A psychological companionship digital human system

Abstract

The present invention belongs to the field of artificial intelligence technology, and specifically discloses a psychological companionship digital human system, which includes: a user information recording module, a user information identification module, a language information confirmation module, a digital human reply generation module, a session end judgment module, a user session collection module, a user session analysis module, a database and a user satisfaction feedback terminal; the present invention further analyzes the user's language and text according to the user's voice emotion type and language expression type, thereby improving the accuracy of understanding the user's needs, thereby improving the digital human personalized customization qualification index, thereby ensuring the effectiveness of psychological companionship, and at the same time, by counting the text similarity between the output questions of each user session and the output questions of the remaining user sessions, thereby confirming the user's digital human usage satisfaction, ensuring the accuracy of user satisfaction confirmation, and thus ensuring the timeliness of discovery of potential problems of the user.

Description

Technical Field

[0001] The invention belongs to the technical field of artificial intelligence and relates to a psychological companionship digital human system. Background Art

[0002] The psychological companion digital human is a virtual partner created using artificial intelligence technology. It can customize its image and provide companionship to users at any time with high confidentiality. At the same time, through natural language processing, machine learning and emotional computing technology, it can understand the user's language information and emotional state, and provide users with emotional support, psychological counseling and daily companionship. It has broad application prospects in mental health care, personal growth and development, and care for special groups.

[0003] For example, the Chinese invention patent with publication number CN117216228B discloses a psychological companion robot and its interaction method, wherein the psychological companion robot includes a robot body and a robot interaction management system, the robot body includes a power supply system and a dialogue system, and the robot interaction management system includes an interaction database, an interaction analysis module, an interaction optimization terminal, and an interaction feedback terminal. The present invention performs in-depth analysis on each input text during each interaction, thereby counting the user's interaction satisfaction and interaction optimization demand, and confirms the adaptation position sorting update frequency of the reply text in the interaction language library accordingly, effectively solving the problem that the fixed reply update frequency is difficult to adjust according to the specific needs of each user, ensuring the interaction quality and interaction effect of the psychological companion robot, improving the interaction accuracy of the psychological companion robot, and realizing the personalized formulation of user interaction optimization.

[0004] The above existing technologies have the following deficiencies: 1. Currently, the main implementers in the field of psychological companionship are psychological companion robots, that is, there is insufficient attention paid to psychological companion digital humans, and thus the personalized needs of users cannot be met, resulting in a lack of uniqueness and pertinence when users interact with digital humans.

[0005] 2. Currently, when conducting interactive analysis of psychological companionship robots, only the input text is deeply analyzed, and no further analysis of the user's emotions is performed. This cannot guarantee the accuracy of understanding user needs, which in turn leads to an insufficient personalized customization qualification index of the robot, thereby limiting the effectiveness of psychological companionship and affecting the user's psychological companionship service conversation experience.

[0006] 3. Currently, interaction satisfaction is confirmed only based on user feedback, and no further user satisfaction analysis is performed based on the user's input text. This makes it impossible to guarantee the accuracy of user satisfaction confirmation, and further makes it impossible to guarantee the timeliness of discovering potential user problems, which results in insufficient targeting of subsequent interaction improvements. Summary of the invention

[0007] In view of this, in order to solve the problems raised in the above background technology, a psychological companionship digital human system is now proposed.

[0008] The purpose of the present invention can be achieved through the following technical solutions: The present invention provides a psychological companionship digital human system, including: a user information recording module, used to record the user's historical conversation information, the historical conversation information includes the conversation text of each historical conversation of the user and the interval time between each interactive question of the user.

[0009] The user information identification module is used to identify the user's language content.

[0010] The language information confirmation module is used to confirm the user's language information according to the user's language content.

[0011] The digital human reply generation module is used to generate the digital human's reply information according to the user's language information.

[0012] The session end judgment module is used to monitor the current session interval length when the user session is stopped, and then judge the user session end to obtain the user session end judgment result.

[0013] The user session collection module is used to collect the session text of the user's current session when the user ends the session judgment result.

[0014] The user conversation analysis module is used to analyze the user's digital human conversation satisfaction based on the conversation text of the user's current conversation, and obtain the user's digital human conversation satisfaction.

[0015] The database is used to store keyword sets of each keyword type.

[0016] The user satisfaction feedback terminal is used to provide corresponding feedback based on the user's digital human conversation satisfaction.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention breaks the lack of attention to the psychological companionship of digital humans by conducting psychological companionship analysis of digital humans, thereby meeting the personalized needs of users, thereby improving the uniqueness and pertinence of users when interacting with digital humans.

[0018] (2) The present invention further analyzes the user's language and text according to the user's voice emotion type and language expression type, breaking the current deficiency of not further analyzing the user's emotions, improving the accuracy of understanding user needs, and further improving the digital human personalized customization qualification index, thereby ensuring the effectiveness of psychological accompaniment and improving the user's conversational experience of psychological accompaniment services.

[0019] (3) The present invention confirms the user's digital human conversation satisfaction by counting the text similarities between the questions output by each user session and the questions output by the remaining users. This avoids the current deficiency of not conducting further user satisfaction analysis based on the user's input text, ensures the accuracy of user satisfaction confirmation, and further ensures the timeliness of discovering potential user problems, thereby improving the pertinence of subsequent interaction improvements. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0021] Figure 1 It is a schematic diagram of the connection of each module of the system of the present invention.

[0022] Figure 2 It is a connection diagram of the user language information confirmation step of the present invention.

[0023] Figure 3 A connection diagram of the steps for determining when a user ends a session according to the present invention. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] See also Figure 1 As shown, the present invention provides a psychological companionship digital human system, which includes: a user information recording module, a user information identification module, a language information confirmation module, a digital human reply generation module, a conversation end judgment module, a user conversation collection module, a user conversation analysis module, a database and a user satisfaction feedback terminal.

[0026] In the above, the language information confirmation module is connected to the user information recording module, the user information identification module, the digital human reply generation module and the database respectively, the session end judgment module is connected to the digital human reply generation module, the user information recording module and the user session collection module respectively, and the user session analysis module is also connected to the user session collection module and the user satisfaction feedback terminal respectively.

[0027] The user information recording module is used to record the user's historical session information, which includes the conversation texts of each historical conversation of the user and the interval length between each interactive question of the user.

[0028] The user information identification module is used to identify the language content of the user.

[0029] The language information confirmation module is used to confirm the user's language information according to the user's language content.

[0030] See also Figure 2 As shown, exemplarily, the confirming of the user's language information includes: G1, extracting the user's voice from the user's language content, and then confirming the user's voice emotion type.

[0031] Furthermore, the confirming of the user's voice emotion type includes: G1-1, extracting the user's voice from the user's language content, and then extracting the user's voice fundamental frequency at each set monitoring time point from the user's voice.

[0032] It should be added that the specific process of extracting the fundamental frequency of the user's voice at each set monitoring time point from the user's voice is as follows: the user's voice is collected by a digital human, and the user's voice is preprocessed to obtain a preprocessed voice signal, and the preprocessed voice signal is framed to divide the voice signal into several short-time segments, and the autocorrelation function of each frame is calculated. The autocorrelation function will have a peak at the fundamental frequency period, and then by finding the peak position of the autocorrelation function, the period of the fundamental frequency is determined, and the fundamental frequency value is calculated.

[0033] G1-2. Calculate the average of the user's voice fundamental frequency at each set monitoring time point, and use the calculation result as the user's voice fundamental frequency.

[0034] G1-3. Compare the user's voice fundamental frequency with the voice fundamental frequency intervals corresponding to each voice fundamental frequency level value to obtain the user's voice fundamental frequency level value.

[0035] It should be added that the speech fundamental frequency intervals corresponding to each speech fundamental frequency level value are as follows: the high speech fundamental frequency level value is 1, the high speech fundamental frequency level value is related to positive emotions, and its corresponding high speech fundamental frequency interval is above 200Hz; the middle speech fundamental frequency level value is 0, the middle speech fundamental frequency level value is regarded as the neutral emotion range, and its corresponding high speech fundamental frequency interval is 100Hz--200Hz; the low speech fundamental frequency level value is -1, the low speech fundamental frequency level value is related to negative emotions, and its corresponding low speech fundamental frequency interval is below 100Hz.

[0036] G1-4. Confirm the user's voice intonation level value according to the user's voice fundamental frequency at each set monitoring time point.

[0037] Furthermore, the confirmation of the user's voice intonation level value includes: G1-4-1, analyzing the user's voice fundamental frequency at each set monitoring time point in the same manner as the user's voice fundamental frequency level analysis method to obtain the user's voice fundamental frequency level value at each set monitoring time point.

[0038] G1-4-2. Combine each set monitoring time point with its next set monitoring time point in pairs to obtain each set monitoring time group.

[0039] G1-4-3, record the fundamental frequencies of the speech at the first set monitoring time point and the second set monitoring time point in a set monitoring time group as and .

[0040] G1-4-4, count the base frequency change rate of the set monitoring time group , , and then obtain the fundamental frequency change rate of each set monitoring time group.

[0041] In a specific embodiment, the calculation formula for the fundamental frequency change rate is: (fundamental frequency value at the later set monitoring time point-fundamental frequency value at the earlier set monitoring time point) / fundamental frequency value at the earlier set monitoring time point. For example, the fundamental frequency values ​​at two adjacent time points are 120 Hz and 130 Hz, respectively, then the fundamental frequency change rate is (130-120) / 120×100%≈8.3%.

[0042] G1-4-5. Match and compare the fundamental frequency change rate of each set monitoring time group with the fundamental frequency change rate interval corresponding to each voice and intonation level value to obtain the voice and intonation level value of each set monitoring time group.

[0043] G1-4-5. Count the number of set monitoring time groups for each voice and intonation level value, and then select the voice and intonation level value corresponding to the maximum number of set monitoring time groups, and use it as the user's voice and intonation level value.

[0044] It should be added that the user's voice intonation is obtained by analyzing the fundamental frequency of the voice signal to understand the ups and downs of the voice. The fundamental frequency of the voice can be extracted through conversational voice analysis tools such as Praat. For example, the fundamental frequency value at each time point in the voice signal is calculated to obtain a curve of the fundamental frequency changing over time.

[0045] According to the results of fundamental frequency analysis, the intonation contour is drawn and the speech signal is divided into several small segments, such as sentences or phrases. Then, the changing trend of the fundamental frequency is observed in each segment. For example, if the fundamental frequency gradually rises in a segment, the segment has a rising intonation; if the fundamental frequency gradually decreases in a segment, it has a falling intonation; if the fundamental frequency remains relatively stable in a segment, it has a flat tone.

[0046] It should be added that the setting process of the user's voice intonation level value is: rising tone (fundamental frequency continues to rise or rises significantly at the end of the interval), falling tone (fundamental frequency continues to fall or falls significantly at the end of the interval) and flat tone (fundamental frequency remains basically stable). Rising tone expresses emotions such as doubt and excitement, so the rising voice intonation level value is assigned to 1. Falling tone expresses emotions such as affirmation and frustration, so the falling voice intonation level value is assigned to -1. Flat tone is a neutral emotion, so the flat voice intonation level value is assigned to 0. For example, if the fundamental frequency in a time interval gradually rises from the beginning to the end, and the increase is large, it can be judged as a rising tone.

[0047] G1-5. Extract the number of words and duration of the user's voice from the user's voice, and then use the ratio of the number of words to the duration of the voice as the user's voice speed, and extract the user's voice energy from the user's voice, and then match the user's voice speed level value and the user's voice energy level value in the same way as the matching method of the user's voice fundamental frequency level value.

[0048] It should be added that speech energy refers to the power carried by the speech signal. Speech energy reflects the strength and intensity of the user's voice. Because the speech energy is different in different emotional states, combining them can more accurately judge the user's emotions. Therefore, when confirming the user's speech emotion type, it is necessary to analyze the user's speech energy.

[0049] For example, when a person is emotional and speaks loudly, the speech energy is relatively high, while when a person speaks softly and is relatively calm, the speech energy is relatively low.

[0050] It should be added that the process of setting the user's voice speed level value is: the speed is determined by calculating the number of words spoken per minute, and the fast speed is set to more than 150 words per minute, the medium speed is 100-150 words per minute, and the slow speed is less than 100 words per minute. The fast speed is related to positive emotions (excitement, eager expression), so the fast speech speed level value is assigned to 1, the medium speed is neutral emotion, so the medium speech speed level value is assigned to 0, and the slow speed represents negative emotions (sadness, contemplation) or cautious expression, so the slow speech speed level value is assigned to -1.

[0051] It should be added that the user's voice speed is obtained by extracting the user's voice duration from the user's voice, converting the voice into text through conversational voice recognition software, and then counting the number of words in the text, and then using the speed calculation formula: speed = (number of words / duration) × 60 to obtain the user's voice speed, where 60 means that when the unit of time is converted from seconds to minutes, 1 minute is equal to 60 seconds, that is, the calculated speed unit is words / minute. For example, if a person speaks 400 words in 2 minutes, then the speed = (400 / 2) × 60 = 200 × 60 = 12000 (words / hour), which is converted into words / minute as 12000 ÷ 60 = 200 (words / minute).

[0052] In a specific embodiment, the duration of the voice segment is 5 seconds, and the number of words is 30, then the speaking speed = (30 / 5) × 60 = 360 (words / minute).

[0053] It should be added that the user's voice energy is obtained by dividing the audio signal into several short-time segments (frames), and then calculating the energy of each frame to obtain the user's voice energy. The commonly used energy calculation formula is: , is the length of the frame, The audio signal sampling points, .

[0054] It should be added that the process of setting the user's voice energy level value is as follows: the root mean square energy of the voice signal is calculated, and the energy values ​​are divided into three levels: high, medium, and low. High energy corresponds to positive emotions, so the high voice energy level value is assigned to 1; medium energy is neutral emotions, so the medium voice energy level value is assigned to 0; low energy corresponds to negative emotions, so the low voice energy level value is assigned to -1.

[0055] G1-6. Extract the number of voice pauses of the user and the duration of each voice pause from the user's voice, and then determine the user's voice pause level value.

[0056] Further, the step of determining the user's voice pause level value includes: G1-6-1, recording the number of the user's voice pauses and the duration of each voice pause as and , Number the number of speech pauses. .

[0057] G1-6-2, from The maximum voice pause duration of the user is filtered out and recorded as , and Calculate the mean value to get the average voice pause duration of the user, recorded as .

[0058] G1-6-3. Count the user's voice pause level , , and They are respectively the set appropriate number of speech pauses and the appropriate duration of speech pauses.

[0059] It should be added that and All of them are extracted from the technical specifications table of the psychological companion digital human.

[0060] It should be added that the setting process of the user's voice pause level value is as follows: the number and duration of pauses in the voice are counted. Long pauses indicate negative emotions (hesitation, tension) or thinking about complex problems, so the long voice pause level value is assigned to -1, medium pauses indicate neutral emotions, so the medium voice pause level value is assigned to 0, and short pauses indicate positive emotions (confidence, fluent expression), so the short voice pause level value is assigned to 1.

[0061] G1-7. Import the user's voice fundamental frequency level value, user's voice intonation level value, user's voice speed level value, user's voice pause level value and user's voice energy level value into the user's voice emotion type judgment model to obtain the user's voice emotion type.

[0062] It should be added that the specific judgment process of the user voice emotion type judgment model is: when the voice level values ​​of each user are 1, the positive emotion is taken as the user's voice emotion type; when the voice level values ​​of each user are 0, the neutral emotion is taken as the user's voice emotion type; when the voice level values ​​of each user are -1, the negative emotion is taken as the user's voice emotion type; when the voice level values ​​of each user are different, the voice emotion type with the largest number of voice level values ​​is selected as the user's voice emotion type; if there are two voice emotion types with the largest number of voice level values, then combined with the user's historical conversation information, the voice emotion type with more historical voice emotion types of the user is selected as the user's voice emotion type.

[0063] In a specific embodiment, the high voice fundamental frequency level value is assigned a value of 1, the middle voice fundamental frequency level value is assigned a value of 0, the low voice fundamental frequency level value is assigned a value of -1, the rising voice intonation level value is assigned a value of 1, the falling voice intonation level value is assigned a value of -1, the flat voice intonation level value is assigned a value of 0, the fast voice speed level value is assigned a value of 1, the middle voice speed level value is assigned a value of 0, the slow voice speed level value is assigned a value of -1, the long voice pause level value is assigned a value of -1 for negative emotions, the middle voice pause level value is assigned a value of 0, the short voice pause level value is assigned a value of 1 for positive emotions, the high voice energy level value is assigned a value of 1, the middle voice energy level value is assigned a value of 0, and the low voice energy level value is assigned a value of -1.

[0064] If the user's voice feature level values ​​are: voice fundamental frequency level value 0, voice intonation level value 0, voice speed level value 0, voice pause level value 0, and voice energy level value 0, then they are imported into the user voice emotion type judgment model, and the judgment result is neutral emotion.

[0065] G2. Convert the user's voice into the user's spoken and written language, extract keywords from the user's spoken and written language, and then confirm the keyword conformity of each language expression type.

[0066] Furthermore, the confirming of the keyword conformity of each language expression type includes: G2-1, extracting a keyword set of each keyword type from a database.

[0067] G2-2. Match and compare each keyword with the keyword set of each keyword type to obtain the keyword type to which each keyword belongs, and then count the number of keywords of each keyword type, recorded as , Number the keyword type. .

[0068] G2-3, record the number of keywords as .

[0069] G2-4, and The ratio of is taken as the keyword ratio of each language expression type and recorded as .

[0070] G2-5. According to the conversation texts of each user's historical conversations, the historical weight factor of each language expression type is analyzed to obtain the historical weight factor of each language expression type, which is recorded as .

[0071] Furthermore, the historical weight factor analysis of each language expression type includes: G2-5-1, extracting each keyword of each user's conversation from the conversation text of each user's historical conversation, and then The analysis method is similar to analyzing the keyword ratio of each language expression type in each user's historical conversation.

[0072] G2-5-2. Sum the keyword ratios of each language expression type in each historical session of the user to obtain the keyword ratios of each language expression type in the historical session of the user.

[0073] G2-5-3. Count the number of historical user sessions, and then use the ratio of the keyword ratio of each language expression type in the user's historical session to the number of historical user sessions as the historical weight factor of each language expression type, and record it as .

[0074] G2-6. Count the keyword matching degree of each language expression type , .

[0075] G3. Filter out the maximum value from the keyword matching of each language expression type, and use the language expression type to which the maximum keyword matching belongs as the user's language expression type.

[0076] It should be added that the language expression types include but are not limited to declarative language, emotional language, demand language and interactive language.

[0077] G4. Use the user's voice emotion type and language expression type as the user's language information.

[0078] The embodiment of the present invention further analyzes the user's language and text according to the user's voice emotion type and language expression type, breaking the current deficiency of not further analyzing the user's emotions, improving the accuracy of understanding user needs, and further improving the digital human personalized customization qualification index, thereby ensuring the effectiveness of psychological companionship and improving the user's conversational experience of the psychological companionship service.

[0079] The digital human reply generation module is used to generate the digital human reply information according to the user's language information.

[0080] It should be added that the specific process of generating the digital person's reply information is as follows: based on the user's language content, the digital person's reply content is generated, and then the user's voice emotion type is matched and compared with the user's voice emotion type corresponding to each digital person's reply emotion type to obtain the digital person's reply emotion type, and at the same time, the digital person's reply expression type is matched in the same way according to the matching method of the digital person's reply emotion type, so that the digital person's reply content is modified according to the digital person's reply emotion type and the digital person's reply expression type, so as to obtain the digital person's reply information.

[0081] The session end judgment module is used to monitor the current session interval length when the user session is stopped, and then judge whether the user has ended the session to obtain a user session end judgment result.

[0082] See also Figure 3 As shown, illustratively, the determination of user session termination includes: Y1, filtering out the longest interval between user interaction questions in history from the intervals between user interaction questions in each user session history.

[0083] Y2. Compare the current session interval duration when the user session is stopped with the user's longest historical session interval duration.

[0084] Y3. If the current session interval duration when the user session is stopped is greater than or equal to the longest session interval duration in the history of the user, the user session termination is regarded as the user session termination judgment result.

[0085] Y4. If the current session interval duration when the user session is stopped is less than the user's longest historical session interval duration, the user's failure to terminate the session is regarded as the result of the user's session termination.

[0086] The user session collection module is used to collect the session text of the user's current session when the user ends the session judgment result is that the user ends the session.

[0087] The user conversation analysis module is used to analyze the user's digital human conversation satisfaction based on the conversation text of the user's current conversation, and obtain the user's digital human conversation satisfaction.

[0088] Exemplarily, the digital human conversation satisfaction analysis of the user includes: R1, extracting the text of each user's conversation output question from the conversation text of the user's current conversation, and generating a word set of each user's conversation output question, and then counting the text similarity between each user's conversation output question and the remaining user's conversation output questions. , Outputs the question number for the user session, , Output the question number for the remaining user session, .

[0089] Furthermore, the method of counting the text similarity between each user session output question and each remaining user session output question includes: R1-1, recording the session output word set of each user session output question as , and the set of session output words for each remaining user session output problem is recorded as .

[0090] R1-2. Count the text similarities between the output questions of each user session and the output questions of each remaining user session , .

[0091] In a specific embodiment, the user session outputs question 1: "What is the weather like today?", the user session outputs question 2: "Will it rain today?", and the user session outputs question 3: "What will the weather be like tomorrow?".

[0092] Construct a set of session output words for user session output questions: for user session output question 1: today, the weather, how, for user session output question 2, today, will it rain, for user session output question 3, tomorrow, the weather, how.

[0093] Calculate the text similarity between user session output question 1 and user session output question 2: .

[0094] Calculate the text similarity between user session output question 1 and user session output question 3: .

[0095] Calculate the text similarity between user session output question 2 and user session output question 3: .

[0096] R2. Match and compare the text similarity between each user session output question and its remaining user session output questions with the user session output question similarity corresponding to each satisfaction influence value to obtain the satisfaction influence value between each user session output question and its remaining user session output questions.

[0097] R3. Calculate the average of the satisfaction impact values ​​of each user's session output question and the remaining user's session output question to obtain the satisfaction impact value of each user's session output question, and then use the average calculation result of the satisfaction impact value of each user's session output question as the user's digital human session satisfaction impact value, and record it as .

[0098] R4. Count users’ satisfaction with digital human conversations , , To set the reference satisfaction impact value, is a natural constant.

[0099] It should be added that It is extracted from the technical specification sheet of the psychological companion digital human.

[0100] The embodiment of the present invention confirms the user's digital human conversation satisfaction by counting the text similarities between the questions output by each user session and the questions output by each remaining user session, thereby avoiding the current deficiency of not performing further user satisfaction analysis based on the user's input text, ensuring the accuracy of user satisfaction confirmation, and further ensuring the timeliness of discovering potential user problems, thereby improving the pertinence of subsequent interaction improvements.

[0101] The database is used to store keyword sets of each keyword type.

[0102] In a specific embodiment, the keyword sets of each keyword type are specifically as follows: the declarative keyword set includes: time words (today, yesterday, tomorrow, etc.), place words (school, park, office, etc.), person words (father, mother, teacher, etc.), thing words (books, computers, mobile phones, etc.), the emotional keyword set includes: happiness (happy, joyful, excited, etc.), sadness (sad, sad, frustrated, etc.), anger (angry, annoyed, angry, etc.), anxiety (worry, nervous, uneasy, etc.), the demand-type keyword set includes: companionship (chatting, listening, etc.), suggestions (what to do, how, etc.), entertainment (games, movies, music, etc.), help (assistance, support, etc.), the interactive keyword set includes: discussion (exchange, discussion, etc.), feedback (response, evaluation, etc.), invitation (together, together, etc.).

[0103] The user satisfaction feedback terminal is used to provide corresponding feedback based on the user's digital human conversation satisfaction.

[0104] The embodiments of the present invention break the lack of attention to the psychological companionship of digital humans by performing psychological companionship analysis of digital humans, thereby meeting the personalized needs of users, thereby improving the uniqueness and pertinence of users when interacting with digital humans.

[0105] The above contents are merely examples and explanations of the concept of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, they shall all fall within the protection scope of the present invention.

Claims

1. A psychological companionship digital human system, characterized in that: include: User information recording module, recording the user's historical session information, including the conversation text of each historical conversation of the user and the interval length between each interactive question of the user; User information identification module, identifying the user's language content; A language information confirmation module confirms the user's language information according to the user's language content; A digital human reply generation module generates a digital human reply message based on the user's language information; The session end judgment module monitors the current session interval length when the user session is stopped, and then judges the user session end to obtain the user session end judgment result; A user session collection module, when the user ends the session, collects the session text of the user's current session; The user conversation analysis module analyzes the user's digital human conversation satisfaction based on the conversation text of the user's current conversation to obtain the user's digital human conversation satisfaction; A database storing a keyword set of each keyword type; User satisfaction feedback terminal, which provides corresponding feedback based on the user's digital human conversation satisfaction; The language information of the confirmed user includes: G1. Extract the user's voice from the user's language content, and then confirm the user's voice emotion type; G2. Convert the user's voice into the user's language and text, extract keywords from it, and then confirm the keyword conformity of each language expression type; G3. Filter out the maximum value from the keyword matching degree of each language expression type, and use the language expression type to which the maximum keyword matching degree belongs as the language expression type of the user; G4. Using the user's voice emotion type and language expression type as the user's language information; The confirmation of the keyword matching degree of each language expression type includes: Extracting a keyword set of each keyword type from a database; Each keyword is matched and compared with the keyword set of each keyword type to obtain the keyword type to which each keyword belongs, and then the number of keywords of each keyword type is counted, which is recorded as , Number the keyword type. ; The number of keywords is recorded as ; Will and The ratio of is taken as the keyword ratio of each language expression type and recorded as ; According to the conversation texts of each user's historical conversations, the historical weight factor analysis of each language expression type is performed to obtain the historical weight factor of each language expression type, which is recorded as ; Statistics on keyword matching of each language expression type , .

2. A psychological companionship digital human system according to claim 1, characterized in that: The step of confirming the user's voice emotion type includes: Extracting the user's voice from the user's language content, and then extracting the user's voice fundamental frequency at each set monitoring time point from the user's voice; Calculate the average of the user's voice fundamental frequency at each set monitoring time point, and use the calculation result as the user's voice fundamental frequency; Compare the user's voice fundamental frequency with the voice fundamental frequency intervals corresponding to each voice fundamental frequency level value to obtain the user's voice fundamental frequency level value; According to the user's voice fundamental frequency at each set monitoring time point, confirm the user's voice intonation level value; Extract the number of words and duration of the user's voice from the user's voice, and then use the ratio of the number of words to the duration of the voice as the user's voice speed, and extract the user's voice energy from the user's voice, and then match the user's voice speed level value and the user's voice energy level value in the same way as the matching method of the user's voice fundamental frequency level value; Extracting the number of voice pauses of the user and the duration of each voice pause from the user's voice, and then determining the user's voice pause level value; The user's voice fundamental frequency level value, user's voice intonation level value, user's voice speed level value, user's voice pause level value and user's voice energy level value are introduced into the user's voice emotion type judgment model to obtain the user's voice emotion type.

3. A psychological companionship digital human system according to claim 2, characterized in that: The step of confirming the user's voice intonation level value includes: The user's voice fundamental frequency at each set monitoring time point is analyzed in the same manner as the user's voice fundamental frequency level analysis method to obtain the user's voice fundamental frequency level value at each set monitoring time point; Combining each set monitoring time point with its next set monitoring time point in pairs to obtain each set monitoring time group; The fundamental frequencies of the speech at the first set monitoring time point and the second set monitoring time point in a set monitoring time group are respectively recorded as and ; Statistics of the base frequency change rate of the set monitoring time group , , and then obtain the fundamental frequency change rate of each set monitoring time group; The fundamental frequency change rate of each set monitoring time group is matched and compared with the fundamental frequency change rate interval corresponding to each voice and intonation level value to obtain the voice and intonation level value of each set monitoring time group; The number of set monitoring time groups of each voice and intonation level value is counted, and then the voice and intonation level value corresponding to the maximum set monitoring time group number is screened out and used as the user's voice and intonation level value.

4. A psychological companionship digital human system according to claim 2, characterized in that: The step of confirming the user voice pause level value includes: The number of voice pauses and the duration of each voice pause are recorded as and , Number the number of speech pauses. ; from The maximum voice pause duration of the user is filtered out and recorded as , and Calculate the mean value to get the average voice pause duration of the user, recorded as ; Statistics of user voice pause levels , , and They are respectively the set appropriate number of speech pauses and the appropriate duration of speech pauses.

5. The psychological companionship digital human system according to claim 1, characterized in that: The historical weight factor analysis of each language expression type includes: Extract the keywords of each user's conversation from the conversation text of each user's history, and then The analysis method is similar to that of analyzing the keyword ratio of each language expression type in each user's historical conversation; Sum the keyword ratios of each language expression type in each historical session of the user to obtain the keyword ratios of each language expression type in the historical session of the user; The number of historical user sessions is counted, and then the ratio of the keyword ratio of each language expression type in the user's historical session to the number of historical user sessions is used as the historical weight factor of each language expression type, and recorded as .

6. The psychological companionship digital human system according to claim 1, characterized in that: The step of determining that the user has terminated the session includes: Y1. Filter out the longest interval between user interaction questions in the user's history from the interval between each user's interaction questions in each user's history session; Y2. Compare the current session interval length when the user session is stopped with the longest session interval length in the history of the user; Y3. If the current session interval duration when the user session is stopped is greater than or equal to the longest session interval duration in the history of the user, the user session termination is regarded as the user session termination judgment result; Y4. If the current session interval duration when the user session is stopped is less than the user's longest historical session interval duration, the user's failure to terminate the session is regarded as the result of the user's session termination.

7. The psychological companionship digital human system according to claim 1, characterized in that: The analysis of the user's digital human conversation satisfaction includes: R1. Extract the text of each user's session output question from the conversation text of the user's current conversation, and generate a word set of each user's session output question, and then count the text similarity of each user's session output question with its remaining user session output questions , Outputs the question number for the user session, , Output the question number for the remaining user session, ; R2, matching and comparing the text similarity between each user session output question and its remaining user session output questions with the user session output question similarity corresponding to each satisfaction influence value, to obtain the satisfaction influence value between each user session output question and its remaining user session output questions; R3. Calculate the average of the satisfaction impact values ​​of each user's session output question and the remaining user's session output question to obtain the satisfaction impact value of each user's session output question, and then use the average calculation result of the satisfaction impact value of each user's session output question as the user's digital human session satisfaction impact value, and record it as ; R4. Count users’ satisfaction with digital human conversations , , To set the reference satisfaction impact value, is a natural constant.

8. The psychological companionship digital human system according to claim 7, characterized in that: The method of counting the text similarity between the output questions of each user session and the output questions of each remaining user session includes: The set of session output words for each user session output question is recorded as , and the set of session output words for each remaining user session output problem is recorded as ; Count the text similarities between each user session output question and the remaining user session output questions , .

Images