An AI-based intelligent marketing and identity authentication method and device

Abstract

The present invention relates to the field of intelligent marketing, and in particular to an AI-based intelligent marketing and identity authentication method and device. The present invention verifies the identity of a user terminal, inquires the user terminal that meets the verification standard through a telephone robot, identifies whether the feedback voice meets a predetermined analysis standard based on emotional keywords and tone features in the feedback voice, analyzes only the feedback voice that meets the analysis standard, determines whether the feedback voice meets a predetermined semantic discrete standard, determines a processing method for the feedback voice based on the analysis result of the feedback voice, screens out sample keywords, selects a reply voice and feeds it back to the user terminal, considers the situation of voice ambiguity in the user feedback process through the above process, analyzes the feedback voice fragmentation, considers the differences between voice segments, finds a voice segment with strong representation as a basis for determining the reply voice, and thereby improves the accuracy of the reply voice to the feedback voice under the premise of ensuring the calculation speed.

Description

Technical Field

[0001] The present invention relates to the field of intelligent marketing, and in particular to an AI-based intelligent marketing and identity authentication method and device. Background Art

[0002] With the development of computer technology and Internet technology, intelligent marketing technology has been widely used in various fields. For example, sales platforms, trading platforms and consulting platforms can automatically reply to users' inquiries through predetermined models or programs, send promotional information to users and automatically reply based on users' feedback.

[0003] For example, Chinese patent publication number: CN114860742A discloses an AI customer service interaction method, device, electronic device and storage medium based on artificial intelligence. The AI ​​customer service interaction method based on artificial intelligence includes: retrieving voice communication text to construct an initial question and answer data set; evaluating the initial question and answer data set according to a preset question value index to obtain a question value score set; filtering the initial question and answer data set based on the question value score set to obtain a preferred question data set; using the voice communication text in the preferred question data set as a question template for AI customer service to interact with customers, and obtaining a customer intention classification set based on the interaction results between AI customer service and customers; updating the question template based on the customer intention classification set to optimize the interaction process between AI customer service and customers. This application can quantitatively evaluate and screen problems that arise in telemarketing, effectively improving the marketing effect and communication efficiency of telemarketing.

[0004] However, the prior art still has the following problems:

[0005] The voice data fed back by the user may be ambiguous and have vague semantic expressions due to different personal language habits. Therefore, the reply data to the feedback voice data may be abnormal. Summary of the invention

[0006] To this end, the present invention provides an AI-based intelligent marketing and identity authentication method and device to overcome the problem in the prior art that the voice data fed back by the user end may be ambiguous and have vague semantic expressions due to different personal language habits, and therefore the reply data to the feedback voice data may be abnormal.

[0007] To achieve the above objectives, the present invention provides an AI-based intelligent marketing and identity authentication method, which comprises:

[0008] Step S1, verifying the identity tag of the user terminal, and inquiring the user terminal that meets the verification criteria through the telephone robot, including outputting a predetermined inquiry voice to the user terminal after the call is connected;

[0009] Step S2, obtaining the feedback voice sent by the user end in response to the inquiry voice, extracting emotional keywords and tone features in the feedback voice to identify whether the feedback voice meets the predetermined analysis criteria, wherein the tone features include fundamental frequency and sound intensity

[0010] Step S3, analyzing only the feedback speech that meets the analysis criteria, including calculating the semantic discrete coefficient according to the conditional probability of each keyword in the feedback speech appearing in the corpus database, and determining whether the feedback speech meets the predetermined semantic discrete criteria according to the semantic discrete coefficient;

[0011] Step S4, determining a processing method for the feedback voice according to the analysis result of the feedback voice, wherein:

[0012] If the feedback speech meets the predetermined semantic discreteness standard, the correlation coefficient between each keyword in the feedback speech and the sample keyword is calculated, and the sample keyword corresponding to the maximum correlation coefficient is extracted.

[0013] If the feedback speech does not meet the predetermined semantic discreteness standard, the feedback speech is cut into several speech segments, the representation keywords in the speech segments are identified, the semantic representation coefficient of the speech segments is calculated according to the conditional probability of the representation keywords of each speech segment appearing in the corpus database, the representation speech segment is selected, the correlation coefficient of each keyword in the representation speech segment and the sample keyword is calculated, and the sample keyword corresponding to the maximum correlation coefficient is extracted;

[0014] Step S5, selecting a reply voice according to the extracted sample keywords, and controlling the telephone robot to send the reply voice to the user end.

[0015] Further, the process of identifying whether the feedback speech meets the predetermined analysis criteria includes:

[0016] If there are no predetermined emotional keywords in the feedback speech and the feedback speech meets the predetermined audio qualification conditions, the feedback speech is determined to meet the predetermined analysis criteria, and the audio qualification conditions include that the sound intensity is less than a predetermined sound intensity threshold and / or the fundamental frequency is less than a predetermined fundamental frequency threshold.

[0017] Furthermore, in step S3, keywords in the feedback speech are extracted, and the semantic discrete coefficient of the feedback speech is calculated according to formula (1):

[0018]

[0019] In formula (1), Pe represents the semantic discrete coefficient, Ai represents the first condition, which includes the i-th keyword of the feedback speech appearing in the sample sentence of the corpus database, B represents the second condition, which is the remaining keywords of the feedback speech appearing in the sample sentence of the corpus database, P(Ai|B) represents the probability of the first condition appearing when the second condition is met, and n represents the number of keywords in the feedback speech.

[0020] Furthermore, in step S3, the process of determining whether the feedback speech meets the predetermined semantic discreteness standard includes:

[0021] The semantic discrete coefficient is compared with a preset semantic discrete coefficient comparison threshold,

[0022] If the semantic discreteness coefficient is greater than or equal to the semantic discreteness coefficient comparison threshold, it is determined that the feedback speech meets the semantic discreteness standard;

[0023] If the semantic discrete coefficient is less than the semantic discrete coefficient comparison threshold, it is determined that the feedback speech does not meet the semantic discrete standard.

[0024] Furthermore, in step S3, the process of screening the representative keywords includes:

[0025] The correlation coefficients between the keywords in the voice clips are calculated one by one, and the keyword groups with correlation coefficients greater than a preset correlation coefficient comparison threshold are screened out to obtain a number of representative keywords.

[0026] Furthermore, in step S3, the semantic representation coefficient of each speech segment is calculated according to formula (2):

[0027]

[0028] In formula (2), PL represents the semantic representation coefficient, AL represents the first difference condition, which is the representation keyword of the speech fragment appearing in the sample sentence of the corpus database, BL represents the second difference condition, which is the representation keyword of the remaining speech fragments appearing in the sample sentence of the corpus database, Pl represents the discrete conditional probability of the speech fragment, P(AL|BL) represents the probability of the first difference condition appearing when the second difference condition is satisfied, and P0 represents the average value of the discrete conditional probabilities of each speech fragment.

[0029] Furthermore, the process of selecting a representation speech segment according to the semantic representation coefficient includes:

[0030] The speech segment corresponding to the maximum semantic representation coefficient is selected as the representative speech segment.

[0031] Furthermore, in step S3, the correlation coefficient between the keyword and the sample keyword is calculated according to formula (3):

[0032]

[0033] In formula (3), NY represents the number of sample sentences containing the keyword or / sample keyword in the corpus database, and NM represents the number of sample sentences containing both the keyword and the sample keyword in the corpus database.

[0034] Furthermore, in step S4, the process of selecting a reply voice according to the extracted keywords includes:

[0035] A plurality of sample keywords and sample answering voices associated with the sample keywords are pre-constructed.

[0036] The sample reply voice associated with the extracted sample keyword is selected as the reply voice.

[0037] Furthermore, a device for applying an AI-based intelligent marketing and identity authentication method is provided, comprising:

[0038] one or more processors;

[0039] Memory;

[0040] and one or more programs,

[0041] The one or more programs are configured to be executed by one or more processors, and the memory includes a storage medium storing a computer program, which can be used to execute an AI-based intelligent marketing and identity authentication method when executed by the processor.

[0042] Compared with the prior art, the present invention verifies the identity of the user terminal, inquires the user terminal that meets the verification criteria through a telephone robot, identifies whether the feedback voice meets the predetermined analysis criteria based on the emotional keywords and tone features in the feedback voice, analyzes only the feedback voice that meets the analysis criteria, determines whether the feedback voice meets the predetermined semantic discrete criteria, determines the processing method for the feedback voice based on the analysis result of the feedback voice, screens out sample keywords, subsequently selects a reply voice based on the sample keywords, controls the telephone robot to send the reply voice to the user terminal, and takes into account the situation of voice ambiguity in the user feedback process through the above process, analyzes the feedback voice fragmentation, considers the differences between the voice segments, finds out the voice segments with strong representation as the basis for determining the reply voice, and thereby improves the accuracy of the reply voice to the feedback voice while ensuring the calculation speed.

[0043] In particular, the present invention only inquires the user terminals that meet the verification standards, thereby improving the inquiry efficiency, and identifies whether the feedback voice meets the predetermined analysis standards based on the emotional keywords and tone features in the feedback voice. In actual situations, the user may be annoyed during the process of inquiring the user terminal, which can be identified by the emotional keywords and specific tone features mixed in the feedback voice. Then, only the feedback voice that meets the analysis standards is subsequently analyzed, and the reply voice is further determined to be fed back to the user terminal, thereby improving the user experience, reducing the waste of computing power, and ensuring the accuracy of the determined reply voice.

[0044] In particular, the present invention considers calculating the semantic discrete coefficient and determining whether the feedback voice meets the predetermined semantic discrete standard. In actual situations, the feedback voice fed back by the user may be semantically ambiguous due to different personal language habits, and the correlation between the keywords is poor, making it difficult to identify the true meaning expressed. Therefore, the present invention calculates the semantic discrete coefficient through the conditional probability of each keyword appearing in the corpus database. The higher the semantic discrete coefficient, the stronger the correlation between the keywords, and the clearer the intention expressed by the feedback voice, which is convenient for accurately determining the reply voice. On the contrary, if the voice discrete coefficient is lower, the intention expressed by the feedback voice is vague, the semantics is relatively vague, and it is difficult to determine the reply voice. The present invention determines whether the feedback voice meets the predetermined voice discrete standard, provides data support for subsequent targeted processing of the feedback voice, thereby reducing the waste of computing power and ensuring the accuracy of the determined reply voice.

[0045] In particular, the present invention fragments the feedback semantics that do not meet the semantic discreteness standard, identifies the characterization keywords in the voice fragments, and the characterization keywords are the characterization keywords with higher correlation coefficients, thereby screening out certain interference data, and, based on the characterization keywords, calculating the semantic characterization coefficients between the characterization voice fragments. The higher the semantic characterization coefficient, the higher the overall correlation between the single voice fragment and the remaining voice fragments, thereby screening out the characterization voice fragments, and subsequently determining the reply voice, thereby reducing the waste of computing power and ensuring the accuracy of the determined reply voice. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A schematic diagram of the steps of an AI-based intelligent marketing and identity authentication method according to an embodiment of the invention;

[0047] Figure 2 A schematic diagram of the decision logic for the semantic discreteness standard of an embodiment of the invention;

[0048] Figure 3 A logical diagram of determining a processing method for feedback speech according to an embodiment of the present invention;

[0049] Figure 4A logical schematic diagram is selected for the keywords representing the embodiments of the invention. DETAILED DESCRIPTION

[0050] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0051] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0052] See also Figures 1 to 4 As shown, Figure 1 This is a schematic diagram of the steps of the AI-based intelligent marketing and identity authentication method according to an embodiment of the invention. Figure 2 This is a schematic diagram of the decision logic for the semantic discreteness standard of an embodiment of the invention. Figure 3 A logical diagram of determining a processing method for feedback speech according to an embodiment of the present invention, Figure 4 The present invention is a logic diagram for selecting keywords representing an embodiment of the invention. The AI-based intelligent marketing and identity authentication method of the present invention includes:

[0053] Step S1, verifying the identity tag of the user terminal, and inquiring the user terminal that meets the verification criteria through the telephone robot, including outputting a predetermined inquiry voice to the user terminal after the call is connected;

[0054] Step S2, obtaining the feedback voice sent by the user end in response to the inquiry voice, extracting emotional keywords and tone features in the feedback voice to identify whether the feedback voice meets the predetermined analysis criteria, wherein the tone features include fundamental frequency and sound intensity

[0055] Step S3, analyzing only the feedback speech that meets the analysis criteria, including calculating the semantic discrete coefficient according to the conditional probability of each keyword in the feedback speech appearing in the corpus database, and determining whether the feedback speech meets the predetermined semantic discrete criteria according to the semantic discrete coefficient;

[0056] Step S4, determining a processing method for the feedback voice according to the analysis result of the feedback voice, wherein:

[0057] If the feedback speech meets the predetermined semantic discreteness standard, the correlation coefficient between each keyword in the feedback speech and the sample keyword is calculated, and the sample keyword corresponding to the maximum correlation coefficient is extracted.

[0058] If the feedback speech does not meet the predetermined semantic discreteness standard, the feedback speech is cut into several speech segments, the representation keywords in the speech segments are identified, the semantic representation coefficient of the speech segments is calculated according to the conditional probability of the representation keywords of each speech segment appearing in the corpus database, the representation speech segment is selected, the correlation coefficient of each keyword in the representation speech segment and the sample keyword is calculated, and the sample keyword corresponding to the maximum correlation coefficient is extracted;

[0059] Step S5, selecting a reply voice according to the extracted sample keywords, and controlling the telephone robot to send the reply voice to the user end.

[0060] Specifically, the present invention does not specifically limit the construction method of the corpus database. The corpus database can be constructed in advance by collecting a number of speech data as samples, or an existing open source corpus database can be used. This is a prior art and will not be described in detail.

[0061] Specifically, the present invention does not limit the setting method of emotional keywords. In this embodiment, emotional keywords can include keywords used to represent anger and boredom, which can be pre-set or extracted by analyzing the corpus through a natural speech model in advance. For example, the long short-term memory network LSTM can capture the temporal information in the text and understand the emotional meaning of the text. The LSTM model can be used to classify the text and extract emotional keywords. Of course, other methods can also be used. This is a prior art and will not be repeated.

[0062] Specifically, the present invention does not limit the specific form of the telephone robot. It only needs to be able to make calls to the user end and answer calls from the user end. This is existing technology and will not be described in detail.

[0063] Specifically, in step S1, the process of determining whether the user terminal meets the verification standard includes:

[0064] The user tags associated with each user terminal are called one by one in the tag database for determination. If there is a preset qualified tag among the user tags associated with the user terminal, it is determined that the user terminal meets the verification standard.

[0065] Specifically, the association relationship between user tags and user terminals can be pre-built, and a single user terminal can be associated with multiple tags. In some possible implementations, user tags include three types: high-intention tags, medium-intention tags, and low-intention tags. The medium-intention tag is set as a qualified tag. If a medium-intention tag exists among the tags associated with the user terminal, the user terminal is determined to meet the verification criteria.

[0066] Specifically, the process of identifying whether the feedback speech meets the predetermined analysis criteria includes:

[0067] If there are no predetermined emotional keywords in the feedback speech and the feedback speech meets the predetermined audio qualification conditions, the feedback speech is determined to meet the predetermined analysis criteria, and the audio qualification conditions include that the sound intensity is less than a predetermined sound intensity threshold and / or the fundamental frequency is less than a predetermined fundamental frequency threshold.

[0068] Specifically, in this embodiment, the fundamental frequency threshold Z0 and the sound intensity threshold R0 are pre-calculated, and several sample speech of anger emotion and sample speech of annoyance emotion are obtained, the pitch features are extracted, the fundamental frequency average value ΔZ and the sound intensity average value ΔR are calculated, and Z0=ΔZ×0.75 and R0=ΔZ×0.85 are set.

[0069] The present invention only inquires the user terminal that meets the verification standard, thereby improving the inquiry efficiency, and identifies whether the feedback voice meets the predetermined analysis standard based on the emotional keywords and tone features in the feedback voice. In actual situations, the user may be annoyed during the process of inquiring the user terminal, which can be identified by the emotional keywords and specific tone features mixed in the feedback voice. Then, only the feedback voice that meets the analysis standard is subsequently analyzed, and the reply voice is further determined to be fed back to the user terminal, thereby improving the user experience, reducing the waste of computing power, and ensuring the accuracy of the determined reply voice.

[0070] Specifically, in step S3, keywords in the feedback speech are extracted, and the semantic discrete coefficient of the feedback speech is calculated according to formula (1):

[0071]

[0072] In formula (1), Pe represents the semantic discrete coefficient, Ai represents the first condition, which includes the i-th keyword of the feedback speech appearing in the sample sentence of the corpus database, B represents the second condition, which is the remaining keywords of the feedback speech appearing in the sample sentence of the corpus database, P(Ai|B) represents the probability of the first condition appearing when the second condition is met, and n represents the number of keywords in the feedback speech.

[0073] Specifically, in step S3, the process of determining whether the feedback speech meets the predetermined semantic discreteness standard includes:

[0074] The semantic discrete coefficient is compared with a preset semantic discrete coefficient comparison threshold,

[0075] If the semantic discreteness coefficient is greater than or equal to the semantic discreteness coefficient comparison threshold, it is determined that the feedback speech meets the semantic discreteness standard;

[0076] If the semantic discrete coefficient is less than the semantic discrete coefficient comparison threshold, it is determined that the feedback speech does not meet the semantic discrete standard.

[0077] Specifically, the semantic discrete coefficient comparison threshold Pe0 is pre-set, wherein a number of feedback speech are randomly extracted in advance, and the semantic discrete coefficient average value ΔPe of the feedback speech is calculated, and Pe0=α×ΔPe is set, wherein α represents the precision coefficient, 0.5<α<0.75.

[0078] The present invention considers calculating the semantic discrete coefficient and determining whether the feedback voice meets the predetermined semantic discrete standard. In actual situations, the feedback voice fed back by the user end may be semantically ambiguous due to different personal language habits, and the correlation between the keywords is poor, making it difficult to identify the true meaning expressed. Therefore, the present invention calculates the semantic discrete coefficient through the conditional probability of each keyword appearing in the corpus database. The higher the semantic discrete coefficient, the stronger the correlation between the keywords, and the clearer the intention expressed by the feedback voice, which is convenient for accurately determining the reply voice. On the contrary, if the voice discrete coefficient is lower, the intention expressed by the feedback voice is vague, the semantics is relatively vague, and it is difficult to determine the reply voice. The present invention determines whether the feedback voice meets the predetermined voice discrete standard, provides data support for subsequent targeted processing of the feedback voice, thereby reducing the waste of computing power and ensuring the accuracy of the determined reply voice.

[0079] Specifically, in step S3, the process of screening the representative keywords includes:

[0080] The correlation coefficients between the keywords in the voice clips are calculated one by one, and the keyword groups with correlation coefficients greater than a preset correlation coefficient comparison threshold are screened out to obtain a number of representative keywords.

[0081] The correlation coefficient between keywords is the ratio of the number of sample sentences in which two keywords exist at the same time in the corpus database to the number of sample sentences in which any one of the keywords exists in the corpus database.

[0082] The correlation coefficient comparison threshold D0 is preset, and the average correlation coefficient ΔD of each keyword is calculated, and D0=ΔD×g is set, where g represents the offset coefficient, 1.15<g<1.3.

[0083] Specifically, in step S3, the semantic representation coefficient of each speech segment is calculated according to formula (2):

[0084]

[0085] In formula (2), PL represents the semantic representation coefficient, AL represents the first difference condition, which is the representation keyword of the speech fragment appearing in the sample sentence of the corpus database, BL represents the second difference condition, which is the representation keyword of the remaining speech fragments appearing in the sample sentence of the corpus database, Pl represents the discrete conditional probability of the speech fragment, P(AL|BL) represents the probability of the first difference condition appearing when the second difference condition is satisfied, and P0 represents the average value of the discrete conditional probabilities of each speech fragment.

[0086] Specifically, the process of selecting a speech segment to represent the speech segment according to the semantic representation coefficient includes:

[0087] The speech segment corresponding to the maximum semantic representation coefficient is selected as the representative speech segment.

[0088] Specifically, in step S3, the correlation coefficient between the keyword and the sample keyword is calculated according to formula (3):

[0089]

[0090] In formula (3), NY represents the number of sample sentences containing the keyword or / sample keyword in the corpus database, and NM represents the number of sample sentences containing both the keyword and the sample keyword in the corpus database.

[0091] The present invention performs fragmentation processing on feedback semantics that do not meet the semantic discreteness standard, identifies characterization keywords in voice segments, and characterization keywords are characterization keywords with higher correlation coefficients, thereby screening out certain interference data, and, based on the characterization keywords, calculates semantic characterization coefficients between characterization voice segments. The higher the semantic characterization coefficient, the higher the overall correlation between a single voice segment and the remaining voice segments, thereby screening out the characterization voice segments, and subsequently determining the reply voice, thereby reducing the waste of computing power and ensuring the accuracy of the determined reply voice.

[0092] Specifically, in step S4, the process of selecting a reply voice according to the extracted keywords includes:

[0093] A plurality of sample keywords and sample answering voices associated with the sample keywords are pre-constructed.

[0094] The sample reply voice associated with the extracted sample keyword is selected as the reply voice.

[0095] Specifically, the specific setting method of the sample keywords and the sample reply voice is not limited, and those skilled in the art can set them according to the application field, for example,

[0096] Taking the bank marketing platform as an example, the sample keyword is set as loan, and the sample reply voice is the voice explaining the loan business processing procedures.

[0097] Specifically, a device for applying an AI-based intelligent marketing and identity authentication method is also provided, comprising:

[0098] one or more processors;

[0099] Memory;

[0100] and one or more programs,

[0101] The one or more programs are configured to be executed by one or more processors, and the memory includes a storage medium storing a computer program, which can be used to execute an AI-based intelligent marketing and identity authentication method when executed by the processor.

[0102] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. An AI-based intelligent marketing and identity authentication method, characterized in that: include: Step S1, verifying the identity tag of the user terminal, and inquiring the user terminal that meets the verification criteria through the telephone robot, including outputting a predetermined inquiry voice to the user terminal after the call is connected; Step S2, obtaining the feedback voice sent by the user end in response to the inquiry voice, extracting emotional keywords and tone features in the feedback voice to identify whether the feedback voice meets the predetermined analysis criteria, wherein the tone features include fundamental frequency and sound intensity Step S3, analyzing only the feedback speech that meets the analysis criteria, including calculating the semantic discrete coefficient according to the conditional probability of each keyword in the feedback speech appearing in the corpus database, and determining whether the feedback speech meets the predetermined semantic discrete criteria according to the semantic discrete coefficient; Step S4, determining a processing method for the feedback voice according to the analysis result of the feedback voice, wherein: If the feedback speech meets the predetermined semantic discreteness standard, the correlation coefficient between each keyword in the feedback speech and the sample keyword is calculated, and the sample keyword corresponding to the maximum correlation coefficient is extracted. If the feedback speech does not meet the predetermined semantic discreteness standard, the feedback speech is cut into several speech segments, the representation keywords in the speech segments are identified, the semantic representation coefficient of the speech segments is calculated according to the conditional probability of the representation keywords of each speech segment appearing in the corpus database, the representation speech segment is selected, the correlation coefficient of each keyword in the representation speech segment and the sample keyword is calculated, and the sample keyword corresponding to the maximum correlation coefficient is extracted; Step S5, selecting a reply voice according to the extracted sample keywords, and controlling the telephone robot to send the reply voice to the user end.

2. The AI-based intelligent marketing and identity authentication method according to claim 1, characterized in that: The process of identifying whether the feedback speech meets the predetermined analysis criteria includes: If there are no predetermined emotional keywords in the feedback speech and the feedback speech meets the predetermined audio qualification conditions, the feedback speech is determined to meet the predetermined analysis criteria, and the audio qualification conditions include that the sound intensity is less than a predetermined sound intensity threshold and / or the fundamental frequency is less than a predetermined fundamental frequency threshold.

3. The AI-based intelligent marketing and identity authentication method according to claim 1, characterized in that: In step S3, keywords in the feedback speech are extracted, and the semantic discrete coefficient of the feedback speech is calculated according to formula (1). In formula (1), Pe represents the semantic discrete coefficient, Ai represents the first condition, which is the i-th keyword of the feedback speech appearing in the sample sentence of the corpus database, B represents the second condition, which is the remaining keywords of the feedback speech appearing in the sample sentence of the corpus database, P(Ai|B) represents the probability of the first condition appearing when the second condition is met, and n represents the number of keywords in the feedback speech.

4. The AI-based intelligent marketing and identity authentication method according to claim 1, characterized in that: In step S3, the process of determining whether the feedback speech meets the predetermined semantic discreteness standard includes: The semantic discrete coefficient is compared with a preset semantic discrete coefficient comparison threshold, If the semantic discreteness coefficient is greater than or equal to the semantic discreteness coefficient comparison threshold, it is determined that the feedback speech meets the semantic discreteness standard; If the semantic discrete coefficient is less than the semantic discrete coefficient comparison threshold, it is determined that the feedback speech does not meet the semantic discrete standard.

5. The AI-based intelligent marketing and identity authentication method according to claim 1, characterized in that: In step S3, the process of screening the representative keywords includes: The correlation coefficients between the keywords in the voice clips are calculated one by one, and the keyword groups with correlation coefficients greater than a preset correlation coefficient comparison threshold are screened out to obtain a number of representative keywords.

6. The AI-based intelligent marketing and identity authentication method according to claim 1, characterized in that: In step S3, the semantic representation coefficient of each speech segment is calculated according to formula (2): In formula (2), PL represents the semantic representation coefficient, AL represents the first difference condition, which is the representation keyword of the speech fragment appearing in the sample sentence of the corpus database, BL represents the second difference condition, which is the representation keyword of the remaining speech fragments appearing in the sample sentence of the corpus database, Pl represents the discrete conditional probability of the speech fragment, P(AL|BL) represents the probability of the first difference condition appearing when the second difference condition is satisfied, and P0 represents the average value of the discrete conditional probabilities of each speech fragment.

7. The AI-based intelligent marketing and identity authentication method according to claim 1, characterized in that: The process of selecting a speech segment to represent the semantic representation coefficient includes: The speech segment corresponding to the maximum semantic representation coefficient is selected as the representative speech segment.

8. The AI-based intelligent marketing and identity authentication method according to claim 1, characterized in that: In step S3, the correlation coefficient between the keyword and the sample keyword is calculated according to formula (3): In formula (3), NY represents the number of sample sentences containing the keyword or / sample keyword in the corpus database, and NM represents the number of sample sentences containing both the keyword and the sample keyword in the corpus database.

9. The AI-based intelligent marketing and identity authentication method according to claim 1, characterized in that: In step S4, the process of selecting a reply voice according to the extracted keywords includes: A plurality of sample keywords and sample answering voices associated with the sample keywords are pre-constructed. The sample reply voice associated with the extracted sample keyword is selected as the reply voice.

10. A device applied to the method according to any one of claims 1 to 9, characterized in that: include: one or more processors; Memory; and one or more programs, The one or more programs are configured to be executed by one or more processors, and the memory includes a storage medium storing a computer program, which can be used to execute an AI-based intelligent marketing and identity authentication method when executed by the processor.

Images