Intelligent interaction system based on customer business data analysis

Abstract

The invention relates to the technical field of intelligent interaction, in particular to an intelligent interaction system based on customer business data analysis. The system comprises a voice authentication unit, an interaction control unit, a log query unit and an analysis callout unit. According to the invention, when a voice authentication unit wakes up an instruction to log in a service query platform and a log query unit is triggered to identify that a current login user has uncompleted query, a signal connection analysis calling unit directly calls client service data at a position corresponding to a timestamp of last query; therefore, the time for the user to search and input again is greatly shortened, the continuity and convenience of user experience are improved, the user does not need to repeatedly input the content inquired before, the system automatically identifies and recovers to the state when the user quits last time, and the operation efficiency of the user is improved; and the operation process is more automatic, the manual operation of the user is reduced, and the intelligence and the automation degree of the system are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent interaction technology, and in particular to an intelligent interaction system based on customer business data analysis. Background Art

[0002] Enterprises have accumulated a large amount of customer business data, such as sales records, customer behavior data, customer feedback, etc. Through in-depth analysis of these data, valuable insights can be mined to provide decision-making support for enterprises;

[0003] For customer business data analysis and query in existing technologies, all-round customer business data including sales records, customer behavior, customer feedback, etc. are collected from various business systems of the enterprise, and the collected data is cleaned, standardized and integrated to build a unified customer business data warehouse as an intelligent interactive platform to analyze data, and output the analysis results to users in the form of charts, reports, etc., to help enterprises better understand customer needs and optimize business decisions;

[0004] However, when querying customer business data, the queryer needs to log in to the intelligent interactive platform first, and then enter the content to be queried in the platform to call up the analysis results. The operation is cumbersome and the intelligence is limited. In addition, the content to be queried needs to be entered again after each re-login, which is not convenient for quickly typing out the data that has not been viewed before. In view of this, we propose an intelligent interactive system based on customer business data analysis. Summary of the invention

[0005] The purpose of the present invention is to provide an intelligent interactive system based on customer business data analysis, which logs in to the business query platform by recognizing the wake-up word features and voiceprint features of voice commands, and directly calls out the query progress when the current user has unfinished business data. When the current business data is completed, other business data are called out through voice control, thereby reducing the tediousness of entering query content and login information multiple times.

[0006] To achieve the above purpose, it includes a voice authentication unit, an interaction control unit, an analysis and call-out unit, and a log query unit;

[0007] The voice authentication unit is used to identify voice commands, which include wake-up commands and control commands. The wake-up commands are used to start the service query platform and detect the current voiceprint to log in to the service query platform. The control commands are used to store voice signals after logging in.

[0008] As a further improvement of the technical solution, the voice authentication unit includes a platform wake-up module and a voiceprint authentication module;

[0009] The platform wake-up module is used to capture the user's audio input in real time, pre-process and extract features of the real-time audio, including wake-up word features and voiceprint features, generate Mel-spectrograms or MFCC features, input the extracted features into the pre-trained speech recognition model, and the model outputs the probability distribution of the wake-up word. According to the output probability distribution, if the probability of the wake-up word exceeds the set threshold, it is determined to be a wake-up word, triggering the start of the business query platform;

[0010] The voiceprint authentication module adds a voiceprint recognition function on the basis of the pre-trained speech recognition model to extract the user's voiceprint features and compares them with the voiceprint samples stored in the database to ensure that only authorized users can log in to the business query platform.

[0011] This is one of the purposes of the present invention. When the user issues a wake-up word, the voice authentication unit not only starts the business query platform, but also automatically logs in to the business platform according to the voiceprint features, freeing up hands and being highly intelligent.

[0012] The interactive control unit is used to identify the control instructions in the voice authentication unit in combination with the voice recognition technology, and to understand the user's query intention using the natural language processing technology;

[0013] The analysis and calling unit is used to execute a command to call out customer business data according to the query intention of the interaction control unit;

[0014] The log query unit is used to record the initial time when the analysis and call-out unit calls out the customer business data to the end time when the query page is closed to determine the query status, and persistently store the query results in the database, the status includes the query progress and the timestamp of the last query;

[0015] As a further improvement of the technical solution, the log query unit includes a status recording module and a progress query module;

[0016] The state recording module is connected to the output end of the analysis and call-out unit, and is used to sense when the analysis and call-out unit executes a call-out command, and obtain the time mark as the initial time, and when the analysis and call-out unit executes a switching command, the time mark is obtained as the termination time;

[0017] The progress query module is used to establish a time threshold based on the historical behavior and feedback of viewing business data.

[0018] The second object of the present invention is that when the voice authentication unit wakes up the command to log in to the business query platform, the trigger log query unit identifies that the current logged-in user has an unfinished query, and the signal connects the analysis and retrieval unit to directly retrieve the customer business data corresponding to the timestamp of the last query;

[0019] When the log query unit identifies that the currently logged-in user has completed the query, the interactive control unit identifies the query intention of the control instruction in the voice authentication unit and drives the analysis and calling unit to execute the control instruction.

[0020] Preferably, the log query unit further includes an interested analysis module;

[0021] The interest analysis module is used to analyze the behavior data in the business data sample of the progress query module, and the behavior data includes the access frequency and the residence time, wherein the residence time is derived from the sum of the difference between the termination time and the initial time obtained by the state recording module in the business data sample, and the comprehensive weight i=α*+β* is determined by comprehensively considering the behavior data, wherein α and β are coefficients for adjusting the weights of different indicators, satisfying α+β=1:

[0022] Interest = Σ(interest of content i

[0023] *Comprehensive weight i) / Σ(weight of all contents)

[0024] According to the order of interest, after the voiceprint authentication module outputs the login signal, the content with high interest is displayed preferentially through a small window.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] 1. In this intelligent interactive system based on customer business data analysis, the business query platform is triggered by capturing the user's audio input in real time; then, the pre-trained speech recognition model extracts the voiceprint features, and the user's identity is determined by calculating the similarity between the voiceprint features and the voiceprint samples. Based on the comparison results, it is determined whether the user is an authorized user. If the similarity is higher than the set threshold, the user is determined to be an authorized user and is allowed to log in, which enhances the security of the system and the user experience.

[0027] 2. In the intelligent interactive system based on customer business data analysis, when the voice authentication unit wakes up the command to log in to the business query platform and the log query unit is triggered to identify that the current logged-in user has an unfinished query, the signal connection analysis and call-out unit directly calls out the customer business data corresponding to the timestamp of the last query, which greatly reduces the time for users to re-search and input, improves the continuity and convenience of user experience, and users do not need to repeatedly input the content they have queried before. The system automatically identifies and restores to the state when they last exited, improving the efficiency of user operations;

[0028] When the log query unit recognizes that the currently logged-in user has completed the query, the interactive control unit recognizes the query intention of the control instruction in the voice authentication unit, and drives the analysis and call-out unit to execute the control instruction. The user can quickly query and call out data through voice commands, which further improves the convenience and efficiency of the operation, and makes the operation process more automated, reduces the user's manual operations, and improves the intelligence and automation of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is the overall structural principle block diagram of the present invention;

[0030] Figure 2 This is a schematic diagram of the pre-trained speech recognition model of the present invention.

[0031] The meaning of each number in the figure is:

[0032] 100, voice authentication unit; 110, platform wake-up module; 120, voiceprint authentication module;

[0033] 200, interactive control unit; 210, text recognition module; 220, word segmentation determination module;

[0034] 300, analysis call-out unit;

[0035] 400, log query unit; 410, status recording module; 420, progress query module; 430, interest analysis module. DETAILED DESCRIPTION

[0036] The following will be combined with the accompanying drawings in 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.

[0037] The terms used in the exemplary embodiments shown in the drawings are not intended to limit the present invention. In the drawings, the same units / elements are given the same reference numerals.

[0038] Unless otherwise specified, the terms (including technical terms) used herein have the commonly understood meanings to those skilled in the art. In addition, it is understood that the terms defined in commonly used dictionaries should be understood to have the same meanings as those in the context of the relevant fields, and should not be understood as idealized or overly formal meanings.

[0039] The following are some definitions of terms:

[0040] The business query platform is a platform for analyzing and storing customer business data. Users can log in to the business query platform at any time to query customer data. The queried customer data includes customer order information, order date and order amount;

[0041] Wake-up words refer to specific words or phrases used to start or activate the business query platform in the voice recognition system, such as "**, hello" and "**, are you there?"

[0042] When querying customer business data, the queryer needs to log in to the intelligent interactive platform and enter the content to be queried in the platform to call up the analysis results. During the user query process, it is usually necessary to record the data, resulting in both hands being occupied, and it is inconvenient to enter the login key and query content. The intelligence is limited. For this reason, the present invention provides an intelligent interactive system based on customer business data analysis, please refer to Figure 1-Figure 2 As shown, it includes a voice authentication unit 100, an interaction control unit 200, an analysis and call-out unit 300 and a log query unit 400;

[0043] The voice authentication unit 100 is used to identify voice commands, which include wake-up commands and control commands. The wake-up commands are used to start the service query platform and detect the current voiceprint to log in to the service query platform. The control commands are used to store voice signals after logging in.

[0044] Preferably, the voice authentication unit 100 includes a platform wake-up module 110 and a voiceprint authentication module 120;

[0045] First, collect a large amount of audio data containing wake-up words. These audio data should contain recordings of different groups of people to cover diverse accents and voice features. Use audio processing tools (such as Librosa or FFmpeg) to denoise the audio data, segment long audio files into short audio clips, each of which contains a complete wake-up word, extract audio features such as Mel-spectrograms and Mel-frequency cepstral coefficients, which are important representations of audio signals, and select appropriate deep learning models, such as convolutional neural networks, long short-term memory networks, or Transformer models, for model training:

[0046] Input layer: The input layer accepts preprocessed audio features (such as MFCC);

[0047] Convolutional layer: The convolutional layer extracts the spatiotemporal features in the audio signal. The convolutional layer can identify local patterns in the audio (such as the features of the wake-up word and the voiceprint features).

[0048] Pooling layer: The output of the convolutional layer is reduced in dimension through the pooling layer (such as maximum pooling or average pooling), retaining important features and reducing computational complexity;

[0049] Connection layer: passes the output of the convolutional layer and the pooling layer to the fully connected layer to further extract high-level features;

[0050] Output layer: The output layer uses the Softmax activation function to convert the network output into the probability distribution of the wake-up word;

[0051] Deploy the trained model to the business query platform, usually using TensorFlow Serving, ONNXRuntime or other inference engines for deployment, and establish a pre-trained speech recognition model on the business query platform, so that the business query platform uses the platform wake-up module 110 to capture the user's audio input in real time, pre-process and extract features of the real-time audio, the features include wake-up word features and voiceprint features, generate Mel spectrum or MFCC features, input the extracted features into the pre-trained speech recognition model, and the model outputs the probability distribution of the wake-up word. According to the output probability distribution, if the probability of the wake-up word exceeds the set threshold, it is determined to be the wake-up word, triggering the start of the business query platform;

[0052] Then, based on the pre-trained speech recognition model, the voiceprint authentication module 120 adds a voiceprint recognition function to extract the user's voiceprint features, and compares them with the voiceprint samples stored in the database to ensure that only authorized users can log in to the business query platform. Specifically, the pre-trained speech recognition model of the platform wake-up module 110 extracts voiceprint features, and the user identity is determined by calculating the similarity between the voiceprint features and the voiceprint samples. Based on the comparison results, it is determined whether the user is an authorized user. If the similarity is higher than the set threshold, the user is judged to be an authorized user and is allowed to log in, thereby enhancing the security of the system and user experience.

[0053] Furthermore, the interactive control unit 200 is used to recognize the control instructions in the voice authentication unit 100 in combination with the voice recognition technology, and to understand the user's query intention using the natural language processing technology;

[0054] The interactive control unit 200 includes a text recognition module 210 and a word segmentation determination module 220;

[0055] The text recognition module 210 is used to pre-process the input voice signal through voice recognition technology, such as noise reduction and normalization, after sensing the login signal from the voiceprint authentication module 120, extract the features of the voice signal, such as MFCC, STFT, etc., map the voice features to phonemes, convert the phoneme sequence into a word sequence according to the grammar and semantics of the language, and decode the most likely text sequence by combining the acoustic model and the language model. The word segmentation determination module 220 is used to use the natural language processing technology to segment and remove stop words on the identified text sequence, and use the naive Bayes classifier to identify the user's query intention, which is expressed as:

[0056]

[0057] Wherein, c is the intent category, d is the input text, P(c) is the category prior probability, P(c|d) is the probability that the text belongs to a certain intent category, P(d|c) is the conditional probability of the text under the category, and P(d) is the overall probability of the text d appearing, which can be used as a normalization constant to further classify the text sequence and transmit it to the analysis and call-out unit 300 to execute the query according to the recognized query intent;

[0058] The analysis and calling unit 300 is used to execute the command of calling out the customer service data and the switching command according to the query intention of the interaction control unit 200;

[0059] Specifically, the analysis and calling unit 300 adopts a SQL query algorithm, including the following steps:

[0060] A query data set is established by writing the query intent corresponding to the SQL query command according to the screening conditions. The screening conditions include customer order information, order date and order amount. The query intent of the word segmentation determination module 220 is input into the query data set, the SQL query command is called out, and the SQL query is executed. For example, ①, the query statement is "query all orders within a specific ab time period". At this time, the screening condition is the ab time period, and all order information within the ab time period is called out; ②, the query statement is "query orders for customer A". At this time, the screening condition is customer A, and all order information for customer A is called out.

[0061] Then use tools such as Matplotlib and Seaborn to draw charts, and use report generation tools such as Tableau, Power BI or programming languages ​​to generate reports;

[0062] When the text recognition module 210 recognizes the voice input again and the query intent recognized by the word segmentation determination module 220 is transmitted to the analysis and call-out unit 300 for query execution, the analysis and call-out unit 300 first executes the switching command, that is, closes the previous page and re-executes the command to call out the customer business data.

[0063] Embodiment 2:

[0064] After the user accidentally exits the query of business data due to other work, after the user logs in to the business query platform through voice, it is necessary to re-enter the content to be queried through the interactive control unit 200 to input the query command in the business query platform through voice, which is not convenient to quickly type out the data that has not been viewed before. Therefore, the log query unit 400 is used to record the initial time when the analysis and call-out unit 300 calls out the customer business data to the end time of closing the query page to determine the query status, and store the query results persistently in the database, and the status includes the query progress and the timestamp of the last query;

[0065] When the voice authentication unit 100 wakes up the command to log in to the business query platform, and the log query unit 400 is triggered to identify that the currently logged-in user has an unfinished query, the signal connection analysis and calling unit 300 directly calls out the customer business data corresponding to the timestamp of the last query, which greatly reduces the time for users to re-search and input, improves the continuity and convenience of user experience, and users do not need to repeatedly input the content they have queried before. The system automatically recognizes and restores to the state when they last exited, thereby improving the efficiency of user operations;

[0066] When the log query unit 400 identifies that the currently logged-in user has completed the query, the interactive control unit 200 identifies the query intention of the control instruction in the voice authentication unit 100, and drives the analysis and call-out unit 300 to execute the control instruction. The user can quickly query and call out data through voice commands, which further improves the convenience and efficiency of the operation, and makes the operation process more automated, reduces the user's manual operation, and improves the intelligence and automation of the system;

[0067] Preferably, the log query unit 400 includes a status recording module 410 and a progress query module 420;

[0068] The state recording module 410 is connected to the output end of the analysis call-out unit 300, and is used to sense when the analysis call-out unit 300 executes the call-out command, obtain the time mark as the initial time, use JavaScript to implement monitoring in the business query platform, and monitor when the analysis call-out unit 300 executes the switching command, and obtain the time mark as the termination time;

[0069] The progress query module 420 is used to establish a time threshold based on the historical behavior and feedback of viewing business data.

[0070]

[0071] The progress query module 420 establishes a time threshold through statistical analysis and uses a confidence interval to determine a threshold range, including the following steps:

[0072] Collect samples of business data viewed by n users in the historical records, calculate the mean T and standard deviation σ of the historical query time data, according to the algorithm formula:

[0073]

[0074] Among them, z is the z value corresponding to the confidence level. For example, at a 95% confidence level, the z value is 1.96. Since the number of users is constantly increasing, the time threshold will change as the value of n continues to increase, thereby improving the accuracy of determining the time threshold.

[0075] Since each user may have different interests when viewing customer service data, in order to automatically recommend the content of interest to the user when logging into the service query platform, the log query unit 400 further includes an interest analysis module 430;

[0076] The interest analysis module 430 is used to analyze the behavior data in the business data sample of the progress query module 420. The behavior data includes access frequency and residence time, wherein the residence time is derived from the sum of the difference between the termination time and the initial time obtained by the state recording module 410 in the business data sample. The behavior data is comprehensively considered to determine the comprehensive weight i=α*user's residence time on content i / the total residence time of the user's access to all content+β*user's access frequency to content i / the total access frequency of the user's access to all content, wherein α and β are coefficients for adjusting the weights of different indicators, satisfying α+β=1:

[0077] Interest = Σ(interest of content i

[0078] *Comprehensive weight i) / Σ(weight of all contents)

[0079] According to the ranking of interest, after the voiceprint authentication module 120 outputs the login signal, the content with high interest is displayed preferentially in a small window;

[0080] For example, a user accesses the following business data content: Content 1 (stay time is 30 minutes, the number of visits is 5 times); Content 2 (stay time is 20 minutes, the number of visits is 10 times); Content 3 (stay time is 50 minutes, the number of visits is 2 times), then:

[0081] Weight 1 = 30 / (30+20+50) = 0.3

[0082] Weight 2 = 20 / (30+20+50) = 0.2

[0083] Weight three = 50 / (30+20+50) = 0.5, which is conducive to recommending content with a higher weighted average interest to users. The calculated weighted average interest is sorted from high to low, and the small window recommendation rules are set: the number of contents displayed in the small window and the display rules are determined. For example, the small window can be set to display a maximum of three recommended contents, and sort them according to the user's interest, giving priority to displaying content with high interest. After the user logs in, after the login signal is output, the content is recommended to the user through the small window according to the sorted content list.

[0084] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. Intelligent interactive system based on customer business data analysis, characterized by: It comprises a voice authentication unit (100), an interaction control unit (200), an analysis and call-out unit (300) and a log query unit (400); The voice authentication unit (100) is used to identify voice instructions, which include wake-up instructions and control instructions. The wake-up instructions are used to start the business query platform and detect the current voiceprint to log in to the business query platform. The control instructions are used to store the voice signal after logging in. The interactive control unit (200) is used to identify control instructions in the voice authentication unit (100) by combining voice recognition technology, and to understand the user's query intention by using natural language processing technology; The analysis and calling unit (300) is used to execute a command to call out customer business data according to the query intention of the interaction control unit (200); The log query unit (400) is used to record the initial time when the analysis and call-out unit (300) calls out the customer business data to the end time when the query page is closed to determine the query status, and to persistently store the query results in the database, the status including the query progress and the timestamp of the last query; When a voice authentication unit (100) wakes up a command to log into the service query platform, triggering the log query unit (400) to identify that the currently logged-in user has an unfinished query, a signal is connected to the analysis and retrieval unit (300) to directly retrieve the customer service data at the location corresponding to the timestamp of the last query; When the log query unit (400) identifies that the currently logged-in user has completed the query, the interactive control unit (200) identifies the query intention of the control instruction in the voice authentication unit (100), and drives the analysis and calling unit (300) to execute the control instruction.

2. The intelligent interactive system based on customer business data analysis according to claim 1, characterized in that: The voice authentication unit (100) comprises a platform wake-up module (110) and a voiceprint authentication module (120); The platform wake-up module (110) is used to capture the user's audio input in real time, pre-process and extract features of the real-time audio, the features include wake-up word features and voiceprint features, generate Mel-spectrograms or MFCC features, input the extracted features into a pre-trained speech recognition model, and the model outputs the probability distribution of the wake-up word. According to the output probability distribution, if the probability of the wake-up word exceeds a set threshold, it is determined to be a wake-up word, triggering the start of the business query platform; The voiceprint authentication module (120) adds a voiceprint recognition function on the basis of the pre-trained speech recognition model to extract the user's voiceprint features and compares them with the voiceprint samples stored in the database to ensure that only authorized users can log in to the business query platform.

3. The intelligent interactive system based on customer business data analysis according to claim 2, characterized in that: The pre-trained speech recognition model includes the following steps during training: Input layer: The input layer receives the preprocessed audio features; Convolutional layer: The convolutional layer is used to extract the spatiotemporal features in the audio signal. The spatiotemporal features include the features of the wake-up word and the voiceprint features. Pooling layer: The output of the convolutional layer is reduced in dimension through the pooling layer; Connection layer: pass the output of the convolutional layer and the pooling layer to the fully connected layer; Output layer: The output layer uses the Softmax activation function to convert the network output into the probability distribution of the wake-up word and output the pre-trained speech recognition model; And deploy the trained pre-trained speech recognition model to the business query platform.

4. The intelligent interactive system based on customer business data analysis according to claim 3 is characterized by: The interactive control unit (200) comprises a text recognition module (210) and a word segmentation determination module (220); The text recognition module (210) is used to, after sensing the login signal from the voiceprint authentication module (120), pre-process the input voice signal through voice recognition technology, extract the characteristics of the voice signal, map the voice characteristics to phonemes, convert the phoneme sequence into a word sequence according to the grammar and semantics of the language, and decode the most likely text sequence in combination with the acoustic model and the language model. The word segmentation determination module (220) is used to use natural language processing technology to perform word segmentation and stop word removal on the identified text sequence, use a naive Bayes classifier to identify the user's query intent, and then classify the text sequence, and transmit it to the analysis and call-out unit (300) to execute the query according to the identified query intent.

5. The intelligent interactive system based on customer business data analysis according to claim 4 is characterized in that: The analysis and calling unit (300) adopts a SQL query algorithm and comprises the following steps: A query data set is established by writing a query intent corresponding to an SQL query command according to screening conditions, wherein the screening conditions include customer order information, order date and order amount, the query intent of the word segmentation determination module (220) is input into the query data set, an SQL query command is called out, and an SQL query is executed.

6. The intelligent interactive system based on customer business data analysis according to claim 1, characterized in that: The log query unit (400) comprises a status recording module (410) and a progress query module (420); The state recording module (410) is connected to the output end of the analysis call-out unit (300), and is used to sense when the analysis call-out unit (300) executes a call-out command, obtain a time mark as an initial time, and monitor when the analysis call-out unit (300) executes a switch command, obtain a time mark as a termination time; The progress query module (420) is used to establish a time threshold based on historical behavior and feedback of viewing business data.

7. The intelligent interactive system based on customer business data analysis according to claim 6, characterized in that: The progress query module (420) establishes a time threshold through statistical analysis and uses a confidence interval to determine a threshold range, including the following steps: Collect n user viewing business data samples in the historical records, calculate the mean T and standard deviation σ of the historical query time data, according to the algorithm formula: Among them, z is the corresponding confidence level value.

8. The intelligent interactive system based on customer business data analysis according to claim 7, characterized in that: The log query unit (400) further includes an interest analysis module (430); The interest analysis module (430) is used to analyze the behavior data in the business data sample of the progress query module (420), the behavior data including access frequency and residence time, wherein the residence time is derived from the sum of the difference between the termination time and the initial time obtained by the state recording module (410) in the business data sample, and the comprehensive weight i=α*(user's residence time for content i / total residence time for users to access all contents)+β*(user's access frequency for content i / total access frequency for users to access all contents) is determined by comprehensively considering the behavior data, wherein α and β are coefficients for adjusting the weights of different indicators, satisfying α+β=1: interest degree=∑(interest degree of content i) *Comprehensive weight i) / Σ(weight of all contents) According to the ranking of interest, after the voiceprint authentication module (120) outputs a login signal, the content with high interest is displayed preferentially through a small window.

Images