Information processing device and program

Abstract

To generate user's persona information.SOLUTION: A processor of an information processing device 1 executes a program to function as: means 31 for receiving utterance data of a real user from a user terminal; means 22 for executing natural language analysis processing to the utterance of the real user; means 27 for generating persona information related to the real user on the basis of a part of speech decomposed by the natural language analysis processing and syntax analysis results; means 30 for transmitting a generation request for an utterance of a virtual user together with history data of respective utterances of the real user and the virtual user, and the persona information to an utterance sentence generation device; means 31 for receiving the utterance data of the virtual user from the generation sentence generation device; and means 32 for transmitting the received utterance data of the virtual user to the user terminal.SELECTED DRAWING: Figure 3

Description

Technical Field

[0001] The present disclosure relates to an information processing apparatus and a program.

Background Art

[0002] Currently, EC sites (online sales) such as apparel have become widespread, and it is expected that in the future, there will be an increasing number of opportunities to open virtual stores in the shopping malls of virtual spaces (metaverses) on the Internet. In such a situation, in EC sites as well as in physical stores, it is important from the perspective of sales promotion to provide one-on-one customer service in which staff introduce recommended products through conversations with customers, dig out customer needs and materialize products that users desire, and respond to various inquiries.

[0003] Generally, the number of users visiting EC sites is much larger than the number of customers visiting physical stores. Therefore, it is not realistic for actual staff to serve each user visiting the EC site individually. Thus, it is assumed that virtual staff (referred to as virtual users or digital staff) are constructed on a computer, and instead of actual staff, digital staff serve customers (referred to as actual users) one-on-one.

[0004] The use of generative AI is expected for creating the speech content for digital staff to have conversations with customers via chat or the like. Along with the generation request, the digital staff sends the history of the conversation between the digital staff and the customer to the generative AI. The generative AI analyzes the conversation history and creates the speech sentence that the digital staff should say next. Since it is used in EC sites, the ultimate goal is sales, and for that purpose, high conversation skills and customer service capabilities like those of excellent actual staff are required. Excellent actual staff are good at having conversations to build an intimate relationship while gaining empathy and trust from customers, whether consciously or unconsciously.

[0005] However, it was not possible to break away from mechanical conversations by generating utterances based solely on the conversation history between the input digital staff and the customer.

Summary of the Invention

Problems to be Solved by the Invention

[0006] The purpose is to generate user persona information.

Means for Solving the Problems

[0007] The information processing apparatus according to the present embodiment includes a memory for storing a program, a processor for executing the program, and a communication interface for communicating with a user terminal and an utterance generation device (generative AI) via an electrical public communication network, and a virtual user conducts a conversation with an actual user. By executing the program, the processor functions as means for receiving data of an actual user's utterance from the user terminal, means for executing natural language analysis processing on the actual user's utterance, means for generating persona information regarding the actual user based on the part-of-speech and syntactic analysis results decomposed by the natural language analysis processing, means for transmitting a generation request for the virtual user's utterance to the utterance generation device together with the data of the conversation history of the utterances of the actual user and the virtual user respectively and the persona information, means for receiving data of the virtual user's utterance from the utterance generation device, and means for transmitting the received data of the virtual user's utterance to the user terminal.

Brief Description of the Drawings

[0008]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

MODE FOR CARRYING OUT THE INVENTION

[0009] Hereinafter, an information processing apparatus according to an embodiment of the present invention will be described with reference to the drawings. The information processing apparatus according to the present embodiment is applied to a scene where a virtual user (virtual user) constructed on a computer has a conversation with an actual user. Here, a scene where a customer who has visited an EC site has a conversation with a virtual staff (digital staff) constructed on a computer in a chat room prepared on the EC site will be described as an example. In this example, the customer corresponds to an actual user, and the digital staff corresponds to a virtual user. Of course, the present invention is not limited to a scene where the actual user corresponds to the customer and the virtual user corresponds to the digital staff.

[0010] As shown in FIG. 1, the information processing apparatus 1 according to the present embodiment is connected to a speech generation apparatus 2 that functions as a generation AI, an EC server 3, and a user terminal 4 via an electrical public communication network 5 such as the Internet. The user terminal 4 is connected to the EC server 3, and a customer (actual user) visits the EC site operated by the EC server 3. A chat room or the like is prepared on the EC site, and the customer purchases a product while conversing with a digital staff (virtual user). The information processing apparatus 1 directly receives the speech data of the customer input from the user terminal 4 from the user terminal 4 or indirectly via the EC server 3, creates a speech of the digital staff in response to the customer's speech, and directly transmits it to the user terminal 4 or indirectly to the user terminal 4 via the EC server 3. By repeating the above process, a conversation between the customer and the digital staff progresses on the chat room. The information processing apparatus 1 transmits a speech generation request to the speech generation apparatus 2 together with the ordered customer's speech and the digital staff's speech, that is, the history of the conversation between the customer and the digital staff.

[0011] In the present embodiment, based on the customer's speech, it is determined whether the customer is positive or negative about the digital staff's speech, and based on the determination result, a sympathy score is calculated to estimate the degree to which the customer shows sympathy for the digital staff's speech, and based on the determination result, a reliability score is calculated to estimate the degree to which the customer shows a sense of trust in the digital staff's speech. Whether to apply the determination result of positive or negative regarding the customer's speech to either the sympathy score or the reliability score is selected according to whether the digital staff's speech follows the theme of the conversation. For example, if the EC site is a clothing store, clothing is applied as the theme of the conversation.

[0012] When it is determined that the customer is positive about the statement of the digital staff, the reliability score or the empathy score is increased, and when it is determined that the customer is negative, the reliability score or the empathy score is decreased. Of course, as the reliability score, the empathy score, and the intimacy score (to be described later) increase, it means that the customer's reliability, empathy, and intimacy with respect to the digital staff increase.

[0013] By repeating this process, it is possible to obtain how the empathy score and the reliability score change as the conversation progresses, and further to identify what kind of statement content of the digital staff caused the change.

[0014] Furthermore, in an actual human relationship, as the intimacy increases, the politeness level of the language use gradually decreases, and there is a tendency to gradually change from honorific language to friendly language. It is considered that the intimacy can be deepened by decreasing the politeness level of the language use and changing from honorific language to friendly language. In other words, if the customer uses friendly language while the digital staff continues to use honorific language, the customer may lose the sense of intimacy.

[0015] Therefore, in this embodiment, the politeness level of the statement of the digital staff generated by the statement generation device 2 is changed according to the situation so as to follow the change in the politeness level of the words spoken by the customer. The politeness level of the words related to the customer's statement is specified, and based on the change in the politeness level of the words related to the specified customer's statement, the politeness level of the digital staff's words is determined. When the politeness level of the words related to the customer's statement decreases, the politeness level of the words related to the digital staff's statement is also decreased by one level from the current politeness level accordingly. When the politeness level of the words related to the customer's statement increases, the politeness level of the words related to the digital staff's statement is also increased by one level from the current politeness level accordingly. In this way, in this embodiment, by changing the politeness level of the digital staff's words according to the politeness level of the customer's words, it is possible to increase the customer's sense of intimacy while suppressing the decrease.

[0016] Also, in the present embodiment, a customer intimacy score for the digital staff is calculated based on a change in the politeness level of the customer's speech. When the politeness level of the customer's speech decreases, the customer intimacy score for the digital staff is increased. When the politeness level of the customer's speech increases, the customer intimacy score for the digital staff is decreased.

[0017] When transmitting a speech generation request to the speech generation device 2, in addition to the conversation history between the customer and the digital staff, by transmitting the transitions of the empathy score, the trust score, and the intimacy score respectively, the speech of the digital staff can be generated so as to follow the flow of the conversation and increase the empathy score and the trust score without decreasing them.

[0018] As shown in FIG. 2, the information processing device 1 as a conversation device has a RAM 12, a ROM 13, a storage unit 14, an input device 15, a display 16, and a communication interface 17 connected to a processor 11 via a system bus 10. The processor 11 is composed of, for example, a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit). The processor 11 executes a conversation program loaded from the storage unit 14 and the ROM 13 into the RAM 12 to execute conversation processing between the customer and the digital staff.

[0019] The RAM 12 functions as a main memory, a work area, etc. of the processor 11. The ROM 13 or the storage unit 14 stores a BIOS (Basic Input Output System) executed by the processor 11, an operating system program (OS), a conversation processing program according to the present embodiment, programs for realizing various other functions, and various data required for those processes.

[0020] The input device 15 consists of a keyboard (KB), a pointing device such as a mouse or a touch panel, etc. The display 19 is typically realized by an LCD (Liquid Crystal Display). In addition to the conversation processing program, the storage unit 14 stores data related to the conversation history between the customer and the digital staff, the persona information of the customer exemplified in FIG. 9, and persona extension information, etc.

[0021] As shown in FIG. 3, by executing the conversation processing program, the processor 11 functions as a control unit 21, a natural language analysis processing unit 22, a sympathy degree estimation processing unit 23, a reliability estimation processing unit 24, an intimacy degree estimation processing unit 25, a purchase possibility estimation processing unit 26, a persona information update unit 27, a persona extension information update unit 28, a diction conversion processing unit 29, a speech generation request processing unit 30, a reception processing unit 31, and a transmission processing unit 32.

[0022] The data of the customer's speech input from the user terminal 4 is received by the reception processing unit 31 via the EC server 3. Note that the data of the customer's speech input from the user terminal 4 may be directly received by the reception processing unit 31 from the user terminal 4 without passing through the EC server 3.

[0023] The natural language analysis processing unit 22 performs natural language analysis processing on the data of the customer's speech received via the reception processing unit 31, decomposes it into parts of speech, extracts in particular the string of nouns, and performs syntactic analysis. Also, the natural language analysis processing unit 22 performs natural language analysis processing on the speech of the digital staff transmitted to the EC server 3 via the transmission processing unit 32, decomposes it into parts of speech, extracts the string of nouns, and performs syntactic analysis. The parts of speech decomposed, the nouns extracted, and the syntactic analysis results by the natural language analysis processing unit 22 are used in the respective processes of the sympathy degree estimation processing unit 23, the reliability estimation processing unit 24, the persona information update unit 27, and the persona extension information update unit 28.

[0024] When the content of the digital staff's speech does not conform to the theme of the conversation, the empathy degree estimation processing unit 23 executes the calculation of the empathy degree score. When the content of the digital staff's speech conforms to the theme of the conversation, the empathy degree score is not calculated. The theme of the conversation is, for example, if the EC site is a clothing store, the clothing field is applied. Specifically, if the noun included in the digital staff's speech is related to clothing, it is determined that the content of the digital staff's speech conforms to the theme of the conversation. If the noun included in the digital staff's speech is not related to clothing and is irrelevant to clothing, it is determined that the content of the digital staff's speech does not conform to the theme of the conversation. This determination process may create an apparel-related term dictionary in advance, store it in the storage unit 14, and check whether the noun included in the digital staff's speech is listed in the dictionary, or it may be estimated using a learned model that has been machine-learned with the appropriateness of the noun and the conversation theme as teacher data.

[0025] When the customer is positive about the digital staff's speech, in order to increase the empathy degree score, a predetermined value is added to the current value of the empathy degree score. When the customer is negative about the digital staff's speech, in order to decrease the empathy degree score, a predetermined value is subtracted from the current value of the empathy degree score. Typically, the addition value and the subtraction value are the same value, but the absolute value of the addition value may be higher or lower than the absolute value of the subtraction value. The determination process related to identifying whether the customer's speech is positive or negative about the digital staff's speech may create a dictionary that covers a plurality of positive expressions and a plurality of negative expressions in advance, store it in the storage unit 14, and determine whether the customer's speech includes any positive expression or negative expression, or it may be estimated using a learned model that has been machine-learned with a plurality of expressions and the possibility of being positive or negative as teacher data.

[0026] When the content of the digital staff's utterance is in line with the conversation topic, the reliability estimation processing unit 24 executes the calculation of the reliability score; when the content of the digital staff's utterance is not in line with the conversation topic, the reliability score calculation is not executed. When the customer makes an affirmative utterance in response to the digital staff's utterance, a predetermined value is added to the current value of the reliability score to increase the reliability score. When the customer makes a negative utterance in response to the digital staff's utterance, a predetermined value is subtracted from the current value of the reliability score to decrease the reliability score. Typically, the addition value and the subtraction value are the same value, but the absolute value of the addition value may be higher or lower than the absolute value of the subtraction value.

[0027] The intimacy estimation processing unit 25 compares the word usage of the customer's utterance with the word usage of the digital staff's utterance, and calculates an intimacy score representing the degree of the customer's intimacy with the digital staff based on the comparison result. Specifically, when the politeness level of the word usage in the customer's utterance decreases, a predetermined value is added to the current value of the intimacy score to increase the intimacy score. When the politeness level of the word usage in the customer's utterance increases, a predetermined value is subtracted from the current value of the intimacy score to decrease the intimacy score. Typically, the addition value and the subtraction value are the same value, but the absolute value of the addition value may be higher or lower than the absolute value of the subtraction value.

[0028] Typically, for the determination of the politeness level of word usage, a dictionary regarding a plurality of inflectional expressions corresponding to each of a plurality of levels with different politeness levels from honorific language to friendly language is created in advance and stored in the storage unit 14, and the inflectional expression of the customer's utterance is collated with the dictionary to identify the politeness level of the corresponding inflectional expression.

[0029] It is assumed that when the empathy score, reliability score, and intimacy score increase, the likelihood of a customer purchasing a product increases. Therefore, the purchase probability estimation unit 26 calculates a purchase probability score (comprehensive score) representing the likelihood of a customer purchasing a product based on the calculated empathy score, reliability score, and intimacy score. For example, the purchase probability score is calculated by adding to the current value of the purchase probability score the change amount of each of the empathy score, reliability score, and intimacy score, or a value obtained by multiplying the change amount by a predetermined coefficient such as 0.5. Note that the purchase probability score may be calculated by simply adding the empathy score, reliability score, and intimacy score, or by weighted adding the empathy score, reliability score, and intimacy score. The calculation method can be arbitrarily changed.

[0030] The persona information update unit 27 generates new persona information regarding the place of origin, hobbies, etc. based on the part-of-speech and syntactic analysis results decomposed by the natural language analysis processing unit 22, and updates the existing persona information regarding the customer by adding it. The persona information is initially the user information registered on the EC site. As the conversation progresses, new persona information is generated and added. The persona information is stored in the storage unit 14.

[0031] The persona extended information update unit 28 generates persona extended information associating nouns with the content of the conversation of the digital staff that is considered to have induced the change in the empathy score and the reliability score. That is, persona extended information is generated as information that can identify whether the nouns in the digital staff's utterance lead to an increase or decrease in the empathy score and the reliability score. For example, as persona extended information related to a certain customer, an increase in the reliability score is associated with "blue" and "jacket" and generated. The persona extended information is appropriately generated as the conversation progresses and accumulated in the storage unit 14.

[0032] The diction conversion processing unit 29 converts the diction of the digital staff's uttered sentence received from the uttered sentence generation device 2 into the diction of the set politeness level. The digital staff's uttered sentences received from the uttered sentence generation device 2 are initially often mechanical, business-like, and dull. As they are, the intimacy level is low. Therefore, a dictionary for converting into diction of multiple levels with different politeness levels from honorific language to friendly language is created in advance and stored in the storage unit 14. The diction conversion processing unit 29 converts the words and phrases of the digital staff's uttered sentence received from the uttered sentence generation device 2 into the set diction using the dictionary.

[0033] Although details will be described later, when the politeness level of the customer's diction is specified and the politeness level changes during the progress of the conversation, the politeness level of the diction used by the digital staff is changed following the change. For example, when the politeness level of the customer's diction decreases, that is, when the intimacy of the customer with respect to the digital staff increases, the politeness level of the diction regarding the digital staff's utterance is decreased by one level from the current politeness level. When the politeness level of the customer's diction increases, that is, when the intimacy of the customer with respect to the digital staff decreases, the politeness level of the diction regarding the digital staff's utterance is increased by one level from the current politeness level. By changing the politeness level of the diction used by the digital staff following the change in the politeness level of the customer's diction in this way, the intimacy of the customer with respect to the digital staff can be increased, and a decrease in the intimacy of the customer with respect to the digital staff can be suppressed.

[0034] The utterance generation request processing unit 30 transmits data on the conversation history between the customer and the digital staff including the customer's previous utterance to the utterance generation device 2 together with a request to generate an utterance for the digital staff to respond to the customer's utterance. Further, in addition to the conversation history data, the utterance generation request processing unit 30 transmits persona information, persona extension information, changes in the empathy score, changes in the reliability score, changes in the intimacy score, and changes in the purchase probability score to the utterance generation device 2. The utterance generation device 2 can generate an utterance for the digital staff so as to follow the flow of the conversation and increase the empathy score and the reliability score without decreasing them by referring to these pieces of information.

[0035] Figure 4 shows the flow of the conversation processing of the information processing apparatus 1. Along with the start of the conversation, the empathy estimation processing unit 23, the reliability estimation processing unit 24, the intimacy estimation processing unit 25, and the purchase probability estimation processing unit 26 initialize their respective scores under the control of the control unit 21 (S1). For example, the empathy score, the reliability score, and the intimacy score are initialized to zero values, and the purchase probability score is initialized to 50 as the median value. When the conversation is repeated with the same customer and digital staff, the values of the empathy score, the reliability score, the intimacy score, and the purchase probability score at the end of the previous conversation may be carried over.

[0036] Data on the customer's utterance is received from the EC server 3 via the reception processing unit 31 (S2). Data on the digital staff's utterance is read from the storage unit 14. The customer's utterance and the digital staff's utterance are subjected to natural language analysis processing (S3). Part-of-speech analysis is performed, nouns are extracted, and syntactic analysis is executed.

[0037] Using the natural language parsing result, the empathy degree estimation processing unit 23 executes empathy degree estimation processing to calculate an empathy degree score (S4), the reliability degree estimation processing unit 24 executes reliability degree estimation processing to calculate a reliability degree score (S5), and the intimacy degree estimation processing unit 25 executes intimacy degree estimation processing to calculate an intimacy degree score (S6). Each processing will be described later. Also, the purchase possibility estimation processing unit 26 calculates a purchase possibility score representing the possibility that a customer purchases a product based on the empathy degree score, the reliability degree score, and the intimacy degree score (S7). The method of calculating a purchase possibility score representing the possibility that a customer purchases a product on an EC site from the empathy degree score, the reliability degree score, and the intimacy degree score is arbitrary. For example, the current value of the purchase possibility score is added with the change amount of each of the empathy degree score, the reliability degree score, and the intimacy degree score, or a value obtained by multiplying the change amount by a predetermined coefficient such as 0.5 to calculate the purchase possibility score.

[0038] Next, the persona extension information update unit 28 generates persona extension information in which nouns in the digital staff's utterance made immediately before the customer's utterance indicating the change are associated with the changes in the empathy degree score, the reliability degree score, and the intimacy degree score respectively (S8). Thereby, when any of the customer's empathy degree, reliability degree, and intimacy degree changes, the nouns in the digital staff's utterance that caused the change can be recognized. For example, when the customer's reliability degree score increases, it can be known that the increase is caused by "blue" and "jacket", in other words, the customer is interested in "blue" and "jacket", or the customer's preference can be specified.

[0039] In the next step S9, the utterance sentence generation request processing unit 30 generates an utterance sentence generation request together with the conversation history, the persona information, the persona extension information, the transition of the empathy degree score, the transition of the reliability degree score, the transition of the intimacy degree score, and the transition of the purchase possibility score. The transmission processing unit 32 transmits the utterance sentence generation request together with the data of the conversation history, the persona information, the persona extension information, the transition of the empathy degree score, the transition of the reliability degree score, the transition of the intimacy degree score, and the transition of the purchase possibility score to the utterance sentence generation device 2 via the communication interface 17 (S9).

[0040] The reception processing unit 31 receives data of the speech of the digital staff from the speech generation device 2 via the communication interface 17 (S10). The diction conversion processing unit 29 converts the diction of the speech of the digital staff into the diction of the set politeness level (S11). The transmission processing unit 32 transmits the data of the speech of the digital staff with the diction converted via the communication interface 17 to the EC server 3 (S12). When the conversation continues (S13, NO), the process returns to step S2 and waits for the next customer's speech. When the conversation has ended (S13, YES), the conversation process ends.

[0041] Fig. 5 shows the flow of the empathy estimation process in step S4 of Fig. 4. Refer to Fig. 10 and Fig. 11 for an example of the transition of the empathy score, reliability score, intimacy score, purchase possibility score, and the change in diction.

[0042] Based on the analysis result of the natural language analysis processing unit 22, the empathy degree estimation processing unit 23 determines whether the content of the digital staff's speech is in line with the conversation theme (S21). For example, if the EC site is a clothing store, the conversation theme is the clothing field. Typically, the nouns included in the digital staff's speech are collated with a dictionary of clothing-related terms to check whether they are listed in the dictionary of clothing-related terms. When the noun included in the digital staff's speech matches any of the terms listed in the dictionary of clothing-related terms, it is determined that the content of the digital staff's speech is in line with the conversation theme (S21, YES). When the noun included in the digital staff's speech does not match any of the terms listed in the dictionary of clothing-related terms, it is determined that the content of the digital staff's speech is not in line with the conversation theme (S21, NO). When it is determined that the content of the digital staff's speech is in line with the conversation theme (S21, YES), it waits for the next digital staff's speech. When it is determined that the content of the digital staff's speech is not in line with the conversation theme (S21, NO), it proceeds to the next step S22 and starts the operation process of the empathy score. Note that when it is determined that the digital staff's speech is in line with the conversation theme, that is, the digital staff's speech is within the scope of the conversation theme, the empathy score is not calculated, but the reliability score is calculated. In other words, when the digital staff's speech is casual conversation, the empathy score is calculated, and when the digital staff's speech is not casual conversation and is in line with the theme to be consulted originally, the reliability score is calculated.

[0043] In step S22, the natural language analysis processing unit 22 analyzes the customer's speech. In step S23, based on the analysis result of the natural language analysis processing unit 22, the empathy degree estimation processing unit 23 differentiates whether the customer's speech is affirmative, negative, or neither with respect to the digital staff's speech. Typically, by collating the predicates included in the customer's speech with a dictionary covering a plurality of pre-created affirmative expressions and a plurality of negative expressions and checking whether they are listed in the dictionary, it is differentiated whether the customer's speech is affirmative, negative, or neither.

[0044] When the customer's statement is positive, the empathy degree estimation processing unit 23 adds a predetermined value I1 to the current value of the empathy degree score to increase the empathy degree score (S24). When the customer's statement is negative, the empathy degree estimation processing unit 23 subtracts a predetermined value D1 from the current value of the empathy degree score to decrease the empathy degree score (S25). When the customer's statement is neither positive nor negative, it is assumed that the customer is indifferent to the statement of the digital staff, and the empathy degree estimation processing unit 23 subtracts a predetermined value D2 whose absolute value is larger than the predetermined value D1 from the current value of the empathy degree score to further decrease the empathy degree score (S26).

[0045] FIG. 6 shows the flow of the reliability estimation process in step S5 of FIG. 4. The reliability estimation processing unit 24 determines whether the content of the statement of the digital staff follows the theme of the conversation based on the analysis result of the natural language analysis processing unit 22 (S31). The determination method is the same as that in step S21. When it is determined that the content of the statement of the digital staff does not follow the theme of the conversation (S31, YES), it waits for the next statement of the digital staff. When it is determined that the content of the statement of the digital staff follows the theme of the conversation (S31, NO), it proceeds to the next step S32 and starts the calculation process of the reliability score.

[0046] In step S32, the natural language analysis processing unit 22 analyzes the customer's statement. In step S33, the reliability estimation processing unit 24 distinguishes whether the customer's statement is positive, negative, or neither with respect to the statement of the digital staff based on the analysis result of the natural language analysis processing unit 22. The method for this distinction is also the same as (S1) S23.

[0047] When the customer's statement is positive, the reliability estimation processing unit 24 adds a predetermined value I11 to the current value of the reliability score to increase the reliability score (S34). When the customer's statement is negative, the reliability estimation processing unit 24 subtracts a predetermined value D11 from the current value of the reliability score to decrease the reliability score (S35). When the customer's statement is neither positive nor negative, assuming that the customer is indifferent to the statement of the digital staff, the reliability estimation processing unit 24 subtracts a predetermined value D12 whose absolute value is larger than the predetermined value D11 from the current value of the reliability score to further decrease the reliability score (S36).

[0048] FIG. 7 shows the flow of the intimacy estimation processing in step S6 of FIG. 4. The intimacy estimation processing unit 25 identifies the politeness level of the customer's statement (S41). For the determination of the politeness level, the ending of the customer's statement is collated with a dictionary regarding a plurality of ending expressions related to each of a plurality of levels of politeness ranging from honorific language to friendly language, and the politeness level of the ending expression listed in the dictionary that matches the ending of the customer's statement is confirmed.

[0049] The intimacy estimation processing unit 25 determines whether or not the identified politeness level of the customer's statement has changed from the politeness level of the previous customer's statement (S42). When the politeness level has not changed (S42, NO), the process returns to step S41 and waits for the next customer's statement. When the politeness level has changed (S42, YES), it is determined whether or not the identified politeness level of the customer's statement has decreased compared to the politeness level of the previous customer's statement (S43).

[0050] When it is determined that the identified politeness level of the customer's statement has decreased compared to the politeness level of the previous customer's statement (S43, YES), a predetermined value I21 is added to the current value of the intimacy score to increase the intimacy score (S44). When it is determined that the identified politeness level of the customer's statement has become higher than the politeness level of the previous customer's statement (S43, NO), a predetermined value D21 is subtracted from the current value of the intimacy score to decrease the intimacy score (S45).

[0051] Figure 8 shows the flow of the speech style change process. The speech style change process is executed in parallel with the conversation process shown in Figure 4. The speech style conversion processing unit 29 first initially sets the speech style of the digital staff's utterance to "5" as, for example, the median of the politeness levels prepared in 11 levels before the conversation (S51). When receiving the customer's utterance, the politeness level of the customer's utterance is specified (S52). Then, the speech style conversion processing unit 29 determines whether the specified politeness level of the customer's utterance has changed from the politeness level of the previous customer's utterance (S53).

[0052] When the politeness level has not changed (S53, NO), the process returns to step S52 and waits for the next customer's utterance. When the politeness level has changed (S53, YES), it is determined whether the specified politeness level of the customer's utterance has decreased compared to the politeness level of the previous customer's utterance (S54).

[0053] When it is determined that the specified politeness level of the customer's utterance has decreased compared to the politeness level of the previous customer's utterance (S54, YES), that is, when the customer's intimacy with the digital staff has increased, the speech style of the digital staff is accordingly changed to a lower politeness level by one step (S55). When it is determined that the specified politeness level of the customer's utterance has increased compared to the politeness level of the previous customer's utterance (S54, NO), that is, when the customer's intimacy with the digital staff has decreased, the speech style of the digital staff is accordingly changed to a higher politeness level by one step (S56).

[0054] When the conversation continues (S57, NO), the process returns to step S52 and waits for the next customer's utterance. When the conversation has ended (S57, YES), the speech style change process ends.

[0055] As described above, according to this embodiment, the empathy score, reliability score of the customer for the speech of the digital staff, and the intimacy score of the customer for the digital staff are calculated, and the changes in these empathy scores and reliability scores are sent to the speech generation device 2 together with the conversation history and the speech generation request. Thus, the speech generation device 2 can generate the speech of the digital staff that enhances the empathy and reliability and suppresses the decrease in empathy and reliability.

[0056] Also, according to this embodiment, by changing the politeness level of the digital staff's language following the change in the politeness level of the customer's speech, the intimacy can be enhanced and the decrease in intimacy can be suppressed.

[0057] Although several embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the gist of the invention. These embodiments and their modifications are included in the scope and gist of the invention, as well as in the invention described in the claims and its equivalent scope.

Explanation of Reference Numerals

[0058] 1... Information processing device, 2... Speech generation device (generative AI), 3... EC server, 4... User terminal, 11... Processor, 10... System bus, 12... RAM, 13... ROM, 14... Storage unit, 15... Input device, 16... Display, 17... Communication interface, 21... Control unit, 22... Natural language analysis processing unit, 23... Empathy estimation processing unit, 24... Reliability estimation processing unit, 25... Intimacy estimation processing unit, 26... Purchase possibility estimation processing unit, 27... Persona information update unit, 28... Persona extension information update unit, 29... Language conversion processing unit, 30... Speech generation request processing unit, 31... Reception processing unit, 32... Transmission processing unit.

Claims

1. An information processing apparatus having a memory for storing a program, a processor for executing the program, and a communication interface for communicating with a user terminal and a speech generation device (generation AI) via an electric public communication network, for a virtual user to have a conversation with an actual user, by executing the program, the processor means for receiving data of the speech of the actual user from the user terminal, means for executing natural language analysis processing on the speech of the actual user, means for generating persona information about the actual user based on the part-of-speech and syntactic analysis results decomposed by the natural language analysis processing, means for transmitting a generation request for the speech of the virtual user to the speech generation device together with the data of the speech history of each of the actual user and the virtual user and the persona information, means for receiving data of the speech of the virtual user from the speech generation device, An information processing apparatus that functions as means for transmitting the received data of the speech of the virtual user to the user terminal.

2. An information processing apparatus having a memory for storing a program, a processor for executing the program, and a communication interface for communicating with a user terminal via an electric public communication network, by executing the program, the processor means for receiving data of the speech of the user from the user terminal, means for executing natural language analysis processing on the speech of the user, An information processing apparatus that functions as means for generating persona information about the user based on the part-of-speech and syntactic analysis results decomposed by the natural language analysis processing.

3. An information processing apparatus having a memory for storing a program, a processor for executing the program, and a communication interface for communicating with a user terminal and a speech generation device (generation AI) via an electric public communication network, for a virtual user to have a conversation with an actual user, means for receiving data of the speech of the actual user from the user terminal, means for executing natural language analysis processing on the speech of the actual user, means for generating persona information about the actual user based on the part-of-speech and syntactic analysis results decomposed by the natural language analysis processing, means for transmitting a request for generating a statement of the virtual user to the statement generation device together with data on the statement history of each of the real user and the virtual user and the persona information; means for receiving data on the statement of the virtual user from the statement generation device; A program that functions as means for transmitting the received data on the statement of the virtual user to the user terminal.

4. An information processing apparatus having a memory for storing a program, a processor for executing the program, and a communication interface for communicating with a user terminal via an electric public communication network, means for receiving data on a statement of a user from the user terminal; means for performing natural language analysis processing on the statement of the user; A program that functions as means for generating persona information about the user based on the part-of-speech and syntactic analysis results decomposed by the natural language analysis processing.

Images