Collection strategy generation method and device, electronic equipment, storage medium and program product

Abstract

The invention discloses a collection strategy generation method and device, electronic equipment, a storage medium and a program product, and relates to the technical field of artificial intelligence. The method comprises the following steps: in a current call collection process, obtaining current call data and historical associated call data; recognizing the current call data and the historical associated call data by adopting a pre-trained target state recognition model and a pre-trained target intention recognition model to obtain a current state and a current intention; and obtaining a preset collection strategy matching rule, matching a current collection strategy in the preset collection strategy matching rule according to the current state and the current intention, and feeding back the current collection strategy to the business personnel, so that the business personnel carries out collection according to the current collection strategy. According to the technical scheme of the embodiment of the invention, automatic generation of the collection strategy is realized, the collection efficiency and the collection quality of business personnel are improved, and the fund security of a financial institution is ensured.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a collection strategy generation method, device, electronic device, storage medium and program product. Background Art

[0002] In recent years, the quality problem of credit card assets has become increasingly serious, and the pressure of post-loan management has become increasingly prominent. Manual debt collection is an important means for financial institutions to recover non-performing assets.

[0003] Manual debt collection personnel usually face a heavy workload and tedious tasks, making it difficult to ensure debt collection efficiency and quality, as well as the financial institutions' financial security. Summary of the invention

[0004] The present invention provides a collection strategy generation method, device, electronic device, storage medium and program product, which take into account the status and intention of the party being collected during the telephone collection process, realize the automatic generation of collection strategies, improve the collection efficiency and quality of business personnel, and thus ensure the financial security of financial institutions.

[0005] According to one aspect of the present invention, a method for generating a collection strategy is provided, the method comprising:

[0006] In the current telephone collection process, current call data and historical associated call data are obtained; wherein the historical associated call data is historical call data associated with the current call data;

[0007] Using a pre-trained target state recognition model and a target intention recognition model, the current call data and the historical associated call data are recognized to obtain a current state and a current intention;

[0008] Obtain preset collection strategy matching rules, match the current collection strategy in the preset collection strategy matching rules according to the current state and the current intention, and provide feedback to business personnel so that the business personnel can perform collection according to the current collection strategy.

[0009] According to another aspect of the present invention, a collection strategy generation device is provided, the device comprising:

[0010] A call data acquisition module, used to acquire current call data and historical associated call data during the telephone collection process; wherein the historical associated call data is historical call data associated with the current call data;

[0011] A state intention recognition module, used to use a pre-trained target state recognition model and a target intention recognition model to recognize the current call data and the historical associated call data to obtain a current state and a current intention;

[0012] The collection strategy generation module is used to obtain preset collection strategy matching rules, match the current collection strategy in the preset collection strategy matching rules according to the current state and the current intention, and provide feedback to business personnel so that the business personnel can collect according to the current collection strategy.

[0013] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0014] at least one processor; and

[0015] a memory communicatively connected to the at least one processor; wherein,

[0016] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the collection strategy generation method described in any embodiment of the present invention.

[0017] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the collection strategy generation method described in any embodiment of the present invention when executed.

[0018] According to another aspect of the present invention, a computer program product is provided. The computer program product includes a computer program. When the computer program is executed by a processor, the method for generating a collection strategy described in any embodiment of the present invention is implemented.

[0019] The technical solution of the embodiment of the present invention obtains the current call data and the historically associated call data during the current telephone collection process, adopts a pre-trained target state recognition model and a target intention recognition model to recognize the current call data and the historically associated call data, obtains the current state and the current intention, obtains the preset collection strategy matching rules, matches the current collection strategy in the preset collection strategy matching rules according to the current state and the current intention, takes into account the state and intention of the party being collected during the telephone collection process, realizes the automatic generation of the collection strategy, and feeds back the current collection strategy to the business personnel so that the business personnel can collect according to the current collection strategy, thereby improving the collection efficiency and quality of the business personnel, thereby ensuring the financial security of the financial institution.

[0020] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 is a flow chart of a collection strategy generation method provided according to the first embodiment of the present invention;

[0023] Figure 2 is a flow chart of a collection strategy generation method provided according to Embodiment 2 of the present invention;

[0024] Figure 3 This is a flow chart of a post-collection quality inspection method provided according to Embodiment 2 of the present invention;

[0025] Figure 4 is a structural schematic diagram of a collection strategy generation device provided according to Embodiment 3 of the present invention;

[0026] Figure 5 It is a structural schematic diagram of an electronic device for implementing the collection strategy generation method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only 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 should fall within the scope of protection of the present invention.

[0028] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0029] Embodiment 1

[0030] Figure 1 A flowchart of a collection strategy generation method provided in Embodiment 1 of the present invention. Embodiments of the present invention are applicable to the case where a collection strategy is automatically generated during a telephone collection process. The method can be executed by a collection strategy generation device, which can be implemented in the form of hardware and / or software. The collection strategy generation device can be configured in an electronic device that carries the collection strategy generation function, such as a collection party terminal or a server.

[0031] See also Figure 1 The collection strategy generation method shown includes:

[0032] S110. In the current telephone collection process, current call data and historical related call data are obtained.

[0033] The current telephone collection process may be the telephone collection process currently being conducted. It can be understood that the business personnel and the party being collected are conducting real-time telephone collection. The current call data may be the call data between the business personnel and the party being collected at the current moment. The historically associated call data may be the historical call data associated with the current call data. The historically associated call data may be used to characterize the context information of the current call data.

[0034] Specifically, in the current telephone collection process, with the authorization of the party being collected, the current call data and historical related call data can be obtained.

[0035] S120: Use a pre-trained target state recognition model and a target intention recognition model to recognize the current call data and the historical associated call data to obtain the current state and current intention of the current telephone collection process.

[0036] The target state recognition model can be used to identify the state of the party being collected during the current telephone collection process. The input data of the target state recognition model includes the current call data and the historical associated call data; the output data is the current state of the party being collected. The target intention recognition model can be used to identify the repayment intention of the party being collected during the current telephone collection process. The input data of the target intention recognition model includes the current call data and the historical associated call data; the output data is the current intention of the party being collected. Exemplarily, the target state recognition model and the target intention recognition model can be machine learning models or deep learning models, such as support vector machines (SVM), naive Bayes, recurrent neural networks (RNN) or long short-term memory networks (LSTM). The current state can be used to characterize the state of the party being collected during the current telephone collection process. Exemplarily, the current state can be the current emotional state. For example, the current state can include active cooperation, neutrality, and passive resistance. Among them, the passive resistance type can include perfunctory, helpless, and rejection. The current intention can be used to characterize the repayment intention of the debtee in the current telephone debt collection process. Exemplarily, the current intention can include high intention, medium intention and low intention.

[0037] Specifically, the current call data and the historical associated call data may be input into a target state recognition model to output the current state. The current call data and the historical associated call data may be input into a target intention recognition model to output the current intention.

[0038] In an optional example, before using a pre-trained target intent recognition model and a target intent recognition model to recognize the current call data and the historical associated call data, and obtaining the current intent and the current intent, obtain historical call samples, historical associated call samples, and historical actual intents in the historical telephone collection process, and obtain an initial intent recognition model. Input the historical call samples and the historical associated call samples into the initial intent recognition model to obtain the historical predicted intent of the historical telephone collection process. According to the difference between the historical predicted intent and the historical actual intent of the historical telephone collection process, adjust the parameters of the initial intent recognition model to obtain the target intent recognition model.

[0039] The initial intent recognition model can be an untrained intent recognition model. The historical actual intent can be used to characterize the real intent of the party being collected during the historical telephone collection process. The historical actual intent of the historical telephone collection process can be used to correct the initial intent recognition model. The historical predicted intent is the predicted intent of the party being collected during the historical telephone collection process, that is, the prediction result of the initial intent recognition model. The target intent recognition model can be a trained intent recognition model. The target intent recognition model can be used to predict the intention of the party being collected during the telephone collection process.

[0040] Specifically, with the authorization of the party being collected, historical call samples, historical associated call samples, and historical actual intentions in the historical telephone collection process can be obtained. An untrained initial intention recognition model can be obtained. The historical call samples and historical associated call samples can be input into the initial intention recognition model to output the historical predicted intentions of the historical telephone collection process. The initial intention recognition model can be adjusted according to the difference between the historical predicted intentions and the historical actual intentions of the historical telephone collection process until the difference between the historical predicted intentions and the historical actual intentions converges, thereby obtaining a target intention recognition model.

[0041] This solution can improve the efficiency and accuracy of debt collection strategy generation by pre-generating a target intent recognition model based on the target intent recognition model.

[0042] S130, obtaining preset collection strategy matching rules, matching the current collection strategy in the preset collection strategy matching rules according to the current state and current intention, and providing feedback to the business personnel so that the business personnel can perform collection according to the current collection strategy during the current telephone collection process.

[0043] The preset collection strategy matching rules may be matching rules between preset status and intention and collection strategy. The preset collection strategy matching rules may be used to quickly determine the collection strategy corresponding to the status and intention. Optionally, the preset collection strategy matching rules may include intention, status, feature analysis and collection strategy. Among them, the feature analysis may be an explanation and example description of the intention and status. The feature analysis may be used to analyze in detail the intention and status of the party being collected. Optionally, the preset collection strategy matching rules may be pre-generated by a technician and stored in the device. Exemplarily, the preset collection strategy matching rules may be a preset collection strategy matching table.

[0044] Table 1 Preset collection strategy matching table

[0045]

[0046] Specifically, the pre-stored preset collection strategy matching rules can be obtained. According to the current state and current intention, the corresponding collection strategy can be queried in the preset collection strategy matching rules to obtain the current collection strategy. The current collection strategy is fed back to the business personnel so that the business personnel can collect according to the current collection strategy during the current call.

[0047] The technical solution of the embodiment of the present invention obtains the current call data and the historically associated call data during the current telephone collection process, adopts a pre-trained target state recognition model and a target intention recognition model to recognize the current call data and the historically associated call data, obtains the current state and the current intention, obtains the preset collection strategy matching rules, matches the current collection strategy in the preset collection strategy matching rules according to the current state and the current intention, takes into account the state and intention of the party being collected during the telephone collection process, realizes the automatic generation of the collection strategy, and feeds back the current collection strategy to the business personnel so that the business personnel can collect according to the current collection strategy, thereby improving the collection efficiency and quality of the business personnel, thereby ensuring the financial security of the financial institution.

[0048] In an optional embodiment of the present invention, the method also includes: after the current telephone collection process is completed, obtaining complete call data; performing feature extraction on the complete call data to obtain collection effect features, communication effect features, call duration features and state change features of the complete call data; using a collection quality scoring model to detect the collection effect features, communication effect features, call duration features and state change features of the complete call data to determine the current collection quality score of the current telephone collection process; and adjusting the preset collection strategy matching rules, target state recognition model or target intent recognition model according to the current collection score of the complete call data.

[0049] The complete call data can be used to characterize the complete call record of the current telephone collection process. Exemplarily, the complete call data may include the current call data, the historical associated call data, the current collection data of the current call data, and the current feedback data of the current collection data. Among them, the current collection data may be the collection data sent by the business personnel to the collection party based on the current collection strategy. The current feedback data may be the feedback data after the collection party receives the current collection data from the business personnel. Exemplarily, the content form of the complete call data may include text form or voice form, etc. The collection effect feature may be used to measure whether the business personnel has successfully collected the debt. The communication effect feature may be used to characterize the effectiveness of the current collection strategy of the business personnel and the language expression ability of the business personnel. For example, whether the business personnel can accurately answer the questions of the collection party, and whether the business personnel can persuade the collection party to repay the debt, etc. The call duration feature may be used to characterize whether the business personnel and the collection party have fully communicated and the corresponding collection effect. It can be understood that a longer call duration feature may indicate that the business personnel and the collection party have fully communicated; and an excessively long call duration may indicate that the collection effect is not good. Optionally, a first preset communication time and a second preset communication time may be set. The first preset communication time is less than the second preset communication time. The first preset communication time may be used to measure whether the business personnel have fully communicated with the party being collected. The second preset communication time may be used to determine whether the collection effect is poor. The state change feature may be used to characterize the state change of the party being collected. It can be understood that if the state of the party being collected changes from an active cooperation type to a passive resistance type, it indicates that there is a problem in the current collection process; if the state of the party being collected changes from a passive resistance type to an active cooperation type, it indicates that the current collection process is properly handled. The collection quality scoring model may be used to evaluate the current telephone collection process. The input data of the collection quality evaluation model may be collection effect features, communication effect features, call duration features, and state change features, and the output result may be the current collection quality score of the current telephone collection process. The training samples of the collection quality scoring model may be collection effect feature samples, communication effect feature samples, call duration feature samples, state change feature samples, and collection quality scoring samples. Among them, the collection quality score samples can be determined and adjusted by technicians based on experience. Optionally, the collection quality score model can be trained in a supervised training mode. Exemplarily, the collection quality assessment model can be a machine learning model or a deep learning model, such as a support vector machine (SVM), Naive Bayes, a recurrent neural network (RNN), or a long short-term memory network (LSTM). The current collection quality score can be used to characterize the accuracy and effectiveness of the current collection strategy in the current collection process.The current collection quality score can be used to feed back the preset collection strategy matching rules, target state recognition model or target intent recognition model to achieve iterative optimization of the preset collection strategy matching rules, target state recognition model or target intent recognition model.

[0050] Specifically, after the current telephone collection process is completed, the complete call data can be obtained with the authorization of the collection party. Principal component analysis (PCA), linear discriminant analysis (LDA), kernel principal component analysis (KPCA) or local linear embedding (LLE) can be used to extract and classify the complete call data to obtain the collection effect features, communication effect features, call duration features and state change features of the complete call data. The collection effect features, communication effect features, call duration features and state change features of the complete call data can be input into a pre-trained collection quality scoring model to output the current collection quality score of the current telephone collection process. When the current collection score of the complete call data is lower than the preset collection quality score threshold, the preset collection strategy matching rule, target state recognition model or target intention recognition model can be adjusted. Among them, the preset collection quality score threshold can be a preset lower limit of the collection quality score. The preset collection quality score can be predetermined and adjusted by the technical staff.

[0051] In an optional example, the content form of the complete call data can be in the form of voice. Before extracting features from the complete call data to obtain the collection effect features, communication effect features, call duration features, and state change features of the complete call data, the complete call data can be processed for accent, speech speed, and background noise. Speech recognition technology can be used to convert the processed complete call data from voice to complete call text. Feature extraction can be performed on the complete call text to obtain the collection effect features, communication effect features, call duration features, and state change features of the complete call data. By preprocessing the complete call data, the completeness and accuracy of the complete call text can be improved, thereby improving the accuracy of the collection quality score.

[0052] After the current telephone collection process is completed, this solution introduces a collection quality scoring model to perform a collection quality scoring on the current telephone collection process, thereby realizing adjustments to the preset collection strategy matching rules, target state recognition model or target intent recognition model. By performing subsequent quality scoring on the current telephone collection process, feedback to the current telephone collection process is realized, and the accuracy of the preset collection strategy matching rules, target state recognition model or target intent recognition model is improved, thereby improving the accuracy and effectiveness of subsequent collection strategies, thereby improving the quality of subsequent telephone collection processes, and further ensuring the financial security of financial institutions.

[0053] In an optional example, after the current telephone collection process is completed, the complete call data can be obtained. The pre-trained target state recognition model and target intent recognition model can be used to identify the complete call data to obtain the secondary detection state and secondary detection intent. The preset collection strategy matching rules can be obtained, and the secondary collection strategy can be matched in the preset collection strategy matching rules according to the secondary detection state and the secondary detection intent, and feedback can be given to the business personnel so that the business personnel can conduct secondary collection on the collection party according to the secondary collection strategy. By generating a secondary collection strategy and providing feedback to the business personnel, the business personnel conduct secondary collection on the collection party according to the secondary collection strategy, which further improves the collection quality and further ensures the financial security of financial institutions.

[0054] Embodiment 2

[0055] Figure 2 A flowchart of a collection strategy generation method provided in the second embodiment of the present invention. Based on the above embodiments, the embodiment of the present invention further adds "obtaining historical call samples and historical related call samples in the process of historical phone collection, and obtaining an initial state recognition model and a state vocabulary; detecting the historical state vocabulary of each alternative state type in the historical call samples, and determining the historical state type word frequency of each alternative state type in the historical call samples; determining the historical phone collection strategy according to the historical state type word frequency of each alternative state type in the historical call samples" before "using a pre-trained target state recognition model and a target intent recognition model to identify the current call data and the historical related call data to obtain the current state and the current intent". The historical actual state of the collection process; the historical call samples and the historical associated call samples are input into the initial state recognition model to obtain the historical predicted state of the historical telephone collection process; according to the difference between the historical predicted state and the historical actual state of the historical telephone collection process, the initial state recognition model is adjusted to obtain the target state recognition model", and a state vocabulary is introduced to determine the historical actual state of the historical telephone collection process, and then the target state recognition model is generated. Compared with the historical actual state labeled based on manual experience, there is a problem that the accuracy of the target state recognition model is too dependent on manual experience, and the detection accuracy of the target state recognition model can be improved. It should be noted that for the parts not described in detail in the embodiments of the present invention, reference can be made to the descriptions of other embodiments.

[0056] See also Figure 2 The collection strategy generation method shown includes:

[0057] S210, obtaining historical call samples and historical associated call samples in the historical telephone collection process, and obtaining an initial state recognition model and a state vocabulary.

[0058] The historical telephone collection process may be a telephone collection process conducted at a historical moment. The historical call sample may be call data conducted between a business person and a collection party at a historical moment. The historical associated call sample may be other historical call samples associated with the historical call sample. The historical associated call sample may be used to characterize the context information of the historical call sample. The initial state recognition model may be an untrained state recognition model. The state vocabulary may be used to record at least one candidate state vocabulary, and the correspondence between the candidate state vocabulary and the candidate state type. The state vocabulary includes candidate state vocabulary corresponding to different candidate state types. The state vocabulary may be pre-generated by a technician and stored in the device. The state vocabulary may be periodically updated by a technician. The candidate state vocabulary may be the state vocabulary recorded in the state vocabulary. The candidate state vocabulary may be used to detect the historical state corresponding to the historical call sample. The candidate state type may be used to characterize the state type corresponding to the candidate state vocabulary. Exemplarily, the candidate state type may include active cooperation type, neutral type and passive resistance type. The candidate state vocabulary may include "willing", "immediately", "affirmative", "good", "reject", "don't want" and "no way". Among them, "willingly", "immediately" and "definitely" are alternative state words of active cooperation type; "okay" is an alternative state word of neutral type; "refuse", "don't want" and "no way" are alternative state words of passive resistance type.

[0059] Specifically, with the authorization of the debt collection party, historical call samples and historical related call samples in the historical telephone debt collection process can be obtained. An untrained initial state recognition model is obtained. A pre-generated and stored state vocabulary is obtained.

[0060] S220: Detect the historical state vocabulary of each candidate state type in the historical call sample, and determine the historical state type word frequency of each candidate state type in the historical call sample.

[0061] The historical state vocabulary may be the candidate state vocabulary existing in the historical call sample. The historical state type word frequency may be the number of all historical state vocabulary corresponding to the candidate state type in the historical call sample.

[0062] Specifically, statistics may be collected on the historical state words of each candidate state type in the historical call samples to determine the number of historical state words included in each candidate state type in the historical call samples, and obtain the word frequency of the historical state type.

[0063] Optionally, the state vocabulary may also include emphasis words; after determining the historical state type word frequency of each candidate state type in the historical call sample, the emphasis word frequency of the historical call sample emphasis words and the historical state words of the negative resistance type may be detected; the historical state type word frequency of the negative resistance type may be enhanced according to the emphasis word frequency; for example, the emphasis word frequency and the historical state type word frequency of the negative resistance type may be directly summed, weighted summed, or multiplied to update the historical state type word frequency of the negative resistance type. When the emphasis words and the historical state words of the negative resistance type appear at the same time, the negative resistance state may be strengthened and the historical state type frequency of the negative resistance type may be enhanced. Exemplarily, the emphasis words may include "absolutely", "fundamentally" or "completely" and the like. This scheme introduces the detection process of emphasis words and the update process of the historical state type word frequency of the negative resistance type based on the emphasis words. It takes into account the influence of emphasis words on the historical state type word frequency of the negative resistance type, improves the accuracy of the historical state type word frequency detection of the negative resistance type, and further improves the accuracy of the determined historical actual state of the historical telephone collection process, thereby improving the accuracy of the target state recognition model.

[0064] S230: Determine the historical actual status of the historical telephone debt collection process according to the historical status type word frequency of each candidate status type in the historical call sample.

[0065] The historical actual state can be used to characterize the real state of the debt collection party in the historical telephone debt collection process. The historical actual state can be used to modify the initial state recognition model.

[0066] Specifically, the historical state type word frequencies of the candidate state types in the historical call samples can be compared, and the candidate state type corresponding to the maximum historical state type word frequency can be determined as the historical actual state of the historical telephone collection process.

[0067] In an optional embodiment of the present invention, the alternative state types include active cooperation type and passive resistance type; accordingly, according to the historical state type word frequency of each alternative state type in the historical call sample, the historical actual state of the historical telephone collection process is determined, including: obtaining negative punctuation marks; detecting the historical negative punctuation mark frequency of negative punctuation marks that appear simultaneously with the historical state words of the negative resistance type in the historical call sample; integrating the historical state type word frequency of the negative resistance type and the historical negative punctuation mark frequency, and updating the historical state type word frequency of the negative resistance type; comparing the historical state type word frequency of the active cooperation type and the historical state type word frequency of the negative resistance type in the historical call sample; and determining the alternative state type corresponding to the maximum value of the historical state type word frequency as the historical actual state of the historical telephone collection process.

[0068] The active cooperation type can be used to characterize the active cooperation attitude of the party being collected during the historical telephone collection process. The passive resistance type can be used to characterize the passive resistance attitude of the party being collected during the historical telephone collection process. Negative punctuation marks can be punctuation marks that characterize the negative resistance attitude of the party being collected. Exemplarily, negative punctuation marks may include exclamation marks and question marks, etc. Negative punctuation marks that appear simultaneously with historical state words of the negative resistance type can be used to enhance the historical state of the negative resistance type. It can be understood that negative punctuation marks that appear simultaneously with historical state words of the negative resistance type can be used to characterize a strong historical state of the negative resistance type. The frequency of historical negative punctuation marks can be used to characterize the frequency of occurrence of negative punctuation marks that appear simultaneously with historical state words of the negative resistance type in historical call samples.

[0069] Specifically, predetermined negative punctuation marks can be obtained. The number of negative punctuation marks that appear simultaneously with negative and resistant historical status words in historical call samples can be counted to determine the historical frequency of negative punctuation marks in historical call samples. The historical status type word frequency of the negative and resistant type can be updated by directly summing, weighted summing or multiplying the historical status type word frequency of the negative and resistant type and the historical negative punctuation mark frequency. The historical status type word frequency of the active cooperation type and the historical status type word frequency of the negative and resistant type in historical call samples can be compared, and the alternative status type corresponding to the maximum value of the historical status type word frequency can be determined as the historical actual state of the historical telephone collection process.

[0070] In an optional embodiment of the present invention, the historical state type word frequency of the negative resistance type and the historical negative punctuation mark frequency are integrated to update the historical state type word frequency of the negative resistance type, including: performing negative sentence structure detection on historical call samples to obtain historical negative sentence structure detection results; integrating the historical state type word frequency of the negative resistance type, the historical negative punctuation mark frequency and the historical negative sentence detection results to update the historical state type word frequency of the negative resistance type.

[0071] Negative sentence structures can be used to imply the negative and resistant attitude of the party being collected during the historical telephone collection process. Exemplarily, negative sentence structures can include short and fragmented sentences or passive sentences. For example, short and fragmented sentences are "No, no, don't urge me!" Compared with complete and organized sentences, short and fragmented sentences can better reflect the negative and resistant attitude of the party being collected. The passive sentence is "I am so annoyed by being urged." Passive sentences may reflect the negative and resistant attitude of the party being collected. The historical negative sentence structure detection results can be used to characterize the detection results of negative sentence structures in historical call samples. Exemplarily. The historical negative sentence structure detection results can be the frequency of occurrence of negative sentence structures in historical call samples. Among them, the frequency of occurrence of negative sentence structures in historical call samples can be represented by the number of occurrences of negative sentence structures in historical call samples.

[0072] Specifically, natural language processing technology can be used to detect negative sentence structures in historical call samples to obtain historical negative sentence structure detection results, such as the frequency of occurrence of historical negative sentence structures. The negative conflict type historical state type word frequency, historical negative punctuation frequency, and historical negative sentence detection results can be directly summed or weighted summed to update the negative conflict type historical state type word frequency.

[0073] This scheme introduces a detection process for negative sentence structures and an update process for the historical state type word frequency of the negative resistance type based on the detection results of the negative sentence structures. It takes into account the influence of negative sentence structures on the historical state type word frequency of the negative resistance type, improves the accuracy of the historical state type word frequency detection of the negative resistance type, and further improves the accuracy of the historical actual state of the determined historical telephone collection process, thereby improving the accuracy of the target state recognition model.

[0074] In an optional embodiment of the present invention, before comparing the historical state type word frequencies of the active cooperation type and the historical state type word frequencies of the passive resistance type in the historical call samples, it also includes: detecting the state change of the historical call samples and the historical associated call samples to obtain the historical state change detection results of the historical call samples; when the historical state change detection result is a change from the passive resistance type to the active cooperation type, integrating the historical state type word frequency of the active cooperation type and the historical state change detection result, and updating the historical state type word frequency of the active cooperation type; when the historical state change detection result is a change from the active cooperation type to the passive resistance type, integrating the historical state type word frequency of the passive resistance type and the historical state change detection result, and updating the historical state type word frequency of the passive resistance type.

[0075] The historical state change detection results can be used to characterize the state difference between the historical call samples and the historical associated call samples. The historical state change detection results can be used to reflect the state change of the collection party from the perspective of context association. Exemplarily, the historical state change detection results may include no change, change from passive resistance type to active cooperation type, and change from active cooperation type to passive resistance type.

[0076] Optionally, any historical actual state in the real-time example of the present invention may be used to determine the first historical actual state corresponding to the historical associated call sample and the second historical actual state corresponding to the historical call sample. The first historical actual state may be the historical actual state corresponding to the historical associated call sample. The second historical actual state may be the historical actual state corresponding to the historical call sample.

[0077] Specifically, the first historical actual state corresponding to the historical associated call sample and the second historical actual state corresponding to the historical call sample can be obtained. The first historical actual state corresponding to the historical call sample and the second historical actual state corresponding to the historical associated call sample can be compared. When the first historical actual state corresponding to the historical call sample and the second historical actual state corresponding to the historical associated call sample are consistent, it is determined that the historical state change detection result of the historical call sample is unchanged. When the first historical actual state corresponding to the historical call sample and the second historical actual state corresponding to the historical associated call sample are inconsistent, and the first historical actual state is an active cooperation type, it is determined that the historical state change detection result of the historical call sample is changed from an active cooperation type to a passive resistance type. When the first historical actual state corresponding to the historical call sample and the second historical actual state corresponding to the historical associated call sample are inconsistent, and the first historical actual state is a passive resistance type, it is determined that the historical state change detection result of the historical call sample is changed from a passive resistance type to an active cooperation type. When the historical state change detection result is a change from a passive resistance type to an active cooperation type, the historical state change detection result is used to enhance the historical state type word frequency of the active cooperation type, such as by increasing the historical state type word frequency of the active cooperation type by a preset number or multiplying the preset number, etc., to update the historical state type word frequency of the active cooperation type. When the historical state change detection result is a change from an active cooperation type to a passive resistance type, the historical state change detection result is used to enhance the historical state type word frequency of the passive resistance type, such as by increasing the historical state type word frequency of the passive resistance type by a preset number or multiplying the preset number, etc., to update the historical state type word frequency of the passive resistance type. Among them, the preset number can be predetermined and adjusted by a technician.

[0078] This scheme introduces a process for detecting the state changes between historical call samples and historical associated call samples, and a process for updating the word frequencies of historical state types of active cooperation or passive resistance based on the state change detection results. It takes into account the impact of the state changes between historical call samples and historical associated call samples, improves the accuracy of the determined historical actual state of the historical telephone collection process, and thus improves the accuracy of the target state recognition model.

[0079] This scheme concretizes the alternative state types into active cooperation type and passive resistance type, introduces the detection process of historical negative punctuation frequency and the update process of historical state type word frequency of negative resistance type based on historical negative punctuation frequency, considers the influence of negative punctuation on the historical state type word frequency of negative resistance type, improves the accuracy of historical state type word frequency detection of negative resistance type, and further improves the accuracy of the determined historical actual state of the historical telephone collection process, thereby improving the accuracy of the target state recognition model.

[0080] S240: Input historical call samples and historical associated call samples into an initial state recognition model to obtain a historical prediction state of the historical telephone debt collection process.

[0081] The historical predicted state is the predicted state of the party being collected during the historical telephone collection process, that is, the historical predicted state can be the prediction result of the initial state recognition model. The historical actual state of the historical telephone collection process can be the actual state of the party being collected during the historical telephone collection process.

[0082] Specifically, the historical call samples and the historical associated call samples may be input into the initial state recognition model to output the historical predicted state of the historical telephone debt collection process.

[0083] S250. According to the difference between the historical predicted state and the historical actual state of the historical telephone collection process, the initial state recognition model is adjusted to obtain the target state recognition model.

[0084] The target state recognition model may be a trained state recognition model. The target state recognition model may be used to predict the state of the debt collection party during the telephone debt collection process.

[0085] Specifically, the initial state recognition model can be adjusted according to the difference between the historical predicted state and the historical actual state of the historical telephone collection process until the difference between the historical predicted state and the historical actual state converges to obtain the target state recognition model.

[0086] S260. In the current telephone collection process, current call data and historical related call data are obtained.

[0087] The historical associated call data is historical call data associated with the current call data.

[0088] S270: Use a pre-trained target state recognition model and a target intent recognition model to recognize the current call data and the historical associated call data to obtain the current state and the current intent.

[0089] S280. Obtain preset collection strategy matching rules, match the current collection strategy in the preset collection strategy matching rules according to the current state and current intention, and provide feedback to the business personnel so that the business personnel can perform collection according to the current collection strategy.

[0090] The technical solution of the embodiment of the present invention obtains historical call samples and historical associated call samples in the historical telephone collection process, and obtains an initial state recognition model and a state vocabulary, detects the historical state vocabulary of each candidate state type in the historical call samples, determines the historical state type word frequency of each candidate state type in the historical call samples, determines the historical actual state of the historical telephone collection process according to the historical state type word frequency of each candidate state type in the historical call samples, inputs the historical call samples and the historical associated call samples into the initial state recognition model, obtains the historical predicted state of the historical telephone collection process, adjusts the parameters of the initial state recognition model according to the difference between the historical predicted state and the historical actual state of the historical telephone collection process, obtains the target state recognition model, introduces the state vocabulary, determines the historical actual state of the historical telephone collection process, and then generates the target state recognition model. Compared with the historical actual state of the historical telephone collection process labeled based on manual experience, there is a problem that the accuracy of the target state recognition model is too dependent on manual experience. The technical solution of the embodiment of the present invention can improve the detection accuracy of the target state recognition model.

[0091] Based on the above embodiments, the present invention also provides a preferred embodiment of a collection strategy generation method. In the current telephone collection process, the system automatically matches the current collection strategy based on the capabilities of the target state recognition model and the target intent recognition model, the call context and the current context, and provides immediate professional support to the business personnel of the collection seat, reducing the time for information search and understanding, thereby improving the call quality and work efficiency of the business personnel.

[0092] The collection strategy generation method includes: in the current telephone collection process, the system obtains the current call content and the historical related call content in real time. By pre-building the target intention recognition and target state recognition models, the current call voice is converted into the current call text in real time, and the processed current call text and the historical related call text are subjected to natural language processing to identify the current intention and current state of the collection party. According to the two dimensions of the current intention and current state of the collection party identified, a preset collection strategy matching table is used to match the corresponding current collection strategy. Among them, the preset collection strategy matching table is shown in Table 1.

[0093] The process of building the target intent recognition model is to first annotate a large number of historical call samples and historical associated call samples, and mark the candidate intent categories of different intent types (such as high intent and low intent). Then use machine learning algorithms (such as support vector machines or naive Bayes, etc.) or deep learning algorithms (such as recurrent neural networks or long short-term memory networks, etc.) for training.

[0094] The construction process of the target state recognition model is to determine the historical actual state of the historical telephone collection process by analyzing the state vocabulary, negative punctuation, negative sentence structure and context association in the historical call samples and the historical related call samples. Specifically, the state vocabulary table and the machine learning algorithm can be combined to classify the alternative state types of the historical call samples and the historical related call samples to obtain the historical actual state of the historical call process. Common alternative state categories can include active cooperation type and passive resistance type. The specific classification steps of the alternative state type include: vocabulary selection and analysis, punctuation use, sentence structure analysis and context association, and the output result is the classification annotation of the historical call samples. Specifically, ① vocabulary selection and analysis. It is necessary to first construct a state vocabulary table with state color, which should include alternative state vocabulary representing active cooperation type (such as "willing", "immediately" and "affirmative", etc.) and alternative state vocabulary representing passive resistance type (such as "reject", "don't want" and "no way", etc.). When identifying historical call texts, the frequency of historical state types of historical state words of different candidate state types can be counted. For example, if there are more negative and resistant historical state words in the historical call text, it may mean that the actual historical state of the party being collected is negative and resistant. Optionally, some emphasis words, such as "absolutely", "fundamentally" or "completely", etc., can be noted. When these emphasis words appear at the same time as negative and resistant historical state words, they may strengthen the negative and resistant state and enhance the frequency of negative and resistant historical state types. ② Utilization of punctuation marks. For example, exclamation marks usually indicate a strong state. If there are more exclamation marks in the historical call text, such as "I really can't pay it back!", it is likely that the party being collected will show an excited or angry state. For example, continuous question marks may represent the doubts or dissatisfaction of the party being collected, such as "Why do you keep urging me? Haven't I said it?" Such expressions may have a questioning tone. The above negative punctuation marks can enhance the frequency of negative and resistant historical state types. ③ Sentence structure analysis. For example, short and fragmented sentences may imply impatience or a bad state, such as "I can't, I can't, don't rush me". Compared with complete and organized sentences, this type of sentence can better reflect the historical actual state of passive resistance. Passive sentences or sentences containing passive voice may reflect the helpless state of the party being collected. For example, "I am so annoyed by being urged." ④ Context association. Consider the previous and next content of the historical call text and the historical related call text to judge the historical actual state of the party being collected during the historical telephone collection process. If the historical related call text has been negotiating the repayment plan, and the party being collected suddenly says "I don't want to talk about this anymore" in the historical call text, combined with the context, it can be inferred that the party being collected may be a little irritable or unwilling to cooperate in the historical actual state of passive resistance.

[0095] In the post-quality inspection phase, that is, after the current telephone collection process is completed, speech recognition technology can be used to convert the complete call data into text, process different accents, speaking speeds and background noise, and ensure the complete record of the content of the complete call data. Through natural language processing technology, the semantics and status in the complete call data can be deeply understood, the secondary detection intention and secondary detection status of the party being collected can be identified, and real-time feedback and secondary collection strategies can be provided to business personnel. The complete call data can also be analyzed to obtain the current collection quality score, objectively evaluate the performance of business personnel in the current telephone collection process, help identify problems with the current collection strategy and provide improvement suggestions.

[0096] Figure 3 The following is a flow chart of a post-collection quality inspection method. Figure 3 As shown in the figure, the post-collection quality inspection methods include:

[0097] A) Start.

[0098] Start the intelligent quality inspection process.

[0099] B) Post-quality inspection: voice recognition is converted into text.

[0100] Use speech recognition technology to convert complete call data into text format.

[0101] C) Identify accent, speaking speed, and background noise.

[0102] The converted complete call text is analyzed to identify the accent, speech speed and background noise. Specifically, the process includes D) accent processing; E) speech speed processing; F) background noise processing.

[0103] Among them, the specific processing process of D) accent processing is G) using accent adaptation technology.

[0104] Specifically, accent adaptation technology can be used to deal with different accents. Accent processing collects speech data from different regions and with different accents, conducts targeted general speech recognition model training, uses transfer learning to migrate the existing general speech recognition model to the recognition task of a specific accent, and fine-tunes the parameters of the general speech recognition model to adapt to accent changes.

[0105] Among them, the specific processing process of E) speech rate processing is H) using speech rate analysis software and algorithm.

[0106] Specifically, a speech recognition model can be established, and the dynamic time warping (DTW) algorithm can be used to process speech at different speech speeds. By segmenting and analyzing the speech signal of the complete call data, the time length and number of syllables of each speech segment can be calculated, and then the speech speed of the party being collected can be determined. The parameters of the speech recognition model can be adjusted according to the different speech speeds of the party being collected.

[0107] Among them, the specific processing process of F) background noise processing is I) using a noise reduction algorithm.

[0108] Specifically, a noise reduction model can be established to remove the background noise of the complete call data. The noise reduction algorithms included in the noise reduction model include noise reduction based on spectrum subtraction, adaptive filtering noise reduction, and noise reduction methods based on deep learning. Among them, spectrum subtraction can subtract the noise channel from the original speech spectrum by estimating the spectrum of the noise. Adaptive filtering, such as the least mean square (LMS) adaptive filter, adjusts the filter coefficients according to the error between the input signal and the expected signal to achieve the purpose of noise reduction. The deep learning method uses a convolutional neural network (CNN) or a recursive neural network (RNN) for noise suppression, and trains the model with a large amount of noisy speech data so that it can automatically remove noise.

[0109] J) Natural Language Processing: Building models for intent and sentiment (i.e., state) recognition.

[0110] Specifically, a target intention recognition model and a target state recognition model are constructed, and natural language processing is performed on the processed complete call text to identify the secondary detection intention and secondary detection state of the collection party. The construction process of the target intention recognition model is the same as above.

[0111] K) Generate response strategies (i.e., secondary collection strategies).

[0112] Specifically, according to the secondary detection intention and secondary detection status of the party being collected, the corresponding secondary collection strategy is matched in the preset collection strategy matching rules, so as to arrange experienced business personnel to conduct secondary collection.

[0113] L) Analyze call data characteristics.

[0114] Specifically, relevant features can be extracted from the complete call data. For example, natural language processing technology can be used to extract text features, such as word frequency, word vector, topic model, etc. For voice features, tone, volume, speech speed, etc. can be extracted to obtain the collection effect features, communication effect features, call duration features and state change features of the complete call data. Among them, the collection effect features can be used as indicators to measure whether the collection is successful. The communication effect features can include the language expression ability of the business personnel and the effectiveness of the current collection strategy. For example, whether the business personnel can accurately answer the questions of the collection party, whether they can effectively persuade the collection party to repay, etc. The call duration feature can be used to characterize whether the business personnel and the collection party have fully communicated and the corresponding collection effect. It can be understood that a longer call duration may indicate that the business personnel and the collection party have fully communicated, but an excessively long call duration may also indicate that the collection effect is not good. The state change feature can be used to characterize the state change of the collection party. It can be understood that if the status of the party being collected changes from active cooperation to passive resistance, it means that there are problems in the current collection process; if the status of the party being collected changes from passive resistance to active cooperation, it means that the current collection process is handled properly.

[0115] N) Construct a collection quality scoring model.

[0116] Specifically, machine learning algorithms such as linear regression, decision trees or random forests can be used to build a collection quality scoring model based on the collection effect characteristics, communication effect characteristics, call duration characteristics and status change characteristics of the extracted complete call data. Among them, linear regression can be used to predict the collection quality score based on multiple features; decision trees and random forests can handle nonlinear relationships and build a scoring model by partitioning the data.

[0117] O) Get a collection quality score.

[0118] Specifically, the complete call data is preprocessed and feature extracted, and the collection effect features, communication effect features, call duration features, and state change features of the complete call data are input into the collection quality scoring model. The model performs calculations based on the input features and outputs the current collection quality score. For example, a trained linear regression model can be used, with the collection effect features, communication effect features, call duration features, and state change features of the complete call data as input, and a comprehensive collection quality score can be calculated based on the weights.

[0119] P) Evaluate the collection quality score content.

[0120] Specifically, the specific content corresponding to the current collection quality score is evaluated, and the current collection quality score is fed back to the preset collection strategy matching table, target state recognition model or target intent recognition model. The model preset collection strategy matching table, target state recognition model or target intent recognition model is iteratively optimized to improve the accuracy and effectiveness of the collection strategy.

[0121] Q) End.

[0122] Complete the intelligent quality inspection process.

[0123] This solution uses the model capability to provide business personnel with accurate collection speech suggestions, greatly enhancing the decision-making ability of business personnel and significantly enhancing decision support; it also improves the professional knowledge level of collection seats, optimizes call quality, standardizes telephone collection behavior, reduces complaints from the collection party, and significantly improves call quality; in addition, it also enhances the overall intelligence and automation of manual collection work and improves collection efficiency; this solution fully utilizes the powerful capabilities of the model by reshaping the work mode of business personnel at collection seats or business personnel who conduct post-quality inspections, optimizes the entire collection business process, and integrates the model's advanced functions in natural language understanding, reasoning summary, and context association. This solution can effectively reduce call time by 10%, while significantly improving the work efficiency of business personnel by between 15% and 20%, directly promoting the effectiveness of collection business and increasing the collection rate, thereby bringing more significant performance improvements to financial institutions.

[0124] Embodiment 3

[0125] Figure 4 This is a schematic diagram of the structure of a collection strategy generation device provided in Embodiment 3 of the present invention. The embodiment of the present invention can be applied to the case where a collection strategy is automatically generated during a telephone collection process. The device can execute a collection strategy generation method. The device can be implemented in the form of hardware and / or software. The device can be configured in an electronic device that carries the collection strategy generation function, such as a collection party terminal or a server.

[0126] See also Figure 4The collection strategy generation device shown includes: a call data acquisition module 410, a state intention recognition module 420 and a collection strategy generation module 430. The call data acquisition module 410 is used to obtain the current call data and the historical associated call data during the telephone collection process; wherein the historical associated call data is the historical call data associated with the current call data; the state intention recognition module 420 is used to use the pre-trained target state recognition model and target intention recognition model to recognize the current call data and the historical associated call data to obtain the current state and current intention of the current telephone collection process; the collection strategy generation module 430 is used to obtain the preset collection strategy matching rules, match the current collection strategy in the preset collection strategy matching rules according to the current state and the current intention, and feedback to the business personnel, so that the business personnel can collect according to the current collection strategy during the current telephone collection process.

[0127] The technical solution of the embodiment of the present invention obtains the current call data and the historically associated call data during the current telephone collection process, adopts a pre-trained target state recognition model and a target intention recognition model to recognize the current call data and the historically associated call data, obtains the current state and the current intention, obtains the preset collection strategy matching rules, matches the current collection strategy in the preset collection strategy matching rules according to the current state and the current intention, takes into account the state and intention of the party being collected during the telephone collection process, realizes the automatic generation of the collection strategy, and feeds back the current collection strategy to the business personnel so that the business personnel can collect according to the current collection strategy, thereby improving the collection efficiency and quality of the business personnel, thereby ensuring the financial security of the financial institution.

[0128] In an optional embodiment of the present invention, the device further includes: a sample acquisition module, which is used to obtain historical call samples and historical associated call samples in the historical telephone collection process before the current state and current intention are identified by using the pre-trained target state recognition model and target intention recognition model to obtain the current call data and the historical associated call data, and to obtain an initial state recognition model and a state vocabulary; wherein the historical associated call samples are other historical call samples associated with the historical call samples; the state vocabulary includes alternative state vocabulary corresponding to different alternative state types; a historical state type word frequency detection module, which is used to detect the historical state of each alternative state type in the historical call samples. The module comprises a vocabulary detection module, which is used to determine the historical state type word frequency of each candidate state type in the historical call sample; a historical actual state determination module, which is used to determine the historical actual state of the historical telephone collection process according to the historical state type word frequency of each candidate state type in the historical call sample; a historical prediction state detection module, which is used to input the historical call sample and the historical associated call sample into the initial state recognition model to obtain the historical prediction state of the historical telephone collection process; a target state recognition model generation module, which is used to adjust the parameters of the initial state recognition model according to the difference between the historical prediction state and the historical actual state of the historical telephone collection process to obtain the target state recognition model.

[0129] In an optional embodiment of the present invention, the alternative state types include active cooperation type and passive resistance type; accordingly, the historical actual state determination module includes: a negative punctuation acquisition unit, used to acquire negative punctuation; a historical negative punctuation frequency detection unit, used to detect the historical negative punctuation frequency of the negative punctuation that appears simultaneously with the historical state vocabulary of the negative resistance type in the historical call sample; a first historical state type word frequency update unit, used to integrate the historical state type word frequency of the negative resistance type and the historical negative punctuation frequency, and update the historical state type word frequency of the negative resistance type; a historical state type word frequency comparison unit, used to compare the historical state type word frequency of the active cooperation type and the historical state type word frequency of the negative resistance type in the historical call sample; a historical actual state determination unit, used to determine the alternative state type corresponding to the maximum value of the historical state type word frequency as the historical actual state of the historical telephone collection process.

[0130] In an optional embodiment of the present invention, the first historical state type word frequency updating unit includes: a historical negative sentence structure detection subunit, which is used to perform negative sentence structure detection on the historical call sample to obtain a historical negative sentence structure detection result; a first historical state type word frequency updating subunit, which is used to integrate the negative and conflicting historical state type word frequency, the historical negative punctuation frequency and the historical negative sentence detection result to update the negative and conflicting historical state type word frequency.

[0131] In an optional embodiment of the present invention, the historical actual state determination module also includes: a historical state change detection unit, which is used to detect the state change of the historical call sample and the historical associated call sample before comparing the historical state type word frequency of the active cooperation type and the historical state type word frequency of the passive resistance type in the historical call sample to obtain the historical state change detection result of the historical call sample; a second historical state type word frequency update unit, which is used to integrate the historical state type word frequency of the active cooperation type and the historical state change detection result when the historical state change detection result is a change from passive resistance type to active cooperation type, and update the historical state type word frequency of the active cooperation type; a third historical state type word frequency update unit, which is used to integrate the historical state type word frequency of the passive resistance type and the historical state change detection result when the historical state change detection result is a change from active cooperation type to passive resistance type, and update the historical state type word frequency of the passive resistance type.

[0132] In an optional embodiment of the present invention, the device also includes: a complete call data acquisition module, which is used to acquire complete call data after the current telephone collection process is completed; wherein the complete call data includes the current call data, the historical associated call data, the current collection data of the current call data, and the current feedback data of the current collection data; a complete call data feature extraction module, which is used to extract features from the complete call data to obtain the collection effect features, communication effect features, call duration features, and state change features of the complete call data; a collection quality scoring module, which is used to use a collection quality scoring model to detect the collection effect features, communication effect features, call duration features, and state change features of the complete call data to determine the current collection quality score of the current telephone collection process; a rule model adjustment module, which is used to adjust the preset collection strategy matching rules, the target state recognition model, or the target intention recognition model according to the current collection score of the complete call data.

[0133] The collection strategy generation device provided in the embodiment of the present invention can execute the collection strategy generation method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0134] In the technical solution of the embodiment of the present invention, the collected information is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant countries and regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0135] Embodiment 4

[0136] According to an embodiment of the present invention, the present invention also provides an electronic device, a readable storage medium and a computer program product.

[0137] Figure 5 A schematic diagram of the structure of an electronic device 500 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0138] like Figure 5 As shown, the electronic device 500 includes at least one processor 501, and a memory connected to the at least one processor 501 in communication, such as a read-only memory (ROM) 502, a random access memory (RAM) 503, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 501 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 502 or the computer program loaded from the storage unit 508 to the random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the electronic device 500 can also be stored. The processor 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0139] Multiple components in the electronic device 500 are connected to the I / O interface 505, including: an input unit 506, such as a keyboard, a mouse, etc.; an output unit 507, such as various types of displays, speakers, etc.; a storage unit 508, such as a disk, an optical disk, etc.; and a communication unit 509, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 509 allows the electronic device 500 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0140] Processor 501 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of processor 501 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. Processor 501 executes the various methods and processes described above, such as a collection strategy generation method.

[0141] In some embodiments, the collection strategy generation method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 508. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 500 via the ROM 502 and / or the communication unit 509. When the computer program is loaded into the RAM 503 and executed by the processor 501, one or more steps of the collection strategy generation method described above may be performed. Alternatively, in other embodiments, the processor 501 may be configured to execute the collection strategy generation method in any other appropriate manner (e.g., by means of firmware).

[0142] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0143] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0144] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0145] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0146] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0147] The computing system may include a debt collector end and a server. The debt collector end and the server are generally remote from each other and usually interact through a communication network. The relationship between the debt collector end and the server is generated by computer programs running on corresponding computers and having a debt collector end-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS (Virtual Private Server) services.

[0148] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0149] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A collection strategy generation method, characterized in that: The method comprises: In the current telephone collection process, current call data and historical associated call data are obtained; wherein the historical associated call data is historical call data associated with the current call data; Using a pre-trained target state recognition model and a target intention recognition model, the current call data and the historical associated call data are recognized to obtain a current state and a current intention; Obtain preset collection strategy matching rules, match the current collection strategy in the preset collection strategy matching rules according to the current state and the current intention, and provide feedback to business personnel so that the business personnel can collect according to the current collection strategy during the current telephone collection process.

2. The method according to claim 1, characterized in that Before the pre-trained target state recognition model and target intention recognition model are used to recognize the current call data and the historical associated call data to obtain the current state and the current intention, the method further includes: Obtaining historical call samples and historical associated call samples in the historical telephone collection process, and obtaining an initial state recognition model and a state vocabulary; wherein the historical associated call samples are other historical call samples associated with the historical call samples; and the state vocabulary includes alternative state vocabulary corresponding to different alternative state types; Detecting the historical state words of each candidate state type in the historical call samples to determine the historical state type word frequency of each candidate state type in the historical call samples; Determining the historical actual state of the historical telephone collection process according to the historical state type word frequency of each candidate state type in the historical call sample; Inputting the historical call samples and the historical associated call samples into the initial state recognition model to obtain the historical predicted state of the historical telephone debt collection process; According to the difference between the historical predicted state and the historical actual state of the historical telephone collection process, the initial state recognition model is adjusted to obtain a target state recognition model.

3. The method according to claim 2, characterized in that The alternative state types include active cooperation type and passive resistance type; Accordingly, determining the historical actual state of the historical telephone collection process according to the historical state type word frequency of each candidate state type in the historical call sample includes: Get negative punctuation; Detecting the historical negative punctuation mark frequency of the negative punctuation mark that appears simultaneously with the historical state words of negative conflict type in the historical call samples; The negative resistance type historical state type word frequency and the historical negative punctuation mark frequency are integrated to update the negative resistance type historical state type word frequency; Comparing the word frequencies of the historical state type of the active cooperation type and the word frequencies of the historical state type of the passive resistance type in the historical call samples; The candidate state type corresponding to the maximum value of the historical state type word frequency is determined as the historical actual state of the historical telephone collection process.

4. The method according to claim 3, characterized in that: The step of integrating the negative conflict type historical state type word frequency and the historical negative punctuation mark frequency to update the negative conflict type historical state type word frequency includes: Performing negative sentence structure detection on the historical call samples to obtain historical negative sentence structure detection results; The negative conflicting historical state type word frequency, the historical negative punctuation mark frequency and the historical negative sentence detection result are integrated to update the negative conflicting historical state type word frequency.

5. The method according to claim 3, characterized in that: Before comparing the historical state type word frequencies of the active cooperation type and the historical state type word frequencies of the passive resistance type in the historical call samples, the method further includes: Performing status change detection on the historical call samples and the historical associated call samples to obtain historical status change detection results of the historical call samples; When the historical state change detection result is a change from a passive resistance type to an active cooperation type, the historical state type word frequency of the active cooperation type and the historical state change detection result are integrated to update the historical state type word frequency of the active cooperation type; When the historical state change detection result is a change from active cooperation type to passive resistance type, the historical state type word frequency of the passive resistance type and the historical state change detection result are integrated to update the historical state type word frequency of the passive resistance type.

6. The method according to claim 1, characterized in that Also includes: After the current telephone collection process is completed, complete call data is obtained; wherein the complete call data includes the current call data, the historical associated call data, the current collection data of the current call data, and the current feedback data of the current collection data; Extracting features from the complete call data to obtain collection effect features, communication effect features, call duration features, and state change features of the complete call data; Using a collection quality scoring model, the collection effect characteristics, communication effect characteristics, call duration characteristics and state change characteristics of the complete call data are detected to determine a current collection quality score of the current telephone collection process; The preset collection strategy matching rule, the target state recognition model or the target intention recognition model is adjusted according to the current collection score of the complete call data.

7. A collection strategy generation device, characterized in that: The device comprises: A call data acquisition module, used to acquire current call data and historical associated call data during the telephone collection process; wherein the historical associated call data is historical call data associated with the current call data; A state intention recognition module, used to use a pre-trained target state recognition model and a target intention recognition model to recognize the current call data and the historical associated call data to obtain a current state and a current intention; The collection strategy generation module is used to obtain preset collection strategy matching rules, match the current collection strategy in the preset collection strategy matching rules according to the current state and the current intention, and provide feedback to the business personnel so that the business personnel can collect according to the current collection strategy during the current telephone collection process.

8. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the collection strategy generation method described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the collection strategy generation method described in any one of claims 1 to 6 when executed.

10. A computer program product, characterized in that The computer program product comprises a computer program, which, when executed by a processor, implements the collection strategy generation method according to any one of claims 1 to 6.

Images