CN112650778B - Information recommendation system, method, device, server and storage medium - Google Patents

Information recommendation system, method, device, server and storage medium Download PDF

Info

Publication number
CN112650778B
CN112650778B CN202011573356.6A CN202011573356A CN112650778B CN 112650778 B CN112650778 B CN 112650778B CN 202011573356 A CN202011573356 A CN 202011573356A CN 112650778 B CN112650778 B CN 112650778B
Authority
CN
China
Prior art keywords
information
account
features
server
feature
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202011573356.6A
Other languages
Chinese (zh)
Other versions
CN112650778A (en
Inventor
黄亮
曹文慧
马宁宁
王嘉晨
李英民
钟辉
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Dajia Internet Information Technology Co Ltd
Original Assignee
Beijing Dajia Internet Information Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Dajia Internet Information Technology Co Ltd filed Critical Beijing Dajia Internet Information Technology Co Ltd
Priority to CN202011573356.6A priority Critical patent/CN112650778B/en
Publication of CN112650778A publication Critical patent/CN112650778A/en
Application granted granted Critical
Publication of CN112650778B publication Critical patent/CN112650778B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2455—Query execution
    • G06F16/24552—Database cache management
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The present disclosure relates to an information recommendation system, method, apparatus, server, and storage medium. The information recommendation system includes: the system comprises a prediction server, a storage server and a calculation cache server; the storage server is used for storing account data; the computing cache server is used for extracting the characteristics of the information stored in the information database, obtaining the information characteristics corresponding to the information and caching the information characteristics; the prediction server is used for responding to the information recommendation request, acquiring corresponding account data from the storage server, acquiring corresponding information features from the information features cached by the calculation cache server, acquiring account features corresponding to the account data, determining the correlation between the account features and the information features, and acquiring recommendation information of corresponding accounts based on the correlation. According to the information recommendation system provided by the disclosure, the prediction server can reduce the calculation process of information feature extraction, so that the calculation loss is reduced.

Description

Information recommendation system, method, device, server and storage medium
Technical Field
The disclosure relates to the technical field of information recommendation, and in particular relates to an information recommendation system, an information recommendation method, an information recommendation device, a server and a storage medium.
Background
With the development of information recommendation technology, a recommendation system for information recommendation constructed by a deep learning method appears, the system can extract account data and information to be recommended corresponding to a request after receiving the recommendation request, and deep representation of the account and the information to be recommended is obtained by using a deep prediction model, so that an information recommendation list is formed.
In the related art, the current recommendation system includes a prediction server, wherein the prediction server performs a process from receiving an external request to performing calculation using a depth prediction model, and for a large-scale recommendation system, the prediction server performs a large amount of calculation tasks, so that the calculation loss of the prediction server in the current recommendation information determining system is too large.
Disclosure of Invention
The disclosure provides an information recommendation system, an information recommendation method, an information recommendation device, a server and a storage medium, so as to at least solve the problem that the calculation loss of a prediction server cluster in the related art is overlarge. The technical scheme of the present disclosure is as follows:
according to a first aspect of an embodiment of the present disclosure, there is provided an information recommendation system, including: the system comprises a prediction server, a storage server and a calculation cache server; wherein,
The storage server is used for storing account data;
the computing cache server is used for extracting characteristics of information stored in the information database, obtaining information characteristics corresponding to the information and caching the information characteristics;
the prediction server is used for responding to an information recommendation request, acquiring corresponding account data from the storage server, acquiring corresponding information features from the information features cached by the calculation cache server, acquiring account features corresponding to the account data, determining correlation between the account features and the information features, and acquiring recommendation information of corresponding accounts based on the correlation.
In an alternative embodiment, the recommendation information determining system further comprises: a training server; the prediction server is provided with a depth prediction model for determining recommendation information; the training server is used for training the model weight parameters of the depth prediction model and storing the model weight parameters obtained by training to the storage server; the calculation cache server is further used for acquiring and caching the model weight parameters from the storage server, acquiring information feature weight parameters for extracting information features from the cached model weight parameters, and performing feature extraction on information stored in an information database according to the information feature weight parameters to obtain information features corresponding to the information; the prediction server is further configured to obtain cached model weight parameters from the calculation cache server, update a local depth prediction model based on the model weight parameters, obtain account features corresponding to the account data through the updated depth prediction model, determine correlation between the account features and the information features, and obtain recommendation information of corresponding accounts based on the correlation.
In an optional embodiment, the calculation cache server is further configured to obtain an information feature weight parameter in the currently cached model weight parameter when the cached model weight parameter is updated and/or information stored in the information database is updated, and perform feature extraction on information currently stored in the information database based on the information feature weight parameter to obtain an information feature corresponding to the information.
In an optional embodiment, the prediction server is further configured to obtain a data vector corresponding to the account data, input the data vector into a first hidden layer of the updated depth prediction model, input the information feature into a second hidden layer of the updated depth prediction model, and obtain recommendation information of the corresponding account according to output of the updated depth prediction model; the first hiding layer is used for extracting features of the data vector to obtain corresponding account features, and the corresponding account features are input to the second hiding layer; the second hidden layer and other hidden layers are used for determining the correlation between the account characteristics and the information characteristics and determining the recommendation information of the corresponding account based on the correlation.
According to a second aspect of the embodiments of the present disclosure, there is provided an information recommendation method, applied to a prediction server, including:
responding to an information recommendation request, respectively acquiring account data corresponding to the information recommendation request from a storage server, and acquiring information features corresponding to the information recommendation request from information features cached by a calculation cache server; the calculation cache server performs feature extraction on information stored in an information database in advance to obtain information features corresponding to the information and caches the information features;
acquiring account characteristics of the account data;
and determining the correlation between the account characteristics and the information characteristics, and obtaining recommendation information of the corresponding account according to the correlation.
In an alternative embodiment, the prediction server is configured with a depth prediction model for determining recommendation information; the determining the correlation between the account feature and the information feature, and obtaining the recommendation information of the corresponding account according to the correlation includes: obtaining cached model weight parameters from the calculation cache server; updating a local depth prediction model based on the model weight parameters, and acquiring account features corresponding to the account data through the updated depth prediction model; and determining the correlation between the account characteristics and the information characteristics, and obtaining the recommendation information of the corresponding account according to the correlation.
In an optional embodiment, the account feature includes a data vector corresponding to the account data; the determining the correlation between the account feature and the information feature, and obtaining the recommendation information of the corresponding account according to the correlation includes: acquiring the data vector, and inputting the data vector into a first hidden layer of the updated depth prediction model; the first hidden layer is used for extracting features of the data vector to obtain corresponding account features; inputting the account feature and the information feature into a second hidden layer of the updated depth prediction model; the second hidden layer and other hidden layers after the second hidden layer are used for determining correlation between the account characteristics and the information characteristics and determining recommendation information of the corresponding account based on the correlation.
According to a third aspect of the embodiments of the present disclosure, there is provided an information recommendation method applied to a computing cache server, including:
extracting features of information stored in an information database to obtain information features corresponding to the information and caching the information features;
if a request for acquiring the information features sent by the prediction server is received, acquiring the information features corresponding to the request for acquiring the information features from the cached information features, and returning the information features to the prediction server, so that the prediction server acquires the account features corresponding to the account data acquired from the storage server, determines the correlation between the account features and the information features, and obtains recommendation information of corresponding accounts based on the correlation.
In an optional embodiment, the feature extraction of the information stored in the information database, to obtain information features corresponding to the information, and caching the information features, includes: the model weight parameters are obtained from the storage server and cached, and the information feature weight parameters for extracting the information features are obtained from the cached model weight parameters; and carrying out feature extraction on the information stored in the information database according to the information feature weight parameters to obtain information features corresponding to the information.
In an optional embodiment, the information recommendation method further includes: when the cached model weight parameters are updated and/or the information stored in the information database is updated, acquiring information characteristic weight parameters in the currently cached model weight parameters; and carrying out feature extraction on the information currently stored in the information database based on the information feature weight parameters to obtain information features corresponding to the information.
According to a fourth aspect of the embodiments of the present disclosure, there is provided an information recommendation apparatus applied to a prediction server, including:
a recommendation request response unit configured to execute, in response to an information recommendation request, obtaining account data corresponding to the information recommendation request from a storage server and obtaining information features corresponding to the information recommendation request from information features cached by a calculation cache server, respectively; the calculation cache server performs feature extraction on information stored in an information database in advance to obtain information features corresponding to the information and caches the information features;
An account feature acquisition unit configured to execute account features;
and the recommendation information acquisition unit is configured to determine the correlation between the account characteristics and the information characteristics and obtain recommendation information of the corresponding account according to the correlation.
In an alternative embodiment, the prediction server is configured with a depth prediction model for determining recommendation information; the recommendation information acquisition unit is further configured to perform acquisition of cached model weight parameters from the calculation cache server; updating a local depth prediction model based on the model weight parameters, and acquiring account features corresponding to the account data through the updated depth prediction model; and determining the correlation between the account characteristics and the information characteristics, and obtaining the recommendation information of the corresponding account according to the correlation.
In an optional embodiment, the account feature includes a data vector corresponding to the account data; the recommendation information acquisition unit is further configured to perform acquisition of the data vector, and input the data vector into a first hidden layer of the updated depth prediction model; the first hidden layer is used for extracting features of the data vector to obtain corresponding account features; inputting the account feature and the information feature into a second hidden layer of the updated depth prediction model; the second hidden layer and other hidden layers after the second hidden layer are used for determining correlation between the account characteristics and the information characteristics and determining recommendation information of the corresponding account based on the correlation.
According to a fifth aspect of the embodiments of the present disclosure, there is provided an information recommendation apparatus applied to a computing cache server, including:
the information characteristic acquisition unit is configured to perform characteristic extraction on information stored in the information database, obtain information characteristics corresponding to the information and cache the information characteristics;
and the information feature sending unit is configured to execute the steps of obtaining the information feature corresponding to the request for obtaining the information feature from the cached information feature and returning the information feature to the prediction server if the request for obtaining the information feature sent by the prediction server is received, so that the prediction server obtains the account feature corresponding to the account data obtained from the storage server, determines the correlation between the account feature and the information feature, and obtains the recommendation information of the corresponding account based on the correlation.
In an optional embodiment, the information feature obtaining unit is further configured to obtain the model weight parameter from the storage server, cache the model weight parameter, and obtain an information feature weight parameter for extracting an information feature from the cached model weight parameter; and carrying out feature extraction on the information stored in the information database according to the information feature weight parameters to obtain information features corresponding to the information.
In an alternative embodiment, the information recommendation device further includes: an information feature updating unit configured to perform updating of the cached model weight parameters, and/or obtain information feature weight parameters in the currently cached model weight parameters when updating of information stored in the information database occurs; and carrying out feature extraction on the information currently stored in the information database based on the information feature weight parameters to obtain information features corresponding to the information.
According to a sixth aspect of embodiments of the present disclosure, there is provided a server comprising: a processor; a memory for storing the processor-executable instructions; wherein the processor is configured to execute the instructions to implement the information recommendation method described in any one of the embodiments of the second or third aspects above.
According to a seventh aspect of embodiments of the present disclosure, there is provided a computer readable storage medium, which when executed by a processor of a server, enables the server to perform the information recommendation method described in any one of the embodiments of the second or third aspects described above.
According to an eighth aspect of embodiments of the present disclosure, there is provided a computer program product comprising a computer program which, when executed by a processor, implements the information recommendation method described in any one of the embodiments of the second or third aspects.
The technical scheme provided by the embodiment of the disclosure at least brings the following beneficial effects:
the information recommendation system provided by the present disclosure includes: the system comprises a prediction server, a storage server and a calculation cache server; the storage server is used for storing account data; the computing cache server is used for extracting the characteristics of the information stored in the information database, obtaining the information characteristics corresponding to the information and caching the information characteristics; the prediction server is used for responding to the information recommendation request, acquiring corresponding account data from the storage server, acquiring corresponding information features from the information features cached by the calculation cache server, acquiring account features corresponding to the account data, determining the correlation between the account features and the information features, and acquiring recommendation information of corresponding accounts based on the correlation. In the information recommendation system provided by the disclosure, the prediction server can directly obtain the cached information characteristics from the calculation cache server, and the calculation process of information characteristic extraction can be reduced, so that the calculation loss of the prediction server is reduced.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the disclosure and together with the description, serve to explain the principles of the disclosure and do not constitute an undue limitation on the disclosure.
Fig. 1 is an application environment diagram illustrating an information recommendation method according to an exemplary embodiment.
Fig. 2 is a schematic diagram showing a structure of an information recommendation system according to an exemplary embodiment.
Fig. 3 is a schematic diagram showing a structure of an information recommendation system according to another exemplary embodiment.
Fig. 4 is a flowchart illustrating an information recommendation method according to an exemplary embodiment.
FIG. 5 is a flowchart illustrating deriving recommendation information for a corresponding account based on relevance, according to an example embodiment.
FIG. 6 is a flowchart illustrating deriving recommendation information for a corresponding account based on relevance according to another exemplary embodiment.
Fig. 7 is a flowchart illustrating an information recommendation method according to another exemplary embodiment.
Fig. 8 is an interactive flow chart illustrating a method of information recommendation according to an exemplary embodiment.
Fig. 9 is a schematic diagram illustrating a structure of a depth prediction model according to an exemplary embodiment.
FIG. 10 is a hidden layer first layer computational decomposition diagram of a depth prediction model, shown according to an exemplary embodiment.
Fig. 11 is a system architecture diagram of an information recommendation system, which is shown according to an exemplary embodiment.
Fig. 12 is a block diagram illustrating an information recommendation apparatus according to an exemplary embodiment.
Fig. 13 is a block diagram illustrating an information recommendation apparatus according to another exemplary embodiment.
Fig. 14 is an internal structural diagram of a server shown according to an exemplary embodiment.
Detailed Description
In order to enable those skilled in the art to better understand the technical solutions of the present disclosure, the technical solutions of the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.
It should be noted that the terms "first," "second," and the like in the description and claims of the present disclosure and in the foregoing figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the disclosure described herein may be capable of operation in sequences other than those illustrated or described herein. The implementations described in the following exemplary examples are not representative of all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with some aspects of the present disclosure as detailed in the accompanying claims.
The information recommendation system provided by the disclosure can be applied to an application environment as shown in fig. 1. Wherein the prediction server 101, the storage server 102, and the calculation cache server 103 interact through a network. Specifically, the storage server 102 stores account data in advance, the calculation cache server 103 stores information features obtained by extracting features of information stored in the information database in advance, after the prediction server 101 receives an information recommendation request, the corresponding account data can be obtained from the storage server 102, the corresponding information features can be obtained from the calculation cache server 103, feature extraction is performed on the received account data, after the account features are obtained, correlation between the account features and the information features is determined, and recommendation information of the account is obtained by using the correlation. The prediction server 101, the storage server 102, and the calculation cache server 103 may be implemented as a stand-alone server or a server cluster formed by a plurality of servers.
Fig. 2 is a schematic structural view of an information recommendation system according to an exemplary embodiment, and as shown in fig. 2, the system includes: a prediction server 201, a storage server 202, and a calculation cache server 203.
The storage server 202 may be a server for storing account data, and the storage server 202 carries an account database, where multiple account data of different accounts are stored in the database, and may include, for example, the age, sex, and the like of the certain account. The calculation cache server 203 is a server for caching information features, and since the information stored in the information database is relatively stable with respect to the account data, the calculation cache server 103 may perform feature extraction on the information in advance to obtain information features corresponding to the information, and cache the obtained information features.
The prediction server 201 is a server for performing recommendation information acquisition, and the server can determine the correlation between account features and information features to obtain recommendation information of corresponding accounts. Specifically, when the prediction server 201 receives the information recommendation request, it may respond to the obtained information recommendation request, obtain account data corresponding to the information recommendation request from the storage server 202, and obtain information features corresponding to the request from the information features cached by the calculation cache server 203, and then, the prediction server 201 may perform feature extraction on the account data obtained from the storage server 202 to obtain corresponding account features, and determine recommendation information of the account based on correlation between the obtained account features and the information features. Since the extraction of the information features is already completed and stored in the calculation cache server 203, the prediction server 201 can reduce the calculation amount of feature extraction of the information data, thereby reducing the calculation loss.
For example, when the prediction server 201 receives an information recommendation request for recommending certain pictures to the account a, account data of the account a may be acquired first from the storage server 202, and relevant picture features of pictures to be recommended may be acquired from the calculation cache server 203. The prediction server 201 may then perform feature extraction on the obtained account data of the account a to obtain account features of the account a, and obtain a final picture to be recommended of the account a by using correlation between the account features of the account a and the relevant picture features of the picture to be recommended.
The information recommendation system includes: a prediction server 201, a storage server 202, and a calculation cache server 203; wherein, the storage server 202 is used for storing account data; the calculation cache server 203 is configured to perform feature extraction on information stored in the information database, obtain information features corresponding to the information, and cache the information features; the prediction server 201 is configured to obtain corresponding account data from the storage server 202 in response to the information recommendation request, obtain corresponding information features from the information features cached by the calculation cache server 203, obtain account features corresponding to the account data, determine correlation between the account features and the information features, and obtain recommendation information of the corresponding account based on the correlation. In the recommendation information determining system provided by the disclosure, the prediction server 201 can directly obtain the cached information features from the calculation cache server 203, and the calculation process of information feature extraction can be reduced, so that the calculation loss of the prediction server 201 is reduced.
In an exemplary embodiment, as shown in fig. 3, the recommendation information determining system, in addition to comprising: the prediction server 301, the storage server 302, the calculation cache server 303, and the training server 304 are further included.
The prediction server 301 is configured with a depth prediction model, and the model is mainly used for extracting features of account data to obtain account features, determining correlation between the account features and information features, and obtaining recommendation information of a corresponding account based on the correlation. The training server 304 is configured to train and update model weight parameters of the depth prediction model configured in the prediction server 301, where the model weight parameters may include account feature weight parameters for reflecting weight parameters corresponding to account features, and information feature weight parameters for reflecting weight parameters corresponding to information features, and the model weight parameters may be a part of model parameters stored in the depth prediction model. The training server 304 may obtain model weight parameters corresponding to the depth prediction model by training the depth prediction model, and store the obtained model weight parameters in the storage server 302.
Then, the calculation cache server 303 may read the model weight parameters obtained by training the training server 304 from the storage server 302 and cache the model weight parameters, and at the same time, the calculation cache server 303 may also read the information feature weight parameters used for feature extraction of the information stored in the information database from the cached model weight parameters, and perform feature extraction of the stored information by using the information feature weight parameters, thereby obtaining corresponding information features and caching the information features.
When the prediction server 301 receives the information recommendation request, the cached model weight parameters may be read from the calculation cache server 303, so as to update the model weight parameters stored in the depth prediction model, obtain the account feature by using the updated depth prediction model, and obtain the recommendation information of the corresponding account by using the correlation between the obtained account feature and the information feature obtained from the calculation cache server 303.
Further, in order to ensure the real-time performance and accuracy of the information features stored in the calculation cache server 303, the calculation cache server 303 may also update the cached information features in real time, specifically, when the calculation cache server 303 detects that the model weight parameters are updated, or when the information stored in the information database is updated, the calculation cache server 303 may update the cached information features accordingly. Specifically, the calculation cache server 303 may read the information feature weight parameter used to reflect the weight parameter corresponding to the information feature in the currently cached model weight parameter, and perform feature extraction on the information currently stored in the information database again by using the information feature weight parameter, so as to implement real-time update of the information feature.
In addition, in order to further reduce the computation loss of the prediction server 301, in an exemplary embodiment, the prediction server 301 may further input a data vector corresponding to the account data into a first hidden layer of the updated depth prediction model, where the first hidden layer performs further feature extraction on the data vector of the input account data, where the data vector may be an account feature obtained after the account data performs the preliminary feature extraction. Specifically, the prediction server 301 may perform preliminary feature extraction on account data to obtain a data vector corresponding to the account data, input the data vector to a first hidden layer of the depth prediction model, and perform feature extraction on the data vector by the first hidden layer to obtain an account feature corresponding to the data vector.
Then, the prediction server 301 may input the account feature corresponding to the data vector and the information feature obtained in the calculation cache server 303 to the second hidden layer of the updated depth prediction model, calculate the correlation between the account feature and the information feature by the second hidden layer and other hidden layers after the second hidden layer, and determine the recommendation information of the corresponding account based on the correlation.
Compared with the prior art, the prediction server needs to perform preliminary feature extraction on the account data and the information in the information database to obtain a data vector corresponding to the account data and a data vector corresponding to the information, and input the data vector corresponding to the account data and the data vector corresponding to the information into the first hidden layer at the same time to calculate the account feature of the data vector corresponding to the account data and the information feature corresponding to the information vector, where the first hidden layer in the prediction server 301 provided in this embodiment is only used to calculate the account feature of the data vector, and the information feature corresponding to the information vector which originally needs to be obtained by the first hidden layer can be calculated in advance by the calculation cache server 303, so that the calculation amount of the depth prediction model in the prediction server 301 can be further reduced, and the calculation loss of the prediction server 301 is further reduced.
In the above embodiment, the training server 304 may train the depth prediction model, so as to update the depth prediction model of the prediction server 301, so as to improve the accuracy of the recommended information obtained by the prediction server 301, and in addition, the calculation cache server 303 may implement real-time update of the cached information feature based on the model weight parameters after training and updating in the training server 304, so as to improve the real-time performance and accuracy of the information feature stored in the calculation cache server 303. In addition, the information feature previously cached by the calculation cache server 303 may be further set to the information feature that needs to be output by the first hidden layer of the prediction server 301, and the calculation amount of the depth prediction model in the prediction server 301 may be further reduced, thereby further reducing the calculation loss of the prediction server 301.
Fig. 4 is a flowchart illustrating an information recommendation method according to an exemplary embodiment, which is used in the prediction server 301 of fig. 3, as shown in fig. 4, and includes the following steps.
In step S401, the prediction server 301 obtains account data corresponding to the information recommendation request from the storage server 302 and obtains information features corresponding to the information recommendation request from the information features cached by the calculation cache server 303, respectively, in response to the information recommendation request; the calculation cache server 303 performs feature extraction on the information stored in the information database in advance, obtains information features corresponding to the information, and caches the information features.
The storage server 302 stores a plurality of account data in advance, and the calculation cache server 303 may perform feature extraction on a plurality of information stored in the information database in advance, so as to obtain a plurality of information features and cache the information features. When the prediction server 301 receives an information recommendation request, account data and information features corresponding to the request may be obtained from the storage server 302 and the calculation cache server 303, respectively, in response to the recommendation request.
For example, the information recommendation request may include an account identifier of an account and an information identifier of information to be acquired, where the identifier may be an account number of the account and an information number of the information, and the information recommendation request is sent to the storage server 302 and the calculation cache server 303, and the storage server 302 and the calculation cache server 303 may find, according to the account identifier and the information identifier, stored corresponding account data and information features, respectively, as account data and information features corresponding to the information recommendation request, and return the account data and the information features to the prediction server 301.
In step S402, the prediction server 301 acquires account characteristics of account data;
in step S403, the prediction server 301 determines the correlation between the account feature and the information feature, and obtains recommendation information of the corresponding account according to the correlation.
Then, the prediction server 301 may perform feature extraction on the obtained account data to obtain corresponding account features, and determine recommendation information of the account based on correlation between the account features obtained by feature extraction and information features returned by the calculation cache server 303. Since the extraction of the information features is already completed and stored in the calculation cache server 203, the prediction server 301 can reduce the calculation amount of feature extraction of the information data, thereby reducing the calculation loss.
In the above information recommendation method, the prediction server 301 responds to the information recommendation request, and obtains account data corresponding to the information recommendation request from the storage server 302 and obtains information features corresponding to the information recommendation request from the information features cached by the calculation cache server 303; the calculation cache server 303 performs feature extraction on the information stored in the information database in advance to obtain information features corresponding to the information and caches the information features; acquiring account characteristics of account data; and determining the correlation between the account characteristics and the information characteristics, and obtaining the recommendation information of the corresponding account according to the correlation. In the information recommendation method provided by the disclosure, the prediction server 301 can directly obtain the cached information features from the calculation cache server 303, so that the calculation process of information feature extraction can be reduced, and the calculation loss of the prediction server 301 is reduced.
In an exemplary embodiment, the prediction server 301 is configured with a depth prediction model for determining recommendation information, and as shown in fig. 5, step S403 may further include the steps of:
in step S501, the prediction server 301 acquires cached model weight parameters from the calculation cache server 303.
The depth prediction model is mainly a prediction model that is used by the prediction server 301 to determine the correlation between account features and information features to obtain recommendation information of a corresponding account, and model weight parameters of a model of the prediction model may be cached in the calculation cache server 303 in advance. When the prediction server 301 needs to perform step S403, the model weight parameters currently stored by the calculation cache server 303 may be read from the calculation cache server 303 to update the depth prediction model.
In step S502, the prediction server 301 updates a local depth prediction model based on the model weight parameter, and obtains account features corresponding to the account data through the updated depth prediction model;
in step S503, the prediction server 301 determines the correlation between the account feature and the information feature, and obtains recommendation information of the corresponding account according to the correlation.
Then, the prediction server 301 may update the locally stored depth prediction model according to the model weight parameter obtained from the calculation cache server 303, and obtain account features corresponding to the account data by using the updated depth prediction model, and calculate correlation between the account features and the information features, so as to obtain recommendation information of the corresponding account.
Further, the account features comprise data vectors corresponding to the account data; as shown in fig. 6, step S503 may further include:
in step S601, the prediction server 301 acquires a data vector, and inputs the data vector into the first hidden layer of the updated depth prediction model; the first hidden layer is used for extracting features of the data vector to obtain corresponding account features.
The data vector refers to a data vector corresponding to the account data, and the data vector may be a part of the account feature, and is obtained by the prediction server 301 through step S402. After the prediction server 301 obtains the account data from the storage server 302, a data vector corresponding to the account data may be obtained in the manner of step S402. Then, the prediction server 301 may input the obtained data vector into the first hidden layer of the updated depth prediction model, where the hidden layer is mainly used to perform further feature extraction on the data vector corresponding to the account data, so as to obtain the account feature corresponding to the data vector.
In step S602, the prediction server 301 inputs the account feature and the information feature into the second hidden layer of the updated depth prediction model; the second hidden layer and other hidden layers are used for determining the correlation between the account characteristics and the information characteristics and determining the recommendation information of the corresponding account based on the correlation.
After the prediction server 301 obtains the account feature corresponding to the data vector through the first hidden layer, the account feature and the information feature obtained from the calculation cache server 303 may be input to the second hidden layer of the depth prediction model, and the correlation between the account feature and the information feature is calculated through the second hidden layer and other hidden layers behind the second hidden layer, and recommendation information of the corresponding account is determined based on the correlation.
In the above embodiment, the prediction server 301 may receive the model weight parameters from the calculation cache server 303 to implement real-time update of the depth prediction model, so as to improve the real-time performance and accuracy of the information features stored in the calculation cache server 303. In addition, the first hidden layer of the prediction server 301 is only used for further feature extraction of the data vector corresponding to the account data, and does not need to perform correlation processing on the information feature, so that the calculation amount of the depth prediction model can be reduced, and the calculation loss of the prediction server 301 is further reduced.
Fig. 7 is a flowchart illustrating an information recommendation method according to an exemplary embodiment, which is used in the calculation cache server 303 of fig. 3, as shown in fig. 7, and includes the following steps.
In step S701, the calculation cache server 303 performs feature extraction on information stored in the information database, obtains information features corresponding to the information, and caches the information features.
The information features refer to features corresponding to information stored in an information database, and the calculation cache server 303 may read information stored in the information database in advance, and perform feature extraction on the information, so as to obtain information features corresponding to the information, and perform cache processing on the information features.
In step S702, when a request for acquiring information features sent by the prediction server 301 is received, the calculation cache server 303 acquires information features corresponding to the request for acquiring information features from the cached information features and returns the information features to the prediction server 301, so that the prediction server 301 acquires account features corresponding to account data acquired from the storage server 302, determines correlation between the account features and the information features, and obtains recommendation information of the corresponding account based on the correlation.
After the calculation cache server 303 completes the caching of the information feature, if the prediction server 301 sends a request for acquiring the information feature to the calculation cache server 303, the calculation cache server 303 may obtain the information feature corresponding to the request from a plurality of information features cached in advance, and return the information feature to the prediction server 301, for example, the request may carry an information identifier corresponding to the information feature required to be acquired, and the calculation cache server 303 may find the corresponding information feature according to the information identifier. Then, the prediction server 301 may obtain account data from the storage server 302, and obtain account features of the account data, and finally, the prediction server 301 may determine recommendation information of the corresponding account by using correlation between the obtained account features and information features returned by the calculation cache server 303.
In the above information recommendation method, the calculation cache server 303 performs feature extraction on information stored in the information database, and obtains and caches information features corresponding to the information; if a request for acquiring information features sent by the prediction server 301 is received, information features corresponding to the request for acquiring information features are acquired from the cached information features and returned to the prediction server 301, so that the prediction server 301 acquires account features corresponding to the account data acquired from the storage server 302, determines correlation between the account features and the information features, and obtains recommendation information of the corresponding account based on the correlation. In the information recommendation method provided by the disclosure, the prediction server 301 can directly obtain the cached information features from the calculation cache server 303, so that the calculation process of information feature extraction can be reduced, and the calculation loss of the prediction server 301 is reduced.
In an exemplary embodiment, step S701 may further include: the calculation cache server 303 acquires and caches the model weight parameters from the storage server 302, and acquires information feature weight parameters for extracting information features from the cached model weight parameters; and carrying out feature extraction on the information stored in the information database according to the information feature weight parameters to obtain information features corresponding to the information.
The model weight parameters refer to model parameters of a depth prediction model configured in the prediction server 301 for determining recommendation information, where the model weight parameters may be obtained by training the depth prediction model by the training server 304 and stored in the storage server 302, and the model weight parameters may include account feature weight parameters for reflecting weight parameters corresponding to account features and information feature weight parameters for reflecting weight parameters corresponding to information features.
Specifically, the calculation cache server 303 may read the model weight parameter stored in the calculation cache server from the storage server 302 and cache the model weight parameter, and then, the calculation cache server 303 may further read the information feature weight parameter from the cached model weight parameter and perform feature extraction on the information stored in the information database by using the information feature weight parameter until the information feature corresponding to the information is obtained.
In addition, the information recommendation method may further include: when the cached model weight parameters are updated and the information stored in the information database is updated, the computing cache server 303 acquires the information feature weight parameters in the currently cached model weight parameters; and carrying out feature extraction on the information currently stored in the information database based on the information feature weight parameters to obtain information features corresponding to the information.
And if the model weight parameters cached by the calculation cache server 303 are updated, or the information stored in the information database is updated, the calculation cache server 303 can correspondingly update the information features stored in the calculation cache server. Specifically, the calculation cache server 303 may read the corresponding information feature weight parameter from the currently cached model weight parameter, and perform feature extraction on the information currently stored in the information database by using the obtained information feature weight parameter, so as to obtain the information feature corresponding to the currently stored information.
In the above embodiment, the calculation cache server 303 may implement real-time update of the cached information features based on the updated model weight parameters or the information stored in the information database, so as to improve the real-time performance and accuracy of the information features stored in the calculation cache server 303.
Fig. 8 is a flowchart illustrating an information recommendation method according to an exemplary embodiment, which is used in the application environment of fig. 1, as shown in fig. 8, and includes the following steps.
In step S801, the prediction server 101 transmits a request for acquiring account data to the storage server 102 and a request for acquiring information features to the calculation cache server 103, respectively, in response to the information recommendation request;
in step S802, the storage server 102 acquires account data corresponding to the request for acquiring account data from the stored account data and returns to the prediction server 101;
in step S803, the calculation cache server 103 acquires information features corresponding to the request for acquiring information features from the cached information features and returns to the prediction server 101;
in step S804, the prediction server 101 acquires account characteristics of account data;
in step S805, the prediction server 101 determines the correlation between the account feature and the information feature, and obtains recommendation information of the corresponding account according to the correlation.
In the information recommendation method, the prediction server 101 can directly obtain the information features from the calculation cache server 103, and the calculation process of information feature extraction can be reduced, so that the calculation loss of the prediction server 101 is reduced.
In an exemplary embodiment, an information recommendation system is provided, which implements information recommendation through a depth prediction model, where the structure of the depth prediction model may be as shown in fig. 9, and fig. 9 is a schematic diagram of the structure of the depth prediction model according to an exemplary embodiment, where the structure has the following features:
1. the input of the first layer of the hidden layer originates from user features and information features, which are independent of each other.
2. The number of hidden layer nodes decreases as the depth of the network increases, so that the calculation amount of the first layer of the hidden layer is the largest.
3. The computation of the first layer of the hidden layer is mainly vector matrix multiplication, and the batch information processing expands the vector matrix multiplication into matrix multiplication.
4. For a batch of information, the corresponding user features in the model calculation are the same.
Meanwhile, according to the matrix multiplication decomposition formula, the calculation of the first layer of the hidden layer may be decomposed as shown in fig. 10, and fig. 10 is a schematic diagram illustrating the calculation decomposition of the first layer of the hidden layer of the depth prediction model according to an exemplary embodiment. Wherein B is the quantity of batch processing information, M is the total length of the user features, N is the total length of the information features, and K is the number of hidden layer output nodes. By this decomposition, the number of multiplications calculated by the first layer of the hidden layer is reduced from B (m+n) K to b+n+m K. The computation of the first layer of hidden layer is thus focused on the matrix computation of the information part.
For a recommendation system, the information and model weight parameters are constant for a time that is much longer than the time for a relatively predicted service. Therefore, the hidden layer first layer result of all information can be calculated in advance. In the prediction process, the first layer calculation result of the information part can be obtained only by performing table lookup operation.
By utilizing the structural feature of the model, the present embodiment reforms the original service architecture by using a calculation cache manner, and the reformed service architecture is shown in fig. 11, and fig. 11 is a system architecture diagram of the information recommendation system according to an exemplary embodiment, where the system may include 4 clusters:
1. training a server cluster, updating weight parameters of a depth prediction model in a prediction server cluster by using training methods such as back propagation and the like, and storing the updated weight parameters in a storage server cluster;
2. the storage server cluster comprises a user data database and a model weight parameter database, wherein the user data database is used for storing user data, and the model weight parameter database is used for storing model weight parameters after training and updating of the training server cluster;
3. the computing cache server cluster comprises an information database, a model parameter cache module and a computing cache module, wherein the information database stores information for information recommendation and a hidden layer first-layer computing result of an information part, the model parameter cache module is used for reading and caching corresponding model parameters from a model weight parameter database in the storage server cluster, and when the information of the information database is updated or the model parameters cached by the model parameter cache module are updated, the hidden layer first-layer computing result of the information part can be recalculated into an updated cache;
4. The prediction server cluster can be used for receiving an external prediction service request, wherein the prediction service request can comprise a user number and a batch of information numbers, then user data can be inquired from the storage server cluster according to the user number, the user characteristics can be obtained by completing the characteristic extraction of the user data, the hidden layer first-layer calculation result of the information part can be obtained from the calculation cache server cluster according to the information number inquiry, the calculation is carried out by utilizing the model weight parameters of the depth prediction model, and finally a recommendation list is formed and the prediction service response is returned. And when the model weight parameters are updated, reading the model weight parameters from a model parameter cache module in the computing cache server cluster to update the depth prediction model.
In the above embodiment, the computing cache module introduced by the computing cache server cluster optimizes the service architecture of the recommendation system, and moves the feature extraction of the information part and the first layer of computation of the hidden layer to the computing cache server cluster, which is favorable for saving the computing consumption of the information part of the prediction server, improving the service capacity, saving the data storage amount and the network transmission bandwidth, and simultaneously improving the resource utilization rate of each server cluster.
Fig. 12 is a block diagram of an information recommendation device, according to an exemplary embodiment. Referring to fig. 12, the apparatus includes a recommendation request response unit 1201, an account feature acquisition unit 1202, and a recommendation information acquisition unit 1203, applied to a prediction server.
A recommendation request response unit 1201 configured to perform, in response to the information recommendation request, acquiring account data corresponding to the information recommendation request from the storage server and acquiring information features corresponding to the information recommendation request from the information features cached by the calculation cache server, respectively; the calculation cache server performs feature extraction on the information stored in the information database in advance to obtain information features corresponding to the information and caches the information features;
an account feature acquisition unit 1202 configured to execute account features;
the recommendation information obtaining unit 1203 is configured to determine a correlation between the account feature and the information feature, and obtain recommendation information of the corresponding account according to the correlation.
In an exemplary embodiment, a depth prediction model for determining recommendation information is configured in the prediction server; a recommendation information acquisition unit 1203 further configured to perform acquisition of cached model weight parameters from the calculation cache server; updating a local depth prediction model based on model weight parameters, and acquiring account characteristics corresponding to account data through the updated depth prediction model; and determining the correlation between the account characteristics and the information characteristics, and obtaining the recommendation information of the corresponding account according to the correlation.
In an exemplary embodiment, the account feature includes a data vector corresponding to the account data; a recommendation information obtaining unit 1203 further configured to perform obtaining a data vector, inputting the data vector into the first hidden layer of the updated depth prediction model; the first hidden layer is used for extracting the characteristics of the data vector to obtain the corresponding account characteristics; inputting account features and information features into a second hidden layer of the updated depth prediction model; the second hidden layer and other hidden layers are used for determining the correlation between the account characteristics and the information characteristics and determining the recommendation information of the corresponding account based on the correlation.
Fig. 13 is a block diagram of an information recommendation device, according to an exemplary embodiment. Referring to fig. 12, the apparatus is applied to a calculation cache server, and includes an information feature acquisition unit 1301 and an information feature transmission unit 1302.
An information feature obtaining unit 1301 configured to perform feature extraction on information stored in the information database, obtain information features corresponding to the information, and cache the information features;
the information feature sending unit 1302 is configured to perform, if a request for obtaining an information feature sent by the prediction server is received, obtaining an information feature corresponding to the request for obtaining the information feature from the cached information feature, and returning the information feature to the prediction server, so that the prediction server obtains an account feature corresponding to the account data obtained from the storage server, determines a correlation between the account feature and the information feature, and obtains recommendation information of the corresponding account based on the correlation.
In an exemplary embodiment, the information feature obtaining unit 1301 is further configured to perform obtaining and caching model weight parameters from the storage server, and obtaining information feature weight parameters for extracting information features from the cached model weight parameters; and carrying out feature extraction on the information stored in the information database according to the information feature weight parameters to obtain information features corresponding to the information.
In an exemplary embodiment, the information recommendation apparatus further includes: the information characteristic updating unit is configured to acquire the information characteristic weight parameters in the currently cached model weight parameters when the cached model weight parameters are updated and the information stored in the information database is updated; and carrying out feature extraction on the information currently stored in the information database based on the information feature weight parameters to obtain information features corresponding to the information.
The specific manner in which the various modules perform the operations in the apparatus of the above embodiments have been described in detail in connection with the embodiments of the method, and will not be described in detail herein.
FIG. 14 is a block diagram illustrating an apparatus 1400 for information recommendation, according to an example embodiment. For example, device S00 may be a server. Referring to fig. 14, the device 1400 includes a processing component 1420 that further includes one or more processors and memory resources, represented by memory 1422, for storing instructions, such as applications, executable by the processing component 1420. The application programs stored in memory 1422 can include one or more modules, each corresponding to a set of instructions. Further, the processing component 1420 is configured to execute instructions to perform the method of information recommendation described above.
The device 1400 may also include a power component 1424 configured to perform power management of the device 1400, a wired or wireless network interface 1426 configured to connect the device 1400 to a network, and an input output (I/O) interface 1428. The device 1400 may operate based on an operating system stored in memory 1422, such as Windows Server, mac OS XTM, unixTM, linuxTM, freeBSDTM, or the like.
In an exemplary embodiment, a storage medium is also provided that includes instructions, such as memory 1422 including instructions, that can be executed by a processor of device 1400 to perform the above-described method. The storage medium may be a non-transitory computer readable storage medium, which may be, for example, ROM, random Access Memory (RAM), CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.
Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This disclosure is intended to cover any adaptations, uses, or adaptations of the disclosure following the general principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.
It is to be understood that the present disclosure is not limited to the precise arrangements and instrumentalities shown in the drawings, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims (18)

1. An information recommendation system, comprising: the system comprises a prediction server, a storage server and a calculation cache server; wherein,
the storage server is used for storing account data;
the computing cache server is used for extracting features of information stored in the information database in advance, obtaining information features corresponding to the information and caching the information features;
the prediction server is used for responding to an information recommendation request, acquiring corresponding account data from the storage server, acquiring corresponding information features from the information features cached by the calculation cache server, acquiring account features corresponding to the account data, determining correlation between the account features and the information features, and acquiring recommendation information of corresponding accounts based on the correlation.
2. The system of claim 1, further comprising: a training server;
the prediction server is provided with a depth prediction model for determining recommendation information;
The training server is used for training the model weight parameters of the depth prediction model and storing the model weight parameters obtained by training to the storage server;
the calculation cache server is further used for acquiring and caching the model weight parameters from the storage server, acquiring information feature weight parameters for extracting information features from the cached model weight parameters, and performing feature extraction on information stored in an information database according to the information feature weight parameters to obtain information features corresponding to the information;
the prediction server is further configured to obtain cached model weight parameters from the calculation cache server, update a local depth prediction model based on the model weight parameters, obtain account features corresponding to the account data through the updated depth prediction model, determine correlation between the account features and the information features, and obtain recommendation information of corresponding accounts based on the correlation.
3. The system of claim 2, wherein the system further comprises a controller configured to control the controller,
the calculation cache server is further configured to obtain an information feature weight parameter in the currently cached model weight parameters when the cached model weight parameters are updated and/or information stored in the information database is updated, and perform feature extraction on information currently stored in the information database based on the information feature weight parameters to obtain information features corresponding to the information.
4. The system of claim 2, wherein the system further comprises a controller configured to control the controller,
the prediction server is further configured to obtain a data vector corresponding to the account data, input the data vector into a first hidden layer of the updated depth prediction model, input the information feature into a second hidden layer of the updated depth prediction model, and obtain recommendation information of a corresponding account according to output of the updated depth prediction model;
the first hiding layer is used for extracting features of the data vector to obtain corresponding account features, and the corresponding account features are input to the second hiding layer; the second hidden layer and other hidden layers are used for determining the correlation between the account characteristics and the information characteristics and determining the recommendation information of the corresponding account based on the correlation.
5. An information recommendation method, applied to a prediction server, comprising:
responding to an information recommendation request, respectively acquiring account data corresponding to the information recommendation request from a storage server, and acquiring information features corresponding to the information recommendation request from information features cached by a calculation cache server; the calculation cache server performs feature extraction on information stored in an information database in advance to obtain information features corresponding to the information and caches the information features;
Acquiring account characteristics of the account data;
and determining the correlation between the account characteristics and the information characteristics, and obtaining recommendation information of the corresponding account according to the correlation.
6. The method of claim 5, wherein the prediction server has a depth prediction model configured therein for determining recommendation information;
the determining the correlation between the account feature and the information feature, and obtaining the recommendation information of the corresponding account according to the correlation includes:
obtaining cached model weight parameters from the calculation cache server;
updating a local depth prediction model based on the model weight parameters, and acquiring account features corresponding to the account data through the updated depth prediction model;
and determining the correlation between the account characteristics and the information characteristics, and obtaining the recommendation information of the corresponding account according to the correlation.
7. The method of claim 6, wherein the account feature comprises a data vector corresponding to the account data;
the determining the correlation between the account feature and the information feature, and obtaining the recommendation information of the corresponding account according to the correlation includes:
Acquiring the data vector, and inputting the data vector into a first hidden layer of the updated depth prediction model; the first hidden layer is used for extracting features of the data vector to obtain corresponding account features;
inputting the account feature and the information feature into a second hidden layer of the updated depth prediction model; the second hidden layer and other hidden layers after the second hidden layer are used for determining correlation between the account characteristics and the information characteristics and determining recommendation information of the corresponding account based on the correlation.
8. An information recommendation method, which is applied to a calculation cache server, comprises the following steps:
extracting features of information stored in an information database in advance to obtain information features corresponding to the information, and caching the information features;
if a request for acquiring the information features sent by the prediction server is received, acquiring the information features corresponding to the request for acquiring the information features from the cached information features, and returning the information features to the prediction server, so that the prediction server acquires the account features corresponding to the account data acquired from the storage server, determines the correlation between the account features and the information features, and obtains recommendation information of corresponding accounts based on the correlation.
9. The method according to claim 8, wherein the feature extraction of the information stored in the information database, obtaining and caching the information features corresponding to the information, includes:
obtaining model weight parameters from the storage server, caching the model weight parameters, and obtaining information feature weight parameters for extracting information features from the cached model weight parameters;
and carrying out feature extraction on the information stored in the information database according to the information feature weight parameters to obtain information features corresponding to the information.
10. The method as recited in claim 9, further comprising:
when the cached model weight parameters are updated and/or the information stored in the information database is updated, acquiring information characteristic weight parameters in the currently cached model weight parameters;
and carrying out feature extraction on the information currently stored in the information database based on the information feature weight parameters to obtain information features corresponding to the information.
11. An information recommendation apparatus, applied to a prediction server, comprising:
a recommendation request response unit configured to execute, in response to an information recommendation request, obtaining account data corresponding to the information recommendation request from a storage server and obtaining information features corresponding to the information recommendation request from information features cached by a calculation cache server, respectively; the calculation cache server performs feature extraction on information stored in an information database in advance to obtain information features corresponding to the information and caches the information features;
An account feature acquisition unit configured to execute account features;
and the recommendation information acquisition unit is configured to determine the correlation between the account characteristics and the information characteristics and obtain recommendation information of the corresponding account according to the correlation.
12. The apparatus of claim 11, wherein the prediction server has a depth prediction model configured therein for determining recommendation information; the recommendation information acquisition unit is further configured to perform acquisition of cached model weight parameters from the calculation cache server; updating a local depth prediction model based on the model weight parameters, and acquiring account features corresponding to the account data through the updated depth prediction model; and determining the correlation between the account characteristics and the information characteristics, and obtaining the recommendation information of the corresponding account according to the correlation.
13. The apparatus of claim 12, wherein the account feature comprises a data vector corresponding to the account data; the recommendation information acquisition unit is further configured to perform acquisition of the data vector, and input the data vector into a first hidden layer of the updated depth prediction model; the first hidden layer is used for extracting features of the data vector to obtain corresponding account features; inputting the account feature and the information feature into a second hidden layer of the updated depth prediction model; the second hidden layer and other hidden layers after the second hidden layer are used for determining correlation between the account characteristics and the information characteristics and determining recommendation information of the corresponding account based on the correlation.
14. An information recommendation apparatus, applied to a calculation cache server, comprising:
the information characteristic acquisition unit is configured to perform characteristic extraction on information stored in an information database in advance, obtain information characteristics corresponding to the information and cache the information characteristics;
and the information feature sending unit is configured to execute the steps of obtaining the information feature corresponding to the request for obtaining the information feature from the cached information feature and returning the information feature to the prediction server if the request for obtaining the information feature sent by the prediction server is received, so that the prediction server obtains the account feature corresponding to the account data obtained from the storage server, determines the correlation between the account feature and the information feature, and obtains the recommendation information of the corresponding account based on the correlation.
15. The apparatus according to claim 14, wherein the information feature acquisition unit is further configured to perform acquisition of model weight parameters from the storage server and caching, and acquire information feature weight parameters for extracting information features from the cached model weight parameters; and carrying out feature extraction on the information stored in the information database according to the information feature weight parameters to obtain information features corresponding to the information.
16. The apparatus as recited in claim 15, further comprising: an information feature updating unit configured to perform updating of the cached model weight parameters, and/or obtain information feature weight parameters in the currently cached model weight parameters when updating of information stored in the information database occurs; and carrying out feature extraction on the information currently stored in the information database based on the information feature weight parameters to obtain information features corresponding to the information.
17. A server, comprising:
a processor;
a memory for storing the processor-executable instructions;
wherein the processor is configured to execute the instructions to implement the information recommendation method of any of claims 5 to 10.
18. A computer readable storage medium, which when executed by a processor of a server, causes the server to perform the information recommendation method of any of claims 5 to 10.
CN202011573356.6A 2020-12-24 2020-12-24 Information recommendation system, method, device, server and storage medium Active CN112650778B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011573356.6A CN112650778B (en) 2020-12-24 2020-12-24 Information recommendation system, method, device, server and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011573356.6A CN112650778B (en) 2020-12-24 2020-12-24 Information recommendation system, method, device, server and storage medium

Publications (2)

Publication Number Publication Date
CN112650778A CN112650778A (en) 2021-04-13
CN112650778B true CN112650778B (en) 2024-03-12

Family

ID=75363178

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011573356.6A Active CN112650778B (en) 2020-12-24 2020-12-24 Information recommendation system, method, device, server and storage medium

Country Status (1)

Country Link
CN (1) CN112650778B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111400615A (en) * 2020-03-19 2020-07-10 腾讯科技（深圳）有限公司 Resource recommendation method, device, equipment and storage medium
CN111881358A (en) * 2020-07-31 2020-11-03 北京达佳互联信息技术有限公司 Object recommendation system, method and device, electronic equipment and storage medium
CN111984864A (en) * 2020-08-14 2020-11-24 北京达佳互联信息技术有限公司 Object recommendation method and device, electronic equipment and storage medium

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10949909B2 (en) * 2017-02-24 2021-03-16 Sap Se Optimized recommendation engine

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111400615A (en) * 2020-03-19 2020-07-10 腾讯科技（深圳）有限公司 Resource recommendation method, device, equipment and storage medium
CN111881358A (en) * 2020-07-31 2020-11-03 北京达佳互联信息技术有限公司 Object recommendation system, method and device, electronic equipment and storage medium
CN111984864A (en) * 2020-08-14 2020-11-24 北京达佳互联信息技术有限公司 Object recommendation method and device, electronic equipment and storage medium

Also Published As

Publication number Publication date
CN112650778A (en) 2021-04-13

Similar Documents

Publication Publication Date Title
US20220391771A1 (en) Method, apparatus, and computer device and storage medium for distributed training of machine learning model
US11941527B2 (en) Population based training of neural networks
CN110458663B (en) Vehicle recommendation method, device, equipment and storage medium
US20200184316A1 (en) Generating discrete latent representations of input data items
EP3579169A1 (en) Learned model provision method, and learned model provision device
CN111967339B (en) Method and device for planning unmanned aerial vehicle path
CN110633796B (en) Model updating method and device, electronic equipment and storage medium
CN114548426B (en) Asynchronous federal learning method, business service prediction method, device and system
CN114036398B (en) Content recommendation and ranking model training method, device, equipment and storage medium
JP2022058328A (en) Apparatus and method for distributed model training, electronic device, storage medium, and computer program
CN112163601A (en) Image classification method, system, computer device and storage medium
JP7285977B2 (en) Neural network training methods, devices, electronics, media and program products
CN115147265A (en) Virtual image generation method and device, electronic equipment and storage medium
CN113641835B (en) Multimedia resource recommendation method and device, electronic equipment and medium
CN112668608B (en) Image recognition method and device, electronic equipment and storage medium
CN114003571A (en) Data processing method and device, electronic equipment and storage medium
CN108319722B (en) Data access method and device, electronic equipment and computer readable storage medium
CN117253287B (en) Action prediction model training method based on domain generalization, related method and product
CN112650778B (en) Information recommendation system, method, device, server and storage medium
WO2023231887A1 (en) Tensor-based continual learning method and device
CN110502715B (en) Click probability prediction method and device
CN110399979B (en) Click rate pre-estimation system and method based on field programmable gate array
CN112749005B (en) Resource data processing method, device, computer equipment and storage medium
JP2022068146A (en) Method for annotating data, apparatus, storage medium, and computer program
CN114357242A (en) Training evaluation method and device based on recall model, equipment and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant