An e-commerce transaction data intelligent analysis system and method based on a large model

Abstract

The application discloses an e-commerce transaction data intelligent analysis system and method based on a large model, relates to the technical field of e-commerce data analysis, and classifies different specification combinations of the same commodity clicked by a user according to core attribute dimensions and value combinations thereof, forms a plurality of specification record groups, and calculates a user decision hesitation index based on switching, returning and staying of the specification record groups in a click sequence; then, a large model input data is constructed in combination with difference features between the specification record groups, user decision auxiliary information is output by a pre-training large model, and therefore, intelligent analysis on a user specification selection intention and a decision difficulty point is realized. The application optimizes defects of the prior art that only relies on overall staying time of a page to roughly judge an intention, and provides a large model with a clear analysis benchmark by positioning specific specification objects of a user to generate purchase dilemmas.

Description

Technical Field

[0001] This invention relates to the field of e-commerce data analysis technology, specifically to an intelligent analysis system and method for e-commerce transaction data based on a large model. Background Technology

[0002] As the number of product specifications on e-commerce platforms increases, the same product often comes with multiple specifications such as color, capacity, size, version, and bundled packages. Users need to choose from multiple complete combinations of product specifications before purchasing. Existing e-commerce data analysis methods typically determine user purchase intent based on page dwell time, total page clicks, browsing history, or purchase history. While these methods can reflect whether a user is paying attention to a particular product page, they struggle to pinpoint which specific combinations of product specifications the user is currently considering. In actual shopping, users may click on a particular combination of product specifications, then switch to other combinations, and then return to the previously left combination, or repeatedly switch between multiple combinations. Existing methods lack the ability to identify this type of click process, resulting in a lack of clear comparison objects for subsequent large-scale model analysis. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent analysis system and method for e-commerce transaction data based on a large model, so as to solve the problems raised in the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an intelligent analysis method for e-commerce transaction data based on a large model, the intelligent analysis method comprising the following steps: Step S1: Obtain the name and specification combination of the products browsed by the user. When the user clicks on any specification combination, generate a click record containing the product name and specification combination and attach a timestamp. Step S2: Capture the click records of the same user according to the preset sliding time window; identify the same product as the target product according to the product name, and group the click records of the same target product into the target product click group; Step S3: Extract the attribute dimensions of each specification combination in the target product click group, count the frequency of each attribute dimension, and determine the core attribute dimension based on the frequency; group all specification combinations with the same value combination under the core attribute dimension and their corresponding click records into the same specification record group, and form a candidate specification comparison set by multiple specification record groups. Step S4: Sort all click records in the candidate specification comparison set according to timestamp to form a click sequence, and calculate the user decision hesitation index for each specification record group based on the click sequence; Step S5: Obtain the difference features between each specification record group, combine the specification combination corresponding to the specification record group, the difference features, and the user decision hesitation index to construct large model input data; input the large model input data into the pre-trained large model to output user decision assistance information.

[0005] Furthermore, the specific steps of step S1 are as follows: The system retrieves the product names and all corresponding specification combinations of the products viewed by the user on the e-commerce platform. These specification combinations are a set of required product specifications that can be independently ordered and purchased, comprising multiple attribute dimensions and their corresponding values. The attribute dimensions are specification categories available for user selection on the product page, used to help users determine the specific product to purchase. Typically, a product requires confirmation of multiple attribute dimensions, and the specific options under each attribute dimension are the attribute values. Different merchants or products may have different ways of classifying attribute dimensions, including but not limited to color, volume, appearance, and quantity. When a user clicks on any specification combination that can be ordered independently, a click record is generated containing the product name, attribute dimension, and value corresponding to the specification combination. A timestamp is generated based on the real-time time when the user clicks the specified combination, and then appended to the click record.

[0006] The specification combination is a complete purchase object composed of multiple attribute dimensions and their values ​​that must be confirmed to complete the order on the product page; by converting each user's click on the complete specification combination into a timestamped click record, the subsequent analysis object is refined from page-level browsing behavior to specification combination-level selection behavior.

[0007] Furthermore, the specific steps of step S2 are as follows: Using the timestamp of the latest generated click record as a dynamic reference point, the preset sliding time window is constructed by tracing back a preset duration, and all click records of the same user within the window are captured; Extract the product name from the click record, perform noise reduction on the product name, and extract feature keywords. The noise reduction process involves removing punctuation marks, special characters, and preset modifiers from the product name. Products with the same characteristic keywords are identified as the same type of product, and the same type of product is identified as the target product; Extract click records for the same target product on the same page and click records across pages and stores; All click records are assigned to the target product click group.

[0008] The preset sliding time window is used to limit the scope of analysis and avoid mixing user clicks that occurred earlier and have lost their continuous decision-making relevance into the current analysis process. When identifying the same product based on the product name, non-core text such as promotional words, punctuation marks, special characters, and delivery instructions can be removed first, and then keywords that can represent the product entity, such as brand, model, capacity, and category, can be retained, thereby reducing the interference of inconsistent product title descriptions across pages and stores on the target product aggregation results.

[0009] Furthermore, the specific steps of step S3 are as follows: Iterate through all click records in the target product click group, read the specification combination contained in each click record, and extract the attribute dimension from the specification combination; Count the frequency of each attribute dimension, and take the attribute dimension with the highest frequency as the core attribute dimension of the target product; Extract the values ​​of each specification combination under the core attribute dimension as the core feature; Group the specification combinations that have identical core features and their corresponding click records into the same specification record group; All specification records of the target product are grouped together to form a candidate specification comparison set.

[0010] Furthermore, the specific steps of step S4 are as follows: Identify all round-trip events in the click sequence and count the number of round trips for each specification record group; a round-trip event refers to the behavior of a user returning to any specific specification combination within a specification record group after switching away from it, including round trips across specification record groups, round trips between different combinations within the same specification record group, and round trips between the same combination within the same specification record group; wherein, the first access to the specification record group does not constitute a round-trip event. The cumulative round-trip dwell time for each specification record group in all round-trip events is calculated. The dwell time calculation interval for a single round-trip event is: from the time the user switches back to the page corresponding to the specification record group to the time the user switches back to leave the page again. The cumulative round-trip dwell time does not include the dwell time of the first visit. Multiply the number of round trips for each specification record group by the corresponding cumulative round trip stay time to obtain the hesitation score for that specification record group; The total score is obtained by summing the hesitation scores of all the specification record groups, and the user decision hesitation index of that specification record group is obtained by dividing the hesitation score of each specification record group by the total score.

[0011] The user decision hesitation index is not simply a count of clicks, but is calculated by combining the user's back-and-forth behavior after switching away from a certain specification record group and returning to that specification record group, as well as the length of time spent after returning. Therefore, the index can distinguish between three types of behavior: accidental clicks, first-time viewing, and repeated comparisons, so that specification record groups that are repeatedly returned to and viewed for a longer period of time receive a higher decision hesitation index.

[0012] Furthermore, the statistical calculation of the cumulative round-trip dwell time for each of the specified record groups across all the round-trip events specifically includes: The cumulative round-trip dwell time is defined as the cumulative round-trip effective active time; the effective active time is the time period during which the user continuously generates interactive behavior on the specification record group page, and the interactive behavior includes at least one of clicking, scrolling, swiping and input; During the process of calculating the effective active duration, if the continuous period of no interaction exceeds the preset silent interruption duration, the timer is paused until the next interaction occurs and the timer is resumed.

[0013] When there are many product specifications and a large amount of page information, the duration of the silent interruption can be increased accordingly to avoid users being mistakenly judged as having stopped working while reading price, inventory, discount, or review information. When there are few product specifications and a small amount of page information, the duration of the silent interruption can be decreased accordingly to reduce the interference of users leaving the page, pausing operations, or non-decision-making stops on the effective activity duration statistics.

[0014] By further limiting the cumulative round-trip dwell time to the cumulative effective active time of round trips, false dwell time caused by users leaving the page, devices being idle, or the page remaining in the background can be eliminated, making the cumulative round-trip dwell time more accurately reflect the user's actual comparison behavior on the specification record group page.

[0015] Furthermore, the specific steps of step S5 are as follows: Obtain real-time business data corresponding to the specification combination, including price, inventory, discount amount, and number of positive reviews; Calculate the price difference, inventory difference, discount difference, and positive review quantity difference for each of the aforementioned specification record groups, and extract the core difference features; The specification combination, core difference features, and user decision hesitation index are structurally concatenated according to a preset prompt word template to generate the input data for the large model. The input data of the large model is input into the pre-trained large model for inference analysis to obtain the user decision-making assistance information output by the pre-trained large model. Based on the user decision-making assistance information, the rendering update of the product details page currently viewed by the user is performed. The rendering update includes adjusting the display order of specification record groups and generating at least one of the following for each specification record group based on the core difference features: price comparison label, inventory status prompt, discount indicator, and positive review recommendation badge.

[0016] The rendering update writes the output of the large model back to the product details page that the user is currently browsing, so that the user can see the difference prompts related to the object they are considering within the original specification selection page, reducing the operation cost of repeatedly switching between multiple specification combinations and manually comparing prices and inventory information.

[0017] Furthermore, the intelligent analysis system includes a data acquisition module, a grouping module, a hesitation analysis module, and a decision support output module; The data acquisition module is used to collect the specification click behavior generated by users when browsing products, and record the corresponding product name, specification combination and click time; The grouping module is used to organize the click records of the same user according to a preset sliding time window, determine the target product according to the product name, divide the click records belonging to the same target product into the target product click group, and divide the specification combinations with consistent values ​​into the corresponding specification record group according to the core attribute dimension that exists in all specification combinations in the target product. The hesitation analysis module is used to calculate the corresponding user decision hesitation index based on the click switching and dwell time between each specification record group. The decision support output module is used to generate large model input data based on specification record groups and user decision hesitation index, and output user decision support information through pre-trained large model.

[0018] Furthermore, the hesitation analysis module includes a round-trip behavior identification unit and an exponential normalization calculation unit; The round-trip behavior identification unit is used to sort the click records in the same candidate specification comparison set according to the click time, identify the user's return viewing behavior between different specification record groups or within the same specification record group, and count the number of round trips and the round-trip dwell time for each specification record group; the index normalization calculation unit is used to calculate the hesitation score of each specification record group according to the number of round trips and the round-trip dwell time, and normalize the hesitation score of each specification record group to obtain the corresponding user decision hesitation index.

[0019] Furthermore, the decision support output module includes an input data construction unit and a page display control unit; The input data construction unit is used to generate large model input data by taking the specification combination, user decision hesitation index and core difference features corresponding to each specification record group according to the preset prompt word template. The page display control unit is used to receive user decision-making assistance information output by the pre-trained large model, adjust the display order of specification record groups in the product details page according to the user decision-making assistance information, and generate at least one of the following for each specification record group based on the core difference features: price comparison label, inventory status prompt, discount indicator, and positive review recommendation badge.

[0020] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention extracts core attribute dimensions of product specification combinations for grouping and accurately identifies user switching behavior between different specification groups based on time series, constructing a quantitative user decision-making hesitation index. This optimizes upon existing technologies that rely solely on overall page dwell time to roughly determine intent, pinpointing the specific specification object causing user purchase indecision and providing a clear analytical benchmark for large-scale models.

[0021] 2. By strictly limiting the round-trip dwell time of specification groups to the effective active time of continuous page interaction, this invention effectively filters out the idle or inactive time of users browsing pages, ensures that the underlying data truly reflects the user's effective attention process, eliminates the interference of dirty data on the algorithm, and improves the accuracy of hesitation index calculation.

[0022] 3. This invention constructs a large model input data by combining the user decision hesitation index with core differentiating features such as price, inventory, discounts, and number of positive reviews, and uses the output results to update the product details page. This can generate decision-making assistance information directly related to the user's current hesitation, reduce the number of times the user repeatedly switches between specification combinations for manual comparison, and improve the efficiency of product specification selection. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the structure of an intelligent analysis method for e-commerce transaction data based on a large model according to the present invention; Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Example: Figure 1As shown, this invention provides a technical solution for intelligent analysis of e-commerce transaction data based on a large model, comprising the following steps: The system is pre-deployed on the e-commerce platform's backend, with a preset sliding time window set to 15 minutes prior to the latest click time. Users repeatedly browse and compare products across multiple stores, and the system obtains the product names and specification combinations viewed by the user. When a user clicks on any specification combination that can be ordered independently, the system generates a click record with an additional timestamp and simultaneously collects page interaction behavior logs including clicks, scrolling, swiping, and input actions.

[0026] The system retrieved three original product names: "Official Direct Sales X Brand Smartphone," "X Brand Smartphone with Free Shipping via SF Express," and "X Brand Smartphone Performance Flagship." Using the most recently generated click record timestamp as a dynamic baseline, the system traced back through the product names within the window, performing noise reduction to remove punctuation, special characters, and modifiers, extracting the common keyword "X Brand Smartphone." The system categorized products with the same keyword as the same target product and grouped all three cross-store click records into the same target product click group.

[0027] Subsequently, the system iterates through all specification combinations within the target product click group, examining their attribute dimensions. The target product click group contains three specific specification click records: the first record specifies white color, 512GB memory, and no screen breakage insurance; the second record specifies white color, 512GB memory, and includes one-year screen breakage insurance; and the third record specifies black color and 256GB memory. The system counts the frequency of each attribute dimension, finding that color appears three times, memory appears three times, and screen breakage insurance appears twice. The system extracts the most frequent attributes, color and memory, as core attribute dimensions and extracts the specific values ​​of each specification combination under these core attribute dimensions as core features. It is determined that the first two records have identical color and memory values ​​under the core attribute dimensions. The system merges these two records into the first specification record group and separately assigns the third record to the second specification record group, thus eliminating interference from additional services on the alignment of the main specifications and forming a candidate specification comparison set.

[0028] Furthermore, the system sorts the click records by timestamp to form a click sequence, identifying round-trip events excluding the first visit within the sequence. The user's click access sequence is as follows: first visit to the first specification record group, first visit to the second specification record group, cross-group return to the first specification record group, switching within the same group to a record with screen breakage risk in the first specification record group, and returning to the second specification record group again. For the first specification record group, there were two round-trip events: one cross-group return and one return within the same group. During the first return, the user stayed for a total of 25 seconds. The system identified through the interaction behavior log that the user remained silent for 8 seconds without any operation, exceeding the preset 5-second silent interruption threshold. After deducting this, the effective active time for a single round-trip was 17 seconds. The effective active time for the second return was 13 seconds, and the cumulative effective active time for round-trips in the first specification record group was 30 seconds. The hesitation score for the first specification record group is the number of round-trips multiplied by the cumulative effective active time for round-trips, which is 60. For the second specification record group, a round-trip event occurred, with the user continuously scrolling through comments without silent interruption, resulting in an effective active duration of 20 seconds and a hesitation score of 20. The system summed the scores of all groups to obtain a total score of 80. The normalized calculation yielded a hesitation index of 0.75 for the first specification record group and 0.25 for the second specification record group.

[0029] Finally, the system acquires real-time business data such as price, inventory, discount amount, and number of positive reviews for each specification record group, calculates the differences, and extracts core difference features. The system backend comparison shows that the first specification record group is 800 yuan more expensive than the second specification record group, but has 3000 more positive reviews, and only 10 units remain in stock. The system generates large-scale model input data based on the above specification combinations, core difference features, and user decision-making hesitation index according to a preset prompt word template. The input data describes a user's high degree of indecision between the first specification record group (white 512G) and the second specification record group (black 256G). The first specification record group has a hesitation index of 0.75, is 800 yuan more expensive, has 3000 more positive reviews, and only 10 units remain in stock. The system requests the output of a sales-driving front-end rendering command. The system then inputs this input data into a pre-trained large-scale model for inference.

[0030] The large model determines that users prefer the first specification group but hesitate due to price, and outputs auxiliary information and front-end rendering instructions. Based on these instructions, the system renders and updates the current product details page, adjusting the display order to place the first specification group first for prominence, and generating comparative text such as price, inventory, and number of positive reviews to alleviate user decision-making hesitation.

[0031] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent analysis of e-commerce transaction data based on a large model, characterized in that: The intelligent analysis method includes the following steps: Step S1: Obtain the name and specification combination of the products browsed by the user. When the user clicks on any specification combination, generate a click record containing the product name and specification combination and attach a timestamp. Step S2: Capture the click records of the same user according to the preset sliding time window; identify the same product as the target product according to the product name, and group the click records of the same target product into the target product click group; Step S3: Extract the attribute dimensions of each specification combination in the target product click group, count the frequency of each attribute dimension, and determine the core attribute dimension based on the frequency; group all specification combinations with the same value combination under the core attribute dimension and their corresponding click records into the same specification record group, and form a candidate specification comparison set by multiple specification record groups. Step S4: Sort all click records in the candidate specification comparison set according to timestamp to form a click sequence, and calculate the user decision hesitation index for each specification record group based on the click sequence; Step S5: Obtain the difference features between each specification record group, combine the specification combination corresponding to the specification record group, the difference features, and the user decision hesitation index to construct large model input data; input the large model input data into the pre-trained large model to output user decision assistance information.

2. The intelligent analysis method for e-commerce transaction data based on a large model according to claim 1, characterized in that: The specific steps of step S1 are as follows: Obtain the product name of the product that the user browses on the e-commerce platform, as well as all the corresponding specification combinations of the product; the specification combination is a set of required specifications of the product that can be ordered independently, consisting of multiple attribute dimensions and their corresponding values; the attribute dimensions are the specification categories that the user can select on the product page, and the specific options under the attribute dimensions are the values ​​of the attributes. When a user clicks on any specification combination that can be ordered independently, a click record is generated containing the product name, attribute dimension, and value corresponding to the specification combination. A timestamp is generated based on the real-time time when the user clicks the specified combination, and then appended to the click record.

3. The intelligent analysis method for e-commerce transaction data based on a large model according to claim 1, characterized in that: The specific steps of step S2 are as follows: Using the timestamp of the latest generated click record as a dynamic reference point, the preset sliding time window is constructed by tracing back a preset duration, and all click records of the same user within the window are captured; Extract the product name from the click record, perform noise reduction on the product name, and extract feature keywords. The noise reduction process involves removing punctuation marks, special characters, and preset modifiers from the product name. Products with the same characteristic keywords are identified as the same type of product, and the same type of product is identified as the target product; Extract click records for the same target product on the same page and click records across pages and stores; All click records are assigned to the target product click group.

4. The intelligent analysis method for e-commerce transaction data based on a large model according to claim 1, characterized in that: The specific steps of step S3 are as follows: Iterate through all click records in the target product click group, read the specification combination contained in each click record, and extract the attribute dimension from the specification combination; Count the frequency of each attribute dimension, and take the attribute dimension with the highest frequency as the core attribute dimension of the target product; Extract the values ​​of each specification combination under the core attribute dimension as the core feature; Group the specification combinations that have identical core features and their corresponding click records into the same specification record group; All specification records of the target product are grouped together to form a candidate specification comparison set.

5. The intelligent analysis method for e-commerce transaction data based on a large model according to claim 1, characterized in that: The specific steps of step S4 are as follows: Identify all round-trip events in the click sequence and count the number of round trips for each specification record group; a round-trip event refers to the behavior of a user returning to any specific specification combination within a specification record group after switching away from it, including round trips across specification record groups, round trips between different combinations within the same specification record group, and round trips between the same combination within the same specification record group; wherein, the first access to the specification record group does not constitute a round-trip event. The cumulative round-trip dwell time for each specification record group in all round-trip events is calculated. The dwell time for a single round-trip event is calculated from the time the user switches back to the page corresponding to the specification record group until the time the user switches back to the page again. The cumulative round-trip dwell time does not include the dwell time of the first visit. Multiply the number of round trips for each specification record group by the corresponding cumulative round trip stay time to obtain the hesitation score for that specification record group; The total score is obtained by summing the hesitation scores of all the specification record groups, and the user decision hesitation index of that specification record group is obtained by dividing the hesitation score of each specification record group by the total score.

6. The intelligent analysis method for e-commerce transaction data based on a large model according to claim 5, characterized in that: The statistical calculation of the cumulative round-trip dwell time for each of the specified record groups across all round-trip events specifically includes: The cumulative round-trip dwell time is defined as the cumulative round-trip effective active time; the effective active time is the time period during which the user continuously generates interactive behavior on the specification record group page, and the interactive behavior includes at least one of clicking, scrolling, swiping and input; During the process of calculating the effective active duration, if the continuous period of no interaction exceeds the preset silent interruption duration, the timer is paused until the next interaction occurs and the timer is resumed.

7. The intelligent analysis method for e-commerce transaction data based on a large model according to claim 1, characterized in that: The specific steps of step S5 are as follows: Obtain real-time business data corresponding to the specification combination, including price, inventory, discount amount, and number of positive reviews; Calculate the price difference, inventory difference, discount difference, and positive review quantity difference for each of the aforementioned specification record groups, and extract the core difference features; The specification combination, core difference features, and user decision hesitation index are structurally concatenated according to a preset prompt word template to generate the input data for the large model. The input data of the large model is input into the pre-trained large model for inference analysis to obtain the user decision-making assistance information output by the pre-trained large model. Based on the user decision-making assistance information, the rendering update of the product details page currently viewed by the user is performed. The rendering update includes adjusting the display order of specification record groups and generating at least one of the following for each specification record group based on the core difference features: price comparison label, inventory status prompt, discount indicator, and positive review recommendation badge.

8. A large-scale model-based intelligent analysis system for e-commerce transaction data, applied to the large-scale model-based intelligent analysis method for e-commerce transaction data as described in any one of claims 1-7, characterized in that: The intelligent analysis system includes a data acquisition module, a grouping module, a hesitation analysis module, and a decision support output module; The data acquisition module is used to collect the specification click behavior generated by users when browsing products, and record the corresponding product name, specification combination and click time; The grouping module is used to organize the click records of the same user according to a preset sliding time window, determine the target product according to the product name, divide the click records belonging to the same target product into the target product click group, and divide the specification combinations with consistent values ​​into the corresponding specification record group according to the core attribute dimension that exists in all specification combinations in the target product. The hesitation analysis module is used to calculate the corresponding user decision hesitation index based on the click switching and dwell time between each specification record group. The decision support output module is used to generate large model input data based on specification record groups and user decision hesitation index, and output user decision support information through pre-trained large model.

9. The intelligent analysis system for e-commerce transaction data based on a large model according to claim 8, characterized in that: The hesitation analysis module includes a round-trip behavior identification unit and an exponential normalization calculation unit; The round-trip behavior recognition unit is used to sort the click records in the same candidate specification comparison set according to the click time, identify the return viewing behavior generated by the user between different specification record groups or within the same specification record group, and count the number of round trips and the round-trip dwell time corresponding to each specification record group. The index normalization calculation unit is used to calculate the hesitation score of each specification record group based on the number of round trips and the round trip stay duration, and to normalize the hesitation score of each specification record group to obtain the corresponding user decision hesitation index.

10. The intelligent analysis system for e-commerce transaction data based on a large model according to claim 8, characterized in that: The decision support output module includes an input data construction unit and a page display control unit; The input data construction unit is used to generate large model input data by taking the specification combination, user decision hesitation index and core difference features corresponding to each specification record group according to the preset prompt word template. The page display control unit is used to receive user decision-making assistance information output by the pre-trained large model, adjust the display order of specification record groups in the product details page according to the user decision-making assistance information, and generate at least one of the following for each specification record group based on the core difference features: price comparison label, inventory status prompt, discount indicator, and positive review recommendation badge.

Images