An information determination method and apparatus, electronic device, and storage medium

Abstract

The application discloses an information determination method and device, electronic equipment and a storage medium. The method comprises: obtaining historical order data of a target user; determining semantic features corresponding to the consumption behavior of the target user based on the historical order data; determining commodity recommendation information and a recommendation time for the target user based on the semantic features and global statistical features, wherein the global statistical features are used to indicate consumption time rules determined based on big data; and outputting the commodity recommendation information when the recommendation time is reached. Since the global statistical features are general consumption time rules determined based on big data, the recommendation is made by comprehensively combining the semantic features of the target user and the global statistical features, which can not only be personalized adapted to the target user, but also provide standardized data reference, reduce the recommendation prediction deviation caused by the abnormal consumption behavior of the target user, and further improve the accuracy and reliability of the recommendation result.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to an information determination method, apparatus, electronic device, and storage medium. Background Technology

[0002] Customer Relationship Management (CRM) refers to the use of information technology and internet technology by enterprises to coordinate interactions with customers in sales, marketing and service in order to improve their core competitiveness. Its ultimate goal is to attract new customers, retain old customers and turn existing customers into loyal customers.

[0003] For example, in the customer relationship management (CRM) context of the retail e-commerce industry, predicting user repeat purchase behavior and related purchase behavior is a core element in achieving product recommendations and unlocking the full lifecycle value of users. Correspondingly, existing recommendation methods targeting this type of behavior mainly rely on manually preset uniform time and behavior thresholds as recommendation trigger conditions, thus failing to meet the needs for refined and personalized recommendations. Summary of the Invention

[0004] This application provides an information determination method, apparatus, electronic device, and storage medium to improve the accuracy and personalization of product recommendations.

[0005] In a first aspect, embodiments of this application provide an information determination method, including: Obtain historical order data from the target user; Based on the historical order data, determine the semantic features corresponding to the target user's consumption behavior; Based on the semantic features and global statistical features, product recommendation information and recommendation timing for the target user are determined, and the global statistical features are used to indicate the consumption time pattern determined based on big data. When the recommended timing is reached, the product recommendation information is output.

[0006] Optionally, determining the product recommendation information and recommendation timing for the target user based on the semantic features and global statistical features includes: The semantic features and the global statistical features are filled into the prompt word template to obtain the prompt words to be processed; The large language model is used to analyze the prompt words to be processed to obtain the product recommendation information and the recommendation timing. The large language model is used to correct the global statistical features based on the semantic features.

[0007] Optionally, the prompt words to be processed include the semantic features, the global statistical features, and preset logical rules; The prompt word to be processed is used to instruct the large language model to generate a correction factor based on the semantic features, to correct the global statistical features based on the correction factor to obtain the corrected features, and to perform conflict detection processing on the corrected features based on the preset logical rules to obtain the recommendation timing.

[0008] Optionally, the global statistical features are obtained through the following steps: A global statistical feature model is constructed, which includes a histogram model and a concept graph model. The histogram model is used to describe the time when different candidate users purchase non-durable goods of the same specification, and the concept graph model is used to describe the time when different candidate users purchase durable goods of different specifications. The global statistical features are extracted based on the histogram model or the concept graph model.

[0009] Optionally, the histogram model is constructed through the following steps: Based on big data statistics, the time difference between two consecutive purchases of the same specification of non-durable products by each of the different candidate users; The histogram model is constructed based on the time difference corresponding to each candidate user.

[0010] Optionally, the concept graph model is constructed through the following steps: Based on big data statistics, the time each candidate user spent purchasing durable products of different specifications was analyzed. A directed graph is constructed as the conceptual graph model. The directed graph includes multiple nodes and edges between the multiple nodes. The multiple nodes are used to represent durable products of different specifications. The weights of the edges between the multiple nodes are used to represent the average conversion time and conversion probability of each candidate user purchasing durable products of different specifications.

[0011] Optionally, determining the semantic features corresponding to the target user's consumption behavior based on the historical order data includes: Natural language processing is performed on the historical order data to obtain descriptive information about the consumption behavior corresponding to the historical order data; The semantic features are generated by processing the descriptive information using a large language model.

[0012] Secondly, embodiments of this application provide an information determining device, comprising: The data acquisition module is used to acquire historical order data of the target user; The semantic feature determination module is used to determine the semantic features corresponding to the consumption behavior of the target user based on the historical order data. The information determination module is used to determine product recommendation information and recommendation timing for the target user based on the semantic features and global statistical features, wherein the global statistical features are used to indicate the consumption time pattern determined based on big data; The information output module is used to output the product recommendation information when the recommendation time is reached.

[0013] Thirdly, embodiments of this application provide an electronic device, the device including: a processor, a memory, and a system bus; The processor and the memory are connected via the system bus; The memory is used to store a program, the program including instructions that, when executed by the processor, cause the processor to perform any of the implementation steps of the information determination method described above.

[0014] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements any of the implementation steps of the information determination method described above.

[0015] As can be seen from the above technical solutions, the embodiments of this application have the following advantages: In this embodiment of the application, after obtaining the historical order data of the target user, the semantic features corresponding to the target user's consumption behavior can be determined based on the historical order data. Then, based on the semantic features and global statistical features, the product recommendation information and recommendation timing for the target user can be determined. Here, the global statistical features can be used to indicate the consumption time pattern determined based on big data. Therefore, when the recommendation timing is reached, the product recommendation information can be output.

[0016] It is evident that analyzing the historical order data of target users allows us to extract corresponding semantic features from fragmented order data, thereby understanding the logic behind their consumption behavior. Furthermore, since global statistical features are based on general consumption time patterns determined by big data, combining the individual consumption behavior characteristics of target users (i.e., the aforementioned semantic features) and the group consumption behavior patterns (i.e., the aforementioned global statistical features) for recommendations not only enables personalized adaptation to target users but also provides standardized data references. This reduces prediction bias caused by a single abnormal consumption behavior of a target user, thereby further improving the accuracy and reliability of the recommendation results. Attached Figure Description

[0017] Figure 1 A flowchart illustrating an information determination method provided in this application embodiment; Figure 2 A schematic diagram illustrating the determination of recommendation timing provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an information determination system provided in an embodiment of this application; Figure 4 This is a schematic diagram of an information determination device provided in an embodiment of this application. Detailed Implementation

[0018] As mentioned earlier, existing recommendation methods for this type of behavior are insufficient to meet the needs for refined and personalized recommendations. Specifically, existing methods mainly fall into the following two categories: First, there's the fixed-rule method. This method uses pre-set, uniform time and behavioral thresholds as recommendation triggers. For example, it might automatically send a repurchase reminder 30 days after a user completes a purchase. Because this method uses standardized, undifferentiated rules, it cannot adapt to the different consumption habits or usage scenarios of various users. For users who hoard goods or use products infrequently, it can cause unnecessary marketing disruptions; while for users with high purchase frequency, the fixed thresholds may lead to delayed repurchase reminders, ultimately affecting marketing conversion efficiency and user experience.

[0019] Second, there are traditional statistical or machine learning methods. These methods rely on collaborative filtering algorithms or neural network models to predict behavior and make recommendations through statistical analysis of users' historical consumption data. However, these methods can only quantitatively analyze surface-level user behavior data, lacking a deep understanding of the underlying business semantics and scenario logic, and cannot identify the phased and correlated patterns of product usage. For example, they cannot determine that a user's switch from a size S to a size M pacifier is a natural consumption path in infant development. Furthermore, these methods heavily depend on sufficient historical consumption data for individual users; for new users with no or very little historical data, there is a serious cold start problem, making it difficult to output effective predictive results.

[0020] It is evident that existing solutions either lack personalized adaptability due to rigid rules, or fail to address the issues of semantic understanding deficiency and cold start for new users by focusing only on surface-level data features and relying on historical user data.

[0021] In order to solve the above problems, this application provides an information determination method, which includes: after obtaining the historical order data of the target user, determining the semantic features corresponding to the target user's consumption behavior based on the historical order data; then, determining the product recommendation information and recommendation timing for the target user based on the semantic features and global statistical features. Here, the global statistical features can be used to indicate the consumption time pattern determined based on big data; therefore, when the recommendation timing is reached, the product recommendation information can be output.

[0022] It is evident that analyzing the historical order data of target users allows us to extract corresponding semantic features from fragmented order data, thereby understanding the logic behind their consumption behavior. Furthermore, since global statistical features are based on general consumption time patterns determined by big data, combining the individual consumption behavior characteristics of target users (i.e., the aforementioned semantic features) and the group consumption behavior patterns (i.e., the aforementioned global statistical features) for recommendations not only enables personalized adaptation to target users but also provides standardized data references. This reduces prediction bias caused by a single abnormal consumption behavior of a target user, thereby further improving the accuracy and reliability of the recommendation results.

[0023] Furthermore, since global statistical data can serve as a standardized data reference, even if the target user has limited historical order data, basic recommendation timing and products can be directly provided based on global statistical features, thereby solving the cold start problem caused by insufficient data.

[0024] It should be noted that the embodiments of this application do not limit the executing entity of the information determination method. For example, the information determination method of this application embodiment can be applied to data processing devices such as servers or terminal devices. The server can be a standalone server, a cluster server, or a cloud server. The terminal device can be an electronic device such as a smartphone, computer, personal digital assistant (PDA), or tablet computer.

[0025] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0026] Figure 1 This is a flowchart illustrating an information determination method provided in an embodiment of this application. (In conjunction with...) Figure 1 As shown, the information determination method provided in this application embodiment may include the following steps S101-S104.

[0027] S101: Obtain historical order data of the target user.

[0028] Historical order data refers to order data corresponding to the target user's historical purchasing behavior.

[0029] In this embodiment, historical order data can be extracted from the e-commerce platform's original order database. This original order database is a data asset for the daily operation of the e-commerce platform and can store all order data of all users on the platform for a long period of time.

[0030] It is understood that, in the specific embodiments of this application, since user privacy-related data such as historical order data are involved, when the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0031] S102: Based on historical order data, determine the semantic features corresponding to the consumption behavior of target users.

[0032] In this embodiment, the semantic features corresponding to the target user's consumption behavior refer to the personalized features associated with the target user's consumption behavior. These semantic features are then used as the basis for personalized adjustments. Through large-scale model inference, global statistical features are dynamically adjusted, transforming the recommendation results from a group average to an individual fit. This perfectly adapts to the usage habits or consumption preferences of different users, improving the accuracy of recommendations and the user experience.

[0033] Based on this, in the specific implementation of the process of determining the aforementioned semantic features, firstly, natural language conversion can be performed on the historical order data to obtain descriptive information of the consumption behavior corresponding to the historical order data. Next, semantic analysis can be performed using a large language model, that is, the descriptive information can be processed through the large language model to generate semantic features.

[0034] For example, structured historical order data can be converted into the following natural language: User A purchased a wide-neck glass baby bottle (high price) on October 1, 2025, and nursing pads on October 15, 2025. Then, through analysis using a large language model, semantic features can be generated: [Price sensitivity: low; Feeding method: breastfeeding; Expected baby age: 3 months]. It is evident that this semantic tag can include multiple dimensions, covering core dimensions such as purchasing power, usage habits, and product usage scenarios. This helps to understand the logic behind the target user's consumption behavior, enabling personalized recommendations tailored to the target user.

[0035] S103: Based on semantic features and global statistical features, determine product recommendation information and timing for target users. Global statistical features are used to indicate consumption time patterns determined based on big data.

[0036] In this embodiment, global statistical features refer to the general consumption time patterns of users extracted from massive historical orders. Product recommendation information indicates the products to be recommended to the target user. Recommendation timing indicates the timing for recommending the products indicated by the aforementioned product recommendation information to the target user.

[0037] In this way, subsequent recommendations based on the individual consumption behavior characteristics of target users (i.e., the semantic features mentioned above) and the consumption behavior patterns of groups (i.e., the global statistical features mentioned above) can not only be personalized to target users, but also provide standardized data references, reducing the bias in recommendation predictions caused by a single abnormal consumption behavior of a target user, thereby further improving the accuracy and reliability of recommendation results.

[0038] Therefore, for ease of understanding, the process of obtaining this global statistical feature will be described below with reference to one possible implementation method.

[0039] As one possible implementation method, for the process of obtaining the above-mentioned global statistical features, in specific implementation, a global statistical feature model can be constructed first, which includes a histogram model and a concept graph model.

[0040] The histogram model is used to describe the time it takes for different candidate users to repeatedly purchase the same type of non-durable goods. Different candidate users refer to users corresponding to massive amounts of historical order data in big data analytics. Non-durable goods are products characterized by short consumption cycles and high purchase frequency, such as a 200-sheet pack of tissues, a 500g bottle of milk powder, or a 700ml bottle cleaner.

[0041] Accordingly, the process of constructing a histogram model for non-durable products of the same specification can first be based on big data statistics of the time difference information of each candidate user purchasing the same non-durable product in two consecutive transactions, and then a histogram model can be constructed based on the time difference information of each candidate user.

[0042] In practice, firstly, orders for the primary target product (i.e., the aforementioned non-durable products of the same specifications) are filtered from the big data. Next, orders from candidate users with at least two purchase records for that primary target product are selected, as a single purchase without repeat purchases is statistically insignificant. Further, for each selected candidate user's order, the time difference between two consecutive purchases of the primary target product is calculated. For example, if candidate user A's most recent purchase of product 1 was at time T... n Candidate user A's last purchase of the same item 1 was on the date T. n-1 Then the time difference information corresponding to candidate user A can be denoted as T. n -T n-1The time when candidate user B last purchased item 1 was T. s Candidate user B's last purchase of the same item 1 was at time T. s-1 Then the time difference information corresponding to candidate user B can be denoted as T. s -T s-1 Then, the time difference information corresponding to all candidate users is used as the horizontal axis of the histogram model, and the number of each time difference information is counted as the vertical axis of the histogram model, thereby describing the time pattern of different candidate users continuously purchasing the same specification of non-durable goods.

[0043] The concept graph model is used to describe the time it takes for different candidate users to purchase durable goods of different sizes. Durable goods refer to products with characteristics such as long consumption cycles and short purchase frequencies, such as small-size baby clothes, large-size baby clothes, small-size pacifiers, medium-size pacifiers, and feeding bowls, which are daily necessities that need to be replaced as the usage scenario or stage changes.

[0044] Accordingly, regarding the construction process of the concept graph model, we can first use big data to statistically analyze the time each candidate user spends purchasing different specifications of durable products. Then, we construct a directed graph as the concept graph model. This directed graph includes multiple nodes and edges between multiple nodes. The multiple nodes represent different specifications of durable products, and the weights of the edges between multiple nodes represent the average conversion time and conversion probability of each candidate user purchasing different specifications of durable products.

[0045] In practical implementation, firstly, orders for the second target product (i.e., the aforementioned durable products of different specifications) are filtered from the big data. Next, orders from candidate users with at least two purchase records for this second target product are selected, as a single purchase without continuous repeat purchases is statistically insignificant. Further, the time information in the orders of each selected candidate user is identified to extract directed conversion behavior. For example, candidate user 3 purchasing item S2 first and then item M2 is a directed conversion behavior. Then, the time is labeled for each directed conversion behavior, and the directed conversion behaviors of all candidate users are summarized. Furthermore, for each directed conversion behavior, its conversion probability and average conversion time are calculated. For example, still using item 2 as an example, the conversion probability can be calculated by counting the number of candidate users who purchased item S2 first and then item M2, and counting the number of all candidate users who purchased item M2; the corresponding conversion probability is obtained based on the quotient of these two. The average conversion time refers to the average time difference between a candidate user purchasing item S2 first and then item M2. Finally, based on these different specifications of durable products and their corresponding average conversion time and conversion probability, a directed graph G(V,E) can be constructed, where V represents different specifications of durable goods and E represents weighted edges, thereby describing the time pattern of different candidate users purchasing different specifications of durable goods.

[0046] Accordingly, after constructing the histogram model and the concept graph model, global statistical features can be extracted based on the histogram model or the concept graph model.

[0047] In practical implementation, for histogram models, the mode can be extracted as a global statistical feature. For concept graph models, the mainstream transition directions, transition probabilities, and average transition times can be extracted as global statistical features.

[0048] As can be seen, after obtaining the global statistical features and semantic features, we can further determine the corresponding product recommendation information and recommendation timing for the target user. This allows the recommendation results to be both personalized and adapted to the target user, while also providing standardized data references. This reduces the prediction bias caused by a single abnormal consumption behavior of the target user, thereby further improving the accuracy and reliability of the recommendation results.

[0049] To facilitate understanding, the process of determining product recommendation information and recommendation timing for target users will be described below with reference to one possible implementation method.

[0050] As one possible implementation method, semantic features and global statistical features can be filled into the prompt word template to obtain the prompt words to be processed.

[0051] Here, the prompt word template refers to the template that guides the large language model to reason based on the provided information to obtain product recommendation information and recommendation timing. It may include preset logical rules and fields to be filled in.

[0052] Preset logic rules refer to the inherent life cycle patterns and usage logic rules of different product categories. For example, matching baby's age with nipple size, the demand substitution relationship between feeding bowls and nipples, the regular usage and consumption rate of bottle cleaner, the upper limit of the replacement cycle of durable goods, or some causal relationship rules across product categories, such as the purchase of weaning aids indicating a decrease in the demand for nipples.

[0053] The field to be filled can be filled with semantic features and global statistical features. After obtaining the prompt words to be processed, the large language model can be used to analyze the prompt words to obtain product recommendation information and recommendation timing.

[0054] Specifically, large language models can be used to refine global statistical features based on semantic features. That is, the large language model first generates a refinement factor based on the semantic features. This refinement factor can include acceleration or deceleration factors; acceleration factors shorten the temporal pattern information provided by the global statistical features, while deceleration factors lengthen the temporal pattern information provided by the global statistical features. Next, the large language model refines the global statistical features based on the refinement factor to obtain the refined features, and then performs conflict detection processing on the refined features based on preset logical rules to obtain consumer recommendation information.

[0055] For example, taking non-durable goods as an example, combined with Figure 2 As shown, the mode or the median of the mode interval displayed in the histogram model can be extracted as a global statistical feature. Accordingly, for user A, if they are determined to be a high-frequency buyer based on their corresponding semantic features, this global statistical feature can be moved forward (i.e., shifted to the left) as the corresponding recommendation timing; for user B, if they are determined to be a low-frequency buyer based on their corresponding semantic features, this global statistical feature can be moved backward (i.e., shifted to the right) as the corresponding recommendation timing.

[0056] For example, taking a durable good like a pacifier as an example, the pre-defined rule logic could be that the faster a baby grows, the earlier the pacifier should be replaced. Correspondingly, assuming that global statistical features indicate a typical 90-day interval between purchasing a size S pacifier and a size M pacifier, and semantic features indicate that the target user has high purchasing power and values ​​hygiene, the large language model can determine the recommended timing of the product recommendation information as 90 × 0.8 = 72 days after the target user purchases a size S pacifier. Then, the large language model performs conflict detection processing based on the pre-defined rule logic to obtain the final product recommendation information. For example, if the target user has not recently purchased a weaning aid or complementary food, the large language model can determine that the recommended product is a size M pacifier. If the target user has recently purchased a weaning aid, it indicates a conflict between the pacifier and the user's weaning needs; therefore, the large language model can determine that the recommended product is a sippy cup.

[0057] S104: When the recommendation time is reached, output product recommendation information.

[0058] In this embodiment, the output format of the product recommendation information is not specifically limited. For example, relevant text messages can be pushed to the target user's terminal device, or relevant notification messages can be pushed through an e-commerce platform.

[0059] Based on the relevant content of steps S101-S104 above, it can be seen that in this embodiment, after obtaining the target user's historical order data, the semantic features corresponding to the target user's consumption behavior can be determined first based on the historical order data. Then, based on the semantic features and global statistical features, product recommendation information and recommendation timing for the target user can be determined. Here, global statistical features can be used to indicate the consumption time pattern determined based on big data. Therefore, when the recommendation timing is reached, product recommendation information can be output. It can be seen that by analyzing the target user's historical order data, the corresponding semantic features can be mined from fragmented order data, thereby understanding the logic behind the target user's consumption behavior. Furthermore, since global statistical features are based on the general consumption time pattern determined by big data, combining the individual consumption behavior characteristics of the target user (i.e., the aforementioned semantic features) and the group consumption behavior pattern (i.e., the aforementioned global statistical features) for recommendation can both personalize the target user and provide standardized data references, reducing the recommendation prediction deviation caused by a target user's abnormal consumption behavior, thereby further improving the accuracy and reliability of the recommendation results.

[0060] Based on the information determination method provided in the above embodiments, this application can also provide an information determination system. The information determination system will now be described in conjunction with embodiments and accompanying drawings.

[0061] Figure 3 This is a schematic diagram of the structure of an information determination system provided in an embodiment of this application. (In conjunction with...) Figure 3 As shown in the embodiments of this application, the information determination system may specifically include a data layer, a statistics layer, an inference layer, and an execution layer.

[0062] The data layer, serving as the system's foundational data input layer, is the data source for all subsequent statistics and inference. It includes the original order database and a data cleaning and standardization module. Specifically, the data representing user orders, such as SQL data or logs in the original order database, can be preprocessed using the data cleaning and standardization module to obtain cleaned, structured data.

[0063] The statistical layer is responsible for extracting global statistical features and classifying basic user profiles, serving as a crucial link between raw data and semantic reasoning. It comprises a global statistical engine and a user profiling engine. Specifically, the global statistical engine generates global statistical features using histogram and concept graph models, while the user profiling engine generates semantic features corresponding to the target user's consumption behavior.

[0064] The inference layer, as the core feature fusion and reasoning layer of the system, can generate parameterized prompt word templates and fill the aforementioned global statistical features and semantic features into inference instructions that the large language model can understand. Then, the large language model can complete logical reasoning according to the prompt word templates to obtain product recommendation information and recommendation timing for the target user.

[0065] As the final execution layer of the system, the execution layer can use the CRM system and dynamic timers to detect the user's consumption time interval in real time and push information to the target user according to the recommended timing, so as to complete the closed loop of the entire repurchase prediction.

[0066] Therefore, by analyzing the historical order data of target users, we can extract corresponding semantic features from fragmented order data, thereby understanding the logic behind the target users' consumption behavior. Furthermore, since global statistical features are based on general consumption time patterns determined by big data, combining the individual consumption behavior characteristics of target users (i.e., the aforementioned semantic features) and the group consumption behavior patterns (i.e., the aforementioned global statistical features) for recommendations not only allows for personalized adaptation to target users but also provides standardized data references. This reduces the prediction bias caused by a single abnormal consumption behavior of a target user, thereby further improving the accuracy and reliability of the recommendation results.

[0067] Based on the information determination method provided in the above embodiments, this application can also provide an information determination device. The information determination device will now be described in conjunction with the embodiments and accompanying drawings.

[0068] Figure 4 This is a schematic diagram of the structure of an information determination device provided in an embodiment of this application. (In conjunction with...) Figure 4 As shown, the information determining device 400 provided in this application embodiment includes: Data acquisition module 401 is used to acquire historical order data of the target user; The semantic feature determination module 402 is used to determine the semantic features corresponding to the consumption behavior of the target user based on the historical order data. The information determination module 403 is used to determine product recommendation information and recommendation timing for the target user based on the semantic features and global statistical features, wherein the global statistical features are used to indicate the consumption time pattern determined based on big data. The information output module 404 is used to output the product recommendation information when the recommendation time is reached.

[0069] Optionally, the information determination module 403 is specifically used for: The semantic features and the global statistical features are filled into the prompt word template to obtain the prompt words to be processed; The large language model is used to analyze the prompt words to be processed to obtain the product recommendation information and the recommendation timing. The large language model is used to correct the global statistical features based on the semantic features.

[0070] Optionally, the prompt words to be processed include the semantic features, the global statistical features, and preset logical rules; The prompt word to be processed is used to instruct the large language model to generate a correction factor based on the semantic features, to correct the global statistical features based on the correction factor to obtain the corrected features, and to perform conflict detection processing on the corrected features based on the preset logical rules to obtain the recommendation timing.

[0071] Optionally, the global statistical features are obtained through the following modules: The model building module is used to build a global statistical feature model, which includes a histogram model and a concept graph model. The histogram model is used to describe the time when different candidate users purchase the same size of non-durable goods, and the concept graph model is used to describe the time when different candidate users purchase different sizes of durable goods. The feature extraction module is used to extract the global statistical features based on the histogram model or the concept graph model.

[0072] Optionally, the histogram model is constructed using the following modules: The time difference statistics module is used to calculate the time difference between two consecutive purchases of the same specification of non-durable products by each candidate user among the different candidate users based on big data. The first model building module is used to build the histogram model based on the time difference corresponding to each candidate user.

[0073] Optionally, the concept graph model is constructed using the following modules: The time statistics module is used to calculate the time each candidate user in the different candidate users purchased durable products of different specifications based on big data. The second model building module is used to construct a directed graph as the conceptual graph model. The directed graph includes multiple nodes and edges between the multiple nodes. The multiple nodes are used to represent durable products of different specifications. The weights of the edges between the multiple nodes are used to represent the average conversion time and conversion probability of each candidate user purchasing durable products of different specifications.

[0074] Optionally, the semantic feature determination module is used to: Natural language processing is performed on the historical order data to obtain descriptive information about the consumption behavior corresponding to the historical order data; The semantic features are generated by processing the descriptive information using a large language model.

[0075] Furthermore, embodiments of this application also provide an electronic device, including: a processor, a memory, and a system bus; The processor and the memory are connected via the system bus; The memory is used to store one or more programs, the one or more programs including instructions that, when executed by the processor, cause the processor to perform any of the implementation steps of the information determination method described above.

[0076] Furthermore, embodiments of this application also provide a computer-readable storage medium storing instructions that, when executed on an electronic device, cause any of the above-described steps of the information determination method to be implemented.

[0077] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that all or part of the steps in the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network communication device such as a media gateway, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application. It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on describing the differences from other embodiments. The same or similar parts between the various embodiments can be referred to mutually.

[0078] The system disclosed in the embodiments is described in a relatively simple manner because it corresponds to the method disclosed in the embodiments. For relevant details, please refer to the method section.

[0079] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0080] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for determining information, characterized in that, include: Obtain historical order data from the target user; Based on the historical order data, determine the semantic features corresponding to the target user's consumption behavior; Based on the semantic features and global statistical features, product recommendation information and recommendation timing for the target user are determined, and the global statistical features are used to indicate the consumption time pattern determined based on big data. When the recommended timing is reached, the product recommendation information is output.

2. The information determination method according to claim 1, characterized in that, The step of determining product recommendation information and recommendation timing for the target user based on the semantic features and global statistical features includes: The semantic features and the global statistical features are filled into the prompt word template to obtain the prompt words to be processed; The large language model is used to analyze the prompt words to be processed to obtain the product recommendation information and the recommendation timing. The large language model is used to correct the global statistical features based on the semantic features.

3. The information determination method according to claim 2, characterized in that, The prompt words to be processed include the semantic features, the global statistical features, and the preset logical rules; The prompt word to be processed is used to instruct the large language model to generate a correction factor based on the semantic features, to correct the global statistical features based on the correction factor to obtain the corrected features, and to perform conflict detection processing on the corrected features based on the preset logical rules to obtain the recommendation timing.

4. The information determination method according to claim 1, characterized in that, The global statistical features are obtained through the following steps: A global statistical feature model is constructed, which includes a histogram model and a concept graph model. The histogram model is used to describe the time when different candidate users purchase non-durable goods of the same specification, and the concept graph model is used to describe the time when different candidate users purchase durable goods of different specifications. The global statistical features are extracted based on the histogram model or the concept graph model.

5. The information determination method according to claim 4, characterized in that, The histogram model is constructed through the following steps: Based on big data statistics, the time difference between two consecutive purchases of the same specification of non-durable products by each of the different candidate users; The histogram model is constructed based on the time difference corresponding to each candidate user.

6. The information determination method according to claim 4, characterized in that, The concept map model is constructed through the following steps: Based on big data statistics, the time each candidate user spent purchasing durable products of different specifications was analyzed. A directed graph is constructed as the conceptual graph model. The directed graph includes multiple nodes and edges between the multiple nodes. The multiple nodes are used to represent durable products of different specifications. The weights of the edges between the multiple nodes are used to represent the average conversion time and conversion probability of each candidate user purchasing durable products of different specifications.

7. The information determination method according to any one of claims 1 to 6, characterized in that, The step of determining the semantic features corresponding to the target user's consumption behavior based on the historical order data includes: Natural language processing is performed on the historical order data to obtain descriptive information about the consumption behavior corresponding to the historical order data; The semantic features are generated by processing the descriptive information using a large language model.

8. An information determining device, characterized in that, include: The data acquisition module is used to acquire historical order data of the target user; The semantic feature determination module is used to determine the semantic features corresponding to the consumption behavior of the target user based on the historical order data. The information determination module is used to determine product recommendation information and recommendation timing for the target user based on the semantic features and global statistical features, wherein the global statistical features are used to indicate the consumption time pattern determined based on big data; The information output module is used to output the product recommendation information when the recommendation time is reached.

9. An electronic device, characterized in that, The device includes: a processor, a memory, and a system bus; The processor and the memory are connected via the system bus; The memory is used to store a program, the program including instructions that, when executed by the processor, cause the processor to perform the steps of the information determination method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the information determination method as described in any one of claims 1 to 7.

Images