Call-out method based on artificial intelligence
By analyzing the historical data of customers and collection specialists, optimizing dialing periods and conversation strategies, and matching the optimal collection specialists, the problem of inefficiency in artificial intelligence outgoing call systems when making calls is solved, and the collection efficiency and customer satisfaction are improved.
Patent Information
- Application Number
- CN202411273513.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-12
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-09-12
AI Technical Summary
The existing artificial intelligence outgoing call system is inefficient when making calls, and the customer's improper answering time leads to reduced efficiency, and lacks personalized dialogue processing strategies, which affects collection efficiency.
By analyzing customer historical call records and historical data of collection specialists, optimizing dialing periods and opening formats, matching the optimal collection specialists, realizing dialogue record matching and business priority allocation, and improving collection efficiency and customer satisfaction.
Improve the collection efficiency and customer answering quality, optimize resource allocation, and ensure efficient debt recovery and customer satisfaction.
Smart Images

Figure CN118972495A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of intelligent outbound calling, and in particular to an outbound calling method based on artificial intelligence. Background Art
[0002] AI outbound calling is an automated telephone calling system implemented using artificial intelligence technology. These systems can automatically make calls and communicate with humans through preset scripts or natural language processing technology. AI outbound calling systems need to understand and generate natural language. Through NLP technology, the system can parse the user's language input, understand their intentions, and generate corresponding responses. ASR technology can convert voice signals into text for further processing by the system. This is a key technology that enables the system to understand human speech. In order to communicate verbally with users, AI outbound calling systems also need to convert text into speech. This process is achieved through TTS technology, making the machine-generated speech clear and natural. AI outbound calling systems integrate multiple AI technologies such as speech recognition, language understanding, and dialogue management, and are widely used in customer service, marketing, information notifications, and survey research.
[0003] The existing outbound calling system requires a large number of calls to be made within a certain period of time, and the calling time is random. When the calling time happens to be when the customer is not convenient to answer the call, the customer development efficiency is reduced. Different opening remarks will also cause different lengths of time for customers to answer the call, which in turn reduces the outbound calling efficiency.
[0004] This solution proposes to determine the set of time periods when the customer answers calls based on the customer's historical call records, count the effectiveness of each opening format, and update the opening format set; at the same time, match the best debt specialist based on the collection specialist's pending business set. Summary of the invention
[0005] The present invention provides an outbound calling method based on artificial intelligence, which is used to promote the solution of the problems mentioned in the above background technology.
[0006] The present invention provides the following technical solution: an outbound calling method based on artificial intelligence, comprising:
[0007] Obtain historical call records of overdue customers;
[0008] According to the historical call records, the best time period strategy is adopted to obtain the calling time period set of overdue customers;
[0009] Get all the opening format words of outbound calls to form an opening format set;
[0010] According to the opening format set and historical call records, an opening format language optimization strategy is adopted to update the opening format set;
[0011] Obtain historical reception data of all debt collection specialists;
[0012] The ranking of debt collectors who record business is the professional ranking;
[0013] Based on historical reception data, a professional scoring strategy is adopted to obtain the professional ranking of each business;
[0014] Obtain the importance levels of all businesses and form a business ranking;
[0015] Record the text of overdue customers and outbound calls as conversation records;
[0016] When an overdue customer needs to be transferred to a collection specialist:
[0017] Form a transfer decision strategy based on the conversation records;
[0018] Execute the transfer judgment strategy and obtain a set of candidate collection specialists;
[0019] Obtain the business of overdue customer consultation when transferring to the collection specialist, which is recorded as the transferred business;
[0020] Obtain the business that has been assigned to each element in the candidate collection specialist set, and obtain the pending business set of each element in the candidate collection specialist set;
[0021] Based on the set of candidate collection specialists and the set of pending businesses, the optimal allocation strategy is adopted to assign the best collection specialist to overdue customers.
[0022] Optionally, the adopting of the optimal time period strategy includes:
[0023] Set the first time, which refers to Monday to Friday;
[0024] Set the second time, which is from Saturday to Sunday;
[0025] The total number of first-time answering times in the historical call records of overdue customers is counted and recorded as the number of first-time answering times;
[0026] The total number of second-time answering calls in the historical call records of overdue customers is counted and recorded as the number of second-time answering calls;
[0027] When the number of first-time answers is greater than the number of second-time answers, the overdue customers are classified as the first group;
[0028] When the number of calls answered at the first time is less than or equal to the number of calls answered at the second time, the overdue customers are classified as the second group of people;
[0029] Set the number of filters for selecting the calling time period;
[0030] Divide 24 hours a day into e time periods, count the number of times overdue customers answer calls in each time period in the historical call records, obtain the time periods with a number greater than the screening number, and form a call time period set;
[0031] When the overdue customer belongs to the first group of people, the overdue customer is called at the first time in the time period included in the calling time period set every day;
[0032] When the overdue customer belongs to the second group of people, a call is made to the overdue customer during a time period included in the dialing time period set every day at the second time.
[0033] Optionally, the opening format optimization strategy includes:
[0034] Set the threshold time for judging whether the opening format words are retained;
[0035] Get any element in the opening format set and record it as the marked opening format;
[0036] The length of the call when the opening format is marked is recorded as the marked time;
[0037] Get all the marked times in the historical call records to form a marked time set;
[0038] Traverse all elements in the marked time set, obtain the number of marked times greater than the threshold time, and record it as the effective number;
[0039] Calculate the number of elements in the marked time set and record it as the number of dialing times;
[0040] Calculate the effective number of times divided by the number of dialing times, and record the result as the effective rate;
[0041] Set the judgment ratio for whether to keep the opening format;
[0042] When the effective rate is less than or equal to the judgment ratio, the marked opening format words are removed from the opening format set;
[0043] Traverse all elements in the opening format set to execute the opening format language optimization strategy and update the opening format set.
[0044] Optionally, the professional scoring strategy may include:
[0045] Get all the businesses and form a business set;
[0046] Get any element in the service set and record it as the positioning service;
[0047] Obtain the scoring data of each collection specialist in the positioning business in the historical reception data, calculate the average of multiple scoring data of each collection specialist, and obtain the average score of each collection specialist in the positioning business;
[0048] The average scores of all collection specialists in the location service are sorted from high to low to obtain the professional ranking of the location service;
[0049] Traverse all elements in the business set, adopt professional scoring strategy, and obtain the professional ranking of each element in the business set.
[0050] Optionally, adopting the optimal allocation strategy includes:
[0051] Get the location of overdue customer phone numbers;
[0052] Obtain the collection specialists whose hometown and place of origin are the same from the candidate collection specialist set, and form a collection specialist set with the same place of origin;
[0053] Acquire all services that are higher in importance than the transfer service in the service level ranking to form a priority service set;
[0054] Acquire all services whose importance level is lower than that of the transfer service in the service level ranking to form a lagging service set;
[0055] The collection specialist whose elements in the pending business set are all in the priority business set is recorded as the discarded collection specialist;
[0056] The collection specialist whose elements in the pending business set are not in the priority business set is the first collection specialist;
[0057] The collection specialist whose elements in the pending business set are in both the priority business set and the delayed business set is recorded as the second collection specialist;
[0058] Traverse the pending business set of each element in the collection specialist set of the same place, obtain all the first collection specialists and the second collection specialists, and form the first collection set and the second collection set respectively;
[0059] Set a limit value to limit the number of businesses that collection specialists can receive;
[0060] Count the number of elements in the pending business set of each collection specialist in the first set to obtain the first set of processed business;
[0061] Compare each element in the first handling number set with the limit value, obtain the collection specialists corresponding to all elements whose value is less than the limit value, and record them as the first preferred specialist set;
[0062] Count the number of elements in the pending business set of each collection specialist in the second set of actions to be taken, and obtain a second set of processing numbers;
[0063] Compare each element in the second handling number set with the limit value, obtain the collection specialists corresponding to all elements whose value is less than the limit value, and record them as the second preferred specialist set.
[0064] Optionally, the adopting of the optimal allocation strategy further includes:
[0065] Set the initial order, which is the order of all pending transactions when the transaction is first assigned to a collection specialist;
[0066] Set the execution order, which is the order of the business among all the businesses to be processed by the collection specialist;
[0067] Set the number of judgments to determine whether the collection specialist is qualified to receive overdue customers;
[0068] Obtain historical data of the execution order of the determination quantity, recorded as a historical execution order set;
[0069] Compare all elements in the historical execution order set with the initial order:
[0070] When the initial order is less than or equal to all elements in the historical execution order set, the collection specialist to which the business belongs is recorded as the marked specialist.
[0071] Optionally, the adopting of the optimal allocation strategy further includes:
[0072] Traverse the pending business set of each element of the first preferred specialist set and the second preferred specialist set, obtain all marked specialists, and form a discarded specialist set;
[0073] Remove elements in the discarded specialist set from the first preferred specialist set and the second preferred specialist set to obtain a first target specialist set and a second target specialist set;
[0074] The collection specialist who responds to overdue customers is recorded as the responding collection specialist;
[0075] Obtain the highest professional ranking of the collection specialist who transfers the business among all the collection specialists who respond, and record him as the responding collection specialist;
[0076] When the responding collection specialist is an element in the first target specialist set, the overdue customer transfer business is assigned to the responding collection specialist, and the initial order of the transfer business is first;
[0077] When the responding debt collection specialist is an element in the second target specialist set;
[0078] Obtain the elements in the pending business set of the answering collection specialist that have a higher importance level than the transferred business, and record them as the response priority business set;
[0079] Get the largest execution order of the response priority service set, recorded as the connection order x;
[0080] The transfer business of overdue customers is assigned to the responding collection specialist, and the initial order of the transfer business is x+1.
[0081] Optionally, the executing the transfer determination strategy includes:
[0082] Extract keywords from the conversation records to form a keyword set;
[0083] Get the number of elements in the keyword set, and record the result as the number of keywords a;
[0084] Distribute the conversation records to all collection agents;
[0085] Obtain the characteristic words extracted from the conversation records of each debt collection specialist;
[0086] A feature matching strategy is adopted for each collection specialist’s feature words, including:
[0087] Obtain the characteristic words of any collection specialist and record them as the positioning characteristic word set;
[0088] Compare the elements in the positioning feature word set and the keyword set;
[0089] Obtaining the same elements in the positioning feature word set and the keyword set to form a matching feature set;
[0090] Calculate the number of elements in the matching feature set, and record the result as the matching number b;
[0091] Calculate the number of matches ÷ the number of keywords, and record the result as the match rate c, c = b ÷ a;
[0092] When the matching rate is equal to 1, the collection specialist corresponding to the positioning feature word set is recorded as the matching collection specialist;
[0093] Get all matching debt collectors to form a candidate debt collector set.
[0094] Optionally, the conversation record includes:
[0095] First, convert the voice input of the debt collection customer into input text, including:
[0096] Audio collection: obtaining the voice input by the debt collection customer, converting the voice into simulated continuous sound waves and recording them as voice signals;
[0097] Preprocessing, obtaining a speech signal, performing pre-emphasis, framing and windowing operations, and recording the result after preprocessing as a speech frame set;
[0098] Extract features, obtain a speech frame set, extract feature parameters of each element in the speech frame set, and form a feature parameter set;
[0099] Apply the acoustic model, obtain a feature parameter set, traverse each element in the feature parameter set, apply the acoustic model to each element, obtain the phoneme corresponding to each speech frame, and form a sound result set;
[0100] Applying the language model to obtain a speech frame set, applying the language model to obtain multiple text sequences corresponding to the speech frame set to form a text result set;
[0101] Search and decode to obtain the sound result set and the text result set, apply the decoding search algorithm, and record the obtained text sequence as the input text;
[0102] Secondly, obtain the tone, expression, and speech speed information of the debt collection customer and record it as implicit information;
[0103] Obtain input text, apply natural language processing algorithms to the input text and implicit information, obtain the emotions and repayment willingness of the collection customers, and record them as communication features;
[0104] Finally, the input text and communication features are obtained, and the reply content is matched and recorded as the output text;
[0105] Perform language conversion operations on the output text, including:
[0106] Standardize text, remove punctuation and format numbers from the output text, and record the result as standard text;
[0107] Prosody modeling: obtain standard text, analyze the intonation, rhythm and emphasis of the standard text, and record the results as text features;
[0108] Apply the acoustic model to convert the standard text into phonemes, and then convert the phonemes into sound wave signals, and the results are recorded as output sound waves;
[0109] Synthesize language, obtain output sound waves and text features, apply sound synthesis technology, and obtain speech output.
[0110] The present invention has the following beneficial effects:
[0111] 1. This AI-based outbound calling method sets the first time and the second time. Count and compare the number of times overdue customers answer calls at these two times. Classify overdue customers according to the time with the most calls. Determine the high-call time period every day and select the calling time period from it. By classifying overdue customers and selecting the time period with high call answering rate, the contact success rate can be increased, thereby improving the collection efficiency. Customized calling strategies can avoid communication during inefficient time periods, save resources and optimize the collection process.
[0112] 2. This AI-based outbound calling method sets a threshold time for call effectiveness, counts the call duration of various opening phrases, evaluates each opening phrase based on its effectiveness, and determines whether to retain it based on its efficiency. Evaluating the effectiveness of opening phrases can help identify and eliminate ineffective conversational methods, ensuring that the opening phrases that are most likely to elicit responses and communication from overdue customers are used. Optimizing the opening format set can improve customer listening quality and satisfaction, which may increase repayment willingness.
[0113] 3. This AI-based outbound calling method obtains the score data of the collection specialists in each business, and each collection specialist calculates the average score based on the score data of the business. The collection specialists are professionally ranked according to the average score. Professional ranking can ensure that each business is handled by the collection specialist who is best at the business, thereby improving the efficiency and quality of collection. The scoring and ranking system can also motivate collection specialists to improve their business processing capabilities.
[0114] 4. This AI-based outbound calling method selects a collection specialist based on the location of the overdue customer's phone call, obtains a collection of collection specialists in the same location, classifies the priority of the business, and assigns collection specialists according to the priority. Regional matching can enhance the communication efficiency between collection specialists and overdue customers, because people in the same region may have better communication fit. By giving priority to high-level business, resource allocation can be optimized and the recovery speed of high-priority debts can be accelerated. Setting a limit on the number of business processing can prevent individual collection specialists from overloading their work and ensure the quality of collection work and the work efficiency of collection specialists.
[0115] 5. This AI-based outbound calling method sets the initial order and execution order of the business in the business to be processed by the collection specialist. Set the number of judgments to determine whether the collection specialist is suitable for receiving overdue customers. Based on the comparison between the historical execution order and the initial order, mark the collection specialist who is suitable for handling the business to avoid the long-term non-processing of low-priority business due to the repeated occurrence of high-priority business. This matching strategy ensures that the business is correctly prioritized according to importance and urgency, while taking into account the situation where low-priority business is not processed for a long time.
[0116] 6. The outbound calling method based on artificial intelligence obtains a set of discarded specialists, removes the elements in the first preferred specialist set and the second preferred specialist set from the discarded specialist set, and obtains a first target specialist set and a second target specialist set, wherein the elements in the target specialist set are qualified to receive overdue customers. The transferred business is allocated according to the response speed and professional ranking of each collection specialist. By accurately selecting collection specialists with outstanding performance, customer satisfaction and collection efficiency can be improved. The transferred business is allocated to the answering collection specialist, and the processing priority of the transferred business is higher than the pending business in the answering collection specialist's pending business set whose business level is lower than the transferred business. Appropriate business allocation ensures efficient processing and reduces processing delays and customer dissatisfaction.
[0117] 7. This AI-based outbound calling method extracts keywords from the conversation and matches them. The most suitable collection specialist is selected to handle a specific conversation based on the matching results. Keyword matching allows collection specialists to focus more on the types of conversations they are best at handling, improving the professionalism and efficiency of case handling. Accurate matching ensures the pertinence and effectiveness of the collection process and reduces ineffective communication.
[0118] 8. This AI-based outbound calling method realizes full-process automation from voice input to text processing and then to voice output. It can quickly respond to user requests, reduce waiting time, and improve processing efficiency, thereby greatly improving user satisfaction and system throughput. BRIEF DESCRIPTION OF THE DRAWINGS
[0119] Figure 1 Schematic diagram of the method of the present invention. DETAILED DESCRIPTION
[0120] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are 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 are within the scope of protection of the present invention.
[0121] Embodiment 1, refer to Figure 1 , an outbound calling method based on artificial intelligence, comprising,
[0122] Obtain historical call records of overdue customers;
[0123] According to the historical call records, the best time period strategy is adopted to obtain the calling time period set of overdue customers;
[0124] Get all the opening format words of outbound calls to form an opening format set;
[0125] According to the opening format set and historical call records, an opening format language optimization strategy is adopted to update the opening format set;
[0126] Obtain historical reception data of all debt collection specialists;
[0127] The ranking of debt collectors who record business is the professional ranking;
[0128] Based on historical reception data, a professional scoring strategy is adopted to obtain the professional ranking of each business;
[0129] Obtain the importance levels of all businesses and form a business ranking;
[0130] Record the text of overdue customers and outbound calls as conversation records;
[0131] When an overdue customer needs to be transferred to a collection specialist:
[0132] Form a transfer decision strategy based on the conversation records;
[0133] Execute the transfer judgment strategy and obtain a set of candidate collection specialists;
[0134] Obtain the business of overdue customer consultation when transferring to the collection specialist, which is recorded as the transferred business;
[0135] Obtain the business that has been assigned to each element in the candidate collection specialist set, and obtain the pending business set of each element in the candidate collection specialist set;
[0136] Based on the set of candidate collection specialists and the set of pending businesses, the optimal allocation strategy is adopted to assign the best collection specialist to overdue customers.
[0137] The optimal time period strategy includes:
[0138] Set the first time, which refers to Monday to Friday;
[0139] Set the second time, which is from Saturday to Sunday;
[0140] The total number of first-time answering times in the historical call records of overdue customers is counted and recorded as the number of first-time answering times;
[0141] The total number of second-time answering calls in the historical call records of overdue customers is counted and recorded as the number of second-time answering calls;
[0142] When the number of first-time answers is greater than the number of second-time answers, the overdue customers are classified as the first group;
[0143] When the number of calls answered at the first time is less than or equal to the number of calls answered at the second time, the overdue customers are classified as the second group of people;
[0144] Set the number of filters for selecting the calling time period;
[0145] Divide 24 hours a day into e time periods, count the number of times overdue customers answer calls in each time period in the historical call records, obtain the time periods with a number greater than the screening number, and form a call time period set;
[0146] When the overdue customer belongs to the first group of people, the overdue customer is called at the first time in the time period included in the calling time period set every day;
[0147] When the overdue customer belongs to the second group of people, a call is made to the overdue customer during a time period included in the dialing time period set every day at the second time.
[0148] Set the first time and the second time. Count and compare the number of times overdue customers answer calls during these two times. Classify overdue customers according to the time with the most calls. Determine the high-call time period every day and select the calling time period from there. By classifying overdue customers and selecting the time period with the high call rate, you can increase the contact success rate and thus improve the collection efficiency. Customized calling strategies can avoid communicating during inefficient time periods, save resources and optimize the collection process.
[0149] The opening format optimization strategy includes:
[0150] Set the threshold time for judging whether the opening format words are retained;
[0151] Get any element in the opening format set and record it as the marked opening format;
[0152] The length of the call when the opening format is marked is recorded as the marked time;
[0153] Obtain all marked times in the historical call records of overdue customers to form a marked time set;
[0154] Traverse all elements in the marked time set, obtain the number of marked times greater than the threshold time, and record it as the effective number;
[0155] Calculate the number of elements in the marked time set and record it as the number of dialing times;
[0156] Calculate the effective number of times divided by the number of dialing times, and record the result as the effective rate;
[0157] Set the judgment ratio for whether to keep the opening format;
[0158] When the effective rate is less than or equal to the judgment ratio, the marked opening format words are removed from the opening format set;
[0159] Traverse all elements in the opening format set to execute the opening format language optimization strategy and update the opening format set.
[0160] Set the threshold time for call effectiveness, count the call duration of various opening phrases, evaluate each opening phrase based on the effectiveness of the call duration, and judge whether to retain the opening phrase based on efficiency. Evaluating the effectiveness of opening phrases can help identify and eliminate ineffective conversation methods, and ensure that the opening phrases that are most likely to elicit responses and communication from overdue customers are used. Optimizing the opening format set can improve customer answering quality and satisfaction, which may increase repayment willingness.
[0161] The professional scoring strategy adopted includes:
[0162] Get all the businesses and form a business set;
[0163] Get any element in the service set and record it as the positioning service;
[0164] Obtain the scoring data of each collection specialist in the positioning business in the historical reception data, calculate the average of multiple scoring data of each collection specialist, and obtain the average score of each collection specialist in the positioning business;
[0165] The average scores of all collection specialists in the location service are sorted from high to low to obtain the professional ranking of the location service;
[0166] Traverse all elements in the business set, adopt professional scoring strategy, and obtain the professional ranking of each element in the business set.
[0167] The collection specialists' score data for each business is obtained, and each collection specialist calculates the average score based on the score data of the business. The collection specialists are professionally ranked according to the average score. Professional ranking can ensure that each business is handled by the collection specialist who is most proficient in that business, thereby improving the efficiency and quality of collection. The scoring and ranking system can also motivate collection specialists to improve their business processing capabilities.
[0168] The optimal allocation strategy is adopted, including:
[0169] Get the location of overdue customer phone numbers;
[0170] Obtain the collection specialists whose hometown and place of origin are the same from the candidate collection specialist set, and form a collection specialist set with the same place of origin;
[0171] Acquire all services that are higher in importance than the transfer service in the service level ranking to form a priority service set;
[0172] Acquire all services whose importance level is lower than that of the transfer service in the service level ranking to form a lagging service set;
[0173] The collection specialist whose elements in the pending business set are all in the priority business set is recorded as the discarded collection specialist;
[0174] The collection specialist whose elements in the pending business set are not in the priority business set is the first collection specialist;
[0175] The collection specialist whose elements in the pending business set are in both the priority business set and the delayed business set is recorded as the second collection specialist;
[0176] Traverse the pending business set of each element in the collection specialist set of the same place, obtain all the first collection specialists and the second collection specialists, and form the first collection set and the second collection set respectively;
[0177] Set a limit value to limit the number of businesses that collection specialists can receive;
[0178] Count the number of elements in the pending business set of each collection specialist in the first set to obtain the first set of processed business;
[0179] Compare each element in the first handling number set with the limit value, obtain the collection specialists corresponding to all elements whose value is less than the limit value, and record them as the first preferred specialist set;
[0180] Count the number of elements in the pending business set of each collection specialist in the second set of actions to be taken, and obtain a second set of processing numbers;
[0181] Compare each element in the second handling number set with the limit value, obtain the collection specialists corresponding to all elements whose value is less than the limit value, and record them as the second preferred specialist set.
[0182] Select a collection specialist based on the location of the overdue customer's phone call, obtain a collection of collection specialists in the same location, classify the priority of the business, and assign collection specialists according to the priority. Regional matching can enhance the communication efficiency between collection specialists and overdue customers, because people in the same region may have better communication fit. By giving priority to high-level business, resource allocation can be optimized and the recovery speed of high-priority debts can be accelerated. Setting a limit on the number of business processing can prevent individual collection specialists from overloading and ensure the quality of collection work and the work efficiency of collection specialists.
[0183] The adopting of the optimal allocation strategy also includes:
[0184] Set the initial order, which is the order of all pending transactions when the transaction is first assigned to a collection specialist;
[0185] Set the execution order, which is the order of the business among all the businesses to be processed by the collection specialist;
[0186] Set the number of judgments to determine whether the collection specialist is qualified to receive overdue customers;
[0187] Obtain historical data of the execution order of the determination quantity, recorded as a historical execution order set;
[0188] Compare all elements in the historical execution order set with the initial order:
[0189] When the initial order is less than or equal to all elements in the historical execution order set, the collection specialist to which the business belongs is recorded as the marked specialist.
[0190] In this embodiment, the determination number is 4;
[0191] Set the initial order and execution order of the business in the collection specialist's pending business. Set the number of judgments to determine whether the collection specialist is suitable for receiving overdue customers. Based on the comparison of the historical execution order and the initial order, mark the collection specialist who is suitable for handling the business to avoid the long-term non-processing of low-priority business due to repeated occurrence of high-priority business. This matching strategy ensures that the business is correctly prioritized according to importance and urgency, while taking into account the situation where low-priority business is not processed for a long time.
[0192] The adopting of the optimal allocation strategy also includes:
[0193] Traverse the pending business set of each element of the first preferred specialist set and the second preferred specialist set, obtain all marked specialists, and form a discarded specialist set;
[0194] In this embodiment, there are two elements in the first preferred specialist set, which are respectively recorded as the first numbered specialist and the second numbered specialist; there are two elements in the second preferred specialist set, which are respectively recorded as the third numbered specialist and the fourth numbered specialist;
[0195] Traverse the pending business sets of the first numbered specialist, the second numbered specialist, the third numbered specialist and the fourth numbered specialist. Among them, the pending business set of the second numbered specialist has 1 element, recorded as the numbered business, and the initial order of the numbered business is 1. In the process of processing the numbered business by the second numbered specialist, the historical data of the execution order are sorted by time as 1, 2, 2, 2 respectively; obtain the historical data of the 4 execution orders to form a historical execution order set, and the elements in the historical execution order set are 1, 2, 2, 2.
[0196] The initial order is less than or equal to all elements in the historical execution order set, indicating that services with higher priority than numbering services are continuously being processed by the second numbering specialist, causing the marking services to be processed later. In this case, the second numbering specialist is the marking specialist.
[0197] Remove elements in the discarded specialist set from the first preferred specialist set and the second preferred specialist set to obtain a first target specialist set and a second target specialist set;
[0198] The collection specialist who responds to overdue customers is recorded as the responding collection specialist;
[0199] Obtain the highest professional ranking of the collection specialist who transfers the business among all the collection specialists who respond, and record him as the responding collection specialist;
[0200] When the responding collection specialist is an element in the first target specialist set, the overdue customer transfer business is assigned to the responding collection specialist, and the initial order of the transfer business is first;
[0201] When the responding debt collection specialist is an element in the second target specialist set;
[0202] Obtain the elements in the pending business set of the answering collection specialist that have a higher importance level than the transferred business, and record them as the response priority business set;
[0203] Get the largest execution order of the response priority service set, recorded as the connection order x;
[0204] The transfer business of overdue customers is assigned to the responding collection specialist, and the initial order of the transfer business is x+1.
[0205] Get the set of discarded specialists, remove the elements in the first preferred specialist set and the second preferred specialist set from the discarded specialist set, and get the first target specialist set and the second target specialist set. The elements in the target specialist set are qualified to receive overdue customers. Allocate the transferred business according to the response speed and professional ranking of each collection specialist. By accurately selecting collection specialists with excellent performance, customer satisfaction and collection efficiency can be improved. Allocate the transferred business to the answering collection specialist, and the processing priority of the transferred business is higher than the pending business in the answering collection specialist's pending business set whose business level is lower than the transferred business. Appropriate business allocation ensures efficient processing and reduces processing delays and customer dissatisfaction.
[0206] The execution of the transfer judgment strategy includes:
[0207] Extract keywords from the conversation records to form a keyword set;
[0208] Get the number of elements in the keyword set, and record the result as the number of keywords a;
[0209] Distribute the conversation records to all collection agents;
[0210] Obtain the characteristic words extracted from the conversation records of each debt collection specialist;
[0211] A feature matching strategy is adopted for each collection specialist’s feature words, including:
[0212] Obtain the characteristic words of any collection specialist and record them as the positioning characteristic word set;
[0213] Compare the elements in the positioning feature word set and the keyword set;
[0214] Obtaining the same elements in the positioning feature word set and the keyword set to form a matching feature set;
[0215] Calculate the number of elements in the matching feature set, and record the result as the matching number b;
[0216] Calculate the number of matches ÷ the number of keywords, and record the result as the match rate c, c = b ÷ a;
[0217] When the matching rate is equal to 1, the collection specialist corresponding to the positioning feature word set is recorded as the matching collection specialist;
[0218] Get all matching debt collectors to form a candidate debt collector set.
[0219] Extract keywords from the conversation and match them. Select the most suitable collection specialist to handle a specific conversation based on the matching results. Keyword matching allows collection specialists to focus more on the types of conversations they are best at handling, improving the professionalism and efficiency of case handling. Accurate matching ensures the pertinence and effectiveness of the collection process and reduces ineffective communication.
[0220] The conversation record includes:
[0221] First, convert the voice input of the debt collection customer into input text, including:
[0222] Audio collection: obtaining the voice input by the debt collection customer, converting the voice into simulated continuous sound waves and recording them as voice signals;
[0223] Preprocessing, obtaining a speech signal, performing pre-emphasis, framing and windowing operations, and recording the result after preprocessing as a speech frame set;
[0224] Pre-emphasis is used to enhance the high-frequency part of the voice signal and compensate for the high-frequency attenuation of the voice signal during transmission; framing divides the voice signal into short time periods, usually 10-30 milliseconds per frame, for subsequent processing; windowing is to reduce the discontinuity at the beginning and end of the frame.
[0225] Extract features, obtain a speech frame set, extract feature parameters of each element in the speech frame set, and form a feature parameter set;
[0226] Apply the acoustic model, obtain a feature parameter set, traverse each element in the feature parameter set, apply the acoustic model to each element, obtain the phoneme corresponding to each speech frame, and form a sound result set;
[0227] Applying the language model to obtain a speech frame set, applying the language model to obtain multiple text sequences corresponding to the speech frame set to form a text result set;
[0228] Search and decode to obtain the sound result set and the text result set, apply the decoding search algorithm, and record the obtained text sequence as the input text;
[0229] The existing ASR algorithm is used to convert the voice input of the debt collection customer into input text. The core of the ASR algorithm is the construction of the acoustic model and the language model. The acoustic model is mainly used to establish the mapping relationship between speech features and phonemes, while the language model is used to calculate the prior probability of word sequences. These two models work together to improve the accuracy and robustness of speech recognition. In this process, the commonly used feature extraction method is the Mel frequency cepstral coefficient, which imitates the auditory characteristics of the human ear and has a certain robustness in a noisy environment.
[0230] Using ASR technology, we can receive and accurately recognize the user's voice input, and accurately capture the user's needs no matter what environment the user is in. ASR technology supports multiple languages and dialects, and can process voice input from all over the world to serve a wider user group.
[0231] Secondly, obtain the tone, expression, and speech speed information of the debt collection customer and record it as implicit information;
[0232] Obtain input text, apply natural language processing algorithms to the input text and implicit information, obtain the emotions and repayment willingness of the collection customers, and record them as communication features;
[0233] Prior Art Natural Language Processing (NLP) is a series of computational methods and techniques used to understand and process human natural language. Natural language is the language used by humans for daily communication, such as English, Chinese, French, etc. The purpose of NLP algorithms is to enable computers to understand, interpret, and generate natural language like humans. It mainly includes the following types of algorithms:
[0234] 1. Lexical analysis algorithm: For example, word segmentation algorithm, which divides continuous text into meaningful units such as words and punctuation marks.
[0235] 2. Syntactic analysis algorithm: determine the structure and grammatical relationships of the sentence.
[0236] 3. Semantic understanding algorithm: tries to understand the meaning of the text.
[0237] 4. Text classification algorithm: classify text into different categories. For example, news articles can be classified into sports, entertainment, finance, etc.
[0238] 5. Sentiment analysis algorithm: Determine whether the sentiment expressed in the text is positive, negative or neutral. For example, analyze whether a comment is a compliment or a complaint about a product.
[0239] 6. Machine translation algorithm: realize conversion between different languages.
[0240] Through NLP technology, it is possible to understand and process user language input, including text preprocessing, lexical analysis, syntactic analysis, etc., to accurately grasp user intentions. Through advanced processing such as sentiment analysis and semantic understanding, it is possible to better interact with users and provide personalized service responses.
[0241] Finally, the input text and communication features are obtained, and the reply content is matched and recorded as the output text;
[0242] Perform language conversion operations on the output text, including:
[0243] Standardize text, remove punctuation and format numbers from the output text, and record the result as standard text;
[0244] Prosody modeling: obtain standard text, analyze the intonation, rhythm and emphasis of the standard text, and record the results as text features;
[0245] Apply the acoustic model to convert the standard text into phonemes, and then convert the phonemes into sound wave signals, and the results are recorded as output sound waves;
[0246] Synthesize language, obtain output sound waves and text features, apply sound synthesis technology, and obtain speech output.
[0247] Existing technology TTS enables intelligent outbound calls to provide feedback to users in a natural and fluent voice form, greatly improving the naturalness of interaction and user satisfaction. It can synthesize voice feedback in different styles and languages according to different users and scenarios, achieving a more personalized service experience.
[0248] The steps of the TTS algorithm include:
[0249] Text normalization: Convert the input text into a format suitable for processing, such as removing punctuation, formatting numbers, etc.
[0250] Prosody modeling: Analyzes the intonation, rhythm, and emphasis in text, which are natural features of speech.
[0251] Acoustic model: Convert text into phonemes and then further into sound wave signals.
[0252] Speech synthesis: Use sound synthesis techniques (such as concatenative synthesis or waveform editing) to generate the final speech output.
[0253] Types of TTS algorithms:
[0254] Concatenative synthesis: simulates human voice changes by changing the pitch, suitable for most text-to-speech applications.
[0255] Waveform Editing: Directly manipulate raw audio waveforms to generate speech, often used in professional sound design.
[0256] Parametric speech synthesis: uses mathematical models to simulate speech characteristics and is suitable for situations where a high degree of customization is required.
[0257] The benefits of intelligent outbound calling include:
[0258] Optimize user experience: Users do not need to face complex interfaces or perform tedious operations, and can complete all their needs through simple voice interaction. Provide barrier-free customer service channels for users with visual impairments or limited mobility, so that they can more conveniently obtain the services they need.
[0259] Enhanced security: Advanced voice recognition technology can provide secure voice services while ensuring user privacy. Integrated security measures can prevent security risks such as voice imitation and ensure stable operation of the system.
[0260] Optimization data drive: Voice data collected through ASR technology can be used to further train and optimize models to make the system more intelligent. Based on big data analysis, the system can automatically adjust and optimize service strategies to continuously improve service quality and efficiency.
[0261] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.
[0262] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.
Claims
1. An outbound calling method based on artificial intelligence, comprising: Obtain historical call records of overdue customers; According to the historical call records, the best time period strategy is adopted to obtain the calling time period set of overdue customers; Get all the opening format words of outbound calls to form an opening format set; According to the opening format set and historical call records, an opening format language optimization strategy is adopted to update the opening format set; Obtain historical reception data of all debt collection specialists; The ranking of debt collectors who record business is the professional ranking; Based on historical reception data, a professional scoring strategy is adopted to obtain the professional ranking of each business; Obtain the importance levels of all businesses and form a business ranking; Record the text of overdue customers and outbound calls as conversation records; When an overdue customer needs to be transferred to a collection specialist: Form a transfer decision strategy based on the conversation records; Execute the transfer judgment strategy and obtain a set of candidate collection specialists; Obtain the business of overdue customer consultation when transferring to the collection specialist, which is recorded as the transferred business; Obtain the business that has been assigned to each element in the candidate collection specialist set, and obtain the pending business set of each element in the candidate collection specialist set; Based on the set of candidate collection specialists and the set of pending businesses, the optimal allocation strategy is adopted to assign the best collection specialist to overdue customers.
2. The outbound calling method based on artificial intelligence according to claim 1, characterized in that: The optimal time period strategy includes: Set the first time, which refers to Monday to Friday; Set the second time, which is from Saturday to Sunday; The total number of first-time answering times in the historical call records of overdue customers is counted and recorded as the number of first-time answering times; The total number of second-time answering calls in the historical call records of overdue customers is counted and recorded as the number of second-time answering calls; When the number of first-time answers is greater than the number of second-time answers, the overdue customers are classified as the first group; When the number of calls answered at the first time is less than or equal to the number of calls answered at the second time, the overdue customers are classified as the second group of people; Set the number of filters for selecting the calling time period; Divide 24 hours a day into e time periods, count the number of times overdue customers answer calls in each time period in the historical call records, obtain the time periods with a number greater than the screening number, and form a call time period set; When the overdue customer belongs to the first group of people, the overdue customer is called at the first time in the time period included in the calling time period set every day; When the overdue customer belongs to the second group of people, a call is made to the overdue customer during a time period included in the dialing time period set every day at the second time.
3. The outbound calling method based on artificial intelligence according to claim 1, characterized in that: The opening format optimization strategy includes: Set the threshold time for judging whether the opening format words are retained; Get any element in the opening format set and record it as the marked opening format; The length of the call when the opening format is marked is recorded as the marked time; Obtain all marked times in the historical call records of overdue customers to form a marked time set; Traverse all elements in the marked time set, obtain the number of marked times greater than the threshold time, and record it as the effective number; Calculate the number of elements in the marked time set and record it as the number of dialing times; Calculate the effective number of times divided by the number of dialing times, and record the result as the effective rate; Set the judgment ratio for whether to keep the opening format; When the effective rate is less than or equal to the judgment ratio, the marked opening format words are removed from the opening format set; Traverse all elements in the opening format set to execute the opening format language optimization strategy and update the opening format set.
4. The outbound calling method based on artificial intelligence according to claim 1, characterized in that: The professional scoring strategy adopted includes: Get all the businesses and form a business set; Get any element in the service set and record it as the positioning service; Obtain the scoring data of each collection specialist in the positioning business in the historical reception data, calculate the average of multiple scoring data of each collection specialist, and obtain the average score of each collection specialist in the positioning business; The average scores of all collection specialists in the location service are sorted from high to low to obtain the professional ranking of the location service; Traverse all elements in the business set, adopt professional scoring strategy, and obtain the professional ranking of each element in the business set.
5. The outbound calling method based on artificial intelligence according to claim 1 is characterized in that: The optimal allocation strategy is adopted, including: Get the location of overdue customer phone numbers; Obtain the collection specialists whose hometown and place of origin are the same from the candidate collection specialist set, and form a collection specialist set with the same place of origin; Acquire all services that are higher in importance than the transfer service in the service level ranking to form a priority service set; Acquire all services whose importance level is lower than that of the transfer service in the service level ranking to form a lagging service set; The collection specialist whose elements in the pending business set are all in the priority business set is recorded as the discarded collection specialist; The collection specialist whose elements in the pending business set are not in the priority business set is the first collection specialist; The collection specialist whose elements in the pending business set are in both the priority business set and the delayed business set is recorded as the second collection specialist; Traverse the pending business set of each element in the collection specialist set of the same place, obtain all the first collection specialists and the second collection specialists, and form the first collection set and the second collection set respectively; Set a limit value to limit the number of businesses that collection specialists can receive; Count the number of elements in the pending business set of each collection specialist in the first set to obtain the first set of processed business; Compare each element in the first handling number set with the limit value, obtain the collection specialists corresponding to all elements whose value is less than the limit value, and record them as the first preferred specialist set; Count the number of elements in the pending business set of each collection specialist in the second set of actions to be taken, and obtain a second set of processing numbers; Compare each element in the second handling number set with the limit value, obtain the collection specialists corresponding to all elements whose value is less than the limit value, and record them as the second preferred specialist set.
6. The outbound calling method based on artificial intelligence according to claim 1, characterized in that: The adopting of the optimal allocation strategy also includes: Set the initial order, which is the order of all pending transactions when the transaction is first assigned to a collection specialist; Set the execution order, which is the order in which the collection specialists place the business in the pending business when they are working; Set the number of judgments to determine whether the collection specialist is qualified to receive overdue customers; Obtain historical data of the execution order of the determination quantity, recorded as a historical execution order set; Compare all elements in the historical execution order set with the initial order: When the initial order is less than or equal to all elements in the historical execution order set, the collection specialist to which the business belongs is recorded as the marked specialist.
7. The outbound calling method based on artificial intelligence according to claim 1, characterized in that: The adopting of the optimal allocation strategy also includes: Traverse the pending business set of each element of the first preferred specialist set and the second preferred specialist set, obtain all marked specialists, and form a discarded specialist set; Remove elements in the discarded specialist set from the first preferred specialist set and the second preferred specialist set to obtain a first target specialist set and a second target specialist set; The collection specialist who responds to overdue customers is recorded as the responding collection specialist; Obtain the highest professional ranking of the collection specialist who transfers the business among all the collection specialists who respond, and record him as the responding collection specialist; When the responding collection specialist is an element in the first target specialist set, the overdue customer transfer business is assigned to the responding collection specialist, and the initial order of the transfer business is first; When the responding debt collection specialist is an element in the second target specialist set; Obtain the elements in the pending business set of the answering collection specialist that have a higher importance level than the transferred business, and record them as the response priority business set; Get the largest execution order of the response priority service set, recorded as the connection order x; The transfer business of overdue customers is assigned to the responding collection specialist, and the initial order of the transfer business is x+1.
8. The outbound calling method based on artificial intelligence according to claim 1, characterized in that: The execution of the transfer judgment strategy includes: Extract keywords from the conversation records to form a keyword set; Get the number of elements in the keyword set, and record the result as the number of keywords a; Distribute the conversation records to all collection agents; Obtain the characteristic words extracted from the conversation records of each debt collection specialist; A feature matching strategy is adopted for each collection specialist’s feature words, including: Obtain the characteristic words of any collection specialist and record them as the positioning characteristic word set; Compare the elements in the positioning feature word set and the keyword set; Obtaining the same elements in the positioning feature word set and the keyword set to form a matching feature set; Calculate the number of elements in the matching feature set, and record the result as the matching number b; Calculate the number of matches ÷ the number of keywords, and record the result as the match rate c, c = b ÷ a; When the matching rate is equal to 1, the collection specialist corresponding to the positioning feature word set is recorded as the matching collection specialist; Get all matching debt collectors to form a candidate debt collector set.
9. The outbound calling method based on artificial intelligence according to claim 1, characterized in that: The conversation record includes: First, convert the voice input of the debt collection customer into input text, including: Audio collection: obtaining the voice input by the debt collection customer, converting the voice into simulated continuous sound waves and recording them as voice signals; Preprocessing, obtaining a speech signal, performing pre-emphasis, framing and windowing operations, and recording the result after preprocessing as a speech frame set; Extract features, obtain a speech frame set, extract feature parameters of each element in the speech frame set, and form a feature parameter set; Apply the acoustic model, obtain a feature parameter set, traverse each element in the feature parameter set, apply the acoustic model to each element, obtain the phoneme corresponding to each speech frame, and form a sound result set; Applying the language model to obtain a speech frame set, applying the language model to obtain multiple text sequences corresponding to the speech frame set to form a text result set; Search and decode to obtain the sound result set and the text result set, apply the decoding search algorithm, and record the obtained text sequence as the input text; Secondly, obtain the tone, expression, and speech speed information of the debt collection customer and record it as implicit information; Obtain input text, apply natural language processing algorithms to the input text and implicit information, obtain the emotions and repayment willingness of the collection customers, and record them as communication features; Finally, the input text and communication features are obtained, and the reply content is matched and recorded as the output text; Perform language conversion operations on the output text, including: Standardize text, remove punctuation and format numbers from the output text, and record the result as standard text; Prosody modeling: obtain standard text, analyze the intonation, rhythm and emphasis of the standard text, and record the results as text features; Apply the acoustic model to convert the standard text into a phoneme set, and then convert the phoneme set into a sound wave signal, and the result is recorded as the output sound wave; Synthesize language, obtain output sound waves and text features, apply sound synthesis technology, and obtain speech output.
Citation Information
Patent Citations
Method and apparatus for processing call center business
CN106375614A
User request allocation method and device for online customer service system and computing device
CN107332902A
Call center seat distribution method, system and device and storage medium
CN110062120A
Call-out strategy determination method and device based on artificial intelligence, equipment and medium
CN115065756A
Debt collection processing method and device, electronic equipment and storage medium
CN115953237A
