Big data intelligent delivery system based on marketing data analysis

Abstract

The invention relates to the technical field of fast moving consumer goods data analysis, and discloses a marketing data analysis-based big data intelligent delivery system which comprises a demand analysis unit, a scene analysis unit, a fusion unit, a division unit and a delivery unit. The method comprises the steps of generating user demand potential energy intensity, dividing the user demand potential energy intensity into three types of circle layers, accurately identifying user groups with potential consumption demands but weak current feedback, adopting differentiated delivery decisions according to types of price sensitivity, information missing degree, decision complexity and the like for main retardation factors of different circle layers, and correcting a strategy in combination with a consumption willingness value. Decision resistance of users in each circle layer can be effectively eliminated, potential consumption potential is converted, resource mismatching is avoided, and the return rate of long-term big data delivery of home fast moving consumer goods is increased.

Description

Technical Field

[0001] The present invention relates to the technical field of fast-moving consumer goods data analysis, and in particular to a big data intelligent delivery system based on marketing data analysis. Background Art

[0002] At present, the core of intelligent coupon delivery lies in determining the delivery targets based on marketing data, and using machine learning algorithms to conduct in-depth analysis of users' purchase history, browsing history, favorites and other data, so as to accurately deliver more resources to user groups with stronger feedback behavior, thereby ensuring the return on delivery.

[0003] However, in the field of household fast-moving consumer goods, there are significant drawbacks in the intelligent delivery of big data. Specifically, in the intelligent delivery of coupons, the existing delivery decisions are overly dependent on users' explicit feedback behavior, resulting in insufficient recognition of user groups in the field of household fast-moving consumer goods who generally have potential consumption demand but currently have weak feedback behavior. It is impossible to actually convert the consumption potential of these users, resulting in a low long-term delivery return rate. Summary of the Invention

[0004] In response to the deficiencies of the existing technology, the present invention provides a big data intelligent delivery system based on marketing data analysis to solve the above problems.

[0005] The above technical objectives of the present invention are achieved through the following technical solutions:

[0006] Big data intelligent delivery system based on marketing data analysis, including:

[0007] The demand analysis unit is used to obtain marketing data of household fast-moving consumer goods, analyze the marketing data, obtain the implicit demand intensity index and consumption willingness value, and fuse the implicit demand intensity index and consumption willingness value to generate the willingness entropy value;

[0008] The scenario analysis unit is used to analyze the intention entropy value, generate the decision block code of time and space constraints, extract marketing data, and obtain the user's scenario adaptability;

[0009] The fusion unit is used to fuse the willingness entropy value, decision block coding and scene adaptability to generate the user demand potential energy intensity;

[0010] The division unit is used to divide users into three circles according to the potential strength of user needs and decision-making blockage codes;

[0011] The delivery unit is used to calculate the main blocking factors of the decision blocking codes for the three types of circles respectively, and adopt different delivery decisions based on the main blocking factors.

[0012] Furthermore, by analyzing the marketing data, we can obtain the implicit demand intensity index and consumption willingness value, including:

[0013] Analyze the user's focus shift at different decision-making stages based on marketing data and generate behavioral entropy values;

[0014] Analyze the semantic similarity between user expressions and product attributes in marketing data to generate semantic coupling coefficients;

[0015] After weighted fusion of behavioral entropy value and semantic coupling coefficient, the demand fluctuation coefficient is obtained;

[0016] The behavioral entropy value, semantic coupling coefficient and demand fluctuation coefficient are integrated to obtain the implicit demand intensity index and consumption willingness value.

[0017] Furthermore, the behavioral entropy value, semantic coupling coefficient, and demand fluctuation coefficient are integrated to obtain the implicit demand intensity index and consumption willingness value, including:

[0018] Map the behavior entropy value, semantic coupling coefficient, and demand fluctuation coefficient into a three-dimensional tensor space to generate a composite feature matrix;

[0019] Transform the composite characteristic matrix to generate a four-dimensional demand potential energy field;

[0020] Analyze the four-dimensional demand potential field and obtain the implicit demand intensity index;

[0021] The four-dimensional demand potential field is spatially divided to generate consumption willingness values, which are divided into four levels: S, A, B and C.

[0022] Furthermore, the implicit demand intensity index is integrated with the consumption intention value to generate the intention entropy value, including:

[0023] Analyze the implicit demand intensity index and consumption willingness value to obtain a dynamic weight matrix, and perform time decay on the dynamic weight matrix to obtain an asymmetric three-dimensional tensor;

[0024] Perform characteristic state superposition on the asymmetric three-dimensional tensor to generate a composite quantum state vector containing the user's true intention;

[0025] Constrained features are extracted from the composite quantum state vector to obtain the intention entropy value with spatiotemporal characteristics.

[0026] Furthermore, the willingness entropy value is analyzed to generate the decision block coding of time and space constraints, including:

[0027] The intention entropy value is mapped to a quantum state vector and then converted to generate a time-space frequency characteristic spectrum;

[0028] Extract the time-space frequency characteristic spectrum to obtain the time attenuation coefficient, spatial thermal value and intention divergence value;

[0029] The time attenuation coefficient, spatial thermal value and intention divergence value are integrated to obtain a multi-dimensional blocking factor matrix;

[0030] The multidimensional blocking factor matrix is ​​analyzed and encoded to obtain a decision blocking code with spatiotemporal constraint characteristics.

[0031] Furthermore, the marketing data is extracted to obtain the user's scenario suitability, including:

[0032] Determine the spatial and temporal density of user activities based on marketing data;

[0033] Combining the spatiotemporal density with the functional attributes of household fast-moving consumer goods, we can obtain the field overlap coefficient;

[0034] Based on the field overlap coefficient, the user demand evolution path is determined and the situation transition probability matrix is ​​generated;

[0035] The spatiotemporal density, field overlap coefficient and situational transition probability matrix are asymmetric tensor fused to generate the user's scene adaptation.

[0036] Furthermore, the willingness entropy, decision block coding, and scenario adaptability are integrated to generate the user demand potential energy intensity, including:

[0037] Calculate the willingness entropy value and decision block code to generate a four-dimensional space-time willingness field matrix;

[0038] According to the scene adaptability, the four-dimensional space-time intention field matrix is ​​modified to generate a modified space-time intention field;

[0039] Analyze the spatiotemporal willingness field and generate the user demand potential matrix;

[0040] Extract the user demand potential energy matrix and obtain the user demand potential energy intensity.

[0041] Furthermore, based on the potential strength of user needs and decision-making blockage coding, users are divided into three circles, including:

[0042] Analyze the four-dimensional space-time will field matrix to determine the dynamic potential energy threshold vector;

[0043] Analyze the decision block coding to obtain the resonance between users and the value proposition of home fast-moving consumer goods;

[0044] Based on the dynamic potential energy threshold vector and the resonance degree, three types of user circles are generated in the four-dimensional space, namely the high resonance layer, the steady-state resonance layer and the low resonance layer.

[0045] Furthermore, for the three types of circles, the main blocking factors of decision blocking coding are calculated respectively, including:

[0046] Extract the decision-making block code to obtain the coupling strength between each factor and the user demand potential;

[0047] Based on the user's circle, differentiated coupling strength thresholds and entropy reduction efficiency weight coefficients are configured;

[0048] Based on the coupling strength threshold and entropy reduction efficiency weight coefficient, the coupling strength is weighted to generate a decision resistance distribution cloud map;

[0049] Calculate the energy entropy reduction gradient of each blocking factor in the decision resistance distribution cloud map and generate an entropy reduction efficiency ranking table;

[0050] The top three factors from the entropy reduction efficiency ranking table are extracted as candidate main blocking factors, and the entropy reduction contribution rate of each candidate factor is calculated respectively;

[0051] The main blocking factor is determined based on the entropy reduction contribution rate.

[0052] Furthermore, different placement decisions are made based on the main blocking factors, including:

[0053] When the main blocking factor type is price sensitivity, the first placement decision is adopted;

[0054] When the main blocking factor type is information loss, the second placement decision is adopted;

[0055] When the main blocking factor type is decision complexity, the third placement decision is adopted.

[0056] In summary, the present invention mainly has the following beneficial effects:

[0057] By mining the implicit demand intensity index and combining it with the decision-making block coding and scenario adaptability of time and space constraints, we generate the potential intensity of user demand, effectively overcoming the drawback of traditional delivery that relies too much on explicit feedback. By analyzing the behavioral entropy and semantic coupling coefficient of users in the four stages of cognition, interest, consideration, and decision-making, we can accurately capture user groups with potential consumption needs but weak current feedback. This multi-dimensional fusion identification method breaks through the limitations of a single data dimension, provides a precise target group for subsequent precise delivery, and solves the problem of insufficient identification of users with implicit needs in existing solutions.

[0058] By dividing users into three circles and formulating differentiated delivery decisions based on the main blocking factors of each circle, the delivery of big data can be adapted to the needs of different users. At the same time, combined with the consumption intention value correction strategy, it ensures that the delivered content is highly matched with user needs, avoids the waste of resources caused by indiscriminate delivery, significantly improves the conversion efficiency of potential users, and especially activates the consumption potential of users in the low resonance layer.

[0059] By activating the consumption potential of potential users and reducing resource mismatch, the long-term return on investment is effectively improved. On the one hand, the potential value is quantified by the potential intensity of user demand to avoid excessive concentration of resources on users with explicit feedback. On the other hand, based on spatiotemporal dynamic analysis and scene adaptability, the investment strategy is adjusted in real time to ensure that the investment decision is synchronized with the evolution path of user demand. It can not only convert current potential users, but also optimize the long-term investment strategy by continuously exploring the evolution laws of user demand, thus ensuring the long-term effectiveness of household fast-moving consumer goods in big data investment. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a schematic diagram of the big data intelligent delivery system based on marketing data analysis of the present invention. DETAILED DESCRIPTION

[0061] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0062] refer to Figure 1 , a big data intelligent delivery system based on marketing data analysis, including:

[0063] The demand analysis unit is used to obtain marketing data of household fast-moving consumer goods, analyze the marketing data, obtain the implicit demand intensity index and consumption intention value, and integrate the implicit demand intensity index and consumption intention value to generate the intention entropy value. The marketing data includes: purchase behavior data, browsing behavior data, collection and add-to-cart data, click interaction data, evaluation and feedback data, household attribute data, and geographic location data, environmental data, and time data;

[0064] The scenario analysis unit is used to analyze the intention entropy value, generate the decision block code of time and space constraints, extract marketing data, and obtain the user's scenario adaptability;

[0065] The fusion unit is used to fuse the willingness entropy value, decision block coding and scene adaptability to generate the user demand potential energy intensity;

[0066] The division unit is used to divide users into three circles according to the potential strength of user needs and decision-making blockage codes;

[0067] The delivery unit is used to calculate the main blocking factors of the decision blocking codes for the three types of circles respectively, and adopt different delivery decisions based on the main blocking factors.

[0068] By mining the implicit demand intensity index, combining the decision-making blockage coding and scenario adaptability of time and space constraints, generating user demand potential intensity and dividing circles, we overcome the drawbacks of over-reliance on explicit feedback, and can accurately identify user groups with potential consumption needs but weak current feedback. By taking differentiated delivery decisions based on the main blockage factors of different circles, we can effectively transform the consumption potential of potential users and improve the long-term return on investment.

[0069] In one case of this embodiment, the marketing data is analyzed to obtain the implicit demand intensity index and the consumption willingness value, including:

[0070] According to the marketing data, the focus of users in different decision-making stages is analyzed to generate behavioral entropy values, including: clarifying the four stages of user decision-making: cognition (first contact with the product), interest (active understanding), consideration (comparison and selection) and decision-making (placing an order to purchase), extracting focus indicators for each stage from the marketing data, such as the number of category page visits in the browsing behavior data of the cognition stage, the number of product collections in the collection and add-to-cart data of the interest stage, the frequency of mentions of competitor comparison keywords in the evaluation and feedback data of the consideration stage, and the stay time on the payment page in the purchase behavior data of the decision-making stage. The indicator data of each stage are counted, and the frequency of occurrence of each focus point in the stage is calculated. For example, in the cognition stage, the three categories of household fast-moving consumer goods pages visited by users are counted, category A is visited 10 times, category B is visited 8 times, category C is visited 2 times, and the total number of visits is 20 times. Then the frequency of occurrence of category A focus points is 10 / 20=0.5, Category B is 0.4, and Category C is 0.1. The entropy value of a single stage is then calculated using the information entropy algorithm. Specifically, the frequencies of all focus points in that stage are multiplied by their logarithms (base 2), and the negative numbers are summed to obtain the entropy value of a single stage. The number of times users shift from focusing on product prices (interest stage indicator) to focusing on materials (consideration stage indicator) from the interest stage to the consideration stage is counted, and the number of times this shift is divided by the total number of shifts from the interest stage to the consideration stage is obtained to obtain the probability of shifting focus points between stages. Weights are set for the four stages (cognition 0.1, interest 0.2, consideration 0.3, and decision-making 0.4). The entropy values ​​of each stage are multiplied by the corresponding weights and then added together. The behavioral entropy value is then obtained by multiplying the transition probability between stages by the weight between stages (the weight of transitions between adjacent stages is 0.5). The higher the behavioral entropy value, the more unstable the changes in user focus points at different stages.

[0071] Analyze the semantic similarity between user expressions and product attributes in marketing data to generate a semantic coupling coefficient. This involves extracting descriptive words mentioned by users (such as durable and non-slip) from the evaluation and feedback data of the marketing data, compiling a list of characteristic words for product attributes (such as product characteristic words such as sturdy material and non-slip surface), and counting the number of matches between user words and product characteristic words in the four stages of cognition, interest, consideration, and decision-making. For example, in the cognition stage, the number of user words with the same meaning as the product characteristic words is obtained. For example, if the user says "durable" and the product characteristic word includes "wear-resistant material", it means that the user word and the product characteristic word are matched. The number of matches between user words and product characteristic words is divided by the total number of words to obtain the stage matching rate. The four stages are assigned weights of 0.1, 0.2, 0.3, and 0.4 respectively. The matching rates of each stage are multiplied and summed. The sum is the semantic coupling coefficient.

[0072] The demand fluctuation coefficient is obtained by weighting the behavioral entropy value and the semantic coupling coefficient. Specifically, the behavioral entropy value and the semantic coupling coefficient are weighted separately, with the behavioral entropy value weighted to 0.6 and the semantic coupling coefficient weighted to 0.4. The behavioral entropy value and the semantic coupling coefficient are multiplied by 0.6 and 0.4, and then added together. The resulting sum is the demand fluctuation coefficient. Among them, in household fast-moving consumer goods, the behavioral entropy value reflects the dynamic shift of users' focus in the four stages of cognition, interest, consideration, and decision-making, and is directly related to the stability of the decision-making process. Its value can immediately reflect the severity of demand fluctuations and is closely related to the real-time behavioral changes that need to be captured in launch decisions. Therefore, the behavioral entropy value is given a higher weight (0.6) to highlight the impact of dynamic decision-making behavior. The semantic coupling coefficient is the static matching result between user expression and product attributes, which is used to reflect the basic demand fit. Its immediate impact on demand fluctuations is weaker than dynamic behavioral changes. Therefore, the semantic coupling coefficient is given a weight (0.4) as a supplement to the basic matching to fit the dynamic characteristics of household fast-moving consumer goods user decision-making.

[0073] The behavioral entropy value, semantic coupling coefficient and demand fluctuation coefficient are integrated to obtain the implicit demand intensity index and consumption willingness value.

[0074] By analyzing the focus shift of users in the four stages of decision-making to generate behavioral entropy values, and combining them with the semantic coupling coefficient to obtain the demand fluctuation coefficient, we can accurately capture potential consumer demand and enhance the ability to identify users with weak feedback but potential demand. By integrating behavioral entropy values, semantic coupling coefficients and demand fluctuation coefficients, we can obtain the implicit demand intensity index and consumption willingness value, which fits the dynamic characteristics of user decision-making in household fast-moving consumer goods. The delivery decisions based on this can reduce excessive reliance on explicit feedback, improve the conversion efficiency of potential users, and help improve the long-term return on investment.

[0075] In one case of this embodiment, the behavior entropy value, the semantic coupling coefficient, and the demand fluctuation coefficient are integrated to obtain the implicit demand intensity index and the consumption willingness value, including:

[0076] The behavioral entropy value, semantic coupling coefficient, and demand fluctuation coefficient are mapped to a three-dimensional tensor space to generate a composite feature matrix. Specifically, the behavioral entropy value, semantic coupling coefficient, and demand fluctuation coefficient are normalized, and each coefficient value is scaled to the interval of 0-1 according to its value range. The three normalized coefficients are used as the x, y, and z axes of the space respectively. The three normalized coefficient values ​​of a single user are mapped to a coordinate point in the space. Then, the coordinate points of multiple users are aggregated to form a composite feature matrix.

[0077] The composite feature matrix is ​​transformed to generate a four-dimensional demand potential field. Specifically, the following steps are performed: sorting the standardized data of the behavioral entropy, semantic coupling coefficient, and demand fluctuation coefficient of all users in the composite feature matrix, calculating the correlation between the three coefficients, and finding the main directions that can reflect these correlations. Each direction represents a set of features that integrate the key information of the three coefficients (the main directions are: comprehensive feature directions that can simultaneously reflect the main changing patterns of the three coefficients. For example, one direction can reflect the common correlation that the semantic coupling coefficient is high when the behavioral entropy value is high, and another direction can reflect the situation where the demand fluctuation coefficient changes alone). The explanatory power of the original data (the degree to which it reflects the overall situation of the data) is ranked according to these comprehensive features, and the top four with the strongest explanatory power are retained. The key features extracted at different time points are sorted in chronological order, and a four-dimensional space is constructed. The first three dimensions correspond to the standardized values ​​of the three coefficients, and the fourth dimension is time. Finally, the key features of each time point are mapped to the specific position in the four-dimensional space. The combination of all positions forms the four-dimensional demand potential field.

[0078] The four-dimensional demand potential energy field is analyzed to obtain the implicit demand intensity index, which specifically includes: extracting the characteristic data corresponding to each user at different time points from the four-dimensional demand potential energy field. These data contain the standardized values ​​of the three coefficients and time information. Users with similar characteristics are grouped into the same group to form multiple demand characteristic clusters. The density of each cluster is calculated, that is, the proportion of the number of users in the cluster to the total number of users. At the same time, the change trend of each cluster in the time dimension is calculated. The density and time change trend weights of the cluster are set to 0.7 and 0.3 respectively. The density of each cluster is multiplied by the corresponding weight, and the time change trend is multiplied by the corresponding weight to obtain the potential energy value within the cluster. The maximum and minimum potential energy values ​​in all clusters are found. The potential energy value of the cluster to which each user belongs is subtracted from the minimum value, and then divided by the difference between the maximum and minimum values ​​to obtain a value between 0 and 1. This value is multiplied by 100 to obtain the implicit demand intensity index in the range of 0-100.

[0079] The four-dimensional demand potential field is spatially divided to generate consumption willingness values, which are divided into four levels: S, A, B, and C. Specifically, all user feature points in the four-dimensional demand potential field are divided into four types of areas, and the implicit demand intensity index of each area center is calculated. The implicit demand intensity index is sorted from high to low, corresponding to S, A, B, and C. Among them, the area where the user is located is the level of his consumption willingness value, and S level corresponds to the implicit demand intensity index interval. , Grade A corresponds to the implicit demand intensity index range , Grade B corresponds to the implicit demand intensity index range , C level corresponds to the implicit demand intensity index range .

[0080] By constructing a three-dimensional tensor space and a four-dimensional demand potential field, user behavior and demand characteristics are analyzed in the time and space dimensions, forming a demand feature cluster and calculating the potential energy value, and accurately extracting the implicit demand intensity index. This breaks through the limitation of relying solely on explicit feedback, and can deeply explore users with weak feedback behavior but strong potential demand, thereby improving the ability to identify potential groups.

[0081] By dividing the consumption willingness levels in four-dimensional space and combining the changing trends in the time dimension, we can dynamically capture the evolution of user needs. Based on the implicit demand intensity index and consumption willingness value, we can convert potential users in a targeted manner, avoid resource mismatch, effectively activate long-term consumption potential, solve the problem of the continued decline in the return on investment of existing investment, and ensure the long-term effectiveness of investment.

[0082] In one case of this embodiment, the implicit demand intensity index is integrated with the consumption willingness value to generate the willingness entropy value, including:

[0083] Analyze the implicit demand intensity index and consumption willingness value to obtain a dynamic weight matrix, and perform time decay on the dynamic weight matrix to obtain an asymmetric three-dimensional tensor. Specifically, for each consumption level under the consumption willingness value, count the number of users in each implicit demand intensity index interval of that level, then divide the number of users in each interval by the total number of users of that level to obtain the correlation ratio between that level and the corresponding index interval. These ratios are arranged by level and implicit demand intensity index interval to form a dynamic weight matrix. Record the collection time corresponding to each data in the dynamic weight matrix. Based on the current time, calculate the time distance of each data point (for example, if the data was collected 10 days ago, the distance is 10). For each weight value in the dynamic weight matrix, decay according to the time distance: for every increase of 1 in the distance, the weight value is multiplied by a fixed decay coefficient (0.95). Due to the different attenuation degrees at different time points, the weight values ​​of each position in the matrix are asymmetric with the time dimension, ultimately forming an asymmetric three-dimensional tensor containing three dimensions: consumption level, index interval, and time.

[0084] The characteristic states of the asymmetric three-dimensional tensor are superimposed to generate a composite quantum state vector containing the user's true intention. Specifically, the following steps are performed: the three dimensions of the asymmetric three-dimensional tensor are clearly defined as consumption level, index interval, and time. The most representative characteristic state in each dimension is extracted, that is, the typical state reflecting the distribution law of the main data in that dimension, such as the most frequently occurring level combination in the consumption level dimension, the interval range where users are concentrated in the index interval dimension, and the key time point of data change in the time dimension. The weight of each characteristic state in its dimension is calculated. The weight is determined by the degree of explanation of the original data by the characteristic state, and the degree of explanation corresponds to the weight (for example, if the degree of explanation is 60%, the weight is 0.6). All characteristic states in the three dimensions are adjusted according to their corresponding weights, and the adjusted characteristic states of each dimension are superimposed and merged. The final vector formed is the composite quantum state vector containing the user's true intention.

[0085] Constrained feature extraction is performed on the composite quantum state vector to obtain the intention entropy with spatiotemporal characteristics. Specifically, for the composite quantum state vector, the time window (nearly 7 days), consumption level, and spatial range of the index interval are used as constraints. The fluctuation amplitude of each feature in the vector is calculated within the constraint range (reflecting the uncertainty of intention), the frequency of occurrence of each fluctuation amplitude is counted, the frequency is multiplied by its logarithm with base 2, and the negative number is summed to obtain the intention entropy with spatiotemporal characteristics.

[0086] By forming an asymmetric three-dimensional tensor through the time decay of the dynamic weight matrix and superimposing characteristic states to generate a composite quantum state vector, it is possible to deeply explore the user's true intentions, extract the intention entropy in combination with time and space constraints, break through the limitations of explicit feedback, accurately capture the true intentions of users with potential consumption needs but weak feedback, and improve their recognition ability. By integrating the implicit demand intensity index and the consumption intention value to generate intention entropy, taking into account the time and space characteristics and the dynamic changes of user intentions, the delivery decision can effectively transform the consumption potential of potential users, avoid resource mismatch, and solve the problem of low long-term delivery return rate.

[0087] In one case of this embodiment, the willingness entropy value is analyzed to generate a decision block code of the spatiotemporal constraint, including:

[0088] The intention entropy value is mapped to a quantum state vector and then converted to generate a time-space frequency characteristic spectrum. Specifically, the following steps are performed: the time window (nearly 7 days), the spatial range of the consumption level and the index range, and the fluctuation characteristics contained in the intention entropy value are mapped to different components of the quantum state vector. The value of each component is determined by the specific data of the corresponding element, and then the mapping is completed. For the time dimension of the quantum state vector, the number of occurrences of each fluctuation feature in the past 7 days is counted, and the frequency of repeated occurrence of the feature at different time intervals is calculated to obtain the time-frequency feature. For the spatial dimension, the distribution of the combination of each consumption level and index range is also counted, and the frequency of repeated occurrence of different combinations within the spatial range is calculated to obtain the spatial frequency feature. Finally, the time and spatial frequency features are integrated according to the corresponding time-space coordinates to form a time-space frequency characteristic spectrum.

[0089] Extract the spatiotemporal frequency characteristic spectrum to obtain the time decay coefficient, spatial heat value, and intention divergence value. Specifically, the following steps are performed: for the time dimension data of the spatiotemporal frequency characteristic spectrum, determine the time frequency characteristic value of each day in the past seven days, calculate the ratio of the daily characteristic value to the characteristic value of the first day, and then average these ratios to obtain the time decay coefficient; count the frequency of occurrence of each consumption level and index interval combination in the spatial dimension of the spatiotemporal frequency characteristic spectrum, and the frequency is the spatial heat value of the combination; calculate the frequency distribution of each feature in the spatiotemporal frequency characteristic spectrum, and then calculate the difference between this distribution and the average frequency distribution of all features, and the difference is the intention divergence value;

[0090] The time decay coefficient, spatial thermal value, and intention divergence value are integrated to obtain a multidimensional blocking factor matrix. Specifically, the weight of the time decay coefficient, spatial thermal value, and intention divergence value are set to 0.4, 0.4, and 0.2, and for each spatiotemporal coordinate point, the time decay coefficient, spatial thermal value, and intention divergence value of the point are multiplied by the corresponding weights and added together. The result is used as the element of the coordinate in the matrix. After all coordinate points are calculated, the matrix formed is the multidimensional blocking factor matrix. Among them, the time decay coefficient reflects the timeliness of the data. User consumption decisions are more affected by recent behavior, so it is given a weight of 0.4. The spatial thermal value reflects the consumption pattern of mainstream users and has a significant impact on decision-making blocks, so it is given a weight of 0.4. The intention divergence value reflects the special needs of users and has a slightly weaker impact, so it is given a weight of 0.2.

[0091] The multidimensional blocking factor matrix is ​​analyzed and encoded to obtain a decision blocking code with spatiotemporal constraint characteristics, specifically including: setting the element value range of the multidimensional blocking factor matrix to 0 1 (0 is no blockage, 1 is complete blockage), 0.3 and 0.7 are taken as thresholds, The element of 0.3 is classified as low resistance and assigned binary 00. The element with a value of 0.7 is classified as medium retardation and assigned 01. The elements with a value of 0.7 are classified as high retardation and assigned 11. According to the matrix space-time coordinates (7 days 3) Sequentially, concatenate the binary codes corresponding to each element and generate a 42-bit code through a 7×3 matrix, which is the spatiotemporal constraint decision blocking code.

[0092] By converting the intention entropy value into a spatiotemporal frequency characteristic spectrum, extracting the time attenuation coefficient, spatial thermal value and intention divergence value and fusing them into a multidimensional blocking factor matrix, we finally generate a decision blocking code with spatiotemporal constraints, which can accurately capture the blocking factors in user consumption decisions, especially identifying the decision-making obstacles in the spatiotemporal dimensions of user groups with strong potential demand but weak current feedback.

[0093] Through in-depth analysis of decision-making blockage coding, the main blockage factors of different user groups can be identified. Targeted delivery strategies can be formulated based on the blockage factors of potential consumer groups, which can effectively reduce their decision-making resistance and activate consumption potential. Compared with the existing delivery method that relies too much on explicit feedback, it can improve the conversion efficiency of users with implicit needs, thereby improving the long-term return on delivery and enhancing the accuracy and long-term effectiveness of home fast-moving consumer goods marketing.

[0094] In one case of this embodiment, extracting marketing data to obtain the user's scenario suitability includes:

[0095] Determine the spatiotemporal density of user activities based on marketing data. This includes extracting user location and time information from the marketing data, dividing the spatiotemporal units into time spans (1 hour) and spatial grids (1 square kilometer), counting the number of user activities within each spatiotemporal unit, and dividing the number by the product of the unit's time span and spatial area to obtain the spatiotemporal density of user activities.

[0096] Combine spatiotemporal density with the functional attributes of household fast-moving consumer goods to derive the field overlap coefficient. Specifically, this includes: determining applicable scenarios based on the functional attributes of household fast-moving consumer goods, such as mops for household cleaning scenarios and tableware for kitchen dining scenarios. For each scenario, set a corresponding time range (e.g., cleaning scenarios are mostly on weekday evenings) and spatial range (e.g., kitchen scenarios correspond to residential kitchen areas). Count the number of user activities within the time and spatial ranges corresponding to the scenarios of each product's functional attributes within the spatiotemporal density of user activities. Divide this number of activities by the total number of user activities to obtain the field overlap coefficient for each product. For all household fast-moving consumer goods, use their relevance to the user as a weight, multiply each product's field overlap ratio by the corresponding weight, and then add up all the products. The resulting sum is the field overlap coefficient.

[0097] Based on the field overlap coefficient, the user demand evolution path is determined and a situation transition probability matrix is ​​generated. Specifically, the following steps are performed: the field overlap coefficients at each time point are sorted in descending order, and the situations corresponding to the highest values ​​are sequentially connected to form the user demand evolution path. The transition frequencies of adjacent situations in the path are counted, and the proportion of this frequency to the total transition frequency of the initial situation is calculated as the transition probability. The transition probabilities of all starting situations and subsequent situations are summarized into a matrix according to the corresponding relationship between rows (starting situations) and columns (subsequent situations), which is the situation transition probability matrix.

[0098] The spatiotemporal density, field overlap coefficient and situational transition probability matrix are asymmetric tensor fused to generate the user's scenario adaptation. Specifically, the following steps are performed: assigning asymmetric weights to the spatiotemporal density, field overlap coefficient and situational transition probability matrix (the spatiotemporal density weight is 0.3, the field overlap coefficient weight is 0.2, and the situational transition probability matrix weight is 0.5). The spatiotemporal density is matched to the situation position in the matrix according to the corresponding spatiotemporal unit and multiplied by the corresponding weight. The field overlap coefficient is directly multiplied by the corresponding weight. The matrix probabilities are multiplied by the corresponding weights and then summarized. The three weighted values ​​are then added together and normalized to 0-100, which is the user's scenario adaptation.

[0099] By extracting the user's spatiotemporal density, combining it with the functional attributes of household fast-moving consumer goods to generate the field overlap coefficient, and then integrating the situational transition probability matrix to obtain the scene adaptability, we can accurately capture the matching degree between users and product functions in specific spatiotemporal scenarios. In particular, for potential users with weak feedback behavior but high scene adaptability, their hidden needs can be explored, which facilitates improving the consumption conversion ability for different users.

[0100] By asymmetric tensor fusion of multi-dimensional data to generate scenario adaptability, taking into account both user activity patterns and product scenario attributes, it can dynamically track the evolution path of demand. The delivery decisions based on this can adapt to the user's real-scene needs, improve the ability to identify and convert potential users, avoid resource mismatch, and effectively solve the problem of insufficient identification of users with hidden needs in existing big data delivery, thereby improving long-term delivery return rate.

[0101] In one case of this embodiment, the willingness entropy value, decision block code, and scenario adaptability are integrated to generate the user demand potential energy intensity, including:

[0102] The willingness entropy value and the decision-making block code are calculated to generate a four-dimensional space-time willingness field matrix. Specifically, the 42-bit binary numbers contained in the decision-making block code are split into groups of two (corresponding to 00, 01, and 11 for low, medium, and high block), and each group of binary numbers is converted into decimal numbers (00 corresponds to 0, 01 corresponds to 1, and 11 corresponds to 3). These decimal numbers together constitute the block base value, and the space-time coordinate correspondence between the willingness entropy value and the block base value is determined. Specifically, the time coordinate corresponds to each day of the past 7 days, and the space coordinate corresponds to the three consumption levels. The combination of the consumption level and the index interval ensures that each consumption level and index interval combination at each time point can match a unique willingness entropy value and a blockage base value. Afterwards, the specific value of the willingness entropy under the corresponding coordinate is multiplied by the blockage base value to obtain the preliminary calculation result under the time-space coordinate. Finally, the preliminary calculation results of all time-space coordinates are arranged in time sequence and space combination sequence to form a matrix containing time and three-dimensional space information. All values ​​in the matrix are then processed so that the range of values ​​is unified between 0 and 1, and the four-dimensional time-space willingness field matrix can be obtained.

[0103] The four-dimensional space-time willingness field matrix is ​​corrected according to the scene adaptability to generate a corrected space-time willingness field, specifically including: converting the scene adaptability (0-100) to the range of 0-1 (dividing by 100), multiplying each element in the four-dimensional space-time willingness field matrix by the converted scene adaptability, and the result is the element value of the corresponding position after correction. The matrix formed by splicing all the corrected elements is the corrected space-time willingness field;

[0104] Analyze the spatiotemporal intention field to generate a user demand potential matrix. Specifically, the following steps are performed: For the modified spatiotemporal intention field, count the element values ​​of the corresponding spatial coordinates at each time point (the past 7 days), use the user activity at each time point as the weight, multiply the element values ​​of the same spatial coordinate at different time points by the corresponding weights, and then add them up to obtain the summary value of each spatial coordinate. These summary values ​​are arranged in order of spatial coordinates. The resulting matrix is ​​the user demand potential matrix;

[0105] Extract the user demand potential energy matrix and obtain the user demand potential energy intensity, specifically including: calculating the sum of the values ​​of all elements in the user demand potential energy matrix, dividing the sum by the total number of elements in the matrix to obtain the average value of the matrix, and normalizing the average value to the range of 0-100 (multiplying the average value by 100). The result is the user demand potential energy intensity.

[0106] By integrating the intention entropy, decision block coding and scenario adaptability to generate the user demand potential intensity, and constructing a four-dimensional space-time intention field matrix and combining it with scenario adaptability correction, it can comprehensively capture the user's potential demand characteristics in the space-time dimension, accurately identify users with weak feedback behavior but high demand potential, and make up for the defect of insufficient identification of users with implicit needs in existing big data delivery. It can also quantify the user's real consumption potential based on the user's demand potential intensity, and then evaluate the user's conversion value, providing an accurate basis for subsequent circle division and big data delivery decisions, helping to formulate effective strategies for potential users, fully transform their consumption potential, solve the problem of low long-term delivery return rate, and improve the overall benefits of smart delivery of home fast-moving consumer goods.

[0107] In one case of this embodiment, users are divided into three circles based on the user demand potential strength and decision-making blockage code, including:

[0108] The four-dimensional space-time willingness field matrix is ​​analyzed to determine the dynamic potential energy threshold vector. Specifically, the following steps are performed: extracting the element values ​​of all space-time coordinates in the four-dimensional space-time willingness field matrix, dividing it into daily sliding windows according to the time dimension, statistically analyzing the distribution of element values ​​in each window, and calculating its 25%, 50%, and 75% quantiles as the low, medium, and high threshold components of the window. The threshold components of all windows are arranged in chronological order to form a dynamic potential energy threshold vector.

[0109] Analyze the decision-making blockage code to obtain the resonance degree between users and the value proposition of household fast-moving consumer goods. Specifically, the following steps are performed: determine the ideal decision state corresponding to the value proposition of household fast-moving consumer goods. For example, if a certain type of product has a high proportion of low and medium blockage, it corresponds to binary 00. Divide the 42 binary bits of the decision-making blockage code into 21 groups with each two bits corresponding to a blockage level of a time-space coordinate (00 is low blockage, 01 is medium blockage, and 11 is high blockage). Count the difference between the actual blockage level and the ideal blockage level in each of the 21 groups. For example, if the actual blockage level is 01 and the ideal blockage level is 00, it is recorded as 1; if the actual blockage level is 11 and the ideal blockage level is 00, it is recorded as 2; if the actual blockage level is consistent with the ideal blockage level, it is recorded as 0. Add all the difference values ​​to obtain the total difference. Then divide the total difference by the maximum possible difference (i.e., 42 when all 21 groups have the maximum difference of 2) to obtain the difference ratio. Subtract the difference ratio from 1. Normalize the obtained result to obtain the resonance degree between users and the value proposition of the product.

[0110] Based on the dynamic potential threshold vector and resonance degree, three types of user circles are generated in the four-dimensional space, including:

[0111] When the user demand potential energy intensity exceeds the high threshold component of the dynamic potential energy threshold vector and the resonance degree 0.85, judged as the first type of circle: high resonance layer;

[0112] When the user demand potential energy intensity is between the middle threshold component and the high threshold component of the dynamic potential energy threshold vector, and falls within the difference between the two Within the 15% range, the resonance degree is between 0.6-0.85, which is judged as the second type of circle: steady-state resonance layer;

[0113] When the user demand potential energy intensity is lower than the middle threshold component of the dynamic potential energy threshold vector and the resonance degree 0.6, judged as the third type of circle: low resonance layer;

[0114] For users who do not meet the conditions of the above three circles, manual intervention will be used to classify them.

[0115] The dynamic potential threshold vector combined with the resonance degree is used to divide user circles into three categories, and the dynamic potential threshold vector is used to capture the spatiotemporal changes in user demand potential. The resonance degree reflects the degree of fit between users and product value propositions. The combination of the two can accurately identify users with high potential but weak feedback. Differentiated strategies are implemented for the three circles. The high-resonance layer can strengthen investment and consolidate loyalty, the steady-state resonance layer needs to reduce decision-making barriers to promote conversion, and the low-resonance layer can tap potential by activating potential demand. This classification method avoids excessive concentration of resources on users with strong explicit feedback, takes into account the conversion of potential users, and can fully release the consumption potential of each circle, improve long-term investment return rate, and enhance the accuracy and sustainability of big data investment in household fast-moving consumer goods.

[0116] In one case of this embodiment, for three types of circles, main blocking factors of decision blocking coding are calculated respectively, including:

[0117] The decision-making blockage code is extracted to obtain the coupling strength between each factor and the user demand potential energy. Specifically, the 42-bit binary code of the decision-making blockage code is divided into 21 groups with two bits per group. Each group corresponds to the blockage factor of a space-time coordinate. Each binary group is then converted to decimal (00 corresponds to 0, 01 corresponds to 1, and 11 corresponds to 3). The potential energy values ​​corresponding to each of the 21 space-time coordinates are found from the user demand potential energy matrix. The correlation between the decimal value of each blockage factor and the potential energy value of the corresponding coordinate is calculated one by one. The correlation degree is the coupling strength of the blockage factor and the user demand potential energy.

[0118] Based on the user's circle, differentiated coupling strength thresholds and entropy reduction efficiency weight coefficients are configured; specifically, they include:

[0119] For high resonance layer users, the coupling strength threshold is set to 0.7, and the entropy reduction efficiency weight coefficient is ;

[0120] For the steady-state resonant layer user, the coupling strength threshold is set to 0.6, and the entropy reduction efficiency weight coefficient is ;

[0121] For low resonance layer users, the coupling strength threshold is set to 0.5, and the entropy reduction efficiency weight coefficient is ;

[0122] Based on the coupling strength threshold and entropy reduction efficiency weight coefficient, the coupling strength is weighted to generate a decision resistance distribution cloud map. Specifically, the following steps are performed: from all coupling strengths, the values ​​that exceed the coupling strength threshold corresponding to the user's circle are selected. Each selected value is split into components according to its corresponding time, space, and intention dimensions. Each dimensional component is then multiplied by the corresponding value in the entropy reduction efficiency weight coefficient. These three products are added together to obtain the weighted value of each selected coupling strength. These weighted values ​​are mapped to the original 21 time and space coordinates, and each coordinate point is colored according to the weighted value. The larger the value, the darker the color. Finally, a decision resistance distribution cloud map is formed.

[0123] The energy entropy reduction gradient of each blocking factor is calculated in the decision resistance distribution cloud map, and an entropy reduction efficiency ranking table is generated. Specifically, the following steps are performed: extracting all the space-time coordinates and weighted values ​​corresponding to each blocking factor from the decision resistance distribution cloud map; calculating the weighted value of each blocking factor at two adjacent time points; specifically, subtracting the weighted value of the previous time point from the weighted value of the latter time point to obtain the difference; if the difference is negative, it means that the entropy is decreasing; dividing this decreased value by the interval between the two time points to obtain the amount of entropy reduction per unit time; based on these unit time reductions, calculating the fastest rate of entropy reduction for each blocking factor, i.e., the energy entropy reduction gradient; arranging all blocking factors from large to small according to their respective energy entropy reduction gradients; and forming a list which is the entropy reduction efficiency ranking table;

[0124] Extract the top three factors from the entropy reduction efficiency ranking table as candidate main blocking factors, and calculate the entropy reduction contribution rate of each candidate factor respectively, specifically including: from the entropy reduction efficiency ranking table, select the top three blocking factors as candidate main blocking factors, calculate the energy entropy reduction gradient of each of these three factors, and then calculate the sum of the three gradients. The gradient of each candidate factor is divided by the sum, and the obtained value is the entropy reduction contribution rate of the factor;

[0125] The main blocking factors are determined based on the entropy reduction contribution rate, including:

[0126] Normalize the contribution rates of all candidate factors so that they are between 0 and 1;

[0127] If the contribution rate of the candidate factor If the value is 0.6, it is directly marked as the main retardation factor to be verified and verified;

[0128] If the contribution rates of all candidate factors are If the value is 0.6, the top two factors with the highest contribution rate are taken and merged into a joint candidate main blocking factor for verification;

[0129] The time validity value is generated by calculating the persistence of the influence of the factor within the user's complete decision cycle based on the time decay coefficient. Specifically, the total duration of the user's complete decision cycle is determined, the cycle is divided into several time periods according to time intervals, the proportion of the duration of each time period to the total duration is used as a weight multiplied by the time decay coefficient corresponding to the time period, and all the products are added together. The result is the time validity value of the factor. The larger the value, the stronger the persistence of the influence.

[0130] Based on the scene adaptation, the correlation density between this factor and the user's high-frequency activity scenes is calculated to generate a spatial adaptation value. Specifically, the following steps are performed: 1. Identify the top five high-frequency activity scenes with the most occurrences of the user, record the specific spatial range and total number of occurrences of each scene, calculate the proportion of the number of occurrences of each high-frequency scene to the total number of all high-frequency scenes as the weight of the scene, then split the scene adaptation according to these high-frequency scenes to obtain the corresponding adaptation of each scene, multiply the adaptation of each scene by the weight of the scene, and then add up all the products to obtain the total. The total is normalized to between 0 and 1. The normalized value is the spatial adaptation value.

[0131] Based on the semantic coupling coefficient, the matching degree between the factor and the user's implicit needs is calculated to generate the intent fit value. Specifically, this includes: extracting descriptive words related to the main blocking factor, compiling a list of characteristic words corresponding to the user's implicit needs, counting the number of matches between the factor descriptive words and the implicit need characteristic words, dividing the number of matches by the total number of words in the two to obtain the basic matching rate, and weighting the basic matching rate and the semantic coupling coefficient with weights of 0.6 and 0.4 to obtain the weighted sum. The sum is then normalized to the range of 0-1 to obtain the intent fit value.

[0132] The time validity value, space adaptation value and intention fit value are integrated to generate a comprehensive verification index. Specifically, for the high resonance layer, the time validity value weight is 0.5, the space adaptation value weight is 0.3, and the intention fit value weight is 0.2. For the steady-state resonance layer, the time validity value weight is 0.4, the space adaptation value weight is 0.4, and the intention fit value weight is 0.2. For the low resonance layer, the time validity value weight is 0.2, the space adaptation value weight is 0.3, and the intention fit value weight is 0.5. The time validity value, space adaptation value and intention fit value are multiplied by the corresponding weights, and the sum is then scaled to the range of 0-1. The result is the comprehensive verification index.

[0133] If the factor is the main blocking factor to be verified and the comprehensive verification index The corresponding circle thresholds (0.7 for high resonance layer, 0.6 for steady-state resonance layer, and 0.5 for low resonance layer) are successfully verified and confirmed as the main blocking factor;

[0134] If the factor is a joint candidate main inhibitor and the comprehensive calibration index Corresponding to the circle threshold (0.7 for high resonance layer, 0.6 for steady-state resonance layer, and 0.5 for low resonance layer), the verification is successful, and the factor with the higher single comprehensive verification index is taken as the main blocking factor;

[0135] If the comprehensive calibration index of the factor If the corresponding circle threshold (high resonance layer 0.7, steady-state resonance layer 0.6, low resonance layer 0.5) is exceeded, the verification fails, and the next candidate factor is extracted from the entropy reduction efficiency ranking table. After further verification, the main blocking factor that meets the conditions can be obtained. If three consecutive candidate factors fail to be verified, it is determined that the user has no clear main blocking factor, and manual intervention is required to make the placement decision.

[0136] For the three types of users in the circle, through differentiated threshold settings and multi-dimensional verification, the main blocking factors of each circle are accurately located. For users in the low resonance layer, potential decision-making obstacles are explored with a lower coupling intensity threshold. Combined with time and space adaptation and implicit demand matching verification, the limitation of relying solely on explicit feedback is broken through, and the problem of insufficient identification of users with potential needs is solved, thereby providing a clear blocking target for converting such users.

[0137] Differentiated delivery based on the main blocking factors can eliminate the decision-making resistance of users in various circles. The high-resonance layer strengthens the advantages and reduces minor obstacles. The steady-state resonance layer focuses on core obstacles to promote conversion. The low-resonance layer accurately breaks through key obstacles to activate potential. This precise policy implementation method can effectively convert potential users, avoid waste of resources, improve the overall delivery conversion efficiency of big data, thereby improving the long-term delivery return rate and enhancing the effectiveness of smart delivery of household fast-moving consumer goods.

[0138] In one case of this embodiment, different delivery decisions are adopted based on the main blocking factor, including:

[0139] If the frequency of price-related keywords in the data associated with the main blocking factor exceeds 60% (such as words like "too expensive" and "not cost-effective"), the type is determined to be price sensitivity.

[0140] If the frequency of user reviews expressing insufficient information exceeds 60% in the data associated with the main blocking factor (for example, words such as "I don't understand the material" or "I want to know the effect of use"), the type is determined to be information missing;

[0141] If, in the data associated with the main blocking factor, the frequency of mentions of competitor comparison keywords (such as comparison words such as which one is better) exceeds 60%, or the dwell time in the consideration phase exceeds the average duration of more than 2 minutes, it is determined to be decision complexity;

[0142] When the main blocking factor type is price sensitivity, the first placement decision is adopted: placing tiered group purchase coupons;

[0143] When the main blocking factor type is information loss, the second placement decision is adopted: placement of experience coupons;

[0144] When the main blocking factor type is decision complexity, the third placement decision is adopted: placing product packaged coupons;

[0145] Modify the first, second, and third placement decisions based on the consumption willingness value. If the consumption willingness value is S or A:

[0146] The first placement decision was revised to: placing high-denomination instant discount coupons;

[0147] The second launch decision was revised to: launch unlimited time trial coupons;

[0148] The third placement decision was revised to: Place personalized packaged coupons;

[0149] If the consumption willingness value is B or C, the original placement decision will be maintained.

[0150] By formulating differentiated delivery decisions for different types of main blocking factors and combining them with consumption intention value correction strategies, we have broken through the current limitations of over-reliance on explicit feedback. By identifying the three main blocking factors of price sensitivity, information lack, and decision complexity, and matching precise tools such as tiered group buying coupons, experience coupons, and product package coupons, we can effectively eliminate the decision-making barriers of users in various circles. At the same time, we dynamically adjust the delivery intensity based on the consumption intention value, upgrade discounts for high-intention users to accelerate conversion, and maintain basic strategies for low-intention users to cultivate demand, which not only guarantees short-term returns but also taps long-term potential. It solves the problem of insufficient conversion of users with implicit needs due to existing big data delivery, and thus can improve the return rate of long-term delivery.

[0151] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A big data intelligent delivery system based on marketing data analysis, characterized by: include: The demand analysis unit is used to obtain marketing data of household fast-moving consumer goods, analyze the marketing data, obtain the implicit demand intensity index and consumption willingness value, and fuse the implicit demand intensity index and consumption willingness value to generate the willingness entropy value; The scenario analysis unit is used to analyze the intention entropy value, generate the decision block code of time and space constraints, extract marketing data, and obtain the user's scenario adaptability; The fusion unit is used to fuse the willingness entropy value, decision block coding and scene adaptability to generate the user demand potential energy intensity; The division unit is used to divide users into three circles according to the potential strength of user needs and decision-making blockage codes; The delivery unit is used to calculate the main blocking factors of the decision blocking codes for the three types of circles respectively, and adopt different delivery decisions based on the main blocking factors.

2. The big data intelligent delivery system based on marketing data analysis according to claim 1 is characterized in that: Analyze marketing data to obtain the implicit demand intensity index and consumption willingness value, including: Analyze the user's focus shift at different decision-making stages based on marketing data and generate behavioral entropy values; Analyze the semantic similarity between user expressions and product attributes in marketing data to generate a semantic coupling coefficient; After weighted fusion of behavioral entropy value and semantic coupling coefficient, the demand fluctuation coefficient is obtained; The behavioral entropy value, semantic coupling coefficient and demand fluctuation coefficient are integrated to obtain the implicit demand intensity index and consumption willingness value.

3. The big data intelligent delivery system based on marketing data analysis according to claim 2 is characterized in that: By integrating the behavioral entropy value, semantic coupling coefficient, and demand fluctuation coefficient, we can obtain the implicit demand intensity index and consumption willingness value, including: Map the behavior entropy value, semantic coupling coefficient, and demand fluctuation coefficient into a three-dimensional tensor space to generate a composite feature matrix; Transform the composite characteristic matrix to generate a four-dimensional demand potential energy field; Analyze the four-dimensional demand potential field and obtain the implicit demand intensity index; The four-dimensional demand potential field is spatially divided to generate consumption willingness values, which are divided into four levels: S, A, B and C.

4. The big data intelligent delivery system based on marketing data analysis according to claim 3 is characterized in that: The implicit demand intensity index is integrated with the consumption willingness value to generate the willingness entropy value, including: Analyze the implicit demand intensity index and consumption willingness value to obtain a dynamic weight matrix, and perform time decay on the dynamic weight matrix to obtain an asymmetric three-dimensional tensor; Perform characteristic state superposition on the asymmetric three-dimensional tensor to generate a composite quantum state vector containing the user's true intention; Constrained features are extracted from the composite quantum state vector to obtain the intention entropy value with spatiotemporal characteristics.

5. The big data intelligent delivery system based on marketing data analysis according to claim 4 is characterized in that: The willingness entropy value is analyzed to generate the decision block coding of time and space constraints, including: The intention entropy value is mapped to a quantum state vector and then converted to generate a time-space frequency characteristic spectrum; Extract the time-space frequency characteristic spectrum to obtain the time attenuation coefficient, spatial thermal value and intention divergence value; The time attenuation coefficient, spatial thermal value and intention divergence value are integrated to obtain a multi-dimensional blocking factor matrix; The multidimensional blocking factor matrix is ​​analyzed and encoded to obtain a decision blocking code with spatiotemporal constraint characteristics.

6. The big data intelligent delivery system based on marketing data analysis according to claim 1 is characterized in that: Extract marketing data to obtain the user's scenario suitability, including: Determine the spatial and temporal density of user activities based on marketing data; Combining the spatiotemporal density with the functional attributes of household fast-moving consumer goods, we can obtain the field overlap coefficient; Based on the field overlap coefficient, the user demand evolution path is determined and the situation transition probability matrix is ​​generated; The spatiotemporal density, field overlap coefficient and situational transition probability matrix are asymmetric tensor fused to generate the user's scene adaptation.

7. The big data intelligent delivery system based on marketing data analysis according to claim 6 is characterized in that: The willingness entropy, decision-making block coding, and scenario adaptability are integrated to generate the user demand potential strength, including: Calculate the willingness entropy value and decision block code to generate a four-dimensional space-time willingness field matrix; According to the scene adaptability, the four-dimensional space-time intention field matrix is ​​modified to generate a modified space-time intention field; Analyze the spatiotemporal willingness field and generate the user demand potential matrix; Extract the user demand potential energy matrix and obtain the user demand potential energy intensity.

8. The big data intelligent delivery system based on marketing data analysis according to claim 7 is characterized in that: Based on the potential strength of user needs and decision-making blockage coding, users are divided into three categories, including: Analyze the four-dimensional space-time will field matrix to determine the dynamic potential energy threshold vector; Analyze the decision block coding to obtain the resonance between users and the value proposition of home fast-moving consumer goods; Based on the dynamic potential energy threshold vector and the resonance degree, three types of user circles are generated in the four-dimensional space, namely the high resonance layer, the steady-state resonance layer and the low resonance layer.

9. The big data intelligent delivery system based on marketing data analysis according to claim 8, characterized in that: For the three types of circles, the main blocking factors of decision blocking coding are calculated respectively, including: Extract the decision-making block code to obtain the coupling strength between each factor and the user demand potential; Based on the user's circle, differentiated coupling strength thresholds and entropy reduction efficiency weight coefficients are configured; Based on the coupling strength threshold and entropy reduction efficiency weight coefficient, the coupling strength is weighted to generate a decision resistance distribution cloud map; Calculate the energy entropy reduction gradient of each blocking factor in the decision resistance distribution cloud map and generate an entropy reduction efficiency ranking table; The top three factors from the entropy reduction efficiency ranking table are extracted as candidate main blocking factors, and the entropy reduction contribution rate of each candidate factor is calculated respectively; The main blocking factor is determined based on the entropy reduction contribution rate.

10. The big data intelligent delivery system based on marketing data analysis according to claim 9 is characterized in that: Different placement decisions are made based on the main blocking factors, including: When the main blocking factor type is price sensitivity, the first placement decision is adopted; When the main blocking factor type is information loss, the second placement decision is adopted; When the main blocking factor type is decision complexity, the third placement decision is adopted.

Images