An item recommendation method, apparatus, device, and medium
By acquiring item sequences and utilizing multidimensional deep learning models and Pareto optimization algorithms, the multidimensional similarity of item combinations is calculated, solving the problem of insufficient accuracy in item recommendation in existing technologies and achieving higher recommendation accuracy and user experience.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DATAGRAND TECH INC
- Filing Date
- 2026-03-30
- Publication Date
- 2026-06-19
AI Technical Summary
In existing technologies, the feature extraction and evaluation methods of single paths cannot fully and accurately express the degree of similarity between items, resulting in low accuracy of item recommendations and difficulty in meeting the actual needs of users.
By acquiring multiple sets of item sequences, a multidimensional deep learning model and Pareto optimization algorithm are used to calculate the multidimensional similarity estimate of the item combination, and multi-objective optimization is performed to determine the optimal similarity of the item combination, so as to improve the effectiveness and accuracy of the similarity.
It improves the accuracy of item recommendations, accurately expresses the similarity between items, enhances the user experience, and adapts to multiple scenarios and business needs.
Smart Images
Figure CN122243615A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of big data technology, and in particular to a method, apparatus, device, and medium for recommending items. Background Technology
[0002] Item recommendation refers to the intelligent technology that various digital systems use to proactively match and present relevant items to users, simplifying the process of finding, accessing, and operating items.
[0003] Currently, the similarity between items is generally determined by a single-path similarity calculation method. Specifically, the single-path similarity calculation method is based on the frequency of user behavior, the basic attributes of items, or a single temporal feature to build a similarity calculation model, and then obtains the similarity result between items through feature extraction and calculation of a single path.
[0004] However, the feature extraction and evaluation method of a single path cannot fully and accurately express the degree of similarity between items. It can only mine the single-dimensional features of item association and cannot take into account the item association features of different dimensions. This can easily lead to insufficient validity of the calculated item similarity results, resulting in low accuracy of associated item recommendations and difficulty in meeting the actual needs of users. Summary of the Invention
[0005] This invention provides a method, apparatus, device, and medium for recommending items, which can effectively improve the effectiveness of similarity between items, enabling the similarity to accurately express the degree of similarity between items, thereby improving the accuracy of item recommendations and enhancing the user experience.
[0006] According to one aspect of the present invention, an item recommendation method is provided, comprising: Based on user behavior data, obtain multiple sets of item sequences, and based on the item sequences, obtain item pair information; Based on the item sequence, item pair information, and pre-established multidimensional deep learning model, obtain multidimensional similarity estimates for each item combination; Based on the Pareto optimization algorithm, the multidimensional similarity estimate of the item combination is optimized in multiple objectives to determine the optimal similarity of the item combination; When a user interacts with a target item, multiple combinations of target items corresponding to the target item are identified, and at least one recommended item is determined based on the optimal similarity of each combination of target items.
[0007] According to another aspect of the present invention, an item recommendation device is provided, comprising: The item pair information acquisition module is used to acquire multiple sets of item sequences based on user behavior data, and to acquire item pair information based on the item sequences; The similarity estimation module is used to obtain the multidimensional similarity estimate of each item combination based on the item sequence, item pair information and the pre-established multidimensional deep learning model. The optimal similarity acquisition module is used to perform multi-objective optimization on the multi-dimensional similarity estimate of the item combination based on the Pareto optimization algorithm, and determine the optimal similarity of the item combination. The recommended item determination module is used to determine multiple combinations of target items corresponding to the target item when a user interacts with the target item, and to determine at least one recommended item based on the optimal similarity of each combination of target items.
[0008] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the item recommendation method according to any embodiment of the present invention.
[0009] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the item recommendation method according to any embodiment of the present invention.
[0010] The technical solution of this invention obtains multiple sets of item sequences based on user behavior data, and obtains item pair information based on the item sequences; obtains multidimensional similarity estimates for each item combination based on the item sequences, item pair information, and a pre-established multidimensional deep learning model; performs multi-objective optimization on the multidimensional similarity estimates of the item combinations using the Pareto optimization algorithm to determine the optimal similarity of the item combinations; when a user interacts with a target item, it determines multiple target item combinations corresponding to the target item, and determines at least one recommended item based on the optimal similarity of each target item combination. This approach effectively improves the effectiveness of similarity between items, enabling the similarity to accurately express the degree of similarity between items, aligning with actual user operation needs, thereby improving the accuracy of item recommendations, enhancing the user experience, and adapting to multiple scenarios and business requirements.
[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description
[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0013] Figure 1 This is a flowchart of an item recommendation method provided according to Embodiment 1 of the present invention; Figure 2 This is a flowchart of another item recommendation method provided according to Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of the structure of an item recommendation device according to Embodiment 3 of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device that implements the item recommendation method of this invention. Detailed Implementation
[0014] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.
[0015] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0016] Example 1 Figure 1This is a flowchart of an item recommendation method provided in Embodiment 1 of the present invention. This embodiment is applicable to situations where item recommendations are made to users when they click, access, or operate on items in the system. This method can be executed by an item recommendation device, which can be implemented in hardware and / or software, and is generally configured in a computer or processor with data processing capabilities. Figure 1 As shown, the method includes: S110. Based on user behavior data, obtain multiple sets of item sequences, and based on the item sequences, obtain item pair information.
[0017] Optionally, user behavior data can refer to a structured data set collected in the business system of authorized users performing various interactive operations on items within a target time interval, including core data items such as user identifier, item identifier, operation type, behavior trigger time, and operating device.
[0018] Optionally, an item sequence can refer to an ordered list of item identifiers arranged according to the order in which users operate on items. The item sequence is divided according to both user identifier and operation type. That is, different operation types of the same user correspond to independent item sequences. For example, the click item sequence and the add item sequence of user A are two independent sequences. The sequence retains the sequential characteristics of the user's operation on items.
[0019] Optionally, item pair information can refer to the combination information of two items generated by Cartesian set operations based on the item sequence. Item pair information can include key information such as item pair identifier, operation type, and the behavior trigger time of each of the two items.
[0020] Before obtaining multiple sets of item sequences based on user behavior data, and before obtaining item pair information based on the item sequences, the process may further include: In the business system, collect the item operation records of each authorized user within the target time interval; Based on multiple pre-specified data items, the validity of each item operation record is identified, and user behavior data is generated based on the valid item operation records.
[0021] Optionally, an authorized user can refer to an operator who has registered in the system and has granted the system permission to collect item operation records.
[0022] Optionally, the item operation record can refer to the original, unfiltered record of user operations on items in the business system, which may include information such as the authorized user's user ID, item ID, behavior trigger time, operation type, operation device, and operation duration.
[0023] Optionally, a data acquisition module can be set up in the business system, pre-define the target time interval and the acquisition range, and then, based on the user authorization agreement, collect the original records of all authorized users' item operations within the target time interval, and store the collected original records in a structured manner.
[0024] Optionally, data items can refer to data items that should be included in the pre-defined item operation record.
[0025] Optionally, the validity of each item operation record can be identified based on multiple pre-specified data items. This may include: checking whether the item operation record contains all the pre-specified data items and whether each data item corresponds to a valid value. If so, the item operation record is determined to be a valid item operation record; otherwise, the item operation record is determined to be an invalid operation record and the item operation record is deleted.
[0026] This includes obtaining multiple sets of item sequences based on user behavior data, and obtaining item pair information based on the item sequences, which may include: Based on user identifiers and behavior types, extract item sequences corresponding to different operation types for each user from user behavior data; For each item sequence of the same operation type, a Cartesian set is generated to obtain the item pairs of that operation type, and the behavior trigger time of each item pair is determined based on user behavior data.
[0027] Optionally, user behavior data can be grouped using user identifier as the primary grouping condition and operation type as the secondary grouping condition to obtain a subset of item operation data for each user corresponding to each operation type. Then, each data subset is sorted in ascending order according to the behavior trigger time, and the item identifier is extracted according to the sorting result to form an ordered list. This list is the item sequence for the user corresponding to the operation type.
[0028] In an optional example, taking user 1 as an example, if the operation type includes click, add to cart, and purchase, then click item sequence [item A, item B, item C, item A, item D], add item sequence [item B, item A, item D], and purchase item sequence [item A, item B] can be generated for user 1 respectively. For multiple authorized users, each authorized user can generate multi-operation type item sequences as shown in user 1. This is only an example for illustration.
[0029] Furthermore, the entire sequence of items can be grouped according to the operation type to obtain a set of item sequences under each operation type. Cartesian set operations can be performed on each item sequence under the same operation type to generate ordered item pairs corresponding to the sequence. All item pairs under the same operation type can be summarized to form a set of item pairs for that operation type.
[0030] Optionally, generating a Cartesian set refers to generating item pairs based on the principles of a unique set of items within a sequence, generating ordered item pairs, and excluding pairings of items themselves. In an optional example, for the item sequence [item B, item A, item D], the generated Cartesian set includes (item B, item A), (item B, item D), (item A, item B), (item A, item D), (item D, item B), and (item D, item A). Each element in the Cartesian set can be regarded as an item pair.
[0031] Optionally, the behavior trigger time can refer to the specific time when a user initiates an interactive operation such as clicking, adding to cart, or purchasing an item. The behavior trigger time of an item pair includes the behavior trigger time corresponding to each of the two items in the item pair.
[0032] S120. Based on the item sequence, item pair information, and the pre-established multidimensional deep learning model, obtain the multidimensional similarity estimate of each item combination.
[0033] Optionally, the multidimensional deep learning model may include an estimation model based on a deep neural network, an estimation model based on an attention mechanism, and an estimation model based on a recurrent neural network. The three models process the item sequence in parallel and output similarity estimates of different dimensions, thereby realizing multidimensional mining of deep features, key behavioral features, and temporal dependency features of the item sequence.
[0034] Optionally, multidimensional similarity estimates can refer to multiple similarity values of a combination of items obtained through different calculation methods or models.
[0035] The process of obtaining multidimensional similarity estimates for each item combination based on the item sequence, item pair information, and a pre-established multidimensional deep learning model may include: Based on the item pair information, calculate the similarity score of each item pair, and determine the first similarity estimate of each item combination based on the similarity score of each item pair. The item sequence is input into an estimation model based on a deep neural network to obtain a second similarity estimate for each item combination; The item sequence is input into an attention-based estimation model to obtain the third similarity estimate of each item combination; The item sequence is input into an estimation model based on a recurrent neural network to obtain the fourth similarity estimate of each item combination.
[0036] Optionally, the similarity score of each item pair can be obtained by calculating multiple influencing factors such as the behavior trigger time, item behavior weight, and time difference. Then, the similarity scores of the set of similar item pairs can be aggregated, weighted averaged, and normalized to obtain the first similarity estimate of each item combination. The first similarity estimate can characterize the temporal association and behavioral value characteristics of the item pairs.
[0037] Optionally, inputting the item sequence into a deep neural network-based estimation model to obtain a second similarity estimate for each item combination may include: encoding the item sequence into feature-defined vector features that the model can recognize; inputting the vector features into a pre-trained deep neural network-based estimation model, performing deep mining and nonlinear mapping on the sequence features through multiple hidden layers, and extracting the hidden layer weights as item vector features; and then mapping the item vector features into similarity values through the output layer, which are the second similarity estimates for each item combination. The second similarity estimate can characterize the deep semantics and association features of the item sequence.
[0038] Optionally, an attention-based estimation model can be used to obtain a third similarity estimate that can characterize the key behaviors and core value features of an item sequence.
[0039] Optionally, inputting the item sequence into an estimation model based on a recurrent neural network to obtain a fourth similarity estimate for each item combination may include: inputting the encoded item sequence into a pre-trained estimation model based on a recurrent neural network to obtain the temporal features of the items; wherein, the temporal features include sequential dependencies and temporal associations; and mapping the temporal features into similarity values through the output layer of the recurrent neural network, which are the fourth similarity estimates for each item combination. The fourth similarity estimate can characterize the temporal dependencies and behavioral trend features of the item sequence.
[0040] S130. Based on the Pareto optimization algorithm, perform multi-objective optimization on the multidimensional similarity estimate of the item combination to determine the optimal similarity of the item combination.
[0041] Optionally, the Pareto optimization algorithm is an algorithm that finds the Pareto optimal solution among multiple conflicting optimization objectives. By using the Pareto optimization algorithm to perform multi-objective optimization on the multi-dimensional similarity estimate, the final output is the optimal similarity of the item combination that takes into account the features of each dimension, which can solve the problem of insufficient expression of single-dimensional similarity.
[0042] Optionally, the multidimensional similarity estimates of the item combination are input into the Pareto optimization algorithm. The similarity estimates of each dimension are used as independent optimization objectives. The Pareto optimization criterion is used to optimize and solve the multi-objective problem. The similarity of one dimension is used as the core optimization objective, and the similarity of the other dimensions is used as the constraint condition. The Pareto optimal solution is selected and determined as the optimal similarity of the item combination. The Pareto optimization algorithm can take into account the feature advantages of the similarity of each dimension and improve the effectiveness of the similarity expression.
[0043] Optionally, optimal similarity can refer to the final similarity value of the item combination obtained by performing multi-objective optimization on the multi-dimensional similarity estimate through the Pareto optimization algorithm. Optimal similarity takes into account the characteristics of similarity in each dimension.
[0044] S140. When a user interacts with a target item, determine multiple combinations of target items corresponding to the target item, and determine at least one recommended item based on the optimal similarity of each combination of target items.
[0045] Optionally, the target item can refer to the item that the user is currently interacting with. It is the benchmark item for recommendations, and the corresponding combination of target items needs to be matched based on this item to complete the recommendation.
[0046] Optionally, the system can monitor users' item interaction behavior in real time. When a user's interaction behavior such as clicking, adding to cart, or purchasing a target item is detected, the system identifies all target item combinations containing the target item from a pre-calculated pool of multiple item combinations, using the target item as a benchmark. The system extracts the optimal similarity of each target item combination, sorts the target item combinations in descending order of optimal similarity, and selects at least one item from the sorted results as a recommended item based on a pre-set recommendation quantity and / or recommendation threshold, and pushes it to the current user.
[0047] The technical solution of this invention obtains multiple sets of item sequences based on user behavior data, and obtains item pair information based on the item sequences; obtains multidimensional similarity estimates for each item combination based on the item sequences, item pair information, and a pre-established multidimensional deep learning model; performs multi-objective optimization on the multidimensional similarity estimates of the item combinations using the Pareto optimization algorithm to determine the optimal similarity of the item combinations; when a user interacts with a target item, it determines multiple target item combinations corresponding to the target item, and determines at least one recommended item based on the optimal similarity of each target item combination. This approach effectively improves the effectiveness of similarity between items, enabling the similarity to accurately express the degree of similarity between items, aligning with actual user operation needs, thereby improving the accuracy of item recommendations, enhancing the user experience, and adapting to multiple scenarios and business requirements.
[0048] Example 2 Figure 2 This is a flowchart of an item recommendation method provided in Embodiment 2 of the present invention. Based on the above embodiments, this embodiment specifically illustrates a method for obtaining multidimensional similarity estimates of item combinations. For example... Figure 2 As shown, the method includes: S210. Based on the user identifier and behavior type, extract the item sequence corresponding to each user's different operation type from the user behavior data.
[0049] S220. Generate Cartesian sets for each item sequence of the same operation type to obtain item pairs of that operation type, and determine the behavior trigger time of each item pair based on user behavior data.
[0050] S230. Based on the item pair information, calculate the similarity score of each item pair, and based on the similarity score of each item pair, determine the first similarity estimate of each item combination.
[0051] Calculating the similarity score for each item pair based on the item pair information may include: Based on the item pair information, determine the behavior trigger times for the first and second items in the item pair respectively; Based on the current time, the behavior trigger time of the first item, and the behavior trigger time of the second item, calculate the item behavior weight of the first item, the item behavior weight of the second item, and the time difference between the item pairs, respectively. The initial similarity score of the item pair is calculated based on the item behavior weight of the first item, the item behavior weight of the second item, and the time difference between the item pairs. Based on the item behavior weight of the second item, the initial similarity score of the item pair is updated, and the updated similarity score of the item pair is obtained.
[0052] Optionally, the first item and the second item are two items belonging to the same item pair. The first item refers to the item that comes first in the item pair, and the second item refers to the item that comes last in the item pair. The time when the item's behavior is triggered can be represented by a timestamp, which can be directly used in subsequent calculations.
[0053] Optionally, the action weight of taking the first item can be calculated based on the current time and the action trigger time of the first item; the action weight of the second item can be calculated based on the current time and the action trigger time of the second item; and the time difference between the actions of the first item and the second item can be calculated based on the action trigger times of the second item.
[0054] Optionally, the behavior weight of an item can be calculated based on the current time and the time when the item's behavior was triggered, using the following formula: ; Where time_weight is the behavior weight of the item being calculated at the current time, t now It is the timestamp of the current calculation time, t a It represents the current time of the item's action trigger. ReLU is a linear rectified function. , that is, between 0 and The maximum value among them is taken, and the item being calculated can be either the first or second item participating in the current calculation.
[0055] Optionally, the time difference between the item pairs can be calculated based on the trigger times of the actions of the first and second items, using the following formula: ; Where diff_ratio is the time difference between the item pairs, t1 is the trigger time of the first item's behavior, and t2 is the trigger time of the second item's behavior.
[0056] Optionally, the time difference between the items is actually calculated using the Sigmoid function, which calculates the difference between the action trigger time of the first item and the action trigger time of the second item. The Sigmoid function can normalize the time difference. The value of the time difference is in the range of 0-1. The larger the value, the closer the operation time of the two items is.
[0057] Optionally, the initial similarity score of the item pair can be calculated based on the item behavior weight of the first item, the item behavior weight of the second item, and the time difference between the item pairs. This can be achieved using the following formula: join_weight=(time_weight1+time_weight2)×diff_ratio; Where join_weight is the initial similarity score of the item pair, time_weight1 is the behavior weight of the first item, time_weight2 is the behavior weight of the second item, and diff_ratio is the time difference between the item pairs containing the first and second items.
[0058] Optionally, the initial similarity score of the item pair can be updated based on the item behavior weight of the second item, and the updated similarity score of the item pair can be obtained. This can be achieved using the following formula: ; Where join_weight' is the similarity score of the updated item pair, join_weight is the initial similarity score of the item pair before the update, and time_weight2 is the behavior weight of the second item.
[0059] The similarity score calculated in the above manner integrates the time decay behavior weight with the time proximity difference factor, achieving a refined calculation of item similarity. By normalizing and non-linearly processing the values using functions such as ReLU and Sigmoid, the influence of excessively large or small values on the calculation results is avoided. At the same time, the initial score is updated by using the second item behavior weight, further strengthening the influence of item behavior value on similarity and improving the accuracy of the similarity score.
[0060] The process of determining the first similarity estimate for each item combination based on the similarity scores of each item pair may include: Identify sets of similar item pairs within each item pair, and determine all similarity scores for item combinations corresponding to the sets of similar item pairs based on the similarity scores of each item pair in the sets of similar item pairs. The first similarity estimate of the item combination is obtained by weighting and normalizing all similarity scores.
[0061] Optionally, the set of identical item pairs can refer to the set of all item pairs in an item combination whose first and second item identifiers are completely identical and in the same order. All item pairs can be grouped according to their item pair identifiers, and all item pairs with completely identical item pair identifiers can be divided into a set, which is the set of identical item pairs. Each set of identical item pairs uniquely corresponds to an item combination. For example, all item pairs of the type (item A, item B) can form a set of identical item pairs, and the item combination corresponding to this set of item pairs is item A + item B.
[0062] Optionally, the similarity scores of all item pairs in each set of similar item pairs are extracted, and these scores are used as the total similarity scores of the corresponding item combination. The weighted average of all similarity scores is calculated, and the weight can be determined according to the behavior trigger frequency of each item. The higher the behavior trigger frequency, the greater the weight. The average similarity score of the item combination is then obtained, and the average similarity score is normalized to obtain the first similarity estimate of the item combination.
[0063] S240. Input the item sequence into the estimation model based on a deep neural network to obtain the second similarity estimate of each item combination.
[0064] S250. Input the item sequence into the attention-based estimation model to obtain the third similarity estimate of each item combination.
[0065] The process of inputting the item sequence into an attention-based estimation model to obtain a third similarity estimate for each item combination may include: By using an attention-based estimation model, linear feature mapping is performed on each target item in the item sequence to obtain the query vector, key vector, and information vector corresponding to each target item. Based on the query vector and key vector of the target item, calculate the attention weight of the target item, and based on the attention weight and information vector, calculate the attention feature vector of the target item. The attention feature vectors of each target item in the item sequence are concatenated to generate the attention fusion feature of the item sequence, and the third similarity estimate of each item combination is obtained based on the attention fusion feature of each item sequence.
[0066] Optionally, the encoded item sequence can be input into an attention-based estimation model. The attention-based estimation model can perform linear feature mapping on each target item element in the sequence, that is, calculate the query vector, key vector and information vector corresponding to the item through three pre-trained weight matrices.
[0067] Optionally, the attention weight of the target item is calculated based on the query vector and key vector of the target item, and the attention feature vector of the target item is calculated based on the attention weight and information vector. This may include: performing an inner product operation on the query vector of the target item and the key vectors of the other items in the item sequence to obtain the attention score of the target item on the other items; dividing the attention score by the square root of the dimension of the key vector to obtain the attention weight; performing a weighted product of the attention weight of the target item and the information vector, and then summing all the product results to obtain the attention feature vector of the target item.
[0068] Optionally, the attention feature vectors of all items in the item sequence are concatenated, and the concatenated vectors are dimensionally compressed and feature fused to obtain the attention fusion feature of the item sequence. This feature retains the key behavioral information of high-weight items in the sequence and weakens the noise information of low-weight items.
[0069] Optionally, the attention fusion features can be input into the fully connected output layer of the model, and the fusion features can be mapped to similarity values in the 0-1 range through linear transformation and activation function; the values are then bound to the corresponding item combinations to obtain the third similarity estimate of each item combination.
[0070] S260. Input the item sequence into the estimation model based on the recurrent neural network to obtain the fourth similarity estimate of each item combination.
[0071] S270. Based on the Pareto optimization algorithm, perform multi-objective optimization on the multidimensional similarity estimate of the item combination to determine the optimal similarity of the item combination.
[0072] S280. When a user interacts with a target item, determine multiple combinations of target items corresponding to the target item, and determine at least one recommended item based on the optimal similarity of each combination of target items.
[0073] The technical solution of this invention obtains multiple sets of item sequences based on user behavior data, and obtains item pair information based on the item sequences; obtains multidimensional similarity estimates for each item combination based on the item sequences, item pair information, and a pre-established multidimensional deep learning model; performs multi-objective optimization on the multidimensional similarity estimates of the item combinations using the Pareto optimization algorithm to determine the optimal similarity of the item combinations; when a user interacts with a target item, it determines multiple target item combinations corresponding to the target item, and determines at least one recommended item based on the optimal similarity of each target item combination. This approach effectively improves the effectiveness of similarity between items, enabling the similarity to accurately express the degree of similarity between items, aligning with actual user operation needs, thereby improving the accuracy of item recommendations, enhancing the user experience, and adapting to multiple scenarios and business requirements.
[0074] Example 3 Figure 3 This is a schematic diagram of the structure of an item recommendation device provided in Embodiment 3 of the present invention. Figure 3 As shown, the device includes: an item pair information acquisition module 310, a similarity estimation module 320, an optimal similarity acquisition module 330, and a recommended item determination module 340.
[0075] The item pair information acquisition module 310 is used to acquire multiple sets of item sequences based on user behavior data, and acquire item pair information based on the item sequences.
[0076] The similarity estimation module 320 is used to obtain the multidimensional similarity estimate of each item combination based on the item sequence, item pair information and the pre-established multidimensional deep learning model.
[0077] The optimal similarity acquisition module 330 is used to perform multi-objective optimization on the multidimensional similarity estimate of the item combination according to the Pareto optimization algorithm, and determine the optimal similarity of the item combination.
[0078] The recommended item determination module 340 is used to determine multiple combinations of target items corresponding to the target item when the user interacts with the target item, and to determine at least one recommended item based on the optimal similarity of each combination of target items.
[0079] The technical solution of this invention obtains multiple sets of item sequences based on user behavior data, and obtains item pair information based on the item sequences; obtains multidimensional similarity estimates for each item combination based on the item sequences, item pair information, and a pre-established multidimensional deep learning model; performs multi-objective optimization on the multidimensional similarity estimates of the item combinations using the Pareto optimization algorithm to determine the optimal similarity of the item combinations; when a user interacts with a target item, it determines multiple target item combinations corresponding to the target item, and determines at least one recommended item based on the optimal similarity of each target item combination. This approach effectively improves the effectiveness of similarity between items, enabling the similarity to accurately express the degree of similarity between items, aligning with actual user operation needs, thereby improving the accuracy of item recommendations, enhancing the user experience, and adapting to multiple scenarios and business requirements.
[0080] Based on the above format example, the item pair information acquisition module 310 can be specifically used for: Based on user identifiers and behavior types, extract item sequences corresponding to different operation types for each user from user behavior data; For each item sequence of the same operation type, a Cartesian set is generated to obtain the item pairs of that operation type, and the behavior trigger time of each item pair is determined based on user behavior data.
[0081] Based on the above format example, the similarity estimation module 320 may include: The first similarity estimation value acquisition unit is used to calculate the similarity score of each item pair based on the item pair information, and determine the first similarity estimate of each item combination based on the similarity score of each item pair. The second similarity estimation unit is used to input the item sequence into an estimation model based on a deep neural network to obtain the second similarity estimation value of each item combination; The third similarity estimation unit is used to input the item sequence into the attention mechanism-based estimation model to obtain the third similarity estimation value of each item combination. The fourth similarity estimation unit is used to input the item sequence into the estimation model based on a recurrent neural network to obtain the fourth similarity estimate of each item combination.
[0082] Based on the above format example, the first similarity estimate acquisition unit can be specifically used for: Based on the item pair information, determine the behavior trigger times for the first and second items in the item pair respectively; Based on the current time, the behavior trigger time of the first item, and the behavior trigger time of the second item, calculate the item behavior weight of the first item, the item behavior weight of the second item, and the time difference between the item pairs, respectively. The initial similarity score of the item pair is calculated based on the item behavior weight of the first item, the item behavior weight of the second item, and the time difference between the item pairs. Based on the item behavior weight of the second item, the initial similarity score of the item pair is updated, and the updated similarity score of the item pair is obtained.
[0083] Based on the above format example, the first similarity estimate acquisition unit can also be specifically used for: Identify sets of similar item pairs within each item pair, and determine all similarity scores for item combinations corresponding to the sets of similar item pairs based on the similarity scores of each item pair in the sets of similar item pairs. The first similarity estimate of the item combination is obtained by weighting and normalizing all similarity scores.
[0084] Based on the above format example, the third similarity estimation unit can be specifically used for: By using an attention-based estimation model, linear feature mapping is performed on each target item in the item sequence to obtain the query vector, key vector, and information vector corresponding to each target item. Based on the query vector and key vector of the target item, calculate the attention weight of the target item, and based on the attention weight and information vector, calculate the attention feature vector of the target item. The attention feature vectors of each target item in the item sequence are concatenated to generate the attention fusion feature of the item sequence, and the third similarity estimate of each item combination is obtained based on the attention fusion feature of each item sequence.
[0085] Based on the above format example, a data acquisition module can also be included for: In the business system, collect the item operation records of each authorized user within the target time interval; Based on multiple pre-specified data items, the validity of each item operation record is identified, and user behavior data is generated based on the valid item operation records.
[0086] The item recommendation device provided in the embodiments of the present invention can execute the item recommendation method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of executing the method.
[0087] Example 4 Figure 4A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.
[0088] like Figure 4 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.
[0089] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.
[0090] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as the item recommendation method described in the embodiments of the present invention. That is: Based on user behavior data, obtain multiple sets of item sequences, and based on the item sequences, obtain item pair information; Based on the item sequence, item pair information, and pre-established multidimensional deep learning model, obtain multidimensional similarity estimates for each item combination; Based on the Pareto optimization algorithm, the multidimensional similarity estimate of the item combination is optimized in multiple objectives to determine the optimal similarity of the item combination; When a user interacts with a target item, multiple combinations of target items corresponding to the target item are identified, and at least one recommended item is determined based on the optimal similarity of each combination of target items.
[0091] In some embodiments, the item recommendation method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the item recommendation method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the item recommendation method by any other suitable means (e.g., by means of firmware).
[0092] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.
[0093] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.
[0094] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.
[0095] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).
[0096] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.
[0097] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.
[0098] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.
[0099] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.
Claims
1. An article recommendation method characterized by comprising: include: Based on user behavior data, obtain multiple sets of item sequences, and based on the item sequences, obtain item pair information; Based on the item sequence, item pair information, and pre-established multidimensional deep learning model, obtain multidimensional similarity estimates for each item combination; Based on the Pareto optimization algorithm, the multidimensional similarity estimate of the item combination is optimized in multiple objectives to determine the optimal similarity of the item combination; When a user interacts with a target item, multiple combinations of target items corresponding to the target item are identified, and at least one recommended item is determined based on the optimal similarity of each combination of target items.
2. The method of claim 1, wherein, Based on user behavior data, multiple sets of item sequences are obtained, and based on the item sequences, item pair information is obtained, including: Based on user identifiers and behavior types, extract item sequences corresponding to different operation types for each user from user behavior data; For each item sequence of the same operation type, a Cartesian set is generated to obtain the item pairs of that operation type, and the behavior trigger time of each item pair is determined based on user behavior data.
3. The method of claim 1, wherein, Based on the item sequence, item pair information, and a pre-established multidimensional deep learning model, obtain multidimensional similarity estimates for each item combination, including: Based on the item pair information, calculate the similarity score of each item pair, and determine the first similarity estimate of each item combination based on the similarity score of each item pair. The item sequence is input into an estimation model based on a deep neural network to obtain a second similarity estimate for each item combination; The item sequence is input into an attention-based estimation model to obtain the third similarity estimate of each item combination; The item sequence is input into an estimation model based on a recurrent neural network to obtain the fourth similarity estimate of each item combination.
4. The method of claim 3, wherein, Based on the item pair information, calculate the similarity score for each item pair, including: Based on the item pair information, determine the behavior trigger time for the first item and the second item in the item pair respectively; Based on the current time, the behavior trigger time of the first item, and the behavior trigger time of the second item, calculate the item behavior weight of the first item, the item behavior weight of the second item, and the time difference between the item pairs, respectively. The initial similarity score of the item pair is calculated based on the item behavior weight of the first item, the item behavior weight of the second item, and the time difference between the item pairs. Based on the item behavior weight of the second item, the initial similarity score of the item pair is updated, and the updated similarity score of the item pair is obtained.
5. The method of claim 3, wherein, Based on the similarity scores of each item pair, determine the first similarity estimate for each item combination, including: Identify sets of similar item pairs within each item pair, and determine all similarity scores for item combinations corresponding to the sets of similar item pairs based on the similarity scores of each item pair in the sets of similar item pairs. The first similarity estimate of the item combination is obtained by weighting and normalizing all similarity scores.
6. The method of claim 3, wherein, The item sequence is input into an attention-based estimation model to obtain a third similarity estimate for each item combination, including: By using an attention-based estimation model, linear feature mapping is performed on each target item in the item sequence to obtain the query vector, key vector, and information vector corresponding to each target item. Based on the query vector and key vector of the target item, calculate the attention weight of the target item, and based on the attention weight and information vector, calculate the attention feature vector of the target item. The attention feature vectors of each target item in the item sequence are concatenated to generate the attention fusion feature of the item sequence, and the third similarity estimate of each item combination is obtained based on the attention fusion feature of each item sequence.
7. The method of claim 1, wherein, Before obtaining multiple sets of item sequences based on user behavior data, and obtaining item pair information based on the item sequences, the process further includes: In the business system, collect the item operation records of each authorized user within the target time interval; Based on multiple pre-specified data items, the validity of each item operation record is identified, and user behavior data is generated based on the valid item operation records. 8.An article recommendation device characterized by comprising: include: The item pair information acquisition module is used to acquire multiple sets of item sequences based on user behavior data, and to acquire item pair information based on the item sequences; The similarity estimation module is used to obtain the multidimensional similarity estimate of each item combination based on the item sequence, item pair information and the pre-established multidimensional deep learning model. The optimal similarity acquisition module is used to perform multi-objective optimization on the multi-dimensional similarity estimate of the item combination based on the Pareto optimization algorithm, and determine the optimal similarity of the item combination. The recommended item determination module is used to determine multiple combinations of target items corresponding to the target item when a user interacts with the target item, and to determine at least one recommended item based on the optimal similarity of each combination of target items.
9. An electronic device, comprising: The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor to enable the at least one processor to perform the article recommendation method according to any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the item recommendation method according to any one of claims 1-7.