KR102377402B1 - Call recommendation system and call recommendation method based on artificial intelligence - Google Patents

Abstract

The present invention provides a call recommendation system based on artificial intelligence. The call recommendation system includes a data collection unit, a matching time prediction unit, a price decision unit, and a final ranking unit. When a service is requested from a service user, the data collection unit collects first past data representing the past location of the service user, first current data representing the current location of the service user, and second past data representing the past location of the service provider and the service provider. Collect second current data indicating the current location. The matching time prediction unit inputs the first and second past data and the first and second current data into the RNN (Recurrent Neural Network) learning model to predict the future location of the service user and the future location of the service provider, and predicts the future location of the service user. When the service provider selects a service by inputting the first prediction data about the location and the second prediction data about the future location of the service provider into a predictive learning model, the service provider will be matched with the next service user after the service provider completes the service. Predict the matching time it will take to complete. The pricing unit determines the price so that the price for the service increases as the matching time increases. The final ranking unit determines a service recommendation ranking among the services requested from the service provider based on the price and preference data indicating the service provider's preference for the service. Each of the RNN learning model and prediction learning model is based on a deep learning algorithm.

Description

Call recommendation system and call recommendation method based on artificial intelligence {CALL RECOMMENDATION SYSTEM AND CALL RECOMMENDATION METHOD BASED ON ARTIFICIAL INTELLIGENCE}

The present invention relates to a call recommendation system, and more specifically to a call recommendation system based on artificial intelligence.

Recently, the transportation service industry using transportation means has become popular. For example, transportation services include a taxi service in which the service provider takes the customer to the destination using his or her own vehicle, a proxy service in which the service provider drives the customer's vehicle to take the customer to the destination, or a quick service that transports various cargoes, etc. There is.

These transportation services require matching between service users who want to use the service and service providers who want to provide the service. Specifically, a service user can request a service from a service provider using a phone, an application, etc. The service provider may receive information about the requested service, such as the service user's destination, the service user's current location, etc., along with the service request. The service provider may decide whether to approve the service request based on the information of the service. When the service provider approves the service request, the service provider and the service requester are matched, and the service requester can receive the service from the service provider.

As the demand for transportation services increases, service providers may receive multiple service requests simultaneously. In this case, the service provider may have difficulty selecting the service request that best suits his or her preferences among multiple service requests. Therefore, there is a need for a system and method that recommends service requests to the service provider according to the service provider's preferences.

The present invention is intended to solve the above-described technical problems, and the purpose of the present invention is to provide a call recommendation system and call recommendation method based on artificial intelligence.

A call recommendation system according to an embodiment of the present invention may include a data collection unit, a matching time prediction unit, a price determination unit, and a final ranking unit. When a service is requested from a service user, the data collection unit collects first past data representing the past location of the service user, first current data representing the current location of the service user, and second past data representing the past location of the service provider and the service provider. Second current data indicating the current location of can be collected. The matching time prediction unit inputs the first and second past data and the first and second current data into the RNN (Recurrent Neural Network) learning model to predict the future location of the service user and the future location of the service provider, and predicts the future location of the service user. When the service provider selects a service by inputting the first prediction data about the location and the second prediction data about the future location of the service provider into a predictive learning model, the service provider will be matched with the next service user after the service provider completes the service. It can be configured to predict the matching time it will take until. The pricing unit may determine the price so that the price for the service increases as the matching time increases. The final ranking unit may determine a service recommendation ranking among services requested from the service provider, based on preference data indicating the service provider's preference for the service and the price. Each of the RNN learning model and prediction learning model may be based on a deep learning algorithm.

The present invention is a call recommendation method implemented on a computer according to another embodiment. When a service is requested from a service user, the present invention provides first past data indicating the past location of the service user and the current location of the service user by the call recommendation system. collecting, by the call recommender system, first current data representing and second historical data representing a past location of the service provider and second current data representing a past location of the service provider; Input the second past data and the second current data into a Recurrent Neural Network (RNN) learning model to predict the future location of the service user and the future location of the service provider, and first predict data regarding the future location of the service user and the service provider. A step of predicting the matching time required from the time the service provider completes the service until the service provider is matched with the next service user by inputting second prediction data about the future location of the service into a predictive learning model, by a call recommendation system. calculating the price so that the price increases as the matching time increases, and by the call recommendation system, selecting a service among the services requested from the service provider based on the price and preference data indicating the service provider's preference for the service. It may include a step of determining the recommendation ranking. RNN learning models and predictive learning models may be based on deep learning algorithms.

The call recommendation system according to an embodiment of the present invention is based on artificial intelligence and can recommend a call that matches the service provider's preferences to the service provider.

Specifically, the call recommendation system can predict the locations of service providers and service users using a Recurrent Neural Network (RNN) algorithm. The call recommendation system can predict the matching time until the next call is matched based on the predicted locations of the service provider and service user. The call recommendation system can dynamically determine the price of a call based on the current call matching situation and predicted matching time. The call recommendation system may determine the ranking of calls requested to the service provider based on the determined price of the call and the preference of the service provider.

Accordingly, the service provider can use the call recommendation system to select a call that suits his or her preferences.

1 is a conceptual diagram for explaining a call recommendation system according to an embodiment of the present invention.
FIG. 2 is a block diagram for explaining the structure of the call recommendation system of FIG. 1.
FIG. 3 is a flowchart for explaining the operations of the call recommendation system of FIG. 2.
FIG. 4 is a flowchart for explaining the operation of the matching time prediction unit of FIG. 2.
FIG. 5 is a flowchart for explaining the operation of the matching time prediction unit of FIG. 2.
FIG. 6 is a conceptual diagram for explaining the operation of the price determination unit and the final ranking unit of FIG. 2.

Hereinafter, embodiments of the present invention will be described clearly and in detail so that a person skilled in the art can easily practice the present invention.

1 is a conceptual diagram for explaining a call recommendation system according to an embodiment of the present invention.

The present invention provides a call recommendation method and a call recommendation system (1000) that can efficiently match service users and service providers in transportation services using transportation means. In this specification, “transportation service” refers to a taxi service in which the service provider uses its own vehicle to take the service user to the destination, a proxy service in which the service provider drives the service user's vehicle to take the service user to the destination, or various services to the service user. It refers to services provided by service providers to service users using transportation methods, such as quick service for transporting cargo. In this specification, “service user” refers to a customer requesting a service, and “service provider” refers to a driver providing the service.

A service user can request a service using the service user application 2200. The service user application 2200 may be a program installed on the service user's electronic device. When a service user requests a service, the service user application 2200 may transmit the service user's service request to the call recommendation system 1000 through the server 2100.

The call recommendation system 1000 may receive a service request from a service user through the server 2100. In this specification, receiving a service request may also be expressed as saying that a call has been received.

Other service users can also request services through service user applications 3200 and 4200, respectively. Servers 3100 and 4100 provide substantially the same operations as server 2100. The service user applications 3200 and 4200 provide substantially the same operations as the service user application 2200, but may be different types of applications. Service provider applications 3300 and 4300 also provide substantially the same operations as service provider application 2300, but may be different types of applications.

When a call is received, the call recommendation system 1000 may receive location information of service users and location information of service providers through the servers 2100, 3100, and 4100. The server 2100 may collect the service user's location information and the service provider's location information using the service user application 2200 and the service provider application 2300. The service user application 2200 and the service provider application 2300 may be programs installed on the electronic device carried by the service user and the electronic device carried by the service provider, respectively. The service user application 2200 and the service provider application 2300 may receive location information from a GPS device installed on the electronic device carried by the service user and the electronic device carried by the service provider.

Additionally, when a call is received, the call recommendation system 1000 may receive preference data of service providers through the servers 2100, 3100, and 4100. Preference data may be information about the service provider's preferred destination, driving time, driving distance, price, etc.

The call recommendation system 1000 may use the location information and preference information received from the servers 2100, 3100, and 4100 to determine the rank of received calls according to the service provider's preference. The call recommendation system 1000 can determine the ranking of received calls based on artificial intelligence. In this specification, “artificial intelligence” refers to technology that realizes human learning ability, reasoning ability, perception ability, and natural language understanding ability through computer programs. As an example, the call recommendation system 1000 may use a deep learning algorithm composed of a neural network structure. The deep learning algorithm used in the call recommendation system 1000 is described in detail with reference to FIGS. 2 to 5.

The call recommendation system 1000 may transmit call ranking information to the service provider application 2300 through the server 2100. The service provider can obtain ranking information through the service provider application 2300. The service provider can select a call that suits his or her preference among calls through the service provider application 2300, based on the ranking information. When the service provider selects a call, the service user who requested the call and the service provider who selected the call may be matched with each other.

When the service provider selects a call, the call recommendation system 1000 may receive information about the call actually selected by the service provider. The call recommendation system 1000 can train an artificial intelligence model used in the call recommendation system 1000 using information about the call actually selected by the service provider. In other words, information about the call actually selected can be used as correct answer information. The structure and operation of the call recommendation system 1000 are described in detail with reference to FIGS. 2 to 6.

The components (1000, 2100-2300, 3100-3300, and 4100-4300) of FIG. 1 may be connected through a network. In this specification, “network” refers to a connection structure that allows information exchange between nodes, such as a plurality of terminals and servers. By way of example, the network may include RF, 3rd Generation Partnership Project (3GPP) network, Long Term Evolution (LTE) network, 5th Generation Partnership Project (5GPP) network, World Interoperability for Microwave Access (WIMAX) network, Internet, LAN ( Local Area Network), Wireless LAN (Wireless Local Area Network), WAN (Wide Area Network), PAN (Personal Area Network), Bluetooth network, NFC network, satellite broadcasting network, analog broadcasting network, DMB (Digital Multimedia Broadcasting) ) network, etc., but is not limited to this.

Electronic devices on which the service user applications 2200, 3200, and 4200 and the service provider applications 2300, 3300, and 4300 are installed may be implemented as electronic devices capable of accessing a remote server or terminal through a network. For example, electronic devices are wireless communication devices that ensure portability and mobility, such as navigation, PCS (Personal Communication System), GSM (Global System for Mobile communications), PDC (Personal Digital Cellular), PHS (Personal Handyphone System), and PDA. (Personal Digital Assistant), IMT (International Mobile Telecommunication)-2000, CDMA (Code Division Multiple Access)-2000, W-CDMA (W-Code Division Multiple Access), Wibro (Wireless Broadband Internet) terminal, smartphone , may include all types of handheld-based wireless communication devices such as smartpads, tablet PCs, etc.

The servers 2100, 3100, and 4100 and the call recommendation system 1000 may be implemented as computers that can access remote servers or terminals through a network. Here, the computer may include, for example, a laptop equipped with a navigation system and a web browser, a desktop, a laptop, etc. Additionally, the call recommendation system 1000 of the present invention may provide services in conjunction with existing applications.

FIG. 2 is a block diagram for explaining the structure of the call recommendation system of FIG. 1.

The call recommendation system 1000 includes a data collection unit 1100, a matching time prediction unit 1200, a price decision unit 1300, a final ranking unit 1400, an answer information collection unit 1500, and an update unit 1600. may include. In this specification, 'part' includes a unit realized by hardware, a unit realized by software, and a unit realized using both. Additionally, one unit may be realized using two or more pieces of hardware, and two or more units may be realized using one piece of hardware. Additionally, the call recommendation system 1000 may additionally include components not shown in FIG. 2 or may not include some of the components 1100 to 1600 shown in FIG. 2 .

The call recommendation system 1000 can receive a call from the server 2100. When a call is received, the data collection unit 1100 may receive the service user's location data and the service provider's location data from the server 2100. The service user's location data may include current data about the service user's current location and past data about the service user's past location. The location data of the service provider may include current data about the current location of the service provider and past data about the past location of the service provider.

As an example, the data collection unit 1100 may receive location information of service users by time. If the time at which the data collection unit 1100 receives the service user's location information is the first time, the current location of the service user may be the service user's location at the first time. In this case, the past location of the service user may be the location of the service user by time before the first time. Specifically, the data collection unit 1100 collects the second time, the third time... Past data regarding the location of the service user at each N time can be received. In this case, the second time, the third time... . The Nth time may be different times before the first time. 'N' may be an integer of 3 or more. The service provider's location data may also be collected in the same form as the service user's location data.

As another example, the data collection unit 1100 may receive location information of service users for each time section. When the time at which the data collection unit 1100 receives the service user's location information is the first time, the current location of the service user may be the average position of the service user in the first time section including the first time. In this case, the past location of the service user may be the average location of the service user for each time section before the first time section. Specifically, the data collection unit 1100 is a second time section, a third time section... Past data regarding the average location of service users in each N time interval can be received. In this case, the second time interval, the third time interval... . The Nth time interval may be different time intervals before the first time interval. 'N' may be an integer of 3 or more. The data collection unit 1100 can collect the service provider's location data in the same format as the service user's location data. First time segment, second time segment... The time length of each N time section and the size of 'N' may be changed by the update unit 1600. Based on the correct answer information, the update unit 1600 creates a first time section, a second time section, etc. to increase the prediction accuracy of the matching time prediction unit 1200. The time length of each N time section and the size of 'N' can be changed.

The data collection unit 1100 may output the location data received from the server 2100 to the matching time prediction unit 1200. Hereinafter, location data is interpreted to include past and present location data of service users and past and present location data of service providers. The matching time prediction unit 1200 can predict the future locations of service users and service providers by inputting location data into a Recurrent Nerual Network (RNN) learning model. The data collection unit 1100 may use an RNN learning model to generate service user prediction data indicating the future location of the service user and service provider prediction data indicating the service provider's future location. Hereinafter, the predicted data is interpreted to include future location data of the service user and future location data of the service provider.

RNN (Recurrent Nerual Network) learning model may be a deep learning algorithm used to predict the future location of service users and service providers based on their past and current locations. As an example, the Recurrent Nerual Network (RNN) learning model may be one of an RNN algorithm, a Convolution Neural Network (CNN)-RNN algorithm, a Long Short Term Memory (LSTM) algorithm, a CNN-LSTM algorithm, etc.

The matching time prediction unit 1200 may predict the future locations of service users and service providers using a CNN-RNN algorithm. The CNN-RNN algorithm may be an artificial intelligence algorithm that predicts the future location of service users and service providers based on location data converted to image data.

Additionally, the matching time prediction unit 1200 may predict the future locations of service users and service providers using the CNN-LSTM algorithm. The CNN-LSTM algorithm may also be an artificial intelligence algorithm that predicts the future location of service users and service providers based on location data converted to image data. The LSTM algorithm may be an improved algorithm of the RNN algorithm. The matching time prediction unit 1200 is not limited to this and may operate based on a deep learning algorithm that can derive future data using past and present data.

The data collection unit 1100 may receive additional data from the server 2100. Additional data may be additional information about service users and service providers. Additional data may include information related to the service requested by the service user, such as the service user's origin, the service user's destination, the time required for the service provider to perform the service, and the driving distance from the origin to the destination. Additionally, the additional data may include time information from a mid- to long-term perspective, such as the time the service was requested, the day of the week the service was requested, whether the date the service was requested was a public holiday, the season of the date the service was requested, etc. Additional data is not limited to this and may also include information about the location and status of the service provider.

The matching time prediction unit 1200 may receive additional data from the data collection unit 1100. The matching time prediction unit 1200 may embed additional data and convert it into vector data. In the following descriptions, “vector data” means data in which additional data has been converted into vector format.

The matching time prediction unit 1200 inputs prediction data and vector data into a prediction learning model to predict the matching time required from performing the selected service to matching with the next service user when the service provider selects a specific service. You can. In this specification, “matching time” means the time taken from performing the currently received call until matching with the next call when the service provider selects the currently received call. However, the present invention is not limited to this, and “matching time” may mean the time taken from when the service provider starts performing the currently received call until it is matched with the next call when the service provider selects the currently received call. there is.

The predictive learning model may be a learning model based on a deep learning algorithm to predict matching time based on the future location of the service user and service provider and additional data. The matching time prediction unit 1200 may rank currently received calls in descending order of matching time, based on a predictive learning model. The matching time prediction unit 1200 may output matching time information of calls to the price determination unit 1300.

The price determination unit 1300 may determine the price of each call by inputting the matching time information into the dynamic pricing model. The price of each call can be dynamically determined based on matching time and waiting time. As an example, the price of a call may increase as the square of the call's matching time increases, and may increase as the call's waiting time increases. In this specification, “waiting time” may mean the time during which a call is not acknowledged by the service provider. In other words, “waiting time” may mean the time during which the service user who received the call is not matched with the service provider. However, the present invention is not limited to this, and the price determination unit 1300 may determine the price of the call so that the price of the call is proportional to the value of the resources used by the service provider to perform the service. The method by which the price determination unit 1300 determines the price of a call is described in detail with reference to FIG. 6.

The price determination unit 1300 may determine the ranking of calls in descending order of price. The price determination unit 1300 may output call price information to the final ranking unit 1400.

The final ranking unit 1400 may receive call price information from the price determination unit 1300. Additionally, the final ranking unit 1400 may receive preference data from the data collection unit 1100. Preference data may indicate a service provider's preference for a service. For example, the service provider may not prefer corporate calls, may not prefer that the service user's vehicle is a foreign car, may not prefer that the distance from the service provider's current location to the service user's origin be more than a certain distance, or that the destination is not more than a certain distance. You may prefer to be close, or you may prefer not to have a passenger. Preference data may include information about the preferences of these service providers.

The final ranking unit 1400 may determine the final ranking of calls based on price information and preference data of the calls. The final ranking unit 1400 may filter calls based on preference data for computational efficiency. For example, the final ranking unit 1400 may determine only the final ranking of the remaining calls after excluding some calls that the service provider does not want.

The final ranking unit 1400 may determine the final score for each of the remaining calls based on the service provider's preference for the call price and call matching time. The final ranking unit 1400 may determine the final ranking of the remaining calls in order of highest final score. The final score of a call can be increased by increasing the price of the call and/or decreasing the matching time of the call. Additionally, the final score of a call may decrease by decreasing the price of the call and/or increasing the matching time of the call.

The final ranking unit 1400 may transmit final ranking information about the final ranking of calls to the service provider application 2300 through the server 2100.

The service provider can receive final ranking information of calls through the service provider application 2300. The service provider can select a call that suits his or her preference among the calls through the service provider application 2300, based on the final ranking information of the calls. The service provider may provide a service corresponding to the selected call to the service requester.

After the service is completed, the correct answer information collection unit 1500 may receive correct answer information from the server 2100. In this specification, “correct answer information” may indicate actual results for situations predicted by the call recommendation system 1000. As an example, if the call recommendation system 1000 predicts the locations of the service provider and service user at the first time, the correct answer information may include information about the actual location of the service provider and service user at the first time. there is. As another example, the correct answer information may include information about the service actually selected by the service provider among the services recommended to the service provider, and the actual matching time it took for the service provider to match with the next service user after the selected service was performed. You can. The correct answer information collection unit 1500 may output the received correct answer information to the update unit 1600.

The update unit 1600 may receive correct answer information from the correct answer information collection unit 1500. Based on the correct answer information, the update unit 1600 can update the data collection unit 1100, matching time prediction unit 1200, price determination unit 1300, etc. to improve the prediction accuracy of the call recommendation system 1000. there is.

The update unit 1600 may change the amount and format of data collected by the data collection unit 1100 based on the correct answer information. For example, when the data collection unit 1100 collects data about the average location of service users in each of the N time intervals from the first time interval, the time intervals (the first The time length of the time interval to the Nth time interval) and/or the size of “N” may be changed.

The update unit 1600 may train deep learning-based learning models (RNN learning model, prediction learning model) of the matching time prediction unit 1200 based on the correct answer information. Due to the update operation of the update unit 1600, the prediction accuracy of the matching time prediction unit 1200 may be improved.

The update unit 1600 may update the dynamic pricing model of the price determination unit 1300 based on the correct answer information. If there is a change in the service provider's preference, the update operation of the update unit 1600 allows the price determination unit 1300 to determine a dynamic price by reflecting the changed service provider's preference.

In other words, the present invention can be used in the field of designated driver service system, personalized recommendation algorithm, demand prediction, and dynamic price determination algorithm. The present invention can predict matching time based on a deep learning algorithm. The present invention can increase service providers' operation rate, individual expected revenue cost, and individual cost efficiency by ranking calls by considering the predicted matching time and the service provider's preference.

In addition, the present invention uses a dynamic price determination algorithm that determines the price by considering the matching time, which can solve the phenomenon of non-preferred calls not being matched, and can help service providers and service users with reasonable price changes according to the laws of supply and demand. can be mutually satisfying. The present invention can increase the matching rate between service providers and service users by recommending calls by reflecting the preferences of service providers.

FIG. 3 is a flowchart for explaining the operations of the call recommendation system of FIG. 2.

In operation S110, the call recommendation system 1000 of FIG. 2 may receive a call from a service user.

In operation S120, when a call is received, the call recommendation system 1000 may collect location data about service users and service providers. Specifically, the recommendation system 1000 may collect past location data and current location data of service users and service providers.

In operation S130, the call recommendation system 1000 may predict the future locations of service users and service providers based on location data and a RNN learning model.

In operation S140, the call recommendation system 1000 may predict the matching time of a service based on the future locations of service users and service providers, additional data, and a predictive learning model. Since “additional data” and “matching time” have been described with reference to FIG. 2, redundant descriptions will be omitted below.

In operation S150, the call recommendation system 1000 may determine a price for each of the services based on a dynamic pricing model. The price of the service may increase as the square of the matching time of the service predicted in operation S140 increases. Additionally, the price of the service may increase as the waiting time of the service increases. Since “waiting time” has been described with reference to FIG. 2, redundant description will be omitted below.

In operation S160, the call recommendation system 1000 may calculate a final score for each service that suits the service provider's preference, based on preference data and the price of each service. Specifically, the call recommendation system 1000 may filter calls that are not preferred by the service provider based on preference data. The call recommendation system 1000 can calculate the final score for each call preferred by the service provider. The final score of a service may increase as the price of the service increases and may decrease as the matching time of the service increases.

In operation S170, the call recommendation system 1000 may determine the final ranking of services in order of highest final score. The service provider may select at least one of the services by referring to the final ranking of the services. The service provider may perform the selected service.

After the service provider performs the selected service, in operation S180, the call recommendation system 1000 may collect correct answer information for the prediction. “Correct answer information” has been described with reference to FIG. 2, so redundant description will be omitted below.

In operation S190, the call recommendation system 1000 may update the configurations 1100 to 1400 of the call recommendation system 1000 based on the correct answer information. Specifically, the call recommendation system 1000 can use the correct answer information to train deep learning-based learning models stored in the call recommendation system 1000. Accordingly, the prediction accuracy of the call recommendation system 1000 can be improved.

FIG. 4 is a flowchart for explaining the operation of the matching time prediction unit of FIG. 2.

The matching time prediction unit 1200 of FIG. 2 may receive location data from the data collection unit 1100. Location data collected from the data collection unit 1100 may include time-dependent location information of service users and time-dependent location information of service providers. The matching time prediction unit 1200 can dataize the locations of service users and service providers from the past to the present in specific time units. Specifically, the location data of the service provider may represent the average location of the service provider in each of the time sections from the first time section to the Nth time section. As an example, the time length of each of the time intervals may be 30 minutes, and 'N' may be 8. In this case, the matching time prediction unit 1200 can predict the location of the service provider 30 minutes later based on the location of the service provider over the past 4 hours. The location data may be in the form of a matrix, and each entry may display the location coordinates of the service provider. One matrix may represent the average location of service providers in a specific time period. The service user's location data may also be in a similar form to the service provider's location data.

The matching time prediction unit 1200 may input location data of service users and location data of service providers into the RNN algorithm. As mentioned above, the present invention is not limited to this, and the matching time prediction unit 1200 may use the LSTM algorithm among the RNN algorithms. The matching time prediction unit 1200 can predict the future location of service users and the future location of service providers using a RNN algorithm.

The matching time prediction unit 1200 may generate vector data by embedding additional data using a vectorization algorithm. Additional data may include service data regarding the service and mid- to long-term data related to the date the service was requested. As an example, service data may include information about the service user's origin, the service user's destination, the time required to perform the service, and the distance from the origin to the destination. Mid- to long-term data may include information about the current time, the day of the week on the date the service is requested, whether the date the service is requested is a public holiday, and the season on the date the service is requested. Mid- to long-term data may be data for finding the location patterns of service providers and service users and inherent patterns of calls from a mid- to long-term (e.g., weekly, monthly, annual) perspective. Additional data may be used as information to improve prediction accuracy for future locations and/or matching times of service users and service providers.

The matching time prediction unit 1200 may input result data and vector data output from the RNN algorithm to the matching time prediction algorithm. The matching time prediction unit 1200 can predict the matching times of calls using a matching time prediction algorithm. The matching time prediction unit 1200 may determine the ranking of calls in the order of shortest predicted matching time.

FIG. 5 is a flowchart for explaining the operation of the matching time prediction unit of FIG. 2.

The matching time prediction unit 1200 of FIG. 2 may receive location data from the data collection unit 1100. Location data collected from the data collection unit 1100 may include time-dependent location information of service users and time-dependent location information of service providers. The method by which the matching time prediction unit 1200 collects location data may be substantially the same as the method described with reference to FIG. 4 . Accordingly, redundant description below will be omitted.

The matching time prediction unit 1200 can convert the locations of service users and service providers from the past to the present into a two-dimensional heat map according to specific time units.

The matching time prediction unit 1200 may use heat map-type location data of service users and location data of service providers as input data for the CNN-RNN algorithm. As mentioned above, the present invention is not limited to this, and the matching time prediction unit 1200 may use the CNN-LSTM algorithm among the CNN-RNN algorithms. The matching time prediction unit 1200 can predict the future location of service users and the future location of service providers using a CNN-RNN algorithm.

In substantially the same way as described with reference to FIG. 4, the matching time prediction unit 1200 can predict the matching times of calls using result data and vector data output from the CNN-RNN algorithm. The matching time prediction unit 1200 may determine the ranking of calls in the order of shortest predicted matching time.

FIG. 6 is a conceptual diagram for explaining the operation of the price determination unit and the final ranking unit of FIG. 2.

The pricing unit 1200 of FIG. 2 can determine the prices of calls using a dynamic pricing algorithm. The price determination unit 1200 can calculate the price of calls using [Equation 1] below.

[Equation 1] may be an equation for finding the price of the service at time 't'. Referring to [Equation 1], “P _t (Di)” may be the price of the service at time ‘t’. “P _t-1 (Di)” means the price of the service at time ‘t-1’. Time 't-1' may mean a time in the past than time 't'. The price determination unit 1300 of FIG. 2 may periodically update the price of the service. In this case, “P _t-1 (Di)” means the price of the service before the service price was updated to “P _t (Di)”. “l _t (Di)” may mean the waiting time when the service is not matched. “W(l _t )” may mean the weight that waiting time will have on the price of the service. In other words, the price of the service (P _t (Di)) can increase as the waiting time (l _t (Di)) of the service increases.

Referring to [Equation 1], “W _t+1 (Di)” may mean the matching time required from when the service provider completes the service until matching with the next service. “w(W _t+1 )” may mean the weight that the matching time will have on the price of the service. The weights (w(l _t ), w(W _t+1 )) may be determined according to the service provider's preference for high prices and preference for short matching times. In other words, the price of the service (P _t (Di)) may increase as the value of the square of the matching time of the service (W _t+1 (Di) ² ) increases.

In other words, the price of the service (P _t (Di)) may increase as the waiting time of the service (l _t (Di)) and the matching time of the service (W _t+1 (Di)) increase. In this specification, the expression that “X” increases in proportion to “Y” means that “Y” also increases as “X” increases. A long service waiting time (l _t (Di)) means that service providers do not prefer the service. Services with a long matching time (W _t+1 (Di)) are also likely to be not preferred by service providers. Therefore, the present invention increases the price of the service (P _t (Di)) as the waiting time (l _t (Di)) and matching time (W _t+1 (Di)) of the service increases, so that the service provider selects This can prevent unanswered starvation calls from occurring. According to the present invention, starvation calls that are not selected by the service provider can be reduced, thereby improving the quality of service of the present invention.

The price determination unit 1200 may determine dynamic prices of services (A, B, C, D, and E). Referring to Figure 6, if the services (A, B, C, D, E) are listed in descending order of dynamic price, “C”, “D”, “E“, “A”, “B” are listed in that order. It can be.

As mentioned above, the price determination unit 1200 may periodically update the dynamic price of the service. In this process, new services may be added to the ranking subject at the request of the service user, and when the service provider selects a service, the selected service may be excluded from the ranking determination subject. Additionally, in this process, the dynamic price of services with increased waiting time and/or matching time may increase. Accordingly, the ranking of services according to dynamic prices may change.

The final ranking unit 1400 of FIG. 2 may receive price information about the dynamic prices of services (A, B, C, D, and E). The final ranking unit 1400 may receive preference data from the data collection unit 1100 of FIG. 2 . The final ranking unit 1400 may filter the service (D) that does not meet the service provider's preference based on preference data. The final ranking unit 1400 can improve calculation efficiency by determining the final ranking of the remaining services (A, B, C, and E) excluding the service (D).

The final ranking unit 1400 can calculate the final score for each of the services (A, B, C, and E) using [Equation 2] below. The final ranking unit 1400 may determine the final ranking of the services (A, B, C, and E) in order of highest final score.

[Equation 2] may be an equation for calculating the final score (Ri) of the service. As explained with reference to [Equation 1], “P _t (Di)” is the price of the service at time ‘t’, and “W _t+1 (Di)” is the price after the service provider completes the service. This may be the matching time required until matching with the next service. “m(P _t )” may mean the weight that the price of the service will have on the final score of the service. “m(W _t+1 )” may mean the weight that the matching time of the service will have on the final score of the service. The weights (m(P _t ), m(W _t+1 )) may be determined according to the service provider's preference for high price and the service provider's non-preference for high matching time.

Referring to [Equation 2], the final score of the service may decrease by decreasing the price and/or increasing the matching time. Additionally, the final score of a service may increase by increasing the price and/or decreasing the matching time. The final score of a service may increase as the price increases and may decrease as the matching time increases. That is, the final score of the service is not limited to the present invention, and may be determined by a mathematical equation configured such that the final score of the service increases as the price increases and decreases as the matching time increases. For example, the final score of the service is It can also be determined by [Equation 3] below.

Referring to [Equation 3], “m ₁ (P _t )” may mean the weight that the price of the service will have on the final score of the service. “m ₁ (W _t+1 )” may mean the weight that the matching time of the service will have on the final score of the service.

The final ranking unit 1400 may determine the final score for each of the services (A, B, C, and E). Referring to FIG. 6, if the services (A, B, C, E) are listed in order of highest final score, they may be listed in the order of “C”, “E“, “A”, and “B”.

The above-described details are specific embodiments for carrying out the present invention. The present invention will include not only the above-described embodiments, but also embodiments that can be simply changed or easily changed in design. In addition, the present invention will also include technologies that can be easily modified and implemented using the embodiments. Therefore, the scope of the present invention should not be limited to the above-described embodiments, but should be determined by the claims and equivalents of the present invention as well as the claims described later.

Claims (10)

When a service is requested from a service user, first past data indicating the past location of the service user and first current data indicating the current location of the service user and second past data indicating the past location of the service provider and the service provider a data collection unit configured to collect second current data indicating the current location of;
The first and second past data and the first and second current data are input into a Recurrent Neural Network (RNN) learning model to predict the future location of the service user and the future location of the service provider, and predict the future location of the service user. When the service provider selects the service by inputting the first prediction data about the future location and the second prediction data about the future location of the service provider into a predictive learning model, the service provider completes the service. a matching time prediction unit configured to predict the matching time it will take for the service provider to match with the next service user;
a price determination unit that determines the price so that the price for the service increases as the matching time increases; and
A final ranking unit that determines a recommendation ranking of the service among services requested from the service provider based on the price and preference data indicating the service provider's preference for the service,
Each of the RNN learning model and the prediction learning model is based on a deep learning algorithm,
The price determination unit adjusts the price if the service user is not matched with a service provider using the call recommendation system after a certain time,
The adjusted price is (the first weight) * (the waiting time when the service user is not matched with the service provider using the call recommendation system) * (the price) + (the second weight) * (the matching time) ^Two- person call recommendation system. According to claim 1,
A call recommendation system in which the matching time prediction unit generates the first prediction data and the second prediction data using the RNN learning model based on a Convolution Neural Network (CNN)-Long-Short Term Memory (LSTM) algorithm. delete According to claim 1,
The matching time prediction unit receives service data about the service and mid- to long-term data about the date on which the service was requested, vectorizes the service data and the mid- to long-term data to generate vector data, and generates vector data and the first prediction. Input data and the second prediction data into the prediction learning model to predict the matching time,
The service data includes information about the departure point of the service user, the destination of the service user, the time required to perform the service, and the distance from the departure point to the destination,
The mid- to long-term data includes information on the current time, the day of the week on the date on which the service is requested, whether the date on which the service is requested is a public holiday, and the season on the date on which the service is requested. According to claim 1,
The final ranking unit determines whether the service is a service suitable for the preference of the service provider based on the preference data, and if the service is a service suitable for the preference, calculates the score of the service according to Equation 2, , A call recommendation system that determines the recommendation ranking of the service according to the score.
(Equation 2)
The calculated score = (first weight)*(the price) - (the second weight)*(the matching time)
(The first weight and the second weight are constants related to the service user's preference for the price and the matching time, respectively) According to claim 1,
The final ranking unit determines whether the service is a service suitable for the preference of the service provider based on the preference data, and if the service is a service suitable for the preference, calculates the score of the service according to Equation 3, , A call recommendation system that determines the recommendation ranking of the service according to the score.
(Equation 3)
The calculated score = ((first weight)*(the price)) / ((second weight)*(the matching time))
(The first weight and the second weight are constants related to the service user's preference for the price and the matching time, respectively) According to claim 1,
The first current data represents the average location of the service user in a first time interval,
the second current data represents the average location of the service provider in the first time interval,
The first prediction data represents the future location where the service user is predicted to be at a second time after the first time interval,
the second predicted data represents the future location where the service provider is predicted to be at the second time,
The place where the service user is actually located at the second time, the place where the service provider is actually located at the second time, the service selected by the service provider among the services, and the next service provider after the selected service is performed. A correct answer information collection unit that collects correct answer information regarding the actual matching time taken to match a service user, etc.; and
A call recommendation system further comprising an update unit configured to update the data collection unit, the matching time prediction unit, the price determination unit, and the final ranking unit based on the correct answer information. According to claim 7,
The first past data represents the average location of the service user in each of a plurality of time intervals,
The second historical data represents an average location of the service provider in each of the plurality of time intervals,
The plurality of time sections are different time sections before the first time section,
The update unit updates the data collection unit so that the number of the plurality of time sections and the time length of each of the plurality of time sections are adjusted based on the correct answer information. According to claim 1,
The matching time prediction unit generates the first prediction data and the second prediction data using the RNN learning model based on a Convolution Neural Network (CNN)-Long-Short Term Memory (LSTM) algorithm, and provides information about the service. Receive service data and mid- to long-term data regarding the date on which the service was requested, vectorize the service data and the mid- to long-term data to generate vector data, and combine the vector data, the first prediction data, and the second prediction data with the vector data. Predict the matching time by inputting it into a predictive learning model,
If the service user is not matched with a service provider using the call recommendation system after a certain time, the price determination unit adjusts the price, and the adjusted price is (first weight) * (the service user Adjusting the price so that waiting time that is not matched with the service provider using a call recommendation system) * (the price) + (the second weight) * (the matching time) ² ,
The final ranking unit determines whether the service is a service suitable for the preference of the service provider based on the preference data, and if the service is a service suitable for the preference, the score of the service is (first weight)*( Calculate the score so that it increases as the adjusted price) increases and decreases as (second weight)*(the matching time) increases, and determines the recommendation ranking of the service according to the score of the service, ,
The service data includes information about the departure point of the service user, the destination of the service user, the time required to perform the service, and the distance from the departure point to the destination,
The mid- to long-term data includes information about the current time, the day of the week on the date on which the service is requested, whether the date on which the service is requested is a public holiday, and the season on the date on which the service is requested,
The first weight and the second weight are determined according to the service user's preference for each of the adjusted price and the matching time,
The first current data represents the average location of the service user in a first time interval,
the second current data represents the average location of the service provider in the first time interval,
The first prediction data represents the future location where the service user is predicted to be at a second time after the first time interval,
the second predicted data represents the future location where the service provider is predicted to be at the second time,
The place where the service user is actually located at the second time, the place where the service provider is actually located at the second time, the service selected by the service provider among the services, and the next service provider after the selected service is performed. A correct answer information collection unit that collects correct answer information regarding the actual matching time taken to match a service user, etc.; and
A call recommendation system further comprising an update unit configured to update the data collection unit, the matching time prediction unit, the price determination unit, and the final ranking unit based on the correct answer information. In a call recommendation system implemented by a computer,
When a service is requested from a service user, first past data indicating the past location of the service user, first current data indicating the current location of the service user, and first data indicating the past location of the service provider are provided by the call recommendation system. 2 collecting historical data and second current data indicating the current location of the service provider;
By the call recommendation system, the first past data, the first current data, the second past data, and the second current data are input into a RNN (Recurrent Neural Network) learning model to determine the future location of the service user and the Predict the future location of the service provider, input first prediction data about the future location of the service user and second prediction data about the future location of the service provider into a predictive learning model so that the service provider provides the service. Predicting the matching time required from completion until the service provider is matched with the next service user;
calculating, by the call recommendation system, the price for the service so that the price for the service increases as the matching time increases; and
determining, by the call recommendation system, a recommendation ranking of the service among services requested from the service provider based on the price and preference data indicating the service provider's preference for the service; and
If the service user is not matched with a service provider using the call recommendation system after a certain period of time, adjusting the price;
The adjusted price is (the first weight) * (the waiting time when the service user is not matched with the service provider using the call recommendation system) * (the price) + (the second weight) * (the matching time) ² people
A call recommendation method in which the RNN learning model and the prediction learning model are based on a deep learning algorithm.

Images