Tobacco consumption behavior conversion analysis method and device, electronic device, and storage medium

Abstract

The application relates to a tobacco consumption behavior conversion analysis method and device, an electronic device and a storage medium. The tobacco consumption behavior conversion analysis method comprises the following steps: dividing a consumer behavior state into different levels, and setting classification standards of different levels; defining different levels as a hidden state set, defining a single purchase quantity of a consumer as an observation set, converting a transaction record into an observation sequence according to the hidden state set and the observation set, iteratively estimating the observation sequence through a hidden Markov model, outputting a state transition probability matrix, calculating a forward probability and a self-loop probability, identifying a state transition lower than a preset threshold in the forward probability as a key conversion bottleneck, and determining an optimal intervention window; for the key conversion bottleneck, a logistic regression model is constructed to quantify the influence of influencing factors on the conversion success probability, and an attribution analysis result is output. Dynamic modeling is performed through the hidden Markov model, and the prediction accuracy of the consumer behavior conversion probability is improved.

Description

Technical Field

[0001] This application relates to the field of tobacco industry consumer behavior analysis technology, and in particular to methods, devices, electronic devices and storage media for analyzing tobacco consumption behavior conversion. Background Technology

[0002] Tobacco consumption behavior refers to a series of consumer behaviors during the tobacco product purchase process, including specific actions such as first-time trial, repeat purchase, and bulk purchase, as well as the evolution between different consumption stages. Tobacco consumption behavior conversion refers to the gradual evolution of consumers from their initial contact with a tobacco product, through repeated and frequent purchases after initial trial. However, traditional indicators, including repurchase rate and retention rate, can only provide an overall result of how many people have left, and cannot provide targeted solutions for consumers leaving at specific stages of tobacco consumption behavior conversion.

[0003] Currently, classic consumer behavior funnels (such as the attention-interest-desire-action model) can hierarchically divide and visualize the conversion process from awareness to purchase. However, consumer behavior funnels assume that each consumer is only in one level at a time, and that the transition between levels is deterministic. This ignores the probabilistic characteristics of consumer behavior and the cycles and jumps between states, failing to truly reflect the dynamic complexity of tobacco consumption behavior conversion in real-world situations, and resulting in low accuracy in predicting the probability of consumer behavior conversion.

[0004] There is currently no effective solution to the problem of low accuracy in predicting the probability of consumer behavior conversion in related technologies. Summary of the Invention

[0005] This embodiment provides a method, apparatus, electronic device, and storage medium for analyzing tobacco consumption behavior conversion, in order to solve the problem of low accuracy in predicting the probability of consumer behavior conversion in related technologies.

[0006] Firstly, this embodiment provides a method for analyzing the transformation of tobacco consumption behavior, including:

[0007] The behavior of tobacco consumers is divided into different levels, and classification standards are set for each level.

[0008] The different levels are defined as a set of hidden states, and the quantity of a consumer's single purchase is defined as an observation set. The consumer's transaction records are converted into an observation sequence based on the set of hidden states and the observation set. The observation sequence is iteratively estimated using a hidden Markov model, and the state transition probability matrix is ​​output.

[0009] Based on the state transition probability matrix, the forward probability and the self-loop probability are calculated. State transitions with forward probabilities below a preset probability threshold are identified as key transformation bottlenecks, and the optimal intervention window is determined. The forward probability is the probability that the consumer will transition from the current state to the next adjacent state. The self-loop probability is the probability that the consumer will remain in the current state and will not transition to the next adjacent state. The key transformation bottleneck includes the predecessor state and the successor state.

[0010] To address the key conversion bottleneck, a logistic regression model is constructed to quantify the impact of tobacco consumption influencing factors on the conversion success probability, and the attribution analysis results of the tobacco consumption influencing factors are output; the conversion success probability refers to the probability that a consumer will successfully advance from the current level to the target level within a preset observation period.

[0011] In some embodiments, after classifying tobacco consumer behavior states into different levels and setting classification criteria for each level, and before defining the different levels as a set of hidden states, defining the quantity of a consumer's single purchase as an observation set, converting each consumer's transaction records into an observation sequence, iteratively estimating the observation sequence using a hidden Markov model, and outputting a state transition probability matrix, the method further includes:

[0012] Obtain consumer transaction data and feedback data, preprocess the transaction data and feedback data to obtain preprocessed data;

[0013] Based on the preprocessed data, the number of qualified consumers at different levels and the conversion rate between different levels are statistically analyzed; the conversion rate refers to the proportion of consumers who move from the current level to the next adjacent level to the total number of consumers at the current level.

[0014] In some embodiments, the preprocessing of the transaction data and feedback data to obtain preprocessed data includes:

[0015] Fill in the missing purchase records;

[0016] Remove records where the quantity of a single purchase exceeds a preset threshold, and also remove records with a value of zero.

[0017] Using the consumer's first purchase date as the base date, the time of the transaction is calculated based on the relative number of days from the base date to obtain the consumer's transaction time series;

[0018] The transaction time sequence of the consumer is traversed by a sliding window of a preset length to determine whether the purchase frequency of the consumer in any consecutive time window meets the preset level judgment criteria.

[0019] The consumer's level is labeled according to the judgment result, and the preprocessed data is obtained.

[0020] In some embodiments, the different levels are defined as a set of hidden states, the quantity of a consumer's single purchase is defined as an observation set, each consumer's transaction record is converted into an observation sequence, and the observation sequence is iteratively estimated using a hidden Markov model to output a state transition probability matrix, including:

[0021] The different levels are defined as the set of hidden states of the hidden Markov model;

[0022] Discretize the quantity of a consumer's single purchase into multiple observation levels, and define them as an observation set;

[0023] Each consumer's transaction records are converted into an observation sequence in chronological order of purchase time; each observation value in the observation sequence is determined based on the observation level in the observation set that matches the purchase quantity of that transaction record.

[0024] The observed sequence of the consumer is iteratively estimated using the Hidden Markov Model until a preset convergence condition is met, and the estimated state transition probability matrix is ​​output.

[0025] In some embodiments, the step of calculating the forward probability and the self-loop probability based on the state transition probability matrix, identifying state transitions with forward probabilities below a preset probability threshold as key transformation bottlenecks, and determining the optimal intervention window includes:

[0026] Calculate the forward probability and self-loop probability of each state from the state transition probability matrix;

[0027] State transitions with probabilities below a preset probability threshold are identified as key transformation bottlenecks.

[0028] The average number of dwell steps is calculated based on the self-loop probability of the key conversion bottleneck.

[0029] Obtain the average behavior interval time of consumers, and multiply the average number of steps on the stand by the average behavior interval time to obtain the average stand duration;

[0030] The optimal intervention window is determined based on the average dwell time.

[0031] In some embodiments, the construction of a logistic regression model quantifies the impact of tobacco consumption influencing factors on the probability of conversion success, and outputs the attribution analysis results of the tobacco consumption influencing factors, including:

[0032] Consumers who are in the preceding state of the key conversion bottleneck will be used as the analysis sample.

[0033] The logistic regression model is constructed by taking whether the analyzed sample advances to the subsequent state of the key transformation bottleneck within the preset observation period as the dependent variable and the factors affecting tobacco consumption as the independent variables.

[0034] Based on the logistic regression model, estimate the regression coefficients of the independent variables on the dependent variable;

[0035] Calculate the odds ratio of the independent variables based on the regression coefficients, and output the attribution analysis results including the odds ratios.

[0036] In some embodiments, determining the tobacco marketing intervention strategy based on the attribution analysis results and the optimal intervention window includes:

[0037] The intervention targets are determined based on the key conversion bottlenecks; the intervention targets are consumers who are in the leading state of the key conversion bottlenecks.

[0038] Determine the timing of intervention based on the optimal intervention window;

[0039] The intervention direction is determined based on the tobacco consumption influencing factors whose odds ratio is higher than the preset odds ratio threshold in the attribution analysis results.

[0040] Based on the intervention target, the intervention timing, and the intervention direction, a tobacco marketing intervention strategy is generated.

[0041] Secondly, this embodiment provides a tobacco consumption behavior transformation analysis device, including: a partitioning iteration module, an analysis and calculation module, and a result output module; wherein:

[0042] The partitioning iteration module is used to divide the tobacco consumer behavior state into different levels and set classification criteria for different levels; the different levels are defined as a set of hidden states, and the quantity of a consumer's single purchase is defined as an observation set. Based on the set of hidden states and the observation set, the consumer's transaction records are converted into an observation sequence. The observation sequence is iteratively estimated using a hidden Markov model, and a state transition probability matrix is ​​output.

[0043] The analysis and calculation module is used to calculate the forward probability and the self-loop probability based on the state transition probability matrix, identify state transitions with forward probabilities below a preset probability threshold as key transformation bottlenecks, and determine the optimal intervention window; the forward probability is the probability that the consumer will transition from the current state to the next adjacent state; the self-loop probability is the probability that the consumer will remain in the current state and will not transition to the next adjacent state; the key transformation bottleneck includes the predecessor state and the successor state.

[0044] The result output module is used to construct a logistic regression model to quantify the impact of tobacco consumption influencing factors on the conversion success probability for the key conversion bottleneck, and output the attribution analysis results of the tobacco consumption influencing factors; the conversion success probability refers to the probability that a consumer will successfully advance from the current level to the target level within a preset observation period.

[0045] Thirdly, this embodiment provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the tobacco consumption behavior conversion analysis method described in the first aspect above.

[0046] Fourthly, this embodiment provides a storage medium storing a computer program that, when executed by a processor, implements the steps of the tobacco consumption behavior conversion analysis method described in the first aspect above.

[0047] Compared with related technologies, the tobacco consumption behavior conversion analysis method, apparatus, electronic device and storage medium provided in this embodiment are different. This method first divides tobacco consumer behavior into different levels and sets classification criteria for each level. Second, it defines different levels as a set of latent states and the quantity of a consumer's single purchase as an observation set. Based on the latent state set and the observation set, the consumer's transaction records are converted into an observation sequence. The observation sequence is iteratively estimated using a Hidden Markov Model, outputting a state transition probability matrix. Next, based on the state transition probability matrix, the forward probability and self-loop probability are calculated. State transitions with forward probabilities below a preset probability threshold are identified as key conversion bottlenecks, and the optimal intervention window is determined. The forward probability is the probability that a consumer will transition from the current state to the next adjacent state; the self-loop probability is the probability that a consumer remains in the current state without transitioning to the next adjacent state; key conversion bottlenecks include predecessor states and successor states. Finally, for key conversion bottlenecks, a logistic regression model is constructed to quantify the impact of tobacco consumption influencing factors on the conversion success probability, outputting the attribution analysis results of tobacco consumption influencing factors. The conversion success probability refers to the probability that a consumer will successfully advance from the current level to the target level within a preset observation period. By constructing a tobacco consumption behavior funnel and introducing a hidden Markov model for dynamic probability modeling, we can directly estimate the transition probability of consumers between different levels, thereby improving the accuracy of predicting the probability of consumer behavior conversion.

[0048] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0049] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0050] Figure 1 This is a hardware structure block diagram of a terminal for a tobacco consumption behavior conversion analysis method according to an embodiment of this application;

[0051] Figure 2 This is a flowchart of a method for analyzing the conversion of tobacco consumption behavior according to an embodiment of this application;

[0052] Figure 3 This is a flowchart of a hidden Markov model parameter estimation according to an embodiment of this application;

[0053] Figure 4 This is a flowchart of a method for analyzing the conversion of tobacco consumption behavior according to another embodiment of this application;

[0054] Figure 5 This is a structural block diagram of a tobacco consumption behavior conversion analysis device according to an embodiment of this application. Detailed Implementation

[0055] To better understand the purpose, technical solution, and advantages of this application, the application is described and illustrated below in conjunction with the accompanying drawings and embodiments.

[0056] Unless otherwise defined, the technical or scientific terms used in this application shall have the general meaning understood by one of ordinary skill in the art to which this application pertains. Words such as “a,” “an,” “an,” “the,” “the,” and “these” used in this application do not indicate quantitative limitation and may be singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps or modules (units) is not limited to the listed steps or modules (units) but may include steps or modules (units) not listed, or may include other steps or modules (units) inherent to these processes, methods, products, or devices. Words such as “connected,” “linked,” and “coupled” used in this application are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. Normally, the character " / " indicates that the objects before and after it are in an "or" relationship. The terms "first," "second," "third," etc., used in this application are merely to distinguish similar objects and do not represent a specific order of objects.

[0057] The method embodiments provided in this example can be executed on a terminal, computer, or similar electronic device with a certain computing power. For example, it can run on a terminal. Figure 1 This is a hardware structure block diagram of a terminal for a tobacco consumption behavior conversion analysis method according to an embodiment of this application. Figure 1 As shown, a terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 and a memory 104 for storing data are also included. The processor 102 may be, but is not limited to, a microprocessor (MCU) or a programmable logic device (FPGA). The terminal may also include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that… Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the terminal described above. For example, the terminal may also include components that are larger than... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown are illustrated.

[0058] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the tobacco consumption behavior conversion analysis method in this embodiment. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, thereby implementing the above-described method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0059] The transmission device 106 is used to receive or send data via a network. This network includes a wireless network provided by the terminal's communication provider. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 can be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0060] This embodiment provides a method for analyzing the transformation of tobacco consumption behavior. Figure 2 This is a flowchart of a tobacco consumption behavior conversion analysis method according to an embodiment of this application, such as... Figure 2 As shown, the process includes the following steps:

[0061] Step S210: Divide the tobacco consumer behavior status into different levels and set classification criteria for different levels.

[0062] Specifically, customer value is assessed using three core indicators: Recency, Frequency, and Monetary Amount, which forms the basis of the RFM model. The RFM model is an analytical tool for measuring customer value and profitability. In this embodiment, drawing on the core principles of the RFM model, consumer behavior characteristics are characterized by capturing the rhythm and value contribution of consumer behavior. Furthermore, considering the progressive behavioral evolution of tobacco consumers from initial contact, habit formation, to bulk consumption, a hierarchical system of behavioral states applicable to tobacco consumption scenarios is constructed. For example, tobacco consumer behavioral states are divided into seven progressive levels from bottom to top, based on the principle of increasing consumption depth and stickiness. Each level is determined using quantifiable operational definitions, as shown in Table 1.

[0063] Table 1

[0064]

[0065] The first tier is trial consumption, operationalized as a consumer's first purchase of the tobacco product (at least one pack). This signifies successful customer acquisition and initial contact between the consumer and the brand. The second tier is repeat purchase consumption, operationalized as two or more purchases. This indicates initial product acceptance, suggesting a willingness to try the product a second time after the initial experience. The third tier is repurchase consumption, operationalized as three or more purchases. This indicates the beginning of a consumption habit, reflecting a consumer's growing preference for the product and the initial formation of repeat purchase behavior. The fourth tier is regular consumption, operationalized as at least one purchase within any consecutive 30 days. This signifies integration into daily consumption, indicating the tobacco product has become part of the consumer's daily choices. The fifth tier is core consumption, operationalized as at least one purchase per month with an average monthly purchase volume greater than or equal to a preset threshold (e.g., five packs). This signifies becoming a primary choice with high stickiness, representing a high frequency and volume of consumption and significantly increased loyalty. The sixth tier is loyalty consumption, which is operationally defined as having a purchase record every month after the product's launch for one year. Its business meaning is long-term retention, reflecting stable purchasing behavior over a considerable period and high customer lifetime value. The seventh tier is viral consumption, operationally defined as an average of two or more purchases per month, with bulk purchases accounting for 70% or more. Its business meaning is social sharing and bulk consumption, representing consumers not only making high-frequency purchases themselves but also encouraging others to participate through bulk purchases, creating a diffusion effect in consumption.

[0066] The above seven levels are arranged in ascending order. The higher the level, the closer the bond between the consumer and the tobacco product, the stronger the behavioral stickiness, and the higher the consumer's consumption value, thus forming a complete consumer behavior evolution path.

[0067] Step S220: Define different levels as hidden state sets, define the quantity of a consumer's single purchase as an observation set, convert the consumer's transaction records into observation sequences based on the hidden state sets and observation sets, iteratively estimate the observation sequences using a hidden Markov model, and output the state transition probability matrix.

[0068] Specifically, the aforementioned seven funnel levels are used as hidden states in the Hidden Markov Model (HMM), corresponding to novelty consumption, repeat purchase consumption, repurchase consumption, regular consumption, core consumption, loyalty consumption, and viral consumption, respectively, to represent the real consumption stages that consumers cannot directly observe. Simultaneously, the quantity of a consumer's single purchase is divided into five observation levels: single package, small batch, medium batch, large batch, and large batch, serving as observable representations of purchasing behavior. Based on this, each consumer's transaction records are converted into an observation sequence in chronological order of purchase. Each observation value in the sequence is determined by matching the corresponding observation level based on the purchase quantity. After the conversion, the parameters of the observation sequence are estimated using a Hidden Markov Model (HMM), outputting a state transition probability matrix. As a probabilistic model of a doubly stochastic process, the HMM can infer the real consumption levels that consumers cannot directly observe from the observable purchase quantity sequence and estimate the transition probabilities between each level, thereby revealing the behavioral evolution patterns of consumers at different consumption stages.

[0069] For example, the parameters of the Hidden Markov Model are defined as λ = (A, B, π), where A is the state transition probability matrix, used to quantify the probability distribution of consumers moving between different consumption levels. This matrix is ​​an N×N square matrix (N is the total number of levels), and each element a... ij The state transition probability matrix represents the probability that a consumer will move from level i to level j, where i and j represent the current level and the next level, respectively. The state transition probability matrix enables a quantitative analysis of the consumer's behavioral evolution path from a macroscopic perspective. For example, matrix A is a 7×7 matrix with a row vector sum of 1, indicating that starting from a given level, a consumer may move to any level, including themselves. B is the observation probability matrix, used to quantify the probability distribution of various observable purchasing behaviors exhibited by a consumer at a specific consumption level. This matrix is ​​an N×M matrix (N is the total number of levels, M is the total number of observation levels), where each element b... j(k) This represents the probability that a consumer's single purchase quantity falls within the k-th observation level when the consumer is at level j. For example, matrix B is a 7×5 matrix, with rows corresponding to 7 consumption levels and columns corresponding to 5 observation levels, and the sum of the row vectors is 1. For instance, a consumer at the core consumption level might have a significantly higher probability of medium and large purchases than single purchases, indicating that core consumers tend to buy in bulk; while a consumer at the trial consumption level might have the highest probability of single purchases, indicating that trial consumers typically buy single purchases first. The observation probability matrix establishes a probability mapping between consumption level and purchase quantity, thus allowing us to deduce the most likely consumption level for a consumer based on the observed purchase quantity sequence. π represents the initial state distribution; since all consumers start from their first purchase, ... The rest are 0. The specific values ​​of the above parameters need to be estimated through unsupervised learning.

[0070] Subsequently, the Baum-Welch algorithm within the expectation-maximization framework can be used for parameter estimation. The Baum-Welch algorithm is a standard method for Hidden Markov Model (HMM) parameter estimation. Its core idea is to find the model parameters that best explain the observed data through iterative optimization, given only the observed data and the unknown state sequence. This algorithm continuously optimizes the model parameters to fit the observed data by repeatedly executing the expectation step (E-step) and the maximization step (M-step) until convergence.

[0071] The E-step estimates the posterior probability distribution of the hidden states based on observed data, given the current model parameters. Specifically, for each consumer's observed sequence O, the E-step, given the current model parameters λ... (old) Given the observed sequence O, two key posterior probabilities are calculated using the forward-backward algorithm:

[0072] ;

[0073] in, Given the entire observation sequence and the current model parameters, this represents the probability that the state is in state i at step t and will transition to state j at step t+1. It is the joint posterior probability of states at adjacent time steps, which is used to estimate the expected number of transitions between states.

[0074] ;

[0075] in, Given the entire observation sequence and the current model parameters, this represents the probability of being in state i at step t, i.e., the posterior probability of a single-point state, which is used to estimate the expected number of stays in each state.

[0076] The M-step, based on the expected statistic calculated in the E-step, updates the model parameters by maximizing the expected value of the likelihood function, enabling the model to better fit the observed data. The update formula for the state transition probability is:

[0077] ;

[0078] in, This represents the updated probability of transitioning from state i to state j. Let i be the expected number of transitions from state i to state j. Let be the expected total number of occurrences of state i. The ratio of these two numbers is the new transition probability estimate. The update formula for the observation probability is:

[0079] ;

[0080] in, V represents the updated observation value in state j. k The probability, To observe V in state j k Expected number of times Let be the expected total number of times state j occurs, and the ratio of the two is the new observation probability estimate; For indicator functions, when O t =V k The value is 1 if it is true, and 0 otherwise.

[0081] ;

[0082] in, This represents the probability of being in state i during the first transaction after the update. The probability of being in state i in step 1 is used as the new initial state distribution estimate. The above E-step and M-step are executed alternately, and the model's interpretability of the observed data continuously improves after each iteration. The iteration terminates when the log-likelihood increment between two adjacent iterations is less than a preset convergence threshold. The model parameters at this point are the optimal parameter estimates that best interpret the observed data.

[0083] Since different consumers have different transaction counts, i.e., different observation sequence lengths T, a multi-sequence Baum-Welch iteration is performed on all consumers' observation sequences. In each iteration, the expected statistics for all sequences are calculated separately and then summed to update the model parameters uniformly. The iteration continues until the log-likelihood increment is less than a preset convergence threshold, at which point the model is considered converged, and the state transition probability matrix is ​​output.

[0084] Figure 3 This is a flowchart illustrating the parameter estimation process for a hidden Markov model according to one embodiment of this application. Figure 3 As shown, the model parameters λ are first initialized. (0) =(A, B, π), where the number of states N=7 and the observation level M=5; then input the observation sequences of all consumers; in the E-step, use the forward-backward algorithm to calculate the posterior probability. and In step M, update the state transition probability, observation probability, and initial state distribution based on the expected statistic; repeat steps E and M until the log-likelihood increment is less than 1e. -4 The iteration terminates when the time is reached, and the estimated model parameters are output. ,in This is the estimated state transition probability matrix.

[0085] Step S230: Calculate the forward probability and self-loop probability based on the state transition probability matrix, identify state transitions with forward probabilities below a preset probability threshold as key transformation bottlenecks, and determine the optimal intervention window; the forward probability is the probability that the consumer will transition from the current state to the next adjacent state; the self-loop probability is the probability that the consumer will remain in the current state and not transition to the next adjacent state; key transformation bottlenecks include predecessor states and successor states.

[0086] Based on the state transition probability matrix obtained in step S220, the forward probability a of each state is extracted. {i,i+1} And the probability of self-loop a {i,i} Wherein, the probability of moving forward is a {i,i+1} This represents the probability that a consumer will advance from the current state i to the next adjacent state i+1, reflecting the conversion efficiency between different levels, i.e., the likelihood that the consumer will successfully jump to a higher level from the current level. The advance probability is an estimated probability of consumer behavior conversion obtained through Hidden Markov Model (HMM) estimation, which can directly quantify the possibility of a consumer transitioning from a certain level to the next. Self-loop probability a {i,i} This represents the probability that a consumer remains in the current state i without undergoing a hierarchical transition, reflecting the stability of each state and the consumer's tendency to remain in that state.

[0087] All extracted forward probabilities are compared, and state transitions with forward probabilities below a preset threshold are identified as key conversion bottlenecks. A key conversion bottleneck refers to the transition between adjacent states with the lowest forward transition probability in the consumer behavior evolution path. This transition represents the most significant resistance and the most likely point of churn or stagnation when consumers move from the current level to the next, and is the core node for implementing marketing interventions to optimize conversion efficiency. This key conversion bottleneck state transition includes a preceding state and a succeeding state. The preceding state is the level the consumer was at before the bottleneck occurred, and the succeeding state is the target level the consumer attempted to advance to but failed to reach effectively.

[0088] After identifying key conversion bottlenecks, the average number of steps on the device is calculated based on the self-loop probability of the preceding states of the bottleneck. This is then combined with the average behavioral interval of consumers to calculate the average dwell time, thereby determining the optimal intervention window. The average number of steps on the device refers to the average number of purchases a consumer makes while in a given state. A higher self-loop probability indicates more average dwell times for consumers, and consequently, more attempts are required to leave that state. Based on this, the average behavioral interval of consumers in that state is obtained—the average number of days between two consecutive purchases. Multiplying the average number of steps on the device by the average behavioral interval yields the average dwell time in that state, quantifying the expected dwell period of consumers in that state. This average dwell time characterizes the expected dwell period of consumers in the preceding states. In practical applications, this average dwell time can also be used to determine the optimal intervention window to optimize the timing of intervention.

[0089] Step S240: Targeting key conversion bottlenecks, construct a logistic regression model to quantify the impact of tobacco consumption influencing factors on the probability of successful conversion, and output the attribution analysis results of tobacco consumption influencing factors; the probability of successful conversion refers to the probability that a consumer will successfully advance from the current level to the target level within a preset observation period.

[0090] For the identified key conversion bottlenecks, a logistic regression model was constructed for attribution analysis. Using consumers in the pre-bottom stage of the bottleneck as the analysis sample, whether they progressed to the subsequent stage within a pre-defined observation period as the dependent variable, and factors from three dimensions—product experience, purchasing behavior, and marketing environment—as independent variables, a logistic regression model was constructed. Specifically, the logistic regression model was built based on the dependent and independent variables. The logistic regression model can estimate the direction and degree of the marginal impact of each independent variable on the probability of conversion success; that is, under the condition that other factors remain constant, a change in a certain factor will cause an increase or decrease in the conversion probability. The logistic regression model was trained using historical data before implementation (i.e., observational data without any strategic intervention) to ensure that the estimated influence relationship reflects the objective correlation between each factor and the conversion probability, rather than the confounding effects of strategic intervention. The form of the logistic regression model is as follows:

[0091] ;

[0092] Where p is the probability of successful conversion. to Let n be the independent variables. For the intercept term, to These are the regression coefficients for each independent variable. The logistic regression model uses maximum likelihood estimation to estimate the parameters and outputs the regression coefficients and their statistical significance levels for each variable.

[0093] Finally, based on the estimation results of the logistic regression model, the regression coefficients of each variable were converted into odds ratios (OR values). The odds ratio visually represents the quantitative impact of each factor on the probability of successful conversion: an OR value greater than 1 indicates that the factor has a positive promoting effect on conversion, with a larger OR value indicating a stronger promoting effect; an OR value less than 1 indicates that the factor has a negative hindering effect; and an OR value equal to 1 indicates that the factor has no significant effect. The output includes the attribution analysis results containing the odds ratios of each variable, clarifying the actual contribution of each tobacco consumption influencing factor to bottleneck breakthrough, and providing a quantitative basis for subsequently developing precise marketing intervention strategies.

[0094] Compared to related technologies, while classic consumer behavior funnels can hierarchically divide and visualize the conversion process from consumer awareness to purchase, these models are essentially static and deterministic analytical frameworks. Their implicit assumption is that each consumer can only be in a specific level of the funnel at any given time, and the evolution between levels follows a fixed unidirectional sequence, without regression, jumps, or cycles. However, real-world tobacco consumption behavior conversion is a complex dynamic process. Consumer behavior is not a simple linear progression; consumers may be on the edge of multiple levels simultaneously, or they may repeatedly linger in the previous stage before making a leap, or even regress from a higher level to a lower level. Such cycles and jumps between states cannot be characterized in static funnel models, thus resulting in low accuracy in predicting the probability of consumer behavior conversion.

[0095] Steps S210 to S240 above involve: First, classifying tobacco consumer behavior states into different levels and setting classification criteria for each level; second, defining different levels as hidden state sets and defining the quantity of a consumer's single purchase as an observation set; converting consumer transaction records into observation sequences based on the hidden state sets and observation sets; iteratively estimating the observation sequences using a hidden Markov model to output a state transition probability matrix; next, calculating the forward probability and self-loop probability based on the state transition probability matrix; identifying state transitions with forward probabilities below a preset probability threshold as key conversion bottlenecks and determining the optimal intervention window; the forward probability is the probability that a consumer will transition from the current state to the next adjacent state; the self-loop probability is the probability that a consumer remains in the current state without transitioning to the next adjacent state; key conversion bottlenecks include predecessor states and successor states; finally, constructing a logistic regression model to quantify the impact of tobacco consumption influencing factors on the conversion success probability for key conversion bottlenecks, and outputting the attribution analysis results of tobacco consumption influencing factors; the conversion success probability refers to the probability that a consumer will successfully advance from the current level to the target level within a preset observation period. By constructing a funnel of tobacco consumption behavior and introducing a hidden Markov model for dynamic probability modeling, the forward probability is calculated based on the state transition probability matrix, thereby realizing the direct estimation of the transition probability of consumers between each level and improving the prediction accuracy of consumer behavior conversion probability.

[0096] Optionally, in one embodiment, after classifying tobacco consumer behavior states into different levels and setting classification criteria for each level, before defining different levels as hidden state sets, defining the quantity of a consumer's single purchase as an observation set, converting each consumer's transaction records into observation sequences, iteratively estimating the observation sequences using a hidden Markov model, and outputting the state transition probability matrix, the method further includes:

[0097] Acquire consumer transaction data and feedback data, preprocess the transaction data and feedback data to obtain preprocessed data; based on the preprocessed data, calculate the number of qualified consumers at different levels and the conversion rate between different levels; the conversion rate refers to the proportion of consumers who move from the current level to the next adjacent level to the total number of consumers at the current level.

[0098] First, two types of raw consumer data are acquired. The first type is transaction data, sourced from the sales terminal system. This data includes anonymized consumer unique identifiers, purchase time, quantity purchased (per pack), and payment amount, used to characterize consumer behavior such as purchase frequency, quantity, and pace. The second type is feedback data, sourced from the terminal's anonymous QR code review system. This data collects consumers' voluntary evaluations of product odor, packaging, and cost-effectiveness, used for constructing variables for the product experience dimension in subsequent attribution analysis. Next, the raw data undergoes preprocessing, including: imputing missing values, removing outliers, aligning time windows based on the first purchase date, and using sliding window detection to determine purchase frequency.

[0099] Then, based on the preprocessed data, each consumer's transaction behavior is labeled according to the aforementioned hierarchical classification criteria to determine their current level and level transition status within the observation period. Based on this, the number of consumers meeting the criteria at each level and the conversion rate between different levels are statistically analyzed. The conversion rate refers to the proportion of consumers who move from the current level to the next adjacent level out of the total number of consumers at the current level, used to measure the conversion efficiency between adjacent levels.

[0100] For example, taking 50,000 consumers who tried a new tobacco product within 12 months of its launch as an example, after preprocessing the raw data, a total of 50,000 valid consumer samples and 820,654 valid transaction records were obtained. Based on the preprocessed data, each consumer was labeled with a status and assigned to a level according to the operational definitions of the above 7 levels. The number of qualified consumers at each level and the conversion rate between adjacent levels were statistically analyzed. The results are shown in Table 2:

[0101] Table 2

[0102]

[0103] As shown in Table 2, there are 50,000 new consumers and 22,500 repeat consumers, so the conversion rate from new consumer to repeat consumer is 45.0%. There are 12,375 repeat consumers, so the conversion rate from repeat consumer to repeat consumer is 55.0%. And so on, we can get the conversion rate of each level (e.g., 48.0% from repeat consumer to regular consumer, 40.0% from regular consumer to core consumer, 20.0% from core consumer to loyalty consumer, and 40.0% from loyalty consumer to referral consumer), as well as the cumulative conversion rate from the initial level to the target level (e.g., 11.88% from new consumer to regular consumer). The conversion rate statistics above can be used to initially identify nodes with high churn rates (e.g., the largest number of churned customers is at the trial-and-go consumer level (27,500) and the highest relative churn rate is at the core consumer level (80%). They also provide a benchmark for the forward probability estimated by the HMM model and can verify the accuracy of the HMM estimation results by comparing the forward probability estimated by the HMM model with the actual conversion rates. Furthermore, based on these statistics, it can be preliminarily determined that although the absolute number of churned customers at the fixed consumer level to the core consumer level is large (3,564), these consumers have formed stable purchasing habits and have high strategic intervention value, providing an initial basis for subsequent precise bottleneck identification and intervention strategy formulation.

[0104] Based on the above preprocessing and statistical results, we can initially identify the differences in conversion efficiency between different levels and the nodes with more serious loss, which provides an initial judgment basis and data foundation for subsequent dynamic probability modeling and accurate bottleneck location.

[0105] In addition, in one embodiment, the transaction data and feedback data are preprocessed to obtain preprocessed data, including: filling in missing purchase records; removing records where the single purchase quantity exceeds a preset quantity threshold and removing records with a zero amount; using the consumer's first purchase date as the base date, calculating the time of the transaction behavior based on the relative number of days of the base date to obtain the consumer's transaction time series; using a sliding window of preset length to traverse the consumer's transaction time series and determine whether the consumer's purchase frequency within any consecutive time window meets a preset level judgment standard; and labeling the consumer's corresponding level according to the judgment result to obtain the preprocessed data.

[0106] First, missing values ​​are handled in the raw data. For missing purchase quantities in transaction records, the median purchase quantity from previous and subsequent transaction records for the same consumer is used for imputation; if the consumer only has one transaction record and the median purchase quantity cannot be calculated, the average purchase quantity of all consumers at the same sales terminal is used for imputation. For missing evaluation information in feedback data, the mode is used for imputation.

[0107] Secondly, outlier removal was performed on the raw data. Records exceeding a preset quantity threshold (e.g., the 99th percentile) were removed, as these records mostly correspond to wholesale or test data and do not fall under the category of normal consumer behavior. Invalid records with zero payment amounts were also removed. After these removal processes, the total removed data accounted for approximately 0.08%, effectively ensuring the quality and reliability of the data used in subsequent analysis.

[0108] Next, the transaction data is aligned to a time window. Using each consumer's first purchase date as the base date (i.e., day 0), the time of all subsequent transactions is recalculated relative to this base date. For example, "purchase on a certain day of a certain month of a certain year" is transformed into "purchase on the Nth day after the first purchase." This preprocessing step eliminates the absolute calendar differences caused by different consumers' first purchase dates, making the behavioral sequences of all consumers comparable in the time dimension and providing a unified time coordinate for subsequent sliding window detection and sequence modeling.

[0109] Finally, a sliding window detection is performed on the transaction data. For tiers that require judgment based on continuous time windows (such as fixed consumption tiers requiring at least one purchase within any consecutive 30 days), a sliding window of a preset length (e.g., 30 days) is used to traverse the transaction time series of each consumer. The sliding step is one transaction day, and the window slides forward one day at a time, recording the purchase frequency within each window. If a consumer reaches the preset purchase frequency (e.g., at least one purchase) within any sliding window, the consumer is determined to meet the state condition of that tier, and the corresponding consumption tier is marked. This achieves the goal of marking the consumer's corresponding tier based on the judgment result, resulting in preprocessed data.

[0110] Through the above preprocessing steps, the raw transaction data and feedback data are converted into a standard format that can be used for subsequent hierarchical statistics and HMM modeling.

[0111] In one embodiment, different levels are defined as a set of hidden states, the quantity of a consumer's single purchase is defined as an observation set, each consumer's transaction record is converted into an observation sequence, and the observation sequence is iteratively estimated using a hidden Markov model to output a state transition probability matrix, including:

[0112] Different levels are defined as the hidden state set of the Hidden Markov Model; the quantity of a consumer's single purchase is discretized into multiple observation levels, defined as the observation set; each consumer's transaction record is converted into an observation sequence according to the purchase time order; each observation value in the observation sequence is determined according to the observation level in the observation set that matches the purchase quantity of that transaction record; the consumer's observation sequence is iteratively estimated using the Hidden Markov Model until the preset convergence condition is met, and the estimated state transition probability matrix is ​​output.

[0113] Specifically, the hidden state set is a set of seven progressive levels (novelty-seeking consumption level, repeat purchase consumption level, repurchase consumption level, regular consumption level, core consumption level, loyalty consumption level, and viral consumption level) that divide tobacco consumer behavior into, representing the actual stage the consumer is currently in. Specifically, the seven funnel levels defined in the preceding steps can be defined as the hidden state set S = {s1, s2, s3, s4, s5, s6, s7} of the Hidden Markov Model, where s1 corresponds to novelty-seeking consumption, s2 to repeat purchase consumption, s3 to repurchase consumption, s4 to regular consumption, s5 to core consumption, s6 to loyalty consumption, and s7 to viral consumption.

[0114] Meanwhile, the observation set is a set of observable purchasing behavior, discretized from the quantity of a consumer's single purchase into five levels: single pack, small batch, medium batch, large batch, and large batch. Specifically, the quantity of a consumer's single purchase can be discretized into five observation levels, defined as the observation set V = {V1, V2, V3, V4, V5}, where V1 represents a single pack purchase (1 pack), V2 represents a small batch purchase (2 to 5 packs), V3 represents a medium batch purchase (6 to 9 packs), V4 represents a large batch purchase (1 item / 10 packs), and V5 represents a large batch purchase (greater than or equal to 11 packs). Based on the above hidden state set and observation set, each consumer's transaction records are converted into an observation sequence according to the purchase time order. The observation sequence is a sequence composed of observation levels, converted from each consumer's transaction records according to the purchase time order, and serves as the input data for the Hidden Markov Model estimation. Specifically, the observation sequence can be set as O = O1, O2, O3, ..., O t , where t is the total number of transactions for the consumer, and each observation in the sequence is determined by matching the corresponding observation level according to the purchase quantity of that transaction.

[0115] After completing the observation sequence transformation, parameters of the observation sequence can be estimated using a Hidden Markov Model (HMM), outputting a state transition probability matrix. The Baum-Welch algorithm is used for iterative estimation of the observation sequence. The Baum-Welch algorithm continuously optimizes the model parameters to fit the observation data by alternately executing E-steps and M-steps. The E-step estimates the probability of the consumer being at each level under the current parameters, and the M-step updates the model parameters based on this estimate. This process is repeated until the log-likelihood increment between two consecutive iterations is less than a preset threshold, at which point the estimated state transition probability matrix is ​​output.

[0116] By mapping consumer behavior levels and purchase quantities to latent states and observed values ​​respectively, the Baum-Welch algorithm is used to automatically estimate the transition probabilities between states under unsupervised conditions. Without the need for manually setting transition rules or thresholds, it can objectively reveal the behavioral evolution patterns of consumers at different consumption stages, providing quantifiable probabilistic basis for subsequent bottleneck identification and intervention strategy formulation.

[0117] In one embodiment, based on the state transition probability matrix, the forward probability and the self-loop probability are calculated. State transitions with forward probabilities below a preset probability threshold are identified as key transformation bottlenecks, and the optimal intervention window is determined, including:

[0118] Calculate the forward probability and self-loop probability of each state from the state transition probability matrix; identify state transitions with forward probabilities below a preset probability threshold as key conversion bottlenecks; calculate the average number of steps on the standby based on the self-loop probability of the key conversion bottlenecks; obtain the average behavior interval time of consumers, multiply the average number of steps on the standby by the average behavior interval time to obtain the average standby duration; determine the optimal intervention window based on the average standby duration.

[0119] All extracted forward probabilities are compared, and state transitions with forward probabilities below a preset probability threshold are identified as key conversion bottlenecks. These state transitions include a predecessor state and a successor state. The predecessor state is the consumer's current level before the key conversion bottleneck occurs, and the successor state is the target level the consumer attempted to advance to but failed to reach. For example, after 147 iterations, convergence occurs, and the key state transition probabilities are shown in Table 3.

[0120] Table 3

[0121]

[0122] As shown in Table 3, the progression probabilities of the seven states estimated based on the Hidden Markov Model (HMM) are as follows: from trial to repurchase: 44.8%; from repurchase to second purchase: 55.1%; from second purchase to regular purchase: 47.9%; from regular purchase to core purchase: 39.8%; from core purchase to loyalty: 20.1%; and from loyalty to referral: 39.7%. The transition from core to loyalty has the lowest probability (20.1%), but this transition requires consumers to maintain monthly purchases over a 12-month period, making the strategy intervention too long and difficult to implement. While the probability of progressing from regular purchase to core purchase is 39.8%, higher than from core to loyalty, it involves 5,940 consumers in the preceding regular purchase state, and these consumers have already formed stable consumption habits (self-loop probability of 45.0%, average dwell time of approximately 27 days), making it more feasible for short-term intervention and possessing higher commercial value. After comprehensive comparison, regular purchase to core purchase can be identified as the primary key conversion bottleneck, with regular purchase as the preceding state and core purchase as the subsequent state.

[0123] After identifying the key conversion bottleneck, the average number of dwell steps is calculated based on the self-loop probability of the precursor state of the key conversion bottleneck. Specifically, the self-loop probability 'a'... {i,i} This represents the probability that a consumer will remain in the current state without moving to the next state. The higher the self-loop probability, the stronger the consumer's tendency to stay in that state and the weaker their mobility. Based on this, the formula for calculating the average number of steps taken is 1 / (1-a). {i,i} The formula states that the higher the self-loop probability, the more purchases a consumer needs to make while remaining in that state, and the more attempts are required to leave that state. Taking the fixed state to the core key conversion bottleneck as an example, the self-loop probability is 45.0%, and substituting into the formula, the average number of steps to stay is approximately 1.82 purchases.

[0124] Based on this, the average behavioral interval time of consumers in this state is obtained, that is, the average number of days between two consecutive purchases. By multiplying the average number of steps taken on the stand by the average behavioral interval time, the average dwell time of consumers in this state can be obtained, which can be used to quantify the expected dwell period of consumers in this state. For example, if the average behavioral interval time of consumers in a fixed state is about 15 days and their average number of steps taken on the stand is 1.82, the average dwell time can be obtained as about 27 days.

[0125] Finally, the optimal intervention window is determined based on the average dwell time. For example, the average dwell time represents approximately 27 days after a consumer enters the "fixed state." If the consumer fails to advance to the core state within this window, there is a high probability they will revert to the repurchase state or churn. Therefore, the optimal intervention window is determined to be between the 15th and 25th day after the consumer enters the fixed state. Intervening too early, before the consumer has formed a stable, fixed consumption habit, may render marketing outreach ineffective; intervening too late, the consumer may have already reverted, and the intervention opportunity will be missed. Therefore, this intervention window represents the optimal time period for implementing marketing intervention in this initial state.

[0126] In one embodiment, a logistic regression model is constructed to quantify the impact of tobacco consumption influencing factors on the probability of successful conversion, and the attribution analysis results of tobacco consumption influencing factors are output, including:

[0127] Consumers in the preceding state of the key conversion bottleneck are used as the analysis sample; whether the analysis sample advances to the subsequent state of the key conversion bottleneck within the preset observation period is used as the dependent variable, and the factors affecting tobacco consumption are used as independent variables to construct a logistic regression model; based on the logistic regression model, the regression coefficients of the independent variables on the dependent variable are estimated; based on the regression coefficients, the odds ratio of the independent variables is calculated, and the attribution analysis results including the odds ratio are output.

[0128] To address the identified key conversion bottlenecks, a logistic regression model was constructed for attribution analysis to identify the core driving and hindering factors influencing their breakthrough. First, the analysis object and target variable of the logistic regression model were determined. All consumers entering the pre-bottom state of the key conversion bottleneck were used as the analysis sample, and their behavior was tracked and observed within a pre-defined observation period. If a consumer successfully progressed to the subsequent state of the key conversion bottleneck within this period, they were considered to have successfully converted; otherwise, they were considered not to have converted. This binarized result was used as the dependent variable of the model, and its probability form represents the probability of successful conversion.

[0129] Secondly, the set of independent variables for the logistic regression model was determined. The independent variables were extracted from three dimensions to identify factors influencing tobacco consumption: product experience, including consumers' subjective evaluations of the product's flavor, consumption rate, and ease of use, derived from anonymous QR code scanning feedback data; purchase behavior, including quantifiable indicators such as the number of purchases and the standard deviation of purchase volume within a preset number of days, derived from consumers' transaction records; and marketing environment, including external intervention information such as whether consumers have been exposed to subscription plan promotions, the intensity of store staff recommendations, and the intensity of in-store promotional activities, derived from in-store records. These three dimensions of variables together constitute the input feature set of the logistic regression model, comprehensively capturing various factors influencing consumer conversion.

[0130] Then, based on the dependent and independent variables, a logistic regression model is constructed. For example, in one embodiment, the logistic regression model takes the form:

[0131] ;

[0132] Where p is the probability of successful conversion. to These represent eight independent variables: product odor absorption evaluation, consumption rate evaluation, ease of use evaluation, number of purchases in the past 30 days, standard deviation of purchase volume in the past 30 days, subscription plan promotion, strength of store staff recommendation, and strength of in-store promotional activities. For the intercept term, to These are the regression coefficients for each independent variable. The logistic regression model uses maximum likelihood estimation to estimate the parameters and outputs the regression coefficients and their statistical significance levels for each variable.

[0133] Based on the estimation results, the regression coefficients of each variable are converted into odds ratios (OR values) to quantify the promoting or hindering effect of each factor on the probability of conversion success. An OR value greater than 1 indicates that the factor has a positive promoting effect on conversion; the larger the OR value, the stronger the promoting effect. An OR value less than 1 indicates that the factor has a negative hindering effect. An OR value equal to 1 indicates that the factor has no significant effect. The output includes the attribution analysis results containing the odds ratios of each variable, providing a quantitative basis for subsequently developing precise marketing intervention strategies.

[0134] Using pre-implementation data (without any strategic intervention), the model was fitted, and the odds ratio (OR) of significant variables (p < 0.05) was output. Taking the attribution analysis of fixing the core as a key conversion bottleneck as an example, the logistic regression results of each variable are shown in Table 4:

[0135] Table 4

[0136]

[0137] Table 4 shows that the two strongest positive drivers of the "fixed to core" conversion bottleneck are the odds ratio (OR) of subscription plan promotion (2.48) and the OR of product flavor evaluation (2.39). This indicates that for consumers who have formed fixed consumption habits, their primary concerns are whether the product meets their flavor requirements and whether there are more economical ways to purchase it. Therefore, subscription plan promotion and product flavor optimization can be identified as the core intervention directions. Furthermore, based on the key conversion bottleneck, the intervention target is identified as consumers in a fixed consumption state, and the optimal intervention window is determined to be from day 15 to day 25 after entering the fixed state. Finally, a marketing intervention strategy targeting this key conversion bottleneck is generated.

[0138] For example, in one embodiment, an independent logistic regression model is constructed for attribution analysis targeting the key conversion bottleneck from core to fission. Using consumers who have entered the core consumption state as the analysis sample, the dependent variable is whether the proportion of bulk purchases within 3 months of becoming a core consumer reaches 70% or more. Independent variables include the intensity of bulk purchase incentives, willingness to share on social media, exposure to bulk trial packs, and receipt of anonymous group-buying information. The logistic regression model outputs the odds ratio (OR) for significant variables (p < 0.05), as shown in Table 5.

[0139] Table 5

[0140]

[0141] Table 5 shows that the advantage ratios for bulk purchases of double points (4.76) and bulk trial packs (2.53) are the main drivers of the conversion from core to viral consumption. This indicates that for consumers who have already formed core consumption patterns, economic incentives and lowering the decision-making threshold are key means to encourage them to convert to bulk viral consumption. The advantage ratios for social sharing (2.18) and anonymous group-buying information (1.95) also have a certain promoting effect. Therefore, bulk purchases of double points incentives and bulk trial pack distribution can be identified as the core intervention directions to address this key conversion bottleneck.

[0142] Attribution analysis using logistic regression models can quantify the marginal contribution of each factor to overcoming key conversion bottlenecks in the form of advantage ratios, effectively improving the accuracy of resource allocation.

[0143] Furthermore, in one embodiment, a tobacco marketing intervention strategy is determined based on attribution analysis results and the optimal intervention window, including:

[0144] The intervention targets are determined based on key conversion bottlenecks; the intervention targets are consumers in the lead-up to key conversion bottlenecks; the intervention timing is determined based on the optimal intervention window; the intervention direction is determined based on the tobacco consumption influencing factors whose odds ratio is higher than the preset odds ratio threshold in the attribution analysis results; and a tobacco marketing intervention strategy is generated based on the intervention targets, intervention timing, and intervention direction.

[0145] The intervention targets are determined based on the key conversion bottleneck. The intervention targets are the consumer groups that are currently in the leading stage of the key conversion bottleneck, that is, those consumers who have reached the level before the key conversion bottleneck and are at the critical point of leaping to the next level. These consumers are most likely to achieve a conversion breakthrough through external intervention.

[0146] The optimal intervention window is used to determine the timing of intervention. This window corresponds to the time interval within which consumers spend an average of their time in the preceding phase of the critical conversion bottleneck. Intervention within this window maximizes the conversion success rate, avoiding both premature and delayed intervention. Furthermore, the direction of intervention is determined based on the tobacco consumption influencing factors with an advantage ratio higher than a preset threshold, as identified in the attribution analysis. These factors have a significant positive impact on the probability of conversion success, with the highest advantage ratios prioritized as the core intervention direction—those influencing consumers' advancement to the next level.

[0147] Based on the intervention target, timing, and direction, a tobacco marketing intervention strategy is generated. This enables the creation of a complete and executable marketing intervention plan encompassing these three dimensions, ensuring that marketing resources are precisely targeted at key points hindering conversion and maximizing intervention effectiveness with limited resources.

[0148] Figure 4 This is a flowchart of a method for analyzing tobacco consumption behavior conversion according to another embodiment of this application. Figure 4 As shown, this method for analyzing tobacco consumption behavior conversion includes the following steps:

[0149] Step S401: Data Fusion and Bottleneck Identification. Consumer transaction data and feedback data within a preset time period after the target tobacco product's launch are acquired through the data acquisition and preprocessing module. Transaction data includes consumer identification, purchase time, and purchase quantity; feedback data includes product flavor evaluation information. The data is processed using a consumer behavior funnel construction module and a hidden Markov model dynamic analysis module to identify the transition probabilities between different states. The state transition with the lowest probability of progress is identified as a key conversion bottleneck. The average dwell time of the preceding state is calculated based on the self-loop probability to determine the optimal intervention window. In one embodiment, data analysis of 5,000 sales terminals and 50,000 new-tasting consumers within 6 months of a new tobacco product's launch shows that the probability of progressing to the core state is 39.8%, with an average dwell time of 27 days, which is identified as a key bottleneck requiring priority intervention.

[0150] Step S402, Attribution Analysis and Strategy Formulation. Targeting key conversion bottlenecks, using consumers in the leading state of this bottleneck as the analysis sample, and variables in product experience, purchase behavior, and marketing environment as independent variables, and whether they advance to the subsequent state within a preset observation period as the dependent variable, a logistic regression model is constructed for attribution analysis, outputting the advantage ratio of each variable. Based on variables with advantage ratios higher than a preset threshold in the attribution analysis results, the intervention direction is determined, the intervention targets are identified based on the key conversion bottleneck, and the intervention timing is determined based on the optimal intervention window, generating a collaborative intervention strategy package. In one embodiment, the attribution analysis results of the logistic regression model show that subscription plan promotion (OR=2.48) and product flavor evaluation (OR=2.39) are the strongest factors driving the conversion from fixed to core. The resulting collaborative strategy could be to send instructions to the consumer operations system, i.e., to push subscription plan promotional information to all consumers in the fixed state, emphasizing the benefits and convenience of bulk orders; or to push reports to the product development and marketing departments, i.e., suggesting small-scale flavor optimization experience activities in areas where fixed-state consumers are concentrated, collecting feedback and iterating the product.

[0151] Step S403, Effect Tracking and Closed-Loop Optimization. After the collaborative intervention strategy package has been implemented for a preset time period, consumer data is re-collected and analyzed to compare the conversion rate changes before and after the strategy implementation and calculate the conversion rate improvement. In one embodiment, two months after the implementation of the collaborative intervention strategy, among the fixed-state consumers who received the subscription plan promotion, the proportion of those who became core consumers increased from 18.5% before the strategy to 34.2%, an increase of 15.7 percentage points; among the fixed-state consumers who participated in the odor-absorbing experience activity, the proportion of those who became core consumers increased from 20.1% to 36.5%, an increase of 16.4 percentage points; the overall conversion rate from fixed to core consumers increased from approximately 40% before the strategy to approximately 52%, higher than the historical level.

[0152] Step S404, Model Iteration. The feedback data after strategy execution is re-inputted into the Hidden Markov Model and Logistic Regression module to update the state transition probabilities and odds ratio estimates, achieving continuous model optimization. In one embodiment, through the aforementioned accurate identification and attribution-driven intervention loop, the number of core consumers for the new product increased by approximately 28% higher than expected in the 9th month after its launch, effectively supporting the product's market growth.

[0153] Steps S401 to S404 above achieve quantitative estimation of conversion probability and automatic bottleneck identification by integrating behavioral funnel and hidden Markov model, shift marketing strategy formulation to data-driven by attribution analysis through logistic regression model, and realize continuous optimization of strategy through closed-loop iteration mechanism, effectively solving the problems of low prediction accuracy of conversion probability and lack of targeted intervention in existing technologies.

[0154] Figure 5 This is a structural block diagram of a tobacco consumption behavior conversion analysis device 50 according to an embodiment of this application, as shown below. Figure 5 As shown, the tobacco consumption behavior transformation analysis device 50 includes: a partitioning and iteration module 52, an analysis and calculation module 54, and a result output module 56; wherein: the partitioning and iteration module 52 is used to divide the tobacco consumer behavior state into different levels and set the classification criteria for different levels; different levels are defined as hidden state sets, and the consumer's single purchase quantity is defined as the observation set; based on the hidden state set and the observation set, the consumer's transaction records are converted into observation sequences; the observation sequences are iteratively estimated using a hidden Markov model, and the state transition probability matrix is ​​output; the analysis and calculation module 54 is used to calculate the forward probability and self-loop probability based on the state transition probability matrix. The probability module identifies state transitions below a preset probability threshold as key conversion bottlenecks and determines the optimal intervention window. The forward probability is the probability that a consumer will transition from the current state to the next adjacent state. The self-loop probability is the probability that a consumer remains in the current state without transitioning to the next adjacent state. Key conversion bottlenecks include predecessor and successor states. The output module 56 is used to construct a logistic regression model to quantify the impact of tobacco consumption influencing factors on the conversion success probability, and outputs the attribution analysis results of these factors. The conversion success probability refers to the probability that a consumer will successfully advance from the current level to the target level within a preset observation period.

[0155] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination.

[0156] This embodiment also provides an electronic device including a memory and a processor, the memory storing a computer program and the processor being configured to run the computer program to perform the steps in any of the above method embodiments.

[0157] Optionally, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.

[0158] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program:

[0159] S1 divides the behavior of tobacco consumers into different levels and sets classification standards for each level;

[0160] S2 defines different levels as hidden state sets and the quantity of a consumer's single purchase as observation set. Based on the hidden state set and observation set, the consumer's transaction records are converted into observation sequences. The observation sequences are iteratively estimated using a hidden Markov model, and the state transition probability matrix is ​​output.

[0161] S3. Based on the state transition probability matrix, calculate the forward probability and the self-loop probability, identify state transitions with forward probabilities below a preset probability threshold as key transformation bottlenecks, and determine the optimal intervention window; the forward probability is the probability that the consumer will transition from the current state to the next adjacent state; the self-loop probability is the probability that the consumer will remain in the current state and will not transition to the next adjacent state; key transformation bottlenecks include the predecessor state and the successor state.

[0162] S4. To address key conversion bottlenecks, a logistic regression model is constructed to quantify the impact of tobacco consumption influencing factors on the probability of successful conversion, and the attribution analysis results of tobacco consumption influencing factors are output. The probability of successful conversion refers to the probability that a consumer will successfully advance from the current level to the target level within a preset observation period.

[0163] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated in this embodiment.

[0164] Furthermore, in conjunction with the tobacco consumption behavior conversion analysis method provided in the above embodiments, this embodiment can also provide a storage medium for implementation. This storage medium stores a computer program; when executed by a processor, the computer program implements any of the tobacco consumption behavior conversion analysis methods in the above embodiments.

[0165] It should be understood that the specific embodiments described herein are merely illustrative of the application and not intended to limit it. All other embodiments derived by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.

[0166] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0167] Obviously, the accompanying drawings are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar situations based on these drawings without any creative effort. Furthermore, it is understood that although the work done in this development process may be complex and lengthy, for those skilled in the art, certain design, manufacturing, or production modifications made based on the technical content disclosed in this application are merely conventional technical means and should not be considered as insufficient disclosure of this application.

[0168] The term "embodiment" in this application refers to a specific feature, structure, or characteristic described in connection with an embodiment that may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily imply the same embodiment, nor does it imply that it is mutually exclusive with or independent of other embodiments. It will be clearly or implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0169] The above embodiments merely illustrate several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of patent protection. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the appended claims.

Claims

1. A method for analyzing the transformation of tobacco consumption behavior, characterized in that, include: The behavior of tobacco consumers is divided into different levels, and classification standards are set for each level. The different levels are defined as a set of hidden states, and the quantity of a consumer's single purchase is defined as an observation set. The consumer's transaction records are converted into an observation sequence based on the set of hidden states and the observation set. The observation sequence is iteratively estimated using a hidden Markov model, and the state transition probability matrix is ​​output. Based on the state transition probability matrix, the forward probability and the self-loop probability are calculated. State transitions with forward probabilities below a preset probability threshold are identified as key transformation bottlenecks, and the optimal intervention window is determined. The forward probability is the probability that the consumer will transition from the current state to the next adjacent state. The self-loop probability is the probability that the consumer will remain in the current state and will not transition to the next adjacent state. The key transformation bottleneck includes the predecessor state and the successor state. To address the key conversion bottlenecks, a logistic regression model is constructed to quantify the impact of tobacco consumption influencing factors on the probability of successful conversion, and the attribution analysis results of the tobacco consumption influencing factors are output. The conversion success probability refers to the probability that a consumer will successfully advance from the current level to the target level within a preset observation period.

2. The method for analyzing the transformation of tobacco consumption behavior according to claim 1, characterized in that, After classifying tobacco consumer behavior into different levels and setting classification criteria for each level, before defining the different levels as a set of hidden states, defining the quantity of a consumer's single purchase as an observation set, converting each consumer's transaction record into an observation sequence, iteratively estimating the observation sequence using a hidden Markov model, and outputting the state transition probability matrix, the following steps are also included: Obtain consumer transaction data and feedback data, preprocess the transaction data and feedback data to obtain preprocessed data; Based on the preprocessed data, the number of qualified consumers at different levels and the conversion rate between different levels are statistically analyzed; the conversion rate refers to the proportion of consumers who move from the current level to the next adjacent level to the total number of consumers at the current level.

3. The method for analyzing the transformation of tobacco consumption behavior according to claim 2, characterized in that, The preprocessing of the transaction data and feedback data to obtain preprocessed data includes: Fill in the missing purchase records; Remove records where the quantity of a single purchase exceeds a preset threshold, and also remove records with a value of zero. Using the consumer's first purchase date as the base date, the time of the transaction is calculated based on the relative number of days from the base date to obtain the consumer's transaction time series; The transaction time sequence of the consumer is traversed by a sliding window of a preset length to determine whether the purchase frequency of the consumer in any consecutive time window meets the preset level judgment criteria. The consumer's level is labeled according to the judgment result, and the preprocessed data is obtained.

4. The method for analyzing the transformation of tobacco consumption behavior according to claim 1, characterized in that, The process involves defining different levels as a set of hidden states, defining the quantity of a consumer's single purchase as an observation set, converting each consumer's transaction record into an observation sequence, iteratively estimating the observation sequence using a hidden Markov model, and outputting a state transition probability matrix, including: The different levels are defined as the set of hidden states of the hidden Markov model; Discretize the quantity of a consumer's single purchase into multiple observation levels, and define them as an observation set; Each consumer's transaction records are converted into an observation sequence in chronological order of purchase time; each observation value in the observation sequence is determined based on the observation level in the observation set that matches the purchase quantity of that transaction record. The observed sequence of the consumer is iteratively estimated using the Hidden Markov Model until a preset convergence condition is met, and the estimated state transition probability matrix is ​​output.

5. The method for analyzing the transformation of tobacco consumption behavior according to claim 1, characterized in that, The step of calculating the forward probability and the self-loop probability based on the state transition probability matrix, identifying state transitions with forward probabilities below a preset probability threshold as key transformation bottlenecks, and determining the optimal intervention window includes: Calculate the forward probability and self-loop probability of each state from the state transition probability matrix; State transitions with probabilities below a preset probability threshold are identified as key transformation bottlenecks. The average number of dwell steps is calculated based on the self-loop probability of the key conversion bottleneck. Obtain the average behavior interval time of consumers, and multiply the average number of steps on the stand by the average behavior interval time to obtain the average stand duration; The optimal intervention window is determined based on the average dwell time.

6. The method for analyzing the transformation of tobacco consumption behavior according to claim 5, characterized in that, The constructed logistic regression model quantifies the impact of tobacco consumption influencing factors on the probability of successful conversion, and outputs the attribution analysis results of the tobacco consumption influencing factors, including: Consumers who are in the preceding state of the key conversion bottleneck will be used as the analysis sample. The logistic regression model is constructed by taking whether the analyzed sample advances to the subsequent state of the key transformation bottleneck within the preset observation period as the dependent variable and the factors affecting tobacco consumption as the independent variables. Based on the logistic regression model, estimate the regression coefficients of the independent variables on the dependent variable; Calculate the odds ratio of the independent variables based on the regression coefficients, and output the attribution analysis results including the odds ratios.

7. The method for analyzing the transformation of tobacco consumption behavior according to claim 6, characterized in that, The step of determining tobacco marketing intervention strategies based on the attribution analysis results and the optimal intervention window includes: The intervention targets are determined based on the key conversion bottlenecks; the intervention targets are consumers who are in the leading state of the key conversion bottlenecks. Determine the timing of intervention based on the optimal intervention window; The intervention direction is determined based on the tobacco consumption influencing factors whose odds ratio is higher than the preset odds ratio threshold in the attribution analysis results. Based on the intervention target, the intervention timing, and the intervention direction, a tobacco marketing intervention strategy is generated.

8. A device for analyzing the transformation of tobacco consumption behavior, characterized in that, include: The module is divided into an iteration module, an analysis and calculation module, and a result output module; among which: The partitioning iteration module is used to divide the tobacco consumer behavior state into different levels and set classification criteria for different levels; the different levels are defined as a set of hidden states, and the quantity of a consumer's single purchase is defined as an observation set. Based on the set of hidden states and the observation set, the consumer's transaction records are converted into an observation sequence. The observation sequence is iteratively estimated using a hidden Markov model, and a state transition probability matrix is ​​output. The analysis and calculation module is used to calculate the forward probability and the self-loop probability based on the state transition probability matrix, identify state transitions with forward probabilities below a preset probability threshold as key transformation bottlenecks, and determine the optimal intervention window; the forward probability is the probability that the consumer will transition from the current state to the next adjacent state; the self-loop probability is the probability that the consumer will remain in the current state and will not transition to the next adjacent state; the key transformation bottleneck includes the predecessor state and the successor state. The result output module is used to construct a logistic regression model to quantify the impact of tobacco consumption influencing factors on the conversion success probability for the key conversion bottleneck, and output the attribution analysis results of the tobacco consumption influencing factors; the conversion success probability refers to the probability that a consumer will successfully advance from the current level to the target level within a preset observation period.

9. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the tobacco consumption behavior conversion analysis method according to 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 tobacco consumption behavior conversion analysis method according to any one of claims 1 to 7.

Images