting and dispatching debts based on a multi-person collaboration mode as claimed in claim 1, characterized in that: Generating a collaborative dispatching strategy according to the communication personnel saturation level and the policy intention, and dispatching the policy according to the collaborative dispatching strategy, includes: Divide the current policy period according to the preset rules to obtain the policy period result; Mapping the policy cycle result to the pre-acquired policy urging cycle grade table to obtain the policy urging cycle grade; Mapping the policy intention to a pre-acquired policy difficulty level table to obtain a policy difficulty level; The result of the combination of the policy difficulty level, communication personnel saturation level, and policy collection cycle level is mapped to a pre-acquired dispatch strategy table, a collaborative dispatch strategy is generated based on the dispatch strategy table, and the policy is dispatched according to the collaborative dispatch strategy.

8. A collection and dispatching device based on a multi-person collaboration mode, characterized in that: include: The recording acquisition module is used to acquire the policy recording; A conversion module, used to convert the policy recording into a recording text; An extraction module, used to extract a high-level semantic feature vector of the recorded text using a preset attention model; A matching calculation module, used for performing matching calculation on the high-level semantic feature vector and a preset intent vector table to obtain the policy intent; A module for obtaining saturation levels is used to obtain the task saturation levels of communication personnel; A generation module, used for generating a collaborative dispatching strategy according to the communication personnel saturation level and the insurance policy intention; The distribution module is used to distribute insurance policies according to the collaborative distribution strategy.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method for debt collection and dispatching based on a multi-person collaborative mode as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, the steps of the method for collecting and dispatching debts based on a multi-person collaborative mode as described in any one of claims 1 to 7 are implemented.

Images