Commodity combination construction methods, apparatus, equipment, media, and products

Abstract

This application relates to a method, apparatus, device, medium, and product for constructing product combinations. The method includes: obtaining a product set for each user; combining product items in the product set with an arbitrary number of items; enumerating all product combinations to form a user combination set; merging all user combination sets after deduplication to form a full combination set; determining the frequency of occurrence of each product combination in the full combination set; determining cohesion based on the mutual information entropy of each product combination in the full combination set, where cohesion characterizes the stability of combinations where multiple product items are accessed by the same user; determining the degree of freedom based on the adjacency entropy of each product combination in the full combination set relative to the entire product set, where the degree of freedom characterizes the richness of combinations where the corresponding product combination is accessed by the same user when paired with other product items; and selecting multiple product combinations based on the frequency, cohesion, and degree of freedom to form a list of high-quality combinations. This application can determine high-quality product combinations for product recommendation services.

Description

Technical Field

[0001] This application relates to e-commerce information processing technology, and in particular to a method for constructing a product combination and its apparatus, equipment, medium, and product. Background Technology

[0002] Product bundle recommendations are a frequently used promotional method on e-commerce platforms, especially those based on independent websites. New products or users on these independent websites may experience a cold start due to a lack of historical data, making it difficult to provide suitable new product recommendations. By identifying high-quality product bundles based on prior knowledge, and combining multiple product items with related access, especially related purchase relationships, into these high-quality bundles, and then recommending other products from the bundle when a user visits one of the bundles, the obstacles of a cold start can be overcome, achieving effective marketing and promotion.

[0003] In traditional technologies, besides manually customizing product combinations, a common approach is to utilize various product similarity algorithms to obtain user behavior data for each product item. Based on the characteristic that product items are purchased by the same users, co-occurrence information between product items is determined, and two product items with a high probability of co-occurrence are identified as a basic combination. The characteristic of this algorithm is that because it identifies based on two product items at a time, it cannot examine the broad connections between product items from a macro perspective, leading to inaccurate combination results. Furthermore, the number of recommended product items that can be provided each time is limited, resulting in a very weak effect of the product combination-based recommendation mechanism.

[0004] In reality, for a product mix to be considered a high-quality mix, it's not enough to consider its usage volume. It's also crucial to consider the stability of the fixed pairings between its various product items, as well as the independence it demonstrates by being able to be widely paired with other product items. Looking at traditional techniques for obtaining product mixes from this perspective reveals that they can only uncover similarities in purchasing behavior between different product items, failing to systematically and comprehensively ensure the reasonableness of the product mix in relation to actual sales scenarios.

[0005] In view of this, there is still room for improvement in data mining techniques for product combinations, and further exploration is needed. Summary of the Invention

[0006] The purpose of this application is to solve the above-mentioned problems by providing a method for constructing a product combination and the corresponding apparatus, equipment, non-volatile readable storage medium, and computer program product.

[0007] According to one aspect of this application, a method for constructing a product combination is provided, comprising the following steps:

[0008] Get each user's product set, which contains multiple product items that the user has visited. Combine the product items in the product set with any number of items, enumerate all product combinations to form a user combination set, and then merge all user combination sets after removing duplicates to form the full combination set.

[0009] Determine the frequency of occurrence of each product combination in the full set of combinations across all user sets;

[0010] The cohesion of a product combination is determined based on the mutual information entropy obtained after arbitrarily dividing each product combination in the full set of combinations. The cohesion characterizes the stability of the combination of multiple product items included in the corresponding product combination being accessed by the same user.

[0011] The degree of freedom of each product combination is determined based on the adjacency entropy of each product combination relative to the product set of all users in the full set of combinations. The degree of freedom represents the richness of the combination of the corresponding product combination with other product items and accessed by the same user.

[0012] Based on the frequency of occurrence, cohesion, and degree of freedom, multiple product combinations are selected from the full set of combinations to form a list of high-quality combinations.

[0013] Optionally, obtain each user's product set, which contains multiple product items that the user has visited. Combine the product items in the product set with any number of items, enumerate all product combinations to form a user combination set, and then deduplicate and merge all user combination sets to form the full combination set, including:

[0014] Obtain access behavior data generated by each user in the same online store in historical access events, and determine from the access behavior data that multiple product items accessed by the user constitute the user's product set. The historical access events include events of adding product items to the shopping cart and / or adding product items to the purchase order.

[0015] For each user's product set, the product items are combined according to any number of product items, and all possible product combinations in the product set are enumerated to form the user's combination set;

[0016] The product combinations in the user combination set of all users are deduplicated and merged to construct the full combination set.

[0017] Optionally, the cohesion of the product combinations is determined based on the mutual information entropy obtained after arbitrarily bisecting each product combination in the full set of combinations, including:

[0018] For each product combination in the full set of combinations, the product combination is arbitrarily bisected by product item to obtain the left set and right set corresponding to each split.

[0019] Calculate the mutual information entropy corresponding to each split based on the left and right sets obtained from each split;

[0020] The lowest mutual information entropy obtained after multiple partitions of each product combination is taken as the cohesion of that product combination.

[0021] Optionally, the degrees of freedom of each product combination in the full set of combinations are determined based on the adjacency entropy of each product combination relative to the product set of all users, including:

[0022] Sort the product items in each user's product collection according to the user's access time;

[0023] For each product combination in the full set of combinations, calculate its left and right adjacency entropy relative to the sorted product sets of each user, and determine the lowest value as the minimum adjacency entropy.

[0024] For each product combination, the lowest adjacency entropy relative to all users is summed as the degree of freedom of that product combination.

[0025] Optionally, based on the frequency of occurrence, cohesion, and degrees of freedom, multiple product combinations from the full set of combinations are selected to form a list of high-quality combinations, including:

[0026] The frequency of occurrence, cohesion, and degree of freedom of each product combination in the full set of combinations are multiplied together to obtain the comprehensive score corresponding to that product combination;

[0027] Based on the comprehensive score, the product combinations in the full set are screened to obtain a list of high-quality product combinations with relatively high comprehensive scores.

[0028] Optionally, after constructing the list of high-quality combinations, include:

[0029] Respond to any access event triggered by a user and retrieve the target product item corresponding to that access event;

[0030] A candidate list is formed by retrieving all product combinations containing the target product item from the list of high-quality combinations.

[0031] Obtain the measured scores of each product combination in the candidate list, and determine the product combination with the highest measured score as the target product combination;

[0032] The product information of all other product items included in the target product combination, excluding the target product item, is pushed to the arbitrary user.

[0033] According to another aspect of this application, a product combination construction device is provided, comprising:

[0034] The data processing module is set to obtain each user's product set, which contains multiple product items that the user has visited, combine the product items in the product set with any number of items, enumerate all product combinations to form a user combination set, and then deduplicate and merge all user combination sets to form a full combination set.

[0035] The frequency determination module is set to determine the frequency of occurrence of each product combination in the full set of combinations in each user combination set.

[0036] The cohesion determination module is configured to determine the cohesion of a product combination based on the mutual information entropy obtained after arbitrarily dividing each product combination in the full set of combinations. The cohesion characterizes the stability of the combination of multiple product items included in the corresponding product combination being accessed by the same user.

[0037] The degree of freedom determination module is configured to determine the degree of freedom of each product combination based on the adjacency entropy of each product combination in the full set of combinations relative to the product sets of all users. The degree of freedom represents the richness of the combination of the corresponding product combination with other product items and accessed by the same user.

[0038] The list construction module is configured to filter multiple product combinations from the full set of combinations based on the frequency of occurrence, cohesion, and degree of freedom, thereby forming a list of high-quality combinations.

[0039] According to another aspect of this application, a product combination construction apparatus is provided, including a central processing unit and a memory, wherein the central processing unit is configured to invoke and run a computer program stored in the memory to perform the steps of the product combination construction method described in this application.

[0040] According to another aspect of this application, a non-volatile readable storage medium is provided, which stores a computer program implemented according to the product combination construction method in the form of computer-readable instructions, wherein the computer program, when invoked by a computer, performs the steps included in the method.

[0041] According to another aspect of this application, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the method described in any embodiment of this application.

[0042] Compared with existing technologies, this application has several technological advantages, including but not limited to:

[0043] First, this application processes the product set, which consists of all historically accessed product items by all users, exhaustively listing all product combinations from the product set to form a user combination set for each user. Then, the user combination sets are deduplicated and merged to obtain the full combination set. Next, by examining the data characteristics of each product combination in the full combination set relative to each user's user combination set and product set, statistical characteristics such as the frequency of occurrence, cohesion, and degree of freedom of each product combination are obtained. Based on these statistical characteristics, a selection of product combinations is selected to form a high-quality combination list. The number of product items in each product combination is not limited by technology. The product combinations in the high-quality combination list are optimized under the combined effect of multiple statistical characteristics, which are more consistent with the global data and can more comprehensively reflect the associated access characteristics between product items. The product items in the list have greater value in mutual combination.

[0044] Secondly, this application analyzes each product combination to demonstrate its popularity by statistically analyzing its frequency of occurrence relative to the global data, to demonstrate the stability of fixed combinations among multiple product items by statistically analyzing cohesion, and to demonstrate the richness of combinations that the product combination can be widely combined with other product items by statistically analyzing degrees of freedom. This multi-dimensional approach provides precise and effective quantification of information value, making the overall information value of each product combination more accurate. Product combinations selected based on these statistical characteristics have higher information quality.

[0045] Furthermore, this application can be applied to independent websites of e-commerce platforms, taking into account the relatively small amount of historical data on a single independent website. It identifies a list of high-quality product combinations to provide product combination recommendations to users visiting the independent website, enabling a cold start for product recommendations and thus possessing high practical value. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a schematic diagram of the network architecture of the application environment for the technical solution of this application;

[0048] Figure 2 This is a flowchart illustrating one embodiment of the product combination construction method of this application;

[0049] Figure 3 This is a schematic diagram illustrating the data processing of user access behavior data in an embodiment of this application;

[0050] Figure 4 This is a flowchart illustrating the process of determining the cohesion of a product mix in an embodiment of this application;

[0051] Figure 5 This is a flowchart illustrating the degree of freedom in determining the combination of goods in an embodiment of this application;

[0052] Figure 6 This is a schematic diagram illustrating the process of using a list of high-quality combinations to promote advertising in an embodiment of this application;

[0053] Figure 7 This is a schematic block diagram of the product combination construction device of this application;

[0054] Figure 8 This is a schematic diagram of a product combination construction device used in this application. Detailed Implementation

[0055] The models cited or potentially cited in this application, including traditional machine learning models or deep learning models, can be deployed on a remote server and invoked remotely on the client, or deployed on a client with the capability to invoke directly, unless explicitly specified in the text. In some embodiments, when running on the client, the corresponding intelligence can be obtained through transfer learning in order to reduce the requirements on the client's hardware operating resources and avoid excessive occupation of the client's hardware operating resources.

[0056] Please see Figure 1 The network architecture adopted in an exemplary application scenario of this application includes a terminal device 80, an independent website 81, and an application server 82. The application server 82 can be used to deploy the product recommendation service of this application. The independent website 81 can be used to deploy and open an online store for e-commerce services. Users on the terminal device 80 can browse product information of product items on the independent website 81 on the page of the online store. The independent website 81 can call the product recommendation service of the application server 82 based on any target product item currently accessed by the user. The product recommendation service determines other product items that match the target product item according to the product combination in the pre-customized high-quality combination list and returns them to the independent website 81. The independent website 81 then recommends the product information of the other product items to the terminal device 80 for the user to browse further, thereby achieving the purpose of product recommendation.

[0057] The product combinations in the high-quality combination list can be determined by the instructions of a computer program product implemented according to the product combination construction method of this application. The relevant instructions are based on various information provided by the independent website 81, including user access behavior data and various product information of product items, and execute the product combination construction method to determine the high-quality product combinations, which constitute the high-quality combination list. Then, the list is stored as the high-quality combination list corresponding to the independent website 81 for the product recommendation service to call.

[0058] It should be noted that the computer program product implemented according to the product combination construction method of this application can run on any computer device, including but not limited to the independent website 81, application server 82 or other servers, as long as it can call the various basic data required to construct the high-quality combination list from the independent website 81.

[0059] Based on the principles revealed above, please refer to Figure 2 According to a product combination construction method provided in this application, in one embodiment, the method includes the following steps:

[0060] Step S1100: Obtain the product set for each user, which contains multiple product items that the user has visited. Combine the product items in the product set with any number of items, enumerate all product combinations to form a user combination set, and remove duplicates from all user combination sets and merge them to form a full combination set.

[0061] In an exemplary application scenario, a list of high-quality product combinations is generated for a standalone website, providing multiple excellent product combinations. This list can be constructed using user browsing behavior data generated from the website's online store as the foundation.

[0062] User access behavior data broadly refers to historical data generated from various user access behaviors. This historical data can be obtained by setting up tracking code on the pages visited by the user. For example, for a user's order purchase behavior, the corresponding order contains the items purchased by the user, thus obtaining the mapping relationship data between the user and the purchased items. Similarly, actions such as adding items to the shopping cart or browsing multiple items during the same session will generate corresponding behavioral data on the independent website. For various behavioral data, one or more types of behavioral data can be selected as the basic data required to construct a high-quality combination list. For example, for ease of understanding, the behavioral data corresponding to a user's order purchase behavior will be used as the basic data in the following text.

[0063] In one embodiment, the time of the required access behavior data can be constrained. For example, access behavior data can be obtained by retracing the current time back to a preset duration. The preset duration can be any value such as three days, one week, one month, one quarter, or half a year. By constraining the time, the access behavior data can better reflect the latest dynamics of the product mix.

[0064] For the same access behavior, each user's historical access behavior data contains mapping data such as access time, accessing user, and accessed product items. Therefore, the product items accessed by each user can be extracted to form a corresponding product set for that user. This product set contains all the product items accessed by that user historically. Considering that the same product item may appear in multiple historical behavior data sets, deduplication can be performed as needed.

[0065] It's easy to understand that each user can obtain their corresponding product set, which contains multiple product items that the user has visited, forming the basic data.

[0066] Furthermore, for each item in each user's item set, according to an exhaustive count of at least two items, multiple items of that count are selected and combined to form item combinations. For example, when there are three items {A; B; C} in the item set, four item combinations can be obtained, namely {AB; BC; AC; ABC}. This process continues, and through enumeration, all possible item combinations for each user are obtained. After enumerating all item combinations for each item set, a user combination set is obtained, which contains all item combinations corresponding to that user.

[0067] Different user sets may contain the same product combinations. To facilitate subsequent data processing, all user sets can be merged into a single full set. For the same product combinations in different user sets, deduplication can be performed so that the same product combination appears only once in the full set.

[0068] Step S1200: Determine the frequency of occurrence of each product combination in the full set of combinations in each user set;

[0069] The total number of times each product combination appears across all user combination sets, i.e., the frequency of occurrence, indicates the popularity of that product combination in historical browsing behavior. The higher the frequency of occurrence, theoretically, the more frequently the product combination has been used by users historically. Therefore, for each product combination in the full set of combinations, its frequency of occurrence can be determined first. This is its first statistical characteristic.

[0070] When calculating the frequency of each product combination in the full set of combinations, the total number of times that product combination appears in all user combinations across all users is counted, thus determining its corresponding frequency. Therefore, it can be seen that the frequency of each product combination in the full set of combinations can be obtained.

[0071] Step S1300: Determine the cohesion of the product combination based on the mutual information entropy obtained after arbitrarily dividing each product combination in the full set of combinations. The cohesion characterizes the stability of the combination of multiple product items included in the corresponding product combination being accessed by the same user.

[0072] To measure the stability of the fixed combinations of multiple product items in each product bundle that are accessed by users, which is actually a characterization of the closeness between the multiple product items, it may be determined by examining the mutual information entropy of the product bundle.

[0073] A product combination is composed of product items, and each product item is a fixed information object. Therefore, when calculating mutual information entropy, for any product combination for which mutual information entropy needs to be calculated, it can be divided into two parts multiple times. Then, according to the mutual information entropy formula, the mutual information entropy obtained in each division case can be calculated.

[0074] The formula for calculating the mutual information entropy is as follows:

[0075]

[0076] in:

[0077] It is the probability of a specific product combination appearing in all the behavior data corresponding to that specific visit behavior across the entire independent website. The probability of appearance can be determined by dividing the frequency of the product combination appearing in all the behavior data corresponding to that specific visit behavior by the total number of behavior data corresponding to that specific visit behavior.

[0078] It represents the probability of a combination of goods appearing in the entire set of combinations. It is the probability of occurrence of the product combination on the left side obtained by dividing the product combination into two parts. It is the probability of the occurrence of the right-hand part of the product combination obtained by dividing the product combination into two parts.

[0079] It should be noted that after a product combination is divided into two parts, the left part and / or the right part may each contain a single product item. This is also allowed and does not affect the calculation of mutual information entropy.

[0080] In one embodiment, for ease of measurement, the logarithmic value of the mutual information entropy can be taken to map its value range to a smoother numerical interval. It is easy to understand that when... and When they are on the same order of magnitude, A value close to 1, which becomes 0 after taking the logarithm, corresponds to a very low mutual information entropy. Taking a two-item combination as an example, if the mutual information entropy of this combination is very low, it means that the first and second items are weakly correlated, or even uncorrelated. and They will be on the same order of magnitude. Conversely, if the mutual information entropy of the product combination is high, the probabilities of products "A", "B", and "AB" are all very close, then... It will be a value much greater than 1.

[0081] As can be seen from the above process, after each product combination is arbitrarily divided multiple times, the mutual information entropy corresponding to each division can be obtained, thus obtaining multiple mutual information entropies. In one embodiment, the mutual information entropy with the lowest value is determined by comparison and used as the cohesion corresponding to the product combination. In another embodiment, the mutual information entropies corresponding to multiple divisions can also be averaged to determine the mean as the cohesion. Since cohesion is essentially mutual information entropy, it is easy to understand that cohesion characterizes the stability of the fixed combination formed by the various product items in the corresponding product combination when accessed together. That is, the greater the cohesion, the higher the stability of the fixed combination formed by the various product items in the product combination; the lower the cohesion, the lower the stability of the combination.

[0082] Based on the above principles, it can be seen that each product combination in the full set of combinations can obtain its corresponding cohesion, thereby obtaining the second statistical characteristic corresponding to each product combination.

[0083] Step S1400: Determine the degree of freedom of each product combination based on the adjacency entropy of each product combination relative to the product set of all users in the full set of combinations. The degree of freedom represents the richness of the combination of the corresponding product combination with other product items and accessed by the same user.

[0084] As mentioned earlier, each user's product set is constructed by extracting product items from the behavioral data corresponding to the user's access behavior. Each behavioral data carries corresponding time information. Therefore, the time information can be used to sort each user's product set so that the sorted product set can be used as reference data to determine the degree of freedom of each product combination in the full set of combinations.

[0085] In one embodiment, the set of goods used to determine the degrees of freedom is used such that even if identical goods exist in behavioral data from different times, they can be retained without deduplication. This allows for direct calculation of adjacency entropy to determine the degrees of freedom, based on the factual data characteristics reflected in the more original behavioral data. Alternatively, deduplication of identical goods from different times within the set and retention of the most recent item is also feasible.

[0086] To determine the degrees of freedom of each product combination in the full set of combinations, we can first refer to the product sets of all users sorted by time information, calculate the adjacency entropy of each product combination in these product sets, and then determine the corresponding degrees of freedom of the product combination based on the sum of the adjacency entropies of the product combination relative to all users.

[0087] Specifically, for each product combination in the full set of combinations, based on its position in the product set sorted by time information for each user, all product items appearing to the left of the product combination and all product items appearing to the right of the product combination are determined, regardless of the position of the product combination within the product set. Then, for the left set consisting of all product items on the left, its left adjacency entropy is calculated; similarly, for the right set consisting of all product items on the right, its right adjacency entropy is calculated. The formula for adjacency entropy is as follows:

[0088]

[0089] in, It is the probability of each item appearing in its corresponding set, such as the left set or the right set. It can be expressed as the number of times it appears divided by the total number of items in its set.

[0090] It can be seen that, regardless of the left or right adjacency entropy, the smaller the value, the lower the degree of freedom, and the more frequent and fixed the combination of the product combination with other product items tends to be. The larger the value, the higher the degree of freedom, indicating that the combination of the product combination with other product items is more diverse and even chaotic. Therefore, both left and right adjacency entropy can represent the degree of freedom and characterize the richness of the combination of the corresponding product combination with other product items and accessed by the same user.

[0091] It's easy to understand that the left and right adjacency entropies of the same product combination relative to the same user's product set may not be consistent. Therefore, in one embodiment, the lowest of the left and right adjacency entropies, i.e., the minimum adjacency entropy, can be selected as the basis for determining the degrees of freedom of the product combination. This approach makes the measurement of degrees of freedom more conservative. In another embodiment, the average of the left and right adjacency entropies can be taken as the minimum adjacency entropy, which can then be used as the basis for determining the degrees of freedom of the product combination. This approach can relatively balance the values ​​of the adjacency entropies of the product combination, avoiding extreme effects in the measurement of degrees of freedom.

[0092] Therefore, for each product combination in the full set of combinations, the minimum adjacency entropy corresponding to each user's product set can be determined. By summing and averaging the minimum adjacency entropies corresponding to all users, the resulting value can be used as the degree of freedom for the final use of that product combination. The degrees of freedom for each product combination are determined using the above process, thus establishing the third statistical characteristic for each product combination.

[0093] In another embodiment that can replace the above embodiments regarding the determination of degrees of freedom, the product sets of all users can be merged into a full product set. Then, according to the time information of the behavioral data, each product item in the set is sorted. Then, based on the full product set, the left adjacency entropy and right adjacency entropy of each product combination in the full combination set are calculated, and the corresponding lowest adjacency entropy is determined and directly used as the degree of freedom, which is the same as the above embodiments.

[0094] Step S1500: Based on the frequency of occurrence, cohesion, and degree of freedom, select multiple product combinations from the full set of combinations to form a list of high-quality combinations.

[0095] The frequency of occurrence of each product combination in the full set of combinations has been determined through the above process. Cohesion and degrees of freedom The frequency of occurrence represents the popularity of the product combination, cohesion represents the stability of the combination of multiple product items in the product combination, and freedom represents the richness of the combination of the product combination with other product items. These provide quantitative standards for product combinations from different dimensions. Therefore, in one embodiment, these quantitative values ​​from different dimensions can be integrated, for example, by directly multiplying the frequency of occurrence, cohesion, and freedom or by weighted summing, to determine the comprehensive score corresponding to each product combination. Subsequently, based on the comprehensive score, a selection of product combinations with relatively high comprehensive scores can be chosen as high-quality product combinations to form a list of high-quality combinations for use by the product recommendation service.

[0096] In another embodiment, a grid search approach can be used. First, an initial threshold is given for the occurrence frequency, cohesion, and degree of freedom. Then, by optimizing each parameter, the corresponding screening threshold for each occurrence frequency, cohesion, and degree of freedom is determined sequentially. Finally, the screening threshold is used to screen the product combinations in the full set of combinations to determine the high-quality product combinations and construct the list of high-quality combinations.

[0097] Once the list of high-quality combinations is determined from the entire set of combinations, the optimal selection of all product combinations is achieved. The selected combinations possess high popularity, strong pairing stability, and a wide variety of pairings, clearly indicating their high quality. When used as the basis for pairing schemes in product recommendation services, it eliminates reliance on historical data, providing effective related promotional information for product recommendation strategies, thereby improving advertising performance indicators and achieving excellent product recommendation results.

[0098] As can be seen from the above embodiments, this application has multiple technical advantages, including but not limited to:

[0099] First, this application processes the product set, which consists of all historically accessed product items by all users, exhaustively listing all product combinations from the product set to form a user combination set for each user. Then, the user combination sets are deduplicated and merged to obtain the full combination set. Next, by examining the data characteristics of each product combination in the full combination set relative to each user's user combination set and product set, statistical characteristics such as the frequency of occurrence, cohesion, and degree of freedom of each product combination are obtained. Based on these statistical characteristics, a selection of product combinations is selected to form a high-quality combination list. The number of product items in each product combination is not limited by technology. The product combinations in the high-quality combination list are optimized under the combined effect of multiple statistical characteristics, which are more consistent with the global data and can more comprehensively reflect the associated access characteristics between product items. The product items in the list have greater value in mutual combination.

[0100] Secondly, this application analyzes each product combination to demonstrate its popularity by statistically analyzing its frequency of occurrence relative to the global data, to demonstrate the stability of fixed combinations among multiple product items by statistically analyzing cohesion, and to demonstrate the richness of combinations that the product combination can be widely combined with other product items by statistically analyzing degrees of freedom. This multi-dimensional approach provides precise and effective quantification of information value, making the overall information value of each product combination more accurate. Product combinations selected based on these statistical characteristics have higher information quality.

[0101] Furthermore, this application can be applied to independent websites of e-commerce platforms, taking into account the relatively small amount of historical data on a single independent website. It identifies a list of high-quality product combinations to provide product combination recommendations to users visiting the independent website, enabling a cold start for product recommendations and thus possessing high practical value.

[0102] Based on any embodiment of this application, please refer to Figure 3 Retrieve each user's product set, which contains multiple product items that the user has visited. Combine the product items in the product set with any number of items, enumerate all product combinations to form a user combination set, and then deduplicate and merge all user combination sets to form the full combination set, including:

[0103] Step S1110: Obtain access behavior data generated by each user in the same online store in historical access events, and determine from the access behavior data that multiple product items accessed by the user constitute the user's product set. The historical access events include events of adding product items to the shopping cart and / or adding product items to the purchase order.

[0104] In this embodiment, the primary consideration is to collect user browsing behavior data based on the same online store, so that the resulting list of high-quality combinations is more suitable for recommending products on the independent website where the online store is located. This is because, when online stores are located across different independent websites, the products they sell may differ significantly, and there may be a lack of necessary correlation between the behavioral data related to these product items. Therefore, limiting the collection of browsing behavior data to the same online store helps ensure that the resulting list of high-quality combinations is more targeted.

[0105] To this end, access behavior data corresponding to a certain access behavior event generated by the online store within a specific time range can be obtained from the user behavior database of the independent website where the online store is located. Then, the access behavior data of each user is parsed to extract the access time and product items. Then, on a user-by-user basis, all the product items that each user has historically accessed are constructed into the same product set. Preferably, the product items in the product set are sorted according to the access time.

[0106] In this application embodiment, the access behavior event is recommended to be the event of adding a product item to the shopping cart and / or the event of adding a product item to the purchase order. The former is mainly because it shows the user's strong desire to purchase the corresponding product item, and the latter is mainly because it shows the user's actual action of making a purchase. It can be seen that both can show the user's need to purchase the corresponding product item, and thus have information reference value as the basic data required.

[0107] Step S1120: For each user's product set, combine the product items in the set according to any number of product items, enumerate all possible product combinations in the product set, and form a user combination set;

[0108] Each user's product set may contain a massive number of product items. By combining any two or more product items of any number, a product combination can be constructed. Following this principle, by enumerating the product combinations of any number of product items in each user's product set, all product combinations formed by combining all product items in each user's product set can be determined, thus forming the user's user combination set.

[0109] Step S1130: Deduplicate and merge the product combinations in the user combination set of all users to construct the full combination set.

[0110] Although there will be no duplicate product combinations in each user combination set, there may be duplicate product combinations in different user combination sets. Considering the need to evaluate each unique product combination, a full combination set can be constructed. All product combinations from all users' user combination sets are added to this full combination set, and for completely identical product combinations, only one of them is retained to achieve deduplication, so that each product combination in the full combination set is unique.

[0111] The above embodiments enable in-depth data processing of access behavior data of all users of the same online store, constructing a user combination set and a full combination set. The user combination set includes an enumeration of all possible product combinations in the historical behavior data, and the number of product items in the product combination is not subject to additional restrictions. This allows for orderly and efficient computational processing when evaluating product combinations in the full combination set, achieving a fast and efficient construction of a high-quality combination list with the optimal amount of computation.

[0112] Based on any embodiment of this application, please refer to Figure 4 The cohesion of a product combination is determined based on the mutual information entropy obtained after arbitrarily bisecting each product combination in the full set of combinations, including:

[0113] Step S1310: For each product combination in the full set of combinations, divide the product combination into two possible parts by product item to obtain the left set and right set corresponding to each division;

[0114] Each product combination in the full set of combinations can be processed individually, and the specific steps of this embodiment can be performed to determine its corresponding cohesion.

[0115] For a product combination, which consists of multiple product items, each product item in the combination can be represented as a PID (Product ID) corresponding to that item. Based on this, the product combination can be bisected along any one product item boundary, dividing it into left and right parts to obtain a left set and a right set, where both the left and right sets contain at least one product item (i.e., they are not empty sets). In other words, for the product combination, this step exhaustively enumerates all possible bisecting schemes, obtaining the left and right sets corresponding to each partition.

[0116] Step S1320: Calculate the mutual information entropy corresponding to each split based on the left and right sets obtained from each split;

[0117] For each segmentation of a product combination, the corresponding left and right sets can be used to calculate the mutual information entropy of the product combination based on its own occurrence probability and the occurrence probabilities of the left and right sets. The specific calculation method has been disclosed in the previous embodiments and will not be repeated here. Thus, each segmentation of a product combination can yield its corresponding mutual information entropy.

[0118] Step S1330: The lowest mutual information entropy obtained after multiple divisions of each product combination is taken as the cohesion of that product combination.

[0119] Each segmentation of each product combination yields a corresponding mutual information entropy. In this embodiment, the lowest mutual information entropy is taken as the cohesion of the product combination, representing the stability of the fixed combination of each product item in the product combination in a relatively conservative way.

[0120] As can be seen from the above embodiments, by performing multiple bisections on each product combination in the full set of combinations to determine the mutual information entropy corresponding to each bisection, and using the lowest information entropy to represent the cohesion of the product combination, it is possible to effectively represent the stability of the fixed combinations of each product item in the entire product combination. It is easy to understand that the higher the cohesion of the product combination, the more often the product items in it are used as stable product combinations, and the more stable such combinations are. Thus, based on historical behavior data, the reference information value of each product combination in terms of the stability of the combinations is effectively represented.

[0121] Based on any embodiment of this application, please refer to Figure 5 The degrees of freedom of each product combination are determined based on the adjacency entropy of each product combination relative to the product set of all users in the full set of combinations, including:

[0122] Step S1410: Sort the product items in each user's product set according to the user's access time;

[0123] Since each user's product set contains access behavior data, and this data includes the time information corresponding to each user's access, the product set can be sorted based on the user's access time, thus more systematically representing the order in which each product item was accessed. It's easy to understand that the product set may contain multiple product items forming product combinations arranged together; therefore, in reality, the product set still represents multiple product combinations based on the adjacency of the product items.

[0124] Step S1420: For each product combination in the full set of combinations, calculate its left adjacency entropy and right adjacency entropy relative to the sorted product sets of each user, and determine the lowest value as the lowest adjacency entropy.

[0125] For each product combination in the full set of combinations, its adjacency entropy can be calculated to represent the richness of combinations that the product combination is paired with other product items and accessed by users. In other words, the lower the adjacency entropy of a product combination, the fewer other product items it is paired with and accessed by users; conversely, the higher the adjacency entropy, the more often it is paired with other product items and accessed by users. The calculation methods for left and right adjacency entropy have been given above and will not be repeated here. For a product combination, after calculating its left and right adjacency entropy, in this embodiment, the lowest value is selected as the minimum adjacency entropy, used to characterize the richness of combinations of that product combination relative to the corresponding user.

[0126] Step S1430: Sum the lowest adjacency entropy of each product combination relative to all users as the degree of freedom of that product combination.

[0127] Following the above process, for all users, each product combination can obtain the minimum adjacency entropy corresponding to each user. To convert this into a representation of degrees of freedom, the minimum adjacency entropies of each user can be summed and averaged to achieve a summary. The summed value is then used as the degrees of freedom of the product combination. Accordingly, each product combination in the full set of combinations can obtain its corresponding degrees of freedom.

[0128] The above embodiments provide a specific method for calculating degrees of freedom. In the calculation process, the granularity is refined to the product set of each user. After determining the minimum adjacency entropy for each user, the degrees of freedom are then aggregated. This method can be refined to the level of each user, so that the obtained degrees of freedom can more accurately represent the richness of each product combination, and make it easier to select high-quality product combinations in the future.

[0129] Based on any embodiment of this application, multiple product combinations are selected from the full set of combinations based on the frequency of occurrence, cohesion, and degree of freedom to form a list of high-quality combinations, including:

[0130] Step S1510: Multiply the frequency of occurrence, cohesion, and degree of freedom of each product combination in the full set of combinations to obtain the comprehensive score corresponding to that product combination;

[0131] In this embodiment, for each product combination in the full set of combinations, its frequency of occurrence is used. Cohesion Degrees of freedom The results are summarized to determine the overall score for each product combination. The formula used to achieve comprehensive quantification is shown below:

[0132]

[0133] As can be seen from the above formula, by strongly correlating the frequency of occurrence, cohesion, and degree of freedom to determine the comprehensive score of the product combination, the comprehensive score can effectively represent the overall quality of the product combination, making the comparison between product combinations based on the comprehensive score more prominent and facilitating screening.

[0134] In a further embodiment, when applying the above formula, corresponding preset weights can be matched to the frequency of occurrence, cohesion, and degree of freedom as needed, which can be flexibly set by those skilled in the art as needed.

[0135] Step S1520: Based on the comprehensive score, filter the product combinations in the full set of combinations to obtain a list of high-quality combinations consisting of a portion of the product combinations with relatively high comprehensive scores.

[0136] After determining the overall score corresponding to each product combination in the full set of combinations, a selection of product combinations with higher overall scores can be chosen as high-quality product combinations to form a list of high-quality combinations.

[0137] In one embodiment, a preset screening threshold can be used to compare the comprehensive scores of all combinations in the full set with the threshold, and the combinations of products with comprehensive scores higher than the screening threshold can be regarded as high-quality product combinations.

[0138] In another embodiment, a preset number of filters can be used. First, all product combinations in the full set are sorted from largest to smallest according to the comprehensive score. Then, the corresponding number of product combinations at the top of the sorted list are selected as high-quality product combinations according to the number of filters.

[0139] As can be seen from the above embodiments, a strong correlation is formed between the frequency of occurrence, cohesion, and degree of freedom to determine the comprehensive score of each product combination. Then, the comprehensive score is used to screen high-quality product combinations. Since the comprehensive score can effectively quantify the quality of each product combination, it can ensure that the list of high-quality combinations formed by the obtained high-quality product combinations can provide effective and high-quality reference information for product recommendation services.

[0140] Based on any embodiment of this application, please refer to Figure 6 After constructing the list of high-quality combinations, it includes:

[0141] Step S1600: Respond to any access event triggered by a user and obtain the target product item corresponding to the access event;

[0142] Once the list of high-quality combinations is determined, it can be invoked by the product recommendation service. The product recommendation service can respond to any access event triggered by a user visiting the online store. Such access events include, for example, a user adding a target product item to their shopping cart, entering the product details page of the target product item, adding the target product item to their order, or completing the payment for the order containing the target product item. Thus, in response to such an access event, the corresponding target product item can be obtained.

[0143] Step S1700: Query the list of high-quality combinations to obtain all combinations of goods containing the target product item to form a candidate list;

[0144] After obtaining the target product item, the product recommendation service can query all product combinations containing the target product item from the list of high-quality combinations. These product combinations, as is easily understood, include not only the target product item but also other product items that form a habitual pairing relationship with it. For the convenience of subsequent calculations, a candidate list is constructed from all product combinations matched from the list of high-quality combinations based on the target product item, i.e., all high-quality product combinations.

[0145] Step S1800: Obtain the measured scores of each product combination in the candidate list, and determine the product combination with the highest measured score as the target product combination;

[0146] The candidate list may contain multiple product combinations. Each time a product is pushed to a user, various strategies can be used to select matching product items. In this case, it's necessary to further optimize each product combination to determine the target product combination, so that other product items can be accurately pushed based on the target combination. An exemplary strategy is to select the most effective product combination as the target product combination, using this as the basis for determining matching product items for that target product. To this end, a preset mechanism can be used to obtain the actual test scores corresponding to each product combination in the candidate list, and then the product combination with the highest actual test score can be selected as the target product combination.

[0147] Regarding the preset mechanism, in one embodiment, each product combination in the list of high-quality combinations can be randomly applied to users across the platform to perform advertising behavior in advance. Then, the conversion rate obtained after advertising these product combinations is used as their corresponding actual test score. In this case, each product combination in the candidate list can be filtered according to its corresponding actual test score, and the product combination with the highest actual test score can be determined as the target product combination.

[0148] Step S1900: Construct product information for all other product items included in the target product combination, excluding the target product item, and push it to the arbitrary user.

[0149] The target product combination determined through the above process includes the target product item and other product items that habitually pair with it. Based on this, product information for all other product items can be obtained, including but not limited to product titles, prices, images, and links. This product information is then encapsulated into a product recommendation list and pushed to any user who triggers the access event. After receiving the product recommendation list on their terminal device, the user's browser parses and encapsulates it into information controls corresponding to each product item for display. For each other product item, a product link is associated, and its image, price, title, and other information are displayed as needed. When the user touches the information control of a certain other product item, they are redirected to the corresponding product details page, effectively promoting other product items that habitually pair with the target product item.

[0150] As can be seen from the above embodiments, the list of high-quality combinations obtained in this application can provide high-quality experience data for product recommendation services, serve as the basis for advertising decisions, make product combination recommendation strategies more effective, and enable cold start advertising promotion for new users.

[0151] Please see Figure 7According to one aspect of this application, a product combination construction apparatus includes a data processing module 1100, a frequency determination module 1200, a cohesion determination module 1300, a degree of freedom determination module 1400, and a list construction module 1500. The data processing module 1100 is configured to acquire a product set for each user, which includes multiple product items visited by the user; combine the product items in the product set with an arbitrary number of items; enumerate all product combinations to form a user combination set; and deduplicate and merge all user combination sets to form a full combination set. The frequency determination module 1200 is configured to determine the frequency of each product combination in the full combination set in each user combination set. The cohesion determination module 1300... The module 1400 is configured to determine the cohesion of a product combination based on the mutual information entropy obtained after arbitrarily bisecting each product combination in the full set of combinations. The cohesion characterizes the stability of the combination of multiple product items included in the corresponding product combination being accessed by the same user. The module 1500 is configured to determine the degree of freedom of a product combination based on the adjacency entropy of each product combination in the full set of combinations relative to the product set of all users. The degree of freedom characterizes the richness of the combination of the corresponding product combination with other product items being accessed by the same user. The module 1500 is configured to filter out multiple product combinations in the full set of combinations based on the frequency of occurrence, cohesion, and degree of freedom to form a list of high-quality combinations.

[0152] Based on any embodiment of this application, the data processing module 1100 includes: a behavior data retrieval unit, configured to acquire access behavior data generated by each user in the same online store during historical access events, and determine from the access behavior data that multiple product items accessed by the user constitute the user's product set, wherein the historical access events include events of adding product items to the shopping cart and / or adding product items to the purchase order; a combination enumeration unit, configured to, for each user's product set, combine the product items in the product set according to any number of product items, enumerate all possible product combinations in the product set, and constitute a user combination set; and a user merging unit, configured to, deduplicate and merge the product combinations in the user combination sets of all users to construct a full combination set.

[0153] Based on any embodiment of this application, the cohesion determination module 1300 includes: a combination segmentation unit, configured to, for each product combination in the full set of combinations, arbitrarily bisegment the product combination by product item to obtain the left set and right set corresponding to each segmentation; an information quantification unit, configured to calculate the mutual information entropy corresponding to each segmentation based on the left set and right set obtained from each segmentation; and a cohesion selection unit, configured to take the lowest mutual information entropy obtained after multiple segmentations of each product combination as the cohesion of the product combination.

[0154] Based on any embodiment of this application, the degree of freedom determination module 1400 includes: a time sorting unit, configured to sort the product items in each user's product set according to the user's access time; an adjacency entropy calculation unit, configured to calculate the left adjacency entropy and right adjacency entropy of each product combination relative to the sorted product sets of each user for each product combination in the full combination set, and determine the lowest value as the minimum adjacency entropy; and a degree of freedom summarization unit, configured to summarize the minimum adjacency entropy of each product combination relative to all users as the degree of freedom of that product combination.

[0155] Based on any embodiment of this application, the list construction module 1500 includes: a fusion scoring unit, configured to multiply the frequency of occurrence, cohesion, and degree of freedom of each product combination in the full set of combinations as the comprehensive score corresponding to that product combination; and a screening construction unit, configured to screen the product combinations in the full set of combinations based on the comprehensive score, and obtain a list of high-quality combinations consisting of a portion of product combinations with relatively high comprehensive scores.

[0156] Based on any embodiment of this application, the product combination construction apparatus of this application further includes: an access response module, configured to respond to an access event triggered by any user and obtain the target product item corresponding to the access event; a combination query module, configured to query all product combinations containing the target product item from the list of high-quality combinations to form a candidate list; a reference selection module, configured to obtain the actual test scores of each product combination in the candidate list and determine the product combination with the highest actual test score as the target product combination; and a result push module, configured to construct and push product information of all other product items included in the target product combination except for the target product item to the arbitrary user.

[0157] Another embodiment of this application also provides a product combination construction device. For example... Figure 8 The diagram shows the internal structure of a product combination construction device. This device includes a processor, a computer-readable storage medium, a memory, and a network interface connected via a system bus. The computer-readable, non-volatile storage medium stores an operating system, a database, and computer-readable instructions. The database stores information sequences, and when executed by the processor, these computer-readable instructions enable the processor to implement a product combination construction method.

[0158] The processor of the product combination construction device provides computing and control capabilities to support the operation of the entire product combination construction device. The memory of the product combination construction device may store computer-readable instructions, which, when executed by the processor, cause the processor to perform the product combination construction method of this application. The network interface of the product combination construction device is used for communication with a terminal.

[0159] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the product combination construction device to which the present application is applied. A specific product combination construction device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0160] In this embodiment, the processor is used to execute... Figure 7 The specific functions of each module are described, and the memory stores the program code and various data required to execute the aforementioned modules or sub-modules. A network interface is used to enable data transmission between user terminals or servers. In this embodiment, the non-volatile readable storage medium stores the program code and data required to execute all modules in the product combination construction device of this application. The server can call the server's program code and data to execute the functions of all modules.

[0161] This application also provides a non-volatile readable storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the product combination construction method of any embodiment of this application.

[0162] This application also provides a computer program product, including a computer program / instructions that, when executed by one or more processors, implement the steps of the method described in any embodiment of this application.

[0163] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a computer-readable storage medium such as a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM).

[0164] In summary, this application selects high-quality product combinations based on the frequency of occurrence, cohesion, and freedom of the product combinations. This ensures that the selected product combinations more accurately reflect the relationships between multiple product items, and there is no limit to the number of product items in the same product combination. This provides accurate product combination information for independent websites, resulting in a higher success rate for product recommendations.

Claims

1. A method for constructing a commodity combination, characterized in that, include: Get each user's product set, which contains multiple product items that the user has visited. Combine the product items in the product set with any number of items, enumerate all product combinations to form a user combination set, and then merge all user combination sets after removing duplicates to form the full combination set. Determine the frequency of occurrence of each product combination in the full set of combinations across all user sets; The cohesion of a product combination is determined based on the mutual information entropy obtained after arbitrarily bisecting each product combination in the full set of combinations. The cohesion characterizes the stability of the association between multiple product items in the corresponding product combination and access by the same user. The determination includes: for each product combination in the full set of combinations, bisecting the product combination arbitrarily by product item to obtain the left and right sets corresponding to each bisecting; calculating the mutual information entropy corresponding to each bisecting based on the left and right sets obtained for each bisecting; and taking the lowest mutual information entropy obtained after multiple bisectings for each product combination as the cohesion of that product combination. The degrees of freedom of each product combination are determined based on the adjacency entropy of each product combination relative to the product sets of all users in the full set of combinations. The degrees of freedom represent the richness of combinations of the corresponding product combination with other product items and accessed by the same user. This includes: sorting the product items in each user's product set according to the user's access time; for each product combination in the full set of combinations, calculating its left adjacency entropy and right adjacency entropy relative to the sorted product sets of each user, and determining the lowest value as the minimum adjacency entropy; summing the minimum adjacency entropy of each product combination relative to all users as the degrees of freedom of that product combination. Based on the frequency of occurrence, cohesion, and degree of freedom, multiple product combinations are selected from the full set of combinations to form a list of high-quality combinations.

2. The product combination construction method according to claim 1, characterized in that, Retrieve each user's product set, which contains multiple product items that the user has visited. Combine the product items in the product set with any number of items, enumerate all product combinations to form a user combination set, and then deduplicate and merge all user combination sets to form the full combination set, including: Obtain access behavior data generated by each user in the same online store in historical access events, and determine from the access behavior data that multiple product items accessed by the user constitute the user's product set. The historical access events include events of adding product items to the shopping cart and / or adding product items to the purchase order. For each user's product set, the product items are combined according to any number of product items, and all possible product combinations in the product set are enumerated to form the user's combination set; The product combinations in the user combination set of all users are deduplicated and merged to construct the full combination set.

3. The product combination construction method according to claim 1 or 2, characterized in that, Based on the frequency of occurrence, cohesion, and degree of freedom, multiple product combinations are selected from the full set of combinations to form a list of high-quality combinations, including: The frequency of occurrence, cohesion, and degree of freedom of each product combination in the full set of combinations are multiplied together to obtain the comprehensive score corresponding to that product combination; Based on the comprehensive score, the product combinations in the full set are screened to obtain a list of high-quality product combinations with relatively high comprehensive scores.

4. The product combination construction method according to claim 1 or 2, characterized in that, After compiling the list of high-quality portfolios, it includes: Respond to any access event triggered by a user and retrieve the target product item corresponding to that access event; A candidate list is formed by retrieving all product combinations containing the target product item from the list of high-quality combinations. Obtain the measured scores of each product combination in the candidate list, and determine the product combination with the highest measured score as the target product combination; The product information of all other product items included in the target product combination, excluding the target product item, is pushed to the arbitrary user.

5. A product combination construction device, characterized in that, include: The data processing module is set to obtain each user's product set, which contains multiple product items that the user has visited, combine the product items in the product set with any number of items, enumerate all product combinations to form a user combination set, and then deduplicate and merge all user combination sets to form a full combination set. The frequency determination module is set to determine the frequency of occurrence of each product combination in the full set of combinations in each user combination set. The cohesion determination module is configured to determine the cohesion of a product combination based on the mutual information entropy obtained after arbitrarily bisecting each product combination in the full set of combinations. The cohesion characterizes the stability of the pairing of multiple product items included in the corresponding product combination when they are accessed by the same user. The module includes: for each product combination in the full set of combinations, arbitrarily bisecting the product combination by product item to obtain the left and right sets corresponding to each bisecting; calculating the mutual information entropy corresponding to each bisecting based on the left and right sets obtained for each bisecting; and taking the lowest mutual information entropy obtained after multiple bisectings for each product combination as the cohesion of that product combination. The degree of freedom determination module is configured to determine the degree of freedom of each product combination based on the adjacency entropy of each product combination in the full set of combinations relative to the product sets of all users. The degree of freedom represents the richness of combinations of the corresponding product combination with other product items and accessed by the same user. This includes: sorting the product items in each user's product set according to the user's access time; for each product combination in the full set of combinations, calculating its left adjacency entropy and right adjacency entropy relative to the sorted product sets of each user, and determining the lowest value as the minimum adjacency entropy; summing the minimum adjacency entropy of each product combination relative to all users as the degree of freedom of that product combination. The list construction module is configured to filter multiple product combinations from the full set of combinations based on the frequency of occurrence, cohesion, and degree of freedom, thereby forming a list of high-quality combinations.

6. The product combination construction device according to claim 5, characterized in that, The data processing module includes: The behavior data retrieval unit is configured to obtain access behavior data generated by each user in the same online store in historical access events, and determine from the access behavior data that multiple product items accessed by the user constitute the user's product set. The historical access events include events of adding product items to the shopping cart and / or adding product items to the purchase order. The combination enumeration unit is set up for each user's product set, and the product items in the set are combined according to any number of product items. All possible product combinations in the product set are enumerated to form the user combination set. The user merging unit is configured to deduplicate and merge the product combinations in the user combination set of all users to construct a full combination set.

7. The product combination construction device according to claim 5 or 6, characterized in that, This device also includes: The access response module is configured to respond to access events triggered by any user and retrieve the target product item corresponding to that access event. The combination query module is configured to retrieve a candidate list of all product combinations containing the target product item from the list of high-quality combinations. The reference selection module is set to obtain the actual scores of each product combination in the candidate list and determine the product combination with the highest actual score as the target product combination. The result push module is configured to push product information of all other product items included in the target product combination, excluding the target product item, to the arbitrary user.

8. A product combination construction device, comprising a central processing unit and a memory, characterized in that, The central processing unit is configured to invoke and run a computer program stored in the memory to perform the steps of the method as described in any one of claims 1 to 4.

9. A non-volatile readable storage medium, characterized in that, It stores a computer program in the form of computer-readable instructions, which, when invoked by a computer, performs the steps included in the method as described in any one of claims 1 to 4.

10. A computer program product, characterized in that, Includes a computer program / instruction, which, when executed by a processor, performs the steps of the method as described in any one of claims 1 to 4.

Images