Recommended item prediction method and device, storage medium and program product

Abstract

The invention discloses a recommendation item prediction method and device, a storage medium and a program product. The method comprises the steps that a historical interaction data set corresponding to an object set is acquired, the object set comprises at least one object, the historical interaction data set comprises a sub-historical interaction data set of the at least one object, and the sub-historical interaction data set comprises at least one piece of interaction behavior data of the object for at least one item; based on the object set and the historical interaction data set, multiple behavior views are constructed, and the behavior views are at least used for representing behavior preference information of the corresponding objects; performing fusion processing on the multiple pieces of behavior preference information corresponding to the multiple behavior views to obtain mixed behavior preference information of the object; and on the basis of the mixed behavior preference information, predicting to-be-recommended items of the object. The technical problem of low accuracy of project recommendation is solved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and more specifically, to a method, device, storage medium, and program product for predicting recommended items. Background Art

[0002] In real-world recommendation scenarios, users interact with items in a variety of ways. For example, in familiar e-commerce scenarios, users can perform various interactions on their favorite items, such as clicking, adding items to carts, adding items to favorites, and purchasing them. Properly utilizing this interaction data can not only effectively alleviate the data sparsity problem in recommendations but also significantly improve the performance of recommendation systems. Consequently, a growing number of research studies are moving towards multi-behavior recommendation.

[0003] Related technologies rely solely on a single method for multi-behavior recommendations, ignoring the complex and ever-changing nature of user preferences. Modeling a user's perspective alone may not be sufficient to enable recommendation platforms to select items that satisfy them. Consequently, these methods suffer from low item recommendation accuracy.

[0004] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0005] The embodiments of the present application provide a method, device, storage medium, and program product for predicting recommended items, so as to at least solve the technical problem of low accuracy of item recommendations.

[0006] According to one aspect of an embodiment of the present application, a method for predicting items to be recommended is provided, which may include: obtaining a historical interaction data set corresponding to an object set, wherein the object set includes at least one object, the historical interaction data set includes a sub-historical interaction data set of at least one object, and the sub-historical interaction data set includes at least one interaction behavior data of the object for at least one item; constructing multiple behavior views based on the object set and the historical interaction data set, wherein the behavior view is at least used to characterize the behavior preference information of the corresponding object; fusing multiple behavior preference information corresponding to the multiple behavior views to obtain mixed behavior preference information of the object; and predicting the items to be recommended for the object based on the mixed behavior preference information.

[0007] Optionally, constructing multiple behavior views based on the object set and the historical interaction dataset includes: retrieving a timestamp corresponding to the historical interaction dataset; and constructing multiple behavior views based on the object set, the historical interaction dataset, and the timestamp.

[0008] Optionally, the behavior view includes at least a graph view and multiple sequence views, the graph view is used to at least represent the association relationship between the interactive behavior data of at least one object, and the sequence view is used to represent the interactive behavior data of the corresponding object in different time periods. Based on the object set, the historical interaction data set and the timestamp, multiple behavior views are constructed, including: based on the object set and the historical interaction data set, constructing a graph view corresponding to the object set; based on the historical interaction data set and the timestamp corresponding to the historical interaction data set, constructing at least one sequence view corresponding to the object.

[0009] Optionally, multiple behavior preference information corresponding to multiple behavior views are fused to obtain mixed behavior preference information of the object, including: determining the first behavior preference information of the object based on the graph view, wherein the first behavior preference information is used to represent the average preference information of the object over a period of time; determining the second behavior preference information of the object based on the sequence view, wherein the second behavior preference is used to represent the dynamic preference information of the object; and fusing the first behavior preference information and at least one second behavior preference information of the object to obtain mixed behavior preference information.

[0010] Optionally, the first behavior preference information and at least one second behavior preference information of the object are fused to obtain mixed behavior preference information, including: calling a feedforward neural network to fuse the first behavior preference information and at least one second behavior preference information to obtain mixed behavior preference information.

[0011] Optionally, the method may further include: training a feedforward neural network based on first behavior preference information samples of the first object under different behavior views and second behavior preference information samples of the second object under different behavior views as training data, wherein the similarity between the behavior preference information of the first object and the behavior preference information of the second object is less than a similarity threshold.

[0012] According to another aspect of an embodiment of the present application, another method for predicting recommended videos is provided, which may include: obtaining a historical interaction data set corresponding to an object set, wherein the object set includes at least one object, the historical interaction data set includes a sub-historical interaction data set of at least one object, and the sub-historical interaction data set includes at least one interaction behavior data of the object for at least one video; constructing multiple behavior views based on the object set and the historical interaction data set, wherein the behavior view is at least used to characterize the behavior preference information of the corresponding object; fusing multiple behavior preference information corresponding to the multiple behavior views to obtain mixed behavior preference information of the object; and predicting a video to be recommended for the object based on the mixed behavior preference information.

[0013] According to another aspect of an embodiment of the present application, a device for predicting items to be recommended is also provided, which may include: an acquisition unit, used to acquire a historical interaction data set corresponding to an object set, wherein the object set includes at least one object, the historical interaction data set includes a sub-historical interaction data set of at least one object, and the sub-historical interaction data set includes at least one interaction behavior data of the object for at least one item; a construction unit, used to construct multiple behavior views based on the object set and the historical interaction data set, wherein the behavior view is at least used to characterize the behavior preference information of the corresponding object; a processing unit, used to fuse multiple behavior preference information corresponding to the multiple behavior views to obtain mixed behavior preference information of the object; and a prediction unit, used to predict the items to be recommended of the object based on the mixed behavior preference information.

[0014] According to another aspect of an embodiment of the present application, a computer-readable storage medium is also provided, which includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the method for predicting items to be recommended in an embodiment of the present application.

[0015] According to another aspect of an embodiment of the present application, a processor is further provided, which is used to run a program, wherein when the program is run by the processor, the method for predicting items to be recommended in an embodiment of the present application is executed.

[0016] According to another aspect of an embodiment of the present application, a computer program product is further provided. The computer program product includes computer instructions, wherein when the computer instructions are executed by a processor, the method for predicting items to be recommended according to an embodiment of the present application is implemented.

[0017] In an embodiment of the present application, a historical interaction dataset corresponding to an object set is obtained, wherein the object set includes at least one object, the historical interaction dataset includes a sub-historical interaction dataset of at least one object, and the sub-historical interaction dataset includes at least one interaction behavior data of the object for at least one item; based on the object set and the historical interaction dataset, multiple behavior views are constructed, wherein the behavior view is at least used to characterize the behavior preference information of the corresponding object; multiple behavior preference information corresponding to the multiple behavior views are fused to obtain mixed behavior preference information of the object; and based on the mixed behavior preference information, the recommended items for the object are predicted. That is, in an embodiment of the present application, the historical interaction data of the object are used to create multiple behavior views, and the behavior preference information corresponding to the multiple behavior views is determined to obtain multiple behavior preference information. In order to further study the common characteristics of the user under multiple views, the multiple behavior preference information can be fused to obtain the mixed behavior preference information corresponding to the object, so that the recommended items for the object can be predicted based on the mixed behavior preference information, thereby achieving the purpose of recommending suitable recommended items to the object based on the object's preferences, solving the technical problem of low accuracy of item recommendation, and achieving the technical effect of improving the accuracy of item recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0019] Figure 1 is a flow chart of a method for predicting items to be recommended according to an embodiment of the present application;

[0020] Figure 2 is a schematic diagram of a diagram according to an embodiment of the present application;

[0021] Figure 3 is a schematic diagram of a sequence view according to an embodiment of the present application;

[0022] FIG4( a ) is a schematic diagram of an improved lightweight graph convolution according to an embodiment of the present application;

[0023] FIG4( b ) is a schematic diagram of multi-behavior interaction modeling between user items according to an embodiment of the present application;

[0024] FIG4( c ) is a schematic diagram of another multi-behavior interaction modeling between user items according to an embodiment of the present application;

[0025] Figure 5 is a schematic diagram of multi-behavior sequence view encoding according to an embodiment of the present application;

[0026] Figure 6 is a schematic diagram of a prediction of a project to be recommended according to an embodiment of the present application;

[0027] Figure 7 is a flowchart of a method for predicting recommended videos according to an embodiment of the present application;

[0028] Figure 8 is a schematic diagram of a device for predicting items to be recommended according to an embodiment of the present application;

[0029] Figure 9 is a structural block diagram of a computer terminal according to an embodiment of the present application;

[0030] Figure 10 The figure is a block diagram of an electronic device for predicting a to-be-recommended item according to an embodiment of the present application. DETAILED DESCRIPTION

[0031] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0032] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0033] First, some nouns or terms that appear in the description of the embodiments of the present application are subject to the following interpretations:

[0034] Graph Neural Networks (GNNs) are deep learning models designed specifically for processing graph-structured data. GNNs iteratively update the representation of each node, taking into account both the node's own information and that of its neighbors, enabling them to learn the node's position and role in the graph.

[0035] A feedforward neural network (FNN) is a neural network model consisting of a series of layers, each containing multiple neurons connected by weights. During training, the FNN adjusts weights to minimize the difference between predicted values and true values, enabling learning and pattern recognition of input data.

[0036] Bayesian Personalized Ranking (BPR) is a recommendation system algorithm that is mainly used to solve the problem of users' preference ranking for different items.

[0037] According to an embodiment of the present application, an embodiment of a method for predicting items to be recommended is provided. The steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0038] In this embodiment, a method for predicting items to be recommended is proposed. The method can use the historical interaction data of the object to create multiple behavior views, determine the behavior preference information corresponding to the multiple behavior views, and obtain multiple behavior preference information. In order to further study the common characteristics of users under multiple views, the multiple behavior preference information can be fused to obtain the mixed behavior preference information corresponding to the object. Based on the mixed behavior preference information, the object's items to be recommended can be predicted, thereby achieving the purpose of recommending suitable items to be recommended to the object based on the object's preferences, solving the technical problem of low accuracy of item recommendation, and achieving the technical effect of improving the accuracy of item recommendation.

[0039] Figure 1 This is a flow chart of a method for predicting items to be recommended according to an embodiment of the present application. Figure 1 As shown, the method may include the following steps:

[0040] Step S101, obtaining a historical interaction dataset corresponding to an object set, wherein the object set includes at least one object, the historical interaction dataset includes a sub-historical interaction dataset of at least one object, and the sub-historical interaction dataset includes at least one interaction behavior data of the object with respect to at least one item;

[0041] Step S102: constructing multiple behavior views based on the object set and the historical interaction data set, wherein the behavior view is at least used to represent the behavior preference information of the corresponding object;

[0042] Step S103, fusing the multiple behavior preference information corresponding to the multiple behavior views to obtain mixed behavior preference information of the object;

[0043] Step S104: predicting and obtaining the recommended items of the object based on the mixed behavior preference information.

[0044] In this embodiment, the above-mentioned object set may include at least one object, which may be a user of a website or a mobile terminal. The above-mentioned historical interaction data set may include a sub-historical interaction data set corresponding to the object. The above-mentioned sub-historical interaction data set may include at least one interactive behavior data of the object on at least one item at different times. The above-mentioned interactive behavior data may include operations such as clicks, orders, add-to-cart, purchases, and clicks. The above-mentioned items may refer to content such as goods or videos. It should be noted that this is only an example, and there is no specific limitation on the type of interactive behavior data, the type of item, or the source of the user.

[0045] Optionally, for each user, the user's past interaction records with items on the platform (such as goods, articles, etc.) can be collected to form at least one sub-historical interaction dataset corresponding to the object. This dataset is not just a simple browsing record, but also includes richer historical behavior data, such as clicks, add-to-cart, favorites, purchases, etc. Each behavior is regarded as a way to express user preferences. The sub-historical interaction datasets of the user on different platforms or different software can be collected to obtain at least one sub-historical interaction dataset corresponding to the object. The sub-historical interaction datasets corresponding to multiple objects in the object set can be constructed to obtain the historical interaction dataset corresponding to the object set.

[0046] In this embodiment, the aforementioned behavior view can be used to represent the behavior preference information of the corresponding object, and can include graph views corresponding to multiple objects, as well as sequence views corresponding to each object. An object set can correspond to one graph view and multiple sequence views. The sequence view has a one-to-one correspondence with the sub-history interaction dataset. The aforementioned behavior preference information can be used to represent the behavior preferences of the object.

[0047] Optionally, based on the collected historical interaction dataset, a "behavioral view" from different perspectives can be constructed for each user. This behavioral view can be used to intuitively display a user's behavioral preferences for different projects, breaking down and representing user behavioral data along multiple dimensions (such as chronological order, behavior type, etc.). For example, a graph view can be constructed to reflect the global association between users and projects, while a sequence view can be constructed to reflect the temporal characteristics of user behavior. Each view carries information from different aspects of the user, helping to fully understand the user's behavioral preferences.

[0048] In this embodiment, the above-mentioned behavioral preference information can be used to determine the object's degree of preference or preference for different items, and may include the object's corresponding general preferences and dynamic preferences. The above-mentioned general preferences can be used to reflect the user's long-term and relatively stable interest tendencies, which can be formed based on all the user's historical interaction behaviors and represent the user's average preferences over a period of time. For example, a person who frequently buys electronic products may have a general preference for electronic products. The above-mentioned dynamic preferences focus more on the user's recent and immediate interest changes, which are inferred based on the user's most recent interaction sequence and can capture short-term fluctuations or trends in the user's interests. For example, if books are frequently searched recently, even if the long-term preference is for electronic products, a dynamic preference for books will be displayed. It should be noted that this is only an example, and there is no specific restriction on the protection content of the behavioral preference information.

[0049] Optionally, after constructing multiple behavioral views, the behavioral preference information of each user under different behavioral views can be determined. The scattered behavioral preference information can be fused to generate a more complete "mixed behavioral preference information" for the user. Machine learning techniques (such as neural networks) can be used to capture and fuse the relationship between behavioral views to ensure that the user's general preferences and dynamic preferences are fully considered. The fusion process may include weighted summation, splicing, and special neural network layers, aiming to extract the comprehensive preference characteristics of the user under all behavioral views. It should be noted that this is only an example and there is no specific limitation on the fusion method.

[0050] Optionally, hybrid behavior preference information can refer to combining graph view preference information and sequence view preference information together, and obtaining the final user preference representation through a series of fusion mechanisms (such as multi-layer feedforward neural networks). It can be used to simultaneously capture the user's long-term stable preferences (general preferences) and short-term dynamically changing preferences (dynamic preferences) to form a comprehensive and detailed user preference portrait.

[0051] For example, suppose a user has long had a preference for technology products (reflected by the graph view preference information), but has recently shown a strong interest in fitness equipment (captured by the sequence view preference information). Then, the hybrid behavior preference information will highlight the importance of fitness equipment in the user's general preferences while also maintaining a focus on technology products. This allows the recommendation system to recommend new fitness equipment to the user while not neglecting the technology products that the user has long been interested in.

[0052] The hybrid behavior preference information construction strategy in this embodiment combines global network features and temporal sequence features to more accurately characterize user interests, improving the accuracy and personalization of recommendations. Furthermore, by introducing self-supervised view comparison learning, it further explores the commonalities and differences in user preferences across different views, enriching the user preferences learned by the model.

[0053] In this embodiment, the above-mentioned items to be recommended can be products, web pages, videos, etc. to be recommended to the object. It should be noted that this is only an example and the types of items to be recommended are not specifically limited.

[0054] Optionally, the integrated "hybrid behavior preference information" can be used to further predict which items are most likely to appeal to the user. This prediction process can be based on recommendation algorithms such as collaborative filtering or deep learning models. It should be noted that this is merely an example and does not impose any specific limitations on the prediction method. The recommendation list generated by this method not only takes into account the user's consistent preferences but also sensitively captures recent changes in their interests, thereby providing more personalized and accurate recommendations, improving user satisfaction and the system's recommendation efficiency.

[0055] In this embodiment, the above method can be applied to recommendation scenarios, such as product recommendation, video recommendation, etc. It should be noted that this is only an example and does not specifically limit the usage scenario of this method.

[0056] Optionally, this embodiment creates the user's historical interaction data into two view structures: a graph view and a sequence view. An improved lightweight graph convolutional network is used to process the user's multi-behavior interaction graph to learn the user's general preferences, and a self-attention network is used to process the user's multi-behavior sequence view to learn the user's dynamic preferences. In order to further study the common characteristics of users under the two views, a self-supervised task is introduced for view comparison learning, and the auxiliary loss generated by the task is combined with the main task loss generated by the dual-view encoding to jointly optimize the model. By introducing the dual view, the user's preference information can be more accurately grasped and effective recommendations can be made for them, which is of great significance to improving user experience and platform satisfaction. The above method can learn the user's preference characteristics based on their historical interactions, and ultimately make accurate recommendations to provide users with a good experience.

[0057] Through the above steps S101 and S104 of the present application, a historical interaction dataset corresponding to an object set is obtained, wherein the object set includes at least one object, the historical interaction dataset includes a sub-historical interaction dataset of at least one object, and the sub-historical interaction dataset includes at least one interaction behavior data of the object for at least one item; based on the object set and the historical interaction dataset, multiple behavior views are constructed, wherein the behavior view is at least used to represent the behavior preference information of the corresponding object; the multiple behavior preference information corresponding to the multiple behavior views are fused to obtain mixed behavior preference information of the object; and based on the mixed behavior preference information, the recommended items for the object are predicted. That is, in an embodiment of the present application, the historical interaction data of the object are used to create multiple behavior views, and the behavior preference information corresponding to the multiple behavior views is determined to obtain multiple behavior preference information. In order to further study the common characteristics of the user under multiple views, the multiple behavior preference information can be fused to obtain the mixed behavior preference information corresponding to the object, so that the recommended items for the object can be predicted based on the mixed behavior preference information, thereby achieving the purpose of recommending suitable recommended items to the object based on the object's preferences, solving the technical problem of low accuracy of item recommendation, and achieving the technical effect of improving the accuracy of item recommendation.

[0058] The above method of this embodiment is further introduced below.

[0059] As an optional implementation, step S102 constructs multiple behavior views based on the object set and the historical interaction dataset, including: retrieving the timestamp corresponding to the historical interaction dataset; and constructing multiple behavior views based on the object set, the historical interaction dataset and the timestamp.

[0060] In this embodiment, the timestamp corresponding to at least one sub-historical interaction dataset in the historical interaction dataset is retrieved. Further, based on the object set, the historical interaction dataset and the timestamp, multiple behavior data can be constructed.

[0061] Optionally, after the historical interaction dataset is obtained, multiple behavior views corresponding to the historical interaction dataset may be constructed based on the timestamp corresponding to at least one sub-historical interaction dataset in the historical interaction dataset.

[0062] Optionally, before constructing multiple behavior views based on the historical interaction dataset and timestamp, the historical interaction dataset and object set can be initialized first, and the initialized historical interaction dataset, object set and timestamp (also called timestamp information) can be used to construct multiple behavior views.

[0063] As an optional implementation, the behavior view includes at least a graph view and multiple sequence views. The graph view is used to at least represent the association relationship between the interactive behavior data of at least one object, and the sequence view is used to represent the interactive behavior data of the corresponding object in different time periods. Based on the object set, the historical interaction data set and the timestamp, multiple behavior views are constructed, including: based on the object set and the historical interaction data set, constructing a graph view corresponding to the object set; based on the historical interaction data set and the timestamp corresponding to the historical interaction data set, constructing at least one sequence view corresponding to the object.

[0064] In this embodiment, the behavior view may include multiple sequence views and graph views. The sequence view may correspond one-to-one with a sub-historical interaction dataset and may be used to represent the interaction behavior data of the corresponding object in different time periods. The graph view may correspond to a historical interaction dataset and may be used to represent the association between the interaction behavior data of at least one object, for example, to determine the preferences of different objects for the same or different items.

[0065] Optionally, the historical interaction data set may include user interaction behavior data for different projects.

[0066] Figure 2 is a schematic diagram of a diagram according to an embodiment of the present application, such as Figure 2 As shown, nodes in a graph view can be used to represent objects or items, and the connections between nodes can be used to represent the interactive behaviors of the objects. For example, a dotted line between nodes can be used to represent clicks, and a solid line between nodes can be used to represent purchases. u can be used to represent different objects, and v can be used to represent items. Figure 3 is a schematic diagram of a sequence view according to an embodiment of the present application, such as Figure 3 As shown, the nodes in the sequence view can be used to represent items selected by the subject at different times.

[0067] Optionally, in this embodiment, u and v can be used to represent users and items respectively, and the object set and historical interaction data set can be represented as U = {u1, u2…u m}、V={v1,v2…v n}. A multi-behavior interaction graph (i.e., graph view) can be represented as Where V represents a node, E represents an edge, and R represents the type of the edge. The user's multi-behavior sequence can be defined as The set of multiple behavior sequences of a single user u∈U is represented as S u =[s1,s2…s x ], which can be composed of the user’s recent x interactive behaviors, where s x Indicates the item information of the xth interaction of user u, and each s x=(v x ,r k ), which contains two parts of information: the item v that the user interacts with x and the type of interaction r k The types of interactive behaviors mentioned above may refer to interactive behavior data, which may include clicks, add-to-cart, purchases, and other behaviors. It should be noted that this is only for illustration and does not impose any specific restrictions on the types of interactive behaviors.

[0068] Optionally, the purpose of initializing the user set, item set, and interaction behavior type is to convert the high-dimensional sparse one-hot vectors of users and items into low-dimensional dense embeddings. The conversion method is to use the encoding vectors corresponding to users and items under certain behaviors (for example, one-hot vectors) and the user embedding matrix Project Embedding Matrix Multiply. d can be used to represent the embedding dimension size. The one-hot vector can be a user or project name (IDentifier, referred to as ID) with the corresponding position element being 1 and the remaining positions being 0. The initial embedding implementation of a single user, project, and user behavior relationship is as follows:

[0069]

[0070] Among them, ID u ∈{0,1} |U| With ID v ∈{0,1} |V| Represent the one-hot vectors of users u and v respectively; represents the embedding matrix; K represents the total number of behavior types; ID r ∈{0,1} K| A one-hot vector representing the row r.

[0071] Optionally, since the user's graph view and the multi-behavior sequence view are two completely different structures, the initial embedding of the user in the multi-behavior view needs to be further processed. The multi-behavior interaction sequence S of user u∈U u ={s1,s2…s x The initial embedding of all interactions in The embedding implementation for each interaction is as follows:

[0072]

[0073] Where, Indicates interaction x =(v x ,r k ) x Initial embedding of Represents interactive behavior r k The initial embedding of ; : represents the multiplication of corresponding elements.

[0074] As an optional implementation, multiple behavior preference information corresponding to multiple behavior views are fused to obtain mixed behavior preference information of the object, including: determining the first behavior preference information of the object based on the graph view, wherein the first behavior preference information is used to represent the average preference information of the object over a period of time; determining the second behavior preference information of the object based on the sequence view, wherein the second behavior preference is used to represent the dynamic preference information of the object; and fusing the first behavior preference information and at least one second behavior preference information of the object to obtain mixed behavior preference information.

[0075] In this embodiment, the object's first behavioral preference information can be determined based on the graph view. This first behavioral preference information can be the user's general preference and can be used to represent the object's average preference information over a period of time. Simultaneously, the object's second behavioral preference information can be determined based on the sequence view. This second behavioral preference information can be used to represent the object's dynamic preference information. Furthermore, the object's first behavioral preference information and the second behavioral preference information can be fused to obtain moderated behavioral preference information.

[0076] Optionally, in the user dual-view encoding stage, two different encoding methods can be used to process the user's graph view and multi-behavior interaction graph respectively to obtain the second behavior preference information corresponding to the sequence view and the first behavior preference information corresponding to the graph view.

[0077] Optionally, a lightweight graph convolutional network is used to process the graph views to learn general preferences corresponding to objects.

[0078] For example, a lightweight graph convolutional network is a general network that encodes the interaction features between users and projects under specific behaviors. In order to integrate the meaning of different behaviors into the preference embedding learning of users and projects, this embodiment draws on the idea of transductive embedding (TransE) in the knowledge graph. In the knowledge graph, (head entity, relationship, tail entity) can be regarded as a triple, and the tail entity can be predicted by combining the head entity and the relationship. For example, if the head entity is Li Bai and the relationship is the occupation, the expected predicted tail entity is the poet. Figure 4(a) is a schematic diagram of an improved lightweight graph convolution according to an embodiment of the present application. As shown in 4(a), in a multi-behavior interaction graph, (user, behavior, project) can also be regarded as a triple. Figure 4(b) is a schematic diagram of multi-behavior interaction modeling between user projects according to an embodiment of the present application. As shown in 4(b), v xCan be used to represent items that users interact with (e.g., v1, v2); u can be used to represent users (i.e., objects); r k It can be used to characterize the type of interactive behavior. By combining users, items, and their corresponding behavioral relationships, multi-behavioral interaction modeling between users and items can be achieved. Furthermore, inspired by knowledge graph entity prediction, this embodiment uses an improved lightweight graph convolutional network to encode the specific behavioral interaction features between users and items, enabling more vivid analysis.

[0079] FIG4( c ) is a schematic diagram of another multi-behavior interaction modeling between users and projects according to an embodiment of the present application. As shown in FIG4( c ), the update rules for users and projects are the same. The update rules for users and projects under the k-th behavior are as follows:

[0080]

[0081] Among them, Nu represents the neighbors of user u under behavior k; Nv represents the neighbors of item v under behavior k; It is a symmetric normalization operation that prevents the embedding scale from increasing with the superposition of graph convolutions; Is a combinatorial function; : corresponds to the element multiplication operation. In addition to this operation, there are also subtraction or neural network operations.

[0082] in,

[0083] After completing the update of user and item preference embeddings, the update rules for behavior embeddings are as follows:

[0084]

[0085] in, Represents the parameter matrix of the k-th behavior in the L-th layer. Its role is to map all behaviors to the same embedding space as users and items and allow them to be used in the next graph convolution layer.

[0086] After encoding users and items under various behaviors, the embeddings of users and items under various behaviors are fused to obtain the embeddings of users and items at each level. The multi-behavior fusion rules of users and items at layer L are as follows:

[0087]

[0088] Finally, after obtaining the preference embeddings of users, items, and each level, we fuse them to obtain the user’s general preference embedding Project Embed v and behavioral embedding The specific implementation is as follows:

[0089]

[0090] Alternatively, considering that the self-attention-based sequence recommendation method is good at incorporating temporal dependencies into the user's dynamic preferences, in order to better learn the user's dynamic preference characteristics, the self-attention mechanism can be used to process the user's multi-behavior interaction sequence. Figure 5 is a schematic diagram of a multi-behavior sequence view encoding according to an embodiment of the present application, such as Figure 5 As shown, we can first add a learnable position embedding matrix P = [p1, p2…p n ] Get the initial embedding matrix with position enhancement in and This is because the self-attention mechanism has no positional relationship, that is, swapping the positions of two elements in the sequence will not affect the final result.

[0091] Then the above obtained item embedding matrix with position enhancement is Send it to the self-attention network to get a new embedding matrix in The specific implementation is as follows:

[0092] B=Self-Attention(Q,K,V)

[0093] Among them, the input Q, K, and V of self-attention are expressed as: The three projection matrices It should also be noted that sequential recommendation predicts the user's interaction behavior at time t based on the user's sequential behavior before time t. Therefore, when calculating self-attention, when j>i, the q of the query vector matrix Q is i k with the key vector matrix K j Perform a masking operation, i.e. q i .(k j ) T =0.

[0094] Next, in order to explore the potential nonlinear characteristics between different dimensions, B can be fed into a two-layer feedforward neural network to obtain in

[0095] z * =FNN(b * )=(W2σ(W1b * +t1)+t2)

[0096] in, are all trainable parameter matrices; σ is the nonlinear activation function ReLu. In addition, in order to prevent overfitting, a residual connection is introduced after the feedforward neural network to obtain in

[0097]

[0098] Finally, in order to obtain more complex features, the above operation is iterated H times, and the result of the Hth iteration is As the final output. The sum of multiple vectors in derives the dynamic preference embedding of user u∈U:

[0099]

[0100] As an optional implementation, the first behavior preference information and at least one second behavior preference information of the object are fused to obtain mixed behavior preference information, including: calling a feedforward neural network to fuse the first behavior preference information and at least one second behavior preference information to obtain mixed behavior preference information.

[0101] In this embodiment, after obtaining the first behavior preference information and the second behavior preference information corresponding to each object, a feedforward neural network can be called to fuse the first behavior preference information and at least one second behavior preference information to obtain mixed behavior preference information.

[0102] Optionally, embed the user's general preferences in the graph view Dynamic preference embedding with sequence view Finally, a two-layer feedforward neural network can be used to fuse the two types of feature preferences to obtain the user's final mixed preference embedding e u , the specific implementation is as follows:

[0103]

[0104] in, σ is the nonlinear activation function ReLu.

[0105] Optionally, the general preference embedding described above can be used to determine general preferences. The dynamic preference embedding described above can be used to determine dynamic preferences.

[0106] As an optional implementation, the method may further include: training a feedforward neural network based on first behavior preference information samples of the first object under different behavior views and second behavior preference information samples of the second object under different behavior views as training data, wherein the similarity between the behavior preference information of the first object and the behavior preference information of the second object is less than a similarity threshold.

[0107] In this embodiment, the feedforward neural network model may be trained based on first behavior preference information samples of a first subject under different behavior views, and second behavior preference information samples of a second subject under different behavior views, wherein the similarity between the first subject's behavior preference information and the second subject's behavior preference information is less than a similarity threshold.

[0108] Optionally, this embodiment can obtain the item preference embedding based on the user's multi-behavior interaction view. v , based on the two views, we get the user's hybrid preference embedding e u Then use the inner product function to predict the probability of the user interacting with the recommended item under the target behavior

[0109]

[0110] Optionally, based on the above possibilities, an item to be recommended corresponding to the object may be determined.

[0111] Optionally, considering that user preferences are not completely isolated and different views can reflect some common characteristics of users, this embodiment uses contrastive learning techniques to learn the commonalities between users' views. Furthermore, in the multi-task joint optimization stage, this embodiment introduces a view contrastive learning task and uses multi-task learning to jointly optimize the model performance by combining the main task loss generated by view encoding and the auxiliary task loss generated by view contrastive learning.

[0112] Contrastive learning focuses on learning the common features between similar instances and distinguishing the differences between non-similar instances. Similarly, in the view contrastive learning task of this invention, the preference embedding of the same user in different views is closer to that of other users. i ∈U preference embedding in different views As a positive sample pair, different users u i 、u j ∈U preference embedding in different views and Then, we use the contrast loss function (infoNCE) introduced in Section 2.5 of this paper as the loss (LossCL) for the dual-view comparison task to maximize the consistency of positive samples and minimize the consistency of negative samples:

[0113]

[0114] Among them, |U| represents the number of users; sim represents the cosine similarity, and τ is a hyperparameter often called the temperature coefficient.

[0115] Furthermore, the losses generated by the dual-view encoding task and the view contrast learning task can be defined as the main task loss and the auxiliary task loss respectively, and they can be jointly learned using a multi-task learning framework:

[0116]

[0117] Where λ is a hyperparameter that controls the contrastive learning task; Θ represents the set of model trainable parameters; μL2 is the normalization coefficient that controls regularization, which is to prevent overfitting; Loss main Can be used to characterize the main task loss.

[0118] Finally, we can use the hit rate HR and the normalized cumulative discount gain NDCG. The hit rate HR emphasizes the accuracy of the recommendation. It is determined by the ratio of the number of records in the recommendation list to the number of records in the test set. The larger the ratio, the higher the recommendation accuracy. The normalized cumulative discount gain NDCG emphasizes the quality of the ranking. The larger the NDCG value, the better the recommendation quality. The higher the ranking, the more likely the user is to like the item.

[0119]

[0120] Here, the numerator represents the number of users whose items in the test set appear in the recommendation list; the denominator represents the total number of users, and K represents the length of the recommendation list.

[0121]

[0122] Among them, NDCG is the result of DCG after normalization; IDCG is the maximum value of DCG, which is the most ideal result of DCG, that is, the items in the recommendation list are all items in the test set; reli represents the correlation of the position in the recommendation list. The items at the front of the recommendation list have relatively higher scores, and the items at the back have relatively lower scores.

[0123] In real-world recommendation scenarios, users interact with items in a variety of ways. For example, in familiar e-commerce scenarios, users can perform various interactions on their favorite items, such as clicking, adding items to carts, adding items to favorites, and purchasing them. Properly utilizing this interaction data can not only effectively alleviate the data sparsity problem in recommendations but also significantly improve the performance of recommendation systems. Consequently, a growing number of research studies are moving towards multi-behavior recommendation.

[0124] In related technologies, multi-behavior recommendation encompasses two types: graph-based and sequence-based. Different recommendation methods have distinct characteristics. Graph-based recommendation methods generally construct a graph based on user-item interaction information and then utilize graph convolution techniques to extract high-level information about the user and items. Sequence-based multi-behavior recommendation differs from the previous approach in that it predicts the next item a user might interact with based on their continuous interactions. Because user sequences are dynamically arranged in chronological order, it aims to learn from users' dynamic preferences. However, a user's preferences are often complex and dynamic, and modeling them solely based on a single user view may not be sufficient for a recommendation platform to select items that satisfy the user. For example, in an e-commerce scenario, if a user's interactions primarily involve mobile phones, computers, and headphones, it can be determined that the user is an electronics enthusiast. This reflects their long-term preferences, so the recommendation platform can frequently recommend electronics products to them. Furthermore, if a user's recent interactions involve books, it can be determined that their recent hobby is likely reading, reflecting their short-term dynamic preferences. Therefore, the recommendation platform can appropriately recommend relevant books or e-readers to them over a period of time, rather than constantly recommending electronics. However, existing graph-based recommendation methods struggle to incorporate users' current dynamic preferences into their preference embeddings, while sequence-based recommendation methods fail to construct a global structure of all user interactions. This prevents the model from learning a more robust preference embedding for users, and thus from making more accurate recommendations. Consequently, all of these methods suffer from the technical issue of low recommendation accuracy.

[0125] To address the aforementioned issues, this embodiment models user preference features based on the user's graph view and sequence view. The goal is to learn more fine-grained user preferences, ultimately enabling more effective recommendations and improving user experience and platform satisfaction. First, based on user data (historical interaction data), multiple behavior views of the user are constructed: a graph view and a sequence view. The user set and item set are initialized. Separate encoders are then used to learn the user's graph view and multi-behavior sequence view, respectively. The user preferences learned from the two views are then combined to form the user's final hybrid preference. A view comparison learning task is then introduced to learn the user's common preferences across the two views. Finally, a multi-task learning mechanism is used to jointly optimize the main task loss generated by view encoding and the auxiliary task loss generated by view comparison learning. This allows the model to be optimized, reflecting user preferences from different perspectives based on different views. These preferences are both distinct and complementary, and share some common characteristics across different views. By combining the user's two multi-behavior views to model user preference features, the upper limit of the multi-behavior recommendation method is raised, thereby improving the accuracy of item recommendations and addressing the technical issue of low item recommendation accuracy.

[0126] Figure 6 is a schematic diagram of a prediction of a project to be recommended according to an embodiment of the present application, such as Figure 6 As shown, the method can include three implementation steps: constructing user multi-behavior views, user dual-view encoding, and multi-task joint optimization. When constructing user multi-behavior views, two views are first constructed based on user interaction information and timestamp information, such as a graph view 601 centered on user u1 and a multi-behavior sequence view 602 based on chronological interactions. During the dual-view encoding stage, an improved lightweight graph convolutional network is used to learn the graph view encoding 603 to obtain the user's general preference embedding. Simultaneously, a self-attention network is used to learn the sequence view encoding 604 to obtain the user's dynamic preference embedding. The two preferences are then fused to obtain the user's final hybrid preference embedding. Subsequently, a FNN is used to make recommendations based on the user's hybrid preference embedding.

[0127] Optionally, in the multi-task joint optimization stage, in order to explore the common preferences of users in the two views, view contrast learning can be introduced to explore the common preferences of users in the two views, and the main task loss generated by multi-task learning combined with view encoding and the auxiliary task loss generated by view contrast learning can be used to jointly optimize the model effect.

[0128] In an embodiment of the present application, the historical interaction data of the object is used to create multiple behavior views, and the behavior preference information corresponding to the multiple behavior views is determined to obtain multiple behavior preference information. In order to further study the common characteristics of users under multiple views, the multiple behavior preference information can be fused to obtain the mixed behavior preference information corresponding to the object. Based on the mixed behavior preference information, the recommended items of the object can be predicted, thereby achieving the purpose of recommending suitable recommended items to the object based on the object's preferences, solving the technical problem of low accuracy of item recommendations, and achieving the technical effect of improving the accuracy of item recommendations.

[0129] According to an embodiment of the present application, a method for predicting recommended videos is also provided. Figure 7 This is a flow chart of a method for predicting recommended videos according to an embodiment of the present application. Figure 7 As shown, the method may include the following steps:

[0130] Step S701: Obtain a historical interaction dataset corresponding to an object set, wherein the object set includes at least one object, the historical interaction dataset includes a sub-historical interaction dataset of at least one object, and the sub-historical interaction dataset includes at least one interaction behavior data of the object for at least one video.

[0131] Step S702 : constructing a plurality of behavior views based on the object set and the historical interaction data set, wherein the behavior view is at least used to represent the behavior preference information of the corresponding object.

[0132] Step S703 : performing fusion processing on the multiple behavior preference information corresponding to the multiple behavior views to obtain mixed behavior preference information of the object.

[0133] Step S704: Based on the mixed behavior preference information, the video to be recommended for the object is predicted.

[0134] In this embodiment, the above-mentioned interactive behavior data may include operations such as clicking, collecting, and liking the video. It should be noted that this is only a specific explanation.

[0135] Optionally, when recommending videos to users, a historical interaction dataset corresponding to an object set can be obtained, and at least one interactive behavior data of the object for at least one video can be determined based on the historical interaction dataset; multiple behavior views can be constructed based on the object set and the historical interaction dataset, wherein the behavior view is at least used to characterize the behavior preference information of the corresponding object; multiple behavior preference information corresponding to the multiple behavior views are fused to obtain mixed behavior preference information of the object; based on the mixed behavior preference information, the video to be recommended for the object is predicted.

[0136] Through the above steps S701 to S704, a historical interaction dataset corresponding to the object set is obtained, wherein the object set includes at least one object, the historical interaction dataset includes a sub-historical interaction dataset of at least one object, and the sub-historical interaction dataset includes at least one interaction behavior data of the object for at least one video; based on the object set and the historical interaction dataset, multiple behavior views are constructed, wherein the behavior view is at least used to characterize the behavior preference information of the corresponding object; the multiple behavior preference information corresponding to the multiple behavior views are fused to obtain mixed behavior preference information of the object; based on the mixed behavior preference information, the video to be recommended for the object is predicted, thereby achieving the technical effect of improving the accuracy of the video to be recommended and solving the technical problem of low accuracy of the video to be recommended.

[0137] According to an embodiment of the present application, a device for predicting items to be recommended is also provided. It should be noted that the device for predicting items to be recommended in this embodiment can be used to execute the method for predicting items to be recommended in the above-mentioned embodiment of the present application.

[0138] Figure 8 Schematic diagram of a device for predicting items to be recommended according to an embodiment of the present application. Figure 8 As shown, the prediction device 80 for items to be recommended may include: an acquisition unit 802 , a construction unit 804 , a processing unit 806 and a prediction unit 808 .

[0139] The acquisition unit 802 is used to acquire a historical interaction dataset corresponding to an object set, wherein the object set includes at least one object, the historical interaction dataset includes a sub-historical interaction dataset of at least one object, and the sub-historical interaction dataset includes at least one interaction behavior data of the object for at least one project.

[0140] The constructing unit 804 is configured to construct a plurality of behavior views based on the object set and the historical interaction data set, wherein the behavior view is at least used to represent the behavior preference information of the corresponding object.

[0141] The processing unit 806 is configured to perform fusion processing on the multiple behavior preference information corresponding to the multiple behavior views to obtain mixed behavior preference information of the object.

[0142] The prediction unit 808 is configured to predict the recommended items of the object based on the mixed behavior preference information.

[0143] The device for predicting items to be recommended in this embodiment obtains a historical interaction data set corresponding to an object set through an acquisition unit, wherein the object set includes at least one object, the historical interaction data set includes a sub-historical interaction data set of at least one object, and the sub-historical interaction data set includes at least one interaction behavior data of the object for at least one item; constructs multiple behavior views based on the object set and the historical interaction data set through a construction unit, wherein the behavior view is at least used to characterize the behavior preference information of the corresponding object; fuses multiple behavior preference information corresponding to the multiple behavior views through a processing unit to obtain mixed behavior preference information of the object; and predicts the item to be recommended of the object based on the mixed behavior preference information through a prediction unit, thereby solving the technical problem of low accuracy of item recommendation and achieving the technical effect of improving the accuracy of item recommendation.

[0144] The embodiment of the present application can provide a computer terminal, which can be any computer terminal device in a computer terminal group. Optionally, in this embodiment, the computer terminal can also be replaced by a terminal device such as a mobile terminal.

[0145] Optionally, in this embodiment, the computer terminal may be located in at least one network device among a plurality of network devices of a computer network.

[0146] In this embodiment, the above-mentioned computer terminal can execute the program code of the following steps in the method for predicting items to be recommended: obtaining an image to be converted whose image content includes an initial biological object, and a reference image whose image content includes a reference biological object, wherein the reference biological image is used to adjust the initial biological object; adjusting the limbs of the initial biological object based on reference limb information of the reference biological object in the reference image to obtain a first biological object; determining first limb attribute information of the first biological object and second limb attribute information of the reference biological object; adjusting the first biological object based on the second limb attribute information and the first limb attribute information to obtain a target biological object; converting the image to be converted based on the target biological object to obtain a target image, wherein the similarity between the biological object in the target image and the reference biological object is higher than a similarity threshold.

[0147] Optionally, Figure 9 is a structural block diagram of a computer terminal according to an embodiment of the present application, such as Figure 9 As shown, the computer terminal 908 may include: one or more (only one is shown in the figure) processors 902 , a memory 904 and a transmission device 906 .

[0148] Among them, the memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the method and device for predicting items to be recommended in the embodiments of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, realizing the above-mentioned method for predicting items to be recommended. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely located relative to the processor, and these remote memories may be connected to the computer terminal 908 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0149] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: obtaining a historical interaction data set corresponding to the object set, wherein the object set includes at least one object, the historical interaction data set includes a sub-historical interaction data set of at least one object, and the sub-historical interaction data set includes at least one interaction behavior data of the object for at least one item; based on the object set and the historical interaction data set, constructing multiple behavior views, wherein the behavior view is at least used to characterize the behavior preference information of the corresponding object; fusing the multiple behavior preference information corresponding to the multiple behavior views to obtain mixed behavior preference information of the object; and predicting the object's to-be-recommended items based on the mixed behavior preference information.

[0150] It can be understood by those skilled in the art that Figure 9 The structure shown is for illustration only, and the computer terminal 908 may also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile Internet device (MID for short), a PAD, or other terminal devices. Figure 9 It does not limit the structure of the computer terminal 908. For example, the computer terminal 908 may also include Figure 9 More or fewer components (such as network interfaces, display devices, etc.) shown in, or with Figure 9 Different configurations shown.

[0151] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0152] According to an embodiment of the present application, a computer-readable storage medium is further provided, which includes a stored program, wherein the program executes the method for predicting items to be recommended in the above embodiment.

[0153] Optionally, in this embodiment, the computer-readable storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.

[0154] Optionally, in this embodiment, the computer-readable storage medium is configured to store program code for executing the following steps: retrieving a timestamp corresponding to a historical interaction dataset; and constructing multiple behavior views based on the object set, the historical interaction dataset, and the timestamp.

[0155] Optionally, the computer-readable storage medium may also execute program code for the following steps: constructing a graph view corresponding to the object set based on the object set and the historical interaction data set; constructing at least one sequence view corresponding to the object based on the historical interaction data set and the timestamp corresponding to the historical interaction data set.

[0156] Optionally, the computer-readable storage medium can also execute the program code of the following steps: based on the graph view, determining the first behavior preference information of the object, wherein the first behavior preference information is used to represent the average preference information of the object over a period of time; based on the sequence view, determining the second behavior preference information of the object, wherein the second behavior preference is used to represent the dynamic preference information of the object; and fusing the first behavior preference information and at least one second behavior preference information of the object to obtain mixed behavior preference information.

[0157] Optionally, the computer-readable storage medium may further execute program code of the following steps: calling a feedforward neural network to fuse the first behavior preference information and at least one second behavior preference information to obtain mixed behavior preference information.

[0158] Optionally, the computer-readable storage medium can also execute program code for the following steps: based on first behavior preference information samples of the first object under different behavior views, and second behavior preference information samples of the second object under different behavior views as training data, a feedforward neural network is trained, wherein the similarity between the behavior preference information of the first object and the behavior preference information of the second object is less than a similarity threshold.

[0159] In this embodiment, the historical interaction data of the object is used to create multiple behavior views, and the behavior preference information corresponding to the multiple behavior views is determined to obtain multiple behavior preference information. In order to further study the common characteristics of users under multiple views, the multiple behavior preference information can be fused to obtain the mixed behavior preference information corresponding to the object. Based on the mixed behavior preference information, the object's to-be-recommended items can be predicted, thereby achieving the purpose of recommending appropriate to-be-recommended items to the object based on the object's preferences, solving the technical problem of low accuracy of item recommendations, and achieving the technical effect of improving the accuracy of item recommendations.

[0160] According to an embodiment of the present application, a processor is further provided, which is used to run a program, wherein when the program is run by the processor, the method for predicting items to be recommended in the above embodiment is executed.

[0161] Optionally, in this embodiment, the computer terminal may be located in at least one network device among a plurality of network devices of a computer network.

[0162] In this embodiment, the above-mentioned computer terminal can execute the program code of the following steps in the method for predicting items to be recommended: obtaining a historical interaction data set corresponding to an object set, wherein the object set includes at least one object, the historical interaction data set includes a sub-historical interaction data set of at least one object, and the sub-historical interaction data set includes at least one interaction behavior data of the object for at least one item; constructing multiple behavior views based on the object set and the historical interaction data set, wherein the behavior view is at least used to characterize the behavior preference information of the corresponding object; fusing multiple behavior preference information corresponding to the multiple behavior views to obtain mixed behavior preference information of the object; and predicting the items to be recommended for the object based on the mixed behavior preference information.

[0163] Among them, the memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the method and device for predicting items to be recommended in the embodiments of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, realizing the above-mentioned method for predicting items to be recommended. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely located relative to the processor, and these remote memories may be connected to the computer terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0164] Optionally, the processor may also execute the following program code: construct multiple behavior views based on the object set and the historical interaction data set, including: retrieving the timestamp corresponding to the historical interaction data set; constructing multiple behavior views based on the object set, the historical interaction data set and the timestamp.

[0165] Optionally, the processor may also execute the program code of the following steps: constructing a graph view corresponding to the object set based on the object set and the historical interaction data set; constructing at least one sequence view corresponding to the object based on the historical interaction data set and the timestamp corresponding to the historical interaction data set.

[0166] Optionally, the processor may also execute the program code of the following steps: determining the first behavior preference information of the object based on the graph view, wherein the first behavior preference information is used to represent the average preference information of the object over a period of time; determining the second behavior preference information of the object based on the sequence view, wherein the second behavior preference is used to represent the dynamic preference information of the object; and fusing the first behavior preference information and at least one second behavior preference information of the object to obtain mixed behavior preference information.

[0167] Optionally, the processor may further execute a program code of the following steps: calling a feedforward neural network to fuse the first behavior preference information and at least one second behavior preference information to obtain mixed behavior preference information.

[0168] Optionally, the processor may also execute the program code of the following steps: training a feedforward neural network based on first behavior preference information samples of the first object under different behavior views and second behavior preference information samples of the second object under different behavior views as training data, wherein the similarity between the behavior preference information of the first object and the behavior preference information of the second object is less than a similarity threshold.

[0169] By using the embodiment of the present application, the historical interaction data of the object is used to create multiple behavior views, and the behavior preference information corresponding to the multiple behavior views is determined to obtain multiple behavior preference information. In order to further study the common characteristics of users under multiple views, the multiple behavior preference information can be fused to obtain the mixed behavior preference information corresponding to the object. Based on the mixed behavior preference information, the recommended items for the object can be predicted, thereby achieving the purpose of recommending suitable recommended items to the object based on the object's preferences, solving the technical problem of low accuracy of item recommendations, and achieving the technical effect of improving the accuracy of item recommendations.

[0170] According to an embodiment of the present application, a computer program product is further provided. The computer program product includes computer instructions, wherein when the computer instructions are executed by a processor, the method for predicting items to be recommended in the above embodiment is implemented.

[0171] An embodiment of the present application may provide an electronic device, which may include a memory and a processor.

[0172] Figure 10 1 is a block diagram of an electronic device for a method of predicting items to be recommended according to an embodiment of the present application. 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 may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present application described and / or claimed herein.

[0173] like Figure 10As shown, the device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1002 or a computer program loaded from a storage unit 1008 into a random access memory (RAM) 1003. Various programs and data required for the operation of the device 1000 can also be stored in the RAM 1003. The computing unit 1001, ROM 1002, and RAM 1003 are connected to each other via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0174] Various components in device 1000 are connected to I / O interface 1005, including: an input unit 1006, such as a keyboard, mouse, etc.; an output unit 1007, such as various types of displays, speakers, etc.; a storage unit 1008, such as a magnetic disk, optical disk, etc.; and a communication unit 1009, such as a network card, modem, wireless communication transceiver, etc. The communication unit 1009 allows device 1000 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0175] The computing unit 1001 can be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 1001 performs the various methods and processes described above, such as the data verification method. For example, in some embodiments, the data verification method can be implemented as a computer software program that is tangibly contained in a machine-readable medium, such as the storage unit 1008. In some embodiments, part or all of the computer program can be loaded and / or installed on the device 1000 via the ROM 1002 and / or the communication unit 1009. When the computer program is loaded into the RAM 1003 and executed by the computing unit 1001, one or more steps of the data verification method described above can be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured to execute the data verification method in any other appropriate manner (for example, by means of firmware).

[0176] According to an embodiment of the present application, a method for predicting items to be recommended is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0177] Various embodiments of the systems and techniques described 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 can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0178] The program code for implementing the methods of the present application can be written in any combination of one or more programming languages. Such program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow charts and / or block diagrams are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0179] In the context of the present application, a machine-readable medium can be a tangible medium that can contain or store a program for use by an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, 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 foregoing.

[0180] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or an 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 can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the 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 acoustic input, voice input, or tactile input).

[0181] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0182] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact through a communication network. The client-server relationship arises through computer programs running on the respective computers and having a client-server relationship with each other. The server may be a cloud server, a server in a distributed system, or a server integrated with a blockchain.

[0183] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0184] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0185] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0186] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected to achieve the purpose of the present embodiment according to actual needs.

[0187] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0188] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM), random access memory (RAM), mobile hard disk, magnetic disk or optical disk, etc., various media that can store program codes.

[0189] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for predicting recommended items, characterized in that: include: Acquire a historical interaction dataset corresponding to an object set, wherein the object set includes at least one object, the historical interaction dataset includes at least one sub-historical interaction dataset of the object, and the sub-historical interaction dataset includes at least one interaction behavior data of the object with respect to at least one item; constructing a plurality of behavior views based on the object set and the historical interaction dataset, wherein the behavior views are at least used to represent the behavior preference information of the corresponding objects; fusing the plurality of behavior preference information corresponding to the plurality of behavior views to obtain mixed behavior preference information of the object; Based on the mixed behavior preference information, items to be recommended for the object are predicted.

2. The method according to claim 1, characterized in that The constructing of multiple behavior views based on the object set and the historical interaction dataset includes: Retrieve the timestamp corresponding to the historical interaction dataset; A plurality of behavior views are constructed based on the object set, the historical interaction dataset, and the timestamp.

3. The method according to claim 2, characterized in that The behavior view includes at least a graph view and multiple sequence views, the graph view is used to represent at least the association relationship between the interaction behavior data of at least one object, and the sequence view is used to represent the interaction behavior data of the corresponding object in different time periods. The multiple behavior views are constructed based on the object set, the historical interaction data set, and the timestamp, including: Constructing the graph view corresponding to the object set based on the object set and the historical interaction dataset; At least one sequence view corresponding to the object is constructed based on the historical interaction dataset and the timestamp corresponding to the historical interaction dataset.

4. The method according to claim 3, characterized in that The fusing the plurality of behavior preference information corresponding to the plurality of behavior views to obtain mixed behavior preference information of the object includes: determining first behavior preference information of the object based on the graph view, wherein the first behavior preference information is used to represent average preference information of the object within a time period; determining second behavior preference information of the object based on the sequence view, wherein the second behavior preference is used to represent dynamic preference information of the object; The first behavior preference information and at least one second behavior preference information of the object are fused to obtain the mixed behavior preference information.

5. The method according to claim 4, characterized in that The fusing the first behavior preference information and at least one second behavior preference information of the object to obtain the mixed behavior preference information includes: A feedforward neural network is called to perform a fusion process on the first behavior preference information and at least one piece of the second behavior preference information to obtain the mixed behavior preference information.

6. The method according to claim 5, characterized in that The method further comprises: The feedforward neural network is trained based on first behavior preference information samples of the first object under different behavior views and second behavior preference information samples of the second object under different behavior views as training data, wherein the similarity between the behavior preference information of the first object and the behavior preference information of the second object is less than a similarity threshold.

7. A method for predicting recommended videos, characterized in that: include: Obtaining a historical interaction dataset corresponding to an object set, wherein the object set includes at least one object, the historical interaction dataset includes at least one sub-historical interaction dataset of the object, and the sub-historical interaction dataset includes at least one interaction behavior data of the object with respect to at least one video; constructing a plurality of behavior views based on the object set and the historical interaction dataset, wherein the behavior views are at least used to represent the behavior preference information of the corresponding objects; fusing the plurality of behavior preference information corresponding to the plurality of behavior views to obtain mixed behavior preference information of the object; Based on the mixed behavior preference information, a video to be recommended for the object is predicted.

8. A device for predicting items to be recommended, characterized in that: include: an acquisition unit, configured to acquire a historical interaction dataset corresponding to an object set, wherein the object set includes at least one object, the historical interaction dataset includes at least one sub-historical interaction dataset of the object, and the sub-historical interaction dataset includes at least one interaction behavior data of the object with respect to at least one item; a construction unit, configured to construct a plurality of behavior views based on the object set and the historical interaction data set, wherein the behavior views are at least used to represent behavior preference information of the corresponding objects; a processing unit, configured to fuse the plurality of behavior preference information corresponding to the plurality of behavior views to obtain mixed behavior preference information of the object; A prediction unit is used to predict the recommended items of the object based on the mixed behavior preference information.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 7.

10. A computer program product, characterized in that The method comprises computer instructions, which implement the method according to any one of claims 1 to 7 when executed by a processor.

Images