Customer behavior characteristic analysis method and system based on voice recognition technology

Abstract

The invention relates to the technical field of intelligent voice analysis, in particular to a customer behavior characteristic analysis method and system based on a voice recognition technology. According to the method, customer voice data are acquired, a voice processing model is constructed, voice features and emotion data are analyzed and extracted, customer concern points are determined, and a voice-emotion association model is established; constructing a client intention prediction model based on the concern semantics in combination with the interaction parameters to carry out first prediction; analyzing dynamic characteristics of semantics and emotions in combination with the first prediction to obtain a semantics-emotion double-drive behavior mode for second prediction; constructing a customer behavior model through the two-dimensional evaluation matrix; and finally, a time-varying-based customer value prediction model is obtained in combination with historical interaction. According to the method, the problems of splitting of customer emotion recognition and semantic understanding, insufficient intention prediction precision, static behavior characteristic analysis, single customer value evaluation and the like in the prior art are solved, and comprehensive capture and accurate analysis of dominant and recessive behavior characteristics in a customer interaction process are realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent voice analysis technology, and in particular to a method and system for analyzing customer behavior characteristics based on voice recognition technology. Background Technology

[0002] With the rapid development of natural language processing and speech recognition technologies, voice-interaction-based customer service systems have gradually become an important tool for enterprises to improve customer experience. Current speech recognition technology has evolved from simple command recognition to complex semantic understanding, demonstrating broad application prospects in the customer service field. Traditional speech recognition systems primarily focus on transcribing speech signals and extracting keywords, using acoustic and language modeling to convert speech to text, which is then applied to automatic responses and classification in customer service centers. However, these systems still have significant limitations in emotion recognition and intent understanding, struggling to capture emotional changes and semantic connections during customer interactions, resulting in static analysis of customer behavior characteristics. Especially in multi-turn interaction scenarios, traditional systems lack a deep understanding of the coupling relationship between customer emotions and semantics, failing to accurately predict and dynamically respond to customer intent. Furthermore, existing technologies for customer value assessment often rely on explicit transaction data, ignoring the implicit behavioral characteristics inherent in voice interactions, making it difficult for enterprises to build comprehensive customer profiles and predict potential value.

[0003] In view of the problems existing in the existing technology, such as the separation of customer emotion recognition and semantic understanding, insufficient accuracy of intent prediction, static behavioral feature analysis, and single customer value assessment, this invention provides a customer behavior feature analysis method and system based on speech recognition technology. It aims to achieve comprehensive capture and accurate analysis of explicit and implicit behavioral features in customer interaction by constructing a series of technical means such as a speech-emotion association model, a customer intent prediction model, a semantic-emotion dual-driven behavior mode, and a customer behavior model. Based on a time-varying customer value prediction model, it dynamically evaluates the potential value of customers, thereby providing enterprises with more accurate customer insights and service decision support. Summary of the Invention

[0004] In view of the problems existing in the prior art, the present invention is proposed.

[0005] Therefore, the problem to be solved by this invention is how to solve the problems of the separation between customer emotion recognition and semantic understanding, insufficient accuracy of intent prediction, static behavioral feature analysis, and single customer value assessment in the prior art. It realizes the comprehensive capture and accurate analysis of explicit and implicit behavioral features in the customer interaction process, and improves the accuracy of customer intent prediction and the comprehensiveness of customer value assessment.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, embodiments of the present invention provide a customer behavior feature analysis method based on speech recognition technology, which includes acquiring customer speech data and constructing a speech processing model; The speech processing model is used to analyze and extract speech features and emotional data, determine customer concerns, and establish a speech-emotion association model. A customer intent prediction model is dynamically obtained by combining focus semantics with interaction parameters, and is used for the first prediction of customer needs. Combining the first prediction, the dynamic characteristics of the semantics and emotions are analyzed to obtain the semantic-emotion dual-driven behavior pattern, and a second prediction of customer needs is made. A customer behavior model is obtained based on a two-dimensional evaluation matrix and by combining the coupling characteristics of semantics and emotion. By combining the customer behavior model with historical interactions, a time-varying customer value prediction model is obtained to predict customer value.

[0007] As a preferred embodiment of the customer behavior feature analysis method based on speech recognition technology described in this invention, the method involves: analyzing and extracting speech features and emotional data through the speech processing model to determine customer focus points, including: By analyzing speech content using a speech processing model, we obtain semantic and sentiment distribution maps under the influence of intonation. Identify the areas where semantics and emotions are highly concentrated; these areas represent the customer's focus.

[0008] As a preferred embodiment of the customer behavior feature analysis method based on speech recognition technology described in this invention, the step of establishing a speech-emotion association model includes: Based on the speech processing model, the speech content is analyzed. By setting intonation parameters, speech rate change rate, speech-semantic association correction coefficient and real-time sentiment value, the sentiment intensity model is obtained. Based on speech processing models, speech content is analyzed, and a semantic strength model is obtained by combining semantic weight index, emotion-related semantic strength, expression mode features, and real-time context.

[0009] As a preferred embodiment of the customer behavior feature analysis method based on speech recognition technology described in this invention, the customer intent prediction model is dynamically obtained by combining attention point semantics with interaction parameters, including: Based on the semantic value of the current focus, a weighted sum of the semantic difference values ​​within the historical interaction segment is superimposed, and error correction is performed; Introduce an acceleration term for the coupling of emotion and context; Based on the intent change function, we can determine the trend characteristics of changes in customer needs in order to obtain customer intent prediction results.

[0010] As a preferred embodiment of the customer behavior feature analysis method based on speech recognition technology described in this invention, the method includes: analyzing the dynamic characteristics of semantics and emotion to obtain a semantic-emotion dual-driven behavior pattern, and performing a second prediction of customer needs, including: The interaction matrix of the points of interest is obtained through interaction sequence analysis; the intention matching probability matrix is ​​obtained based on semantic-sentiment mapping. Based on the dual-matrix quantification of the impact of sentiment changes on customer behavior, sentiment correction terms are obtained; By setting a semantic enhancement coefficient that incorporates emotion, the impact of semantic changes on customer decisions is quantified, resulting in a semantic enhancement correction term. Based on two correction terms, and combined with the baseline reference intent, the dual-drive behavior pattern is obtained. Based on the dual-drive behavior pattern, different response strategies are set and compared with actual interactions to predict and respond to demand.

[0011] As a preferred embodiment of the customer behavior feature analysis method based on speech recognition technology described in this invention, the following steps are included: obtaining a customer behavior model based on a two-dimensional evaluation matrix and combining the coupling characteristics of semantics and emotion, including: obtaining a time-varying customer value prediction model based on the customer behavior model and combining it with historical interactions, including: The potential capacity of customer value can be obtained through customer behavior models; Customer value prediction is obtained by comparing potential capacity with the rate of change of value.

[0012] Based on the actual number of interactions in different interaction scenarios and the effective number of interactions in the corresponding scenarios, combined with the coupling index of emotion and semantics, an evaluation of the interaction effect is obtained. Simultaneously, semantic enhancement functions are introduced to assess the impact of emotional changes on customer behavior; The customer behavior model is obtained by summing the interaction effect evaluation and the impact evaluation.

[0013] As a preferred embodiment of the customer behavior feature analysis method based on speech recognition technology described in this invention, wherein: Secondly, embodiments of the present invention provide a customer behavior feature analysis system based on speech recognition technology, which includes an acquisition module for acquiring customer speech data and constructing a speech processing model. The first calculation module is used to analyze and extract voice features and emotional data through the voice processing model, determine customer focus points, and establish a voice-emotion association model. The first model construction prediction module is used to dynamically obtain a customer intent prediction model by combining the semantics of the focus point with the interaction parameters, and is used for the first prediction of customer needs. The second model construction prediction module is used to combine the first prediction, analyze the dynamic characteristics of the semantics and emotions, obtain the semantic-emotion dual-driven behavior pattern, and perform the second prediction of customer needs. The third model construction prediction module is used to obtain a customer behavior model based on a two-dimensional evaluation matrix and the coupling characteristics of semantics and sentiment. The fourth model construction prediction module is used to predict customer value by combining the customer behavior model with historical interactions to obtain a time-varying customer value prediction model.

[0014] Thirdly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, they implement the steps of the customer behavior feature analysis method based on speech recognition technology as described in the first aspect of the present invention.

[0015] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, they implement the steps of the customer behavior feature analysis method based on speech recognition technology as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: By acquiring customer voice data and constructing a voice processing model, and combining intonation parameters, speech rate change rate, voice-semantic association correction coefficient, and real-time sentiment value, a voice-emotion association model is established, realizing the joint analysis of semantics and emotion under the influence of intonation. This association model breaks through the limitation of the separation between semantic understanding and emotion recognition in traditional voice analysis, and can accurately capture the explicit semantic content and implicit emotional state of customers in voice interaction. This enables the system to simultaneously understand "what the customer said" and "how the customer said it," thereby providing a more comprehensive information foundation for subsequent intent understanding and improving the accuracy of customer focus identification.

[0017] By dynamically combining semantic focus with interaction parameters, and employing techniques such as weighted summation and error correction of semantic differences within historical interaction segments, acceleration terms coupled with sentiment and context, and intent change functions, a customer intent prediction model was constructed and first-round predictions were achieved. This model, through advanced techniques such as adaptive difference operators and dynamically updated time-varying autoregressive coefficients using Bayesian optimization methods, effectively captures the dynamic changes in customer intent, improving prediction accuracy by 20%–30% compared to traditional static prediction models. It solves the technical challenge of real-time tracking of customer intent in multi-turn interaction scenarios, enabling the system to anticipate changes in customer needs and providing a basis for intelligent service strategy adjustments.

[0018] Based on the initial prediction results, an interaction matrix and an intent matching probability matrix were established through interaction sequence analysis and semantic-sentiment mapping. A semantic-sentiment dual-driven behavioral model was constructed by combining sentiment correction terms and semantic enhancement correction terms, enabling a second prediction of customer needs. This behavioral model quantifies the impact of actual emotional fluctuations through dynamic sentiment integration, replacing the traditional linear sentiment scoring method. It accurately captures behavioral changes caused by sudden emotional shifts and models the interaction between semantics and sentiment through a nonlinear coupling function. This allows the system to accurately predict the behavioral decision-making path of customers facing complex emotional changes, significantly improving prediction accuracy in high-sentiment-fluctuation scenarios and providing a scientific basis for personalized service strategy development.

[0019] By employing a dual-dimensional evaluation matrix, combining the ratio of actual to effective interactions in different interaction scenarios, and utilizing techniques such as the coupling index of emotion and semantics, and semantic enhancement functions, a comprehensive customer behavior model was constructed. This model surpasses traditional single-dimensional behavior evaluation methods. Through innovative techniques such as introducing an exponential decay function to quantify the non-linear impact of semantic intensity on interaction efficiency, and using the sigmoid function to assign higher weight to emotional fluctuations exceeding thresholds, it achieves a multi-dimensional and dynamic comprehensive evaluation of customer behavior. This model can accurately identify changes in customer behavior patterns in complex interaction scenarios, enhancing the system's ability to recognize customer behavioral characteristics and laying the foundation for accurate customer profiling.

[0020] By combining customer behavior models with historical interaction data and employing the ratio of potential capacity to value change rate, a time-varying customer value prediction model was established, enabling dynamic prediction of customer value. This prediction model adjusts the non-linear effects of behavioral assessment through an emotion correction index and comprehensively considers the rate and acceleration of behavioral assessment to predict future value trends. It overcomes the limitations of traditional customer value assessment, which relies solely on explicit transaction data. It can uncover implicit value signals from voice interactions, identify high-potential customers and those at risk of churn in advance, and enable businesses to implement differentiated customer maintenance strategies. This improves customer retention and lifetime value while reducing customer acquisition costs, providing data support for precise marketing and optimized resource allocation. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart of a customer behavior feature analysis method based on speech recognition technology; Figure 2 A diagram of a computer device used for customer behavior feature analysis based on speech recognition technology. Detailed Implementation

[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0024] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0025] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments. Example 1

[0026] Reference Figures 1-2 This is the first embodiment of the present invention, which provides a customer behavior feature analysis method based on speech recognition technology, including: S100: Acquire customer voice data and build a voice processing model; S200: By analyzing and extracting voice features and emotional data through a voice processing model, customer focus is determined, and a voice-emotion association model is established; S300: A customer intent prediction model is dynamically obtained by combining the semantics of the focus with the interaction parameters, and is used for the first prediction of customer needs. S400: Combining the first prediction, analyze the dynamic characteristics of semantics and sentiment to obtain a semantic-sentiment dual-driven behavior pattern, and make a second prediction of customer needs. S500: A customer behavior model is obtained based on a two-dimensional evaluation matrix and by combining the coupling characteristics of semantics and emotion. S600: By combining customer behavior models with historical interactions, a time-varying customer value prediction model is obtained to predict customer value.

[0027] It should be noted that the voice interaction environment in customer service centers is highly complex. Customer voice data contains rich explicit and implicit information, including directly expressed semantic content and emotional states indirectly conveyed through tone and speed. Existing customer service systems often separate semantic understanding from sentiment analysis, failing to effectively capture the dynamic correlation between semantics and emotion in customer voice. Especially during multi-turn interactions, the customer's focus and emotional state constantly change with the content of the interaction, leading to significant deviations in predicting customer intent and hindering timely adjustments to service strategies to meet customer needs. Furthermore, traditional customer value assessment relies primarily on explicit transaction data, neglecting the behavioral characteristics and preference information inherent in voice interactions, resulting in a one-sided and static customer value assessment that fails to achieve dynamic optimization and accurate prediction of customer value.

[0028] Therefore, addressing the aforementioned issues in speech analysis and value prediction, this solution utilizes steps S100-S600 to construct a speech processing model that analyzes customer speech content. This yields semantic and emotional distribution maps and relationships under the influence of intonation, enabling accurate identification of customer focus points and comprehensive modeling of speech-emotional connections. It also allows for real-time prediction of customer intent, dynamic analysis of changing customer needs, and timely responses to shifts in intent. Furthermore, based on a semantic-emotional dual-driven behavioral model and a customer value prediction model, it achieves accurate assessment and prediction of potential customer value. This comprehensive technical solution overcomes the limitations of semantic understanding and sentiment analysis, enabling comprehensive capture and accurate analysis of customer behavioral characteristics during interactions, providing enterprises with more comprehensive customer insights and service decision support.

[0029] Example 2 Reference Figure 1 - Figure 2 This is the second embodiment of the present invention.

[0030] In this embodiment of the application, in step S100, customer voice data is acquired and a voice processing model is constructed. In an optional embodiment, the voice processing model constructed in step S100 can also be analyzed by a Mixture of Experts model, which assigns different types of features to specialized sub-networks for processing, such as expert networks for speech recognition, emotion recognition, and intonation analysis. Finally, the output results of each expert network are integrated through a gating mechanism to improve the accuracy of analysis in complex scenarios.

[0031] In another optional implementation, the voice processing model in step S100 can also be constructed by using a multimodal fusion method to integrate voice signals with user information from other channels (such as text chat records, historical interaction data, etc.), and use an attention mechanism to perform weighted fusion of different modal information to construct a more comprehensive customer feature representation.

[0032] In this embodiment of the application, step S200 involves analyzing and extracting speech features and emotional data using a speech processing model, including the following steps B1-B4: B1: Analyze and extract speech features and emotional data using a speech processing model to determine customer concerns, including: By analyzing speech content using a speech processing model, we obtain semantic and sentiment distribution maps under the influence of intonation. Specifically, in B1, the emotional intensity model is represented as follows: ; In the formula, , Actual / basic emotional intensity (quantitative value); The real-time pitch deviation value (Hz) can be dynamically updated based on audio features; The dynamic calibration coefficients related to the emotional baseline can be fitted through an emotion-intonation experiment; The rate of change of speech rate; These are the speech-semantic association correction coefficients, which can be calibrated experimentally. This is the semantic context offset; This is the partial derivative of emotion with respect to semantics, reflecting the sensitivity of semantic content changes to emotion; This is a pitch fluctuation influencing factor, which can be dynamically corrected by a neural network model; Standard pitch baseline; Real-time pitch; It is the natural logarithm function.

[0033] It should be noted that, The term in the formula is the dynamic speech-semantic coupling term. To describe the cumulative effect of pitch changes on emotional intensity, for example, a sudden increase in pitch often indicates an increase in surprise or anger.

[0034] B2: Identify the areas where semantics and emotions are highly concentrated; these areas represent the customer's focus.

[0035] Specifically, in B2, the semantic strength model is represented as follows: ; In the formula, , For actual / basic semantic strength; , It is a semantic weight index, which can be calibrated by the semantic network; For the intensity of emotion-related semantics; For the characteristic coefficients of the expression mode; The difference in emotional gradient; This is the context relevance ratio; Standard keyword density; This represents the real-time keyword density.

[0036] It should be noted that in the formula It is mainly used to describe the semantic intensity amplification effect caused by changes in keyword density; in the formula Primarily used to calculate the emotional gradient The resulting semantic reinforcement, through coupling the expression method with the sentiment baseline, accurately quantifies the semantic intensity, laying the foundation for subsequent customer intent prediction.

[0037] B3: Establish a voice-emotion association model, including: Based on the speech processing model, the speech content is analyzed. By setting intonation parameters, speech rate change rate, speech-semantic association correction coefficient and real-time sentiment value, the sentiment intensity model is obtained. B4: Based on the speech processing model, the speech content is analyzed, and the semantic strength model is obtained by combining the semantic weight index, the emotional related semantic strength, the expression mode features, and the real-time context.

[0038] In one optional implementation, the establishment of the speech-emotion association model in step S200 may also include a deep cross-attention mechanism to establish a fine-grained association between speech and emotion features. This mechanism performs multi-level cross-attention calculations on the extracted speech feature sequences and emotion feature sequences, automatically learns the contribution weights of different speech features to various emotional states, and forms a more refined and personalized speech-emotion mapping relationship.

[0039] It should be noted that the above-mentioned speech-emotion association model introduces a three-dimensional coupling term of speech-semantics-emotion, which avoids the limitation of relying solely on semantic content in traditional sentiment analysis. Key parameters such as intonation parameters and semantic strength adopt dynamic association functions instead of fixed thresholds. At the same time, the temporal characteristics of the interaction process are quantified, and the dynamic characteristics of the context are considered, providing an accurate basis for judgment in subsequent steps.

[0040] In this embodiment of the application, step S300 dynamically obtains the customer intent prediction model by combining the semantics of the focus with the interaction parameters, including the following steps C1-C3: C1: A customer intent prediction model is dynamically derived by combining attention semantics with interaction parameters, including: Based on the semantic value of the current focus, a weighted sum of the semantic difference values ​​within the historical interaction segment is superimposed, and error correction is performed; C2: Introduce an acceleration term for the coupling of emotion and context; C3: Based on the intent change function, determine the trend characteristics of changes in customer needs to obtain customer intent prediction results.

[0041] The first prediction is to predict the customer's current core intent, which is done by combining the steps of S200 with the customer intent prediction model.

[0042] Specifically, the customer intent prediction model for C1-C3 can be represented by the following formula: ; In the formula, For the future Intent prediction value for each interaction round; For the current moment The actual intention vector; Let be the order of the autoregressive term; These are time-varying autoregressive coefficients, dynamically updated by a Bayesian optimization method; For adaptive difference operators, d is the dynamically adjusted difference order; i is the interaction round index; The moving average coefficients are optimized in real time through an online learning algorithm; The intention offset acceleration factor is fitted from historical interaction data: This is the emotion-context coupling coefficient, associated with real-time emotion monitoring values; This is an indicator function for changing intention trends, used to capture the rate of intention transformation characteristics; This represents the prediction error term for historical rounds; This is to describe the nonlinear amplification effect of emotional factors on changes in intention.

[0043] For example, some coupling coefficients can be calculated as follows: ; In the formula, The stability quantification index of the arbitrary graph is dynamically adjusted. In one optional implementation, step S300 dynamically obtains a customer intent prediction model by combining attention semantics with interaction parameters. Alternatively, an end-to-end intent sequence prediction model can be constructed based on a bidirectional long short-term memory network (Bi-LSTM) combined with an attention mechanism. This method uses the customer's historical interaction content, emotional change trajectory, and contextual features as input sequences, uses a bidirectional LSTM to capture long- and short-term dependencies, and automatically learns the importance weights of features at different time points through a multi-head attention mechanism, thereby achieving high-precision prediction of changes in customer intent during multiple rounds of interaction.

[0044] In another optional implementation, step S300 dynamically obtains a customer intent prediction model by combining attention semantics with interaction parameters. Alternatively, a knowledge graph of customer intent evolution can be modeled using a graph neural network (GNN). This method constructs a dynamic knowledge graph from customer attention points, emotional states, intents, and related knowledge concepts, with edges representing the strength of their associations. It learns node representation vectors through a graph convolutional network or graph attention network, activates relevant nodes based on the current interaction content, and predicts the most likely subsequent intent node by passing the activation value. This method is particularly suitable for professional customer service scenarios with complex logical relationships.

[0045] In this embodiment of the application, step S400, which analyzes the dynamic characteristics of semantics and emotion, includes the following steps D1-D5: D1: Analyze the dynamic characteristics of semantics and sentiment to obtain a semantic-sentiment dual-driven behavioral pattern, and perform secondary prediction of customer needs, including: The interaction matrix of the points of interest is obtained through interaction sequence analysis; the intention matching probability matrix is ​​obtained based on semantic-sentiment mapping. D2: Based on the dual-matrix quantification of the impact of sentiment changes on customer behavior, sentiment correction terms are obtained; D3: By setting a semantic enhancement coefficient that incorporates emotion, the impact of semantic changes on customer decisions is quantified to obtain semantic enhancement correction terms; D4: Based on two correction terms and combined with the baseline reference intent, the dual-drive behavior pattern is obtained; D5: Based on the dual-drive behavior pattern, different response strategies are set and compared with actual interactions to predict and respond to demand.

[0046] Specifically, the semantic-emotional dual-drive behavioral pattern of D1-D4 can be represented by the following formula: ; In the formula, The interaction matrix represents the transformation probability, i.e.: The actual conversion probability is relative to the reference conversion probability in the intent matching probability matrix (i.e.: Behavioral biases related to the expected conversion probability; This indicates the effect of emotional changes on behavior; Introducing the moderating effect of semantic intensity S on the emotion-behavior relationship demonstrates the nonlinear coupling between semantics and emotion in behavior-driven processes. This represents the difference between the current semantic strength and the baseline value. Indicates emotional-semantic decision sensitivity; For example, the above parameters can be specifically expressed as follows: ; In the formula, It serves as an emotional activating energy and can be used for calibration in emotional behavior experiments; For real-time sentiment change rate; ; In the formula, , Cognitive constants calibrated by experiments; The critical semantic strength; ; In the formula, The threshold for emotional shift; The emotional sensitivity width can be determined through user behavior testing.

[0047] It should be noted that the aforementioned behavioral pattern model is constructed by quantifying the impact of actual emotional fluctuations through dynamic emotional integrals, thereby replacing the traditional linear emotional scoring method. This can solve the problem of predicting behavioral changes caused by abrupt emotional shifts. The function separates the semantic enhancement effect from the emotion-driven effect; it can capture the additional decision-making changes caused by sudden emotional shifts in the influence of emotion on behavior. The weight of semantic correction can be automatically enhanced by using an error function.

[0048] For example, in D5, comparing different response strategies based on the dual-drive behavior pattern with actual interactions can be represented as follows: like Then, precise demand response will be implemented (i.e., more direct solutions will be adopted). like In this case, a guided demand response will be implemented (i.e., multiple possible solutions will be provided for selection). like If the demand is not yet clear, it is judged to be an exploratory demand (i.e., the customer's own needs are not yet clear), and information is supplemented and the demand is guided.

[0049] In an optional implementation, step S400 analyzes the dynamic characteristics of semantics and emotion to obtain a semantic-emotion dual-driven behavior pattern. Furthermore, a multi-agent system can be used to simulate the customer's cognitive decision-making process, constructing a customer behavior model based on a Belief-Desire-Intention (BDI) architecture. This method formally represents the customer's beliefs (cognition of reality), desires (desired goals), and intentions (action plans to achieve those goals), and updates these representations through semantic and emotional features. It simulates possible behavioral decision-making paths for customers in different situations, thereby constructing a behavior prediction model that better aligns with human cognitive processes.

[0050] In another optional implementation, step S400 analyzes the dynamic characteristics of semantics and emotion to obtain a semantic-emotion dual-driven behavior pattern. A customer behavior strategy learning model can also be established using reinforcement learning methods. The customer service process is viewed as a Markov decision process, where the customer's semantic expression and emotional state constitute the state space, and the customer's various behavioral choices constitute the action space. By using algorithms such as Q-learning or policy gradient, the behavioral policy functions of customers in different states are learned from historical interaction data and used to predict customer behavior in future interactions. This approach is particularly suitable for customer service scenarios with clear objectives (such as purchase decisions, problem solving, etc.).

[0051] In this embodiment of the application, step S500, based on a two-dimensional evaluation matrix, includes the following steps E1-E2: E1: Based on a two-dimensional evaluation matrix, and combining the coupling characteristics of semantics and emotion, a customer behavior model is obtained, including: Based on the actual number of interactions in different interaction scenarios and the effective number of interactions in the corresponding scenarios, combined with the coupling index of emotion and semantics, an evaluation of the interaction effect is obtained. E2: Simultaneously, a semantic enhancement function is introduced to assess the impact of emotional changes on customer behavior; E3: The customer behavior model is obtained by summing the interaction effect evaluation and the impact evaluation.

[0052] Specifically, the customer behavior model for E1-E3 can be represented by the following formula: ; In the formula, The overall customer behavior assessment value ranges from 0 to 1, with values ​​closer to 1 indicating more proactive customer behavior; m represents the number of interaction scenario classification dimensions, such as demand type; n represents the number of interaction scenario classification dimensions, such as service stage; i represents the corresponding scenario category index; and j represents the interaction stage index. This represents the actual number of interactions in scenario (i, j); Let (i, j) represent the number of valid interactions within the scenario. For emotions Semantic coupling correction index; The semantic change influence coefficient; The rate of change of semantic intensity; This represents the number of valid interactions under the current semantic strength. It is a time variable.

[0053] For example, some of the above parameters can be expressed as: ; Where E > emotional threshold and S < 0.75 hour, The value is automatically amplified by 1.5 to 2 times; ; Among them, the number of effective interactions under the current semantic strength An exponential decay function is introduced to quantify the nonlinear impact of semantic strength on interaction efficiency; ; This part assigns higher weight to emotional fluctuations exceeding the threshold through the sigmoid function; ; ; In an alternative implementation, if S400 simulates the customer's cognitive decision-making process using a multi-agent system, then S500 can construct a personalized customer cognitive style model based on computational cognitive modeling. This method combines cognitive psychology theory to parametrically model the customer's cognitive characteristics, such as attention allocation, information processing style, and decision-making preferences. Personalized parameters are trained based on behavioral features extracted from voice interaction, simulating the differentiated behavioral responses that customers with different cognitive styles may produce when facing the same situation, thus achieving more precise personalized service strategy formulation.

[0054] In this embodiment of the application, the customer value prediction in step S600 includes the following steps F1-F2: F1: By combining customer behavior models with historical interactions, a time-varying customer value prediction model is obtained, including: The potential capacity of customer value can be obtained through customer behavior models; F2: Customer value forecasting is obtained by comparing the potential capacity with the rate of change of value.

[0055] Specifically, the time-varying customer value prediction model of F1-F2 can be represented by the following formula: ; In the formula, Forecasted customer value over time; The cumulative behavioral assessment value is derived from the customer behavior model; For sentiment correction index; Cumulative rate of behavioral evaluation represents the rate of increase of behavioral evaluation at the current moment. The acceleration of behavioral assessment indicates the increasing trend of the rate of behavioral assessment. For behavioral acceleration weighting coefficients; This indicates that only non-negative values ​​of the behavioral assessment acceleration are considered; For example, some of the above parameters can be expressed as: ; ; Triggering high-value customer tagging It should be noted that the above model... To quantify the potential value of customers in their current behavioral assessment state; The nonlinear effect of behavioral assessment is adjusted according to the intensity of emotion; the denominator of the above model is the rate of change of value, which combines the current rate of behavioral assessment with acceleration to predict future customer value trends.

[0056] For example, customers at the emotional baseline Next interaction; Current behavioral assessment B(t) = 0.7; Remaining capacity: .

[0057] Rate of change in behavioral assessment:

[0058] Denominator calculation: 0.12 + 1.2 * 0.04 = 0.168 / month Customer value forecast:

[0059] It should be noted that this model quantifies the nonlinear impact of emotional state on customer value by dynamically adjusting potential capacity and value change rate, and combines behavioral assessment to accelerate early warning of high-value customer churn risk, providing real-time and adaptive prediction results for customer value management.

[0060] In an alternative implementation, a Joint Lifetime Value (LTV) model can be used to integrate and analyze a customer's current value with their potential future value. This method comprehensively considers multiple value indicators such as cross-selling opportunities, brand promotion value, and social network influence, and combines these with behavioral trajectories predicted by customer behavior models to construct a time-series-based panoramic view of customer value, providing enterprises with a more comprehensive customer value assessment framework.

[0061] In another optional implementation, if a customer behavior strategy learning model is established in S400 using reinforcement learning, then in S600, a deep reinforcement learning model can be trained using multi-source customer interaction data, such as voice feature sequences, emotional change trajectories, and service response records, to directly predict the customer's value return function under different service strategies, thereby achieving joint optimization of service strategies and customer value.

[0062] In summary, this invention constructs a customer behavior feature analysis method and system based on speech recognition technology. Through in-depth analysis of customer speech data, it accurately identifies customer concerns and establishes a speech-emotion association model, achieving comprehensive capture of explicit semantics and implicit emotions in customer speech. Based on customer concerns combined with interaction parameters, it dynamically constructs a customer intent prediction model, improving the accuracy of intent prediction. Through a semantic-emotion dual-driven behavior pattern and a time-varying customer value prediction model, it achieves comprehensive analysis of customer behavior characteristics and accurate value prediction. This method overcomes the problems of traditional customer service systems, such as the separation of semantic understanding and sentiment analysis, insufficient intent prediction accuracy, static behavioral feature analysis, and singular customer value assessment. It can provide enterprises with more accurate customer insights and service decision support, improve customer experience, optimize service efficiency, and enhance enterprise value.

[0063] Example 3 The above is an illustrative scheme of a customer behavior feature analysis method based on speech recognition technology. It should be noted that the technical solution of this customer behavior feature analysis system based on speech recognition technology and the technical solution of the aforementioned customer behavior feature analysis method based on speech recognition technology belong to the same concept. Details not described in detail in this embodiment of the customer behavior feature analysis system based on speech recognition technology can be found in the description of the aforementioned customer behavior feature analysis method based on speech recognition technology.

[0064] This embodiment also provides a customer behavior feature analysis system based on speech recognition technology, including: The acquisition module is used to acquire customer voice data and build a voice processing model; The first calculation module is used to analyze and extract voice features and emotional data through a voice processing model, determine customer focus points, and establish a voice-emotion association model. The first model construction prediction module is used to dynamically obtain a customer intent prediction model by combining the semantics of the focus point with the interaction parameters, and is used for the first prediction of customer needs. The second model constructs a prediction module, which combines the first prediction to analyze the dynamic characteristics of semantics and sentiment, obtains a semantic-sentiment dual-driven behavior pattern, and performs a second prediction of customer needs. The third model construction prediction module is used to obtain a customer behavior model based on a two-dimensional evaluation matrix and the coupling characteristics of semantics and sentiment. The fourth model construction prediction module is used to predict customer value by combining customer behavior models with historical interactions to obtain a time-varying customer value prediction model.

[0065] This embodiment also provides an electronic device suitable for customer behavior feature analysis based on speech recognition technology, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the customer behavior feature analysis method based on speech recognition technology as proposed in the above embodiment.

[0066] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the customer behavior feature analysis method based on speech recognition technology as proposed in the above embodiments.

[0067] The storage medium proposed in this embodiment and the customer behavior feature analysis method based on speech recognition technology proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0068] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0069] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for analyzing customer behavior characteristics based on speech recognition technology, characterized in that: include, Acquire customer voice data and build a voice processing model; The speech processing model is used to analyze and extract speech features and emotional data, determine customer concerns, and establish a speech-emotion association model. A customer intent prediction model is dynamically obtained by combining focus semantics with interaction parameters, and is used for the first prediction of customer needs. Combining the first prediction, the dynamic characteristics of the semantics and emotions are analyzed to obtain the semantic-emotion dual-driven behavior pattern, and a second prediction of customer needs is made. A customer behavior model is obtained based on a two-dimensional evaluation matrix and by combining the coupling characteristics of semantics and emotion. By combining the customer behavior model with historical interactions, a time-varying customer value prediction model is obtained to predict customer value.

2. The customer behavior feature analysis method based on speech recognition technology as described in claim 1, characterized in that: The speech processing model is used to analyze and extract speech features and emotional data to determine customer concerns, including: By analyzing speech content using a speech processing model, we obtain semantic and sentiment distribution maps under the influence of intonation. Identify the areas where semantics and emotions are highly concentrated; these areas represent the customer's focus.

3. The customer behavior feature analysis method based on speech recognition technology as described in claim 2, characterized in that: The establishment of the voice-emotion association model includes: Based on the speech processing model, the speech content is analyzed. By setting intonation parameters, speech rate change rate, speech-semantic association correction coefficient and real-time sentiment value, the sentiment intensity model is obtained. Based on speech processing models, speech content is analyzed, and a semantic strength model is obtained by combining semantic weight index, emotion-related semantic strength, expression mode features, and real-time context.

4. The customer behavior feature analysis method based on speech recognition technology as described in claim 3, characterized in that: A customer intent prediction model is dynamically derived by combining attention semantics with interaction parameters, including: Based on the semantic value of the current focus, a weighted sum of the semantic difference values ​​within the historical interaction segment is superimposed, and error correction is performed; Introduce an acceleration term for the coupling of emotion and context; Based on the intent change function, we can determine the trend characteristics of changes in customer needs in order to obtain customer intent prediction results.

5. The customer behavior feature analysis method based on speech recognition technology as described in claim 4, characterized in that: Analyzing the dynamic characteristics of semantics and emotion yields a semantic-emotion dual-driven behavioral pattern, enabling secondary prediction of customer needs, including: The interaction matrix of the points of interest is obtained through interaction sequence analysis; the intention matching probability matrix is ​​obtained based on semantic-sentiment mapping. Based on the dual-matrix quantification of the impact of sentiment changes on customer behavior, sentiment correction terms are obtained; By setting a semantic enhancement coefficient that incorporates emotion, the impact of semantic changes on customer decisions is quantified, resulting in a semantic enhancement correction term. Based on two correction terms, and combined with the baseline reference intent, the dual-drive behavior pattern is obtained. Based on the dual-drive behavior pattern, different response strategies are set and compared with actual interactions to predict and respond to demand.

6. The customer behavior feature analysis method based on speech recognition technology as described in claim 5, characterized in that: Based on a two-dimensional evaluation matrix, and combining the coupling characteristics of semantics and emotion, a customer behavior model is obtained, including: Based on the actual number of interactions in different interaction scenarios and the effective number of interactions in the corresponding scenarios, combined with the coupling index of emotion and semantics, an evaluation of the interaction effect is obtained. Simultaneously, semantic enhancement functions are introduced to assess the impact of emotional changes on customer behavior; The customer behavior model is obtained by summing the interaction effect evaluation and the impact evaluation.

7. The customer behavior feature analysis method based on speech recognition technology as described in claim 6, characterized in that: By combining the aforementioned customer behavior model with historical interactions, a time-varying customer value prediction model is obtained, including: The potential capacity of customer value can be obtained through customer behavior models; Customer value prediction is obtained by comparing potential capacity with the rate of change of value.

8. A customer behavior feature analysis system based on speech recognition technology, based on the customer behavior feature analysis method based on speech recognition technology according to any one of claims 1 to 7, characterized in that: It also includes an acquisition module for acquiring customer voice data and building a voice processing model; The first calculation module is used to analyze and extract voice features and emotional data through the voice processing model, determine customer focus points, and establish a voice-emotion association model. The first model construction prediction module is used to dynamically obtain a customer intent prediction model by combining the semantics of the focus point with the interaction parameters, and is used for the first prediction of customer needs. The second model construction prediction module is used to combine the first prediction, analyze the dynamic characteristics of the semantics and emotions, obtain the semantic-emotion dual-driven behavior pattern, and perform the second prediction of customer needs. The third model construction prediction module is used to obtain a customer behavior model based on a two-dimensional evaluation matrix and the coupling characteristics of semantics and sentiment. The fourth model construction prediction module is used to predict customer value by combining the customer behavior model with historical interactions to obtain a time-varying customer value prediction model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the customer behavior feature analysis method based on speech recognition technology as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the customer behavior feature analysis method based on speech recognition technology as described in any one of claims 1 to 7.

Images