Online sales system with order expectation calculation function

Abstract

It is possible to have a dialogue that understands the customer's intentions, and humans can easily take over the response. [Solution] An online sales system with an order expectation calculation function comprising a customer reception unit for identifying and accepting customers, a presentation implementation unit, a dialogue unit having a verbal information transmission unit for conveying verbal information to the customer, and a verbal response acquisition unit for acquiring the verbal responses of the customer, an order score calculation unit for calculating an order score using the verbal responses of the customer, a response history memory unit for storing response history with the customer, and an administrator communication unit for transmitting the response history to a system administrator, wherein the dialogue unit conducts a first dialogue and calculates a first order score with the customer, and after calculating the first order score, communicates second linguistic information to the customer that changes depending on the first order score, and conducts a second dialogue with the customer to acquire a second verbal response, which is the customer's response to the second linguistic information.

Description

[Technical field]

[0001] The present invention relates to an online sales system for use on websites related to the sale of services and goods. [Background technology]

[0002] In recent years, orders for services and products via the Internet have become more common. In addition, administrators of websites and other websites that accept orders for services and products may provide real-time customer service to customers who access the websites via the Internet, in the same way as they provide real-time customer service to customers who visit a brick-and-mortar store. Currently, methods that enable real-time customer service to customers who have obtained information from websites and other websites include a method in which trained human receptionists communicate with customers by telephone, chat, or online conference system.

[0003] On the other hand, by using large-scale language models that utilize so-called AI (artificial intelligence), it has become possible for a computer to generate words according to the situation (see Patent Document 1, etc.). By using a computer's language generation function, it may be possible for a computer to carry out some of the sales activities that are usually carried out by trained humans. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 7487986 specification Summary of the Invention [Problem to be solved by the invention]

[0005] However, there is a problem that it is difficult for a computer to carry out sales activities that take into account the intentions of customers as humans do. Also, when a human takes over a sales activity after a part of the sales activity is carried out by a computer, there is a problem that it is difficult to obtain important information such as the customer's order expectation.

[0006] The present invention has been made in consideration of such problems, and relates to an online sales system with an order expectation calculation function that is capable of understanding the intentions of customers who are the target of sales, and that allows humans to easily take over that information. [Means for solving the problem]

[0007] The online sales system with order expectation calculation function according to the present invention comprises: a customer reception unit for identifying and receiving customers who have visited a predetermined website; a presentation implementation unit for conveying information about a sales target to the customer; a linguistic information transmission unit for transmitting linguistic information to the customer, and a linguistic response acquisition unit for acquiring the linguistic response of the customer to the linguistic information transmitted to the customer, and a dialogue unit for conducting dialogue to explore at least the customer's expectation of ordering the sales object; an order score calculation unit that calculates an order score, which is an index regarding the customer's order expectation for the sales object, by using at least the verbal response of the customer; a response history storage unit that stores a response history for the customer including at least the linguistic information conveyed to the customer, the linguistic reaction of the customer to the linguistic information, and the order score; a manager communication unit that communicates at least a part of the response history to a system manager, the dialogue unit communicates first linguistic information, which is one of the linguistic information, to the customer, and conducts a first dialogue with the customer to obtain a first linguistic response, which is a reaction of the customer to the first linguistic information and is one of the linguistic responses; The order score calculation unit calculates a first order score, which is one of the order scores, in consideration of the first linguistic response after the first dialogue; After calculating the first order score, the dialogue unit conveys to the customer second linguistic information which is another of the linguistic information to be conveyed to the customer and which changes depending on the first order score, and conducts a second dialogue with the customer to obtain a second linguistic response which is the customer's reaction to the second linguistic information and is the other of the linguistic responses.

[0008] Such an online sales system with an order expectation calculation function can have a dialogue with a customer while understanding the customer's intentions in the form of an order score or the like, by having a presentation implementation unit, a dialogue unit, and an order score calculation unit. That is, the dialogue unit can change the second linguistic information to be conveyed to the customer in the second dialogue according to the order score calculated after the first dialogue, thereby making it possible to have a dialogue according to the customer's reaction. In addition, the response history including the order score is stored in the response history storage unit, so that a human manager can easily take over the sales activities performed by the online sales system while obtaining important information such as the customer's order expectation. In addition, the manager can easily deduce the grounds and reasons for the customer response performed by the online sales system by checking the order score in the response record, which makes it easy to tune the online sales system.

[0009] For example, the sales system according to the present invention includes a first machine learning unit that performs machine learning to generate the second linguistic information based on the first order score using learning data including the first order score as an explanatory variable and the second linguistic information as a target variable for a large-scale language model, The dialogue unit may, in the second dialogue, cause a large-scale language model trained by the first machine learning unit to generate the second linguistic information, and communicate the second linguistic information generated by the large-scale language model to the customer.

[0010] The dialogue unit generates the second linguistic information from a large-scale language model trained by the first machine learning unit, and thus the online sales system according to the present invention can conduct more diverse and natural dialogue closer to human dialogue than a dialogue system that prepares a multiple-choice script, for example. Such natural dialogue can improve sales effectiveness.

[0011] Further, for example, the sales system according to the present invention includes a second machine learning unit that performs machine learning of the second linguistic response and the second order score based on the first order score and the second linguistic information using learning data including the first order score and the second linguistic information as explanatory variables and having the second linguistic response and a second order score, which is another one of the order scores calculated taking the second linguistic response into consideration, as objective variables for a large-scale language model; The dialogue unit may, in the second dialogue, convey to the customer the second linguistic information that the second machine learning unit has determined to be relatively likely to increase the second order score compared to the first order score.

[0012] The dialogue unit uses the large-scale language model trained by the second machine learning unit to conduct dialogue, and the online sales system according to the present invention can search for second linguistic information that effectively increases the second order score and convey it to the customer. Such an online sales system can conduct dialogue that effectively increases the order score and leads to a deal.

[0013] In addition, for example, the online sales system according to the present invention may further include a non-verbal reaction acquisition unit that acquires a non-verbal reaction of the customer during the transmission of information by the presentation implementation unit and during the dialogue by the dialogue unit, The order score calculation unit may calculate the order score using the non-verbal response in addition to the verbal response.

[0014] Such an online sales system can calculate an order score using the customer's non-verbal responses, allowing sales activities to be conducted toward the customer while more appropriately understanding the customer's intentions.

[0015] Also, for example, the system may include a history and attribute storage unit that stores history and attribute information of the customer, The order score calculation unit may calculate the order score using the history and attribute information of the customer in addition to the verbal response.

[0016] Such an online sales system can calculate an order score using customer history and attribute information, taking into account each customer's individuality and more appropriately understanding their intentions when conducting sales activities toward them.

[0017] Also, for example, the presentation implementation unit may convey information about one or more sales objects selected from a plurality of sales objects to the customer.

[0018] The presentation implementation unit may be a single-target type that implements one presentation for one sales target, or a multiple-target type that prepares different presentations for multiple sales targets. The customer may select / decide which of the multiple sales targets to present to, or the online sales system may select / decide according to the customer. By enabling sales to multiple sales targets, such a sales system can carry out more diverse sales activities. [Brief description of the drawings]

[0019] [Figure 1] FIG. 1 is a conceptual diagram showing a communication network including an online sales system according to a first embodiment of the present invention. [Diagram 2] FIG. 2 is a functional block diagram showing a schematic structure of the online sales system shown in FIG. [Diagram 3]FIG. 3 is a conceptual diagram showing the internal structures of the customer reception unit, the presentation storage unit, and the history / attribute information storage unit shown in FIG. [Figure 4] FIG. 4 is a conceptual diagram showing the internal structure of the history / attribute information storage unit shown in FIG. [Diagram 5] FIG. 5 is a conceptual diagram showing an internal structure of the response history storage unit shown in FIG. [Figure 6] FIG. 6 is a conceptual diagram showing the internal structure of the linguistic information transfer unit and the linguistic response acquisition unit shown in FIG. [Figure 7] FIG. 7 is a conceptual diagram showing the internal structure of the non-verbal reaction acquisition unit and the order score calculation unit shown in FIG. [Figure 8] FIG. 8 is a flow chart showing an example of online sales by the online sales system shown in FIGS. [Figure 9] FIG. 9 is a conceptual diagram illustrating a first dialogue carried out in the online sales system shown in FIGS. [Figure 10] FIG. 10 is a conceptual diagram for explaining how the next linguistic information to be conveyed to a customer changes depending on the order score in an online sales system. [Figure 11] FIG. 11 is a conceptual diagram illustrating a second machine learning unit included in the online sales system. [Figure 12] FIG. 12 is a graph showing a first example of the order score calculated by the order score calculation unit. [Figure 13] FIG. 13 is a graph showing a second example of the order score calculated by the order score calculation unit. [Figure 14] FIG. 14 is a functional block diagram showing a schematic structure of an online sales system according to the second embodiment of the present invention. [Figure 15] FIG. 15 is a functional block diagram showing a schematic structure of an online sales system according to the third embodiment of the present invention. [Figure 16] FIG. 16 is a conceptual diagram showing an example of information transmitted to a system administrator by the administrator transmission unit of the online sales system shown in FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0020] First embodiment 1 is a conceptual diagram showing an online sales system 10 with an order expectation calculation function according to a first embodiment of the present invention (hereinafter, simply referred to as the "online sales system") and its peripheral devices. The online sales system 10 is composed of a server device and the like, and is communicatively connected to a plurality of customer terminals 94 via a network communication network 96 such as the Internet.

[0021] The online sales system 10 is also communicatively connected, via a network 96 such as the Internet, to an administrator terminal 92, which is a communication terminal of an administrator who manages the online sales system 10. However, the administrator who manages the online sales system 10 can also read out information about the online sales system 10 directly via a user interface that the online sales system 10 has, without going through the administrator terminal 92.

[0022] As described later with reference to FIG. 2, the online sales system 10 is composed of a server device having a customer reception unit 16, a presentation implementation unit 20, a dialogue unit 30, etc. The server device constituting the online sales system 10 has an internal configuration and functions similar to those of a general computer, such as a calculation unit, a storage unit, a display unit, an input unit, and a communication unit. In the explanation of the online sales system 10, the commonalities with a general computer will be omitted. In addition, the server device constituting the online sales system 10 may be a cloud type or an on-premise type, and is not particularly limited.

[0023] 1 is an information communication terminal that communicates with the online sales system 10 and is used by a customer to receive information about a specific product from the online sales system 10. Examples of the customer terminal 94 include communication terminals such as home personal computers, tablets, smartphones, car navigation systems, and telephones, but are not particularly limited as long as they are terminals that can communicate with the online sales system 10 via the network communication network 96. The administrator terminal 92 is similar to the customer terminal 94.

[0024] Fig. 2 is a functional block diagram showing a schematic structure of the online sales system shown in Fig. 1. As shown in Fig. 2, the online sales system 10 has a customer reception unit 16, a presentation implementation unit 20, a presentation information storage unit 14, a dialogue unit 30, an order score calculation unit 40, a history / attribute storage unit 66, a manager communication unit 56, a response history storage unit 50, etc. The online sales system 10 also has a first machine learning unit 71, a second machine learning unit 72, a non-verbal reaction acquisition unit 60, a learning data acquisition unit 76, etc.

[0025] The customer reception unit 16 shown in Fig. 2 identifies customers who access a location on the website of the online sales system 10 in real time (real-time type) via a communication network, and accepts visits to the online sales system 10. Note that in the first embodiment, an example will be described in which the customer and the online sales system 10 interact in real time. However, as in the online sales system 110 according to the second embodiment described later, the customer reception unit 116 receives data from the customer by email or the like, and the online sales system 110 also transmits data to the customer by email or the like, thereby interacting with the customer in non-real-time (time-shift type), is also included in the embodiment.

[0026] Fig. 3(a) is a conceptual diagram showing the internal structure of the customer reception unit 16 shown in Fig. 2. The customer reception unit 16 includes a customer identification unit 17, a machine learning determination unit 18, a presentation selection unit 19, and the like.

[0027] The customer identification unit 17 identifies a customer who accesses the online sales system 10 via the customer terminal 94. For example, the customer identification unit 17 identifies the customer by requesting the customer terminal 94 to input an ID and a password. The machine learning determination unit 18 can select whether or not to perform machine learning on a dialogue to be performed by the dialogue unit 30 with the identified customer. For example, the online sales system 10 can sequentially import and learn the contents of the dialogue of the customer by the dialogue unit 30 as learning data by the machine learning determination unit 18 determining that machine learning will be performed. On the other hand, the online sales system 10 can suppress the occurrence of delays in response processing in dialogue by the machine learning determination unit 18 determining that machine learning will not be performed during times of concentrated access, etc.

[0028] The presentation selection unit 19 in the customer reception unit 16 shown in Fig. 3 selects the contents of the presentation to be performed by the presentation execution unit 20 shown in Fig. 2. The presentation selection unit 19 can determine the contents of the presentation to be performed by the presentation execution unit 20 from the input results from the customer terminal 94 acquired before and after customer identification and the customer history and attribute information 66a stored in the history and attribute storage unit 66 (see Fig. 2).

[0029] 2 conveys information about products and services that are the subject of sale by the online sales system 10 to customers received by the customer reception unit 16. The presentation implementation unit 20 conveys information about the sales subject to the customer, for example, by sending video (including moving and still images) data including audio to the customer terminal 94, and the customer terminal 94 plays the video data sent from the presentation implementation unit 20. Note that the information about the sales subject conveyed to the customer by the presentation implementation unit 20 may be data only of images or videos, or data only of audio, and the format of the information is not particularly limited.

[0030] The presentation implementation unit 20 conveys information about sales objects to customers by reading out video data and the like stored in the presentation information storage unit 14 of the online sales system 10. Fig. 3(b) is a conceptual diagram explaining information stored in the presentation information storage unit 14 shown in Fig. 2. As shown in Fig. 3(b), the presentation information storage unit 14 stores information about multiple sales objects, such as product A, product B, product C, and service D, service E, and service F.

[0031] The presentation implementation unit 20 can communicate to the customer information on one or more sales objects selected by the presentation selection unit 19 of the customer reception unit 16 from among the multiple sales objects stored in the presentation information storage unit 14. The presentation (transmission of information on the sales object) stored in the presentation information storage unit 14 and implemented by the presentation implementation unit 20 is not limited to a non-interactive presentation in which information is communicated to the customer one-way. In contrast, the presentation implementation unit 20 may implement an interactive presentation in which language information is received from the customer by, for example, operating a dialogue unit 30 (described later) during the presentation.

[0032] 2 interacts with a customer who has accessed the online sales system 10 via a customer terminal 94 by exchanging words (linguistic information). The content of the interaction between the interaction unit 30 and the customer is not particularly limited, but at least includes an interaction that explores the customer's degree of expectation of ordering for the sales item for which a presentation has been given to the customer.

[0033] As shown in Fig. 2, the dialogue unit 30 has a linguistic information transmission unit 31 that transmits linguistic information (words to the customer) to the customer, and a linguistic response acquisition unit 36 ​​that acquires the customer's linguistic response to the linguistic information transmitted to the customer (words from the customer). Fig. 6 is a conceptual diagram showing an example of the functions of the linguistic information transmission unit 31 and the linguistic response acquisition unit 36. The linguistic information transmission unit 31 may transmit the linguistic information to the customer in text, for example, in a chat format, or may transmit audio information to the customer terminal 94 and have the speaker of the customer terminal 94 play back the linguistic information to the customer by voice.

[0034] Like the linguistic information transmission unit 31, the linguistic response acquisition unit 36 ​​may acquire linguistic responses from text data input via the customer terminal 94 and transmitted to the online sales system 10, or may acquire linguistic responses based on voice data input via a microphone or the like of the customer terminal 94.

[0035] As described later, the linguistic information (words to the customer) conveyed to the customer by the dialogue unit 30 may be generated by a machine (computer) using a large-scale language model, or may be selected and determined based on a script prepared in advance by an administrator, etc. Furthermore, the linguistic response (words from the customer) acquired by the dialogue unit 30 from the customer may include standard information that the customer selects from a binary option of YES / NO or multiple options.

[0036] However, from the viewpoint of enhancing the customer's experience of the online sales system 10, it is preferable that at least a part of the linguistic responses (words from the customer) acquired by the dialogue unit 30 includes information on words and sentences that the customer freely thought up, such as free character input or voice input. Specific examples of dialogue between the dialogue unit 30 and the customer will be explained later.

[0037] The order score calculation unit 40 shown in FIG. 2 uses at least the verbal reactions of the customer acquired by the dialogue unit 30 to calculate an order score, which is an index of the customer's expectation of ordering a sales item about which information has been conveyed to the customer by the presentation implementation unit 20 or the like. The order score calculation unit 40 can calculate the order score by using not only the verbal reactions of the customer acquired by the dialogue unit 30, but also the non-verbal reactions of the customer acquired by the non-verbal reaction acquisition unit 60 (see FIG. 2) during or before and after the presentation, and the history and attribute information 66a of the customer stored in the history and attribute storage unit 66 (see FIG. 2). Note that whether the order score calculation unit 40 uses the non-verbal reactions and the history and attribute information 66a of the customer when calculating the order score can be arbitrarily set by the administrator of the online sales system 10 or the like.

[0038] The order score calculated by the order score calculation unit 40 may be any index related to the customer's expectation of ordering the sales item. For example, one example of the order score is a numerical value representing the likelihood that the customer will order the sales item, using a point-adding (and point-deducting) method for each item. Another example of the order score is a numerical value representing the likelihood that the customer will place an order for a customer who has given a predetermined verbal response, based on big data related to past response records and customer interaction records stored by the online sales system 10, which is calculated by an artificial intelligence.

[0039] FIG. 7(b) is a conceptual diagram showing an example of scoring items used by the order score calculation unit 40 shown in FIG. 2 to calculate the order score. The scoring items used by the order score calculation unit 40 shown in FIG. 7(b) include current case items that evaluate the customer's reaction during access at the time of calculating the order score, and customer history items that are information on the customer acquired before the access at the time of calculating the order score. The current case items used by the order score calculation unit 40 include the customer's verbal reaction acquired by the dialogue unit 30, as well as the customer's non-verbal reaction 61 such as the viewing time of the presentation and the time the customer was away from the desk during the presentation. The customer history items used by the order score calculation unit 40 include the customer's transaction history in the online sales system 10, the customer's online site browsing history, and customer attribute information such as age, gender, and occupation.

[0040] The history and attribute storage unit 66 shown in FIG. 2 stores history and attribute information 66a (see FIG. 4) of customers who have accessed the online sales system 10. FIG. 3(c) is a conceptual diagram showing the configuration of the history and attribute storage unit 66. As shown in FIG. 3(c), the history and attribute storage unit 66 stores the history and attribute information 66a of customers who have accessed the online sales system 10, distinguishing between each customer. Furthermore, the information stored in the history and attribute storage unit 66 of the order score calculation unit 40 shown in FIG. 2 is read out as appropriate by the dialogue unit 30 and the order score calculation unit 40 shown in FIG. 2.

[0041] Fig. 4 is a conceptual diagram showing an example of history and attribute information stored in the history and attribute storage unit 66 shown in Fig. 3(c). Fig. 4 shows an example of history and attribute information 66a of customer A shown in Fig. 3(c).

[0042] The history and attribute information 66a of customer A includes a customer score indicating customer A's unique score, case information that stores information for each case (cases A to D in FIG. 4) about cases that customer A has inquired about to the online sales system 10 up to now, site browsing information for that customer, and customer-specific information such as the customer's affiliation, age, sex, and area of ​​residence.

[0043] As shown in FIG. 4, the case information stored in the history and attribute memory unit 66 for customer A includes the date the case was received, the target product or service, whether the case is in progress or completed, the final result of the case, the response history, the current order score (if in progress), etc.

[0044] 2 stores a response history for a customer who has accessed the online sales system 10. Fig. 5 is a conceptual diagram showing an example of a response history 51 stored in the response history storage unit 50 shown in Fig. 2.

[0045] As shown in Figure 5, the response history 51 stored in the response history memory unit 50 includes at least a dialogue history 51c including a linguistic information history 51ca relating to the linguistic information conveyed to the customer and a linguistic response history 51cb relating to the customer's linguistic response to the linguistic information, and an order score 51f calculated during (and after) the response with the customer.

[0046] In addition, the response history stored in the response history memory unit 50 may include the start and end times 51a of the customer's access to the online sales system 10, information on the presentation 51b shown to the customer, non-verbal response history 51d, which is a history of the customer's non-verbal responses 61 acquired by the non-verbal response acquisition unit 60 (see Figure 2), and the time progression of the customer's order score during the response (see Figures 12 and 13).

[0047] The response history storage unit 50 can store the response history of the customer in a manner that allows classification by response date or response time in the form of a daily report, etc. Also, a part or all of the response history stored in the response history storage unit 50 is stored in the history / attribute storage unit 66 as history / attribute information 66a for each customer, or is associated with the history / attribute information 66a.

[0048] 2 transmits at least a part of the response history stored in the response history storage unit 50 to the system administrator of the online sales system 10. The administrator communication unit 56 transmits the response history to the system administrator by, for example, displaying the information stored in the response history storage unit 50 on the display unit of the administrator terminal 92 shown in FIG.

[0049] 2 acquires nonverbal reactions of the customer during the transmission of information by the presentation implementation unit 20 and during dialogue by the dialogue unit 30. The nonverbal information acquired by the nonverbal reaction acquisition unit 60 is stored in the response history storage unit 50, the history and attribute storage unit 66, etc.

[0050] 7(a) is a conceptual diagram showing an example of a non-verbal reaction of a customer acquired by the non-verbal reaction acquisition unit 60. Examples of non-verbal reactions include the movement of the customer's eyes during the conversation, the tone of voice and speech rate of the customer, the viewing time of a presentation, and the access time to the online sales system 10 (time spent on the website). The non-verbal reaction acquisition unit 60 can acquire the non-verbal reaction of the customer by, for example, receiving an image from the camera of the customer terminal 94 that captures the customer himself / herself, receiving audio information from the microphone of the customer terminal 94 that captures the sound of the customer's staying space, and recording the viewing time and access time.

[0051] 2 performs machine learning on the large-scale language model to generate linguistic information based on the order score using learning data that includes the order score of a customer as an explanatory variable and has linguistic information to be communicated to a customer with that order score as a target variable. Since the order score is an index related to the customer's order expectation, the large-scale language model becomes able to generate natural and appropriate words and sentences as linguistic information to be communicated to a customer with a predetermined order score by repeatedly learning the large-scale language model using the first machine learning unit 71.

[0052] 2 uses linguistic information generated by the large-scale language model trained by the first machine learning unit 71 as linguistic information to be conveyed to the customer, thereby enabling more diverse and natural dialogue closer to human dialogue and improving sales effectiveness to the customer. For example, in the online sales system 10, the dialogue unit 30 dialogues with the customer in which linguistic information and linguistic responses are exchanged, while the order score calculation unit 40 calculates the order score of the customer in parallel.

[0053] More specifically, first, the dialogue unit 30 conveys first linguistic information 31a, which is one type of linguistic information, to the customer, and acquires a first linguistic response 36a, which is one type of linguistic response, as the customer's reaction to the first linguistic information 31a, thereby carrying out a first dialogue with the customer (see FIG. 9). Next, after the first dialogue, the order score calculation unit 40 calculates a first order score 41a, which is one type of order score, taking into account the acquired first linguistic response 36a (see FIG. 10).

[0054] Furthermore, after calculating the first order score, the dialogue unit 30 conveys to the customer second linguistic information 31c which is another piece of linguistic information to be conveyed to the customer and which changes depending on the first order score 41a, and further conducts a second dialogue with the customer to obtain a second linguistic reaction which is the customer's reaction to the second linguistic information 31c and is another piece of the linguistic reactions.

[0055] In the processing of the dialogue unit 30 and the order score calculation unit 40, the first machine learning unit 71 performs machine learning on a large-scale language model to generate second linguistic information 31c based on the first order score 41a using learning data that includes the first order score 41a as an explanatory variable and has the second linguistic information as a target variable. Then, in the second dialogue, the dialogue unit 30 can convey to the customer the second linguistic information 31c generated by the first machine learning unit 71 in the trained large-scale language model.

[0056] The second machine learning unit 72 shown in Figure 2 uses learning data that includes the customer's order score and the linguistic information to be conveyed next to the customer as explanatory variables, and has the customer's linguistic reaction when that linguistic information is conveyed and the order score calculated next taking that linguistic reaction into consideration as objective variables, to machine-train a large-scale language model to determine the customer's linguistic reaction based on the current order score and the linguistic information to be conveyed next, and the next order score taking this into consideration.

[0057] By repeatedly learning the large-scale language model by such a second machine learning unit 72, the large-scale language model can deepen its understanding of what combination of a given order score and linguistic information to be conveyed to the customer will bring about the most favorable combination of the customer's linguistic response and the next calculated order score. Therefore, the dialogue unit 30 can search for linguistic information to be conveyed to the customer next that can effectively improve the order score, using the large-scale language model trained by the second machine learning unit 72. Then, by conveying to the customer the linguistic information searched for by the large-scale language model, the dialogue unit 30 can effectively increase the possibility of receiving an order from the customer.

[0058] 11, in the processing of the dialogue unit 30 and the order score calculation unit 40 described above, the second machine learning unit 72 uses learning data including the first order score 41a and the second linguistic information 31c as explanatory variables and the second linguistic response and the second order score as objective variables to machine-learn the second linguistic response 36c and the second order score 41b based on the first order score 41a and the second linguistic information 31c in a large-scale language model. Then, in the second dialogue, the dialogue unit 30 can convey to the customer the second linguistic information 31c that can effectively improve the order score and that is searched for by the large-scale language model trained by the second machine learning unit 72.

[0059] 2 acquires learning data (teacher data) used when the first machine learning unit 71 and the second machine learning unit 72 learn a large-scale language model. For example, a system administrator or a developer of an online sales system can cause the first machine learning unit 71 and the second machine learning unit 72 to learn a large-scale language model by providing learning data to the first machine learning unit 71 and the second machine learning unit 72 via the learning data acquisition unit 76. Examples of the learning data acquired by the learning data acquisition unit 76 include hypothetical interaction scenarios with order scores that are likely to occur between a salesperson and a customer, interaction records regarding sales that have actually taken place in the past between a human salesperson and a customer to which an order score has been assigned, and interaction records with customers acquired in the past by the online sales system 10.

[0060] The online sales system 10 shown in FIG. 1 to FIG. 7 will be described in detail below with specific examples. FIG. 8 is a flow chart showing an example of online sales by the online sales system 10. In step 001 shown in FIG. 8, the customer reception unit 16 of the online sales system 10 receives a customer who visits a predetermined website for accessing the online sales system 10. More specifically, the customer reception unit 16 identifies the customer by prompting the customer who accesses via the customer terminal 94 to input an ID and a password. The customer reception unit 16 may accept the customer as an unspecified visitor without identifying the customer by an ID or the like. When the customer is accepted as a visitor, the customer reception unit 16 can identify the customer accepted as a visitor during or after the dialogue.

[0061] Next, in step S002, the customer reception unit 16 of the online sales system 10 identifies a sales object about which information is to be presented to the customer received in step S001, and a presentation (such as a video file) that conveys the information about the sales object. The sales object and presentation may be selected by the customer via the customer terminal 94, or may be selected by the customer reception unit 16.

[0062] In step S003 shown in Fig. 8, the presentation implementation unit 20 shown in Fig. 2 starts a presentation to convey information about the sales object to the customer. The presentation implementation unit 20 conveys the information about the sales object by making a presentation, for example, by displaying a video file or the like on the customer terminal 94. The sales object for which the presentation implementation unit 20 makes a presentation may be either a product or a service, and may be either a paid item or a free item.

[0063] When the presentation starts in step S003, the non-verbal reaction acquisition unit 60 shown in Fig. 2 starts acquiring non-verbal reactions of customers. The non-verbal reaction acquisition unit 60 acquires information such as the customer's eye movement, whether the customer has left the seat, and the viewing time of the presentation. In the example shown in Fig. 8, the non-verbal reaction acquisition unit 60 continues acquiring the customer's non-verbal reactions until the customer service ends (step S016).

[0064] 8, the transmission of information about the target product by the presentation implementation unit 20 is completed. The presentation by the presentation implementation unit 20 may end when the video file or the like constituting the presentation is played to the end, or may end when the playback of the video file or the like is stopped by an operation from the customer terminal 94 or the like.

[0065] As shown in FIG. 8, after the presentation is finished, the dialogue unit 30 of the online sales system 10 starts a dialogue with the customer. In step S006 shown in FIG. 8, the dialogue unit 30 starts a first dialogue with the customer. In step S007, the linguistic information transmission unit 31 transmits first linguistic information 31a (see FIG. 9) to the customer, and then in step S008, the linguistic reaction acquisition unit 36 ​​acquires a first linguistic reaction 36a, which is a reaction (reply) from the customer to the first linguistic information 31a. In this way, words are exchanged back and forth between the dialogue unit 30 and the customer, and the first dialogue ends (step S009).

[0066] Fig. 9 is a conceptual diagram showing an example of the first dialogue carried out in steps S006 to S009. As shown in Fig. 9, in the first dialogue carried out after the presentation, the linguistic information transmission unit 31 conveys to the customer words such as "Do you think the product will be useful to you?" as the first linguistic information 31a. The first linguistic information 31a may be selected by the dialogue unit 30 from words (script collection) prepared in advance, or may be generated in a large-scale language model in real time.

[0067] The customer responds to the first linguistic information 31a transmitted from the linguistic information transmission unit 31, and the first linguistic response 36a, which is the customer's response, is acquired by the linguistic response acquisition unit 36. For example, the linguistic response acquisition unit 36 ​​acquires first linguistic responses 36a such as "I need it now," "It depends on the price," "I want to compare it with competing products," and "I don't need it now."

[0068] The contents of the first dialogue by the dialogue unit 30 and the first linguistic information 31a to be conveyed to the customer are not particularly limited, but include at least contents to explore the customer's expectation of ordering the sales item for which a presentation has been made. Examples of dialogues to explore the customer's expectation of ordering include contents directly asking about the necessity of the sales item and contents indirectly exploring the expectation of ordering, such as confirming whether or not there are any questions about the presentation.

[0069] As shown in FIG. 8, in step S010, after the first dialogue ends, the order score calculation unit 40 calculates a first order score 41a (see FIGS. 10 to 13). The first order score 41a is calculated taking into consideration the first linguistic response 36a acquired in the first dialogue. In the example shown in FIG. 9, among the first linguistic responses 36a, a high first order score 41a is calculated for "needed right now," a low first order score 41a is calculated for "not needed right now," and an intermediate first order score 41a is calculated for "depends on price" and "comparison with competing products."

[0070] As shown in Fig. 8, after the first order score is calculated, the dialogue unit 30 starts the second dialogue (step S011). In step S012, the linguistic information transmission unit 31 transmits the second linguistic information 31c (see Fig. 9) to the customer, and then in step S013, the linguistic response acquisition unit 36 ​​acquires the second linguistic response 36c, which is the customer's response (reply) to the second linguistic information 31c. In this way, similar to the first dialogue, the dialogue unit 30 and the customer exchange words back and forth, and the second dialogue ends (step S014).

[0071] Here, the second linguistic information 31c conveyed to the customer by the linguistic information conveyance unit 31 of the dialogue unit 30 changes according to the first order score 41a, which is the previously calculated order score. Fig. 10 is a conceptual diagram showing the relationship between the first order score 41a and the second linguistic information 31c conveyed to the linguistic information conveyance unit 31 in the second dialogue.

[0072] As shown in FIG. 10, when the first order score 41a is high and the customer's order expectation is high, the linguistic information transmission unit 31 of the dialogue unit 30 can transmit to the customer, as the second linguistic information 31c, a content proposing the submission of a quote, such as "Shall I provide you with a quote?". On the other hand, when the first order score 41a is low and the customer's order expectation is low, the linguistic information transmission unit 31 of the dialogue unit 30 can transmit to the customer, as the second linguistic information 31c, a content confirming whether other product information is required, such as "Are there any other products that interest you?". In addition, when the first order score 41a is an intermediate value, the linguistic information transmission unit 31 of the dialogue unit 30 can transmit to the customer, as the second linguistic information 31c, a content such as a notification of a limited-time campaign or a confirmation of whether additional product information is required.

[0073] As shown in FIG. 8, in step S015, after the second dialogue is completed, the order score calculation unit 40 calculates the second order score 41b (see FIGS. 10 to 13). The second order score 42b is calculated in consideration of the second linguistic response 36c (see FIG. 12, etc.) acquired in the second dialogue. In the example shown in FIG. 10, when the second linguistic response 36c to the second linguistic information 31c is a content approving the quotation submission proposal, the order score calculation unit 40 calculates a high second order score 41b. On the other hand, when the second linguistic response 36c to the second linguistic information 31c is a no-response or ambiguous response to the confirmation of the necessity of information on other products, the order score calculation unit 40 calculates a low second order score 41b.

[0074] As shown in Fig. 8, after the calculation of the second order score, the dialogue unit 30 repeats dialogue with the customer and calculation of the order score as necessary. Finally, in step S016, the dialogue with the customer by the dialogue unit 30 ends, and the online sales system 10 ends the customer service accepted in step S001. Note that the response history including information such as the content of the dialogue between the customer and the dialogue unit 30 described using the flowchart in Fig. 8, the calculated order score, the content of the presentation, and the response time is stored in the response history storage unit 50 shown in Fig. 2.

[0075] 10, the second linguistic information 31c to be conveyed to the customer in the second dialogue changes according to the first order score 41a, which is the order score calculated immediately before. The dialogue unit 30 may prepare a plurality of pieces of second linguistic information 31c that change in response to the first order score 41a, and change the second linguistic information 31c by selecting, from the plurality of pieces of second linguistic information 31c, the information that corresponds to the first order score 41a.

[0076] Also, differently from this, in the second dialogue, the dialogue unit 30 can cause the large-scale language model trained by the first machine learning unit 71 to generate the second linguistic information 31c, and can convey the second linguistic information 31c generated by the large-scale language model to the customer (see FIG. 10). In this case, the first machine learning unit 71 trains the large-scale language model to generate the second linguistic information 31c based on the first order score 41a by machine learning using learning data including the first order score 41a as an explanatory variable and the second linguistic information 31c as a target variable. The online sales system 10 using such a first machine learning unit 71 and large-scale language model can carry out more diverse and natural dialogue closer to human dialogue than a dialogue unit using a multiple-choice script, and can improve sales effectiveness through natural dialogue.

[0077] Furthermore, in the second dialogue, the dialogue unit 30 can also convey to the customer the second linguistic information 31c searched by the large-scale language model trained by the second machine learning unit 72 as shown in Fig. 11. The second machine learning unit 72 trains the large-scale language model by machine learning the second linguistic response 36c and the second order score 41b based on the first order score 41a and the second linguistic information 31c using learning data including the first order score 41a and the second linguistic information 31c as explanatory variables and having the second linguistic response 36c and the second order score 41b calculated taking into account the second linguistic response 36c as objective variables.

[0078] By repeatedly learning the large-scale language model by such a second machine learning unit 72, the large-scale language model can learn what combination of a predetermined order score and linguistic information to be conveyed to the customer can effectively improve the next calculated order score. In the second dialogue, the dialogue unit 30 causes the large-scale language model trained by the second machine learning unit 72 to search for second linguistic information 31c that is relatively likely to increase the second order score 41b relative to the first order score 41a, and conveys the second linguistic information 31c to the customer, thereby improving sales effectiveness.

[0079] FIG. 12 is a graph showing an example of the time transition of the order score during customer service in the online sales system 10. The order score calculation unit 40 shown in FIG. 2 calculates a pre-interaction order score 43 (one of the order scores) after the presentation ends, and then starts a first dialogue. Furthermore, the order score calculation unit 40 acquires the first linguistic response 36a and calculates a first order score 41a taking this into consideration, and the dialogue unit 30 conveys to the customer second linguistic information 31c according to the first order score 41a, thereby starting a second dialogue. Furthermore, the order score calculation unit 40 calculates a second order score 41b after the second dialogue, and continues the dialogue.

[0080] Such an online sales system 10 with an order expectation calculation function can have a dialogue with a customer while understanding the customer's intention in the form of an order score or the like, as shown in FIG. 12. That is, by changing the linguistic information to be conveyed to the customer in the next dialogue according to the order score calculated by the dialogue unit 30 after a predetermined dialogue, it becomes possible to have a dialogue according to the customer's reaction. In addition, by storing the response history 51 including the order score in the response history storage unit 50, a human manager can easily take over the sales activities performed by the online sales system while obtaining important information such as the customer's order expectation. In addition, by checking the order score in the response history 51 via the manager transmission unit 256, the manager can easily estimate the grounds and reasons for the customer response performed by the online sales system 10, which makes it easy to tune the online sales system 10.

[0081] In addition, the online sales system 10 can have more diverse and natural dialogues closer to human dialogues by having the dialogue unit 30 generate second linguistic information 31c in a large-scale language model trained by the first machine learning unit 71 as shown in FIG. 10, thereby improving sales effectiveness. In addition, in general, when using a large-scale language model, if the explanatory variables and the objective variables are mainly linguistic information, it may be difficult to analyze the learning progress and problems in the generated language. However, in the online sales system 10, in which the large-scale language model is trained by the first machine learning unit 71 including the order score as an explanatory variable, it is easy to grasp problems in the large-scale language model and to tune it.

[0082] Moreover, in the online sales system 10, the dialogue unit 30 causes a large-scale language model trained by a second machine learning unit as shown in Fig. 11 to search for second linguistic information 31c that maximizes the next calculated order score, thereby raising the order score during customer interaction as shown in Fig. 12, and enabling a dialogue leading to a deal. Moreover, as shown in Fig. 13, the online sales system 10 can calmly and stably deal with customers even when dealing with customers whose order score cannot be raised. Therefore, the online sales system 10 can reduce the burden on salesmen who currently deal with such customers.

[0083] The online sales system 10 may use both the first machine learning unit 71 and the second machine learning unit 72, or may use either one of them. The online sales system 10 may be able to arbitrarily select whether to operate or stop the operation of the first machine learning unit 71 and the second machine learning unit 72 so that an administrator of the online sales system 10 can reduce the load on the server or collect data on a large-scale language model in a specific learning state.

[0084] Second embodiment 14 is a functional block diagram showing a schematic configuration of an online sales system 110 with an order expectation calculation function according to a second embodiment of the present invention. The online sales system 110 according to the second embodiment is similar to the online sales system 10 according to the first embodiment, except that the customer reception unit 116, the presentation implementation unit 120, the dialogue unit 130, and the non-verbal reaction acquisition unit 160 are partially different. In the explanation of the online sales system 110, only the differences with the online system 10 will be explained, and explanations of the commonalities will be omitted.

[0085] In the online sales system 110 shown in FIG. 14, the customer reception unit 116 receives data from the customer by e-mail or the like, and the online sales system 110 also transmits data to the customer by e-mail or the like, thereby interacting with the customer in a non-real-time manner. However, in customer interaction by the online sales system 110, real-time interaction and time-shift interaction may be mixed, such as shifting to real-time communication as described in the online sales system 10 midway. Furthermore, in customer interaction by the online sales system 110, the customer may be able to select an interaction method such as a text-based interaction or a voice-based interaction, and the online sales system 110 side may also selectively use a text-based interaction or a voice-based interaction. Furthermore, the online sales system 110 can convey facial expressions (smile, serious expression, etc.) and gestures (bow, gestures) to the customer using an avatar or the like in association with the interaction with the customer. In addition, the online sales system 110 itself may be capable of switching between computer-based interaction and sales staff interaction (human interaction) so that the sales staff can take over customer interaction as necessary.

[0086] The customer reception unit 116 receives requests for product explanations and questions about services from customers via electronic information transmission means such as e-mail. In this case, the customer identification unit 119 identifies the customer by the e-mail address, the title of the e-mail, and text information written in the body of the e-mail. The customer identification unit 119 also has an automatic customer determination function that determines how to handle the customer based on the received information.

[0087] For example, when the customer identification unit 119 determines from information such as an email address that the inquiry is from a valid customer, it responds to the customer with an email using text generated by the dialogue unit 130 and dialogues with the customer. On the other hand, when the customer identification unit 119 determines from information such as an email address that invalid information such as so-called spam has been received, it does not respond to the customer. Furthermore, the customer reception unit 116 can identify, from the contents of the email body, that the customer wishes to be handled by a sales staff member over the phone. When the customer reception unit 116 recognizes information that the customer wishes to be handled by a human, the online sales system 110 can convey this to the system administrator via the administrator communication unit 56.

[0088] The presentation implementation unit 120 shown in FIG. 14 performs a presentation to a customer by attaching electronic presentation materials or video playback files to a reply mail, or by attaching a link to a website where the presentation materials can be played to a reply mail. The dialogue unit 130 can dialogue with a customer by exchanging linguistic information with the customer through email, an inquiry form, chat, or the like. Unlike the real-time type, the non-verbal reaction acquisition unit 160 cannot acquire information such as the customer's attitude while watching a presentation using a camera or the like, but can acquire other information in the same way as the non-verbal reaction acquisition unit 60 shown in FIG. 7. The online sales system 110 can calculate and use the order score in the same way as the online sales system 10 of the first embodiment.

[0089] In addition, the online sales system 110 according to the second embodiment has the same effects as the online sales system 10 according to the first embodiment.

[0090] Third embodiment 15 is a functional block diagram showing a schematic configuration of an online sales system 210 with an order expectation degree calculation function according to a third embodiment of the present invention. The online sales system 210 according to the third embodiment is similar to the online sales systems 10, 110 according to the first or second embodiment, except that it has a customer approach unit 205 and has the additional function of a manager communication unit 256. Explanation of commonalities will be omitted.

[0091] The customer approach unit 205 transmits information that motivates potential customers so as to increase the number of accesses by customers to the customer reception unit 10 and to increase inquiries by customers to the customer reception unit 116. For example, the customer approach unit 205 can use the information stored in the response history storage unit 50 to transmit information (advertisements and promotions) about products and services in which the customer is presumed to be interested to a specific customer.

[0092] In addition, the customer approach unit 205 calculates the correlation between the attributes of the customers stored in the history and attribute information storage unit 66 and the customer's response records stored in the response history storage unit 50, recognizes potential customers who are highly interested in a particular product or service, and can send advertisements targeting those potential customers. Examples of advertisements sent by the customer approach unit 205 include, but are not limited to, SNS advertisements, listing advertisements, banner displays, and affiliate advertisements.

[0093] The customer approach unit 205 can also send direct mail to specific potential customers and conduct telephone sales. When generating text for direct mail to specific potential customers or when generating words to convey to potential customers in telephone sales, the customer approach unit 205 can do so in conjunction with the dialogue unit 130. Furthermore, the customer approach unit 205 can also send emails, phone calls, etc. to customers as customer follow-ups regarding cases that have already been proposed and are being implemented.

[0094] The manager communication unit 256 shown in Fig. 15 can communicate to the system manager not only the response records for each customer as shown in Fig. 12 and Fig. 13, but also the summary information and analysis information of sales activities related to a specific product / service or a group of products / services. Fig. 16 is a conceptual diagram showing an example of sales information of the online sales system 210 that the manager communication unit 256 communicates to the system manager.

[0095] As shown in FIG. 16, the administrator communication unit 256, for example, displays a graph on the administrator terminal 92 shown in FIG. 1, which is a compilation of sales statuses for each month for a specific product sold by the online sales system 210. In FIG. 16, the sales statuses for each month for a specific product are shown as a bar graph indicating the cumulative number of customers belonging to three stages, namely, "approaching," "proposal being implemented," and "order received." The administrator communication unit 256, for example, counts the number of customers to whom the customer approach unit 205 has sent out advertisements as the number of "approaching" customers, the number of customers to whom the customer reception unit 116 has received information such as emails from customers as the number of "proposal being implemented" customers, and the number of customers to whom the dialogue unit 130 has received information indicating that they have received orders as the number of "order received" customers, and counts the number of customers.

[0096] In addition, the manager communication unit 256 can display sales information in various formats, such as displaying the percentage of order scores for customers with "proposal in progress" and displaying in the form of a customer list which sales stage each individual customer belongs to. A system manager who uses such an online sales system 10 can easily recognize issues and areas for improvement in the sales process. Furthermore, the system manager can carry out effective sales activities by solving the recognized issues and making improvements.

[0097] The administrator communication unit 256 may also have an automatic management function of analyzing data from the dialogue unit 130, the presentation storage unit 14, the customer approach unit 205, the response history storage unit 50, etc., and instructing and managing each unit on actions to optimize customer service provided by the online sales system 210. Examples of the automatic management function performed by the administrator communication unit 256 include analyzing the contents of the response history storage unit 50 to optimize an advertising strategy provided by the customer approach unit 205, improving the contents of the presentation stored in the presentation information storage unit 14, and optimizing a policy for customer follow-up (automatic follow-up) provided by the customer approach unit 205.

[0098] In addition, the online sales system 210 according to the third embodiment has the same effects as the online sales systems 10 and 110 according to the first and second embodiments. [Explanation of symbols]

[0099] 10. Online sales system 14…Presentation information storage section 16…Customer Reception Department 17…Customer Identification Section 18…Machine learning judgment section 19…Presentation Selection Section 20…Presentation Department 30…Dialogue section 31…Linguistic Information Transmission Section 31a…First linguistic information 31c…Second linguistic information 36…Verbal response acquisition section 36a…First verbal response 36c…Second verbal response 40…Order score calculation section 41a…First order score 41b…Second Order Score 43…Pre-interaction order score 50…Response history memory section 51...Response history 51a…Access start time and access end time 51b…Displayed presentation 51c…Dialogue history (language information) 51d…Non-verbal response history 51f…Order score 56…Administrator Communication Department 71…First Machine Learning Division 72…Second Machine Learning Division 76…Learning data acquisition section 60…Non-verbal response acquisition section 61...Non-verbal reactions 66…History and attribute memory section 66a…History and attribute information 92...Administrator terminal 94…Customer terminal 96…Network communication network

Claims

1. a customer reception unit for identifying and receiving a customer; a presentation implementation unit for conveying information about a sales target to the customer; a linguistic information transmission unit for transmitting linguistic information to the customer, and a linguistic response acquisition unit for acquiring the linguistic response of the customer to the linguistic information transmitted to the customer, and a dialogue unit for conducting dialogue to explore at least the customer's expectation of ordering the sales object; an order score calculation unit that calculates an order score, which is an index regarding the customer's order expectation for the sales object, by using at least the verbal response of the customer; a response history storage unit that stores a response history for the customer including at least the linguistic information conveyed to the customer, the linguistic reaction of the customer to the linguistic information, and the order score; a manager communication unit that communicates at least a part of the response history to a system manager, the dialogue unit communicates first linguistic information, which is one of the linguistic information, to the customer, and conducts a first dialogue with the customer to obtain a first linguistic response, which is a reaction of the customer to the first linguistic information and is one of the linguistic responses; The order score calculation unit calculates a first order score, which is one of the order scores, in consideration of the first linguistic response after the first dialogue, The online sales system with order expectation calculation function, wherein after calculating the first order score, the dialogue unit conveys to the customer second linguistic information which is another of the linguistic information to be conveyed to the customer and which changes depending on the first order score, and conducts a second dialogue with the customer to obtain a second linguistic response which is the customer's reaction to the second linguistic information and is the other of the linguistic responses.

2. a first machine learning unit that performs machine learning to generate the second linguistic information based on the first order score using learning data including the first order score as an explanatory variable and the second linguistic information as a target variable for a large-scale language model; The online sales system of claim 1, wherein the dialogue unit, in the second dialogue, causes a large-scale language model trained by the first machine learning unit to generate the second linguistic information, and conveys the second linguistic information generated by the large-scale language model to the customer.

3. a second machine learning unit that performs machine learning of the second linguistic response and the second order score based on the first order score and the second linguistic information using learning data including the first order score and the second linguistic information as explanatory variables and having the second linguistic response and a second order score, which is another one of the order scores calculated taking the second linguistic response into consideration, as an objective variable for a large-scale language model; The online sales system of claim 1, wherein the dialogue unit, in the second dialogue, causes the large-scale language model trained by the second machine learning unit to search for the second linguistic information that is relatively likely to increase the second order score relative to the first order score, and communicates the second linguistic information to the customer.

4. a non-verbal reaction acquisition unit that acquires a non-verbal reaction of the customer during the transmission of information by the presentation implementation unit and during the dialogue by the dialogue unit; The online sales system according to claim 1 , wherein the order score calculation unit calculates the order score using the non-verbal response in addition to the verbal response.

5. a history / attribute storage unit for storing history / attribute information of the customer; The online sales system according to claim 1 , wherein the order score calculation unit calculates the order score using the history and attribute information of the customer in addition to the verbal response.

6. 2. The online sales system according to claim 1, wherein the presentation implementation unit conveys information on one or more sales objects selected from a plurality of sales objects to the customer.

Images