CN113742590A - Recommendation method and device, storage medium and electronic equipment - Google Patents
Recommendation method and device, storage medium and electronic equipment Download PDFInfo
- Publication number
- CN113742590A CN113742590A CN202111045547.XA CN202111045547A CN113742590A CN 113742590 A CN113742590 A CN 113742590A CN 202111045547 A CN202111045547 A CN 202111045547A CN 113742590 A CN113742590 A CN 113742590A
- Authority
- CN
- China
- Prior art keywords
- target
- recommendation
- probability
- recommended
- fusion
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000007499 fusion processing Methods 0.000 claims abstract description 23
- 230000004927 fusion Effects 0.000 claims description 83
- 230000006399 behavior Effects 0.000 claims description 38
- 230000006870 function Effects 0.000 claims description 25
- 238000012549 training Methods 0.000 claims description 24
- 238000000605 extraction Methods 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 9
- 238000012163 sequencing technique Methods 0.000 claims description 9
- 239000013598 vector Substances 0.000 description 26
- 238000010586 diagram Methods 0.000 description 10
- 230000008569 process Effects 0.000 description 9
- 238000006243 chemical reaction Methods 0.000 description 6
- 230000003287 optical effect Effects 0.000 description 6
- 238000013475 authorization Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 239000013307 optical fiber Substances 0.000 description 2
- 230000002093 peripheral effect Effects 0.000 description 2
- 230000000644 propagated effect Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000008707 rearrangement Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The embodiment of the invention discloses a recommendation method, a recommendation device, a storage medium and electronic equipment. The method comprises the following steps: acquiring user behavior data and recommended object data, and extracting characteristic information based on the user behavior data and the recommended object data; inputting the characteristic information into a multi-target recommendation model to obtain a multi-target recommendation probability of a recommended object; determining multi-target weights based on the characteristic information, and performing fusion processing on the multi-target recommendation probability of the recommended object based on the multi-target weights to obtain the target recommendation probability of the recommended object; and recommending based on the target recommendation probability of each recommendation object. In the embodiment, the output weight of each target task is determined in an individualized manner, the multi-target recommendation probability is further subjected to fusion processing based on the multi-target weight to obtain the target recommendation probability meeting the user specific requirement, and the recommended object is recommended based on the target recommendation probability, so that the recommendation accuracy of the recommended object is improved.
Description
Technical Field
The embodiment of the invention relates to the technical field of deep learning, in particular to a recommendation method, a recommendation device, a storage medium and electronic equipment.
Background
Under the scene of information recommendation, in order to better improve the income of an information recommendation platform, the income of an information delivery party and the user experience, the demand of accurate recommendation to a user exists.
However, in the process of implementing the present invention, the inventors found that at least the following technical problems exist in the prior art: in the current recommendation scheme, the recommendation accuracy and the recommendation flexibility are poor, and personalized recommendation cannot be performed according to the difference of users.
Disclosure of Invention
The embodiment of the invention provides a recommendation method, a recommendation device, a storage medium and electronic equipment, and aims to realize high-precision recommendation.
In a first aspect, an embodiment of the present invention provides a recommendation method, including:
acquiring user behavior data and recommended object data, and extracting characteristic information based on the user behavior data and the recommended object data;
inputting the characteristic information into a multi-target recommendation model to obtain a multi-target recommendation probability of a recommended object;
determining multi-target weights based on the characteristic information, and performing fusion processing on the multi-target recommendation probability of the recommended object based on the multi-target weights to obtain the target recommendation probability of the recommended object;
and recommending based on the target recommendation probability of each recommendation object.
Optionally, the determining multiple target weights based on the feature information, and performing fusion processing on the multiple target recommendation probabilities of the recommended object based on the multiple target weights to obtain the target recommendation probability of the recommended object includes:
and inputting the characteristic information and the multi-target recommendation probability into a fusion model to obtain the target recommendation probability output by the fusion model.
Optionally, the fusion model includes a weight determination module and a fusion module, wherein;
the weight determination module is used for generating multi-target weights based on the characteristic information;
and the fusion module is used for carrying out fusion processing on the multi-target recommendation probability based on the multi-target weight to obtain the target recommendation probability of the recommendation object.
Optionally, the training method of the fusion model includes:
acquiring sample data and extracting characteristic information of the sample data;
processing the characteristic information based on a fusion model to be trained, and obtaining a prediction probability based on a multi-target recommendation probability output by a multi-target recommendation model;
generating a loss function based on the prediction probability, a plurality of target labels corresponding to the sample data and a preset loss fusion hyper-parameter, and training the fusion model to be trained based on the loss function.
Optionally, the determining the multi-target weights based on the feature information includes:
and inputting the characteristic information into a weight identification model to obtain the multi-target weight output by the weight identification model.
Optionally, the recommending based on the target recommendation probability of each recommendation object includes:
determining the recommended objects with the target recommendation probability meeting a preset value as recommended objects to be recommended, sorting the screened recommended objects based on the target recommendation probability, determining a recommendation sequence, and recommending based on the recommendation sequence; or,
and sequencing the plurality of recommended objects based on the target recommendation probability, and recommending the recommended objects in a preset sequencing range based on a sequencing order.
Optionally, the multi-objective recommendation model includes: a plurality of target networks, a plurality of gating networks, a plurality of expert networks, and a plurality of fusion modules, wherein,
each expert network is used for extracting features of input feature information from one dimension, each gate control network is used for outputting weights of a target task under multiple different dimensions, each fusion module is used for fusing the features extracted under the multiple different dimensions and output by the multiple expert networks according to the weights of the target task under the multiple different dimensions, and each target network is used for predicting the recommendation probability under a target task.
Optionally, the multi-target recommendation probability includes a click rate and a conversion rate.
In a second aspect, an embodiment of the present invention further provides a recommendation apparatus, including:
the characteristic information extraction module is used for acquiring user behavior data and recommended object data and extracting characteristic information based on the user behavior data and the recommended object data;
the multi-target recommendation probability determining module is used for inputting the characteristic information into a multi-target recommendation model to obtain the multi-target recommendation probability of the recommended object;
the target recommendation probability determining module is used for determining multi-target weights based on the characteristic information, and performing fusion processing on the multi-target recommendation probability of the recommended object based on the multi-target weights to obtain the target recommendation probability of the recommended object;
and the recommending module is used for recommending based on the target recommending probability of each recommending object.
In a third aspect, an embodiment of the present invention further provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor executes the computer program to implement the recommendation method provided in any embodiment of the present invention.
In a fourth aspect, the embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the recommendation method according to any embodiment of the present invention.
According to the technical scheme, the multi-target recommendation probability and the multi-target recommendation probability are determined based on the characteristic information by extracting the characteristic information from the acquired user behavior data and the acquired recommended object data, and the multi-target recommendation probability of the recommended object is subjected to fusion processing based on the multi-target weight to obtain the target recommendation probability of the recommended object. The multi-target recommendation probability is comprehensively predicted from the perspective of multiple dimensions and multiple targets, comprehensiveness and pertinence of probability prediction are improved, meanwhile, current scenes and user intentions are identified in feature information obtained based on user behavior data and recommendation object data, weights of all target tasks are determined in a targeted mode, fusion processing is further conducted on the multi-target recommendation probability based on the multi-target weights, target recommendation probability meeting specific requirements of users is obtained, recommendation objects are recommended based on the target recommendation probability, and recommendation accuracy of the recommended objects is improved.
Drawings
Fig. 1 is a schematic flowchart of a recommendation method according to an embodiment of the present invention;
FIG. 2 is a schematic structural diagram of a multi-objective recommendation model according to an embodiment of the present invention;
FIG. 3 is a flow chart of determining a target recommendation probability provided by an embodiment of the invention;
FIG. 4 is a schematic diagram of a process for recommending probabilities according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of a recommendation device according to a second embodiment of the present invention;
fig. 6 is a schematic structural diagram of an electronic device according to a third embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and are not limiting of the invention. It should be further noted that, for the convenience of description, only some of the structures related to the present invention are shown in the drawings, not all of the structures.
Example one
Fig. 1 is a flowchart of a recommendation method according to an embodiment of the present invention, where the present embodiment is applicable to a case of recommending an item to a user, the method may be executed by a recommendation apparatus provided in an embodiment of the present invention, the recommendation apparatus may be implemented by software and/or hardware, and the recommendation apparatus may be configured on a terminal such as a mobile phone, a tablet computer, and the like, and an electronic computing device such as a computer or a server.
The method specifically comprises the following steps:
s110, obtaining user behavior data and recommendation object data, and extracting characteristic information based on the user behavior data and the recommendation object data.
And S120, inputting the characteristic information into a multi-target recommendation model to obtain the multi-target recommendation probability of the recommended object.
S130, determining multi-target weights based on the characteristic information, and fusing the multi-target recommendation probability of the recommended object based on the multi-target weights to obtain the target recommendation probability of the recommended object.
And S140, recommending based on the target recommendation probability of each recommendation object.
In this embodiment, a recommendation request is received, and in response to the recommendation request, the recommendation method of this embodiment is executed to recommend an object to a user. The recommendation request may be generated when the recommended application is detected to be started or logged in, or when a refresh operation in the recommended application is detected, or when the recommendation control or the recommendation case is detected to be clicked, or when a recommendation instruction input in a voice manner is received. The recommendation request generation method is not limited, and any type of recommendation request can be received.
And responding to the recommendation request, and acquiring user behavior data and recommendation object data, wherein the user behavior data can be behavior data of a login user or behavior data of a current operation user. Optionally, the behavior data of the user is determined based on the user identification. Optionally, the biometric feature of the current operating user is obtained, and the current operating user is identified by the biometric feature, it should be noted that the user identifier and the biometric feature may be obtained by authorization of the user. And matching in a preset identification library through one or more of the biological characteristics to determine the corresponding user. Correspondingly, different users correspond to different behavior data respectively. In the historical operation process, the behavior data and the user identification are stored in a correlation mode, and the corresponding behavior data can be extracted according to the user identification of the user.
Optionally, the user behavior data includes, but is not limited to, user browsing data, user click data, user transaction data, user evaluation data, user forwarding data, user collection data, user sharing data, and the like within a first preset time range. The first preset time range may be one week, 15 days, or one month before the current time, which is not limited herein.
The recommendation object may be any type of object that makes a recommendation, and is not limited in this respect. The recommended object data includes, but is not limited to, an object type, an object attribute, browsing data of the object within a second preset time range, click data, transaction data, evaluation data, forwarding data, collection data, sharing data, and the like, where the second preset time range may be one week, 15 days, or one month, and the like, before the current time, and is not limited thereto. It should be noted that the user behavior data and the recommendation object data may be obtained by authorization of the user.
The feature extraction is performed on the user behavior data and the recommendation object data, and may be performed, for example, on the basis of a preset feature extraction model. The feature extraction model may be a machine learning model, such as a neural network model, a deep neural network model, or the like. Optionally, the user behavior data and the recommended object data are respectively input to a feature extraction model, so as to obtain user feature information and recommended object feature information output by the feature extraction model. Optionally, the user behavior data and the recommended object data are synchronously input to the feature extraction model, so as to obtain the comprehensive feature information output by the feature extraction model.
In some embodiments, the obtained feature information is used as input information of the multi-target recommendation model, wherein the feature information is a feature vector.
In some embodiments, the input information for the multi-objective recommendation model includes: the system comprises a user interest vector, a recommended object vector and a dense feature vector, wherein the user interest vector is a feature vector obtained by feature extraction based on user behavior data and recommended object data, the recommended object vector is a feature vector obtained by feature extraction based on the recommended object data, and the dense feature vector comprises a user portrait, a recommended object portrait and interaction behavior features of a user and the recommended object.
And inputting the input information determined by any one of the embodiments into the multi-target recommendation model to obtain the multi-target recommendation probability output by the multi-target recommendation model. The multi-target recommendation model can respectively perform targeted processing according to a plurality of recommended target tasks, so that the recommendation probability of each target task is obtained, and the comprehensiveness and the pertinence of recommendation are improved compared with the recommendation model of a single target. In this embodiment, the Multi-target recommendation Model may be, but is not limited to, an MMOE (Multi-gate Mixture-of-Experts) Model, an ESMM (Whole Space Multi-task Model), and is not limited thereto.
Optionally, the multi-objective recommendation model includes: the system comprises a plurality of target networks, a plurality of gate control networks, a plurality of expert networks and a plurality of fusion modules, wherein each expert network is used for extracting features of input feature information from one dimension, each gate control network is used for outputting the weight of a target task under a plurality of different dimensions, each fusion module is used for fusing the extracted features of the expert networks under the different dimensions according to the weight of the target task under the different dimensions, and each target network is used for predicting the recommendation probability under a target task.
For example, referring to fig. 2, fig. 2 is a schematic structural diagram of a multi-target recommendation model according to an embodiment of the present invention. Fig. 2 includes four expert networks, two gate control networks, two fusion modules, and two target networks, where the two target networks correspond to a click task and a conversion task, respectively. The input information is synchronously input into each expert network and each gating network, wherein each expert network respectively extracts the features of different dimensions of the input information and respectively outputs the extracted feature vectors, the gating network A is used for generating the fusion weight of a target network corresponding to the click task, and the gating network B is used for generating the fusion weight of the target network corresponding to the conversion task. The first fusion network performs fusion processing on the feature vectors of the expert networks based on the fusion weight generated by the gate control network A, the fused feature vectors are input to target networks corresponding to the click task, the second fusion network performs fusion processing on the feature vectors of the expert networks based on the fusion weight generated by the gate control network B, the fused feature vectors are input to the target networks corresponding to the conversion task, the two target networks respectively input recommendation probabilities to form multi-target recommendation probabilities, and the multi-target recommendation probabilities are output in the form of probability vectors.
In fig. 2, the multi-objective recommendation probability includes click through rate and conversion rate. It should be noted that fig. 2 is only an example, and in other embodiments, the number of the expert networks, the number of the target networks, and the number of the gating networks in the multi-target recommendation model may be set according to requirements, which is not limited in this respect. Correspondingly, corresponding target tasks can be increased according to recommendation requirements, and the multi-target recommendation probability corresponds to the recommendation probability comprising the multi-target recommendation probability.
The recommendation probabilities of a plurality of target tasks are obtained through the multi-target recommendation model, and the multi-target recommendation probabilities output by the multi-target recommendation model cannot meet the difference recommendation requirements of different users aiming at the conflict among the target tasks and different user intentions.
In this embodiment, multi-target weights are determined based on the user behavior data and feature information extracted from the recommended object data, and fusion processing is performed on the multi-target recommendation probability of the recommended object based on the multi-target weights to obtain the target recommendation probability of the recommended object. The multi-target weight comprises weights corresponding to a plurality of target tasks respectively, and the weight processing is carried out on the basis of the weight of each target task and the recommendation probability of each target task to obtain the target recommendation probability. Optionally, the multi-target weight may be output in a form of a weight vector, and accordingly, the multi-target recommendation probability of the recommended object is fused based on the multi-target weight, and the multi-target weight and the multi-target recommendation probability may be vector-multiplied to obtain the target recommendation probability.
In this embodiment, the current scene and the user intention may be identified based on the input feature information, and the multi-target weight that meets the current scene and the user intention may be output, that is, the multi-target weight is a user-customized weight, and the multi-target weights of different users are different. And fusing the multi-target recommendation probability according with the multi-target weight of the current scene and the user intention to obtain the target recommendation probability according with the user characteristic, and recommending a plurality of recommendation objects based on the target recommendation probability to realize difference recommendation aiming at the user requirements.
In some optional embodiments, determining the multi-objective weights based on the feature information includes: and inputting the characteristic information into a weight identification model to obtain the multi-target weight output by the weight identification model. The weight identification model may be a preset network model, and may exemplarily include a plurality of fully connected layers, which are used for processing the input feature information, identifying the current scene and the user intention, and outputting the weight of each target task.
The recommendation method includes recommending a plurality of recommendation objects based on a target recommendation probability, selecting recommendation objects to be recommended based on the target recommendation probability, and exemplarily determining recommendation objects with recommendation probabilities meeting a preset value as recommendation objects to be recommended, sorting the selected recommendation objects based on the target recommendation probability, and determining a recommendation sequence. For example, the plurality of recommended objects may be sorted based on the target recommendation probability, for example, the recommended objects are sorted according to a sorting mode that the target recommendation probability is from large to small, and the recommended objects in a preset sorting range are recommended based on a sorting order. The preset sorting range may be N bits before sorting, where N may be preset, or determined according to a display requirement of the user terminal, for example, the number of first screen displays of the user terminal.
It should be noted that the two steps of inputting the feature information into the multi-target recommendation model to obtain the multi-target recommendation probability of the recommended object and determining the multi-target weight based on the feature information may be executed synchronously or in any order, which is not limited to this.
According to the technical scheme of the embodiment, the target recommendation probability of the recommended object is obtained by extracting the characteristic information from the acquired user behavior data and the acquired recommended object data, determining the multi-target recommendation probability and the multi-target fusion weight based on the characteristic information, and performing fusion processing on the multi-target recommendation probability of the recommended object based on the multi-target weight. The multi-target recommendation probability is comprehensively predicted from the perspective of multiple dimensions and multiple targets, comprehensiveness and pertinence of probability prediction are improved, meanwhile, current scenes and user intentions are identified from feature information obtained based on user behavior data and recommendation object data, weights of all target tasks are determined in a targeted mode, fusion processing is further conducted on the multi-target recommendation probability based on the multi-target weights, target recommendation probability meeting specific requirements of users is obtained, recommendation is conducted on the recommendation object based on the target recommendation probability, recommendation accuracy of the recommendation object is improved, and personalized recommendation is achieved.
On the basis of the above embodiment, determining multi-target weights based on the feature information, and performing fusion processing on the multi-target recommendation probabilities of the recommended objects based on the multi-target weights to obtain the target recommendation probabilities of the recommended objects, includes: and inputting the characteristic information and the multi-target recommendation probability into a fusion model to obtain the target recommendation probability output by the fusion model.
The feature information and the multi-target recommendation probability are subjected to fusion processing through a fusion model, and the target recommendation probability is directly output.
In some embodiments, the fusion model comprises a weight determination module and a fusion module, wherein; the weight determination module is used for generating multi-target weights based on the characteristic information; and the fusion module is used for carrying out fusion processing on the multi-target recommendation probability based on the multi-target weight to obtain the target recommendation probability of the recommendation object.
In some embodiments, the weight determination module may be a gated network, and may be composed of three layers of fully-connected networks. Illustratively, the network parameter is set to 256 × 128 × n _ task, where n _ task is the target task number, the activation function is the RELU function, and the fusion weight vector is obtained by the Softmax function. The specific structure of the weight determination module is not limited, and may be set as needed.
Exemplarily, referring to fig. 3, fig. 3 is a flow chart for determining a target recommendation probability according to an embodiment of the present invention. FIG. 3 is a diagram showing input information of feature information determined based on user behavior data and recommended object data, the feature information being respectively input into a fusion model and a multi-objective recommendation model, a weight determination module in the fusion model outputting multi-objective weights based on the feature information and inputting the multi-objective weights to the fusion module, the multi-objective recommendation probability output by the multi-objective recommendation model being input into the fusion module in the fusion model, and the fusion module performing fusion processing on the multi-objective weights and the multi-objective recommendation probability to obtain the objective recommendation probability of the recommended object.
On the basis of the above embodiment, the fusion model and the multi-objective recommendation model may be trained independently. The method comprises the steps of obtaining sample data in a training process of a multi-target recommendation model, iterating training characteristic information based on the sample data, processing the training characteristic information based on the multi-target recommendation model to be trained to obtain multi-target recommendation probability, obtaining labels of all target tasks in the sample data, generating a loss function based on the multi-target recommendation probability and the labels of all the target tasks, and adjusting parameters of the multi-target recommendation model based on the loss function. And the training process is iteratively executed until the training condition is met. Wherein the loss function may be:
wherein L ismmoeFor the loss function, n is the target task number, yiFor the output of the target task i,
is a true tag of the target task i, αiFusion weight superparameter, function for loss of target task i
Cross entropy loss is calculated. Wherein alpha isiMay be preset.
For the fusion model, the training method of the fusion model comprises the following steps: acquiring sample data and extracting characteristic information of the sample data; processing the characteristic information based on a fusion model to be trained, and obtaining a prediction probability based on a multi-target recommendation probability output by a multi-target recommendation model; generating a loss function based on the prediction probability, a plurality of target labels corresponding to the sample data and a preset loss fusion hyper-parameter, and training the fusion model to be trained based on the loss function. And the training process is iteratively executed until the training condition is met.
In this embodiment, the training processes of the fusion model and the multi-target recommendation model are alternately executed, and the required sample data is the same, that is, the training feature information corresponding to the sample data is respectively input into the fusion model to be trained and the multi-target recommendation model to be trained. And outputting the multi-target recommendation probability by the multi-target recommendation model to be trained, and obtaining a loss function of the multi-target recommendation model based on the multi-target recommendation probability and the plurality of target labels corresponding to the sample data. The fusion model receives the multi-target recommendation probability output by the multi-target recommendation model at the same time and outputs a prediction probability yfinalAnd y · w, wherein y is the multi-target recommendation probability, and w is the multi-target weight generated by the weight determination module in the fusion model. Generating a loss function of a fusion model based on the prediction probability, a plurality of target labels corresponding to the sample data and a preset loss fusion hyper-parameter, wherein the loss function of the multi-target recommendation model trains the multi-target recommendation model, the loss function of the fusion model trains the fusion model, and the multi-target recommendation model recommendsThe training process of the model and the fusion model has no mutual influence.
Correspondingly, the generating of the loss function based on the prediction probability, the plurality of target labels corresponding to the sample data, and the preset loss fusion hyper-parameter may be obtained by the following formula:
where n is the number of target tasks, yfinalThe predicted probability that is output for the fusion model,
is a true tag of the target task i, αiAnd fusing the weight hyperparameters for the losses of the target task i.
The training conditions of the fusion model and the multi-target recommendation model are the same, and for example, the training conditions may be the same iteration number or preset accuracy of the prediction probability. It should be noted that the fusion model and the multi-objective recommendation model share input information, but in the back propagation process, the fusion model only updates parameters of the fusion model itself, and extraction of characteristic information is not affected. The purpose of this is to avoid the fusion model and the multi-target recommendation model from influencing the model parameters due to different learning targets. The fusion model and the multi-target recommendation model adopt an alternate training mode to update parameters, and because partial network parameters of the fusion model are simple, the invention can set the multi-target recommendation model for 10 steps of iterative training and set the fusion model for 1 step of iterative training.
A preferred example is further provided on the basis of the above embodiment, for example, referring to fig. 4, fig. 4 is a schematic diagram of a processing procedure of recommending a probability provided by the embodiment of the present invention. In the preferred embodiment, a data set containing a user behavior sequence (namely user behavior data), candidate commodities (namely recommended objects) and dense features is constructed, and a target task of model learning is defined, wherein the target task can be a click task CTR and a conversion task CVR. The data stream in the data set is vectorized by the ID and interest extraction module (for example, the characteristics can also beExtraction model), for example, one-hot encoding is performed on recommendation object data (e.g., attribute information such as categories, brands, etc.), and then low-dimensional embedding is performed through end-to-end training to obtain a recommendation object vector xi. The interest extraction network utilizes an attention mechanism to perform interest vector x on the user according to the inflow recommendation object data and the user behavior sequenceuAnd modeling. Dense feature xdense(including user portrait, commodity portrait and user and commodity interactive behavior characteristics) and recommendation object vector xiUser interest vector xuAnd splicing to obtain input information x.
And (3) obtaining the output of each task by the input vector x through a multi-target recommendation model MMoE, and recording as a multi-target output vector, namely an output vector y of multi-target recommendation probability.
The input vector x is input to a gated network GATEX in the fusion model, which identifies the current scene and user intent from user behavior, candidate goods and some contextual features, and outputs a set of fusion weight vectors ω, i.e., multi-objective weight terms.
Performing dot product on the multi-target recommendation probability y and the multi-target weight item omega, and calculating final output yfinalAs a basis for recommendation of the target object.
Example two
Fig. 5 is a schematic structural diagram of a recommendation device according to a second embodiment of the present invention, where the recommendation device includes:
a feature information extraction module 210, configured to obtain user behavior data and recommended object data, and extract feature information based on the user behavior data and the recommended object data;
the multi-target recommendation probability determining module 220 is configured to input the feature information into a multi-target recommendation model to obtain a multi-target recommendation probability of a recommended object;
a target recommendation probability determining module 230, configured to determine multi-target weights based on the feature information, and perform fusion processing on the multi-target recommendation probabilities of the recommended objects based on the multi-target weights to obtain target recommendation probabilities of the recommended objects;
and the recommending module 240 is used for recommending based on the target recommending probability of each recommending object.
On the basis of the foregoing embodiment, optionally, the multi-target recommendation probability determination module is configured to:
and inputting the characteristic information and the multi-target recommendation probability into a fusion model to obtain the target recommendation probability output by the fusion model.
Optionally, the fusion model includes a weight determination module and a fusion module, wherein;
the weight determination module is used for generating multi-target weights based on the characteristic information;
and the fusion module is used for carrying out fusion processing on the multi-target recommendation probability based on the multi-target weight to obtain the target recommendation probability of the recommendation object.
Optionally, the apparatus further comprises: the fusion model training module is used for:
acquiring sample data and extracting characteristic information of the sample data;
processing the characteristic information based on a fusion model to be trained, and obtaining a prediction probability based on a multi-target recommendation probability output by a multi-target recommendation model;
generating a loss function based on the prediction probability, a plurality of target labels corresponding to the sample data and a preset loss fusion hyper-parameter, and training the fusion model to be trained based on the loss function.
On the basis of the foregoing embodiment, optionally, the target recommendation probability determining module is configured to:
and inputting the characteristic information into a weight identification model to obtain the multi-target weight output by the weight identification model.
On the basis of the foregoing embodiment, optionally, the multi-target recommendation model includes:
a plurality of target networks, a plurality of gating networks, a plurality of expert networks, and a plurality of fusion modules, wherein,
each expert network is used for extracting features of input feature information from one dimension, each gate control network is used for outputting weights of a target task under multiple different dimensions, and each fusion module is used for fusing the features extracted under the multiple different dimensions and output by the multiple expert networks according to the weights of the target task under the multiple different dimensions.
On the basis of the above embodiment, optionally, the recommending module 240 is configured to:
determining the recommended objects with the target recommendation probability meeting a preset value as recommended objects to be recommended, sorting the screened recommended objects based on the target recommendation probability, determining a recommendation sequence, and recommending based on the recommendation sequence; or,
and sequencing the plurality of recommended objects based on the target recommendation probability, and recommending the recommended objects in a preset sequencing range based on a sequencing order.
The recommendation device provided by the embodiment of the invention can execute the recommendation method provided by any embodiment of the invention, and has the corresponding functional modules and beneficial effects of executing the recommendation method.
EXAMPLE III
Fig. 6 is a schematic structural diagram of an electronic device according to a third embodiment of the present invention. FIG. 6 illustrates a block diagram of an electronic device 12 suitable for use in implementing embodiments of the present invention. The electronic device 12 shown in fig. 6 is only an example and should not bring any limitation to the function and the scope of use of the embodiment of the present invention. The device 12 is typically an electronic device that undertakes image classification functions.
As shown in FIG. 6, electronic device 12 is embodied in the form of a general purpose computing device. The components of electronic device 12 may include, but are not limited to: one or more processors 16, a memory device 28, and a bus 18 that connects the various system components (including the memory device 28 and the processors 16).
A program 36 having a set (at least one) of program modules 26 may be stored, for example, in storage 28, such program modules 26 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which examples or some combination thereof may include an implementation of a gateway environment. Program modules 26 generally perform the functions and/or methodologies of the described embodiments of the invention.
The processor 16 executes various functional applications and data processing, for example, implementing the recommended methods provided by the above-described embodiments of the present invention, by executing programs stored in the storage device 28.
Example four
A fourth embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the recommendation method provided by the embodiment of the present invention.
Of course, the computer program stored on the computer-readable storage medium provided by the embodiment of the present invention is not limited to the method operations described above, and may also execute the recommendation method provided by any embodiment of the present invention.
Computer storage media for embodiments of the invention may employ any combination of one or more computer-readable media. The computer readable medium may be a computer readable signal medium or a computer readable storage medium. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.
A computer readable signal medium may include a propagated data signal with computer readable source code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.
Source code embodied on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
Computer source code for carrying out operations for aspects of the present invention may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C + +, or the like, as well as conventional procedural programming languages, such as the "C" programming language or similar programming languages. The source code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of gateway, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider).
It is to be noted that the foregoing is only illustrative of the preferred embodiments of the present invention and the technical principles employed. It will be understood by those skilled in the art that the present invention is not limited to the particular embodiments described herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the invention. Therefore, although the present invention has been described in greater detail by the above embodiments, the present invention is not limited to the above embodiments, and may include other equivalent embodiments without departing from the spirit of the present invention, and the scope of the present invention is determined by the scope of the appended claims.
Claims (10)
1. A recommendation method, comprising:
acquiring user behavior data and recommended object data, and extracting characteristic information based on the user behavior data and the recommended object data;
inputting the characteristic information into a multi-target recommendation model to obtain a multi-target recommendation probability of a recommended object;
determining multi-target weights based on the characteristic information, and performing fusion processing on the multi-target recommendation probability of the recommended object based on the multi-target weights to obtain the target recommendation probability of the recommended object;
and recommending based on the target recommendation probability of each recommendation object.
2. The method of claim 1, wherein the determining multi-objective weights based on the feature information, and performing fusion processing on the multi-objective recommendation probabilities of the recommended objects based on the multi-objective weights to obtain the target recommendation probabilities of the recommended objects comprises:
and inputting the characteristic information and the multi-target recommendation probability into a fusion model to obtain the target recommendation probability output by the fusion model.
3. The method of claim 2, wherein the fusion model comprises a weight determination module and a fusion module, wherein;
the weight determination module is used for generating multi-target weights based on the characteristic information;
and the fusion module is used for carrying out fusion processing on the multi-target recommendation probability based on the multi-target weight to obtain the target recommendation probability of the recommendation object.
4. The method of claim 2, wherein the training method of the fusion model comprises:
acquiring sample data and extracting characteristic information of the sample data;
processing the characteristic information based on a fusion model to be trained, and obtaining a prediction probability based on a multi-target recommendation probability output by a multi-target recommendation model;
generating a loss function based on the prediction probability, a plurality of target labels corresponding to the sample data and a preset loss fusion hyper-parameter, and training the fusion model to be trained based on the loss function.
5. The method of claim 1, wherein determining multi-objective weights based on the feature information comprises:
and inputting the characteristic information into a weight identification model to obtain the multi-target weight output by the weight identification model.
6. The method of claim 1, wherein the recommending based on the target recommendation probability of each recommended object comprises:
determining the recommended objects with the target recommendation probability meeting a preset value as recommended objects to be recommended, sorting the screened recommended objects based on the target recommendation probability, determining a recommendation sequence, and recommending based on the recommendation sequence; or,
and sequencing the plurality of recommended objects based on the target recommendation probability, and recommending the recommended objects in a preset sequencing range based on a sequencing order.
7. The method of claim 1, wherein the multi-objective recommendation model comprises:
a plurality of target networks, a plurality of gating networks, a plurality of expert networks, and a plurality of fusion modules, wherein,
each expert network is used for extracting features of input feature information from one dimension, each gate control network is used for outputting weights of a target task under multiple different dimensions, each fusion module is used for fusing the features extracted under the multiple different dimensions and output by the multiple expert networks according to the weights of the target task under the multiple different dimensions, and each target network is used for predicting the recommendation probability under a target task.
8. A recommendation device, comprising:
the characteristic information extraction module is used for acquiring user behavior data and recommended object data and extracting characteristic information based on the user behavior data and the recommended object data;
the multi-target recommendation probability determining module is used for inputting the characteristic information into a multi-target recommendation model to obtain the multi-target recommendation probability of the recommended object;
the target recommendation probability determining module is used for determining multi-target weights based on the characteristic information, and performing fusion processing on the multi-target recommendation probability of the recommended object based on the multi-target weights to obtain the target recommendation probability of the recommended object;
and the recommending module is used for recommending based on the target recommending probability of each recommending object.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the recommendation method as claimed in any one of claims 1-7 when executing the program.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the recommendation method according to any one of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111045547.XA CN113742590A (en) | 2021-09-07 | 2021-09-07 | Recommendation method and device, storage medium and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111045547.XA CN113742590A (en) | 2021-09-07 | 2021-09-07 | Recommendation method and device, storage medium and electronic equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113742590A true CN113742590A (en) | 2021-12-03 |
Family
ID=78736824
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111045547.XA Pending CN113742590A (en) | 2021-09-07 | 2021-09-07 | Recommendation method and device, storage medium and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113742590A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116385070A (en) * | 2023-01-18 | 2023-07-04 | 中国科学技术大学 | Multi-target prediction method, system, equipment and storage medium for short video advertisement of E-commerce |
| CN117408296A (en) * | 2023-12-14 | 2024-01-16 | 深圳须弥云图空间科技有限公司 | Sequence recommendation depth ordering method and device for multitasking and multi-scene |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014056408A1 (en) * | 2012-10-08 | 2014-04-17 | 腾讯科技（深圳）有限公司 | Information recommending method, device and server |
| CN109408731A (en) * | 2018-12-27 | 2019-03-01 | 网易（杭州）网络有限公司 | A kind of multiple target recommended method, multiple target recommended models generation method and device |
| CN109785147A (en) * | 2018-10-24 | 2019-05-21 | 中国平安人寿保险股份有限公司 | Insurance kind sort method and device, electronic equipment and computer readable storage medium |
| CN112183818A (en) * | 2020-09-02 | 2021-01-05 | 北京三快在线科技有限公司 | Recommendation probability prediction method and device, electronic equipment and storage medium |
| CN112232546A (en) * | 2020-09-09 | 2021-01-15 | 北京三快在线科技有限公司 | Recommendation probability estimation method and device, electronic equipment and storage medium |
| CN112699305A (en) * | 2021-01-08 | 2021-04-23 | 网易传媒科技（北京）有限公司 | Multi-target recommendation method, device, computing equipment and medium |
-
2021
- 2021-09-07 CN CN202111045547.XA patent/CN113742590A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014056408A1 (en) * | 2012-10-08 | 2014-04-17 | 腾讯科技（深圳）有限公司 | Information recommending method, device and server |
| CN109785147A (en) * | 2018-10-24 | 2019-05-21 | 中国平安人寿保险股份有限公司 | Insurance kind sort method and device, electronic equipment and computer readable storage medium |
| CN109408731A (en) * | 2018-12-27 | 2019-03-01 | 网易（杭州）网络有限公司 | A kind of multiple target recommended method, multiple target recommended models generation method and device |
| CN112183818A (en) * | 2020-09-02 | 2021-01-05 | 北京三快在线科技有限公司 | Recommendation probability prediction method and device, electronic equipment and storage medium |
| CN112232546A (en) * | 2020-09-09 | 2021-01-15 | 北京三快在线科技有限公司 | Recommendation probability estimation method and device, electronic equipment and storage medium |
| CN112699305A (en) * | 2021-01-08 | 2021-04-23 | 网易传媒科技（北京）有限公司 | Multi-target recommendation method, device, computing equipment and medium |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116385070A (en) * | 2023-01-18 | 2023-07-04 | 中国科学技术大学 | Multi-target prediction method, system, equipment and storage medium for short video advertisement of E-commerce |
| CN116385070B (en) * | 2023-01-18 | 2023-10-03 | 中国科学技术大学 | Multi-target prediction method, system, equipment and storage medium for short video advertisement of E-commerce |
| CN117408296A (en) * | 2023-12-14 | 2024-01-16 | 深圳须弥云图空间科技有限公司 | Sequence recommendation depth ordering method and device for multitasking and multi-scene |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20220262032A1 (en) | Systems and Methods for Geolocation Prediction | |
| CN110555469B (en) | Method and device for processing interactive sequence data | |
| CN111966914B (en) | Content recommendation method and device based on artificial intelligence and computer equipment | |
| CN109740018B (en) | Method and device for generating video label model | |
| CN112115257B (en) | Method and device for generating information evaluation model | |
| CN109145828B (en) | Method and apparatus for generating video category detection model | |
| CN108062377A (en) | The foundation of label picture collection, definite method, apparatus, equipment and the medium of label | |
| CN108564102A (en) | Image clustering evaluation of result method and apparatus | |
| CN111708876A (en) | Method and device for generating information | |
| CN112149699B (en) | Method and device for generating model and method and device for identifying image | |
| CN113742590A (en) | Recommendation method and device, storage medium and electronic equipment | |
| CN109784243B (en) | Identity determination method and device, neural network training method and device, and medium | |
| CN109902681B (en) | User group relation determining method, device, equipment and storage medium | |
| CN112989146A (en) | Method, apparatus, device, medium, and program product for recommending resources to a target user | |
| CN114860892B (en) | Hierarchical category prediction method, device, equipment and medium | |
| CN111104874A (en) | Face age prediction method, training method and device of model and electronic equipment | |
| CN114842411A (en) | Group behavior identification method based on complementary space-time information modeling | |
| CN113140012B (en) | Image processing method, device, medium and electronic equipment | |
| CN112861474B (en) | Information labeling method, device, equipment and computer readable storage medium | |
| CN113705293A (en) | Image scene recognition method, device, equipment and readable storage medium | |
| CN117312845A (en) | Sample labeling method, medium and electronic equipment | |
| CN116257704A (en) | Point-of-interest recommendation method based on user space-time behaviors and social information | |
| CN113672807B (en) | Recommendation method, recommendation device, recommendation medium, recommendation device and computing equipment | |
| CN113255819B (en) | Method and device for identifying information | |
| CN112651942B (en) | Layout detection method and device |
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 |