CN111680221A

CN111680221A - Information recommendation method, device, equipment and computer readable storage medium

Info

Publication number: CN111680221A
Application number: CN202010803019.5A
Authority: CN
Inventors: 胡乐
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2020-08-11
Filing date: 2020-08-11
Publication date: 2020-09-18
Anticipated expiration: 2040-08-11
Also published as: CN111680221B

Abstract

The embodiment of the application provides an information recommendation method, an information recommendation device, information recommendation equipment and a computer-readable storage medium, wherein the method comprises the following steps: responding to a received information recommendation request, and acquiring a user identifier carried in the information recommendation request; acquiring user characteristics corresponding to the user identification, a plurality of pieces of information to be recommended and information characteristics of the plurality of pieces of information to be recommended; obtaining object characteristics of recommended objects carried in the information to be recommended, wherein the object characteristics at least comprise price characteristics of the recommended objects; determining target recommendation information from the plurality of information to be recommended based on the user characteristics, the information characteristics and the object characteristics; and sending the recommendation response carrying the target recommendation information to a user terminal. By the method and the device, the user characteristics, the information characteristics and the object characteristics including the price characteristics are combined to perform information recommendation, so that the accuracy of information recommendation can be improved, and accurate recommendation of information is realized.

Description

Information recommendation method, device, equipment and computer readable storage medium

Technical Field

The embodiment of the application relates to the technical field of information recommendation, and relates to but is not limited to an information recommendation method, an information recommendation device, information recommendation equipment and a computer-readable storage medium.

Background

With the development of internet technology and the popularization of mobile terminals, people can acquire various information anytime and anywhere by using various application programs installed in the mobile terminals. The method for recommending information which may be interesting to users in a targeted manner has become one of the services which are of great interest to many network platforms nowadays, for example, a news platform needs to recommend news which may be interesting to users, an audio and video resource playing platform needs to recommend audio and/or video resources which may be interesting to users, a shopping platform needs to recommend commodity information which may be interesting to users, and the like.

At present, generally, a ranking model in a recommendation system is used, the click rate of information to be recommended is predicted and ranked based on user characteristics and recommendation information characteristics, and a high score is used as a priority recommendation object.

Disclosure of Invention

The embodiment of the application provides an information recommendation method, an information recommendation device, information recommendation equipment and a computer-readable storage medium, and the accuracy of information recommendation can be improved by combining user characteristics, information characteristics and object characteristics including price characteristics to perform information recommendation.

The technical scheme of the embodiment of the application is realized as follows:

the embodiment of the application provides an information recommendation method, which comprises the following steps:

responding to a received information recommendation request, and acquiring a user identifier carried in the information recommendation request;

acquiring user characteristics corresponding to the user identification, a plurality of pieces of information to be recommended and information characteristics of the plurality of pieces of information to be recommended;

obtaining object characteristics of recommended objects carried in the information to be recommended, wherein the object characteristics at least comprise price characteristics of the recommended objects;

inputting the user characteristics, the information characteristics and the object characteristics into a trained ranking model to obtain a ranking result of the plurality of information to be recommended;

determining at least one piece of target recommendation information from the plurality of pieces of information to be recommended based on the sorting result; and sending the recommendation response carrying the target recommendation information to a user terminal.

An embodiment of the present application provides an information recommendation device, including:

the first obtaining module is used for responding to the received information recommendation request and obtaining a user identifier carried in the information recommendation request;

the second obtaining module is used for obtaining the user characteristics corresponding to the user identification, a plurality of pieces of information to be recommended and the information characteristics of the plurality of pieces of information to be recommended;

a third obtaining module, configured to obtain object features of recommended objects carried in the multiple pieces of information to be recommended, where the object features at least include price features of the recommended objects;

the input module is used for inputting the user characteristics, the information characteristics and the object characteristics into a trained ranking model to obtain a ranking result of the plurality of pieces of information to be recommended;

the first determining module is used for determining at least one piece of target recommendation information from the plurality of pieces of information to be recommended based on the sorting result;

and the sending module is used for sending the recommendation response carrying the target recommendation information to the user terminal.

In some embodiments, the apparatus further comprises:

a fourth obtaining module, configured to obtain historical behavior data, user attribute features, and user context features corresponding to the user identifier;

the second determining module is used for determining user portrait characteristics and target object identifications corresponding to the user identifications based on the historical behavior data, wherein the target object identifications are object identifications of recommended objects clicked by users or converted by the users;

a fifth obtaining module, configured to obtain a historical object feature corresponding to the target object identifier;

and the third determination module is used for determining the user characteristics corresponding to the user identification based on the user attribute characteristics, the user context characteristics, the user portrait characteristics and the historical object characteristics.

In some embodiments, the third obtaining module is further configured to:

acquiring object identifications of recommended objects corresponding to information to be recommended;

acquiring object information corresponding to the object identification from a standard object information base;

and extracting the object characteristics of the recommended object from the object information.

In some embodiments, the third obtaining module is further configured to:

extracting at least an object identifier, a standard type and an object brand of the recommended object from the object information;

acquiring the object price of the recommended object and price classification information corresponding to the standard type;

determining a price index of the recommended object based on the object price and the price classification information, wherein the price index is used for representing the degree of the object price in the same class;

and determining the price of the object and the price index as the price characteristic of the recommended object.

In some embodiments, the apparatus further comprises:

a sixth obtaining module, configured to obtain a sample object library and a preset candidate type set, where the sample object library includes multiple sample objects and sample types corresponding to the multiple sample objects;

a fourth determining module, configured to determine a union of the sample type set and the candidate type set included in the sample object library as a standard type set;

a seventh obtaining module, configured to obtain target sample information of each sample object in the sample object library, where the target sample information is sample information other than a sample type;

and the training module is used for training the object classification model by utilizing the target sample information and the standard type set to obtain a trained object classification model.

In some embodiments, the apparatus further comprises:

an eighth obtaining module, configured to obtain an original object information base provided by a provider, where the original object information base includes original object information of multiple recommended objects;

a fifth determining module, configured to determine, as each predicted object information, object information obtained by removing the object type of each recommended object from each piece of original object information;

the classification module is used for inputting the information of each predicted object into a trained object classification model to obtain the standard type of each recommended object;

and the information base construction module is used for constructing a standard object information base based on the standard type and the original object information of each recommended object.

In some embodiments, the apparatus further comprises:

a ninth obtaining module, configured to obtain object information corresponding to each standard type in a standard object information base;

a sixth determining module, configured to determine, based on the object information corresponding to each standard type, a lowest price and a highest price corresponding to each standard type;

and the seventh determining module is used for determining the price classification information of each standard type based on the lowest price, the highest price and the preset interval number of each standard type.

In some embodiments, the first determining module is further configured to:

acquiring information recommendation duration corresponding to the information recommendation request;

and determining at least one piece of target recommendation information from the plurality of pieces of information to be recommended based on the information recommendation duration and the sorting result.

a memory for storing executable instructions; and the processor is used for realizing the method when executing the executable instructions stored in the memory.

Embodiments of the present application provide a computer-readable storage medium storing executable instructions for causing a processor to implement the above-mentioned method when executed.

The embodiment of the application has the following beneficial effects:

after receiving the information recommendation request, the information recommendation server acquires the user characteristics corresponding to the user identification carried in the information recommendation request and the information characteristics of the plurality of pieces of information to be recommended, and acquires the object characteristics of the recommendation objects in the plurality of pieces of information to be recommended, especially the price characteristics, so that the target recommendation information is determined from the plurality of pieces of information to be recommended by combining the user characteristics, the information characteristics and the object characteristics, and thus the target recommendation information can be ensured to not only accord with the user characteristics, but also accord with the consumption habits of the user, the accuracy of information recommendation can be improved, and the starting rate and the conversion rate of information recommendation can be improved.

Drawings

FIG. 1 is a schematic diagram illustrating a flow of advertisement prediction using a ranking model in the related art;

fig. 2A is a schematic network architecture diagram of an information recommendation system 20 provided in an embodiment of the present application;

fig. 2B is an alternative structural diagram of the information recommendation system 20 applied to the blockchain system according to the embodiment of the present application;

FIG. 2C is an alternative block diagram according to an embodiment of the present disclosure;

fig. 3 is a schematic structural diagram of an information recommendation server 300 provided in an embodiment of the present application;

fig. 4 is a schematic flow chart of an implementation of an information recommendation method according to an embodiment of the present application;

FIG. 5 is a schematic diagram of an implementation process for building a standard object information base according to an embodiment of the present application;

fig. 6 is a schematic flow chart of still another implementation of the information recommendation method according to the embodiment of the present application;

fig. 7A is a schematic flowchart of another implementation process of the information recommendation method according to the embodiment of the present application;

FIG. 7B is a schematic diagram of an interface for an advertiser to set commodity data according to an embodiment of the present disclosure;

FIG. 7C is an interface schematic of a merchandise detail page;

FIG. 8 is a schematic flow chart illustrating implementation of ranking prediction using a ranking model according to an embodiment of the present disclosure;

FIG. 9 is a flowchart illustrating mapping a plurality of commodity libraries to a unified category system according to an embodiment of the present application;

FIG. 10 is a schematic diagram of building a commodity knowledge base and a uniform category according to an embodiment of the present application;

FIG. 11A is a diagram of the results of a ranking model prediction with only commodity price features added;

FIG. 11B is a diagram illustrating the prediction results of the ranking model with added price index features;

fig. 12 is a schematic diagram of a variation curve of an advertisement volume rate when information recommendation is performed by using the information recommendation method provided in the embodiment of the present application.

Detailed Description

In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be described in further detail with reference to the attached drawings, the described embodiments should not be considered as limiting the present application, and all other embodiments obtained by a person of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

In the following description, reference is made to "some embodiments" which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiments of the present application belong. The terminology used in the embodiments of the present application is for the purpose of describing the embodiments of the present application only and is not intended to be limiting of the present application.

Scientific terms related to the embodiments of the present application are explained below.

1) A sequencing model: the most core part of the information recommendation system is a model for sequencing all recommendation information according to the user and the context condition in a given recommendation information queue;

2) and (3) commodity advertisement: a special advertisement form endows the advertisement with commodity attributes, so that the advertisement has real commodity meaning.

3) Click Through Rate (CTR, Click-Through-Rate) is an important index for measuring the effect of internet advertisements, and refers to the Click arrival Rate of network advertisements (picture advertisements/text advertisements/keyword advertisements/ranking advertisements/video advertisements, etc.), i.e. the actual Click times of the advertisements are divided by the advertisement display amount (Show content);

4) conversion Rate (CVR), which is an index for measuring the effectiveness of CPA advertisements, refers to the Conversion Rate from the time when a user clicks an advertisement to the time when the user becomes an active user or even a paid user;

5) and the advertisement recall refers to a process of acquiring the advertisement according to the input keyword group.

In order to better understand the information recommendation method provided in the embodiment of the present application, first, an information recommendation method in the related art is described:

at present, a recommendation system uses a ranking model (CTR prediction model and CVR prediction model) and uses user characteristics (user portrait, interest intention, context) and advertisement characteristics (advertisement material, title) as input, and obtains the prediction scores of the user and the advertisement pairwise through deep network model prediction, and the higher the prediction score is under the same bid condition, the higher the ranking of the advertisement is.

Fig. 1 is a schematic flow chart illustrating an implementation of advertisement prediction using a ranking model in the related art, as shown in fig. 1, the ranking model includes two sides (also called towers). The left tower inputs user information, the right tower inputs commodity information, and the implementation process comprises the following steps:

in step S001, feature information is input to the vector (embedding) layer.

Here, when the step S001 is implemented, inputting each user feature (such as gender, age, city, occupation, region, and the like) into the vector layer of the left tower, mapping each user feature into one vector, and splicing vectors corresponding to all user features to obtain a user feature vector; inputting each advertisement characteristic (such as advertisement id, advertiser information, advertisement material, advertisement pattern and the like) to a vector layer of the right tower, mapping each advertisement characteristic into a vector, and splicing vectors corresponding to all advertisement characteristics to obtain an advertisement characteristic vector.

And step S002, inputting the user characteristic vector into the full-connection network of the left tower to obtain the user characteristic vector with the preset dimension, and inputting the advertisement characteristic vector into the full-connection network of the right tower to obtain the advertisement characteristic vector with the preset dimension.

Here, as shown in fig. 1, the fully-connected network of the left tower and the fully-connected network of the right tower are both three-layer fully-connected networks, and when the step S002 is implemented, the user feature vector may be fully-connected by using the three-layer fully-connected network of the left tower, so as to obtain a 32-dimensional user feature vector; and carrying out full connection processing on the advertisement characteristic vector by utilizing a three-layer full connection network of a right tower to obtain a 32-dimensional advertisement characteristic vector.

And S003, calculating an inner product of the user characteristic vector with the preset dimension and the advertisement characteristic vector with the preset dimension to obtain an advertisement score.

Here, an inner product is calculated by using the user feature vector of the preset dimension and the advertisement feature vector of the preset dimension, the obtained value is the score of the advertisement to the user, and the higher the score is, the higher the recommendation degree of the commodity is.

Existing ranking models only consider the scores of advertisements on a particular user, while for commercial advertisements, the attributes on the advertisement do not fully reflect the characteristics of the commercial product, such as the price, category, etc. of the commercial product. Especially for different commodity advertisements with huge price difference, similar advertisement attributes are possible, and the predicted scores are almost the same under the existing model; in fact, the advertisement has great tendency and diversity for different people due to different commodities on the advertisement.

Under the effect of the existing model, the commercial scores are not dominant (the winning rate is low) in the sorting queue of the advertisement, and for particularly low-priced commercial advertisements (below 90 yuan), the cost expected by the advertiser is mostly below 40 yuan, which is lower than or equal to the conversion cost of most common advertisements. Due to the combination of the above two reasons, the advertisement of low-priced goods has a difficult role in the advertisement system.

Based on the above problems, in order to enable the ranking model to predict more accurate scores according to different commodities, in the embodiment of the present application, commodity features (including price features) are added to the ranking model to enhance the model effect.

An exemplary application of the information recommendation device provided in the embodiment of the present application is described below, and the information recommendation device provided in the embodiment of the present application may be implemented as a server. Next, an exemplary application when the information recommendation apparatus is implemented as an information recommendation server will be described.

Referring to fig. 2A, fig. 2A is a schematic diagram of a network architecture of an information recommendation system 20 according to an embodiment of the present application. As shown in fig. 2A, the information recommendation system 20 includes a user terminal 100, an application server 200, an information recommendation server 300, and a provider terminal 400. Communication connection is established between the user terminal 100 and the application server 200 through a network, and communication connection is established between the information recommendation server 300 and the application server 200, and between the provider terminal 400 through a network.

The user terminal 100 and the provider terminal 400 may be any terminals with an on-screen display function, such as a notebook computer, a tablet computer, a desktop computer, a mobile device (e.g., a mobile phone, a portable music player, a personal digital assistant, and a portable game device), and the user terminal 100 and the provider terminal 400 are illustrated as a smart phone and a desktop computer in fig. 2A, but the specific types of the two are not limited thereto. Various applications (apps) are installed in the user terminal 100, for example, an instant messaging App, a shopping App, a video playing App, etc., various apps are also installed in the provider terminal 400, in the embodiment of the present application, an advertisement delivery App may be installed in the supplier terminal, and the supplier of the recommended object may deliver the object information of the recommended object through the advertisement delivery App on the supplier terminal 400, and recommendation information corresponding to the recommended objects is uploaded to the information recommendation server 300, the information recommendation server 300 extracts object features based on the object information, and extracts information features based on the recommendation information, the information recommendation server 300 also extracts user features of the user in advance, and the object characteristics comprise price characteristics, and the user characteristics comprise the object characteristics of the recommended objects which are clicked historically or converted by the user besides the basic attribute characteristics of the user.

When a user watches a video by using a video playing App installed on a user terminal 100, or when the user starts a certain shopping App, an information recommendation request is triggered, the information recommendation request is sent to an information recommendation server 300 through an application server 200, after the information recommendation server 300 receives the information recommendation request, one or more target recommendation information is determined from a plurality of information to be recommended by combining user characteristics, information characteristics of the information to be recommended and object characteristics of a recommendation object in the information to be recommended, then the information recommendation server 300 sends the determined target recommendation information to the user terminal 100 through the application server 200, and the target recommendation information is presented in the user terminal 100.

In this embodiment, the application server 200 and the information recommendation server 300 may be independent physical servers, may also be a server cluster or a distributed system formed by a plurality of physical servers, and may also be cloud servers providing basic cloud computing services such as cloud services, a cloud database, cloud computing, a cloud function, cloud storage, network services, cloud communication, middleware services, domain name services, security services, and a big data and artificial intelligence platform, where the servers may be directly or indirectly connected in a wired or wireless communication manner, and the application is not limited herein.

The artificial intelligence cloud Service is also generally called AI as a Service (AIaaS). The method is a service mode of an artificial intelligence platform, and particularly, the AIaaS platform splits several types of common AI services and provides independent or packaged services at a cloud. This service model is similar to the one opened in an AI theme mall: all developers can access one or more artificial intelligence services provided by the platform through an API (application programming interface) interface, and part of the qualified developers can also use the AI framework and the AI infrastructure provided by the platform to deploy and operate and maintain the own dedicated cloud artificial intelligence services.

The information recommendation system 20 related To the embodiment of the present application may also be a distributed system 201 of a blockchain system, referring To fig. 2B, where fig. 2B is an optional structural schematic diagram of the information recommendation system 20 provided in the embodiment of the present application, where the distributed system 201 may be a distributed node formed by a plurality of nodes 202 (any form of computing devices in an access network, such as a server and a user terminal) and a client 203, a Peer-To-Peer (P2P, Peer To Peer) network is formed between the nodes, and the P2P Protocol is an application layer Protocol operating on a Transmission Control Protocol (TCP). In a distributed system, any machine, such as a server or a terminal, can join to become a node, and the node comprises a hardware layer, a middle layer, an operating system layer and an application layer.

It should be noted that, in the distributed system 201, each node 202 corresponds to a user terminal, and on the user terminal of each user, the interaction information and the interaction data of the user (for example, the type of information clicked by the user, the number of clicks, and the like) are collected, and further, the user portrait is depicted to determine the preference and habit of the user, so that, in combination with the method of the embodiment of the present application, the target recommendation information corresponding to each user terminal is accurately determined, so as to implement information recommendation for the user terminal.

In the block chain system, interactive information and interactive data of each user are recorded and can not be changed, and along with the updating of the interactive information and the interactive data of the users, the data stored in the block chain can be updated, so that portrait of the users can be updated in time, user characteristics can be updated in time, and further, when information recommendation is carried out, target recommendation information more suitable for the users can be matched based on the portrait of the depicted users (namely habits and preferences of the users), and accurate and efficient recommendation of the users is achieved.

Referring to the functions of each node in the blockchain system shown in fig. 2B, the functions related to each node in the blockchain system will be described in detail as follows:

1) routing, a basic function that a node has, is used to support communication between nodes. Besides the routing function, the node may also have the following functions:

2) the application is used for being deployed in a block chain, realizing specific services according to actual service requirements, recording data related to the realization functions to form recording data, carrying a digital signature in the recording data to represent a source of task data, and sending the recording data to other nodes in the block chain system, so that the other nodes add the recording data to a temporary block when the source and integrity of the recording data are verified successfully. For example, the services implemented by the application include: 2.1) wallet, for providing the function of transaction of electronic money, including initiating transaction (i.e. sending the transaction record of current transaction to other nodes in the blockchain system, after the other nodes are successfully verified, storing the record data of transaction in the temporary blocks of the blockchain as the response of confirming the transaction is valid; of course, the wallet also supports the querying of the electronic money remaining in the electronic money address. And 2.2) sharing the account book, wherein the shared account book is used for providing functions of operations such as storage, query and modification of account data, record data of the operations on the account data are sent to other nodes in the block chain system, and after the other nodes verify the validity, the record data are stored in a temporary block as a response for acknowledging that the account data are valid, and confirmation can be sent to the node initiating the operations. 2.3) Intelligent contracts, computerized agreements, which can enforce the terms of a contract, implemented by codes deployed on a shared ledger for execution when certain conditions are met, for completing automated transactions according to actual business requirement codes, such as querying the logistics status of goods purchased by a buyer, transferring the buyer's electronic money to the merchant's address after the buyer signs for the goods; of course, smart contracts are not limited to executing contracts for trading, but may also execute contracts that process received information.

3) And the Block chain comprises a series of blocks (blocks) which are mutually connected according to the generated chronological order, new blocks cannot be removed once being added into the Block chain, and recorded data submitted by nodes in the Block chain system are recorded in the blocks.

4) Consensus (Consensus), a process in a blockchain network, is used to agree on transactions in a block among a plurality of nodes involved, the agreed block is to be appended to the end of the blockchain, and the mechanisms for achieving Consensus include Proof of workload (PoW, Proof of Work), Proof of rights and interests (PoS, Proof of equity (DPoS), Proof of granted of shares (DPoS), Proof of Elapsed Time (PoET, Proof of Elapsed Time), and so on.

Referring to fig. 2C, fig. 2C is an optional schematic diagram of a Block Structure (Block Structure) provided in this embodiment, each Block includes a hash value of a transaction record (hash value of the Block) stored in the Block and a hash value of a previous Block, and the blocks are connected by the hash values to form a Block chain. The block may include information such as a time stamp at the time of block generation. A block chain (Blockchain), which is essentially a decentralized database, is a string of data blocks associated by using cryptography, and each data block contains related information for verifying the validity (anti-counterfeiting) of the information and generating a next block.

Referring to fig. 3, fig. 3 is a schematic structural diagram of an information recommendation server 300 according to an embodiment of the present application, where the information recommendation server 300 shown in fig. 3 includes: at least one processor 310, memory 350, at least one network interface 320, and a user interface 330. The various components in the information recommendation server 300 are coupled together by a bus system 340. It will be appreciated that the bus system 340 is used to enable communications among the components connected. The bus system 340 includes a power bus, a control bus, and a status signal bus in addition to a data bus. For clarity of illustration, however, the various buses are labeled as bus system 340 in fig. 3.

The Processor 310 may be an integrated circuit chip having Signal processing capabilities, such as a general purpose Processor, a Digital Signal Processor (DSP), or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, or the like, wherein the general purpose Processor may be a microprocessor or any conventional Processor, or the like.

The user interface 330 includes one or more output devices 331, including one or more speakers and/or one or more visual display screens, that enable presentation of media content. The user interface 330 also includes one or more input devices 332, including user interface components to facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.

The memory 350 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid state memory, hard disk drives, optical disk drives, and the like. Memory 350 optionally includes one or more storage devices physically located remote from processor 310. The memory 350 may include either volatile memory or nonvolatile memory, and may also include both volatile and nonvolatile memory. The nonvolatile Memory may be a Read Only Memory (ROM), and the volatile Memory may be a Random Access Memory (RAM). The memory 350 described in embodiments herein is intended to comprise any suitable type of memory. In some embodiments, memory 350 is capable of storing data, examples of which include programs, modules, and data structures, or subsets or supersets thereof, as exemplified below, to support various operations.

An operating system 351 including system programs for processing various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks;

a network communication module 352 for communicating to other computing devices via one or more (wired or wireless) network interfaces 320, exemplary network interfaces 320 including: bluetooth, wireless compatibility authentication (WiFi), and Universal Serial Bus (USB), etc.;

an input processing module 353 for detecting one or more user inputs or interactions from one of the one or more input devices 332 and translating the detected inputs or interactions.

In some embodiments, the apparatus provided by the embodiments of the present application may be implemented in software, and fig. 3 illustrates an information recommendation apparatus 354 stored in the memory 350, where the information recommendation apparatus 354 may be an information recommendation apparatus in the information recommendation server 300, and may be software in the form of programs and plug-ins, and includes the following software modules: the first obtaining module 3541, the second obtaining module 3542, the third obtaining module 3543, the input module 3544, the first determining module 3545, and the sending module 3546, which are logical and thus may be arbitrarily combined or further separated depending on the functionality implemented. The functions of the respective modules will be explained below.

In other embodiments, the apparatus provided in the embodiments of the present Application may be implemented in hardware, and for example, the apparatus provided in the embodiments of the present Application may be a processor in the form of a hardware decoding processor, which is programmed to execute the information recommendation method provided in the embodiments of the present Application, for example, the processor in the form of the hardware decoding processor may be one or more Application Specific Integrated Circuits (ASICs), DSPs, Programmable Logic Devices (PLDs), Complex Programmable Logic Devices (CPLDs), Field Programmable Gate Arrays (FPGAs), or other electronic components.

The information recommendation method provided by the embodiment of the present application will be described below with reference to an exemplary application and implementation of the information recommendation server 300 provided by the embodiment of the present application. Referring to fig. 4, fig. 4 is a schematic flow chart of an implementation of an information recommendation method provided in an embodiment of the present application, where the information recommendation method is applied to an information recommendation server, and the following description is provided with reference to the steps shown in fig. 4.

Step S101, responding to the received information recommendation request, and acquiring the user identification carried in the information recommendation request.

Here, the information recommendation request may be sent by a user terminal, where the user terminal triggers the information recommendation request when opening a certain application program or when opening a certain video, the information recommendation request is sent to an application server corresponding to the application program first, and the application server sends the information recommendation request to the information recommendation server. The information recommendation request carries a user identifier, where the user identifier may be an identifier of a user when the user logs in an application program, and the user identifier has uniqueness. For example, for instant messaging applications, the user identification may be a user ID, rather than a nickname.

After receiving the information recommendation request, the information recommendation server obtains the user identifier carried in the information recommendation request, so as to determine user preference, user attributes and the like through the user identifier.

Step S102, obtaining a user characteristic corresponding to the user identification, a plurality of pieces of information to be recommended and information characteristics of the plurality of pieces of information to be recommended.

Here, the user characteristics may include user attribute characteristics, for example, a user name, an age, and a location corresponding to the user identifier, and also include user context characteristics, where the user context characteristics are context characteristics when the user initiates the recommendation request of this time, for example, a network type to which the user terminal belongs, a type of the user terminal, and an opportunity to trigger the recommendation request. The network types may include mobile network networks, wired networks, wireless networks, and the like; the types of user terminals may include smart phones, tablet computers, desktop computers, and the like; the opportunity for triggering the recommendation request can be to start App, watch videos, and the like.

The plurality of recommendation information may be recalled recommendation information. The recommendation information may be image recommendation information or video recommendation information, and the information characteristics of the recommendation information may include: information titles, information material, etc.

In the embodiment of the present application, the user characteristics may further include a history object characteristic, where the history object characteristic may be an object characteristic of a recommended object that the user clicks or has a conversion, and may include, for example, an object identifier, an object price, and the like. In the embodiment of the application, the historical object features are added to the user features, so that the consideration on the consumption habits of the user can be increased when information recommendation is carried out, and the accuracy of the information recommendation is further improved.

In some embodiments, the user characteristics may be extracted from the user attribute information and the historical behavior data of the user by the information recommendation server before receiving the information recommendation request.

Step S103, obtaining object characteristics of the recommendation objects carried in the information to be recommended.

Here, the recommendation object may be a commodity, and the object feature at least includes a price feature of the recommendation object, where the price feature includes an absolute price feature of the recommendation object and a relative price feature, where the absolute price feature represents a height of an absolute price of the recommendation object, and the relative price feature represents a height of a relative price of the recommendation object in a same type of product; in some embodiments, the object characteristics may also include object identification, object type, object brand, and the like.

And step S104, inputting the user characteristics, the information characteristics and the object characteristics into a trained ranking model to obtain a ranking result of the plurality of information to be recommended.

Here, the order model may be a neural network model, for example, a deep learning neural network model, and the order model may include at least a vector layer and a fully connected layer, and in a specific implementation, may include two towers, i.e., a left tower and a right tower, and include a vector layer and a fully connected layer at the two towers, respectively.

In step S102 and step S103, the obtained user characteristics, information characteristics, and object characteristics are also in an original format, for example, the user identifier may be 012345, the gender may be women, and the location may be beijing, etc. When the step S104 is implemented, the user features may be input to a vector layer of a tower, so as to obtain vector representation of the user features, and the vector representation of the user features is input to a full-link layer of the tower, so as to obtain a first vector of a preset dimension output by the full-link layer; inputting the information characteristics and the object characteristics into a vector layer of another tower to obtain vector representations of the information characteristics and the object characteristics, splicing the vector representations of the information characteristics and the object characteristics to obtain spliced vectors, inputting the spliced vectors into a full-connection layer of the tower to obtain a second vector with a preset dimension, performing inner product calculation on the first vector and the second vector to obtain prediction scores of the information to be recommended, and performing ascending or descending arrangement on the basis of the prediction scores of each piece of the information to be recommended to obtain a sorting result.

Step S105, determining at least one piece of target recommendation information from the plurality of pieces of information to be recommended based on the sorting result.

Here, when implemented, the step S105 may be to obtain, as the target recommendation information, the previous one or the previous ones of the information to be recommended from the sorting result based on the information recommendation time length corresponding to the information recommendation request, and the total playing time length of the at least one target recommendation information is greater than or equal to the information recommendation time length.

When the target recommendation information is screened for the plurality of pieces of information to be recommended, not only are the user characteristics and the information characteristics considered, but also the object characteristics of the recommendation object are fused, so that the recommendation information more suitable for the user can be provided.

And step S106, sending the recommendation response carrying the target recommendation information to the user terminal.

Here, when step S106 is implemented, the information recommendation server may send the recommendation response to the application server, and then the application server sends the recommendation response to the user terminal. In some embodiments, the user terminal obtains the target recommendation information carried in the recommendation response and presents the target recommendation information.

In the information recommendation method provided by the embodiment of the application, after receiving an information recommendation request, an information recommendation server obtains a user characteristic corresponding to a user identifier carried in the information recommendation request and information characteristics of a plurality of pieces of information to be recommended, and obtains object characteristics of recommended objects in the plurality of pieces of information to be recommended, especially price characteristics, so that target recommendation information is determined from the plurality of pieces of information to be recommended by combining the user characteristic, the information characteristics and the object characteristics, and thus, the target recommendation information can be ensured to conform to not only the user characteristics but also consumption habits of users, the accuracy of information recommendation can be improved, and the starting rate and the conversion rate of information recommendation can be improved.

In some embodiments, the step S105 "determining at least one target recommendation information from the plurality of information to be recommended based on the sorting result" may be implemented by:

step S1051, obtaining the information recommendation duration corresponding to the information recommendation request.

Here, the information recommendation request may carry an information recommendation duration, where the information recommendation duration may be determined based on a trigger timing of the information recommendation request, for example, when the user starts a certain App to trigger the information recommendation request, the recommendation duration may be 3 to 5 seconds, and when the user watches a video to trigger the information recommendation request, the recommendation duration may be 30 seconds to 90 seconds, for example, 65 seconds.

Step S1052, determining at least one target recommendation information from the plurality of information to be recommended based on the information recommendation duration and the sorting result.

Here, when the step S1052 is implemented, the first N pieces of recommendation information may be selected from the sorting result according to the recommendation time length, where the sum of the play time lengths of the first N-1 pieces of recommendation information is less than the information recommendation time length, the sum of the play time lengths of the first N pieces of recommendation information is greater than or equal to the recommendation time length, and the sum of the play time lengths of the current N pieces of recommendation information is greater than the information recommendation time length, the play speed of the last piece of recommendation information may be adjusted, so that the N pieces of recommendation information are played within the information recommendation time length.

In some embodiments, the training data of the ranking model during training also include user features, object features and information features, so that the ranking model can sense prices of different recommended objects, and therefore, when the trained ranking model is used for determining target recommendation information from information to be recommended, a ranking effect can be optimized.

In some embodiments, before step S101, the information recommendation server needs to extract user features corresponding to each user identifier and information features of each piece of recommendation information in advance, so that when an information recommendation request is received, target recommendation information can be provided to the user in time. The information features can be obtained by performing feature extraction on the recommended information; the extraction process of the user features can be realized by the following steps:

step S201, obtaining historical behavior data, user attribute characteristics, and user context characteristics corresponding to the user identifier.

Here, the user attribute characteristics may include a user identification, a gender, an age, a location, and the like, the user context characteristics may include a type of the user terminal, a type of a network used by the user terminal, a trigger timing of an information recommendation request, and the like, and the historical behavior data may include an article browsed by the user, a keyword searched for, an article purchased, recommendation information viewed, and the like.

Step S202, determining a user portrait characteristic and a target object identification corresponding to the user identification based on the historical behavior data.

Here, in step S202, when implemented, data analysis may be performed on the historical behavior data, so as to determine preference information of the user, so as to determine an image feature of the user, and the target object identifier is an object identifier of a recommended object that the user clicks or has a conversion, and by analyzing the historical behavior data, which recommended information the user clicks or purchases a recommended object corresponding to which recommended information, so as to determine the target object identifier.

Step S203, obtaining a history object feature corresponding to the target object identifier.

Here, in some embodiments, the information recommendation server may perform feature extraction on object information of recommendation objects stored in the information recommendation server, obtain object features of the recommendation objects, and store the object features of the recommendation objects in the feature repository. In implementation, step S203 may obtain corresponding historical object features from the feature repository based on the target object identifier. The historical object characteristics may include object identification, object type, and the like, and in some embodiments, the historical object characteristics may also include price characteristics of the recommended object.

And step S204, determining the user characteristics corresponding to the user identification based on the user attribute characteristics, the user context characteristics, the user portrait characteristics and the historical object characteristics.

Through the steps S201 to S204, the history object feature of the recommendation object clicked or converted by the user is added to the user feature, and the consumption level and consumption habit of the user can be determined through the history object feature, so that the accuracy of information recommendation is improved.

In some embodiments, the step S103 "obtaining the object features of the recommendation object carried in the plurality of pieces of information to be recommended" may be implemented by:

step S1031, obtaining object identifiers of recommendation objects corresponding to the information to be recommended.

Here, the object identification may be an object name or an object ID.

Step S1032, the object information corresponding to the object identifier is acquired from the standard object information base.

The standard object information base comprises standard types of the recommended objects, the standard types are obtained by performing type prediction on the recommended information except the object types in the recommended information of the recommended objects through a trained object classification model, and the trained object classification model can map the object types of the recommended objects provided by different suppliers to a unified category, so that the prices of similar recommended objects can be compared relatively.

Step S1033, extracts an object feature of the recommended object from the object information.

Here, step S1033 may be implemented by:

step S331, extracting at least an object identifier, a standard type, and an object brand of the recommended object from the object information.

Here, in implementation, the object identifier, the standard type, and the object brand of the recommended object may be extracted from fields corresponding to the object identifier, the standard type, and the object brand in the object information in step S331.

Step S332, obtaining the object price of the recommended object and the price classification information corresponding to the standard type.

Here, the object price of the recommended object may be obtained from a price field in the object information, and the object price may be an integer or a floating point number. The price classification information corresponding to each standard type may be determined according to the object prices of all recommended objects belonging to the same standard type in the standard object information base. The price classification information may be a correspondence between a price index and a price interval.

Step S333, based on the object price and the price classification information, determining the price index of the recommended object.

Here, the price index is used to represent the degree of the price of the object being expensive in the same category, and in the embodiment of the present application, the price index may be classified into 10 classes, where a higher price index indicates that the recommended object is more expensive in the same category, for example, the recommended object with the price index of 1 is cheap in the same category, and the recommended object with the price index of 10 is expensive in the same category.

Step S334, determining the object price and the price index as the price characteristic of the recommended object.

Here, the object price may be an absolute price of the recommended object; the price index may characterize the relative price of the recommended objects.

In some embodiments, the information recommendation server may extract in advance the object features of all recommended objects included in the standard object information base and store the object features in the object feature base, and at this time, in step S103, when implementing, the object identifier of the recommended object corresponding to each piece of information to be recommended may be first obtained, and then the object feature corresponding to the object identifier may be obtained from the object feature base.

In some embodiments, in order to determine the price characteristics of a recommended object more accurately, the recommended object provided by different suppliers needs to be mapped into a unified category system, the unified category system includes a plurality of standard types, and when the recommended object is mapped into the unified category system, the type of the recommended object can be predicted through a trained object classification model, so as to obtain the standard type of the recommended object.

In some embodiments, the trained object classification model can be obtained through the following steps S401 to S404, which are described below in conjunction with the respective steps.

In step S401, a sample object library and a preset candidate type set are obtained.

Here, the sample object library includes a plurality of sample objects and sample types corresponding to the plurality of sample objects. The sample object library can be a large platform object library with a complete commodity system, and the sample type can be an object category path to which the sample object belongs.

Step S402, determining a union of the sample type set and the candidate type set included in the sample object library as a standard type set.

Here, the candidate type set may include candidate types specified by a worker corresponding to the information recommendation server. The candidate type is generally a type that is not included in the set of sample types and is used to supplement the set of sample types.

Step S403, obtaining target sample information of each sample object in the sample object library.

Here, the target sample information is other sample information than the sample type.

And S404, training the object classification model by using the target sample information and the standard type set to obtain a trained object classification model.

Here, the object classification model may be a neural network model, and when the step S404 is implemented, the sample type of the target sample information may be determined first, then the target sample information is input into a preset object classification model for prediction processing to obtain a prediction type of the target sample information, and the object classification model is trained according to the sample type and the prediction type, that is, parameters of the object classification model are adjusted until the training target is reached, so as to obtain the trained object classification model.

Through the steps S401 to S404, the trained object classification model can be obtained, so that the trained object classification model can be used to map object libraries provided by different suppliers into a standard object library, so that all recommended objects are mapped to the same category system.

In some embodiments, based on the trained object classification model, a standard object information base can be constructed through steps S501 to S504 shown in fig. 5, which will be described below in conjunction with the steps.

In step S501, an original object information base provided by a supplier is acquired.

Here, the original object information base includes original object information of a plurality of recommended objects; in some embodiments, the original object information may be object information set on the information recommendation App by the supplier person through the supplier terminal, and may include an object identifier, an object brand, a material, a color, a price, a picture, and the like. After object information is set by a supplier through a supplier terminal, the supplier personnel can upload the object information to an information recommendation server, the information recommendation server establishes an object information base for each supplier, and the information recommendation server stores the received object information into the corresponding object information base.

Step S502, the object information obtained by removing the object type of each recommended object from each piece of original object information is determined as each piece of predicted object information.

Here, since the standard type of each recommended object needs to be predicted, the object type of the recommended object needs to be deleted from the original object information in step S502, so as to avoid interference of the original object type with the prediction result.

Step S503, inputting the information of each predicted object into the trained object classification model to obtain the standard type of each recommended object.

Step S504, a standard object information base is constructed based on the standard type and the original object information of each recommended object.

Here, when the step S504 is implemented, the standard type of each recommended object may be added to the original object information of each recommended object to obtain standard object information, and the standard object information of all the recommended objects constitutes a standard object information base.

By using the steps S501 to S504, the original object information bases provided by different suppliers can be mapped into the standard object information base, so that all types of the recommended objects in the information recommendation server are uniformly mapped into the same category system, and the subsequent barrel processing of the prices of the recommended objects of the same standard type is facilitated.

In some embodiments, after the standard object information base is constructed, the price classification information of each standard type included in the standard object information base can be determined by the following steps:

step S505, acquiring object information corresponding to each standard type in the standard object information base.

Here, step S505, when implemented, may acquire all object information belonging to each standard type, for example, acquire all object information of a standard type of "car"; all object information of the standard type of "shampoo" is acquired, and so on.

Step S506, the lowest price and the highest price corresponding to each standard type are determined based on the object information corresponding to each standard type.

Here, the price of each recommended object is included in the object information acquired in step S505, so that after all the object information corresponding to each standard type is determined, the highest price and the lowest price of all the recommended objects belonging to the standard type can be determined.

And step S507, determining the price classification information of each standard type based on the lowest price, the highest price and the preset interval number of each standard type.

Here, the number of intervals of different standard types is the same, and is, for example, K, which is an integer greater than 2, and may be 10, 15, 20, or the like. But each price interval may be different. In the embodiment of the present application, the lowest price and the highest price may be divided into K-2 price intervals, and 0 to the lowest price and the highest price to infinity are respectively used as two price intervals, so as to obtain K price intervals, where the K price intervals correspond to price indexes from 1 to K, and a correspondence between the K price intervals and the K price indexes is determined as the standard type of price classification information.

Through the steps S505 to S507, price indexes of the prices of the recommended objects belonging to the same category in the standard object information base can be determined, and the price indexes reflect the relative price levels of the products under the specific categories, so that the price index features are merged into the ranking model during information recommendation, and the ranking model can sense the relative prices of different products and optimize the ranking effect.

Based on the foregoing embodiment, an information recommendation method is further provided in an embodiment of the present application, and is applied to the network architecture shown in fig. 2A, fig. 6 is a schematic diagram of a further implementation flow of the information recommendation method provided in the embodiment of the present application, and as shown in fig. 6, the method includes:

in step S601, the provider terminal acquires object information and recommendation information set for a recommendation object in response to a setting operation.

Here, the supplier terminal may have installed therein an information recommendation App by which the supplier person can set a recommendation target to be recommended, such as clothes, bags, or online courses, and set target information and recommendation information of the recommendation target, such as advertisements, by the information recommendation App.

And step S602, when the supplier terminal receives the information uploading operation, uploading the set object information and the recommendation information to the information recommendation server.

Here, when the supplier person completes setting of the object information and the recommendation information of the recommendation object through the information recommendation App, the set object information and the recommendation information may be uploaded to the information recommendation server by making an upload operation.

In step S603, the information recommendation server extracts the object features of the received object information and extracts the information features of the recommendation information.

After receiving the object information and the recommendation information uploaded by the supplier, the information recommendation server firstly determines whether the original information base of the supplier is stored in the information recommendation server, wherein the original information base and the recommendation information base are included, if the original information base of the supplier is stored in the information recommendation server, the received object information is stored in the original object information base, the standard type of the recommendation object is determined by using a trained object classification model, the standard type and the object information of the recommendation object are stored in the standard object information base, and the recommendation information of the recommendation object is stored in the recommendation information base of the supplier.

When the step S603 is implemented, the information recommendation server extracts object features based on the object information of the recommendation object in the standard object information base, and extracts information features of the recommendation information based on the recommendation information in the recommendation information base.

Step S604, the information recommendation server obtains user characteristics corresponding to each user identifier.

Here, the information recommendation server obtains historical behavior data corresponding to each user identifier, thereby determining a user image feature and a target object identifier based on the historical behavior data, and obtaining a user attribute feature corresponding to each user identifier. The target object identification is the object identification of the recommended object which is clicked or converted by the user.

Step S605, the information recommendation server screens the information to be recommended corresponding to each user identifier from all the recommendation information based on the user characteristics corresponding to each user identifier.

Here, the information to be recommended may be recommendation information recalled corresponding to each user identifier determined based on each user characteristic.

In step S606, the user terminal receives a trigger operation that triggers the information recommendation request.

Here, the trigger operation may be an operation of clicking an icon of a certain App to start the App, or may be a click operation for a certain video in a video playback App.

In step S607, the user terminal sends an information recommendation request to the information recommendation server via the application server in response to the trigger operation.

Here, the information recommendation request carries a user identifier, and the information recommendation request also carries a type of the trigger operation.

Step S608, the information recommendation server obtains the user identifier carried in the information recommendation request.

Step S609, the information recommendation server obtains the user characteristics corresponding to the user identifier, the plurality of information to be recommended, and the information characteristics of the plurality of information to be recommended.

Step S610, the information recommendation server obtains object features of recommendation objects carried in the multiple pieces of information to be recommended.

Here, the object characteristics include at least price characteristics of the recommended object, and the price characteristics may include an absolute price of the recommended object and a price index of the recommended object.

Step S611, the information recommendation server inputs the user characteristics, the information characteristics, and the object characteristics to the trained ranking model to obtain a ranking result of the plurality of pieces of information to be recommended.

Step S612, the information recommendation server obtains the information recommendation duration corresponding to the information recommendation request.

Step S613, the information recommendation server determines at least one piece of target recommendation information from the plurality of pieces of information to be recommended based on the information recommendation duration and the sorting result.

And step S614, the information recommendation server sends the recommendation response carrying the target recommendation information to the user terminal.

Step S615, the user terminal presents the target recommendation information.

Here, when the target recommendation information is a video, step S615 may be implemented when the user terminal plays the target recommendation information.

It should be noted that, the same concepts and steps as those of the embodiments of the present application are explained with reference to the descriptions of the other embodiments.

In the information recommendation method provided by the embodiment of the application, a supplier edits the object information and the recommendation information of a recommendation object needing to be promoted through a supplier terminal and uploads the object information and the recommendation information to an information server, the information server maps the object information provided by different suppliers into a standard object information base so as to level the recommendation objects provided by different suppliers, and calculates price indexes for all recommendation objects under each standard type so as to reflect the relative price of the recommendation objects under the same type; after receiving the information recommendation request, the information recommendation server obtains user characteristics corresponding to a user identifier carried in the information recommendation request and information characteristics of a plurality of pieces of information to be recommended, and obtains object characteristics of recommended objects in the plurality of pieces of information to be recommended, especially obtains price characteristics, so that target recommendation information is determined from the plurality of pieces of information to be recommended by combining the user characteristics, the information characteristics and the object characteristics, and thus the target recommendation information can be ensured to not only accord with the user characteristics, but also accord with consumption habits of users, and the accuracy rate of information recommendation can be improved.

Next, an exemplary application of the embodiment of the present application in a practical application scenario will be described. In the embodiment of the present application, recommendation information is described as an example of an advertisement.

Fig. 7A is a schematic view of a further implementation flow of the information recommendation method provided in the embodiment of the present application, and as shown in fig. 7A, the flow includes:

in step S701, product data provided by an advertiser is acquired.

Here, the advertiser may set the item data through the interface shown in fig. 7B, and may present the item detail interface shown in fig. 7C by clicking or touching the item detail button control 721 in fig. 7B.

Step S702, the characteristics of the goods are mined based on the goods data provided by the advertiser.

Step S703, a commodity library is constructed based on the commodity data provided by the advertiser and the mined commodity features.

In step S704, the advertiser selects a commodity through the delivery terminal.

Step S705 stores the advertisement corresponding to the selected product in the advertisement library.

Steps S701 to S705 shown in fig. 7A are a process of placing a product advertisement, and in the process of placing an advertisement, it is necessary to select an appropriate product from the product library as a part of the product advertisement.

Step S706, storing the commodity features of each commodity extracted from the commodity library into a feature warehouse, and storing the advertisement features of each advertisement extracted from the advertisement library into the feature warehouse;

step S707, obtaining reflow effect data;

step S708, log conversion is carried out on the effect data;

step S709, correlating the converted data with the commodity characteristics and the advertisement characteristics in the characteristic warehouse to obtain a training data stream;

step S710, extracting characteristics based on the training data stream;

step S711, performing model training by using the extracted features to obtain a trained sequencing model;

step S706 to step S711 are a model training part, and the feature attributes related to the advertisement and the commodity are written into a training data stream through a feature repository to perform model training.

Step S712, inputting the user characteristics provided by the user portrait, the advertisement characteristics of the recalled advertisement and the corresponding commodity characteristics into the ranking model to obtain a ranking result.

Through step S712, the online model can be pre-estimated, and through the ranking model trained through the above steps, the recalled advertisements are ranked according to the user characteristics, the commodity characteristics, and the advertisement characteristics provided by the user portrait, so as to obtain a ranking result.

In some embodiments, after obtaining the ranking results, it may be determined that one or more advertisements are to be provided to the user based on the ranking results.

Fig. 8 is a schematic diagram of a process for implementing ranking prediction by using a ranking model according to an embodiment of the present application, where as shown in fig. 8, the ranking model also includes a left tower and a right tower, and the left tower and the right tower include a fully connected network with at least three layers, and the process includes:

step S801, splicing the commodity foreword features, the user features and the foreword features to obtain a first splicing vector, and splicing the advertisement features and the corresponding commodity features to obtain a second splicing vector.

Here, the commodity preamble features may be commodity features, such as a commodity ID, a category ID, a brand ID, and the like, of a screened user having a click and a converted commodity based on historical behavior data of the user, and in some embodiments, the commodity preamble features may further include commodity price and a commodity price index, which are used to represent a preference of the user for the commodity price, so that identification accuracy of the user intention can be improved.

The commodity feature concatenated with the advertisement feature is a commodity feature of a commodity included in the advertisement, for example, an advertisement about a black sesame ball, and then the commodity feature corresponding to the advertisement is the commodity feature of the black sesame ball. The commodity features spliced with the advertisement features certainly include price features, for example, commodity prices and price indexes.

Step S802, inputting the first splicing vector into the full-connection network of the left tower to obtain a 32-dimensional first vector, and inputting the second splicing vector into the full-connection network of the right tower to obtain a 32-dimensional second vector.

In step S803, an inner product of the first vector and the second vector is calculated to obtain an advertisement prediction score.

In the embodiment of the present application, before step S801, it is further required to extract product features, such as a product name, a product ID, a category ID, a brand ID, a price, and the like, where the product name is referred to as a character string type feature, the product ID, the category ID, and the brand ID are all integer type features, and the price is a floating point type feature.

In the embodiment of the application, the commodity name, the commodity ID, the category ID and the brand ID are all processed by the original form access characteristic embedding layer. Due to the particularity of the price characteristic, the direct processing of the price characteristic by an ID or a character string is easy to cause loss, so that a barrel processing method is adopted. The upper and lower bounds of the barrel are connected end to end, and the assumption is that each sub-barrel is sequentially: [1,10,20,30,40,50,60,70,80,90,100,120,140,160,180,200,250,300,350,400,500,600,700,800,900,1000,2000,3000,4000,5000,6000,7000,8000,9000]

Correspondingly, the barrel numbers are 1, 2, 3, 4 … … in sequence, if the price of a certain commodity is 25 yuan, it falls on the 3 rd barrel, and its extracted price characteristic is 3.

In the embodiment of the application, after the commodity characteristics (including price characteristics) are added, the AUC of the whole e-commerce advertisement is improved by 1.16%, wherein the AUC of the commodity advertisement with the price is improved by 2.08%, and the commodity characteristics greatly improve the prediction accuracy of the commodity advertisement in the model.

However, if the absolute price is input into the ranking model, the ranking model only learns the absolute price, but does not accurately know whether the commodity is cheap or not. For example, a cheap car has a price of 10 ten thousand yuan, and a famous watch has a price of 9 ten thousand yuan. The absolute price of a car is higher than that of a watch, but obviously the relative price (grade) of the watch is higher than that of the car.

In order to make the ranking model have better prediction on the relative price of the commodity, the method for calculating the price index of the commodity is provided in the embodiment of the application.

When the method is implemented, all commodities in all commodity libraries (including large platforms and small advertiser main merchants) are mapped to the same category system through the commodity knowledge base. Thus, all the commodities are classified under respective categories.

As shown in fig. 9, all commodities in the commodity library a 901, the commodity library B902, and the commodity library C903 are mapped to the same category system, and the category system includes a uniform category 1911, a uniform category 2912, uniform categories 3913, …, and a uniform category N91N.

And then, carrying out bucket separation on all commodity prices under each category, wherein each uniform category has respective price bucket separation. For example, the price of each category of goods may be divided into 10 levels (price indices).

Table 1 shows the price level division table for the car and the watch, and as shown in table 1, the price index of the 10 ten thousand yuan car is 3, the price index of the 9 ten thousand yuan nameplate table is 10, and the price grade difference between the two products is obvious.

In the embodiment of the application, the price index features are also added into the ranking model for training.

Fig. 10 is a schematic diagram of building a commodity knowledge base and unified categories provided in the embodiment of the present application, and as shown in fig. 10, a large-platform commodity base 1001 with a complete commodity system and standard industry categories 1002 formulated by itself are first used as training samples to perform training, so as to obtain a commodity knowledge base model 1003.

Then, all commodity libraries (including a small advertiser commodity library and a large platform E-commerce commodity library) are predicted through a commodity knowledge base model, and a unified category system 1004 is obtained.

For example, the "bag" of goods under advertiser A and the "nameplate bag" of goods under advertiser B are both categorized into the category of "clothing bag".

The technical effects of the information recommendation method provided by the embodiment of the present application are graphically described below by way of examples.

First, the price index is graphically demonstrated to reflect the relative price of the goods, and in the embodiment of the application, two groups of goods, namely teapots and hair dye cream, are found from real goods advertisements. Table 2 is a comparison table of commercial characteristics of teapots and hair dye cream.

As can be seen from table 2, the overall price of the hair dye creams (C, D) is significantly lower than that of the teapots (a, B), and the price of the 138 hair dye cream D is not very expensive, but is already very expensive in the high school; the pot a, which is 199 in price, is absolutely expensive, but is inexpensive.

Therefore, the model with the price index features merged into it is compared with the original model: the noble hair dye cream D has higher similarity with the noble teapot B; the precious hair dye cream D has a lower similarity to the cheap teapot a.

The following is visually demonstrated in conjunction with fig. 11A and 11B:

model prediction embedding feature vectors are used for A, B, D three groups of advertisements and 500 randomly extracted noise advertisements, and the models are marked in a prediction result graph through a TSNE dimension reduction visualization method. The closer the distance between a point in the figure and a point is, the higher the similarity between the two points is.

Fig. 11A is a schematic diagram of the prediction result of the ranking model with only the price features of the product added, and it can be seen from fig. 11A that the product D1103 and the product a 1101 belong to the same product with low absolute price, the distance is short, and the similarity is high; the product D1103 and the product B1102 belong to high-grade products, but the distance is long and the similarity is low.

Fig. 11B is a schematic diagram of the prediction result of the ranking model with added price index features, and as can be seen from fig. 11B, the product D1103 and the product a 1101 belong to the absolute low-price product, the distance between the price grades is far, the similarity is low, and the distance between the product D1013 and the product B1102 belong to the high-grade product, and the similarity is high.

Fig. 12 is a schematic diagram of a variation curve of an advertisement start rate when information recommendation is performed by using the information recommendation method provided in the embodiment of the present application, and as shown by a start rate curve 1201 in fig. 12, after information recommendation is performed by using the information recommendation method provided in the embodiment of the present application from 12 months and 24 days in 2019, the start rate of a product advertisement is gradually increased from 10% to 24%.

It should be noted that the merchandise features on the merchandise side and the merchandise pre-text features on the user side may add up to a consumption promotion of 1.1% for the system large disk, wherein there is a consumption promotion of 10.1% for the merchandise advertisement.

In the embodiment of the application, commodity characteristics, particularly price characteristics are integrated into an advertisement ranking model to enhance the discrimination and accuracy of the ranking model in predicting advertisements with different prices. Based on the price of the commodity given by the advertiser, the commodity knowledge base technology is utilized to level and process the commodities of each advertiser, the commodities are mapped to a unified category system, and price indexes are calculated for all the commodities under each category so as to reflect the relative price of the commodities under the specific category. The price index features are integrated into the sequencing model, so that the model can sense the relative prices of different commodities and optimize the sequencing effect.

Continuing with the exemplary structure of the information recommendation device 354 implemented as a software module provided in the embodiments of the present application, in some embodiments, as shown in fig. 3, the software module stored in the information recommendation device 354 of the memory 350 may be an information recommendation device in the server 300, including:

a first obtaining module 3541, configured to obtain, in response to a received information recommendation request, a user identifier carried in the information recommendation request;

a second obtaining module 3542, configured to obtain a user characteristic corresponding to the user identifier, multiple pieces of information to be recommended, and an information characteristic of the multiple pieces of information to be recommended;

a third obtaining module 3543, configured to obtain object features of recommended objects carried in the multiple pieces of information to be recommended, where the object features at least include price features of the recommended objects;

an input module 3544, configured to input the user characteristics, the information characteristics, and the object characteristics to a trained ranking model, so as to obtain a ranking result of the multiple pieces of information to be recommended;

a first determining module 3545, configured to determine at least one piece of target recommendation information from the plurality of pieces of information to be recommended based on the sorting result;

a sending module 3546, configured to send the recommendation response carrying the target recommendation information to the user terminal.

In some embodiments, the apparatus further comprises:

In some embodiments, the third obtaining module is further configured to:

In some embodiments, the apparatus further comprises:

In some embodiments, the first determining module is further configured to:

It should be noted that the description of the apparatus in the embodiment of the present application is similar to the description of the method embodiment, and has similar beneficial effects to the method embodiment, and therefore, the description is not repeated. For technical details not disclosed in the embodiments of the apparatus, reference is made to the description of the embodiments of the method of the present application for understanding.

Embodiments of the present application provide a storage medium having stored therein executable instructions, which when executed by a processor, will cause the processor to perform the methods provided by embodiments of the present application, for example, the methods as illustrated in fig. 4, 5 and 6.

In some embodiments, the storage medium may be a computer-readable storage medium, such as a Ferroelectric Random Access Memory (FRAM), a Read Only Memory (ROM), a Programmable Read Only Memory (PROM), an Erasable Programmable Read Only Memory (EPROM), a charged Erasable Programmable Read Only Memory (EEPROM), a flash Memory, a magnetic surface Memory, an optical disc, or a Compact disc Read Only Memory (CD-ROM), among other memories; or may be various devices including one or any combination of the above memories.

In some embodiments, executable instructions may be written in any form of programming language (including compiled or interpreted languages), in the form of programs, software modules, scripts or code, and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

By way of example, executable instructions may correspond, but do not necessarily have to correspond, to files in a file system, and may be stored in a portion of a file that holds other programs or data, such as in one or more scripts in a hypertext markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, sub-programs, or portions of code). By way of example, executable instructions may be deployed to be executed on one computing device or on multiple computing devices at one site or distributed across multiple sites and interconnected by a communication network.

The above description is only an example of the present application, and is not intended to limit the scope of the present application. Any modification, equivalent replacement, and improvement made within the spirit and scope of the present application are included in the protection scope of the present application.

Claims

1. An information recommendation method, comprising:

determining at least one piece of target recommendation information from the plurality of pieces of information to be recommended based on the sorting result;

and sending the recommendation response carrying the target recommendation information to a user terminal.

2. The method of claim 1, further comprising:

acquiring historical behavior data, user attribute characteristics and user context characteristics corresponding to the user identification;

determining user portrait characteristics and target object identification corresponding to the user identification based on the historical behavior data, wherein the target object identification is object identification of a recommended object clicked by the user or converted by the user;

acquiring historical object characteristics corresponding to the target object identification;

and determining a user characteristic corresponding to the user identification based on the user attribute characteristic, the user context characteristic, the user portrait characteristic and the historical object characteristic.

3. The method according to claim 1, wherein obtaining the object features of the recommendation objects carried in the plurality of information to be recommended comprises:

4. The method of claim 3, wherein the extracting the object feature of the recommended object from the object information comprises:

5. The method of claim 4, further comprising:

acquiring a sample object library and a preset candidate type set, wherein the sample object library comprises a plurality of sample objects and sample types corresponding to the sample objects;

determining a union set of a sample type set and a candidate type set included in the sample object library as a standard type set;

acquiring target sample information of each sample object in the sample object library, wherein the target sample information is other sample information except for sample types;

and training an object classification model by using the target sample information and the standard type set to obtain the trained object classification model.

6. The method of claim 5, further comprising:

acquiring an original object information base provided by a supplier, wherein the original object information base comprises original object information of a plurality of recommended objects;

determining object information obtained by removing the object type of each recommended object from each original object information as each predicted object information;

inputting the information of each predicted object into a trained object classification model to obtain the standard type of each recommended object;

and constructing a standard object information base based on the standard type and the original object information of each recommended object.

7. The method as recited in claim 6, wherein said method further comprises:

acquiring object information corresponding to each standard type in a standard object information base;

determining the lowest price and the highest price corresponding to each standard type based on the object information corresponding to each standard type;

and determining the price classification information of each standard type based on the lowest price, the highest price and the preset interval number of each standard type.

8. The method of claim 1, wherein the determining at least one target recommendation information from the plurality of information to be recommended based on the sorting result comprises:

9. An information recommendation apparatus, comprising:

10. An information recommendation apparatus characterized by comprising:

a memory for storing executable instructions; a processor for implementing the method of any one of claims 1 to 8 when executing executable instructions stored in the memory.

11. A computer-readable storage medium having stored thereon executable instructions for causing a processor, when executed, to implement the method of any one of claims 1 to 8.