Information recommendation method and apparatus, device, and storage medium

Abstract

The embodiments of the present application relate to the technical field of artificial intelligence, and provide an information recommendation method and apparatus, a device, and a storage medium. The method comprises: performing training on the basis of historical user material interaction information to obtain a deep semantic recall model and a deep semantic sorting model; and by means of the deep semantic recall model, recalling a plurality of candidate materials corresponding to a user identifier, and by means of the deep semantic sorting model, determining and sorting preference scores of the plurality of candidate materials and recalled materials of other paths, so as to recommend materials, thereby eliminating dependence on operation experience, and improving the recommendation effect. The training enables the deep semantic recall model to learn the text semantic representation of the user material interaction information, so that when materials are recalled, the literal semantics of the materials is more concerned, thereby improving the accuracy of the recalled materials. The deep semantic sorting model is enabled to learn the behavior semantic representation of the user material interaction information, so that when the recalled materials are subjected to preference scoring and sorting, the user behavior information of the material feedback level is more concerned, thereby improving the user click rate.

Description

Information recommendation method, device, equipment and storage medium

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to the Chinese patent application filed with the State Intellectual Property Office of the People's Republic of China on July 21, 2023, with application number 202310904850.3 and invention name "An information recommendation method, device, equipment and storage medium", the entire contents of which are incorporated by reference into this application. Technical Field

[0003] The embodiments of the present application relate to the field of artificial intelligence technology, and in particular to an information recommendation method, apparatus, device, and storage medium. Background Art

[0004] In the long-term operation of carousels, in order to improve the click-through conversion rate of carousels, related technologies are based on historical experience in carousel recommendation operations to recommend carousels. Specifically, recommended items are manually selected based on factors such as preferred, active, discounted, and popular.

[0005] However, the recommendation effect of the above method depends on manual operation experience, and operation experience is highly dependent on the popularity of materials and the head effect; if there are few popular products and insufficient interactive evaluation, the recommendation effect will be poor.

[0006] Summary of the Invention

[0007] The embodiments of the present application provide an information recommendation method, apparatus, device, and storage medium for improving the efficiency and accuracy of material recommendations.

[0008] In one aspect, an embodiment of the present application provides an information recommendation method, the method comprising:

[0009] Receive an information recommendation request carrying a user ID sent by a business system;

[0010] Recalling multiple candidate materials corresponding to the user identifier through a deep semantic recall model;

[0011] Determining preference scores for the plurality of candidate materials and recalled materials of other routes associated with the user identifier using a deep semantic ranking model, wherein the deep semantic recall model and the deep semantic ranking model are trained based on historical user-material interaction information;

[0012] At least one recommended material is determined from the plurality of candidate materials and recalled materials on other routes based on the obtained preference scores.

[0013] In one aspect, an embodiment of the present application provides an information recommendation device, comprising:

[0014] A receiving module, configured to receive an information recommendation request carrying a user identifier sent by a business system;

[0015] A recall module, configured to recall multiple candidate materials corresponding to the user identifier through a deep semantic recall model;

[0016] a ranking module, configured to determine preference scores for each of the plurality of candidate materials and recalled materials of other routes associated with the user identifier using a deep semantic ranking model, wherein the deep semantic recall model and the deep semantic ranking model are trained based on historical user-material interaction information;

[0017] The recommendation module is configured to determine at least one recommended material from the plurality of candidate materials and the recalled materials from other routes based on the obtained preference scores.

[0018] Optionally, the recall module is further configured to:

[0019] Before determining the preference scores of the multiple candidate materials and the recalled materials of other routes associated with the user identifier through a deep semantic sorting model, material recall is performed based on timeliness strategy, popularity strategy and relevance strategy to obtain recalled materials of other routes associated with the user identifier.

[0020] Optionally, the sorting module is specifically configured to:

[0021] sorting the plurality of candidate materials and recalled materials from other routes based on the obtained preference scores to obtain a first sorting result;

[0022] Reordering the first sorting result based on a preset business policy to obtain a second sorting result;

[0023] Based on the second sorting result, at least one recommended material is selected from the multiple candidate materials and the recalled materials of other routes.

[0024] Optionally, the sorting module is specifically configured to:

[0025] sorting the plurality of candidate materials and recalled materials from other routes based on the obtained preference scores to obtain a first sorting result;

[0026] Based on the first sorting result, at least one recommended material is selected from the multiple candidate materials and the recalled materials of other routes.

[0027] Optionally, the recommendation module is further configured to:

[0028] Based on the obtained preference scores, after determining at least one recommended material from the multiple candidate materials and recalled materials from other routes, the material identification list corresponding to the at least one recommended material is sent to the business system, so that the business system obtains corresponding material detail information from the material system based on the material identification list, and sends the obtained material detail information to the terminal device for display.

[0029] Optionally, the historical user-material interaction information includes: attribute information of multiple materials with which the user has historically interacted, and / or multiple historical behavior information of the user interacting with the materials.

[0030] Optionally, the text semantic representation includes: user latent semantic representation and material latent semantic representation; the behavior semantic representation includes: user preference latent semantic representation and material interest point latent semantic representation.

[0031] On the one hand, an embodiment of the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned information recommendation method when executing the program.

[0032] On the one hand, an embodiment of the present application provides a computer-readable storage medium storing a computer program executable by a computer device. When the program is run on the computer device, the computer device executes the steps of the above-mentioned information recommendation method.

[0033] On the one hand, an embodiment of the present invention provides a computer program product, which includes a computer program stored on a computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer performs the steps of the above-mentioned information recommendation method.

[0034] In an embodiment of the present application, a deep semantic recall model and a deep semantic sorting model are pre-trained based on historical user-material interaction information; an information recommendation request carrying a user identifier sent by a business system is received, and multiple candidate materials corresponding to the user identifier are automatically recalled through the deep semantic recall model; the preference scores of the multiple candidate materials and other recalled materials associated with the user identifier are automatically determined and sorted through the deep semantic sorting model, and then materials are recommended based on the sorting results, eliminating dependence on operational experience and enhancing the recommendation effect.

[0035] In the process of training the deep semantic recall model, the deep semantic recall model learns to make textual semantic representations of historical user-material interaction information. Therefore, when the deep semantic recall model is used to recall materials, more attention is paid to the literal semantics of the materials themselves, so that the candidate materials can be better expanded in terms of literal semantics, which can improve the recalled materials. In the process of training the deep semantic ranking model, the deep semantic ranking model learns to make behavioral semantic representations of historical user-material interaction information. Therefore, when the deep semantic ranking model is used to perform preference scoring and ranking for each recalled material, more attention is paid to user behavior information at the material feedback level, which can better improve user click-through rate. In this application, the two semantic representations act at different stages, can complement each other, eliminate dependence on operational experience, enhance recommendation effects, and achieve personalized recommendation effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0037] FIG1 is a schematic diagram of a system architecture provided in an embodiment of the present application;

[0038] FIG2A is a schematic diagram of a recommendation strategy architecture provided in an embodiment of the present application;

[0039] FIG2B is a schematic diagram of a system architecture provided in an embodiment of the present application;

[0040] FIG3 is a flow chart of an information recommendation method provided in an embodiment of the present application;

[0041] FIG4 is a schematic diagram of a network architecture of a deep semantic recall model provided in an embodiment of the present application;

[0042] FIG5 is a schematic diagram of a network architecture of a deep semantic ranking model provided in an embodiment of the present application;

[0043] FIG6 is a schematic diagram of the structure of an information recommendation device provided in an embodiment of the present application;

[0044] FIG7 is a schematic structural diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solutions and beneficial effects of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0046] For ease of understanding, the terms involved in the embodiments of the present invention are explained below.

[0047] ANN: Approximate Nearest Neighbor, which refers to a set of K vectors in a vector library that are close to the query vector.

[0048] TDM: Tree-based Deep Match, deep tree matching model.

[0049] CB: Content-base, based on content, for example, content-based recommendation algorithm.

[0050] GNN: Graph Neural Network, graph neural network.

[0051] NCF: Neural Collaborative Filtering, a collaborative filtering algorithm based on a neural network model.

[0052] DCN: Deep&Cross Network, deep recommendation model.

[0053] DIN: Deep Interest Network, deep interest network.

[0054] DNN: Deep Neural Network.

[0055] ESMM: Entire Space Multi-task model, multi-task model.

[0056] MMOE: Multi-gate Mixture-of-Experts model, multi-task model.

[0057] It should be noted that in the following implementation methods of this application, related data such as user information and behavioral information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0058] Refer to Figure 1, which is a system architecture diagram applicable to an embodiment of the present application. The system architecture at least includes a recommendation system 101, a business system 102, a material system 103, and a terminal device 104.

[0059] Terminal device 104 is pre-installed with an application for displaying materials, including but not limited to text, images, and videos. Applications include client applications, web applications, mini-program applications, and the like. Terminal device 104 may be, but is not limited to, a smartphone, tablet computer, laptop computer, desktop computer, smart home appliance, intelligent voice interaction device, or smart in-vehicle device.

[0060] The material system 103 is used to store detailed information of all recommended materials. The recommendation system 101 and the business system 102 can be independent physical servers, or they can be server clusters or distributed systems composed of multiple physical servers. They can also be cloud servers that provide basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. The terminal device 104 can be directly or indirectly connected to the business system 102 through wired or wireless communication, and this application does not impose any restrictions on this.

[0061] The terminal device 104 sends an information recommendation request to the business system 102 . The information recommendation request includes a user ID. The information recommendation request is used to request detailed information of a recommended material corresponding to the user ID from the business system 102 .

[0062] The business system 102 sends an information recommendation request carrying the user ID to the recommendation system 101 . The information recommendation request is used to request the recommendation system 101 for a list of material IDs of recommended materials corresponding to the user ID.

[0063] The recommendation system 101 adopts the recommendation strategy architecture shown in FIG2A , performs multi-channel material recall, recall fusion, fine sorting, and re-sorting, and then sends the material identification list of the topN recommended materials to the business system 102 .

[0064] The business system 102 obtains corresponding material detail information from the material system 103 based on the material identification list, and sends the obtained material detail information to the terminal device 104 .

[0065] The terminal device 104 displays the material details in the installed application interface.

[0066] In some embodiments, referring to FIG2B , the recommendation system 101 includes a recommendation operation management platform 105, an online recommendation service platform 106, and an offline data processing platform 107, wherein the recommendation operation management platform 105 includes: system services, data services, recommendation services, and analysis services; the online recommendation service platform 106 includes: multi-channel recall, recall fusion, precise sorting, re-sorting, and recommendation central control; the offline data processing platform 107 includes data synchronization, data sets, incremental training, vector construction, and recall queues, wherein data synchronization includes material synchronization and user behavior synchronization, and material synchronization is achieved by obtaining basic information of materials from the material system 103, and user behavior synchronization is achieved by collecting behavioral information of users' applications on terminal devices.

[0067] In practical applications, the information recommendation method in the embodiments of the present application can be applied to various recommendation scenarios such as carousel recommendations, mall recommendations, member recommendations, and advertising recommendations.

[0068] Based on the system architecture diagram shown in FIG1 , an embodiment of the present application provides a process of an information recommendation method, as shown in FIG3 . The process of the method is executed by a computer device, which may be the recommendation system 101 shown in FIG1 , and includes the following steps:

[0069] Step S301: receiving an information recommendation request carrying a user identifier sent by a business system.

[0070] Specifically, materials include but are not limited to text, images, and videos. The information recommendation request is used to request a list of material identifiers of recommended materials corresponding to the user identifier from the recommendation system.

[0071] Step S302: Recall multiple candidate materials corresponding to the user identifier through a deep semantic recall model.

[0072] Specifically, the deep semantic recall model is trained based on historical user-material interaction information. During the training process, the deep semantic recall model learns to perform textual semantic representation of historical user-material interaction information, where the textual semantic representation includes user latent semantic representation and material latent semantic representation.

[0073] Historical user-item interaction information includes: attribute information of multiple items with which users have historically interacted, and / or information on multiple historical user-item interactions. Item attribute information includes textual information such as item titles and item identifiers. Historical behavior information includes user identifiers, recent behavior sequences, and long-term behavior sequences. Behaviors include item clicks, item stays, and item reviews. The network architecture of the Deep Semantic Recall Model is shown in Figure 4. Deep learning algorithms applicable to the Deep Semantic Recall Model include GNN, TDM, Gru4Rec, Word2Vec, NCF, and YouTube.

[0074] In some embodiments, in the process of recalling multiple candidate materials corresponding to a user identifier through a deep semantic recall model, input information of the trained deep semantic recall model is first obtained based on the user identifier (for example, user-material interaction information associated with the user identifier).

[0075] The deep semantic recall model extracts textual semantic features from input information, achieving a deep representation of users and items in a unified vector space. Specifically, textual semantic features include both latent user and item semantic representations. After extracting these features, the model can recall multiple candidate items based on factors such as complementarity, group identity, and similarity, thereby expanding the set of candidate items to be recalled.

[0076] The following are some of the candidate materials that can be expanded:

[0077] The first category is expandable complementary materials, such as core materials and dependent materials. Users prefer mobile phones (core materials) and can recall mobile phone accessories (dependent materials) through semantics.

[0078] The second category is expandable group common materials. For example, the common materials of the white-collar group are digital products / cars. Then the materials that the white-collar group has common preferences for can be recalled from each other through semantics.

[0079] The third category can be expanded to display dissimilar materials. For example, beer and diapers do not have any similarities in terms of material attributes, but similarities can be found from user behavior feedback, which can be achieved through semantic recall.

[0080] The fourth category is to expand the preference materials of users with similar behavior sequences. For example, users with similar behaviors will express similarity in semantics, so the preference materials of other users with similar behaviors can be expanded.

[0081] In the embodiment of the present application, users and materials are deeply represented in a unified vector space based on user-material pairs, which can deeply mine hidden correlation semantics and make recall more abundant.

[0082] In some embodiments, user-material interaction information includes: attribute information of multiple materials with which the user has historically interacted; extracting the semantic feature representations corresponding to the attribute information of the multiple materials through a deep semantic recall model, and performing averaging processing on the obtained multiple semantic feature representations to obtain text semantic features.

[0083] Specifically, material attribute information includes textual information such as the material title and its logo. This textual information is semantically represented using a deep semantic recall model. This model can be a pre-trained model that uses a massive corpus for self-supervised learning. Because it utilizes a sufficient amount of observed data, it can provide a good semantic representation of text, achieving excellent semantic representation results in carousel recommendations. The pre-trained model can then be fine-tuned to achieve the same results as the pre-trained model.

[0084] Step S303: Determine preference scores of the multiple candidate materials and the recalled materials of other routes associated with the user identifier through a deep semantic ranking model.

[0085] Specifically, the deep semantic sorting model is obtained by training based on historical user-material interaction information. During the training process of the deep semantic sorting model, the deep semantic sorting model learns to perform behavioral semantic representation of historical user-material interaction information, where the behavioral semantic representation includes the latent semantic representation of user preferences and the latent semantic representation of material interest points.

[0086] The network architecture of the deep semantic ranking model is shown in Figure 5. Using pre-collected historical user-item interaction information from multiple users as training samples, the model is learned end-to-end using either single-objective or multi-objective learning algorithms. Single-objective learning primarily considers user behavior in relation to a single objective, such as learning a model solely for clicks. Multi-objective learning considers user behavior in relation to multiple objectives, such as simultaneously learning models for both clicks and purchases. Learning across different objectives allows for a richer and more robust representation of user preferences for different objectives. For example, the learned latent vector represents a user's click preference, or a learned latent vector represents a user's click-and-purchase preference. Single-objective learning algorithms include DCN, DIN, and DNN, while multi-objective algorithms include ESMM and MMOE. Neural network algorithms, through multi-layered learning, can represent users' long-term and short-term item preferences and points of interest in a latent semantic manner.

[0087] After the training is completed, the input information of the trained deep semantic sorting model is obtained based on the user ID (for example, the user-material interaction information associated with the user ID), the behavioral semantic features are extracted, and based on the behavioral semantic features, the preference scores of multiple candidate materials and recalled materials of other routes are determined.

[0088] In some embodiments, material recall is performed based on a timeliness strategy, a popularity strategy, and a relevance strategy, respectively, to obtain recalled materials from other channels associated with the user identifier. Of course, recalled materials from other channels are not limited to recalls based on the aforementioned strategies; they can also be obtained through new and hot recall, collaborative filtering, content-based model recall, and other means, which are not specifically limited in this application.

[0089] In some embodiments, the deep semantic sorting model includes a material vector query table; the user-material interaction information includes at least: multiple historical behavior information of the user interacting with the material; from the material vector query table, the semantic feature representations corresponding to each of the multiple historical behavior information are queried and the multiple semantic feature representations obtained are averaged to obtain behavioral semantic features.

[0090] Specifically, after training data through a deep semantic ranking model, an intermediate product, namely an item vector lookup table, is generated. This item vector lookup table is then queried based on multiple historical behavior information to obtain multiple semantic features. These features are then averaged to calculate behavioral semantic features. These behavioral semantic features include latent semantic representations of user preferences and item points of interest.

[0091] Step S304: Based on the obtained preference scores, at least one recommended material is determined from the multiple candidate materials and the recalled materials of other routes.

[0092] In some embodiments, multiple candidate materials and recalled materials from other routes are ranked based on the obtained preference scores to obtain a first ranking result. The first ranking result is re-ranked based on a preset business policy to obtain a second ranking result. Based on the second ranking result, at least one recommended material is selected from the multiple candidate materials and recalled materials from other routes.

[0093] Specifically, multiple candidate materials and recalled materials from other routes are sorted in descending order of preference score to obtain a first sorting result. The preset business policy can be adjusted based on actual business needs. After re-sorting the first sorting result based on the preset business policy to obtain a second sorting result, the top N materials in the second sorting result are selected as recommended materials, where N is a positive integer greater than 0.

[0094] In some embodiments, based on the obtained preference scores, multiple candidate materials and recalled materials from other routes are sorted to obtain a first sorting result. Then, based on the first sorting result, at least one recommended material is selected from the multiple candidate materials and recalled materials from other routes.

[0095] Specifically, multiple candidate materials and recalled materials from other routes are sorted in descending order of preference scores to obtain a first sorting result. The first N materials in the first sorting result are used as recommended materials.

[0096] In an embodiment of the present application, a deep semantic recall model and a deep semantic sorting model are pre-trained based on historical user-material interaction information; an information recommendation request carrying a user identifier sent by a business system is received, and multiple candidate materials corresponding to the user identifier are automatically recalled through the deep semantic recall model; the preference scores of the multiple candidate materials and other recalled materials associated with the user identifier are automatically determined and sorted through the deep semantic sorting model, and then materials are recommended based on the sorting results, eliminating dependence on operational experience and enhancing the recommendation effect.

[0097] During the training of the deep semantic recall model, the deep semantic recall model learns to represent historical user-item interaction information in a textual semantic manner. Therefore, when recalling items using the deep semantic recall model, it pays more attention to the literal semantics of the items themselves. This allows candidate items to be better expanded in terms of literal semantics, which can improve the recalled items. During the training of the deep semantic ranking model, the deep semantic ranking model learns to represent historical user-item interaction information in a behavioral semantic manner. Therefore, when using the deep semantic ranking model to perform preference scoring and ranking of each recalled item, it pays more attention to user behavior information at the item feedback level, which can better improve user click-through rate. In this application, the two semantic representations act at different stages and can complement each other, eliminating the reliance on operational experience, enhancing the recommendation effect, and achieving personalized recommendation effects. Secondly, the recommendation method of this application uses a deep semantic model to conduct a deep semantic understanding of past user-item interaction behaviors, thereby learning the semantic representation of user preferences and the semantic representation of item points of interest. Finally, it calculates the semantic matching between user preferences and item points of interest to complete accurate recommendations. This method can better mine long-tail items, meet niche personalized needs, and thus create higher business conversion rates. In addition, the recommendation method of this application eliminates the reliance on explicit user / material portraits, and instead makes personalized recommendations based on implicit semantic portrait features such as text semantic features and behavioral semantic features, making the representation of users and materials in specified scenarios more accurate and timely.

[0098] Based on the same technical concept, an embodiment of the present application provides a schematic structural diagram of an information recommendation device, as shown in FIG6 . The device 600 includes:

[0099] Receiving module 601, configured to receive an information recommendation request carrying a user identifier sent by a business system;

[0100] A recall module 602 is configured to recall multiple candidate materials corresponding to the user identifier using a deep semantic recall model;

[0101] A sorting module 603 is configured to determine preference scores for the plurality of candidate items and recalled items of other routes associated with the user identifier using a deep semantic sorting model, wherein the deep semantic recall model and the deep semantic sorting model are trained based on historical user-item interaction information;

[0102] The recommendation module 604 is configured to determine at least one recommended material from the multiple candidate materials and the recalled materials on other routes based on the obtained preference scores.

[0103] Optionally, the recall module 602 is further configured to:

[0104] Before determining the preference scores of the multiple candidate materials and the recalled materials of other routes associated with the user identifier through a deep semantic sorting model, material recall is performed based on timeliness strategy, popularity strategy and relevance strategy to obtain recalled materials of other routes associated with the user identifier.

[0105] Optionally, the sorting module 603 is specifically configured to:

[0106] sorting the plurality of candidate materials and recalled materials from other routes based on the obtained preference scores to obtain a first sorting result;

[0107] Reordering the first sorting result based on a preset business policy to obtain a second sorting result;

[0108] Based on the second sorting result, at least one recommended material is selected from the multiple candidate materials and the recalled materials of other routes.

[0109] Optionally, the sorting module 603 is specifically configured to:

[0110] sorting the plurality of candidate materials and recalled materials from other routes based on the obtained preference scores to obtain a first sorting result;

[0111] Based on the first sorting result, at least one recommended material is selected from the multiple candidate materials and the recalled materials of other routes.

[0112] Optionally, the recommendation module 604 is further configured to:

[0113] Based on the obtained preference scores, after determining at least one recommended material from the multiple candidate materials and recalled materials from other routes, the material identification list corresponding to the at least one recommended material is sent to the business system, so that the business system obtains corresponding material detail information from the material system based on the material identification list, and sends the obtained material detail information to the terminal device for display.

[0114] The historical user-material interaction information includes: attribute information of multiple materials with which the user has historically interacted, and / or multiple historical behavior information of the user's interaction with the materials.

[0115] Optionally, the text semantic representation includes: user latent semantic representation and material latent semantic representation; the behavior semantic representation includes: user preference latent semantic representation and material interest point latent semantic representation.

[0116] In an embodiment of the present application, a deep semantic recall model and a deep semantic sorting model are pre-trained based on historical user-material interaction information; an information recommendation request carrying a user identifier sent by a business system is received, and multiple candidate materials corresponding to the user identifier are automatically recalled through the deep semantic recall model; the preference scores of the multiple candidate materials and other recalled materials associated with the user identifier are automatically determined and sorted through the deep semantic sorting model, and then materials are recommended based on the sorting results, eliminating dependence on operational experience and enhancing the recommendation effect.

[0117] In the process of training the deep semantic recall model, the deep semantic recall model learns to make textual semantic representations of historical user-material interaction information. Therefore, when the deep semantic recall model is used to recall materials, more attention is paid to the literal semantics of the materials themselves, so that the candidate materials can be better expanded in terms of literal semantics, which can improve the recalled materials. In the process of training the deep semantic ranking model, the deep semantic ranking model learns to make behavioral semantic representations of historical user-material interaction information. Therefore, when the deep semantic ranking model is used to perform preference scoring and ranking for each recalled material, more attention is paid to user behavior information at the material feedback level, which can better improve user click-through rate. In this application, the two semantic representations act at different stages, can complement each other, eliminate dependence on operational experience, enhance recommendation effects, and achieve personalized recommendation effects.

[0118] Based on the same technical concept, an embodiment of the present application provides a computer device, which may be the recommendation system shown in Figure 1. As shown in Figure 7, the computer device includes at least one processor 701 and a memory 702 connected to the at least one processor. The specific connection medium between the processor 701 and the memory 702 is not limited in the embodiment of the present application. In Figure 7, the processor 701 and the memory 702 are connected via a bus as an example. Buses can be divided into address buses, data buses, control buses, etc.

[0119] In the embodiment of the present application, the memory 702 stores instructions that can be executed by at least one processor 701. The at least one processor 701 can perform the steps of the above-mentioned information recommendation method by executing the instructions stored in the memory 702.

[0120] Among them, the processor 701 is the control center of the computer device, which can use various interfaces and lines to connect various parts of the computer device, and realize information recommendation by running or executing instructions stored in the memory 702 and calling data stored in the memory 702. Optionally, the processor 701 may include one or more processing units. The processor 701 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application programs, etc., and the modem processor mainly processes wireless communications. It is understandable that the above-mentioned modem processor may not be integrated into the processor 701. In some embodiments, the processor 701 and the memory 702 may be implemented on the same chip. In some embodiments, they may also be implemented separately on independent chips.

[0121] The processor 701 can be a general-purpose processor, such as a central processing unit (CPU), a digital signal processor, an application-specific integrated circuit (ASIC), a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component, and can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present application. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of the present application can be directly embodied as being executed by a hardware processor, or can be executed by a combination of hardware and software modules in the processor.

[0122] The memory 702 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules. The memory 702 may include at least one type of storage medium, such as a flash memory, a hard disk, a multimedia card, a card-type memory, a random access memory (Random Access Memory, RAM), a static random access memory (Static Random Access Memory, SRAM), a programmable read-only memory (Programmable Read Only Memory, PROM), a read-only memory (Read Only Memory, ROM), an electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, EEPROM), a magnetic memory, a disk, an optical disk, etc. The memory 702 is any other medium that can be used to carry or store a desired program code in the form of an instruction or data structure and can be accessed by a computer device, but is not limited thereto. The memory 702 in the embodiment of the present application can also be a circuit or any other device that can realize a storage function, for storing program instructions and / or data.

[0123] Based on the same inventive concept, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program that can be executed by a computer device. When the program runs on the computer device, the computer device executes the steps of the above-mentioned information recommendation method.

[0124] Based on the same inventive concept, an embodiment of the present application provides a computer program product, which includes a computer program stored on a computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer device, the computer device performs the steps of the above-mentioned information recommendation method.

[0125] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0126] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer device or other programmable data processing device produce a device for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0127] These computer program instructions may also be stored in a computer-readable memory that can direct a computer device or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0128] These computer program instructions can also be loaded onto a computer device or other programmable data processing device, so that a series of operating steps are executed on the computer device or other programmable device to produce a process implemented by the computer device, so that the instructions executed on the computer device or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.

[0129] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0130] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. An information recommendation method, wherein: include: Receiving an information recommendation request carrying a user identifier sent by a business system; Recalling multiple candidate materials corresponding to the user identifier through a deep semantic recall model; Determining preference scores of the plurality of candidate materials and recalled materials of other routes associated with the user identifier by using a deep semantic sorting model, wherein the deep semantic recall model and the deep semantic sorting model are obtained by training based on historical user-material interaction information; Based on the obtained preference scores, at least one recommended material is determined from the multiple candidate materials and the recalled materials of other routes.

2. The method of claim 1, wherein: Before determining the preference scores of the plurality of candidate materials and the recalled materials of other routes associated with the user identifier by using the deep semantic sorting model, the method further includes: Material recall is performed based on the timeliness strategy, the popularity strategy and the correlation strategy respectively, and the recalled materials of other roads associated with the user identifier are obtained.

3. The method of claim 1, wherein: The step of determining at least one recommended material from the plurality of candidate materials and the recalled materials of other routes based on the obtained preference scores comprises: Based on the obtained preference scores, the plurality of candidate materials and the recalled materials of other routes are sorted to obtain a first sorting result; Based on a preset business policy, reorder the first sorting result to obtain a second sorting result; Based on the second sorting result, at least one recommended material is selected from the multiple candidate materials and the recalled materials of other routes.

4. The method of claim 1, wherein: The step of determining at least one recommended material from the plurality of candidate materials and the recalled materials of other routes based on the obtained preference scores comprises: Based on the obtained preference scores, the plurality of candidate materials and the recalled materials of other routes are sorted to obtain a first sorting result; Based on the first sorting result, at least one recommended material is selected from the multiple candidate materials and the recalled materials of other routes.

5. The method of claim 1, wherein: After determining at least one recommended material from the plurality of candidate materials and the recalled materials of other routes based on the obtained preference scores, the method further includes: The material identification list corresponding to the at least one recommended material is sent to the business system, so that the business system obtains corresponding material detail information from the material system based on the material identification list, and sends the obtained material detail information to the terminal device for display.

6. The method according to any one of claims 1 to 5, wherein: Also includes: In the process of training the deep semantic recall model, the deep semantic recall model is enabled to learn to perform text semantic representation on the historical user-material interaction information; In the process of training the deep semantic sorting model, the deep semantic sorting model is enabled to learn to perform behavioral semantic representation of the historical user-material interaction information.

7. The method of claim 6, wherein: The historical user-material interaction information includes: attribute information of multiple materials with which the user has historically interacted, and / or multiple historical behavior information of the user's interaction with the materials.

8. The method of claim 6, wherein: The text semantic representation includes: user latent semantic representation and material latent semantic representation; the behavior semantic representation includes: user preference latent semantic representation and material interest point latent semantic representation.

9. An information recommendation device, wherein: include: A receiving module, used for receiving an information recommendation request carrying a user identifier sent by a business system; A recall module, used to recall multiple candidate materials corresponding to the user identifier through a deep semantic recall model; A sorting module, used to determine the preference scores of the plurality of candidate materials and the recalled materials of other routes associated with the user identifier by using a deep semantic sorting model, wherein the deep semantic recall model and the deep semantic sorting model are obtained by training based on historical user-material interaction information; The recommendation module is used to determine at least one recommended material from the multiple candidate materials and the recalled materials of other routes based on the obtained preference scores.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of any one of the methods of claims 1 to 8 are implemented.

11. A computer-readable storage medium, wherein: It stores a computer program executable by a computer device. When the program is run on the computer device, the computer device executes the steps of any method described in claims 1 to 8.

12. A computer program product, wherein: The computer program product comprises a computer program stored on a computer-readable storage medium, wherein the computer program comprises program instructions, and when the program instructions are executed by a computer, the computer is caused to execute the steps of the method according to any one of claims 1 to 8.

Images