A partitioned output conversion rate optimization system and method

Abstract

The application relates to the technical field of commodity sample optimization, and discloses a partition sample output conversion rate optimization system and method, which comprises the following modules: a data acquisition and statistics module, which is used for collecting sample output records, sales records and display position information, and generating sample output conversion rate statistical data and position observation data; a differential partition weight module, which is used for constructing a differential partition set and calculating position weights; a commodity relationship calculation module, which is used for determining demand transfer coefficients, complementary gain coefficients and an adjacency rule set; a group constraint generation module, which is used for generating candidate groups and setting category ratio constraints; a conversion target calculation module, which is used for calculating basic conversion values and weighted conversion target values; a combination optimization solution module, which is used for solving sample mapping under multiple constraint conditions; and an iterative update control module, which is used for triggering parameter updating and re-solution based on statistical window recalculation results. The application realizes conversion rate optimization of a commodity sample output scheme under the condition of limited sample output positions.

Description

Technical Field

[0001] This invention belongs to the field of product sampling optimization technology, specifically relating to a conversion rate optimization system and method for zoned sampling. Background Technology

[0002] In retail store operations, the way merchandise is displayed directly impacts sales conversion rates. Factors such as display location, product mix, and the relationship between adjacent products all influence customer attention and purchasing decisions to varying degrees. As stores expand and product varieties increase, the number of display locations available for merchandise becomes limited. How to rationally arrange merchandise within these limited display spaces has long been a challenge in store operations.

[0003] Existing product display management methods often rely on manual experience or adjustments based on single-dimensional indicators, such as historical sales, single-item conversion rates, or fixed display rules. These methods typically ignore differences between stores in terms of location, business district type, and customer flow structure, and rarely consider the objective differences in exposure, dwell time, and traffic flow between different display locations within a store. This makes it difficult to develop a unified and reusable display optimization strategy across different stores or regions.

[0004] On the other hand, while some existing technologies incorporate data analysis methods to evaluate product sales performance, they are mostly focused on single products or simple combinations, failing to differentiate and model the substitution and synergy relationships between products, and lacking mechanisms to comprehensively consider product conversion capabilities and location characteristics at specific display locations. Furthermore, after the display plan is generated, the relevant technologies typically lack control measures for recalculation and adjustment based on new rounds of actual execution data, making it difficult to adapt to changes in the store's operating environment and customer behavior over time. Summary of the Invention

[0005] This invention provides a conversion rate optimization system and method for zoned sampling, which solves the technical problem in related technologies that, when the number of sampling locations is limited, there is a lack of unified modeling and optimization of sampling schemes that combine store differences, sampling location characteristics and product relationships.

[0006] This invention provides a conversion rate optimization system for partitioned sampling, comprising: Data collection and statistics module 1 is used to determine the preset statistics window, collect the store's sample records, sales records and display location information within the preset statistics window to obtain fact details, and calculate the sample conversion rate statistics and location observation data from the fact details; Micro-zone weight module 2 is used to construct a set of micro-zones based on store profile elements and display location types, and to calculate location weights based on location observation data. The commodity relationship calculation module 3 is used to determine the demand transfer coefficient, complementary gain coefficient and adjacency rule set based on the sample conversion rate statistics within the micro-partition set; The grouping constraint generation module 4 is used to generate candidate groups based on the complementary gain coefficient and complementary gain threshold, the demand transfer coefficient and the upper limit threshold of demand transfer, and to set category ratio constraints. The conversion target calculation module 5 is used to determine the basic conversion value based on the sample conversion rate statistics, and to determine the weighted conversion target value based on the location weight, basic conversion value, complementary gain coefficient and demand transfer coefficient. The combinatorial optimization solution module 6 is used to construct a deterministic combinatorial optimization model based on the weighted transformation target value, and solves the sampling mapping under the constraints of unique position, non-repeating products, complete candidate groups, adjacency rule set and category ratio. The iterative update control module 7 is used to recalculate the weighted transformation target value in the new preset statistical window and compare it with the previous preset statistical window. When the comparison result is lower than the preset difference threshold, the demand transfer coefficient, complementary gain coefficient and candidate group return to the combination optimization solution module are updated.

[0007] This invention provides a method for optimizing the conversion rate of partitioned sampling, comprising the following steps: Step 91: Determine the preset statistical window, collect the store's sample records, sales records and display location information within the preset statistical window to obtain fact details, and calculate the sample conversion rate statistics and location observation data from the fact details; Step 92: Construct a set of micro-zones based on store profile elements and display location types, and calculate location weights based on location observation data; Step 93: Within the micro-partition set, determine the demand transfer coefficient, complementary gain coefficient, and adjacency rule set based on the sample conversion rate statistics. Step 94: Generate candidate clusters based on the complementary gain coefficient and complementary gain threshold, the demand transfer coefficient and the upper limit threshold of demand transfer, and set category matching constraints; Step 95: Determine the basic conversion value based on the sample conversion rate statistics, and determine the weighted conversion target value based on the location weight, basic conversion value, complementary gain coefficient and demand transfer coefficient; Step 96: Construct a deterministic combinatorial optimization model based on the weighted conversion target value, and solve for the sample mapping under the constraints of unique location, non-repeating products, complete candidate groups, adjacency rule set and category ratio. Step 97: Recalculate the weighted transformation target value in the new preset statistical window and compare it with the previous preset statistical window. When the comparison result is lower than the set threshold, update the demand transfer coefficient, complementary gain coefficient and candidate group and return to step 96.

[0008] The beneficial effects of this invention are as follows: By introducing a zoning sampling modeling mechanism, this invention unifies the processing of store profile elements, display location characteristics, and product sampling conversion data, constructing a sampling analysis framework based on micro-zoning at the store level. This allows the differences between different stores and different sampling locations to be effectively characterized and participate in optimization calculations. By distinguishing and modeling the demand transfer relationships and complementary relationships between products, and combining candidate grouping and category ratio constraints, the sampling plan maintains structural consistency and executability. Using the weighted conversion target value as a unified optimization benchmark, the sampling mapping is deterministically combined and solved under multiple business and structural constraints, avoiding the instability of sampling configuration caused by single indicators or human experience. By introducing a recalculation and comparison mechanism based on statistical windows, product relationship parameters and sampling plans can be conditionally updated according to actual execution data, thereby maintaining the continuity and stability of the sampling optimization process in a dynamic operating environment. Overall, this invention achieves systematic and reproducible conversion rate optimization of product sampling plans under limited sampling resources. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of a conversion rate optimization system for partitioned sampling according to the present invention. Detailed Implementation

[0010] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0011] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in one or more embodiments of the present invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0012] like Figure 1 As shown, a conversion rate optimization system for partitioned sampling includes: Data collection and statistics module 1 is used to determine the preset statistics window, collect the store's sample records, sales records and display location information within the preset statistics window to obtain fact details, and calculate the sample conversion rate statistics and location observation data from the fact details; Micro-zone weight module 2 is used to construct a set of micro-zones based on store profile elements and display location types, and to calculate location weights based on location observation data. The commodity relationship calculation module 3 is used to determine the demand transfer coefficient, complementary gain coefficient and adjacency rule set based on the sample conversion rate statistics within the micro-partition set; The grouping constraint generation module 4 is used to generate candidate groups based on the complementary gain coefficient and complementary gain threshold, the demand transfer coefficient and the upper limit threshold of demand transfer, and to set category ratio constraints. The conversion target calculation module 5 is used to determine the basic conversion value based on the sample conversion rate statistics, and to determine the weighted conversion target value based on the location weight, basic conversion value, complementary gain coefficient and demand transfer coefficient. The combinatorial optimization solution module 6 is used to construct a deterministic combinatorial optimization model based on the weighted transformation target value, and solves the sampling mapping under the constraints of unique position, non-repeating products, complete candidate groups, adjacency rule set and category ratio. The iterative update control module 7 is used to recalculate the weighted transformation target value in the new preset statistical window and compare it with the previous preset statistical window. When the comparison result is lower than the preset difference threshold, the demand transfer coefficient, complementary gain coefficient and candidate group return to the combination optimization solution module are updated.

[0013] In one embodiment of the present invention, a preset statistical window is determined, and the sample display records, sales records, and display location information of the store within the preset statistical window are collected to obtain fact details. The sample conversion rate statistics and location observation data are calculated from the fact details, including: Step 11: Determine a preset statistical window, defined by a start time and an end time, with the start time earlier than the end time. This preset statistical window serves as the time boundary for all data collection, statistics, and summarization in this embodiment, constraining the effective range of sample records, sales records, display location information, and location observation data. For sample records that cross the preset statistical window, only the time period falling within the window is retained; for sales records that do not fall within the window, they are not included in subsequent statistical processing, thus avoiding interference with conversion rate calculation results caused by mixing data from different time scales.

[0014] Step 12: Within the preset statistical window, the system collects the corresponding sample display records, sales records, and display location information for each store. The sample display records represent the actual display status of a product at a specific location within the store; the sales records represent the sales quantity of the product within the time window; and the display location information represents the specific sample display location of the product. The system uses store identifier, sample display location identifier, product identifier, and timestamp as unified association fields to associate the above sample display records, sales records, and display location information, thereby forming a factual detail. This factual detail describes the sample display and sales status of each store, each sample display location, and each product at a specific point in time within the preset statistical window.

[0015] Step 13: After generating the fact details, the system aggregates and processes the fact details according to the statistical days. For each combination of store identifier, display location identifier, and product identifier, the system counts whether the product was displayed at the corresponding display location on that day, and accumulates the statistical days in which it was displayed to obtain the number of display days for that combination within the preset statistical window. Simultaneously, the system counts the sales volume generated by this combination within each statistical day and accumulates it to obtain the sales volume. Based on this, the system determines the ratio of sales volume to display days as the display conversion rate of the product at the corresponding store and display location, and generates display conversion rate statistics accordingly. Through this method, the display conversion rate can accurately reflect the conversion from display to sale of a product under actual display conditions, without being affected by non-display time or data outside the statistical window.

[0016] Step 14: Within the preset statistical window, the system summarizes location observation data related to each sampling location identifier. This location observation data includes exposure counts, dwell time, and path pass counts. Exposure counts represent the frequency with which the sampling location is visible to customers; dwell time represents the time customers spend at that location; and path pass counts represent the number of times customers pass through the area where the sampling location is located. After generating the location observation data, the system performs a consistency check on the sampling location identifiers in the sampling conversion rate statistics to ensure that the corresponding sampling location identifier exists in the location observation data and that the exposure counts, dwell time, and path pass counts are all valid values ​​not less than zero. The data that passes the check is encapsulated and output as the basic input for subsequent location weight calculations and zoning sampling optimization.

[0017] Through the above implementation methods, the present invention simultaneously collects and standardizes the store's display behavior, sales behavior, and customer flow characteristics at the display location within a unified preset statistical window. This results in a clear and consistent set of factual details, display conversion rate statistics, and location observation data. This provides a reliable data foundation for subsequent conversion rate optimization based on store zoning, differences in display location, and product characteristics, thereby supporting the accuracy and stability of the conversion rate optimization process under zoned display conditions.

[0018] In one embodiment of the present invention, a micro-zone set is constructed based on store profile elements and display location types, and location weights are calculated based on location observation data, including: Step 21: Obtain the corresponding store profile elements for each store identifier. These elements describe the store's external operating environment and overall attributes, including region, business district type, city level, and store type. Region distinguishes different geographical areas, business district type represents the functional attributes of the commercial area where the store is located, city level distinguishes city development levels, and store type distinguishes different business formats or operating models. The system performs value range validation on the store profile elements, identifying store identifiers whose values ​​do not fall into a preset enumeration set as invalid data, thereby preventing non-standard or abnormal profile data from interfering with subsequent partitioning results.

[0019] Step 22: Obtain the corresponding display location information for each display location identifier under each store identifier, and map the display location information to a predefined display location type. The display location type describes the display attributes of the display location within the store's spatial structure, including endcaps, prime locations, traffic hotspots, and less popular locations. Endcaps represent display locations located at the ends of aisles, prime locations represent display locations at visual or tactile height, traffic hotspots represent locations on customers' main walking paths, and less popular locations represent locations with relatively low exposure and dwell time. For display location identifiers that cannot be mapped to the above display location types, the system determines them as invalid display location type data.

[0020] Step 23 involves removing invalid store identifier records corresponding to invalid store profile elements and invalid display location identifier records corresponding to invalid display location types. For the remaining valid data, the system takes the store profile element corresponding to the store identifier and the display location type corresponding to the display location identifier and combines them in an orderly manner to generate corresponding micro-zone identifiers for each store identifier and display location identifier. By summarizing the different combination results, a set of micro-zones is formed, and a correspondence table between store identifiers, display location identifiers, and micro-zone identifiers is generated simultaneously. The micro-zones are used as the basic analysis units for subsequent conversion rate statistics, product relationship analysis, and display optimization.

[0021] Step 24: Calculate the location weight of each sampling location based on location observation data. Specifically, within the range of each store identifier, the system summarizes the corresponding exposure count, dwell time, and path passage count for each sampling location identifier. To eliminate the influence of differences in store size and customer traffic on the indicator values, the system normalizes the corresponding indicators by using the maximum value of exposure count, dwell time, and path passage count within the same store identifier. Subsequently, the system performs a weighted summation of the normalized indicator results according to preset weights to obtain the location weight reflecting the comprehensive visibility and accessibility of the sampling location. Finally, the system binds and outputs the location weights with the correspondence between store identifiers, sampling location identifiers, and micro-zone identifiers.

[0022] Through the above implementation methods, the present invention combines the external profile features of the store with the spatial display features inside the store to construct a set of micro-partitions with unified semantics and clear boundaries. It also quantifies the relative importance of different sampling locations within the same store through location observation data, providing fine-grained and comparable basic conditions for subsequent sampling combination analysis and conversion rate optimization at the partition level, thereby supporting the refined implementation of the conversion rate optimization process under partitioned sampling conditions.

[0023] In one embodiment of the present invention, within a micro-partition set, determining the demand transfer coefficient, complementary gain coefficient, and adjacency rule set based on sample conversion rate statistics includes: Step 31: Based on the correspondence table between store identifiers, display location identifiers, and micro-zone identifiers, the display conversion rate statistics are mapped to the corresponding micro-zone identifiers. Subsequently, the system constructs a statistical day sample set using micro-zone identifiers, store identifiers, and statistical days as dimensions. This statistical day sample set describes the display and conversion status of products within a specific micro-zone, in a specific store, and on a specific natural day. Display conversion rate statistics that cannot be mapped to any micro-zone identifier are considered invalid data and discarded by the system to ensure that subsequent analysis is based only on data with clear zoning semantics.

[0024] Step 32: Within the same micro-region identifier, for each pair of first and second product identifiers, determine the sampling relationship of the two products on the same statistical day based on the statistical day sample set. When both the first and second product identifiers are sampled on the same statistical day, the corresponding statistical day sample is assigned to the simultaneously sampled sample set; when the second product identifier is sampled on the statistical day but the first product identifier is not sampled, the corresponding statistical day sample is assigned to the non-simultaneously sampled sample set. Through this method, the system distinguishes between common sampling scenarios and individual sampling scenarios between products within the micro-region.

[0025] Step 33: For each pair of first and second product identifiers, the conversion rates of the second product identifier in the simultaneous and non-simultaneous sampling sample sets are summarized to calculate the average conversion rate of the second product identifier under simultaneous sampling conditions and the average conversion rate of the second product identifier under non-simultaneous sampling conditions. The system determines the ratio of the aforementioned two average conversion rates as the retention ratio, and the result of subtracting the retention ratio is determined as the demand transfer coefficient. The demand transfer coefficient is used to characterize the degree of substitution or transfer to the conversion performance of the second product identifier when the first product identifier is sampled within the same micro-region.

[0026] Step 34: Within the same micro-region identifier, based on the statistical day sample set, count the number of statistical days in which the first product identifier and the second product identifier appear simultaneously, as well as the number of statistical days in which the first product identifier appears and the number of statistical days in which the second product identifier appears. Normalize these counts using the total number of statistical days to obtain the co-occurrence ratio and the individual product occurrence ratio of the two products. Subtract one from the ratio of the co-occurrence ratio to the individual product occurrence ratios to determine the complementary gain coefficient. The complementary gain coefficient characterizes the degree of synergistic gain generated by simultaneous sampling of two products relative to independent sampling within the same micro-region.

[0027] After calculating the demand transfer coefficient and the complementary gain coefficient, the system performs a joint judgment on the two types of coefficients. When the complementary gain coefficient is not less than a preset complementary gain threshold and the demand transfer coefficient is not greater than a preset demand transfer upper limit threshold, the system writes the corresponding first product identifier and second product identifier into the adjacency rule set. The adjacency rule set is used to characterize the product pairs that are allowed to be adjacent or jointly sampled within the same micro-partition.

[0028] Through the above implementation methods, the present invention introduces a commodity relationship analysis mechanism based on statistical daily samples at the micro-partition level, which quantifies and describes the substitution and synergy relationships between commodities and expresses them through adjacency rule sets. This enables the subsequent partition sampling and conversion rate optimization process to not only consider the conversion performance of individual commodities, but also to comprehensively consider the impact of commodity combination sampling on the conversion rate, thereby improving the rationality and stability of conversion rate optimization under partition sampling conditions.

[0029] In one embodiment of the present invention, candidate clusters are generated based on complementary gain coefficients and complementary gain thresholds, demand transfer coefficients and demand transfer upper limit thresholds, and category allocation constraints are set, including: Step 41: Within the same micro-region identifier, the system iterates through all combinations of the first and second product identifiers. For each product combination, the system reads the corresponding complementary gain coefficient and demand transfer coefficient. When the complementary gain coefficient is not less than a preset complementary gain threshold and the demand transfer coefficient is not greater than a preset demand transfer upper limit threshold, the first and second product identifiers are identified as a product pair in the pairable set. The pairable set describes product combination relationships within the same micro-region that have a positive synergistic relationship and controlled substitution risk. For product pairs lacking complementary gain coefficients or demand transfer coefficients, the system considers them incomplete data and removes them to ensure the reliability of the pairable set.

[0030] Step 42: After obtaining the pairable set, the system constructs a product association graph based on this set. Product identifiers are used as nodes in the graph, and product pairs from the pairable set are used as edges between nodes to represent allowed pairing relationships between products. The system performs connectivity analysis on the product association graph to identify interconnected sets of nodes, thus obtaining several initial clusters. Initial clusters represent candidate combinations formed by multiple products connected through pairable relationships within the same micro-partition. Subsequently, the system obtains a preset lower limit and a preset upper limit for cluster size, and performs screening and splitting processing on the initial clusters: initial clusters smaller than the preset lower limit are eliminated; initial clusters larger than the preset upper limit are decomposed according to connectivity relationships, ultimately obtaining candidate clusters that meet the size constraints.

[0031] Step 43: After obtaining the candidate clusters, category ratio constraints are set at the micro-region level. Specifically, the system obtains the product category identifiers corresponding to each product identifier within the candidate cluster, and obtains the preset lower limit and upper limit of the category ratio. Within the same micro-region identifier, the system counts the number of products corresponding to each product category identifier in the candidate cluster, and limits the number of products to the range specified by the preset lower limit and upper limit of the category ratio, thereby forming a category ratio constraint for that micro-region. The category ratio constraint is used to describe the allowed quantity structure of different product categories at the candidate cluster level. Finally, the system archives and stores the formed category ratio constraints according to the micro-region identifier.

[0032] Through the above implementation methods, within the micro-region, the present invention further organizes the pairable relationships obtained based on the product relationship coefficient into structured candidate groups, and standardizes the candidate groups through group size constraints and category ratio constraints, so that the subsequent conversion rate optimization process of regional sampling can be carried out under the premise of taking into account the product synergy relationship and the rationality of the category structure, thereby enhancing the adaptability and stability of regional sampling combinations in different stores and different sampling areas.

[0033] In one embodiment of the present invention, step 43 further includes: Step 51: For each candidate group, obtain the product category identifier corresponding to each product identifier within the group. For product identifiers lacking product category identifiers, the system considers them as abnormal data that cannot participate in the category matching verification and removes the product identifier from the candidate group, thereby ensuring that subsequent category statistics and constraint judgments are based on complete and valid category information.

[0034] Step 52: Within the same micro-partition identifier, count the number of products corresponding to each product category identifier in the candidate group. When the number of products corresponding to a certain product category identifier exceeds the preset category ratio limit, the candidate group is not directly eliminated as a whole. Instead, the product identifiers are sorted according to the number of times the product appears in the pairable set. The number of times a product identifier appears in the pairable set reflects the degree of association of the product in the product relationship network. The system eliminates the corresponding product identifiers one by one in ascending order of the number of appearances until the number of products corresponding to the product category identifier meets the preset category ratio limit.

[0035] Step 53: After correcting for the number of categories exceeding the upper limit, the system further verifies the lower limit of category ratio for candidate groups. For candidate groups where the number of products corresponding to a product category identifier is lower than the preset lower limit of category ratio, the system removes the entire candidate group and removes it from subsequent sample combination optimization. For candidate groups that simultaneously meet both the preset lower limit and the preset upper limit of category ratio requirements, the system archives and stores them according to the micro-partition identifier, serving as a set of valid candidate groups within that micro-partition that can be used for subsequent sample mapping solutions.

[0036] Through the above implementation methods, the present invention further refines the category ratio constraints at the candidate group level, introduces a deterministic elimination rule based on the occurrence frequency of pairwise sets, so that while satisfying the category structure constraints, the candidate groups retain as many products as possible that have a high degree of correlation in the product relationship network, thereby providing candidate sampling units with stable structure, clear constraints and reasonable combination for the conversion rate optimization process of partitioned sampling.

[0037] In one embodiment of the present invention, a basic conversion value is determined based on sample conversion rate statistics, and a weighted conversion target value is determined based on location weight, basic conversion value, complementary gain coefficient, and demand transfer coefficient, including: Step 61: Based on the correspondence table between store identifiers, display location identifiers, and micro-zone identifiers, the display conversion rate statistics are mapped to the corresponding micro-zone identifiers. Subsequently, the display conversion rate statistics are summarized along the dimensions of micro-zone identifiers and product identifiers. To avoid deviations in the basic value calculation caused by extreme outliers, the system processes the summarized results according to the same outlier removal rules as the display conversion rate statistics, and determines the basic conversion value of each product identifier within its corresponding micro-zone based on the outlier-removed statistical results. The basic conversion value characterizes the average conversion level of a product within that micro-zone, without considering the specific display location or the influence of adjacent products.

[0038] Step 62: After determining the basic conversion value, the system introduces factors related to the display location and product relationships. Specifically, based on the correspondence table between store identifiers, display location identifiers, and micro-zone identifiers, the system binds the location weights to these identifiers, forming a location binding view. This location binding view describes the relative importance of each display location within its respective micro-zone. Simultaneously, within the same micro-zone identifier, the system merges the complementary gain coefficients and demand transfer coefficients corresponding to the first and second product identifiers, forming a product pair effect view. This product pair effect view centrally describes the positive synergistic and negative substitution relationships generated by product pairs when they are displayed simultaneously.

[0039] Step 63: For each display location under each store identifier, read the corresponding location weight and micro-zone identifier in the location binding view, and further read the basic conversion value of the corresponding product identifier under the micro-zone identifier. The system determines the product of the location weight and the basic conversion value as the basic contribution value of the display location and product combination, which is used to characterize the basic contribution of the location and product combination to the overall conversion target without considering the influence of adjacent display locations.

[0040] The system determines the adjacency relationships between sampling locations based on the adjacency rule set, and then identifies adjacent sampling location identifier pairs based on this. For each adjacent sampling location identifier pair, the system obtains the corresponding first and second product identifiers, and reads the complementary gain coefficient and demand transfer coefficient of the product pair from the product pair effect view. The system determines the positive term by multiplying the complementary gain coefficient by a preset fusion coefficient, and the negative term by multiplying the demand transfer coefficient by the preset fusion coefficient. The result of subtracting the negative term from the positive term is taken as the adjacent effect contribution value of the adjacent product pair. The preset fusion coefficient is a pre-configured fixed parameter used to adjust the influence intensity of the product pair effect in the overall target.

[0041] Finally, the system summarizes the basic contribution value and the adjacent effect contribution value for all sampling location identifiers under each store identifier, forming the weighted conversion target value for that store under the current sampling mapping conditions. This weighted conversion target value serves as the sole objective quantity for the subsequent deterministic combinatorial optimization model, used to solve the matching relationship between sampling locations and products while satisfying constraints on candidate clusters, adjacency rule sets, and category matching.

[0042] Through the above implementation methods, based on the micro-regional sampling conversion rate statistics, the present invention uniformly quantifies and expresses the product's own conversion ability, sampling location differences, and adjacent product sampling relationships, and constructs a weighted conversion target value that can reflect the true conversion performance of regional sampling, providing a clear, comparable, and solvable target basis for subsequent regional sampling conversion rate optimization.

[0043] In one embodiment of the present invention, a deterministic combinatorial optimization model is constructed based on a weighted transformation objective value. Under constraints of unique location, non-repeating products, complete candidate groups, adjacency rule set, and category ratio, the sample mapping is obtained, including: Step 71: Using the store identifier as the boundary, obtain all display location identifiers under that store identifier. Based on the correspondence table between store identifiers, display location identifiers, and micro-zone identifiers, determine the micro-zone identifier and location weight corresponding to each display location identifier. For each micro-zone identifier, the system further obtains the basic conversion value, candidate groups, adjacency rule set, and category matching constraints associated with that micro-zone identifier. If any of the above data is missing from the micro-zone identifier corresponding to a certain display location identifier, the system considers that display location identifier as data that cannot participate in optimization and removes it to ensure that the data entering the model is complete and consistent.

[0044] Step 72: After completing data preparation and filtering, a deterministic combinatorial optimization model is constructed using the weighted conversion target value as the sole objective. This model compares the weighted conversion target values ​​corresponding to different sample mappings under the condition of satisfying the model constraint set, and selects the sample mapping with the largest weighted conversion target value as the model output. The model constraint set includes location uniqueness constraint, product non-repetition constraint, candidate group integrity constraint, adjacency rule set constraint, and category ratio constraint. Specifically, the location uniqueness constraint limits each sample location to only one product; the product non-repetition constraint limits the same product from being sampled repeatedly within the same store; the candidate group integrity constraint limits the sample scheme to meet the structural requirements of candidate groups; the adjacency rule set constraint limits the product combination relationship between adjacent sample locations; and the category ratio constraint limits the quantity range of different product categories in the sample scheme.

[0045] Step 73: After the deterministic combinatorial optimization model is established, the system first outputs an initial solution that satisfies the model constraint set. Then, within the same store identifier range, the system selects two sampling location identifiers and swaps their corresponding product identifiers. The system re-verifies whether the model constraint set is still satisfied after the swap, and recalculates the weighted conversion target value after the swap. When the weighted conversion target value after the swap is not less than the weighted conversion target value before the swap, and the model constraint set remains satisfied, the system executes the swap operation. The system repeats the above swap and verification process until there are no more swap operations that meet the conditions, thereby obtaining a stable sampling mapping. The sampling mapping is used to describe the final correspondence between each sampling location identifier and product identifier within the current store.

[0046] Through the above implementation methods, the present invention integrates product conversion capability, sampling location differences, product adjacency relationships, and category structure constraints into a deterministic combination optimization framework in the scenario of partitioned sampling. Using the weighted conversion target value as a unified comparison benchmark, the sampling scheme is solved under multiple constraints, thereby providing a technical solution with a clear structure, logical closed loop, and repeatable execution for conversion rate optimization under partitioned sampling conditions.

[0047] In one embodiment of the present invention, the weighted transformation target value is recalculated within a new preset statistical window and compared with the previous preset statistical window. When the comparison result is lower than a preset difference threshold, the demand transfer coefficient, complementary gain coefficient, and candidate cluster return step 6 are updated, including: Step 81: Obtain the start and end times of the new preset statistical window, and verify that the start time is earlier than the end time to ensure that the new preset statistical window has a valid time boundary. The new preset statistical window does not overlap or only partially overlaps with the previous preset statistical window in time, and is used to reflect the latest round of store display and sales behavior. Within the new preset statistical window, the system collects the corresponding store display records, sales records, and display location information to form new fact details. Subsequently, the system filters the new fact details based on the obtained display mapping, retaining only records where the store identifier and display location identifier match the display mapping, and removing records that do not match, thereby ensuring that the recalculation process is based on the actually executed display plan.

[0048] Step 82: After generating new fact details, the system, based on these new fact details and using the same methodology as the aforementioned statistical phase, recalculates the sample conversion rate statistics and location observation data. Simultaneously, the system reuses existing store profile elements and display location types to reconstruct a micro-partition set for the new fact details, and recalculates the location weights accordingly. Within this micro-partition set, the system further recalculates the demand transfer coefficient, complementary gain coefficient, adjacency rule set, and candidate clusters based on the new statistical results. It also reuses product category matching constraints and, based on this, recalculates the basic conversion value and weighted conversion target value. Through this method, the system ensures that the product relationship coefficients and conversion target values ​​reflect the latest sample and sales data while maintaining consistency in the partitioning and constraint structure.

[0049] Step 83: After completing the recalculation of the weighted conversion target value, the system compares the recalculated weighted conversion target value with the weighted conversion target value corresponding to the previous preset statistical window within the same store identifier range. The preset difference threshold is used to limit the allowable range of change between the two weighted conversion target values. When the comparison result is lower than the preset difference threshold, the system determines that the current sampling scheme needs adjustment within the corresponding micro-partition, and limits the update range to the corresponding micro-partition identifier. Within this range, the system updates the demand transfer coefficient, complementary gain coefficient, and candidate clusters, and then returns to execute the step of establishing a deterministic combinatorial optimization model based on the weighted conversion target value and solving for the sampling mapping. When the comparison result is not lower than the preset difference threshold, the system determines that the current sampling scheme remains stable, maintains the sampling mapping unchanged, and does not trigger a re-solution.

[0050] Through the above implementation methods, the present invention introduces a recalculation and comparison mechanism based on statistical windows in the process of optimizing the regional sampling conversion rate. This enables the sampling scheme to be selectively updated based on the latest sampling and sales data while maintaining the existing structure and constraints. In this way, the continuity, stability and controllability of the regional sampling conversion rate optimization process are maintained in a dynamic business environment.

[0051] This invention provides a method for optimizing the conversion rate of partitioned sampling, comprising the following steps: Step 91: Determine the preset statistical window, collect the store's sample records, sales records and display location information within the preset statistical window to obtain fact details, and calculate the sample conversion rate statistics and location observation data from the fact details; Step 92: Construct a set of micro-zones based on store profile elements and display location types, and calculate location weights based on location observation data; Step 93: Within the micro-partition set, determine the demand transfer coefficient, complementary gain coefficient, and adjacency rule set based on the sample conversion rate statistics. Step 94: Generate candidate clusters based on the complementary gain coefficient and complementary gain threshold, the demand transfer coefficient and the upper limit threshold of demand transfer, and set category matching constraints; Step 95: Determine the basic conversion value based on the sample conversion rate statistics, and determine the weighted conversion target value based on the location weight, basic conversion value, complementary gain coefficient and demand transfer coefficient; Step 96: Construct a deterministic combinatorial optimization model based on the weighted conversion target value, and solve for the sample mapping under the constraints of unique location, non-repeating products, complete candidate groups, adjacency rule set and category ratio. Step 97: Recalculate the weighted transformation target value in the new preset statistical window and compare it with the previous preset statistical window. When the comparison result is lower than the set threshold, update the demand transfer coefficient, complementary gain coefficient and candidate group and return to step 96.

[0052] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0053] The embodiments of the present invention have been described above, but the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of the present embodiments, all of which are within the protection scope of the present embodiments.

Claims

1. A conversion rate optimization system for partitioned sampling, characterized in that, include: Data collection and statistics module 1 is used to determine the preset statistics window, collect the store's sample records, sales records and display location information within the preset statistics window to obtain fact details, and calculate the sample conversion rate statistics and location observation data from the fact details; Micro-zone weight module 2 is used to construct a set of micro-zones based on store profile elements and display location types, and calculate the location weights based on location observation data. The location weights are used to reflect the overall visibility and accessibility of the sample location. The commodity relationship calculation module 3 is used to determine the demand transfer coefficient, complementary gain coefficient and adjacency rule set based on the sample conversion rate statistics within the micro-partition set; Cluster constraint generation module 4 is used to generate candidate clusters based on complementary gain coefficients and complementary gain thresholds, demand transfer coefficients and demand transfer upper limit thresholds, and to set category allocation constraints, including: Step 41: Traverse the combination of the first product identifier and the second product identifier within the same micro-partition identifier. When the complementary gain coefficient is not less than the preset complementary gain threshold and the demand transfer coefficient is not greater than the preset demand transfer upper limit threshold, determine the first product identifier and the second product identifier as a product pair in the pairable set, and remove product pairs that lack complementary gain coefficient or demand transfer coefficient. Step 42: Construct a product association graph based on the pairable set, where product identifiers are used as nodes and product pairs in the pairable set are used as edges; perform connectivity analysis on the product association graph to obtain initial clusters, and obtain the preset lower limit and preset upper limit of cluster size; remove the initial clusters according to the preset lower limit of cluster size and decompose them according to the preset upper limit of cluster size to obtain candidate clusters; Step 43: Obtain the product category identifier corresponding to each product identifier in the candidate group, obtain the preset category ratio lower limit and preset category ratio upper limit, count the number of products for each product category identifier in the same micro-partition identifier, and limit the number of products to the range specified by the preset category ratio lower limit and preset category ratio upper limit to form a category ratio constraint, and archive the category ratio constraint according to the micro-partition identifier. The conversion target calculation module 5 is used to determine the basic conversion value based on the sample conversion rate statistics, and to determine the weighted conversion target value based on the location weight, basic conversion value, complementary gain coefficient and demand transfer coefficient. The combinatorial optimization solution module 6 is used to construct a deterministic combinatorial optimization model based on the weighted transformation target value, and solves the sampling mapping under the constraints of unique position, non-repeating products, complete candidate groups, adjacency rule set and category ratio. The iterative update control module 7 is used to recalculate the weighted transformation target value in the new preset statistical window and compare it with the previous preset statistical window. When the comparison result is lower than the preset difference threshold, the demand transfer coefficient, complementary gain coefficient and candidate group return to the combination optimization solution module are updated.

2. The conversion rate optimization system for partitioned sampling according to claim 1, characterized in that, Define a preset statistical window, collect the store's sample display records, sales records, and display location information within the preset statistical window to obtain factual details, and calculate sample conversion rate statistics and location observation data from the factual details, including: Step 11: Obtain the start time and end time of the preset statistical window, verify that the start time is earlier than the end time, and use the preset statistical window as the boundary for collecting and summarizing sample records, sales records, display location information and location observation data. Sampling records that cross the preset statistical window are truncated according to the boundary, and sales records that do not fall into the preset statistical window are considered invalid data. Step 12: Collect sample records, sales records and display location information in the preset statistics window, and use store identifier, sample location identifier, product identifier and timestamp as unified association fields. Based on the unified association fields, associate the sample records, sales records and display location information to obtain fact details. Step 13: Collect the fact details by statistical day. For each combination of store logo, display location logo and product logo, count whether the product was displayed on the day and accumulate them to get the number of display days. Count the sales volume on the day and accumulate it to get the sales volume. Use the ratio of sales volume to display days as the display conversion rate and generate display conversion rate statistics. Step 14: In the preset statistics window, summarize the number of exposures, dwell time and movement of each sampling location identifier to form location observation data by counting. Verify that the sampling location identifier in the sampling conversion rate statistics exists in the location observation data and that the number of exposures, dwell time and movement are not less than zero. Encapsulate and output the data that passes the verification.

3. The conversion rate optimization system for partitioned sampling according to claim 1, characterized in that, A set of micro-zones is constructed based on store profile elements and display location types. Location weights are calculated based on location observation data, including: Step 21: Obtain the store profile elements corresponding to each store identifier. The store profile elements include region, business district type, city level and store type. Perform value range verification on the store profile elements. Store identifiers whose values ​​are not in the preset enumeration set are determined to be invalid data of store profile elements. Step 22: Obtain the display location information corresponding to each display location under each store identifier, and map the display location information to the display location type. The display location type includes endcap, prime shelf, traffic hotspot, and cold spot. Display location identifiers that cannot be mapped to the display location type are determined to be invalid data of the display location type. Step 23: Remove the store identifier records corresponding to invalid data of store profile elements and the display location identifier records corresponding to invalid data of display location type. For each remaining store identifier and display location identifier, take the store profile element corresponding to the store identifier and the display location type corresponding to the display location identifier and combine them in an orderly manner to obtain micro-zone identifiers. Summarize the different micro-zone identifiers to obtain a micro-zone set, and generate a correspondence table between store identifiers, display location identifiers and micro-zone identifiers. Step 24: Based on the location observation data, summarize the number of exposures, dwell time and movement pass counts for each sampling location under each store identifier. Normalize the maximum value of the number of exposures, dwell time and movement pass counts within the same store identifier, and weight the normalization results to obtain the location weight. Then bind the location weight with the store identifier, sampling location identifier and micro-zone identifier in the corresponding relationship table and output it.

4. The conversion rate optimization system for partitioned sampling according to claim 1, characterized in that, Within the micro-partition set, the demand transfer coefficient, complementary gain coefficient, and adjacency rule set are determined based on sample conversion rate statistics, including: Step 31: Based on the correspondence table between store identifier, sample location identifier and micro-zone identifier, map the sample conversion rate statistics to the micro-zone identifier, and construct a statistical day sample set with micro-zone identifier, store identifier and statistical day, and remove the sample conversion rate statistics that cannot be mapped to the micro-zone identifier. Step 32: Within the same micro-partition identifier, for each first product identifier and second product identifier, determine the sample set for simultaneous sampling based on whether the two products appear simultaneously in the statistical day sample set; determine the sample set for non-simultaneous sampling based on whether the second product identifier appears in the statistical day sample set and the first product identifier does not appear. Step 33: Within the same micro-region identifier, for each first product identifier and second product identifier, summarize the sample conversion rate of the second product identifier in the simultaneous sample set and the non-simultaneous sample set, respectively, to obtain the average sample conversion rate of the second product identifier under the simultaneous sample set and the average sample conversion rate of the second product identifier under the non-simultaneous sample set. The ratio of the former to the latter is used as the retention ratio, and the result of subtracting the retention ratio is determined as the demand transfer coefficient. Step 34: Within the same micro-partition identifier, for each first product identifier and second product identifier, based on the statistical day sample set, count the number of statistical days in which the two products appear simultaneously, the number of statistical days in which the first product identifier appears, and the number of statistical days in which the second product identifier appears, and normalize them to obtain the co-occurrence ratio and the co-occurrence ratio of the two products; subtract one from the product of the co-occurrence ratio and the co-occurrence ratio of the two products to determine the complementary gain coefficient; and when the complementary gain coefficient is not less than the preset complementary gain threshold and the demand transfer coefficient is not greater than the preset demand transfer upper limit threshold, write the corresponding first product identifier and second product identifier into the adjacency rule set.

5. The conversion rate optimization system for partitioned sampling according to claim 1, characterized in that, Step 43 further includes: Step 51: Obtain the product category identifier corresponding to each product identifier in the candidate group, and remove product identifiers that are missing product category identifiers from the candidate group; Step 52: Within the same micro-partition identifier, count the number of products for each product category identifier in the candidate group. For product category identifiers that exceed the preset category ratio limit, remove product identifiers in ascending order of the number of times the product identifier appears in the pairable set until the preset category ratio limit is met. Step 53: Eliminate candidate groups corresponding to product category identifiers that are below the preset category ratio lower limit, and archive candidate groups that meet the preset category ratio lower limit and preset category ratio upper limit according to micro-region identifiers.

6. The conversion rate optimization system for partitioned sampling according to claim 1, characterized in that, The basic conversion value is determined based on sample conversion rate statistics. The weighted conversion target value is then determined based on location weight, basic conversion value, complementary gain coefficient, and demand transfer coefficient, including: Step 61: Based on the correspondence table between store identifier, display location identifier and micro-zone identifier, map the display conversion rate statistics to the micro-zone identifier, and summarize the display conversion rate statistics with the micro-zone identifier and product identifier. Based on the summary results and in accordance with the extreme value removal rule that is consistent with the display conversion rate statistics, determine the basic conversion value. Step 62: Based on the correspondence table between store identifiers, display location identifiers, and micro-zone identifiers, bind the location weights to the store identifiers, display location identifiers, and micro-zone identifiers to form a location binding view; and within the same micro-zone identifier, combine the complementary gain coefficient and demand transfer coefficient of the first product identifier and the second product identifier to form a product effect view. Step 63: For each display location identifier under each store identifier, read the location weight and corresponding micro-zone identifier in the location binding view, read the basic conversion value of the product identifier under the micro-zone identifier, and use the product of the location weight and the basic conversion value as the basic contribution value; determine the adjacent display location identifier pairs according to the adjacency rule set, read the complementary gain coefficient and demand transfer coefficient in the product pair effect view for the first and second product identifiers on the adjacent display location identifier pairs, use the product of the complementary gain coefficient and the preset fusion coefficient as the positive term, use the product of the demand transfer coefficient and the preset fusion coefficient as the negative term, subtract the negative term from the positive term to obtain the adjacent effect contribution value, and summarize the basic contribution value and the adjacent effect contribution value to obtain the weighted conversion target value.

7. The conversion rate optimization system for partitioned sampling according to claim 1, characterized in that, A deterministic combinatorial optimization model is constructed based on the weighted conversion objective value. Under the constraints of unique location, non-repeating products, complete candidate clusters, adjacency rule set, and category ratio, the sample mapping is obtained, including: Step 71: Obtain all display location identifiers under the store identifier as the boundary, and determine the micro-zone identifier and location weight corresponding to each display location identifier according to the correspondence table of store identifier, display location identifier and micro-zone identifier. Obtain the basic conversion value, candidate group, adjacency rule set and category ratio constraint within the micro-zone identifier, and remove the display location identifier corresponding to the micro-zone identifier that is missing any data. Step 72: Using the weighted conversion target value as the sole optimization objective of the combinatorial optimization model, a deterministic combinatorial optimization model is established. The deterministic combinatorial optimization model is used to compare the weighted conversion target values ​​corresponding to different sampled mappings under the condition of satisfying the model constraint set, and select the sampled mapping with the largest weighted conversion target value as the model output. The model constraint set includes: unique position constraint, non-repeating product constraint, complete candidate group constraint, adjacency rule set constraint, and category matching constraint. Step 73: Output the initial solution that satisfies the model constraint set under the deterministic combinatorial optimization model, and select two sampling location identifiers within the same store identifier to exchange their corresponding product identifiers and then re-examine the model constraint set; when the weighted conversion target value after the exchange is not less than the weighted conversion target value before the exchange and the model constraint set is satisfied, perform the exchange, repeat the exchange until there is no exchange operation that satisfies the conditions, and obtain and output the sampling mapping.

8. The conversion rate optimization system for partitioned sampling according to claim 1, characterized in that, Within the new preset statistical window, recalculate the weighted transformation target value and compare it with the previous preset statistical window. When the comparison result is lower than the preset difference threshold, update the demand transfer coefficient, complementary gain coefficient, and candidate clusters, and return to step 6, including: Step 81: Obtain the start and end times of the new preset statistics window and verify that the start time is earlier than the end time; collect the sample records, sales records and display location information in the new preset statistics window to obtain new fact details, and filter the records that match the store identifier and the sample location identifier according to the sample mapping, and remove the records that do not match. Step 82: Based on the new fact details, recalculate the sample conversion rate statistics and location observation data according to the established criteria, and reuse the store profile elements and display location types to construct a micro-zone set and recalculate the location weight; within the micro-zone set, recalculate the demand transfer coefficient, complementary gain coefficient, adjacency rule set and candidate group, and reuse the category matching constraints to recalculate the basic conversion value and weighted conversion target value. Step 83: Within the same store identifier, compare the recalculated weighted conversion target value with the weighted conversion target value corresponding to the previous preset statistical window. When the comparison result is lower than the preset difference threshold, limit the update range to the corresponding micro-partition identifier and update the demand transfer coefficient, complementary gain coefficient and candidate group. Then return to execute the step of establishing a deterministic combination optimization model based on the weighted conversion target value and solving to obtain the sample mapping. When the comparison result is not lower than the preset difference threshold, keep the sample mapping unchanged.

9. A method for optimizing conversion rate in partitioned sampling, characterized in that, The conversion rate optimization system for partitioned sampling as described in any one of claims 1-8 includes: Step 91: Determine the preset statistical window, collect the store's sample records, sales records and display location information within the preset statistical window to obtain fact details, and calculate the sample conversion rate statistics and location observation data from the fact details; Step 92: Construct a set of micro-zones based on store profile elements and display location types, and calculate location weights based on location observation data; Step 93: Within the micro-partition set, determine the demand transfer coefficient, complementary gain coefficient, and adjacency rule set based on the sample conversion rate statistics. Step 94: Generate candidate clusters based on the complementary gain coefficient and complementary gain threshold, the demand transfer coefficient and the upper limit threshold of demand transfer, and set category matching constraints; Step 95: Determine the basic conversion value based on the sample conversion rate statistics, and determine the weighted conversion target value based on the location weight, basic conversion value, complementary gain coefficient and demand transfer coefficient; Step 96: Construct a deterministic combinatorial optimization model based on the weighted conversion target value, and solve for the sample mapping under the constraints of unique location, non-repeating products, complete candidate groups, adjacency rule set and category ratio. Step 97: Recalculate the weighted transformation target value in the new preset statistical window and compare it with the previous preset statistical window. When the comparison result is lower than the set threshold, update the demand transfer coefficient, complementary gain coefficient and candidate group and return to step 96.

Images