Intelligent financial customer service interaction optimization method and system

Abstract

The invention relates to the technical field of financial science and technology, and discloses an intelligent financial customer service interaction optimization method and system, and the method comprises the steps: collecting a real-time session data flow and a session path when a target customer carries out the session interaction with an intelligent customer service system, calculating a satisfaction degree attenuation value of the intelligent customer service system, and evaluating an interaction quality index of the session path; extracting speech spectrum features and text semantic vectors in the real-time session data stream, drawing an emotion fluctuation curve of the target customer in the session process, extracting a wave crest semantic fragment corresponding to an emotion wave crest in the emotion fluctuation curve, and calculating an emotion entropy value of the target customer; calculating the service integrating degree of the intelligent customer service system in the interaction channel; generating a session strategy optimization scheme of the target customer; and executing interaction processing about the target client, calculating a client journey conversion rate corresponding to the session strategy optimization scheme, and performing optimization adjustment on the session strategy optimization scheme to obtain a final session strategy. According to the invention, the accuracy of financial customer service interaction optimization can be improved.

Description

Technical Field

[0001] The present invention relates to an intelligent financial customer service interaction optimization method and system, belonging to the technical field of financial technology. Background Art

[0002] With the rapid development of the financial industry, customers' demands for financial services are becoming increasingly diversified and personalized. Intelligent financial customer service, as a key tool to improve customer service efficiency and quality, has been widely used in financial institutions.

[0003] At present, intelligent financial customer service interactions mainly rely on rule-based systems and simple natural language processing technologies. First, financial institutions will sort out common business problems and customer consultation scenarios and develop corresponding rule bases. When customers initiate consultations, the system matches and parses customer input according to preset rules to give corresponding answers.

[0004] However, this traditional intelligent financial customer service interaction method has many limitations. On the one hand, the rule-based system has poor flexibility and is difficult to cope with complex and changeable customer problems and constantly updated financial business scenarios. Once the customer's consultation exceeds the scope of the preset rules, the system cannot accurately understand and answer, resulting in a poor customer experience. On the other hand, simple natural language processing technology has limited depth of understanding of semantics and cannot accurately grasp the implicit intentions and emotional tendencies in customer discourse. For example, customers may express their concerns about a certain financial product in a more obscure way. The existing system is difficult to accurately identify and provide targeted comfort and solutions, which reduces customer satisfaction with financial institutions. Therefore, a method is needed to improve the accuracy of financial customer service interaction optimization. Summary of the invention

[0005] The present invention provides an intelligent financial customer service interaction optimization method and system, the main purpose of which is to improve the accuracy of financial customer service interaction optimization.

[0006] To achieve the above objectives, the present invention provides an intelligent financial customer service interaction optimization method, comprising:

[0007] Collecting real-time conversation data streams and conversation paths when target customers interact with the intelligent customer service system, parsing conversation node sequences in the real-time conversation data streams, dividing interaction feature dimensions corresponding to the conversation node sequences, calculating satisfaction attenuation values of the intelligent customer service system based on the interaction feature dimensions, and evaluating interaction quality indexes of the conversation paths based on the satisfaction attenuation values;

[0008] Extract the voice spectrum features and text semantic vectors from the real-time session data stream, combine the voice spectrum features and the text semantic vectors, draw the emotional fluctuation curve of the target customer during the session, extract the peak semantic segments corresponding to the emotional peaks in the emotional fluctuation curve, and calculate the emotional entropy value of the target customer based on the peak semantic segments;

[0009] Query the type of interaction channel currently used by the target customer, collect the environmental noise data, interface operation heat map, and multi-task switching frequency of the interaction channel type, and calculate the service fit degree of the intelligent customer service system in the interaction channel by combining the environmental noise data, the interface operation heat map, and the multi-task switching frequency;

[0010] Based on the interaction quality index, the emotional entropy value, and the service fit degree, construct the dynamic optimization weight matrix of the intelligent customer service system, and generate the session strategy optimization plan for the target customer based on the dynamic optimization weight matrix;

[0011] Based on the session strategy optimization plan, perform interaction processing on the target customer, collect customer interaction data during the interaction processing, calculate the customer journey conversion rate corresponding to the session strategy optimization plan based on the customer interaction data, and optimize and adjust the session strategy optimization plan based on the customer journey conversion rate to obtain the final session strategy.

[0012] Optionally, the calculating the satisfaction attenuation value of the intelligent customer service system based on the real-time session data stream includes:

[0013] Perform data preprocessing on the real-time session data stream to obtain a standardized session data stream;

[0014] Perform session node division on the standardized session data stream to obtain a session node sequence;

[0015] Extract features from the session node sequence to obtain interaction feature dimensions;

[0016] Calculate the satisfaction score of each session node in the session node sequence based on the interaction feature dimensions;

[0017] Calculate the satisfaction attenuation value of the intelligent customer service system based on the satisfaction scores.

[0018] Optionally, the calculating the satisfaction score of each session node in the session node sequence based on the interaction feature dimensions includes:

[0019] Allocate the dimension weights corresponding to the interaction feature dimensions, and perform quantization processing on the interaction feature dimensions to obtain dimension feature values;

[0020] Normalize the dimension eigenvalue to obtain a normalized dimension eigenvalue, and query the dimension ideal value corresponding to the interaction feature dimension;

[0021] Combine the normalized dimension eigenvalue to calculate the dimension average value and dimension standard deviation corresponding to the interaction feature dimension;

[0022] Combine the dimension weight, the normalized dimension eigenvalue, the dimension ideal value, the dimension average value and the dimension standard deviation, and calculate the satisfaction score of each session node in the session node sequence through the following formula:

[0023]

[0024] where A represents the satisfaction score of each session node in the session node sequence, and β a represents the dimension weight corresponding to the a-th dimension in the interaction feature dimension, and B ia represents the normalized dimension eigenvalue corresponding to the a-th dimension of the i-th node in the session node sequence, represents the dimension average value of the a-th dimension, and α a represents the fluctuation adjustment coefficient of the a-th dimension, represents the dimension average value of the a-th dimension, and σ a represents the dimension standard deviation of the a-th dimension, a represents the serial number of the interaction feature dimension, q represents the number of interaction feature dimensions, and i represents the serial number of the session node sequence.

[0025] Optionally, the extraction of the voice spectrum feature and text semantic vector in the real-time session data stream includes:

[0026] Extract the session voice signal corresponding to the real-time session data stream, perform time-domain conversion processing on the session voice signal to obtain a voice frequency-domain signal;

[0027] Extract the frequency-domain signal feature corresponding to the voice frequency-domain signal;

[0028] Identify the session text information in the real-time session data stream, and extract the key session text in the session text information;

[0029] Perform semantic parsing on the key session text to obtain the key session text semantics;

[0030] Perform vectorization processing on the key session text semantics to obtain a session text semantic vector;

[0031] Perform dimensionality reduction processing on the session text semantic vector to obtain a text semantic vector.

[0032] Optionally, calculating the emotional entropy value of the target customer based on the peak semantic segment includes:

[0033] Extract semantic key characters from the peak semantic segment;

[0034] Based on the semantic key characters, perform emotional classification on the peak semantic segment to obtain the emotional state of the semantic segment;

[0035] Calculate the emotional probability density corresponding to the emotional state of the semantic segment, and count the number of emotional categories corresponding to the emotional state of the semantic segment;

[0036] Combining the emotional probability density and the number of emotional categories, calculate the emotional entropy value of the target customer through the following formula:

[0037]

[0038] where D represents the emotional entropy value of the target customer, and E b represents the emotional probability density of the b-th emotion in the emotional state of the semantic segment, b represents the serial number of the emotional state of the semantic segment, and w represents the number of emotional states of the semantic segment.

[0039] Optionally, combining the environmental noise data, the interface operation heat map, and the multi-task switching frequency to calculate the service fit degree of the intelligent customer service system in the interaction channel includes:

[0040] Calculate the decibel value of the environmental noise data to obtain the environmental noise level;

[0041] Divide the interface operation heat map into regions to obtain functional operation regions;

[0042] Count the operation heat values corresponding to each region in the functional operation regions;

[0043] Perform time window statistics on the multi-task switching frequency to obtain the task switching frequency value;

[0044] Combine the environmental noise level, the operation heat value, and the task switching frequency value to calculate the service fit degree of the intelligent customer service system in the interaction channel.

[0045] Optionally, combining the environmental noise level, the operation heat value, and the task switching frequency value to calculate the service fit degree of the intelligent customer service system in the interaction channel includes:

[0046] Perform standardization processing on the environmental noise level, the operation heat value, and the task switching frequency value to obtain the standard noise level value, the standard operation heat value, and the standard switching frequency value;

[0047] Combined with the standard noise level value, the standard operation heat value, and the standard switching frequency value, calculate the service fit of the intelligent customer service system in the interaction channel through the following formula:

[0048]

[0049] Wherein, F represents the service fit of the intelligent customer service system in the interaction channel, G represents the standard noise level value, H represents the standard operation heat value, L represents the standard switching frequency value, and ∈ represents the smoothing constant.

[0050] Optionally, calculating the customer journey conversion rate corresponding to the session policy optimization plan based on the customer interaction data includes:

[0051] Perform data cleaning on the customer interaction data to obtain target customer interaction data;

[0052] Identify the interaction data tags corresponding to the target customer interaction data, and perform stage division on the target customer interaction data to obtain customer journey stage data;

[0053] Based on the interaction data tags, identify the key journey nodes in the customer journey stage data;

[0054] Count the number of node customers corresponding to the key journey nodes from the customer journey stage data;

[0055] Calculate the customer journey conversion rate corresponding to the session policy optimization plan based on the number of node customers.

[0056] Optionally, identifying the key journey nodes in the customer journey stage data based on the interaction data tags includes:

[0057] Identify the customer behavior identifiers in the interaction data tags, and extract the customer behavior characteristics corresponding to the customer behavior identifiers;

[0058] Calculate the feature similarity between the customer behavior characteristics;

[0059] Based on the feature similarity, perform feature clustering processing on the customer behavior characteristics to obtain clustered behavior characteristics;

[0060] Analyze the key behaviors corresponding to the clustered behavior characteristics, and perform node mapping on the key behaviors to obtain the key journey nodes in the customer journey stage data.

[0061] To solve the above problems, the present invention also provides an intelligent financial customer service interaction optimization system, and the system includes:

[0062] The interaction quality index evaluation module is used to collect the real-time session data stream and session path when the target customer conducts a session interaction with the intelligent customer service system, parse the session node sequence in the real-time session data stream, divide the interaction feature dimensions corresponding to the session node sequence, calculate the satisfaction decay value of the intelligent customer service system based on the interaction feature dimensions, and evaluate the interaction quality index of the session path based on the satisfaction decay value;

[0063] The emotional entropy value calculation module is used to extract the voice spectrum features and text semantic vectors in the real-time session data stream, combine the voice spectrum features and the text semantic vectors to draw the emotional fluctuation curve of the target customer during the session, extract the peak semantic segments corresponding to the emotional peaks in the emotional fluctuation curve, and calculate the emotional entropy value of the target customer based on the peak semantic segments;

[0064] The service fit calculation module is used to query the type of interaction channel currently used by the target customer, collect the environmental noise data, interface operation heat map and multi-task switching frequency of the interaction channel type, and calculate the service fit of the intelligent customer service system in the interaction channel by combining the environmental noise data, the interface operation heat map and the multi-task switching frequency;

[0065] The session strategy optimization plan generation module is used to construct a dynamic optimization weight matrix of the intelligent customer service system based on the interaction quality index, the emotional entropy value and the service fit, and generate a session strategy optimization plan for the target customer based on the dynamic optimization weight matrix;

[0066] The plan optimization and adjustment module is used to perform interaction processing on the target customer based on the session strategy optimization plan, collect customer interaction data during the interaction processing, calculate the customer journey conversion rate corresponding to the session strategy optimization plan based on the customer interaction data, and optimize and adjust the session strategy optimization plan based on the customer journey conversion rate to obtain the final session strategy.

[0067] Compared with the problems described in the background art, the present invention calculates the satisfaction attenuation value of the intelligent customer service system based on the real-time session data stream, which can insight into the dynamic changes of the satisfaction of the target customer during the interaction process, and further lays an important basis for the evaluation of the interaction quality index of the subsequent session path. Further, the present invention can accurately capture the emotional changes of the target customer during the session by extracting the voice spectrum features and text semantic vectors in the real-time session data stream, providing a strong basis for in-depth understanding of the needs of the target customer and optimizing the intelligent customer service system. The present invention calculates the service fit degree of the intelligent customer service system in the interaction channel by combining the environmental noise data, the interface operation heat map and the multi-task switching frequency, and can understand the adaptation degree of the intelligent customer service system in the interaction channel through the service fit degree, thereby improving the construction accuracy of the dynamic optimization weight matrix of the subsequent intelligent customer service system. Further, the present invention constructs the dynamic optimization weight matrix of the intelligent customer service system based on the interaction quality index, the emotional entropy value and the service fit degree, which can dynamically adjust the service strategy of the intelligent customer service system, improve the satisfaction and interaction experience of the target customer, and provide a scientific basis for the optimization of the subsequent session strategy. Further, the present invention executes the interaction processing on the target customer based on the session strategy optimization scheme and collects the customer interaction data during the interaction processing, which can insight into the behavior patterns, demand preferences and pain points of the target customer during the interaction process, providing a solid data basis for accurately calculating the customer journey conversion rate in the future. Therefore, the intelligent financial customer service interaction optimization method and system provided by the embodiments of the present invention can improve the accuracy of financial customer service interaction optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 FIG. is a schematic flow chart of an intelligent financial customer service interaction optimization method provided by an embodiment of the present invention;

[0069] Figure 2 FIG. is a schematic diagram of a module for implementing the intelligent financial customer service interaction optimization method provided by an embodiment of the present invention.

[0070] The implementation, functional features and advantages of the present invention will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0071] 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.

[0072] An embodiment of the present application provides an intelligent financial customer service interaction optimization method. The execution subject of the intelligent financial customer service interaction optimization method includes, but is not limited to, at least one of electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided by the embodiment of the present application. In other words, the intelligent financial customer service interaction optimization method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to: a single server, a server cluster, a cloud server, or a cloud server cluster, etc.

[0073] Embodiment 1:

[0074] Referring to Figure 1 As shown, it is a flowchart of the intelligent financial customer service interaction optimization method provided by an embodiment of the present invention. In this embodiment, the intelligent financial customer service interaction optimization method includes:

[0075] S1. Collect the real-time session data stream and session path when the target customer interacts with the intelligent customer service system. Based on the real-time session data stream, calculate the satisfaction decay value of the intelligent customer service system. Based on the satisfaction decay value, evaluate the interaction quality index of the session path.

[0076] By calculating the satisfaction decay value of the intelligent customer service system based on the real-time session data stream, the present invention can insight into the dynamic change of the target customer's satisfaction during the interaction process, and thus lay an important basis for the subsequent evaluation of the interaction quality index of the session path. It should be explained that the intelligent customer service system is a system that provides services to customers with the help of artificial intelligence technology. The real-time session data stream is an information flow containing text, voice, etc. generated in real time when the target customer interacts with the intelligent customer service system. The session path is the operation step process experienced by the target customer when interacting with the intelligent customer service system. The satisfaction decay value is a quantitative index of the degree of reduction of the customer's satisfaction in the intelligent customer service system over time. Further, the real-time session data stream and session path when the target customer interacts with the intelligent customer service system can be collected through a data collection tool.

[0077] Specifically, calculating the satisfaction decay value of the intelligent customer service system based on the real-time session data stream includes:

[0078] Perform data preprocessing on the real-time session data stream to obtain a standardized session data stream;

[0079] Perform session node division on the standardized session data stream to obtain a session node sequence;

[0080] Extract features from the session node sequence to obtain an interaction feature dimension;

[0081] Calculate the satisfaction score of each session node in the session node sequence based on the interaction feature dimension;

[0082] Calculate the satisfaction attenuation value of the intelligent customer service system based on the satisfaction score.

[0083] It should be explained that the session node sequence is an ordered arrangement formed by key information nodes identified from the standardized session data stream according to specific rules in chronological order. The interaction feature dimension is an aspect reflecting the interaction characteristics divided after extracting different types of features from the session node sequence. The satisfaction score is a value representing the customer's satisfaction with the interaction of each session node in the session node sequence, which is calculated by weighted comparison of the performance of each session node in each interaction feature dimension with the preset ideal value.

[0084] Furthermore, the data preprocessing of the real-time session data stream can be implemented through a data cleaning tool; the session node division can be implemented through natural language processing technology; the feature extraction can be implemented through feature engineering methods; the calculation of the satisfaction attenuation value can be implemented through time series analysis methods, such as using the moving average method (MA) to smooth the time series data of the customer satisfaction score over a period of time, and calculating the average attenuation degree of satisfaction per unit time according to the change trend of the score.

[0085] Furthermore, as an optional embodiment of the present invention, calculating the satisfaction score of each session node in the session node sequence based on the interaction feature dimension includes:

[0086] Allocate the dimension weights corresponding to the interaction feature dimension, and perform quantization processing on the interaction feature dimension to obtain dimension feature values;

[0087] Perform normalization processing on the dimension feature values to obtain normalized dimension feature values, and query the dimension ideal values corresponding to the interaction feature dimension;

[0088] Combine the normalized dimension feature values to calculate the dimension average value and dimension standard deviation corresponding to the interaction feature dimension;

[0089] Combine the dimension weights, the normalized dimension feature values, the dimension ideal values, the dimension average values, and the dimension standard deviations, and calculate the satisfaction score of each session node in the session node sequence through the following formula:

[0090]

[0091] where A represents the satisfaction score of each session node in the session node sequence, and β a represents the dimension weight corresponding to the a-th dimension in the interaction feature dimension, and Bia It represents the normalized dimensional eigenvalue corresponding to the a-th dimension of the i-th node in the session node sequence. It represents the dimensional average value of the a-th dimension, α a It represents the fluctuation adjustment coefficient of the a-th dimension. It represents the dimensional average value of the a-th dimension, σ a It represents the dimensional standard deviation of the a-th dimension. a represents the serial number of the interaction feature dimension, q represents the number of interaction feature dimensions, and i represents the serial number of the session node sequence.

[0092] Furthermore, the fluctuation adjustment coefficient is used to measure the fluctuation of the data in this dimension. It is obtained by calculating the coefficient of variation (the ratio of the standard deviation to the mean) of this dimension in historical data. The larger the coefficient of variation, the greater the fluctuation of the data in this dimension, and the greater the value of the fluctuation adjustment coefficient, and the more obvious the adjustment effect on the satisfaction score. For example, if the coefficient of variation of a certain interaction feature dimension is 0.3, then the fluctuation adjustment coefficient can take the value of 0.3.

[0093] It should be explained that the dimensional weight is the proportion of the importance of the corresponding interaction feature dimension in the comprehensive evaluation. The dimensional eigenvalue is the data representation obtained from the actual observation or calculation of the interaction feature dimension. The normalized dimensional eigenvalue is the value obtained by normalizing the dimensional eigenvalue for easy unified comparison and operation. The dimensional ideal value is the best standard numerical value that the corresponding interaction feature dimension is expected to reach. The dimensional average value and the dimensional standard deviation are respectively the average level numerical value and the measure numerical value of the data dispersion degree of the corresponding interaction feature dimension in a large number of data samples.

[0094] Furthermore, the dimensional weight corresponding to the interaction feature dimension can be assigned through expert experience combined with mathematical methods such as the Analytic Hierarchy Process (AHP). The interaction feature dimension can be quantitatively processed through a data analysis tool based on a specific algorithm to obtain the dimensional eigenvalue; the dimensional eigenvalue can be normalized through common normalization algorithms such as maximum-minimum normalization to obtain the normalized dimensional eigenvalue. The dimensional ideal value corresponding to the interaction feature dimension can be queried by referring to industry standards and enterprise internal service target setting documents; combined with the normalized dimensional eigenvalue, the dimensional average value and the dimensional standard deviation corresponding to the interaction feature dimension can be calculated through statistical software or statistical functions in programming.

[0095] Based on the satisfaction attenuation value, the present invention evaluates the interaction quality index of the session path, which can intuitively reflect the service effectiveness of the intelligent customer service system during the entire session, enabling the operator to clearly grasp the changing trend of the customer experience, so as to accurately locate system problems, optimize and improve them targeted, and enhance the overall service quality. It should be explained that the interaction quality index is an evaluation result of the interaction quality presented as a quantitative value that comprehensively reflects the multi-dimensional interaction characteristics of the customer and the intelligent customer service system during the entire interaction process in the session path, such as response time, problem-solving rate, and dialogue fluency. Further, based on the satisfaction attenuation value, the interaction quality index of the session path is evaluated. If the satisfaction attenuation value exceeds a specific threshold (such as 20%), it is determined that the interaction quality is low, and the interaction strategy of the intelligent customer service system needs to be optimized. For example, when the satisfaction attenuation value is within 10%, the interaction quality is determined to be excellent; when the satisfaction attenuation value is between 10% and 20%, the interaction quality is determined to be good; when the satisfaction attenuation value exceeds 20%, the interaction quality is determined to be poor. At this time, the response speed and problem-solving efficiency of the intelligent customer service system can be considered for optimization.

[0096] S2. Extract the voice spectrum features and text semantic vectors from the real-time session data stream, combine the voice spectrum features and the text semantic vectors, draw the emotional fluctuation curve of the target customer during the session, extract the peak semantic segment corresponding to the emotional peak in the emotional fluctuation curve, and calculate the emotional entropy value of the target customer based on the peak semantic segment.

[0097] By extracting the voice spectrum features and text semantic vectors from the real-time session data stream, the present invention can accurately capture the emotional changes of the target customer during the session, providing a strong basis for deeply understanding the needs of the target customer and optimizing the intelligent customer service system. It should be explained that the voice spectrum features refer to a set of numerical values extracted from the customer's voice data that reflect characteristics such as the voice frequency distribution and energy change, such as the energy proportion in different frequency bands and the fundamental frequency change. The text semantic vector is to transform the text content input by the customer into a vector representation in a high-dimensional space through a specific algorithm, and this vector can reflect the semantic information of the text.

[0098] Specifically, the extraction of the voice spectrum features and text semantic vectors from the real-time session data stream includes:

[0099] Extract the session voice signal corresponding to the real-time session data stream, perform time-domain conversion processing on the session voice signal to obtain a voice frequency-domain signal;

[0100] Extract the frequency-domain signal features corresponding to the voice frequency-domain signal;

[0101] Identify the session text information in the real-time session data stream, and extract the key session text from the session text information;

[0102] Perform semantic parsing on the key session text to obtain the key session text semantics;

[0103] Perform vectorization processing on the key session text semantics to obtain the session text semantic vector;

[0104] Perform dimensionality reduction processing on the session text semantic vector to obtain the text semantic vector.

[0105] It should be explained that the session voice signal is the original signal of the voice part corresponding to the real-time session data stream, the voice frequency-domain signal is the signal obtained by converting the session voice signal from the time domain and showing the energy distribution with frequency as the independent variable, the frequency-domain signal feature is the numerical value corresponding to the voice frequency-domain signal that can reflect voice characteristics such as the energy ratio of different frequency bands and the fundamental frequency change, the session text information is the text content input by the customer in the real-time session data stream, the key session text is the text segment in the session text information that is of great significance for analyzing the customer's emotions, intentions, etc., the key session text semantics is the meaning expressed by the key session text, and the session text semantic vector is the vector representation of the key session text semantics in a high-dimensional space through a specific natural language processing algorithm.

[0106] Furthermore, the session voice signal corresponding to the real-time session data stream can be extracted through the audio acquisition module of the speech recognition software; the session voice signal can be subjected to time-domain conversion processing through the Fourier transform algorithm to obtain the voice frequency-domain signal; the frequency-domain signal features corresponding to the voice frequency-domain signal can be extracted through a dedicated speech feature extraction tool such as Praat software; the session text information in the real-time session data stream can be identified through the text extraction function in the natural language processing toolkit (such as NLTK); the key session text in the session text information can be extracted through a screening algorithm based on keyword matching and semantic understanding; semantic parsing techniques such as dependency syntax analysis and semantic role labeling can be used to perform semantic parsing on the key session text to obtain the key session text semantics; vectorization algorithms such as word embedding (such as Word2Vec, GloVe) can be used to perform vectorization processing on the key session text semantics to obtain the session text semantic vector; dimensionality reduction algorithms such as principal component analysis (PCA) and linear discriminant analysis (LDA) can be used to perform dimensionality reduction processing on the session text semantic vector to obtain the text semantic vector.

[0107] The present invention draws the emotion fluctuation curve of the target customer during the conversation by combining the speech spectrum features and the text semantic vectors, which can intuitively show the fluctuations of the customer's emotions in the time dimension, provide a visual basis for the intelligent customer service system to adjust the service strategy in time, and extract the peak semantic fragments corresponding to the emotion peaks in the emotion fluctuation curve, so as to accurately locate the key nodes of the customer's emotional outburst and the related text content, and help the operation personnel to deeply analyze the customer's pain points and improve the service process in a targeted manner. It should be explained that the emotion fluctuation curve is a curve intuitively showing the fluctuations of the customer's emotions during the conversation of the target customer, with time as the horizontal axis and the emotion intensity as the vertical axis, based on the speech spectrum features and the text semantic vectors; the peak semantic fragment is a text fragment corresponding to the emotion peak in the emotion fluctuation curve, which is input by the customer at the moment when the emotion intensity reaches the peak, and contains strong emotional expression; In the first step, the speech spectrum features and the text semantic vectors can be combined, and the data of the two can be input into an emotion intensity assessment model based on machine learning training to generate an emotion intensity value, thereby drawing the emotion fluctuation curve of the target customer during the conversation. For example, a large amount of speech spectrum features and text semantic vector data with labeled emotion intensity can be trained using a support vector machine algorithm to construct an assessment model that can accurately map the relationship between the two and emotion intensity. After inputting real-time speech spectrum features and text semantic vectors, the model will output the emotion intensity value corresponding to each time point. These values can be connected in sequence with time as the horizontal axis and emotion intensity as the vertical axis, so that the emotion fluctuation curve of the target customer during the conversation can be clearly drawn, which intuitively reflects the ups and downs of the customer's emotions. The peak point can be located through a curve analysis algorithm, and the peak semantic segment corresponding to the emotion peak in the emotion fluctuation curve can be extracted from the original conversation data according to the timestamp.

[0108] The present invention calculates the emotional entropy value of the target customer based on the peak semantic fragment, so as to quantify the emotional complexity of the target customer at the emotional peak moment, thereby laying an important basis for the subsequent construction of the dynamic optimization weight matrix of the intelligent customer service system. It should be explained that the emotional entropy value is a numerical value used to quantify the emotional complexity and uncertainty of the target customer.

[0109] In detail, the calculating the emotion entropy value of the target customer based on the peak semantic segment includes:

[0110] Extracting semantic key characters in the peak semantic segment;

[0111] Based on the semantic key characters, the emotion classification of the peak semantic segment is performed to obtain the emotion of the semantic segment;

[0112] Calculate the emotional probability density corresponding to the emotion of the semantic segment, and count the number of emotional categories corresponding to the emotion of the semantic segment;

[0113] Combine the emotional probability density and the number of emotional categories, and calculate the emotional entropy value of the target customer through the following formula:

[0114]

[0115] where D represents the emotional entropy value of the target customer, and E b represents the emotional probability density of the b-th emotion in the emotions of the semantic segment, b represents the serial number of the emotion of the semantic segment, and w represents the number of emotions of the semantic segment.

[0116] It should be explained that the semantic key character is the character in the peak semantic segment that can accurately convey the core semantics and plays a key role in understanding the customer's emotion; the emotion of the semantic segment is the customer's emotional tendency inferred through text analysis corresponding to the peak semantic segment; the emotional probability density is the distribution of the possibility of this emotion appearing in different intensity intervals corresponding to the emotion of the semantic segment. Further, the semantic key characters in the peak semantic segment can be extracted through a text mining algorithm based on rule matching and word frequency statistics; based on the semantic key characters, the peak semantic segment can be classified by a trained deep neural network emotion classification model to obtain the emotion of the semantic segment; the emotional probability density corresponding to the emotion of the semantic segment can be calculated through non-parametric statistical methods such as kernel density estimation, and the number of emotional categories corresponding to the emotion of the semantic segment can be counted through a simple counting algorithm.

[0117] S3. Query the type of interaction channel currently used by the target customer, collect the environmental noise data, interface operation heat map and multi-task switching frequency of the interaction channel type, and combine the environmental noise data, the interface operation heat map and the multi-task switching frequency to calculate the service fit degree of the intelligent customer service system in the interaction channel.

[0118] The present invention calculates the service fit degree of the intelligent customer service system in the interaction channel by combining the environmental noise data, the interface operation heat map and the multi-task switching frequency, and can understand the adaptation degree of the intelligent customer service system in the interaction channel through the service fit degree, thereby improving the construction accuracy of the dynamic optimization weight matrix of the subsequent intelligent customer service system.

[0119] It should be noted that the types of interaction channels cover different ways for customers to interact with the intelligent customer service, such as mobile phone apps, web pages, social media platforms, etc. The environmental noise data reflects the noise level of the physical environment where the customer is located when using the interaction channel, which will affect the clarity of voice interaction and the customer's attention. The interface operation heat map intuitively shows the operation behavior of the customer on the interaction interface, including the frequency and regional distribution of operations such as clicks, swipes, and stays, which helps to understand the customer's usage preferences and operation habits for interface functions. The multi-task switching frequency refers to the frequency of switching between this interaction channel and other applications or tasks during the customer's use of the intelligent customer service, which reflects the complexity of the customer's usage scenario and the potential demand for the timeliness of the customer service response. The service fit degree represents the adaptation degree of the intelligent customer service system in the interaction channel. Further, the query of the type of interaction channel currently used by the target customer can be achieved through the device information of the target customer; the collection of the environmental noise data of the interaction channel type can be achieved through a microphone device; the collection of the interface operation heat map of the interaction channel type can be achieved through the user interface log; the collection of the multi-task switching frequency of the interaction channel type can be achieved through the operating system log of the user device.

[0120] Specifically, calculating the service fit degree of the intelligent customer service system in the interaction channel by combining the environmental noise data, the interface operation heat map, and the multi-task switching frequency includes:

[0121] Calculating the decibel value of the environmental noise data to obtain the environmental noise level;

[0122] Dividing the interface operation heat map into regions to obtain functional operation regions;

[0123] Counting the operation heat values corresponding to each region in the functional operation regions;

[0124] Performing time window statistics on the multi-task switching frequency to obtain the task switching frequency value;

[0125] Combining the environmental noise level, the operation heat value, and the task switching frequency value to calculate the service fit degree of the intelligent customer service system in the interaction channel.

[0126] It should be noted that the environmental noise level is a quantitative representation of the noise level of the environment where the customer is located in the interaction channel after calculating the decibel value of the environmental noise data; the functional operation area is the operation area obtained after dividing the interface operation heat map, representing different functional modules of the intelligent customer service system; the operation heat value is the value corresponding to each area in the functional operation area, which is statistically obtained based on data such as the number of operations and reflects the frequency of user operations in that area; the task switching frequency value is a quantitative indicator of the frequency of customer switching between different tasks within a specific time window after performing a time window statistics on the multi-task switching frequency.

[0127] Furthermore, the decibel value of the environmental noise data can be calculated through an audio analysis algorithm to obtain the environmental noise level; the interface operation heat map can be divided into areas through an image recognition and segmentation algorithm to obtain the functional operation area; the operation heat value corresponding to each area in the functional operation area can be statistically obtained by setting an event listening mechanism within the functional operation area; the multi-task switching frequency can be statistically analyzed with a time window by recording the timestamps of system task state changes and setting a reasonable time window to obtain the task switching frequency value.

[0128] Furthermore, as an optional embodiment of the present invention, calculating the service fit degree of the intelligent customer service system in the interaction channel by combining the environmental noise level, the operation heat value, and the task switching frequency value includes:

[0129] Performing a normalization process on the environmental noise level, the operation heat value, and the task switching frequency value to obtain a standard noise level value, a standard operation heat value, and a standard switching frequency value;

[0130] Combining the standard noise level value, the standard operation heat value, and the standard switching frequency value, and calculating the service fit degree of the intelligent customer service system in the interaction channel through the following formula:

[0131]

[0132] where F represents the service fit degree of the intelligent customer service system in the interaction channel, G represents the standard noise level value, H represents the standard operation heat value, L represents the standard switching frequency value, and ∈ represents a smoothing constant.

[0133] Among them, the standard noise level value, the standard operation heat value, and the standard switching frequency value are the values obtained after normalizing the environmental noise level, the operation heat value, and the task switching frequency value respectively. Furthermore, the smoothing constant is an extremely small constant (such as 0.001) used to avoid the denominator being zero, and e is the base of the natural logarithm (approximately equal to 2.71828).

[0134] S4. Based on the interaction quality index, the emotional entropy value, and the service fit degree, construct a dynamic optimization weight matrix for the intelligent customer service system, and based on the dynamic optimization weight matrix, generate an optimized session strategy plan for the target customer.

[0135] By constructing a dynamic optimization weight matrix for the intelligent customer service system based on the interaction quality index, the emotional entropy value, and the service fit degree, the present invention can dynamically adjust the service strategy of the intelligent customer service system, improve the satisfaction and interaction experience of the target customer, and provide a scientific basis for the optimization of subsequent session strategies.

[0136] It should be explained that the dynamic optimization weight matrix is a matrix used to dynamically adjust the weights of the interaction quality index, the emotional entropy value, and the service fit degree. The session strategy optimization plan is a specific strategy for optimizing the session experience of the target customer based on the dynamic optimization weight matrix. Further, based on the interaction quality index, the emotional entropy value, and the service fit degree, the initial weights of each index can be determined by using the fuzzy analytic hierarchy process, and then combined with the change trend of real-time data, the weights are dynamically updated by using the adaptive weighted adjustment algorithm, so as to construct the dynamic optimization weight matrix of the intelligent customer service system; based on the dynamic optimization weight matrix, the scores of each strategy in the session strategy library established in advance in terms of interaction quality, emotional entropy value, and service fit degree can be weighted and calculated, and the strategy with the highest comprehensive score is selected and fine-tuned according to the actual situation to generate the optimized session strategy plan for the target customer.

[0137] S5. Based on the optimized session strategy plan, perform interaction processing on the target customer, collect customer interaction data during the interaction processing, calculate the customer journey conversion rate corresponding to the optimized session strategy plan based on the customer interaction data, and optimize and adjust the optimized session strategy plan based on the customer journey conversion rate to obtain the final session strategy.

[0138] By performing interaction processing on the target customer based on the optimized session strategy plan and collecting customer interaction data during the interaction processing, the present invention can gain insights into the behavior patterns, demand preferences, and pain points of the target customer during the interaction process, providing a solid data foundation for accurately calculating the customer journey conversion rate in the future. It should be explained that the customer interaction data is the data recorded by the target customer during the interaction processing. Further, the customer interaction data during the interaction processing can be collected through page tagging scripts.

[0139] Based on the customer interaction data, the present invention calculates the customer journey conversion rate corresponding to the session strategy optimization plan, which can measure the effectiveness of the session strategy optimization plan in promoting customers to achieve key business goals, clearly insight into the conversion bottlenecks in the customer behavior path, and provide strong data support for subsequent optimization and adjustment of the session strategy optimization plan. It should be noted that the customer journey conversion rate is corresponding to the session strategy optimization plan. When interacting with target customers using this optimization plan, the proportion of the number of customers who complete preset key business goals such as purchasing products, successfully solving problems and expressing satisfaction, etc., from the start of participating in the interaction to the end, to the total number of customers participating in the interaction.

[0140] Specifically, calculating the customer journey conversion rate corresponding to the session strategy optimization plan based on the customer interaction data includes:

[0141] Clean the customer interaction data to obtain target customer interaction data;

[0142] Identify the interaction data tags corresponding to the target customer interaction data, and divide the target customer interaction data into stages to obtain customer journey stage data;

[0143] Based on the interaction data tags, identify the key journey nodes in the customer journey stage data;

[0144] Count the number of node customers corresponding to the key journey nodes from the customer journey stage data;

[0145] Calculate the customer journey conversion rate corresponding to the session strategy optimization plan based on the number of node customers.

[0146] It should be noted that the target customer interaction data is a data subset obtained by removing noise, duplicate and error data from the customer interaction data and can be used for accurate analysis; the interaction data tag is an identifier used to mark key information such as customer behavior, business process stage, consultation content type, etc. for the target customer interaction data; the customer journey stage data is a set of customer interaction data divided into different stages according to the business process logic of the target customer interaction data; the key journey node is a specific link or event in the customer journey stage data that represents the significance in the business process advancement of customers and can measure the conversion effect; the number of node customers is the specific number of customers who reach the key node corresponding to the key journey node statistically in the customer journey stage data.

[0147] Furthermore, the customer interaction data can be cleaned by the box plot method to obtain the target customer interaction data; the interaction data tags corresponding to the target customer interaction data can be identified by a tag recognition tool, and the tag recognition tool is compiled by a scripting language; the target customer interaction data can be divided into stages according to the business process to obtain the customer journey stage data; the number of node customers corresponding to the key journey nodes can be counted from the customer journey stage data by a statistical method; based on the number of node customers, the customer journey conversion rate corresponding to the session strategy optimization plan is calculated, the starting number of customers and the ending number of customers of the key journey nodes are determined according to the number of node customers, the ratio of the starting number of customers and the ending number of customers is calculated to obtain the node journey conversion rate, and the average value corresponding to the node journey conversion rate is calculated to obtain the customer journey conversion rate corresponding to the session strategy optimization plan.

[0148] Furthermore, as an alternative embodiment of the present invention, the identifying the key journey nodes in the customer journey stage data based on the interaction data tags includes:

[0149] Identify the customer behavior identifiers in the interaction data tags, and extract the customer behavior characteristics corresponding to the customer behavior identifiers;

[0150] Calculate the feature similarity between the customer behavior characteristics;

[0151] Based on the feature similarity, perform feature clustering processing on the customer behavior characteristics to obtain the clustered behavior characteristics;

[0152] Analyze the key behaviors corresponding to the clustered behavior characteristics, and perform node mapping on the key behaviors to obtain the key journey nodes in the customer journey stage data.

[0153] It should be explained that the customer behavior identifier is the part of the interaction data tag that is used to clearly identify the specific behavior actions, operations or behavior tendencies of the customer; the customer behavior characteristic is the specific characteristics and manifestation forms that can reflect the attributes, methods, frequencies, times, etc. of the customer behavior in the customer behavior identifier; the feature similarity is a quantitative index obtained by calculating and comparing the same or similar degrees of the attributes, methods, etc. between the customer behavior characteristics; the clustered behavior characteristic is a set of behavior characteristics with similar characteristics formed by grouping and classifying the customer behavior characteristics according to the feature similarity; the key behavior is the representative behavior in the clustered behavior characteristic set that has an important impact on the customer journey, business goal achievement, etc. and plays a key role.

[0154] Furthermore, the customer behavior identifiers in the interaction data tags can be recognized through a predefined behavior tag dictionary and a text matching algorithm. The customer behavior features corresponding to the customer behavior identifiers can be extracted from the relevant interaction data by establishing a feature extraction rule and a data mapping relationship corresponding to the customer behavior identifiers. The feature similarity between the customer behavior features can be calculated through common similarity measurement methods such as the cosine similarity algorithm and the Euclidean distance algorithm, in combination with the data structure and attribute characteristics of the customer behavior features. When the feature similarity is greater than a preset threshold, the customer behavior features are subjected to feature clustering processing to obtain clustered behavior features. The preset threshold can be set to 0.8 or can be set according to the actual business scenario. The behaviors that play a key role in promoting the business process and customer conversion can be analyzed and determined by performing frequency statistics, business impact assessment, and expert experience judgment on the clustered behavior features, so as to analyze the key behaviors corresponding to the clustered behavior features. By sorting out the business process framework, a correspondence table between the key behaviors and each stage of the customer journey is established. Based on this table, the key behaviors are matched to the corresponding customer journey stages, and then the key behaviors are subjected to node mapping to obtain the key journey nodes in the customer journey stage data.

[0155] The present invention optimizes and adjusts the session strategy optimization plan based on the value of the customer journey conversion rate to obtain a final session strategy. For example, when the conversion rate is low, the conversion bottlenecks of the key nodes in each stage are deeply analyzed and improved from aspects such as speech optimization and process simplification; when the conversion rate is high, the advantageous links are strengthened to further improve the overall effect, thereby improving the accuracy of financial customer service interaction optimization.

[0156] Compared with the problems described in the background art, the present invention calculates the satisfaction attenuation value of the intelligent customer service system based on the real-time session data stream, which can insight into the dynamic changes in the satisfaction of the target customer during the interaction process, and further lays an important basis for the evaluation of the interaction quality index of the subsequent session path. Further, the present invention can accurately capture the emotional changes of the target customer during the session by extracting the voice spectrum features and text semantic vectors in the real-time session data stream, providing a strong basis for in-depth understanding of the needs of the target customer and optimizing the intelligent customer service system. The present invention calculates the service fit degree of the intelligent customer service system in the interaction channel by combining the environmental noise data, the interface operation heat map and the multitask switching frequency, and can understand the adaptation degree of the intelligent customer service system in the interaction channel through the service fit degree, thereby improving the construction accuracy of the dynamic optimization weight matrix of the subsequent intelligent customer service system. Further, the present invention constructs the dynamic optimization weight matrix of the intelligent customer service system based on the interaction quality index, the emotional entropy value and the service fit degree, which can dynamically adjust the service strategy of the intelligent customer service system, improve the satisfaction and interaction experience of the target customer, and provide a scientific basis for the optimization of the subsequent session strategy. Further, the present invention executes the interaction processing on the target customer based on the session strategy optimization scheme and collects the customer interaction data during the interaction processing, which can insight into the behavior patterns, demand preferences and pain points of the target customer during the interaction process, providing a solid data basis for accurately calculating the customer journey conversion rate in the future. Therefore, the intelligent financial customer service interaction optimization method and system provided by the embodiments of the present invention can improve the accuracy of financial customer service interaction optimization.

[0157] Embodiment 2:

[0158] As Figure 2 shown, it is a functional module diagram of an intelligent financial customer service interaction optimization system of the present invention.

[0159] The intelligent financial customer service interaction optimization system 200 of the present invention can be installed in an electronic device. According to the functions to be realized, the intelligent financial customer service interaction optimization system may include an interaction quality index evaluation module 201, an emotional entropy value calculation module 202, a service fit degree calculation module 203, a session strategy optimization scheme generation module 204 and a scheme optimization adjustment module 205. The modules of the present invention may also be referred to as units, which refer to a series of computer program segments that can be executed by a processor of an electronic device and can complete fixed functions, and are stored in the memory of the electronic device.

[0160] In the embodiments of the present invention, the functions of each module / unit are as follows:

[0161] The interactive quality index evaluation module 201 is used to collect the real-time session data stream and session path when the target customer conducts a session interaction with the intelligent customer service system, parse the session node sequence in the real-time session data stream, divide the interactive feature dimensions corresponding to the session node sequence, calculate the satisfaction attenuation value of the intelligent customer service system based on the interactive feature dimensions, and evaluate the interactive quality index of the session path based on the satisfaction attenuation value;

[0162] The emotional entropy value calculation module 202 is used to extract the voice spectrum features and text semantic vectors in the real-time session data stream, combine the voice spectrum features and the text semantic vectors to draw the emotional fluctuation curve of the target customer during the session, extract the peak semantic segments corresponding to the emotional peaks in the emotional fluctuation curve, and calculate the emotional entropy value of the target customer based on the peak semantic segments;

[0163] The service fit degree calculation module 203 is used to query the type of interactive channel currently used by the target customer, collect the environmental noise data, interface operation heat map and multi-task switching frequency of the interactive channel type, and calculate the service fit degree of the intelligent customer service system in the interactive channel by combining the environmental noise data, the interface operation heat map and the multi-task switching frequency;

[0164] The session strategy optimization scheme generation module 204 is used to construct a dynamic optimization weight matrix of the intelligent customer service system based on the interactive quality index, the emotional entropy value and the service fit degree, and generate a session strategy optimization scheme for the target customer based on the dynamic optimization weight matrix;

[0165] The scheme optimization and adjustment module 205 is used to perform interactive processing on the target customer based on the session strategy optimization scheme, collect customer interaction data during the interactive processing, calculate the customer journey conversion rate corresponding to the session strategy optimization scheme based on the customer interaction data, and optimize and adjust the session strategy optimization scheme based on the customer journey conversion rate to obtain the final session strategy.

[0166] Specifically, each module in the intelligent financial customer service interaction optimization system 200 in the embodiments of the present invention adopts the same technical means as those in the above Figure 1 in the intelligent financial customer service interaction optimization method described, and can produce the same technical effects, which will not be elaborated here.

[0167] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention.

[0168] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An intelligent financial customer service interaction optimization method, characterized in that, The method includes: Collecting the real-time session data stream and session path during the session interaction between the target customer and the intelligent customer service system, parsing the session node sequence in the real-time session data stream, dividing the interaction feature dimensions corresponding to the session node sequence, calculating the satisfaction attenuation value of the intelligent customer service system based on the interaction feature dimensions, and evaluating the interaction quality index of the session path based on the satisfaction attenuation value; Extracting the voice spectrum features and text semantic vectors in the real-time session data stream, combining the voice spectrum features and the text semantic vectors, plotting the emotional fluctuation curve of the target customer during the session, extracting the peak semantic segments corresponding to the emotional peaks in the emotional fluctuation curve, and calculating the emotional entropy value of the target customer based on the peak semantic segments; Querying the type of interaction channel currently used by the target customer, collecting the environmental noise data, interface operation heat map, and multi-task switching frequency of the interaction channel type, and calculating the service fit degree of the intelligent customer service system in the interaction channel by combining the environmental noise data, the interface operation heat map, and the multi-task switching frequency; Constructing a dynamic optimization weight matrix for the intelligent customer service system based on the interaction quality index, the emotional entropy value, and the service fit degree, and generating an optimized session strategy plan for the target customer based on the dynamic optimization weight matrix; Performing interaction processing on the target customer based on the optimized session strategy plan, collecting customer interaction data during the interaction processing, calculating the customer journey conversion rate corresponding to the optimized session strategy plan based on the customer interaction data, and optimizing and adjusting the optimized session strategy plan based on the customer journey conversion rate to obtain the final session strategy.

2. The intelligent financial customer service interaction optimization method according to claim 1, wherein The calculating the satisfaction attenuation value of the intelligent customer service system based on the real-time session data stream includes: Performing data preprocessing on the real-time session data stream to obtain a standardized session data stream; Performing session node division on the standardized session data stream to obtain a session node sequence; Performing feature extraction on the session node sequence to obtain interaction feature dimensions; Calculating the satisfaction score of each session node in the session node sequence based on the interaction feature dimensions; Calculating the satisfaction attenuation value of the intelligent customer service system based on the satisfaction score.

3. The intelligent financial customer service interaction optimization method according to claim 2, characterized in that, The calculating the satisfaction score of each session node in the session node sequence based on the interaction feature dimensions includes: Assigning the dimension weights corresponding to the interaction feature dimensions and performing quantization processing on the interaction feature dimensions to obtain dimension feature values; Performing normalization processing on the dimension feature values to obtain normalized dimension feature values, and querying the dimension ideal values corresponding to the interaction feature dimensions; Calculating the dimension average value and dimension standard deviation corresponding to the interaction feature dimensions by combining the normalized dimension feature values; Calculating the satisfaction score of each session node in the session node sequence through the following formula by combining the dimension weights, the normalized dimension feature values, the dimension ideal values, the dimension average value, and the dimension standard deviation: Among them, A represents the satisfaction score of each session node in the session node sequence, and β a represents the dimension weight corresponding to the a-th dimension in the interaction feature dimension, and B ia represents the normalized dimension eigenvalue corresponding to the a-th dimension of the i-th node in the session node sequence, represents the dimension average value of the a-th dimension, and α a represents the fluctuation adjustment coefficient of the a-th dimension, represents the dimension average value of the a-th dimension, and σ a represents the dimension standard deviation of the a-th dimension. a represents the serial number of the interaction feature dimension, q represents the number of interaction feature dimensions, and i represents the serial number of the session node sequence.

4. The intelligent financial customer service interaction optimization method according to claim 1, wherein Extracting the speech spectral features and text semantic vectors from the real-time session data stream includes: Extracting the session voice signal corresponding to the real-time session data stream, performing time-domain conversion processing on the session voice signal to obtain a voice frequency-domain signal; Extracting the frequency-domain signal features corresponding to the voice frequency-domain signal; Identifying the session text information in the real-time session data stream, and extracting the key session text in the session text information; Performing semantic parsing on the key session text to obtain the key session text semantics; Performing vectorization processing on the key session text semantics to obtain a session text semantic vector; Performing dimensionality reduction processing on the session text semantic vector to obtain a text semantic vector.

5. The intelligent financial customer service interaction optimization method according to claim 1, wherein Calculating the emotional entropy value of the target customer based on the peak semantic segment includes: Extracting the semantic key characters in the peak semantic segment; Based on the semantic key characters, performing emotional classification on the peak semantic segment to obtain the semantic segment emotion; Calculating the emotional probability density corresponding to the semantic segment emotion, and counting the number of emotional categories corresponding to the semantic segment emotion; Combining the emotional probability density and the number of emotional categories, calculating the emotional entropy value of the target customer through the following formula: Among them, D represents the emotional entropy value of the target customer, and E b represents the emotional probability density of the b-th emotion in the semantic segment emotion, b represents the serial number of the semantic segment emotion, and w represents the number of semantic segment emotions.

6. The intelligent financial customer service interaction optimization method according to claim 1, wherein Combining the environmental noise data, the interface operation heat map, and the multi-task switching frequency to calculate the service fitness of the intelligent customer service system in the interaction channel includes: Calculating the decibel value of the environmental noise data to obtain the environmental noise level; Performing region division on the interface operation heat map to obtain a function operation region; Counting the operation heat values corresponding to each region in the function operation region; Performing time window statistics on the multi-task switching frequency to obtain a task switching frequency value; Combining the environmental noise level, the operation heat value, and the task switching frequency value to calculate the service fitness of the intelligent customer service system in the interaction channel.

7. The intelligent financial customer service interaction optimization method according to claim 6, characterized in that Combining the environmental noise level, the operation heat value, and the task switching frequency value to calculate the service fitness of the intelligent customer service system in the interaction channel includes: Performing normalization processing on the environmental noise level, the operation heat value, and the task switching frequency value to obtain a standard noise level value, a standard operation heat value, and a standard switching frequency value; Combining the standard noise level value, the standard operation heat value, and the standard switching frequency value, calculating the service fitness of the intelligent customer service system in the interaction channel through the following formula: Where, F represents the service fitness of the intelligent customer service system in the interaction channel, G represents the standard noise level value, H represents the standard operation heat value, L represents the standard switching frequency value, and ∈ represents a smoothing constant.

8. The intelligent financial customer service interaction optimization method according to claim 1, characterized in that, Calculating the customer journey conversion rate corresponding to the session strategy optimization plan based on the customer interaction data includes: Performing data cleaning on the customer interaction data to obtain target customer interaction data; Identifying the interaction data labels corresponding to the target customer interaction data, and performing stage division on the target customer interaction data to obtain customer journey stage data; Based on the interaction data tags, identify the key journey nodes in the customer journey stage data; Count the number of customer nodes corresponding to the key journey nodes from the customer journey stage data; Based on the number of customer nodes, calculate the customer journey conversion rate corresponding to the session strategy optimization plan.

9. The intelligent financial customer service interaction optimization method according to claim 8, wherein, The step of identifying the key journey nodes in the customer journey stage data based on the interaction data tags includes: Identify the customer behavior identifiers in the interaction data tags, and extract the customer behavior characteristics corresponding to the customer behavior identifiers; Calculate the feature similarity between the customer behavior characteristics; Based on the feature similarity, perform feature clustering processing on the customer behavior characteristics to obtain clustered behavior characteristics; Analyze the key behaviors corresponding to the clustered behavior characteristics, and perform node mapping on the key behaviors to obtain the key journey nodes in the customer journey stage data.

10. An intelligent financial customer service interaction optimization system, characterized in that, The system includes: An interaction quality index evaluation module, which is used to collect the real-time session data stream and session path when the target customer conducts a session interaction with the intelligent customer service system, parse the session node sequence in the real-time session data stream, divide the interaction feature dimensions corresponding to the session node sequence, and calculate the satisfaction attenuation value of the intelligent customer service system based on the interaction feature dimensions. Based on the satisfaction attenuation value, evaluate the interaction quality index of the session path; An emotional entropy value calculation module, which is used to extract the voice spectrum features and text semantic vectors in the real-time session data stream, combine the voice spectrum features and the text semantic vectors, draw the emotional fluctuation curve of the target customer during the session, extract the peak semantic segments corresponding to the emotional peaks in the emotional fluctuation curve, and calculate the emotional entropy value of the target customer based on the peak semantic segments; A service fit calculation module, which is used to query the type of interaction channel currently used by the target customer, collect the environmental noise data, interface operation heat map and multi-task switching frequency of the interaction channel type, and calculate the service fit of the intelligent customer service system in the interaction channel by combining the environmental noise data, the interface operation heat map and the multi-task switching frequency; A session strategy optimization plan generation module, which is used to construct a dynamic optimization weight matrix of the intelligent customer service system based on the interaction quality index, the emotional entropy value and the service fit, and generate a session strategy optimization plan for the target customer based on the dynamic optimization weight matrix; A plan optimization and adjustment module, which is used to perform interaction processing on the target customer based on the session strategy optimization plan, collect customer interaction data during the interaction processing, calculate the customer journey conversion rate corresponding to the session strategy optimization plan based on the customer interaction data, and optimize and adjust the session strategy optimization plan based on the customer journey conversion rate to obtain the final session strategy.

Images