Inbound call volume forecasting method, apparatus, device and medium for customer service

Abstract

The disclosure provides an incoming call volume prediction method for customer service, belonging to the field of artificial intelligence. The method comprises the following steps: determining a prediction day and a prediction period; extracting a first incoming call volume feature of the prediction period in the prediction day; selecting a historical reference day of the prediction day based on an expert rule to obtain a first reference day; extracting a second incoming call volume feature of the prediction period in the first reference day; constructing a first feature vector based on the first incoming call volume feature and the second incoming call volume feature; and taking the first feature vector as an input of a trained incoming call volume prediction model, obtaining an output of the incoming call volume prediction model, and obtaining an incoming call volume of the prediction period in the prediction day. The disclosure also provides an incoming call volume prediction device for customer service, a computer device, a storage medium and a program product.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence, and in particular to a method and apparatus for predicting inbound call volume for customer service, computer equipment, storage media, and program products. Background Technology

[0002] Currently, most customer service (e.g., remote banking, logistics consulting, e-commerce after-sales service) relies primarily on manual methods for predicting inbound call volume. These manual methods mainly depend on rules and experience, selecting actual inbound call volume from a historical period comparable to the current prediction timeframe, and scaling it according to certain principles to forecast the current inbound call volume.

[0003] In the process of realizing the present invention, the inventors found that the prior art has at least the following problems: (1) In the process of manually predicting the volume of incoming calls, the selection of historical time periods is somewhat random, which brings unavoidable errors to the actual prediction and the prediction accuracy needs to be improved; (2) Moreover, manual operation is extremely dependent on subjective experience, time-consuming and labor-intensive, with limited factors to consider and little room for optimization, and the level of intelligence is low; (3) The prediction results of manual prediction are greatly affected by sudden hot events, which affects the subsequent work arrangements of customer service personnel, which can easily lead to resource waste or shortage, and there is a bottleneck in improving customer service satisfaction. Summary of the Invention

[0004] In order to solve at least one of the above-mentioned problems in the prior art, this disclosure provides a method and apparatus, computer equipment, storage medium and program product for predicting inbound call volume for customer service, so as to realize intelligent prediction of customer service inbound call volume, improve the prediction accuracy to at least a certain extent, and assist in the rational arrangement of agent work.

[0005] A first aspect of this disclosure provides a method for predicting inbound call volume for customer service, comprising: determining a prediction day and a prediction period; extracting a first inbound call volume feature from the prediction period within the prediction day; selecting a historical reference day for the prediction day based on expert rules to obtain a first reference day; extracting a second inbound call volume feature from the prediction period within the first reference day; constructing a first feature vector based on the first and second inbound call volume features; and using the first feature vector as input to a trained inbound call volume prediction model to obtain the output of the inbound call volume prediction model, thereby obtaining the inbound call volume for the prediction period within the prediction day.

[0006] According to embodiments of this disclosure, the first inbound call volume characteristic includes at least one of the following: a business time period characteristic determined based on the business content of the predicted date; a daily granular time period characteristic determined based on the time interval of the predicted time period within a day; a historical time series characteristic obtained based on the time series of inbound calls during the predicted time period within a time period that meets preset conditions before the predicted date; or a time window characteristic obtained based on the statistical value of inbound calls within historical dates that have a second periodic relationship with the predicted date.

[0007] According to embodiments of this disclosure, the aforementioned historical time series characteristics include at least one of the following: the statistical value of the inbound call volume during the forecast period within the first period prior to the forecast date; or the inbound call volume during the forecast period on the most recent R dates prior to the forecast date that have the same attributes as the forecast date; wherein R is a positive integer.

[0008] According to embodiments of this disclosure, the aforementioned time window features include at least one of the following: statistical values ​​of incoming calls within historical dates that have a second periodic relationship with the aforementioned prediction date; statistical values ​​of incoming calls within the aforementioned prediction period within historical dates that have a second periodic relationship with the aforementioned prediction date; and statistical values ​​of incoming calls within historical dates that have a second periodic relationship with the aforementioned prediction date and are all either working days or non-working days; wherein the aforementioned second periodic relationship includes at least one of a weekly periodic relationship, a monthly periodic relationship, a quarterly periodic relationship, or an annual periodic relationship.

[0009] According to an embodiment of this disclosure, the extraction of the second inbound call volume feature in the prediction period of the first reference day includes: obtaining the historical inbound call volume of the first reference day in the prediction period; scaling the historical inbound call volume according to a predetermined scaling rule to obtain the second inbound call volume feature.

[0010] According to an embodiment of this disclosure, the above-mentioned selection of a historical reference day based on expert rules to obtain a first reference day includes: obtaining attribute information of the prediction day, the attribute information including M attributes, where M is a positive integer; when the first attribute of the prediction day has a first attribute value, searching for a date that matches the first attribute value within a first time period before the prediction day, wherein the first attribute is one of the M attributes; when the first attribute of the prediction day is a second attribute value, searching for a date that matches the second attribute value within a second time period before the prediction day.

[0011] According to an embodiment of this disclosure, when M is greater than or equal to 2, the above-mentioned selection of historical reference days for the prediction date based on expert rules to obtain the first reference day includes: preferentially selecting a date that matches all of the above-mentioned M attributes of the prediction date as the first reference day; and when there is no date that matches all of the above-mentioned M attributes of the prediction date, selecting a date that partially matches the above-mentioned M attributes of the prediction date as the first reference day.

[0012] According to embodiments of this disclosure, the aforementioned inbound traffic prediction model is trained as follows: Inbound traffic data for N time periods within at least one historical date is obtained, where the N time periods include the prediction period, and N is a positive integer; a third inbound traffic feature is extracted from the prediction period within the historical date, in the same manner as the extraction of the first inbound traffic feature; a second reference date is obtained by selecting a historical reference date based on the expert rules; a fourth inbound traffic feature is extracted from the prediction period within the second reference date, in the same manner as the extraction of the second inbound traffic feature; a second feature vector is constructed based on the third and fourth inbound traffic features; and training sample data is obtained by using the second feature vector as the input sample of the inbound traffic prediction model and the inbound traffic data for the prediction period within the historical date as the output sample of the inbound traffic prediction model; the inbound traffic prediction model is trained using the training sample data.

[0013] According to embodiments of this disclosure, the training process of the aforementioned inbound traffic prediction model further includes: constructing multiple pending inbound traffic prediction models using different algorithms; training the aforementioned inbound traffic prediction model using the aforementioned training sample data includes: training the multiple pending inbound traffic prediction models respectively using the aforementioned training sample data; comparing the errors of the multiple pending inbound traffic prediction models in a test set after training the multiple pending inbound traffic prediction models; and selecting the model with the smaller error from the multiple pending inbound traffic prediction models as the aforementioned inbound traffic prediction model.

[0014] A second aspect of this disclosure provides an inbound call volume prediction device for customer service, comprising: a determination module, a first inbound call volume feature extraction module, a reference day selection module, a second inbound call volume feature extraction module, a feature vector construction module, and a prediction module. The determination module is used to determine a prediction day and a test period; the first inbound call volume feature extraction module is used to extract a first inbound call volume feature from the prediction period within the prediction day; the reference day selection module is used to select a historical reference day for the prediction day based on expert rules to obtain a first reference day; the second inbound call volume feature extraction module is used to extract a second inbound call volume feature from the prediction period within the first reference day; the feature vector construction module is used to construct a first feature vector based on the first and second inbound call volume features; and the prediction module is used to use the first feature vector as input to a trained inbound call volume prediction model, obtain the output of the inbound call volume prediction model, and thus obtain the inbound call volume for the prediction period within the prediction day.

[0015] According to embodiments of this disclosure, the inbound traffic prediction device further includes a training module, wherein the training module specifically includes: an acquisition submodule, a third inbound traffic feature extraction submodule, a reference day selection submodule, a fourth inbound traffic feature extraction submodule, a feature vector construction submodule, a first training submodule, and a second training submodule. The acquisition submodule is used to acquire the inbound call volume of N time periods in at least one historical date, wherein the N time periods include the prediction time period, and N is a positive integer; the third inbound call volume feature extraction submodule is used to extract the third inbound call volume feature of the prediction time period in the historical date in the same way as the extraction of the first inbound call volume feature; the reference day selection submodule is used to select a historical reference day for the historical date based on the expert rules to obtain a second reference day; the fourth inbound call volume feature extraction submodule is used to extract the fourth inbound call volume feature of the prediction time period in the second reference day in the same way as the extraction of the second inbound call volume feature; the feature vector construction submodule is used to construct a second feature vector based on the third inbound call volume feature and the fourth inbound call volume feature; the first training submodule is used to use the second feature vector as the input sample of the inbound call volume prediction model and the inbound call volume of the prediction time period in the historical date as the output sample of the inbound call volume prediction model to obtain training sample data; the second training submodule is used to train the inbound call volume prediction model using the training sample data.

[0016] According to embodiments of this disclosure, the second training submodule further includes: an algorithm construction unit, a prediction model training unit, a comparison unit, and a model selection unit. The algorithm construction unit is used to construct multiple undetermined inbound call volume prediction models using different algorithms; the prediction model training unit is used to train the multiple undetermined inbound call volume prediction models using the training sample data; the comparison unit is used to compare the errors of the multiple undetermined inbound call volume prediction models on a test set after training; and the model selection unit is used to select the model with the smaller error from the multiple undetermined inbound call volume prediction models as the inbound call volume prediction model.

[0017] A third aspect of this disclosure provides a computer device including: one or more processors, and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the call volume prediction method for customer service as described above.

[0018] A fourth aspect of this disclosure provides a computer-readable storage medium storing computer program instructions that, when executed by a processor, are used to implement the inbound call volume prediction method for customer service as described above.

[0019] A fifth aspect of this disclosure provides a computer program product including computer program instructions that, when executed by a processor, implement the inbound call volume prediction method for customer service as described above.

[0020] The above one or more embodiments have the following advantages or beneficial effects: using machine learning algorithms to predict inbound call volume improves the intelligence of inbound call volume prediction. The input features used for prediction include not only inbound call volume features from historical reference days with the same attributes as the prediction day, but also inbound call volume features from the prediction period within the prediction day. This allows for intelligent prediction of inbound call volume to take into account both historically similar situations and the specific circumstances of the current prediction period, at least partially avoiding the subjectivity, randomness, and limitations of manually selected features, thus improving the accuracy of inbound call volume prediction. Attached Figure Description

[0021] The foregoing contents, other objects, features, and advantages of this disclosure will become clearer from the following description of embodiments with reference to the accompanying drawings, in which several embodiments of this disclosure are illustrated by way of example and not limitation, wherein:

[0022] Figure 1This illustration schematically depicts an application scenario of an inbound call volume prediction method and apparatus for customer service, computer equipment, media, and program products according to embodiments of the present disclosure.

[0023] Figure 2 A flowchart illustrating an inbound call volume prediction method for customer service according to an embodiment of the present disclosure is shown schematically.

[0024] Figure 3 A flowchart illustrating a training method for an inbound traffic prediction model according to an embodiment of the present disclosure is shown.

[0025] Figure 4 The diagram illustrates an inbound traffic prediction model constructed using the random forest algorithm according to an embodiment of the present disclosure.

[0026] Figure 5 The diagram illustrates an inbound traffic prediction model constructed using the Gradient Boosting Tree (GBRT) algorithm according to an embodiment of the present disclosure.

[0027] Figure 6 A block diagram of an inbound call volume prediction device for customer service according to an embodiment of the present disclosure is shown schematically; and

[0028] Figure 7 A block diagram of an electronic device according to an embodiment of the present disclosure is shown schematically. Detailed Implementation

[0029] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0030] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0031] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0032] When using expressions such as "at least one of A, B, and C", they should generally be interpreted in accordance with the meaning that is commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B, and C, etc.).

[0033] It should be understood in this document that any number of elements in the specification and figures are for illustrative purposes and not for limitation, and that any naming (e.g., first, second) is for distinction only and has no limiting meaning.

[0034] The inventors studied and summarized manual prediction methods, coded them, and combined them with machine learning algorithms to propose an intelligent solution for predicting customer service call volume, thereby improving prediction accuracy and better assisting in the rational allocation of agent work.

[0035] Specifically, embodiments of this disclosure provide a method, apparatus, computer device, medium, and program product for predicting inbound call volume for customer service. The method includes: first, determining the prediction date and prediction period; then, extracting a first inbound call volume feature for the corresponding prediction period within the prediction date. Next, selecting a historical reference date for the prediction date according to expert rules to obtain a first reference date; simultaneously, extracting a second inbound call volume feature for the corresponding prediction period within the first reference date. Then, constructing a first feature vector based on the extracted first and second inbound call volume features. Finally, using the constructed first feature vector as input to a trained inbound call volume prediction model, obtaining the output of the inbound call volume prediction model to obtain the inbound call volume for the corresponding prediction period within the prediction date.

[0036] According to embodiments of this disclosure, based on preset expert rules, a historical reference day with the same attributes as the prediction day is selected. Features of the prediction period are extracted from both the prediction day and the historical reference day to construct a feature vector, which is then input into a machine learning algorithm to achieve intelligent prediction of inbound call volume. This intelligent prediction of inbound call volume takes into account both historically similar call volume patterns and the specific characteristics of the current prediction period. This avoids the subjectivity, randomness, and limitations of manual selection, thus improving the accuracy of inbound call volume prediction.

[0037] It should be noted that the methods, devices, computer equipment, media and program products for predicting inbound call volume for customer service as determined in the embodiments of this disclosure can be used in the financial field (e.g., remote banking), or in any field other than the financial field (e.g., express delivery tracking in logistics, pre-sales or after-sales service in the retail industry, remote business processing in telecommunications outlets, etc.). This disclosure does not limit the application field.

[0038] Figure 1 This illustration schematically depicts an application scenario of the inbound call volume prediction method and apparatus for customer service, computer equipment, media, and program products according to embodiments of this disclosure; it should be noted that... Figure 1 The illustrations shown are merely examples of system architectures that can be applied to the embodiments of this disclosure, in order to help those skilled in the art understand the technical content of this disclosure, but do not mean that the embodiments of this disclosure cannot be used in other devices, systems, environments or scenarios.

[0039] like Figure 1 As shown, the application scenario 100 may include a client 101, a network 102, and a customer service center 103. The network 102 is used to transmit signals between the client 101 and the customer service center 103. For example, when a customer contacts customer service, the network 102 transmits the customer's incoming call signal to the customer service center 103.

[0040] Client 101 may include landlines, mobile phones, and / or IP phones, video phones, etc., accessed via the public network.

[0041] Network 102 may include telephone network, Internet or mobile network, and can transmit signals through different types of networks depending on the different channels in the client 101 used by the customer.

[0042] Customer service center 103 may include a customer service center server (cluster) 31 and multiple terminals 32. The server (cluster) 31 can manage the multiple terminals 32 and assign access requests received from clients 101 to one of the multiple terminals 32. It is understood that... Figure 1 The example of a telephone as a terminal 32 is merely exemplary; in remote banking, terminal 32 can be any type of terminal device with a display screen.

[0043] According to embodiments of this disclosure, server (cluster) 31 can execute the call volume prediction method for customer service provided in the disclosed embodiments, predict the call volume for subsequent time periods, and manage multiple terminals 32 based on the prediction results. Accordingly, the call volume prediction device, computer equipment, media, and program products for customer service provided in the embodiments of this disclosure can be configured in server (cluster) 31.

[0044] It is understood that the inbound call volume prediction method for customer service provided in this disclosure embodiment can also be executed by other servers communicating with server (cluster) 31. Accordingly, the inbound call volume prediction device, computer equipment, media, and program products for customer service provided in this disclosure embodiment can also be located in other servers communicating with server (cluster) 31.

[0045] It should be understood that Figure 1 The types and numbers of terminal devices, networks, and servers shown are merely illustrative. Depending on implementation needs, any type and number of terminal devices, networks, and servers can be included.

[0046] The following will be based on Figure 1 The described application scenario, taking remote banking as an example, involves... Figures 2-5 The present disclosure provides a detailed description of the inbound call volume prediction method for customer service, according to embodiments of the present disclosure.

[0047] Figure 2 A flowchart illustrating an inbound call volume prediction method for customer service according to an embodiment of the present disclosure is shown.

[0048] like Figure 2 As shown, the inbound call volume prediction method may include operations S210 to S260.

[0049] In operation S210, determine the forecast date and forecast period.

[0050] Inbound call volume for remote banking centers refers to the volume of calls made by customers after they have completed relevant banking transactions. On the one hand, the bank will send a text message with the details of the transaction on the relevant business day, and the customer will call back to the bank to inquire after receiving the message. On the other hand, customers will proactively call the bank when they need to conduct certain related business.

[0051] A forecast period is a specific time interval within a forecast day. For example, if you want to forecast the inbound call volume every 15 minutes from morning to night for the next week, then each day of the next week is a forecast day. Each 15 minutes within each day is a forecast period, such as the 15 minutes from 8:00 AM to 8:15 AM. Of course, the starting time of the forecast period can be determined according to actual needs, and there is no limitation here.

[0052] The prediction date and prediction period can be determined based on user operations, such as the administrator configuring the prediction date and prediction period through the configuration terminal and uploading it to the server (cluster) 31; or it can be automatically run in the server (cluster) 31 through a preset script program, such as setting the server (cluster) 31 to predict the business call volume of each prediction period in the next day or several days during the business off-peak period in the evening, so as to help the server (cluster) 31 manage multiple customer service terminals 32.

[0053] In operation S220, extract the first inbound call feature for the predicted period of the predicted day.

[0054] The first characteristic of inbound call volume during the forecast period within the forecast day can be determined based on factors that influence inbound call volume forecasting. These factors may include, for example, whether the forecast day is a weekday or a non-weekday, the end of the month or the middle of the month, whether it is a special holiday or business day, and / or the time interval of the day in which the forecast period falls, such as noon, evening, or morning, as inbound call volume typically varies significantly across different time intervals of the day.

[0055] According to embodiments of this disclosure, the extracted first inbound call feature may include, but is not limited to, at least one of the following: business period feature, daily granularity period feature, historical time series feature, and time window feature. The business period feature is a feature determined based on the business content of the prediction day. The daily granularity period feature is a feature determined based on the time interval within a day in which the prediction period falls. The historical time series feature is a feature obtained based on the time sequence of inbound calls for the prediction period within a time period that meets preset conditions prior to the prediction day. The time window feature is a feature obtained based on the statistical value of inbound calls within historical dates that have a second periodic relationship with the prediction day.

[0056] According to embodiments of this disclosure, the value of each feature among business time period features, daily granular time period features, historical time series features and / or time window features can be obtained, and then combined into a vector to obtain the first inbound call volume feature.

[0057] The following will use the credit card business of a remote bank as an example to further explain each of the features listed above as the first incoming call volume features, but this does not limit the extractable first incoming call volume features to the scope of the embodiments.

[0058] Regarding the determination of business time period characteristics based on the business content of the predicted date. In the actual operation of credit card business inbound calls at remote banking centers, the daily inbound call volume is greatly affected by the business content of that day. In some embodiments, for the sake of unified management, banks may designate certain fixed dates to specifically handle specific business content (e.g., repayment, bill SMS sending, repayment reminder SMS sending, or interest SMS sending, etc.). Therefore, business time period characteristics include, but are not limited to, repayment date, bill SMS sending date, repayment reminder SMS sending date, or interest SMS sending date. The business time period characteristics of the predicted date can be determined through the calendar data of the predicted date (i.e., month or the date of each month, etc.). Thus, in some embodiments, the value of the business time period characteristics of the predicted date can be the date of the predicted date. Or in other embodiments, for example, corresponding values ​​(e.g., 1 to 4) can be assigned to the repayment date, bill SMS sending date, repayment reminder SMS sending date, or interest SMS sending date, and other days can be assigned other values ​​(e.g., 0), so that the business time period to which the predicted date belongs can be determined based on the date of the predicted date, and mapped to a specific value, thereby obtaining the value of the business time period characteristics of the predicted date.

[0059] Regarding the daily-level time period characteristics determined based on the time interval of the predicted period within a day: In the inbound call volume of credit card services at remote banking centers, the fluctuation trend of inbound calls in different time periods each day is basically the same, exhibiting a certain regularity. Therefore, daily-level time period characteristics can be extracted. In one embodiment, the time intervals of a day can be divided as follows: 00:00:00-06:00:00, 06:00:00-11:00:00, 11:00:00-14:00:00, 14:00:00-16:00:00, and 16:00:00-23:59:59. Similarly, corresponding values ​​can be assigned to each time interval. When the predicted period falls within a certain time interval, the value corresponding to that time interval is used as the value of the daily-level time period characteristic.

[0060] Regarding historical time-series characteristics: Historical time-series characteristics are derived from the time series of inbound calls during the forecast period, based on preset conditions within a timeframe prior to the forecast date. These characteristics reflect the fluctuations and trends in inbound call volume over a period preceding the forecast date, thus providing trend information for the inbound call volume during the forecast period on the forecast date.

[0061] According to some embodiments of this disclosure, historical time-series characteristics may include: statistical values ​​of inbound call volume during the prediction period within the first cycle prior to the prediction date.

[0062] The length of the first period can be set based on practical experience. For example, the first period can be one week, one month, one quarter, and / or one year. Considering the time-series correlation characteristics of inbound calls, in some embodiments, statistical values ​​can be calculated for different lengths of the first period, and these statistical values ​​can all be used as the values ​​of the historical time-series characteristics. For example, the inbound call statistics for 8:00 to 8:15 in the month before the forecast date can be calculated at a monthly frequency, the inbound call statistics for 8:00 to 8:15 in the quarter before the forecast date can be calculated at a quarterly frequency (where a quarter is 3 months), and the inbound call statistics for 8:00 to 8:15 in the year before the forecast date can be calculated at an annual frequency, and then all three statistical values ​​can be used as the values ​​of the historical time-series characteristics.

[0063] In this embodiment of the disclosure, the statistical values ​​of the incoming volume may include, for example, the mean, standard deviation, maximum or minimum value, etc.

[0064] According to embodiments of this disclosure, historical time-series features may include: predicting inbound call volume for a time period based on the most recent R (R is a positive integer) dates prior to the prediction date that have the same attributes as the prediction date.

[0065] For example, in some embodiments, the discrete values ​​of inbound call volume for the prediction period within the 5 most recent statistical dates (where R = 5, just an example) closest to the prediction date can be listed in the historical time series characteristics. In some embodiments, the 5 statistical dates can be dates with the same calendar information as the prediction date before the prediction date, such as the same day of each month, so the 5 most recent statistical dates are the 5 dates within the 5 most recent months. Accordingly, the discrete values ​​of the 5 statistical dates are the inbound call volume for each time period (e.g., 8:00 to 8:15) of the aforementioned 5 dates. In other embodiments, the 5 statistical dates can be 5 statistical days with the same business time period characteristics or the same business content as the prediction date before the prediction date, for example, when the prediction date is a repayment date, the 5 statistical dates are the 5 repayment dates before the prediction date.

[0066] Regarding time window features. According to embodiments of this disclosure, time window features are obtained based on statistical values ​​of inbound call volume within historical dates that have a second periodic relationship with the prediction date. The second periodic relationship includes at least one of weekly, monthly, quarterly, or annual periodic relationships. The term "second period" is used herein to distinguish it from the aforementioned "first period," implying that the period lengths used in determining the historical time series features and time window features are determined independently.

[0067] A time window feature reflects the inbound call volume on a specific day prior to the prediction date. In real life, some situations tend to occur periodically, exhibiting a certain cyclical recurrence characteristic. For example, e-commerce companies hold promotional events on fixed dates each year. Time window features can be used to introduce cyclical factors that influence inbound call volume.

[0068] According to one embodiment of this disclosure, the time window feature may include: statistical values ​​of incoming calls within historical dates that have a second periodic relationship with the prediction date. For example, it may be the statistical value of incoming calls on a day that is exactly 3 months, 6 months, and / or 12 months away from the prediction date; or it may be the statistical value of incoming calls on a day or several days that are exactly 3 months, 6 months, and / or 12 months away from the prediction date and belong to the same ten-day period as the prediction date, wherein each month is divided into six half-ten-day periods, with the 31st day included in the sixth half-ten-day period.

[0069] According to another embodiment of this disclosure, the time window feature may include: statistical values ​​of inbound calls during the prediction period within historical dates that have a second cycle relationship with the prediction date. Taking a prediction period of 8:00 to 8:15 as an example, the time window feature may be: statistical values ​​of inbound calls during the 8:00 to 8:15 period on a day exactly 3 months, 6 months, and / or 12 months from the prediction date; or, the time window feature may be: statistical values ​​of inbound calls during the 8:00 to 8:15 period on a day or several days that are exactly 3 months, 6 months, and / or 12 months from the prediction date and belong to the same date cycle as the prediction date; or, the time window feature may also be: statistical values ​​of inbound calls during the 8:00 to 8:15 period on a day or several days that are exactly 3 months, 6 months, and / or 12 months from the prediction date and belong to the same week as the prediction date.

[0070] According to another embodiment of this disclosure, the time window feature may include: statistical values ​​of inbound calls within historical dates that have a second periodic relationship with the prediction date and are all either weekdays or non-weekdays. When the prediction date is a weekday, the time window feature may, for example, be statistical values ​​of inbound calls on the first weekday after a full 3, 6, and / or 12 months have passed since the prediction date. When the prediction date is a non-weekday, the time window feature may, for example, be statistical values ​​of inbound calls on the first non-weekday after a full 3, 6, and / or 12 months have passed since the prediction date.

[0071] The above lists various scenarios for time window features. It is understood that the time window features according to embodiments of this disclosure include, but are not limited to, the scenarios listed above, and will not be elaborated upon further here.

[0072] In operation S230, a historical reference day is selected based on expert rules to obtain the first reference day. The first reference day is a historical reference day with the same attributes as the prediction day. The attributes of the prediction day can be the business content of the prediction day (e.g., SMS day, billing day), the calendar information of the prediction day (beginning of the month, end of the month, middle of the month, or a certain day of the month, etc.), or the holiday information of the prediction day in daily production and life (e.g., working day, non-working day, long holiday, short holiday, adjusted workday, etc.).

[0073] According to one embodiment of this disclosure, in operation S230, attribute information of the prediction date can be obtained first. The attribute information includes M attributes, where M is a positive integer. Then, when the first attribute of the prediction date has a first attribute value, a date matching the first attribute value is searched within a first time period before the prediction date, where the first attribute is one of the M attributes. Alternatively, when the first attribute of the prediction date is a second attribute value, a date matching the second attribute value is searched within a second time period before the prediction date. In this way, the found first reference date can have the same first attribute as the prediction date.

[0074] According to some other embodiments of this disclosure, when M is greater than or equal to 2, the process of selecting a historical reference day in operation S230 may be: preferentially selecting a date that matches all M attributes of the prediction day as the first reference day; and when there is no date that matches all M attributes of the prediction day, selecting a date that partially matches the M attributes of the prediction day as the first reference day.

[0075] Table 1 illustrates an example of an expert rule for selecting a historical reference day in operation S230. According to embodiments of this disclosure, a historical reference day can be selected based on the expert rules listed in Table 1 below. In Table 1, if the preferred reference day is empty, alternative 1, alternative 2, alternative 3, and alternative 4 can be considered sequentially as reference days. For example, a table for selecting reference days can be shown in Table 1.

[0076] Further explanation of Table 1 is as follows: Note 1) Broad weekdays refer to the use of Tuesday, Wednesday, Thursday, and Friday, dividing the week into four categories: Monday, Tuesday, Wednesday, Thursday, Friday, Saturday, and Sunday. Note 2) Each month consists of 5-day periods, totaling 6 periods. The 31st is included in the sixth period.

[0077] Table 1:

[0078]

[0079] Refer to Table 1 above. When the aforementioned first attribute is attribute 1 of the prediction date, the first attribute value can be, for example, an SMS date, and the second attribute value can be, for example, a non-SMS date.

[0080] When the forecast date has multiple attributes as shown in Table 1, the date that matches all of the forecast date's attributes is selected first as the first reference date, exemplified by the preferred reference date in Table 1. When no date matches all of the forecast date's attributes, the dates that partially match the forecast date's attributes are considered sequentially as the first reference date, exemplified by Alternate 1, Alternate 2, Alternate 3, and Alternate 4 in Table 1.

[0081] As can be seen, the historical reference date can be selected in operation S230 by setting one or more attributes in the expert rules for selecting the historical reference date, and by setting the rules and conditions for selecting the historical reference date based on these attributes.

[0082] In operation S240, the second inbound call feature for the forecast period in the first reference day is extracted. The second inbound call feature can be the inbound call data for the forecast period in the first reference day, or it can be the scaled data of the inbound call data for the forecast period in the first reference day.

[0083] According to one embodiment of this disclosure, in operation S240, the historical incoming call volume of the first reference day during the prediction period can be obtained first, and then the historical incoming call volume can be scaled according to a predetermined scaling rule to obtain the second incoming call volume feature.

[0084] For example, after determining the historical reference day for the prediction date based on the aforementioned preset expert rules, the first reference day is obtained. Then, the historical inbound call volume of the first reference day during the prediction period is acquired. Finally, the inbound call volume of the first reference day is scaled according to the ratio of the inbound call volume of the past dates in the month of the prediction date to the inbound call volume of the corresponding dates in the month of the first reference day.

[0085] For example, historical inbound call volume can be scaled in one of the following four different ways: scaled by "total inbound call volume of the month containing the reference date / total volume of the nearest neighboring month"; scaled by "total inbound call volume of the month containing the reference date / total volume of the nearest neighboring month with the same number of days"; scaled by "average daily inbound call volume of the month containing the reference date / average daily inbound call volume of the nearest neighboring month"; scaled by "average daily inbound call volume of the month containing the reference date / average daily inbound call volume of the nearest neighboring month with the same number of days".

[0086] In operation S250, a first feature vector is constructed based on the first and second incoming call volume features. The first and second incoming call volume features can be combined to obtain the first feature vector. In one embodiment, the content of the first feature vector can be, for example,: {business time period feature, daily granular time period feature, historical time series feature, time window feature, second incoming call volume feature}. The value of each feature in the first feature vector can be a single numerical value or multiple numerical values ​​listed in a predetermined order (for example, the historical time series feature described above can include multiple numerical values).

[0087] In operation S260, the first feature vector is used as the input of the trained inbound traffic volume prediction model, and the output of the inbound traffic volume prediction model is obtained to obtain the inbound traffic volume for the predicted period of the day.

[0088] Therefore, it can be seen that when using machine learning algorithms to intelligently predict inbound call volume, the embodiments of this disclosure can take into account the specific characteristics of the prediction period for inbound call volume under similar historical circumstances, avoiding the subjectivity, randomness and limitations of consideration factors in manual selection, and improving the accuracy of inbound call volume prediction.

[0089] Figure 3 A flowchart illustrating a training method for an inbound traffic prediction model according to an embodiment of the present disclosure is shown.

[0090] like Figure 3As shown, the training method for the inbound call volume prediction model may include operations S310 to S370.

[0091] In operation S310, the incoming call volume for N time periods in at least one historical date is obtained, where the N time periods include the predicted time periods, and N is a positive integer.

[0092] For example, relevant historical data within a historical period (one or two years) close to the prediction date can be obtained as the basis for training the inbound call volume prediction model.

[0093] In remote banking credit card services, historical inbound call volume data, calendar data, and the number of new credit card accounts opened daily over the past two years, divided by prediction period, can be used as the base data for inbound call volume prediction models. For example, when predicting inbound call volume every 15 minutes of each day, historical inbound call volume data at the 15-minute granularity of each day can be selected from the base data.

[0094] In operation S320, the third inbound call feature for the predicted time period from the historical dates is extracted in the same way as the first inbound call feature extraction. The process of extracting the third inbound call feature can be referred to the previous description of extracting the first feature vector in operation S220.

[0095] In operation S330, a historical reference day is selected based on expert rules to obtain the second reference day. The process of selecting the second reference day can be referred to the previous section on selecting the first reference day in operation S230.

[0096] In operation S340, the fourth inbound volume feature for the second reference day's mid-prediction period is extracted in the same manner as the extraction of the second inbound volume feature. The extraction process for the fourth inbound volume feature can be referred to the previous description of extracting the second feature vector in operation S220.

[0097] In operation S350, a second feature vector is constructed based on the third and fourth inbound volume features. Similar to the first feature vector, the third and fourth inbound volume features can be combined into a vector to obtain the second feature vector.

[0098] In operation S360, the second feature vector is used as the input sample of the inbound traffic prediction model, and the inbound traffic volume of the prediction period in the historical date is used as the output sample of the inbound traffic prediction model to obtain training sample data.

[0099] In operating the S370, an inbound call volume prediction model is trained using training sample data. During training, the second feature vector is input into the inbound call volume prediction model, and the inbound call volume data output by the model is obtained. Then, based on the inbound call volume data output by the model and the actual inbound call volume data for the predicted period in historical dates, the prediction error is obtained. Based on this prediction error, the inbound call volume prediction model is optimized using a backpropagation algorithm. This is how the inbound call volume prediction model is trained.

[0100] According to one embodiment of this disclosure, multiple pending inbound call volume prediction models can be constructed using different algorithms. Then, using the training sample data obtained in operation S360, each of these multiple pending inbound call volume prediction models is trained. After training the multiple pending inbound call volume prediction models, their errors are compared on a test set. The model with the smallest error is then selected as the pending inbound call volume prediction model. In this way, a better algorithm model can be selected to construct the inbound call volume prediction model, thereby improving the prediction accuracy of the inbound call volume prediction model.

[0101] Figure 4 The diagram illustrates a call volume prediction model constructed using the random forest algorithm according to an embodiment of the present disclosure.

[0102] Figure 5 The illustration shows a schematic diagram of an inbound traffic prediction model constructed using the Gradient Boosting Tree (GBRT) algorithm according to an embodiment of the present disclosure.

[0103] Combination Figure 4 and Figure 5 For example, a portion of the collected basic data can be used as a training set and the other portion as a validation set. For instance, a random forest model and a gradient boosting decision tree (GBRT) model can be constructed separately for comparison and validation.

[0104] In the random forest model algorithm, different decision trees are uncorrelated; the final prediction value is obtained by averaging the predictions of all decision trees. The gradient boosting tree model algorithm (GBRT) is an additive model that sequentially trains a set of regression trees and finally sums the predictions of all regression trees to obtain a strong learner. Each new tree fits the negative gradient direction of the current loss function.

[0105] After training the two exemplary inbound call volume prediction models described above, the errors of the two models are compared on a test set. In one embodiment, the validation results on the test set are shown in Table 2 below, where the prediction period is the inbound call volume every 15 minutes:

[0106] Table 2:

[0107]

[0108] It should be noted that the error rate calculation formula in Table 2 is as follows:

[0109]

[0110] Table 2 compares the error rates of the results from manual prediction in the existing technology with those from the two undetermined inbound call volume prediction models mentioned above. Clearly, the accuracy of the prediction results from both models is improved compared to the manual prediction results.

[0111] From the two proposed inbound call volume prediction models, the model with the smaller error was selected as the inbound call volume prediction model. According to the verification results in Table 2, the Gradient Boosting Tree (GBRT) model has the smaller error and can be selected as the final inbound call volume prediction model.

[0112] Based on the above-described method for predicting inbound call volume for customer service according to embodiments of this disclosure, this disclosure also provides an apparatus for predicting inbound call volume for customer service according to embodiments of this disclosure. The following will be combined with... Figure 6 The device is described in detail.

[0113] Figure 6 A block diagram of an inbound call volume prediction device for customer service according to an embodiment of the present disclosure is shown schematically.

[0114] like Figure 6 As shown, according to an embodiment of this disclosure, the inbound traffic prediction device 600 may include a determination module 610, a first inbound traffic feature extraction module 620, a reference day selection module 630, a second inbound traffic feature extraction module 640, a feature vector construction module 650, and a prediction module 660. According to another embodiment of this disclosure, the inbound traffic prediction device 600 may further include a training module 670.

[0115] The determination module 610 is used to determine the forecast date and forecast period. In one embodiment, the determination module 610 can be used to perform the operation S210 described above.

[0116] The first incoming call feature extraction module 620 is used to extract the first incoming call features for the predicted midday forecast period. In one embodiment, the first incoming call feature extraction module 620 can be used to perform the operation S220 described above.

[0117] The reference day selection module 630 is used to select a historical reference day for the prediction day based on expert rules, thereby obtaining a first reference day. In one embodiment, the reference day selection module 630 can be used to perform the operation S230 described above.

[0118] The second incoming volume feature extraction module 640 is used to extract the second incoming volume features for the first reference day mid-prediction period. In one embodiment, the second incoming volume feature extraction module 640 can be used to perform the operation S240 described above.

[0119] The feature vector construction module 650 is used to construct a first feature vector based on a first inhalation volume feature and a second inhalation volume feature. In one embodiment, the feature vector construction module 650 can be used to perform the operation S250 described above.

[0120] The prediction module 660 is used to take the first feature vector as input to the trained inbound call volume prediction model, obtain the output of the inbound call volume prediction model, and thus obtain the inbound call volume for the predicted period during the day. In one embodiment, the prediction module 660 can be used to perform the operation S260 described above.

[0121] According to another embodiment of this disclosure, the inbound traffic prediction device further includes a training module 670. Specifically, the training module 670 includes: an acquisition submodule, a third inbound traffic feature extraction submodule, a reference day selection submodule, a fourth inbound traffic feature extraction submodule, a feature vector construction submodule, a first training submodule, and a second training submodule.

[0122] The acquisition submodule is used to acquire the inbound call volume for N time periods from at least one historical date, where the N time periods include the prediction time period, and N is a positive integer.

[0123] The third incoming call feature extraction submodule is used to extract the third incoming call feature for the predicted time period from historical dates in the same way as the first incoming call feature extraction.

[0124] The reference date selection submodule is used to select a historical reference date based on expert rules to obtain a second reference date.

[0125] The fourth incoming volume feature extraction submodule is used to extract the fourth incoming volume feature for the second reference day mid-prediction period in the same way as the second incoming volume feature extraction.

[0126] The feature vector construction submodule is used to construct a second feature vector based on the third and fourth inbound volume features.

[0127] The first training submodule is used to obtain training sample data by using the second feature vector as the input sample of the inbound traffic prediction model and the inbound traffic volume of the prediction period in the historical date as the output sample of the inbound traffic prediction model.

[0128] The second training submodule is used to train an inbound call volume prediction model using training sample data.

[0129] As an optional embodiment, the second training submodule further includes an algorithm construction unit, a prediction model training unit, a comparison unit, and a model selection unit.

[0130] The algorithm construction unit is used to build multiple prediction models for undetermined inbound call volume using different algorithms.

[0131] The prediction model training unit is used to train multiple prediction models for undetermined inbound call volume using training sample data.

[0132] The comparison unit is used to compare the errors of multiple pending inbound call volume prediction models on the test set after training.

[0133] The model selection unit is used to select the model with the smaller error from multiple pending inbound call volume prediction models as the inbound call volume prediction model.

[0134] Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure, or at least part of the functions of any one or more of them, can be implemented in one module. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be implemented by dividing them into multiple modules. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be at least partially implemented as hardware circuitry, such as Field Programmable Gate Arrays (FPGAs), Programmable Logic Arrays (PLAs), Systems-on-Chip, Systems-on-Substrate, Systems-on-Package, Application-Specific Integrated Circuits (ASICs), or implemented in hardware or firmware by any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be at least partially implemented as computer program modules, which, when run, can perform corresponding functions.

[0135] For example, any and multiple modules of the following can be implemented in one module: determination module 610, first incoming volume feature extraction module 620, reference day selection module 630, second incoming volume feature extraction module 640, feature vector construction module 650, prediction module 660, training module 670, acquisition submodule, third incoming volume feature extraction submodule, reference day selection submodule, fourth incoming volume feature extraction submodule, feature vector construction submodule, first training submodule, second training submodule, algorithm construction unit, prediction model training unit, comparison unit, and model selection unit. Alternatively, any one of these modules can be split into multiple modules. Or, at least some of the functionality of one or more of these modules can be combined with at least some of the functionality of other modules and implemented in one module. According to embodiments of this disclosure, at least one of the following modules can be at least partially implemented as hardware circuits, such as Field Programmable Gate Array (FPGA), Programmable Logic Array (PLA), System-on-Chip, System-on-Substrate, System-on-Package, Application-Specific Integrated Circuit (ASIC), or any other reasonable means of integrating or packaging circuits, or implemented in hardware or firmware, or implemented in software, hardware, or firmware, or in any appropriate combination of software, hardware, and firmware implementations. Alternatively, at least one of the following modules can be implemented, at least partially, as a computer program module: the determination module 610, the first incoming volume feature extraction module 620, the reference day selection module 630, the second incoming volume feature extraction module 640, the feature vector construction module 650, the prediction module 660, the training module 670, the acquisition submodule, the third incoming volume feature extraction submodule, the reference day selection submodule, the fourth incoming volume feature extraction submodule, the feature vector construction submodule, the first training submodule, the second training submodule, the algorithm construction unit, the prediction model training unit, the comparison unit, and the model selection unit. When the computer program module is run, it can perform the corresponding functions.

[0136] Figure 7 A block diagram of an electronic device 700 according to an embodiment of the present disclosure is shown schematically.

[0137] like Figure 7As shown, an electronic device 700 according to an embodiment of the present disclosure includes a processor 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage portion 708 into a random access memory (RAM) 703. The processor 701 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 701 may also include onboard memory for caching purposes. The processor 701 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0138] RAM 703 stores various programs and data required for the operation of electronic device 700. Processor 701, ROM 702, and RAM 703 are interconnected via bus 704. Processor 701 performs various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 702 and / or RAM 703. It should be noted that the programs may also be stored in one or more memories other than ROM 702 and RAM 703. Processor 701 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in said one or more memories.

[0139] According to embodiments of this disclosure, the electronic device 700 may further include an input / output (I / O) interface 705, which is also connected to a bus 704. The electronic device 700 may also include one or more of the following components connected to the I / O interface 705: an input section 706 including a keyboard, mouse, etc.; an output section 707 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as needed. A removable medium 711, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 710 as needed so that computer programs read from it can be installed into the storage section 708 as needed.

[0140] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0141] According to embodiments of this disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, such as, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this disclosure, the computer-readable storage medium may include ROM 702 and / or RAM 703 and / or one or more memories other than ROM 702 and RAM 703 described above.

[0142] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to cause the computer system to implement the methods provided in the embodiments of this disclosure.

[0143] When the computer program is executed by the processor 701, it performs the functions defined in the system / apparatus of this disclosure embodiments. According to embodiments of this disclosure, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0144] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 709, and / or installed from a removable medium 711. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0145] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 709, and / or installed from the removable medium 711. When the computer program is executed by the processor 701, it performs the functions defined in the system of this disclosure embodiment. According to embodiments of this disclosure, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0146] According to embodiments of this disclosure, program code for executing the computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can execute entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0147] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0148] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0149] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of this disclosure is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A method for predicting inbound call volume for customer service, characterized in that, include: Determine the forecast date and forecast period; Extract the first inbound call characteristics of the predicted time period within the predicted day; The first reference day is obtained by selecting a historical reference day for the prediction date based on expert rules. Extract the second inbound call volume feature for the predicted time period in the first reference day; Based on the first inhalation volume feature and the second inhalation volume feature, a first feature vector is constructed; as well as Using the first feature vector as the input to the trained inbound traffic volume prediction model, the output of the inbound traffic volume prediction model is obtained to obtain the inbound traffic volume of the predicted period in the predicted day. The historical reference date selected based on expert rules to obtain the first reference date includes: Obtain the attribute information of the predicted date, the attribute information including M attributes, where M is a positive integer; When the first attribute of the predicted date has a first attribute value, a date that matches the first attribute value is searched within a first time period before the predicted date, wherein the first attribute is one of the M attributes; When the first attribute of the predicted date is the value of the second attribute, search for a date that matches the value of the second attribute within a second time range before the predicted date; When M is greater than or equal to 2, the step of selecting the historical reference day for the prediction day based on expert rules to obtain the first reference day includes: The date that matches all M attributes of the predicted date is preferentially selected as the first reference date; and If there is no date that matches all M attributes of the predicted date, select a date that partially matches the M attributes of the predicted date as the first reference date.

2. The method according to claim 1, characterized in that, The first inhalation volume characteristic includes at least one of the following: Business time period characteristics determined based on the business content of the predicted day; The daily-granular time period characteristics are determined based on the time interval in which the predicted time period falls within a day; Historical time series characteristics are obtained based on the time series of inbound calls during the prediction period within a time period that meets preset conditions prior to the prediction date; or The time window features are obtained based on the statistical values ​​of inbound calls within historical dates that have a second cycle relationship with the predicted date.

3. The method according to claim 2, characterized in that, The historical time-series features include at least one of the following: The statistical value of inbound calls during the forecast period within the first cycle prior to the forecast date; or The inbound call volume for the predicted period is calculated based on the R most recent dates that have the same attributes as the predicted date, prior to the predicted date; where R is a positive integer.

4. The method according to claim 2, characterized in that, The time window features include at least one of the following: Statistical values ​​of inbound calls within historical dates that have a second periodic relationship with the predicted date; Statistical values ​​of inbound calls during the predicted period within historical dates that have a second periodic relationship with the predicted date; The statistical values ​​of inbound calls within historical dates that have a second periodic relationship with the predicted date and are all either working days or non-working days; in, The second cyclical relationship includes at least one of the following: weekly cyclical relationship, monthly cyclical relationship, quarterly cyclical relationship, or annual cyclical relationship.

5. The method according to claim 1, wherein, The extraction of the second inbound call feature for the prediction period in the first reference day includes: Obtain the historical inbound traffic volume of the first reference day during the prediction period; The historical inbound call volume is scaled according to a predetermined scaling rule to obtain the second inbound call volume feature.

6. The method according to claim 1, characterized in that, The inbound call volume prediction model is trained in the following manner: Obtain the inbound call volume for N time periods from at least one historical date, wherein the N time periods include the predicted time period, and N is a positive integer; Extract the third inbound call feature from the predicted time period in the historical dates in the same manner as extracting the first inbound call feature; Based on the expert rules, a historical reference day is selected for the historical date to obtain the second reference day; The fourth inbound volume feature for the prediction period in the second reference day is extracted in the same manner as the second inbound volume feature extraction. Based on the third and fourth inbound volume features, a second feature vector is constructed; and The second feature vector is used as the input sample of the inbound traffic prediction model, and the inbound traffic during the prediction period in the historical date is used as the output sample of the inbound traffic prediction model to obtain training sample data. The inbound call volume prediction model is trained using the training sample data.

7. The method according to claim 6, characterized in that, The training process of the inbound call volume prediction model also includes: Multiple prediction models for undetermined inbound call volume were constructed using different algorithms; The step of training the inbound traffic volume prediction model using the training sample data includes: training the plurality of pending inbound traffic volume prediction models respectively using the training sample data. After training the multiple undetermined inbound call volume prediction models, the errors of the multiple undetermined inbound call volume prediction models are compared on a test set; and The model with the smaller error among the multiple undetermined inbound call volume prediction models is selected as the inbound call volume prediction model.

8. A call volume prediction device for customer service, characterized in that, The device includes: The determination module is used to determine the forecast date and forecast period; The first inbound call volume feature extraction module is used to extract the first inbound call volume features of the predicted period in the predicted day; The reference day selection module is used to select a historical reference day for the prediction day based on expert rules, thus obtaining the first reference day. The second incoming volume feature extraction module is used to extract the second incoming volume features of the prediction period in the first reference day; The feature vector construction module is used to construct a first feature vector based on the first inhalation volume feature and the second inhalation volume feature; and The prediction module is used to take the first feature vector as the input of the trained inbound traffic volume prediction model, obtain the output of the inbound traffic volume prediction model, and obtain the inbound traffic volume of the prediction period in the prediction day. The historical reference date selected based on expert rules to obtain the first reference date includes: Obtain the attribute information of the predicted date, the attribute information including M attributes, where M is a positive integer; When the first attribute of the predicted date has a first attribute value, a date that matches the first attribute value is searched within a first time period before the predicted date, wherein the first attribute is one of the M attributes; When the first attribute of the predicted date is the value of the second attribute, search for a date that matches the value of the second attribute within a second time range before the predicted date; When M is greater than or equal to 2, the step of selecting the historical reference day for the prediction day based on expert rules to obtain the first reference day includes: The date that matches all M attributes of the predicted date is preferentially selected as the first reference date; and If there is no date that matches all M attributes of the predicted date, select a date that partially matches the M attributes of the predicted date as the first reference date.

9. A computer device, comprising: One or more processors; Memory, used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors perform the method according to any one of claims 1-7.

10. A computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the method of any one of claims 1-7.

11. A computer program product comprising computer program instructions that, when executed by a processor, implement the method of any one of claims 1-7.

Images