Information prediction method and device, equipment and storage medium
By using interactive behavior data and pre-trained classification network model, the user's intentions' age-group attribute information is automatically predicted, and the problem of time-consuming and labor-intensive and inaccurate judgment is solved, and efficient and accurate information prediction is achieved.
Patent Information
- Application Number
- CN202310666301.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-06
- Publication Date
- 2024-12-06
AI Technical Summary
In the prior art, the age-level information of the item that manually determines the user's intentions is time-consuming and labor-intensive, which increases labor costs and is not accurate enough.
The target item sequence and the target item vector sequence are determined based on the interactive behavior data of the target user, and the information prediction is performed using the pre-trained target classification network model to automatically predict the item age group attribute information of the user's intention.
It realizes automatic prediction of the age-level information of the item of user intention, reduces manual participation, and improves the accuracy of information prediction.
Smart Images

Figure CN119090537A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present invention relate to computer technology, and in particular to an information prediction method, apparatus, device and storage medium. Background Art
[0002] With the rapid development of Internet technology, more and more users like to shop online. In order to achieve precision marketing, it is often necessary to accurately judge the user's shopping needs and intentions.
[0003] Currently, the age attribute information of the items that the user intends to buy is usually manually determined based on the items that the user has purchased in the past. For example, the user currently wants to buy items suitable for the age group of 0 to 6 months.
[0004] However, in the process of implementing the present invention, the inventors found that there are at least the following problems in the prior art:
[0005] Manually judging the age attribute information of items that users are interested in is time-consuming and labor-intensive, and increases labor costs. Summary of the invention
[0006] The embodiments of the present invention provide an information prediction method, apparatus, device and storage medium to realize automatic prediction of age group attribute information of items intended by users and ensure the accuracy of information prediction.
[0007] In a first aspect, an embodiment of the present invention provides an information prediction method, comprising:
[0008] Based on the target user's interactive behavior data, determining a target item sequence in which the target user has interactive behavior;
[0009] Determining a target item vector sequence corresponding to the target user based on a target item representation vector corresponding to each target item in the target item sequence;
[0010] Based on the target item vector sequence and the pre-trained target classification network model, information prediction is performed on the age group attribute of the item intended by the target user to obtain the age group attribute information of the item intended by the target user.
[0011] In a second aspect, an embodiment of the present invention further provides an information prediction device, including:
[0012] A target item sequence determination module, used to determine a target item sequence in which the target user has an interactive behavior based on the interactive behavior data of the target user;
[0013] A target item vector sequence determination module, configured to determine a target item vector sequence corresponding to the target user based on a target item representation vector corresponding to each target item in the target item sequence;
[0014] The information prediction module is used to predict the age group attribute of the item intended by the target user based on the target item vector sequence and the pre-trained target classification network model, so as to obtain the age group attribute information of the item intended by the target user.
[0015] In a third aspect, an embodiment of the present invention further provides an electronic device, the electronic device comprising:
[0016] one or more processors;
[0017] A memory for storing one or more programs;
[0018] When the one or more programs are executed by the one or more processors, the one or more processors implement the information prediction method provided by any embodiment of the present invention.
[0019] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements an information prediction method as provided in any embodiment of the present invention.
[0020] One embodiment of the above invention has the following advantages or beneficial effects:
[0021] By determining the target item sequence with which the target user has interactive behavior based on the target user's interactive behavior data, and determining the target item vector sequence corresponding to the target user based on the target item representation vector corresponding to each target item in the target item sequence, the target item vector sequence can be input into a pre-trained target classification network model to predict the information on the age group attributes of the target user's intended items, thereby realizing automatic prediction of the age group attribute information of the user's intended items without human intervention, and the target classification network model is a pre-trained model, thereby ensuring the accuracy of information prediction using the target classification network model. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0023] Figure 1 is a flow chart of an information prediction method provided by an embodiment of the present invention;
[0024] Figure 2is a flow chart of another information prediction method provided by an embodiment of the present invention;
[0025] Figure 3 is an example diagram of a sliding division target item sequence involved in an embodiment of the present invention;
[0026] Figure 4 is an example diagram of the architecture of a Swim-transformer encoder involved in an embodiment of the present invention;
[0027] Figure 5 is a flow chart of another information prediction method provided by an embodiment of the present invention;
[0028] Figure 6 is an example diagram of the architecture of a target classification network model involved in an embodiment of the present invention;
[0029] Figure 7 is a schematic diagram of the structure of an information prediction device provided by an embodiment of the present invention;
[0030] Figure 8 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0031] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.
[0032] It should be noted that the collection, collection, updating, analysis, processing, use, transmission, storage and other aspects of user personal information involved in the technical solution of this disclosure are in compliance with the provisions of relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken for user personal information to prevent illegal access to user personal information data and maintain the security of user personal information, network security and national security.
[0033] Figure 1 This is a flowchart of an information prediction method provided by an embodiment of the present invention. This embodiment can be applied to the case of predicting the age attribute information of items intended by users, such as the case of predicting the age attribute information of maternal and infant items intended by users in a maternal and infant scenario. The method can be executed by an information prediction device, which can be implemented by software and / or hardware and integrated into an electronic device. Figure 1 As shown, the method specifically comprises the following steps:
[0034] S110: Determine a target item sequence with which the target user has interactive behavior based on the interactive behavior data of the target user.
[0035] The target user refers to a user who currently needs to perform information prediction. The interactive behavior data may refer to the item interaction information of the target user generating the preset interactive behavior before the current moment. The preset interactive behavior may include but is not limited to: at least one of the following: adding to a shopping cart behavior, clicking behavior, browsing behavior, searching behavior, and placing an order to purchase behavior. The item interaction information may include: item interaction time, item interaction times, and item interaction quantity. For example, the interactive behavior data may include the item purchase time, item purchase times, and item purchase quantity corresponding to each item purchased by the target user in the current recent period of time.
[0036] The target item sequence may refer to a sequence consisting of multiple target items. Each target item may be identified by the minimum stock keeping unit SKU (Stock Keeping Unit) to distinguish different target items. Each target item in the target item sequence may be an item with an applicable age group attribute, that is, the target item is not universal for all age groups, but can only be used by users in a specific age group. For example, the target item is a stage 3 formula milk powder suitable for 12-36 months.
[0037] Specifically, the target user's interaction behavior data in the current recent period can be obtained based on the target user's interaction log, so as to more accurately identify the age attribute information of the target user's intended items based on the recent interaction behavior data. The historical items with which the target user has interacted in the interaction behavior data are extracted, and the target item sequence is determined based on the historical items in the following three ways, but not limited to:
[0038] As a first implementation method, each extracted historical item can be directly used as a target item, and the items are arranged in sequence according to the order of their interaction time to obtain a target item sequence.
[0039] As a second implementation method, it is possible to detect whether each extracted historical item has an applicable age group attribute, and use the historical items with the applicable age group attribute as target items. These target items are arranged in chronological order according to the time of item interaction to obtain a target item sequence. At this time, the target item sequence with the applicable age group attribute can more accurately represent the item age group attribute information that the target user previously intended, thereby improving the accuracy of predicting the item age group attribute information that the target user currently intends.
[0040] As a third implementation method, it is possible to detect whether each historical item is an item under a preset item category, and use the historical items belonging to the preset item category as target items, and arrange these target items in order of item interaction time to obtain a target item sequence. Among them, the preset item category may refer to a category with applicable age group attributes, such as a preset item category that may be, but is not limited to, a maternal and child category. By obtaining a target item sequence under a preset item category, the target user's current intended item age group attribute information under the preset item category can be more accurately predicted, so that information prediction can be performed only for a certain category to meet different business needs.
[0041] S120 . Determine a target item vector sequence corresponding to a target user based on a target item representation vector corresponding to each target item in the target item sequence.
[0042] The target item characterization vector may be a feature vector used to characterize the target item introduction information. The target item introduction information may be, but is not limited to, target item title information and / or target item image information. The target item title information and target item image information are concise summary information of the target item attribute characteristics, and both contain most of the item information. Each target item may correspond to one or more target item characterization vectors. A target item vector sequence may refer to a sequence composed of target item characterization vectors corresponding to each item.
[0043] Specifically, the target item introduction information corresponding to each target item in the target item sequence can be encoded in real time to obtain the target item representation vector corresponding to each target item, or the item introduction information corresponding to each item can be encoded in advance to obtain the item representation vector corresponding to each item, so that based on the item representation vector corresponding to each item obtained in advance, the target item representation vector corresponding to each target item in the target item sequence can be quickly obtained, thereby improving the information prediction efficiency. Each target item in the target item sequence can be replaced with a corresponding target item representation vector to obtain a target item vector sequence. If each target item corresponds to at least two target item representation vectors, all target item representation vectors corresponding to each target item can be first vector spliced to obtain a complete target item representation vector, and then each target item in the target item sequence can be replaced with a corresponding complete target item representation vector to obtain a target item vector sequence. By splicing multiple target item representation vectors, more dimensional information can be provided, thereby further improving the accuracy of information prediction.
[0044] S130, based on the target item vector sequence and the pre-trained target classification network model, predict the age group attribute of the item intended by the target user to obtain the age group attribute information of the item intended by the target user.
[0045] Among them, the age group attribute information of the target user's intended items may refer to the age group attribute information of the target user's interest, that is, the user wants to buy items that are suitable for this age group attribute information. The target classification network model may be a neural network model for classifying and predicting the age group attribute information of the user's intended items. A plurality of age group attribute information of the items may be divided in advance based on business needs, and each divided age group attribute information of the items may be used as an information category classified by the target classification network model. For example, if the age group attribute information of the items is divided into 4 age groups, namely: 0 to 6 months, 6 months to 12 months, 12 months to 3 years old, and over 3 years old, then the target classification network model also corresponds to 4 classification categories, so as to predict the age group attribute information of the user's intended items by classification.
[0046] The target classification network model is a classification network model with time series processing capability. For example, the target classification network model may include a time cycle network sub-model and a fully connected network sub-model. The time cycle network sub-model may be, but not limited to, a long short-term memory network (LSTM). The fully connected network sub-model may be, but not limited to, a multi-layer perceptron (MLP). The target classification network model is obtained by pre-training the model based on the sample item vector sequence corresponding to the sample user and the item age attribute information intended by the sample user, so as to ensure the accuracy of the prediction of the target classification network model.
[0047] Specifically, since the lengths of the target item vector sequences corresponding to different users may be inconsistent, the target item vector sequences can be subjected to element deletion or completion processing based on the preset sequence length to obtain the target item vector sequence under the preset sequence length. For example, if the length of the target item vector sequence is greater than the preset sequence length, the redundant front elements or rear elements in the target item vector sequence can be deleted to obtain the target item vector sequence under the preset sequence length. If the length of the target item vector sequence is less than the preset sequence length, the target item vector sequence can be padded with 0 to obtain the target item vector sequence under the preset sequence length. The target item vector sequence under the preset sequence length is input into the target classification network model obtained by pre-training for information prediction, which can ensure the prediction accuracy of the target classification network model. In the target classification network model, the relevant information in the target item vector sequence can be processed by the time cycle network sub-model and the information can be passed on, even the information of the earlier step length can be passed to the back, thereby overcoming the defect of memory disappearance. The target feature information sequence output by the time cycle network sub-model is input into the fully connected network sub-model for full connection processing, and the predicted probability value corresponding to the attribute information of each item age group is determined and output. Based on the predicted probability value corresponding to each item age group attribute information output by the target classification network model, the item age group attribute information intended by the target user is determined. Since the target user can intend to have one or more item age group attribute information at the same time, each item age group attribute information with a predicted probability value greater than a preset threshold can be used as the item age group attribute information currently intended by the target user, so that the target classification network model can be used to automatically predict the item age group attribute information intended by the user, and the accuracy of the information prediction can be guaranteed.
[0048] It should be noted that the age attribute information of the items that the user intends can change dynamically over time. For example, in the mother-child scenario, as the child grows up, the age attribute information of the items that the user intends will also increase, so that the order of items in the target item vector sequence can indirectly reflect the growth stage of the user's child, and the target item vector sequence is characterized by the granularity of items, which can better represent the user's intention. The target classification network model can fully extract the time series information in the target item vector sequence, so that the age attribute information of the item that the user currently intends can be accurately predicted.
[0049] The technical solution of this embodiment determines the target item sequence with which the target user has interactive behavior based on the target user's interactive behavior data, and determines the target item vector sequence corresponding to the target user based on the target item representation vector corresponding to each target item in the target item sequence. The target item vector sequence can be input into a pre-trained target classification network model to predict the information of the age group attribute of the target user's intended item, thereby realizing automatic prediction of the age group attribute information of the user's intended item without human intervention, and the target classification network model is a pre-trained model, thereby ensuring the accuracy of information prediction using the target classification network model.
[0050] Based on the above technical solution, step S110 may include: based on the target user's interaction behavior data, determining the target items with which the target user has interaction behavior and the latest interaction time corresponding to each target item; based on the latest interaction time, sorting the target items to obtain a target item sequence.
[0051] The most recent interaction time may be an interaction time closest to the current moment. If a user interacts with the same item multiple times, the user will have multiple interaction times, and the most recent interaction time is the latest interaction time among these interaction times.
[0052] Specifically, all historical items in which the target user has interactive behavior in the interactive behavior data can be extracted, and these historical items can be deduplicated to obtain each target item after deduplication, and one or more item interaction times corresponding to each target item can be determined, and the latest item interaction time can be used as the latest interaction time of the corresponding target item. All target items are sorted in the order of the latest interaction time to obtain a target item sequence, and each target item will only appear once in the target item sequence. Compared with the target item sequence obtained by sorting all historical items, the target item sequence after deduplication can extract information more fully and save space.
[0053] Exemplarily, determining the target items with which the target user has interactive behavior based on the target user's interactive behavior data may include: determining the target items with which the target user has interactive behavior under a preset item category based on the target user's interactive behavior data, wherein the items under the preset item category have applicable age group attribute information.
[0054] Among them, the preset item category can be a category with applicable age group attributes that is pre-set based on business needs. For example, the preset item category can be but not limited to the maternal and child category or the clothing category. Specifically, all historical items in which the target user has interactive behavior in the interactive behavior data can be extracted, and each historical item can be detected whether it is an item under the preset item category, and the historical items belonging to the preset item category can be used as the target item. The target items are sorted in the order of the most recent interaction time corresponding to each target item, and the target item sequence of the target user under the preset item category is obtained, so that the target user's current intended item age group attribute information under the preset item category can be predicted in a targeted manner. For example, based on the target item sequence of the target user under the maternal and child category, the user's current intended maternal and child item age group attribute information can be obtained, so that the person-goods matching can be performed based on the applicable age group attribute information of each maternal and child item itself, and the maternal and child items currently intended by the user can be obtained, and then these maternal and child items can be pushed to the target user, realizing precision marketing.
[0055] Figure 2 A flowchart of another information prediction method provided for an embodiment of the present invention. Based on the above embodiments, this embodiment optimizes the step of "determining the target item vector sequence corresponding to the target user based on the target item characterization vector corresponding to each target item in the target item sequence", and on this basis, also optimizes the step of "predicting the age group attribute of the target user's intended item based on the target item vector sequence and the pre-trained target classification network model to obtain the target user's intended item age group attribute information". The explanations of the terms that are the same or corresponding to the above embodiments are not repeated here.
[0056] See also Figure 2 Another information prediction method provided in this embodiment specifically includes the following steps:
[0057] S210: Determine a target item sequence with which the target user has interactive behavior based on the interactive behavior data of the target user.
[0058] S220: Based on a preset sliding step length and a preset sequence length, the target item sequence is slidingly divided to obtain a plurality of target item subsequences.
[0059] The preset sliding step size may refer to the size of the sliding window, that is, the number of items that the sliding window slides each time. The preset sequence length may refer to the length of the target item subsequence, that is, the number of target items contained in the item subsequence. The length of each target item subsequence is the preset sequence length. The target item subsequence may refer to a sequence composed of part of the target items.
[0060] Specifically, by sliding on the target item sequence multiple times based on a preset sliding step and a preset sequence length, the target item sequence can be divided into multiple target item subsequences of equal length. The sliding division method can keep the sequence lengths corresponding to different users consistent, thereby ensuring that users with different interaction frequencies are compared on the same dimension, and fully extracting user information with shorter sequences, thereby improving the accuracy of information prediction.
[0061] Exemplarily, S220 may include: sliding on the target item sequence with a preset sliding step length, selecting a target item subsequence of a preset sequence length each time the sliding is performed; if the number of target items selected by the last sliding does not meet the preset sequence length, filling the last target item subsequence with a preset value.
[0062] Specifically, when sliding for the first time, the sliding window is located on the first target item in the target item sequence, so that the sliding division is performed from the first target item. Each sliding operation will select target items of a preset sequence length from the window position to form a target item subsequence. If the target item subsequence selected by the last sliding operation does not meet the preset sequence length, it will be filled with a preset value, such as 0, to obtain all target item subsequences of the same length. For example, Figure 3 An example diagram of sliding partitioning of target item sequence is given, such as Figure 3 As shown, the preset sliding step size is 2, the preset sequence length is 4, and the target item sequence of length 7 corresponding to the target user (ie, user A) can be slidably divided into 3 target item subsequences of length 4, and the last target item subsequence, i.e., sequence 3, does not meet the preset sequence length, and is filled with 0. Figure 3 Each target item in is represented by an item serial number.
[0063] S230 . Determine a plurality of target item vector subsequences corresponding to the target user based on the target item representation vector corresponding to each target item in each target item subsequence.
[0064] Specifically, the target item introduction information corresponding to each target item in each target item subsequence can be encoded in real time to obtain the target item representation vector corresponding to each target item, or the item introduction information corresponding to each item can be encoded in advance to obtain the item representation vector corresponding to each item, so that based on the item representation vector corresponding to each item obtained in advance, the target item representation vector corresponding to each target item in each target item subsequence can be quickly obtained, thereby improving the information prediction efficiency. After determining the target item representation vector corresponding to each target item, each target item in each target item subsequence can be replaced with the corresponding target item representation vector to obtain the corresponding target item vector subsequence.
[0065] Exemplarily, S230 may include: determining a title representation vector and / or an image representation vector corresponding to each target item in each target item sub-sequence; performing information replacement on each target item in each target item sub-sequence based on the title representation vector and / or the image representation vector to obtain a target item vector sub-sequence corresponding to each target item sub-sequence.
[0066] Among them, the title representation vector can be obtained by encoding the target item title information corresponding to the target item. The image representation vector can be obtained by encoding the target item image information corresponding to the target item. Specifically, each target item in each target item subsequence can be replaced with a corresponding title representation vector or image representation vector to obtain a corresponding target item vector subsequence, or the title representation vector and the image representation vector can be vector spliced to obtain a complete vector, and each target item in each target item subsequence can be replaced with a corresponding complete vector to obtain a corresponding target item vector subsequence. By splicing the title representation vector and the image representation vector that contain most of the information about the item, the item-related information can be more comprehensively mined, thereby further improving the accuracy of information prediction.
[0067] For example, the title representation vector can be obtained by encoding the target item title information corresponding to the target item based on the encoder Transformer. Compared with sequence models such as Recurrent Neural Network (RNN), the encoder Transformer can not only consider the positional relationship of elements in the sequence through forward positional encoding, but also apply the structure extraction information of the multi-head attention mechanism Multi-Head Attention. The existence of the multi-head attention mechanism makes the text information extraction more sufficient. Moreover, the encoder Transformer can capture global information, realize parallel computing, save encoding time, and improve encoding efficiency.
[0068] For example, the image representation vector can be obtained by encoding the target object image information corresponding to the target object based on the Swim-transformer encoder. Unlike text information, images have larger scale differences, higher pixel resolutions, and different sequence lengths, so the encoding computational complexity is higher. The Swim-transformer encoder is an improvement on the Transformer encoder. Figure 4 A sample diagram of the architecture of a Swim-transformer is given. Figure 4 As shown in the figure, one feature of Swim-transformer is the moving window, which can solve the problem of high global self-attention complexity by using self-attention in a small window. At the same time, the moving operation allows cross-window connection, so that the model can take into account the global situation. Another feature of Swim-transformer is the layered architecture. The existence of the layered architecture makes modeling more flexible under images of various scales, making the computational complexity linearly related to the image size. By using Swim-transformer, the image information of the target object can be encoded more accurately, which improves the encoding effect.
[0069] S240, based on multiple target item vector subsequences corresponding to the target user and a pre-trained target classification network model, predict the age group attribute of the item intended by the target user to obtain the age group attribute information of the item intended by the target user.
[0070] Specifically, all target item subsequences under a preset sequence length can be input into a pre-trained target classification network model for information prediction. When the target classification network model includes a time cycle network sub-model and a fully connected network sub-model, each target item vector sub-sequence can be input into the time cycle network sub-model for time series information processing to obtain a target feature information sequence corresponding to each target item vector sub-sequence, and all target feature information sequences corresponding to all target item vector sub-sequences are input into the fully connected network sub-model for full connection processing to obtain and output the predicted probability corresponding to each item age group attribute information. Based on the predicted probability value corresponding to each item age group attribute information output by the target classification network model, each item age group attribute information with a predicted probability value greater than a preset threshold can be used as the item age group attribute information currently intended by the target user, thereby realizing the automatic prediction of the item age group attribute information intended by the user and ensuring the accuracy of information prediction.
[0071] The technical solution of this embodiment, by slidingly dividing the target item sequence based on a preset sliding step size and a preset sequence length, obtains multiple target item subsequences of the same sequence length, and inputs the target item vector subsequences corresponding to the target item subsequences of the same length into the target classification network model for information prediction, which can more fully extract sequence information and further improve the accuracy of information prediction.
[0072] Figure 5 A flowchart of another information prediction method provided by an embodiment of the present invention. Based on the above embodiments, this embodiment optimizes the step of "predicting the age attribute of the target user's intended item based on multiple target item vector subsequences corresponding to the target user and the pre-trained target classification network model, and obtaining the target user's intended item age attribute information". The explanations of the terms that are the same or corresponding to the above embodiments are not repeated here.
[0073] See also Figure 5 Another information prediction method provided in this embodiment specifically includes the following steps:
[0074] S510: Determine a target item sequence with which the target user has interactive behavior based on the interactive behavior data of the target user.
[0075] S520: Based on a preset sliding step length and a preset sequence length, the target item sequence is slidingly divided to obtain a plurality of target item subsequences.
[0076] S530 : Determine a plurality of target item vector subsequences corresponding to the target user based on the target item representation vector corresponding to each target item in each target item subsequence.
[0077] S540: Determine a target interaction information subsequence corresponding to each target item vector subsequence based on the interaction information corresponding to each target item.
[0078] The interaction information corresponding to each target item may include but is not limited to information on the number of interactions and / or information on the number of interactions. Information on the number of interactions may refer to the number of interactions within the current recent period of time, such as the number of times a user purchases the same item. Information on the number of interactions may be determined based on the number of items at each interaction. For example, the number of items at each interaction may be directly determined as the interaction number information, or, if there are multiple interactions with the same item, the number of items at each interaction may be added together, and the obtained sum may be used as the interaction number information corresponding to the item. A target interaction information subsequence may refer to a sequence consisting of interaction information corresponding to each target item.
[0079] Specifically, each target item representation vector in each target item vector subsequence is replaced with corresponding interaction information to obtain a corresponding target interaction information subsequence.
[0080] Exemplarily, S540 may include: determining a target interaction number subsequence corresponding to each target item vector subsequence based on interaction number information corresponding to each target item; and / or determining a target interaction item quantity subsequence corresponding to each target item vector subsequence based on interaction item quantity information corresponding to each target item.
[0081] Specifically, when the interaction information includes the number of interactions information, each target item representation vector in each target item vector subsequence is replaced with the corresponding number of interactions information, and the obtained target interaction information subsequence includes the target number of interactions subsequence. When the interaction information includes the number of interactions information, each target item representation vector in each target item vector subsequence is replaced with the corresponding number of interactions information, and the obtained target interaction information subsequence includes the target number of interactions subsequence.
[0082] S550, inputting the target item vector subsequence and the corresponding target interaction information subsequence into the pre-trained target classification network model for information prediction.
[0083] It should be noted that since the target user's interaction information for different items is different, the interaction information can indirectly reflect the user's intention. For example, if a user purchases the same item multiple times or purchases multiple identical items at a time, it indicates that the user is more interested in the item. Therefore, the target interaction information subsequence can be used to weight the corresponding target item vector subsequence so that information can be extracted more fully and the accuracy of information prediction can be further improved.
[0084] Specifically, each target item vector subsequence can be concatenated with the corresponding target interaction information subsequence, or input into the pre-trained target classification network model by bitwise multiplication, so that the target classification network model can perform information prediction based on the target item vector subsequence and the target interaction information subsequence at the same time, further improving the accuracy of information prediction.
[0085] For example, Figure 6 An example diagram of the architecture of a target classification network model is given, such as Figure 6As shown, the target classification network model may include: a time cycle network sub-model, a multi-head attention network sub-model and a fully connected network sub-model. Among them, the time cycle network sub-model can be any time series processing network architecture. For example, the time cycle network sub-model can be but not limited to a long short-term memory network LSTM. The multi-head attention network sub-model can be an attention network model that extracts information from multiple angles, i.e., Multi Head Attention. The fully connected network sub-model can be a network model for implementing classification. For example, the fully connected network sub-model can be but not limited to a multi-layer perceptron MLP.
[0086] For example, for Figure 6 Given the target classification network model, S550 may include: inputting each target item vector subsequence and the corresponding target interaction information subsequence into the time cycle network sub-model for time series information processing to obtain a first feature information sequence corresponding to each target item vector subsequence; inputting the first feature information sequence corresponding to each target item vector subsequence into the multi-head attention network sub-model for multi-head attention processing to obtain a second feature information sequence; inputting the second feature information sequence into the fully connected network sub-model for full connection processing to obtain the predicted probability corresponding to the age attribute information of each item.
[0087] Specifically, each target item vector subsequence and the corresponding target interaction information subsequence can be concatenated, or the subsequence obtained by bitwise multiplication can be input into the time cycle network sub-model for time series information processing to obtain the first feature information sequence corresponding to each target item vector subsequence, and the first feature information sequence corresponding to each target item vector subsequence can be input into the multi-head attention network sub-model to extract important sequence information from multiple angles to obtain the extracted second feature information sequence, and all the second feature information sequences corresponding to all the target item vector subsequences can be input into the fully connected network sub-model for full connection processing to obtain the prediction probability corresponding to the age attribute information of each item and output it. By using the multi-head attention network sub-model to extract important sequence information from multiple angles, the accuracy of the target classification network model prediction can be further improved.
[0088] S560: Based on the output of the target classification network model, determine the age group attribute information of the item that the target user intends.
[0089] Specifically, based on the predicted probability value corresponding to each item age group attribute information output by the target classification network model, each item age group attribute information with a predicted probability value greater than a preset threshold can be used as the item age group attribute information currently intended by the target user, thereby realizing automatic prediction of the item age group attribute information intended by the user and further improving the accuracy of information prediction.
[0090] The technical solution of this embodiment can further improve the accuracy of information prediction by determining the target interaction information subsequence corresponding to each target item vector subsequence based on the interaction information corresponding to each target item, and using the target interaction information subsequence to perform weighted prediction on the corresponding target item vector subsequence.
[0091] The following is an embodiment of an information prediction device provided by an embodiment of the present invention. The device and the information prediction methods of the above-mentioned embodiments belong to the same inventive concept. For details not described in detail in the embodiment of the information prediction device, reference can be made to the embodiments of the above-mentioned information prediction methods.
[0092] Figure 7 This is a schematic diagram of the structure of an information prediction device provided by an embodiment of the present invention. This embodiment can be applied to the case where the age attribute information of items intended by users is predicted, such as the case where the age attribute information of maternal and child items intended by users is predicted in a maternal and child scenario. Figure 7 As shown, the device specifically includes: a target item sequence determination module 710 , a target item vector sequence determination module 720 and an information prediction module 730 .
[0093] Among them, the target item sequence determination module 710 is used to determine the target item sequence with which the target user has interactive behavior based on the interactive behavior data of the target user; the target item vector sequence determination module 720 is used to determine the target item vector sequence corresponding to the target user based on the target item representation vector corresponding to each target item in the target item sequence; the information prediction module 730 is used to perform information prediction on the age group attribute of the item intended by the target user based on the target item vector sequence and the pre-trained target classification network model, so as to obtain the age group attribute information of the item intended by the target user.
[0094] The technical solution of this embodiment determines the target item sequence with which the target user has interactive behavior based on the target user's interactive behavior data, and determines the target item vector sequence corresponding to the target user based on the target item representation vector corresponding to each target item in the target item sequence. The target item vector sequence can be input into a pre-trained target classification network model to predict the information of the age group attribute of the target user's intended item, thereby realizing automatic prediction of the age group attribute information of the user's intended item without human intervention, and the target classification network model is a pre-trained model, thereby ensuring the accuracy of information prediction using the target classification network model.
[0095] Optionally, the target item sequence determination module 710 includes:
[0096] A target item determination unit, configured to determine, based on the target user's interaction behavior data, the target items with which the target user has interaction behavior and the most recent interaction time corresponding to each target item;
[0097] The target item sequence determining unit is used to sort the target items based on the most recent interaction time to obtain a target item sequence.
[0098] Optionally, the target item determination unit is specifically used to: determine, based on the target user's interaction behavior data, the target item with which the target user has interaction behavior under a preset item category, wherein the items under the preset item category have applicable age group attribute information.
[0099] Optionally, the target item vector sequence determination module 720 includes:
[0100] a target item subsequence determination unit, configured to perform sliding division on the target item sequence based on a preset sliding step length and a preset sequence length to obtain a plurality of target item subsequences;
[0101] a target item vector subsequence determining unit, configured to determine a plurality of target item vector subsequences corresponding to the target user based on a target item representation vector corresponding to each target item in each target item subsequence;
[0102] The information prediction module 730 is specifically used to: based on multiple target item vector subsequences corresponding to the target user and a pre-trained target classification network model, perform information prediction on the age group attribute of the item intended by the target user to obtain the age group attribute information of the item intended by the target user.
[0103] Optionally, the target item subsequence determination unit is specifically used to: slide on the target item sequence with a preset sliding step length, and select a target item subsequence of a preset sequence length each time; if the number of target items selected by the last sliding does not meet the preset sequence length, fill the last target item subsequence with a preset value.
[0104] Optionally, the target item vector subsequence determination unit is specifically used to:
[0105] Determine a title representation vector and / or an image representation vector corresponding to each target item in each target item sub-sequence; perform information replacement on each target item in each target item sub-sequence based on the title representation vector and / or the image representation vector to obtain a target item vector sub-sequence corresponding to each target item sub-sequence.
[0106] Optionally, the title representation vector is obtained by encoding the target item title information corresponding to the target item based on an encoder Transformer;
[0107] The image representation vector is obtained by encoding the target object image information corresponding to the target object based on a Swim-transformer.
[0108] Optionally, the information prediction module 730 specifically includes:
[0109] a target interaction information subsequence determination unit, configured to determine a target interaction information subsequence corresponding to each target item vector subsequence based on the interaction information corresponding to each target item;
[0110] An information prediction unit, used for inputting the target item vector subsequence and the corresponding target interaction information subsequence into a pre-trained target classification network model for information prediction;
[0111] The attribute information determination unit is used to determine the age group attribute information of the item intended by the target user based on the output of the target classification network model.
[0112] Optionally, the target interaction information subsequence determination unit is specifically configured to:
[0113] Based on the interaction count information corresponding to each target item, determine the target interaction count subsequence corresponding to each target item vector subsequence; and / or, based on the interactive item quantity information corresponding to each target item, determine the target interactive item quantity subsequence corresponding to each target item vector subsequence.
[0114] Optionally, the target classification network model includes: a time cycle network sub-model, a multi-head attention network sub-model and a fully connected network sub-model;
[0115] The information prediction unit is specifically used for:
[0116] Each of the target item vector subsequences and the corresponding target interaction information subsequence is input into the time recurrent network sub-model for time series information processing to obtain a first feature information sequence corresponding to each of the target item vector subsequences; the first feature information sequence corresponding to each of the target item vector subsequences is input into the multi-head attention network sub-model for multi-head attention processing to obtain a second feature information sequence; the second feature information sequence is input into the fully connected network sub-model for full connection processing to obtain the predicted probability corresponding to the age attribute information of each item.
[0117] The information prediction device provided in the embodiment of the present invention can execute the information prediction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of executing the information prediction method.
[0118] It is worth noting that in the embodiment of the above-mentioned information prediction device, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.
[0119] Figure 8 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. Figure 8 A block diagram of an exemplary electronic device 12 suitable for use in implementing embodiments of the present invention is shown. Figure 8 The electronic device 12 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.
[0120] like Figure 8 As shown, the electronic device 12 is in the form of a general purpose computing device. The components of the electronic device 12 may include, but are not limited to: one or more processors or processing units 16, a system memory 28, and a bus 18 that connects various system components (including the system memory 28 and the processing unit 16).
[0121] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus architectures. By way of example, these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.
[0122] The electronic device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the electronic device 12, including volatile and non-volatile media, removable and non-removable media.
[0123] The system memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The electronic device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be used to read and write non-removable, non-volatile magnetic media ( Figure 8 not shown, usually called a "hard drive"). Although Figure 8Not shown in the figure, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, a DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The system memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present invention.
[0124] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in system memory 28, such program modules 42 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. Program modules 42 generally perform the functions and / or methods of the embodiments described herein.
[0125] The electronic device 12 may also communicate with one or more external devices 14 (e.g., keyboards, pointing devices, displays 24, etc.), may communicate with one or more devices that enable a user to interact with the electronic device 12, and / or may communicate with any device that enables the electronic device 12 to communicate with one or more other computing devices (e.g., network cards, modems, etc.). Such communication may be performed via an input / output (I / O) interface 22. Furthermore, the electronic device 12 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 20. As shown, the network adapter 20 communicates with other modules of the electronic device 12 via a bus 18. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 12, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.
[0126] The processing unit 16 executes various functional applications and data processing by running the program stored in the system memory 28, for example, implementing the steps of an information prediction method provided by the embodiment of the present invention, the method comprising:
[0127] Based on the target user's interactive behavior data, determining a target item sequence in which the target user has interactive behavior;
[0128] Determining a target item vector sequence corresponding to the target user based on a target item representation vector corresponding to each target item in the target item sequence;
[0129] Based on the target item vector sequence and the pre-trained target classification network model, information prediction is performed on the age group attribute of the item intended by the target user to obtain the age group attribute information of the item intended by the target user.
[0130] Of course, those skilled in the art can understand that the processor can also implement the technical solution of the information prediction method provided by any embodiment of the present invention.
[0131] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of the information prediction method provided in any embodiment of the present invention are implemented. The method includes:
[0132] Based on the target user's interactive behavior data, determining a target item sequence in which the target user has interactive behavior;
[0133] Determining a target item vector sequence corresponding to the target user based on a target item representation vector corresponding to each target item in the target item sequence;
[0134] Based on the target item vector sequence and the pre-trained target classification network model, information prediction is performed on the age group attribute of the item intended by the target user to obtain the age group attribute information of the item intended by the target user.
[0135] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to: an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.
[0136] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.
[0137] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
[0138] Computer program code for performing the operations of the present invention may be written in one or more programming languages or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0139] It should be understood by those skilled in the art that the modules or steps of the present invention described above can be implemented by a general-purpose computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, optionally, they can be implemented by a program code executable by a computer device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.
[0140] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.
Claims
1. An information prediction method, characterized in that: include: Based on the target user's interactive behavior data, determining a target item sequence in which the target user has interactive behavior; Determining a target item vector sequence corresponding to the target user based on a target item representation vector corresponding to each target item in the target item sequence; Based on the target item vector sequence and the pre-trained target classification network model, information prediction is performed on the age group attribute of the item intended by the target user to obtain the age group attribute information of the item intended by the target user.
2. The method according to claim 1, characterized in that The determining, based on the target user's interactive behavior data, a target item sequence in which the target user has interactive behavior comprises: Based on the target user's interaction behavior data, determine the target items with which the target user has interaction behavior and the most recent interaction time corresponding to each target item; The target items are sorted based on the most recent interaction time to obtain a target item sequence.
3. The method according to claim 2, characterized in that The determining, based on the target user's interactive behavior data, of a target item with which the target user has an interactive behavior includes: Based on the target user's interactive behavior data, target items with which the target user has interactive behavior under a preset item category are determined, wherein the items under the preset item category have applicable age group attribute information.
4. The method according to claim 1, characterized in that The determining, based on the target item representation vector corresponding to each target item in the target item sequence, a target item vector sequence corresponding to the target user includes: Based on a preset sliding step length and a preset sequence length, the target item sequence is slidingly divided to obtain a plurality of target item subsequences; Determine a plurality of target item vector subsequences corresponding to the target user based on the target item representation vector corresponding to each target item in each target item subsequence; The step of predicting the age attribute of the item intended by the target user based on the target item vector sequence and the pre-trained target classification network model to obtain the age attribute information of the item intended by the target user includes: Based on multiple target item vector subsequences corresponding to the target user and a pre-trained target classification network model, information prediction is performed on the age group attribute of the item intended by the target user to obtain the age group attribute information of the item intended by the target user.
5. The method according to claim 4, characterized in that The step of slidingly dividing the target item sequence based on a preset sliding step length and a preset sequence length to obtain a plurality of target item subsequences includes: Slide on the target item sequence with a preset sliding step length, and select a target item subsequence of a preset sequence length each time; If the number of target items selected by the last slide does not meet the preset sequence length, the last target item subsequence is filled and completed using the preset value.
6. The method according to claim 4, characterized in that The determining, based on the target item representation vector corresponding to each target item in each target item subsequence, a plurality of target item vector subsequences corresponding to the target user includes: Determine a title representation vector and / or an image representation vector corresponding to each target item in each target item sub-item sequence; Based on the title representation vector and / or the image representation vector, information replacement is performed on each target item in each target item subsequence to obtain a target item vector subsequence corresponding to each target item subsequence.
7. The method according to claim 6, characterized in that The title representation vector is obtained by encoding the target item title information corresponding to the target item based on the encoder Transformer; The image representation vector is obtained by encoding the target object image information corresponding to the target object based on a Swim-transformer.
8. The method according to claim 4, characterized in that The method of predicting the age group attribute of the item intended by the target user based on the multiple target item vector subsequences corresponding to the target user and the pre-trained target classification network model to obtain the age group attribute information of the item intended by the target user includes: Based on the interaction information corresponding to each target item, determine a target interaction information subsequence corresponding to each target item vector subsequence; Inputting the target item vector subsequence and the corresponding target interaction information subsequence into a pre-trained target classification network model for information prediction; Based on the output of the target classification network model, the age group attribute information of the item intended by the target user is determined.
9. The method according to claim 8, characterized in that The step of determining a target interaction information subsequence corresponding to each target item vector subsequence based on the interaction information corresponding to each target item includes: Based on the interaction count information corresponding to each target item, determine a target interaction count subsequence corresponding to each target item vector subsequence; and / or Based on the interactive item quantity information corresponding to each target item, a target interactive item quantity subsequence corresponding to each target item vector subsequence is determined.
10. The method according to claim 8, characterized in that The target classification network model includes: a time cycle network sub-model, a multi-head attention network sub-model and a fully connected network sub-model; The step of inputting the target item vector subsequence and the corresponding target interaction information subsequence into a pre-trained target classification network model for information prediction includes: Input each of the target item vector subsequences and the corresponding target interaction information subsequence into the time recurrent network submodel for time series information processing to obtain a first feature information sequence corresponding to each of the target item vector subsequences; Inputting the first feature information sequence corresponding to each of the target item vector subsequences into the multi-head attention network sub-model for multi-head attention processing to obtain a second feature information sequence; The second feature information sequence is input into the fully connected network sub-model for full connection processing to obtain the predicted probability corresponding to the age attribute information of each item.
11. An information prediction device, characterized in that: include: A target item sequence determination module, used to determine a target item sequence in which the target user has an interactive behavior based on the interactive behavior data of the target user; A target item vector sequence determination module, configured to determine a target item vector sequence corresponding to the target user based on a target item representation vector corresponding to each target item in the target item sequence; The information prediction module is used to predict the age group attribute of the item intended by the target user based on the target item vector sequence and the pre-trained target classification network model, so as to obtain the age group attribute information of the item intended by the target user.
12. An electronic device, characterized in that: The electronic device comprises: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the information prediction method as described in any one of claims 1-10.
13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the information prediction method as described in any one of claims 1 to 10 is implemented.
