CN113836406A - Information flow recommendation method and device - Google Patents
Information flow recommendation method and device Download PDFInfo
- Publication number
- CN113836406A CN113836406A CN202111064025.4A CN202111064025A CN113836406A CN 113836406 A CN113836406 A CN 113836406A CN 202111064025 A CN202111064025 A CN 202111064025A CN 113836406 A CN113836406 A CN 113836406A
- Authority
- CN
- China
- Prior art keywords
- information flow
- sample data
- network
- label
- flow recommendation
- 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
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
- G06F18/25—Fusion techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Biology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The disclosure relates to the technical field of big data processing, in particular to an information flow recommendation method and device. An information flow recommendation method, comprising: acquiring user characteristics and information flow characteristics; inputting the user characteristics and the information flow characteristics into the trained information flow recommendation network to obtain a recommendation result output by the information flow recommendation network; the information flow recommendation network is obtained by pre-training based on a multi-target fusion label, and the multi-target fusion label is obtained by processing according to label values of at least two dimensions of sample data. The method simplifies the network structure and the training process, and improves the network effect and robustness.
Description
Technical Field
The disclosure relates to the technical field of big data processing, in particular to an information flow recommendation method and device.
Background
The information flow recommendation is a method for constructing and training a learning network by mining features so as to match and deduce information flows which a user may like. In an information flow recommendation scenario, targets of multiple dimensions are often considered, so that the structure of a recommendation network is complex and the effect is not good.
Disclosure of Invention
In order to solve the technical problem, the embodiments of the present disclosure provide an information flow recommendation method and apparatus, an information flow recommendation network training method and apparatus, an electronic device, and a storage medium.
In a first aspect, an embodiment of the present disclosure provides an information flow recommendation method, including:
acquiring user characteristics and information flow characteristics;
inputting the user characteristics and the information flow characteristics into a trained information flow recommendation network to obtain a recommendation result output by the information flow recommendation network; the information flow recommendation network is obtained by pre-training based on a multi-target fusion label, and the multi-target fusion label is obtained by processing according to label values of at least two dimensions of sample data.
In some embodiments, the training process of the information flow recommendation network includes:
acquiring a sample data set; the sample data set comprises a plurality of sample data and the multi-target fusion label corresponding to each sample data;
inputting the sample data set into an untrained information flow recommendation network to obtain an output value output by the information flow recommendation network;
and adjusting the network parameters of the information flow recommendation network based on the difference between the output value corresponding to each sample data and the multi-target fusion label until a convergence condition is met, and obtaining the trained information flow recommendation network.
In some embodiments, each of the sample data includes tag values of at least two dimensions, and the determining the multi-target fusion tag corresponding to the sample data includes:
obtaining tag values of the at least two dimensions included in the sample data;
and performing fusion processing on the label values of the at least two dimensions to obtain the multi-target fusion label corresponding to the sample data.
In some embodiments, the fusing the tag values of the at least two dimensions to obtain the multi-target fusion tag corresponding to the sample data includes:
and performing Gaussian smoothing processing on the label values of the at least two dimensions to obtain the multi-target fusion label.
In some embodiments, the information stream includes multimedia information, and the tag types corresponding to the tag values include a first tag type and a second tag type, where the first tag type represents an operation condition of a user on the multimedia information, and the second tag type represents a playing condition of the user on the multimedia information.
In a second aspect, an embodiment of the present disclosure provides a method for training an information flow recommendation network, including:
the method comprises the steps of obtaining a sample data set, wherein the sample data set comprises a plurality of sample data, and each sample data comprises at least two-dimensional label values;
for each sample data, performing fusion processing based on the label values of the at least two dimensions of the sample data to obtain a multi-target fusion label corresponding to the sample data;
inputting the sample data set into an untrained information flow recommendation network to obtain an output value output by the information flow recommendation network;
and adjusting the network parameters of the information flow recommendation network based on the difference between the output value corresponding to each sample data and the multi-target fusion label until a convergence condition is met, and obtaining the trained information flow recommendation network.
In a third aspect, an embodiment of the present disclosure provides an information flow recommendation apparatus, including:
an acquisition module configured to acquire user characteristics and information flow characteristics;
the recommending module is configured to input the user characteristics and the information flow characteristics into a trained information flow recommending network to obtain a recommending result output by the information flow recommending network; the information flow recommendation network is obtained by pre-training based on a multi-target fusion label, and the multi-target fusion label is obtained by processing label values of at least two dimensions of sample data
In some embodiments, the information flow recommendation apparatus of embodiments of the present disclosure further includes a network training module configured to:
acquiring a sample data set; the sample data set comprises a plurality of sample data and the multi-target fusion label corresponding to each sample data;
inputting the sample data set into an untrained information flow recommendation network to obtain an output value output by the information flow recommendation network;
and adjusting the network parameters of the information flow recommendation network based on the difference between the output value corresponding to each sample data and the multi-target fusion label until a convergence condition is met, and obtaining the trained information flow recommendation network.
In some embodiments, each said sample data comprises tag values in at least two dimensions, the apparatus further comprising a tag determination module configured to:
obtaining tag values of the at least two dimensions included in the sample data;
and performing fusion processing on the label values of the at least two dimensions to obtain the multi-target fusion label corresponding to the sample data.
In a fourth aspect, an embodiment of the present disclosure provides a training apparatus for recommending a network by an information flow, including a network training module, where the network training module is configured to:
the method comprises the steps of obtaining a sample data set, wherein the sample data set comprises a plurality of sample data, and each sample data comprises at least two-dimensional label values;
for each sample data, performing fusion processing based on the label values of the at least two dimensions of the sample data to obtain a multi-target fusion label corresponding to the sample data;
inputting the sample data set into an untrained information flow recommendation network to obtain an output value output by the information flow recommendation network;
and adjusting the network parameters of the information flow recommendation network based on the difference between the output value corresponding to each sample data and the multi-target fusion label until a convergence condition is met, and obtaining the trained information flow recommendation network.
In a fifth aspect, the present disclosure provides an electronic device, including:
a processor; and
a memory storing computer instructions readable by the processor, the processor performing the method of any of the embodiments of the first aspect or the second aspect when the computer instructions are read.
In a sixth aspect, the embodiments of the present disclosure provide a storage medium for storing computer-readable instructions for causing a computer to execute the method according to any one of the embodiments of the first aspect or the second aspect.
The information flow recommendation method of the embodiment of the disclosure comprises the steps of obtaining user characteristics and information flow characteristics, inputting the user characteristics and the information flow characteristics into a trained information flow recommendation network, and obtaining a recommendation result output by the information flow recommendation network; the information flow recommendation network is obtained by pre-training based on the multi-target fusion label, and the multi-target fusion label is obtained by processing according to label values of at least two dimensions of sample data. In the embodiment of the disclosure, the multi-target fusion label is used for training the information flow recommendation network, a plurality of networks do not need to be constructed and trained, and the network structure is simplified. And because the multi-target fusion label fuses label values of multiple dimensions, the network can learn the relevant characteristics of multiple targets and the associated characteristics among the targets in the iterative training process, so that the network training efficiency is improved, and the network prediction effect is improved. In addition, the information flow recommendation network output of the embodiment of the disclosure can be directly used as a final prediction result, and does not depend on artificial experience, so that the network expression capability and robustness are improved.
Drawings
In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present disclosure, and other drawings can be obtained by those skilled in the art without creative efforts.
Fig. 1 is a flow chart of a method of information flow recommendation in some embodiments according to the present disclosure.
Fig. 2 is a flow chart of a method of information flow recommendation in some embodiments according to the present disclosure.
Fig. 3 is a flow chart of a method of information flow recommendation in some embodiments according to the present disclosure.
Fig. 4 is a flow chart of a method of information flow recommendation in some embodiments according to the present disclosure.
Fig. 5 is a flow diagram of a method of training an information flow recommendation network in accordance with some embodiments of the present disclosure.
FIG. 6 is a schematic diagram of a training method for an information flow recommendation network in accordance with some embodiments of the present disclosure.
Fig. 7 is a block diagram of an information flow recommendation device according to some embodiments of the present disclosure.
Fig. 8 is a block diagram of an information flow recommendation device according to some embodiments of the present disclosure.
FIG. 9 is a block diagram of an electronic device suitable for implementing the disclosed method.
Detailed Description
The technical solutions of the present disclosure will be described clearly and completely with reference to the accompanying drawings, and it is to be understood that the described embodiments are only some embodiments of the present disclosure, but not all embodiments. All other embodiments, which can be derived by one of ordinary skill in the art from the embodiments disclosed herein without making any creative effort, shall fall within the protection scope of the present disclosure. In addition, technical features involved in different embodiments of the present disclosure described below may be combined with each other as long as they do not conflict with each other.
In an information flow recommendation scenario, targets of multiple dimensions are often required to be considered. Taking multimedia information recommendation as an example, when a recommendation system intelligently recommends appropriate content to a user, multi-dimensional system targets such as video types, click rates, play completion rates, praise amount, and comment amount need to be considered.
In the related art, an intelligent recommendation system based on machine learning needs to construct and train a model for each target individually, and then fusion is performed on output layers of a plurality of models to meet the requirements of multi-target tasks. However, for scenes such as multimedia information recommendation, the network structure of the recommendation system is complex due to the fact that the system targets are numerous and the model is built independently for each target, and the incidence relation among multiple targets cannot be fused due to the fact that the model is built and trained independently. In addition, when outputs of a plurality of models are fused, a weight needs to be set for an output value of each model depending on manual experience, so that the recommendation system is greatly influenced by the manual experience, and the robustness and the prediction effect are poor.
Based on the defects in the related art, the embodiments of the present disclosure provide an information flow recommendation method and apparatus, an information flow recommendation network training method and apparatus, an electronic device, and a storage medium.
In a first aspect, the embodiments of the present disclosure provide an information flow recommendation method, which is applicable to an electronic device, and the electronic device may be any type of device, such as a computer, a server, a wearable device, a mobile terminal, and the like, and the disclosure is not limited thereto.
As shown in fig. 1, in some embodiments, an information flow recommendation method of an example of the present disclosure includes:
and S110, acquiring user characteristics and information flow characteristics.
And S120, inputting the user characteristics and the information flow characteristics into the trained information flow recommendation network to obtain a recommendation result output by the information flow recommendation network.
Specifically, the user characteristics refer to characteristics of multiple dimensions for a target user, and reflect individual information, preferences, habits, and the like of the user. Taking the multimedia information recommendation scenario as an example, the user characteristics may include user information (e.g., gender, age, school calendar, location, etc.), historical browsing records, and the like.
The information flow characteristics refer to characteristics of multiple dimensions of the target information flow, and reflect types, audience conditions and the like of the target information flow. Still taking the multimedia information recommendation scenario as an example, the information flow characteristics may include information flow type (e.g., laugh, love, gourmet, etc.), creator, actor, etc.
In the embodiment of the disclosure, the input of the information flow recommendation network is the user characteristic and the information flow characteristic, and the output is the recommendation result, and the recommendation result indicates whether the information flow corresponding to the information flow characteristic is suitable for being recommended to the user.
In some embodiments, the information flow recommendation network may be any pre-trained based on a machine learning model or a neural network model, and may be applied to a unit module for performing ranking recommendation on information flows. For example, the information recommendation network may adopt a neural network with architecture such as deep FM, LR, FM, etc., which is not limited by the present disclosure.
In some embodiments, the recommendation result output by the information flow recommendation network may be a matching probability between the target information flow recommendation and the target user, so that when actually deployed, a corresponding probability threshold may be set, and the information flow with the matching probability greater than the probability threshold is recommended to the target user. In other embodiments, the recommendation result output by the information flow recommendation network may also be a ranking of matching probabilities of the plurality of information flows and the target user, so that Top-k information flows ranked in the Top may be recommended to the target user in actual deployment. This is understood and fully implemented by those skilled in the art, and the present disclosure will not be described in detail herein.
In the embodiment of the disclosure, the information flow recommendation network is obtained by pre-training based on the multi-target fusion tag. It can be understood that the purpose of network training of the information flow recommendation network is to make the output value of the information flow recommendation network as close as possible to the label value of the sample data, so that the information flow recommendation network has a good prediction effect. However, based on the foregoing, for an information flow recommendation scenario, the targets trained by the network often include multiple dimensions, and therefore, in the prior art, a sub-network needs to be separately constructed and trained for the targets of each dimension, so that each sub-network can learn the relevant features of the targets of one dimension.
However, in the embodiment of the present disclosure, a sub-network does not need to be separately constructed and trained for each target, but a multi-target fusion tag corresponding to sample data is obtained by performing fusion processing on tag values of multiple dimensions of the sample data, and an information flow recommendation network is trained by using the multi-target fusion tag. It can be understood that the multi-objective fusion label fuses label values of multiple dimensions, so that the information flow recommendation network can learn the relevant features of multiple dimensional targets and the associated features between the multiple dimensional targets in the training process.
Taking a multimedia information recommendation scene as an example, the click rate and the play completion rate are common training targets for multimedia information recommendation. Taking a sample data as an example, the sample data includes a user characteristic of the user a, an information flow characteristic of the multimedia information B, and a tag value corresponding to the click rate and the play-out rate. The click rate represents the click condition of the user A on the multimedia information B, and the completion play rate represents the watching condition of the user A on watching the multimedia information B.
In the prior art, two sub-networks need to be respectively constructed, wherein one sub-network takes the click-through rate as a training target, and the other sub-network takes the completion rate as a training target. And then, in the application stage of the information flow recommendation network, performing weighted fusion on the outputs of the two sub-networks based on artificial experience to obtain a prediction result of the recommendation network.
In the embodiment of the disclosure, during network training, sample data is processed first, and for each sample data, the click rate and the end broadcast rate tag value included in the sample data are fused to obtain a multi-target fusion tag fusing the click rate and the end broadcast rate, where the multi-target fusion tag represents the fusion feature of the user a for clicking and viewing the multimedia information B. And performing network training by taking the multi-target fusion label as a label value of the sample data, so that the information flow recommendation network can learn the fusion characteristics of a plurality of dimensional targets in one round of iterative training.
It should be noted that the foregoing examples only illustrate the information flow recommendation method of the present disclosure, and do not limit the present disclosure, and in other embodiments, the information flow may be an information flow in other forms, and the tag value corresponding to the information flow may also be a tag value in other dimensions, which is not limited by the present disclosure.
According to the method and the device, the multi-target fusion label is obtained through the fusion processing of the label values of multiple dimensions based on the sample data, the information flow recommendation network is trained through the multi-target fusion label, a plurality of networks do not need to be constructed and trained, and the network structure is simplified. In addition, as the multi-target fusion label fuses label values of multiple dimensions, the network can learn relevant characteristics of multiple targets and correlation characteristics among the targets in the iterative training process, so that the network training efficiency is improved, and the network prediction effect is improved. In addition, the information flow recommendation network output of the embodiment of the disclosure can be directly used as a final prediction result, and does not depend on artificial experience, so that the network expression capability and robustness are improved.
As shown in fig. 2, in some embodiments, in an information flow recommendation method according to an example of the present disclosure, a process of training an information flow recommendation network includes:
and S210, acquiring a sample data set.
And S220, inputting the sample data set into an untrained information flow recommendation network to obtain an output value output by the information flow recommendation network.
And S230, adjusting network parameters of the information flow recommendation network based on the difference between the output value corresponding to each sample data and the multi-target fusion label until a convergence condition is met, and obtaining the trained information flow recommendation network.
In particular, the sample data set is a set of training samples, which includes a plurality of sample data. In some embodiments, the sample data may be purged based on historical data. In network training, each sample data has a corresponding label (label) representing the true value of the sample data, and the purpose of network training is to make the output value of the network as close as possible to the true value represented by the label.
In the embodiment of the present disclosure, based on the foregoing, the label of each sample data includes label values of multiple dimensions, and each label value represents a true value of the sample data in the corresponding dimension.
Taking a multimedia information recommendation scene as an example, in an exemplary sample data, tag types corresponding to tag values of the sample data may include a first type tag and a second type tag, where the first type tag may represent an operation condition of a user on multimedia information, such as a click rate, a like rate, a comment rate, and the like. The second type tag can represent the playing condition of the multimedia information by the user, such as the playing completion rate, the playing progress and the like. In one example, the tag values of the sample data include a first tag value of a click rate dimension and a second tag value of an end rate dimension. The first label value is 1, and the user clicks the video corresponding to the sample data; the second label value is 0.95, which represents that the user watches 95% of the content of the video corresponding to the sample data.
It can be understood that the tag value of the sample data may be obtained by manual tagging or may be determined by monitoring the historical operation record of the user, which is not limited by the present disclosure.
In the embodiment of the disclosure, before training the information flow recommendation network, a sample data set needs to be processed to obtain a multi-target fusion tag corresponding to each sample data. As shown in fig. 3, in some embodiments, the process of determining a multi-target fusion tag comprises:
s211, obtaining the label values of at least two dimensions included by the sample data.
S212, fusion processing is carried out on the label values of at least two dimensions, and a multi-target fusion label corresponding to the sample data is obtained.
Based on the foregoing, each sample data has label values of multiple dimensions. In the embodiment of the disclosure, for each sample data in the sample data set, fusion processing is performed based on label values of multiple dimensions corresponding to the sample data, so as to obtain a multi-target fusion label fused with multiple dimension features.
In some embodiments, a gaussian function may be used to perform tag smoothing on tag values of multiple dimensions of sample data, so as to fuse the tag values of multiple dimensions into a multi-target fusion tag. Those skilled in the art can understand that specific parameters of the gaussian function can be set according to a specific service scenario, and the variables of the gaussian function can be adjusted according to data corresponding to different services, so that the variables conform to data distribution. This can be understood and fully implemented by those skilled in the art, and the present disclosure will not be described in detail herein.
It is to be understood that the above embodiments are merely exemplary illustrations of the present disclosure, and in other embodiments, the algorithm for the multidimensional tag value fusion process is not limited to the gaussian smoothing process, but may be any other fusion algorithm suitable for implementation, such as an exponential function, a trigonometric function, and the like, which is not limited by the present disclosure.
After the label values of the sample data are subjected to fusion processing, each sample data in the sample data set corresponds to a multi-target fusion label, and network training is performed on the information flow recommendation network based on the sample data set.
Firstly, sample data in a sample data set can be input into an untrained information flow recommendation network, the information flow recommendation network can be built based on a deep FM framework, and for the construction of the information flow recommendation network, a person skilled in the art can understand and fully realize the construction based on the related technology, and the description of the disclosure is omitted.
In some embodiments, a sigmoid function may be employed by an activation layer of the information flow recommendation network, so that the information flow recommendation network predicts and inputs a corresponding output value based on the user characteristics of the sample data and the information flow characteristics. The sigmoid value domain is positioned in [0,1], so that the output value can be well restricted in the [0,1] range. The output value represents the matching degree between the user characteristic predicted and output by the information flow recommendation network and the information flow characteristic, and the higher the output value is, the higher the probability of recommending the information flow to the user is.
After the output value of the information flow recommendation network is obtained, the difference between the output value and the multi-target fusion label obtained in advance can be determined. In some embodiments, the cross entropy loss function can be used to calculate a loss value between the output value and the multi-target fusion tag, which is the difference between the two. The larger the loss value is, the larger the difference between the predicted value and the true value of the network is, the worse the network effect is, and vice versa. Therefore, the network parameters of the information flow recommendation network can be adjusted according to the back propagation of the loss value, and the process of continuously optimizing and adjusting the network parameters is the process of network training in the loop iteration.
And stopping the training process of the information flow recommendation network until the information flow recommendation network meets the convergence condition, so as to obtain the trained information flow recommendation network. The convergence condition of the network may be set according to a specific training requirement, for example, when the prediction accuracy of the network meets the requirement, the convergence condition may be determined to be satisfied, and for example, when the iteration training turns meet a preset turn, the convergence condition may be determined to be satisfied, which is not limited in this disclosure.
According to the method and the device, the multi-target fusion label is obtained through the fusion processing of the label values of multiple dimensions based on the sample data, the information flow recommendation network is trained through the multi-target fusion label, a plurality of networks do not need to be constructed and trained, and the network structure is simplified. In addition, as the multi-target fusion label fuses label values of multiple dimensions, the network can learn relevant characteristics of multiple targets and correlation characteristics among the targets in the iterative training process, so that the network training efficiency is improved, and the network prediction effect is improved. In addition, the information flow recommendation network output of the embodiment of the disclosure can be directly used as a final prediction result, and does not depend on artificial experience, so that the network expression capability and robustness are improved.
Fig. 4 shows some embodiments of the information recommendation method of the present disclosure, in the embodiment of fig. 4, an information flow recommendation scene takes multimedia information recommendation as an example, and the following description is made with reference to fig. 4.
As shown in fig. 4, in some embodiments, an information flow recommendation method of an example of the present disclosure includes:
s401, acquiring a sample data set.
In some implementations, sample data in the sample data set may be obtained based on a user historical browsing history. Each sample data in the sample data set includes user characteristics (e.g., user age, gender, scholarship, etc.) and information flow characteristics (e.g., video type, amount of approval, creator, etc.).
Each sample data corresponds to a plurality of label values of dimensions, such as any number of broadcast completion rates, click rates, approval rates, comment rates, forwarding rates, and the like. The click rate shows the condition that the user clicks the video, the play completion rate shows the proportion of the video watched by the user, the praise rate shows the condition that the user praise the video, the comment amount shows the condition that the user commends the video or sends a barrage, and the forwarding rate shows the condition that the user forwards the video. Of course, the sample data may also include tag values for other dimensions, which are not enumerated by this disclosure.
S402, fusing label values of multiple dimensions included in the sample data to obtain a multi-target fusion label corresponding to the sample data.
In some embodiments, each sample data includes tag values of two dimensions, namely a first tag value of the click rate dimension and a second tag value of the end play rate dimension. In one example, a gaussian smoothing algorithm may be used to perform fusion processing on the first tag value and the second tag value, so as to obtain a multi-target fusion tag corresponding to sample data.
For example, in an exemplary sample data, the first tag value of the click-through rate dimension is 1, indicating that the user clicked on the video. The second label value of the end play rate dimension is 0.95, indicating that the user watched 0.95 of the video content. Therefore, the multi-target fusion label obtained by performing fusion processing on the first label value and the second label value by using gaussian smoothing may be 0.85, which means that the multi-target fusion label corresponding to the sample data is 0.85.
And S403, inputting the sample data set into an untrained information flow recommendation network to obtain an output value output by the information flow recommendation network.
S404, adjusting network parameters of the information flow recommendation network based on the difference between the output value corresponding to each sample data and the multi-target fusion label until a convergence condition is met, and obtaining the trained information flow recommendation network.
Specifically, taking the sample data of the above example as an example, the sample data is input into an untrained information flow recommendation network, and an output value obtained by activating a sigmoid function at an output layer is 0.3.
It can be understood that the output value 0.3 represents a predicted value of the information flow recommendation network, and the multi-target fusion tag 0.85 represents a true value, so that a loss value between the two is obtained by using a cross entropy function, and the original network parameters are adjusted by using the loss value. And repeating the iteration until a convergence condition is met to obtain the trained information flow recommendation network.
S405, inputting the user characteristics and the information flow characteristics into the trained information flow recommendation network to obtain a recommendation result output by the information flow recommendation network.
After the training of the information flow recommendation network is completed, the information flow recommendation network can be used for information flow recommendation. Inputting the obtained user characteristics and the information flow characteristics into the trained information flow recommendation network, wherein the output value of the information flow recommendation network is a recommendation result, the recommendation result represents the matching degree between the user and the information flow, the higher the matching degree is, the higher the probability of recommending the video to the user is, and the opposite is true.
According to the method and the device, the multi-target fusion label is obtained through the fusion processing of the label values of multiple dimensions based on the sample data, the information flow recommendation network is trained through the multi-target fusion label, a plurality of networks do not need to be constructed and trained, and the network structure is simplified. In addition, as the multi-target fusion label fuses label values of multiple dimensions, the network can learn relevant characteristics of multiple targets and correlation characteristics among the targets in the iterative training process, so that the network training efficiency is improved, and the network prediction effect is improved. In addition, the information flow recommendation network output of the embodiment of the disclosure can be directly used as a final prediction result, and does not depend on artificial experience, so that the network expression capability and robustness are improved.
In a second aspect, the embodiments of the present disclosure provide a training method for an information flow recommendation network, where the method is applicable to an electronic device, and the electronic device may be any device suitable for deploying an information flow recommendation network, such as a computer, a server, a wearable device, a mobile terminal, and the like, and the disclosure is not limited thereto.
As shown in fig. 5, in some embodiments, a training method of an information flow recommendation network of an example of the present disclosure includes:
and S510, acquiring a sample data set.
And S520, performing fusion processing on each sample data based on the label values of at least two dimensions of the sample data to obtain a multi-target fusion label corresponding to the sample data.
S530, inputting the sample data set into the untrained information flow recommendation network to obtain an output value output by the information flow recommendation network.
And S540, adjusting network parameters of the information flow recommendation network based on the difference between the output value corresponding to each sample data and the multi-target fusion label until a convergence condition is met, and obtaining the trained information flow recommendation network.
In some embodiments, fig. 6 illustrates the principle of recommending network training for information flows in the embodiments of the present disclosure, and the following description is made with reference to fig. 6.
As shown in fig. 6, in the embodiment of the present disclosure, the network training process includes two parts, namely, an information flow recommendation network 700 and a fusion processing module 800. Taking a sample data as an example, taking the user characteristic and the information flow characteristic of the sample data as the input of the information flow recommendation network 700, the hidden layer performs higher-dimensional characteristic extraction and processing on the user characteristic and the information flow characteristic, and outputs a corresponding output value through the sigmoid activation layer, wherein the output value represents a predicted value of the information flow recommendation network 700.
Meanwhile, the fusion processing module 800 performs fusion processing on a plurality of tag values of the sample data. As shown in fig. 6, the tag values corresponding to the sample data include a first tag value and a second tag value, and the fusion processing module 800 may perform fusion processing on the first tag value and the second tag value based on, for example, a gaussian smoothing algorithm to obtain a multi-target fusion tag, where the multi-target fusion tag represents a true value of the sample data.
In the example of fig. 6, the cross entropy loss function is used to calculate the loss value between the multi-target fusion tag and the output value, and then the back propagation is used to adjust the network parameters of the information flow recommendation network 700. And repeating the iteration until the network convergence condition is met, finishing the training of the information flow recommendation network 700, and obtaining the trained information flow recommendation network 700.
Where not described in detail, those skilled in the art will no doubt understand and will fully appreciate the foregoing embodiments and any further description of the disclosure is deemed necessary.
According to the method and the device, the multi-target fusion label is obtained through the fusion processing of the label values of multiple dimensions based on the sample data, the information flow recommendation network is trained through the multi-target fusion label, a plurality of networks do not need to be constructed and trained, and the network structure is simplified. In addition, as the multi-target fusion label fuses label values of multiple dimensions, the network can learn relevant characteristics of multiple targets and correlation characteristics among the targets in the iterative training process, so that the network training efficiency is improved, and the network prediction effect is improved. In addition, the information flow recommendation network output of the embodiment of the disclosure can be directly used as a final prediction result, and does not depend on artificial experience, so that the network expression capability and robustness are improved.
In a third aspect, the disclosed embodiments provide an information flow recommendation apparatus, which may be deployed in an electronic device, where the electronic device may be any type of device, such as a computer, a server, a wearable device, a mobile terminal, and the like, and the disclosure is not limited thereto.
As shown in fig. 7, in some embodiments, an information flow recommendation apparatus of an example of the present disclosure includes:
an obtaining module 10 configured to obtain user characteristics and information flow characteristics;
the recommending module 20 is configured to input the user characteristics and the information flow characteristics into the information flow recommending network to obtain a recommending result output by the information flow recommending network; the information flow recommendation network is obtained by pre-training based on the multi-target fusion label, and the multi-target fusion label is obtained by processing according to label values of at least two dimensions of sample data.
According to the method and the device, the multi-target fusion label is obtained through the fusion processing of the label values of multiple dimensions based on the sample data, the information flow recommendation network is trained through the multi-target fusion label, a plurality of networks do not need to be constructed and trained, and the network structure is simplified. In addition, as the multi-target fusion label fuses label values of multiple dimensions, the network can learn relevant characteristics of multiple targets and correlation characteristics among the targets in the iterative training process, so that the network training efficiency is improved, and the network prediction effect is improved. In addition, the information flow recommendation network output of the embodiment of the disclosure can be directly used as a final prediction result, and does not depend on artificial experience, so that the network expression capability and robustness are improved.
As shown in fig. 8, in some embodiments, the information flow recommendation apparatus of the disclosed example further includes a network training module 30 and a label determination module 40.
In some embodiments, network training module 30 is configured to:
acquiring a sample data set; the sample data set comprises a plurality of sample data and a multi-target fusion label corresponding to each sample data;
inputting the sample data set into an untrained information flow recommendation network to obtain an output value output by the information flow recommendation network;
and adjusting the network parameters of the information flow recommendation network based on the difference between the output value corresponding to each sample data and the multi-target fusion label until the convergence condition is met, and obtaining the trained information flow recommendation network.
In some embodiments, each sample data comprises tag values in at least two dimensions, the tag determination module 40 being configured to:
obtaining label values of at least two dimensions included in sample data;
and carrying out fusion processing on the label values of at least two dimensions to obtain a multi-target fusion label corresponding to the sample data.
In a fourth aspect, the disclosed embodiments provide a training apparatus for an information flow recommendation network. In some embodiments, the training apparatus comprises a network training module configured to:
the method comprises the steps of obtaining a sample data set, wherein the sample data set comprises a plurality of sample data, and each sample data comprises at least two dimensionality label values;
for each sample data, performing fusion processing based on label values of at least two dimensions of the sample data to obtain a multi-target fusion label corresponding to the sample data;
inputting the sample data set into an untrained information flow recommendation network to obtain an output value output by the information flow recommendation network;
and adjusting the network parameters of the information flow recommendation network based on the difference between the output value corresponding to each sample data and the multi-target fusion label until the convergence condition is met, and obtaining the trained information flow recommendation network.
According to the method and the device, the multi-target fusion label is obtained through the fusion processing of the label values of multiple dimensions based on the sample data, the information flow recommendation network is trained through the multi-target fusion label, a plurality of networks do not need to be constructed and trained, and the network structure is simplified. In addition, as the multi-target fusion label fuses label values of multiple dimensions, the network can learn relevant characteristics of multiple targets and correlation characteristics among the targets in the iterative training process, so that the network training efficiency is improved, and the network prediction effect is improved. In addition, the information flow recommendation network output of the embodiment of the disclosure can be directly used as a final prediction result, and does not depend on artificial experience, so that the network expression capability and robustness are improved.
In a fifth aspect, the present disclosure provides an electronic device, including:
a processor; and
a memory storing computer instructions readable by a processor, the processor performing the method of any of the above embodiments when the computer instructions are read.
In a sixth aspect, the embodiments of the present disclosure provide a storage medium for storing computer-readable instructions for causing a computer to execute the method according to any one of the above embodiments.
Specifically, fig. 9 shows a schematic structural diagram of an electronic device 600 suitable for implementing the method of the present disclosure, and the corresponding functions of the processor and the storage medium can be implemented by the electronic device shown in fig. 9.
As shown in fig. 9, the electronic device 600 includes a processor 601 that can perform various appropriate actions and processes according to a program stored in a memory 602 or a program loaded from a storage section 608 into the memory 602. In the memory 602, various programs and data necessary for the operation of the electronic apparatus 600 are also stored. The processor 601 and the memory 602 are connected to each other by a bus 604. An input/output (I/O) interface 605 is also connected to bus 604.
The following components are connected to the I/O interface 605: an input portion 606 including a keyboard, a mouse, and the like; an output portion 607 including a display such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and the like, and a speaker; a storage section 608 including a hard disk and the like; and a communication section 609 including a network interface card such as a LAN card, a modem, or the like. The communication section 609 performs communication processing via a network such as the internet. The driver 610 is also connected to the I/O interface 605 as needed. A removable medium 611 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 610 as necessary, so that a computer program read out therefrom is mounted in the storage section 608 as necessary.
In particular, the above method processes may be implemented as a computer software program according to embodiments of the present disclosure. For example, embodiments of the present disclosure include a computer program product comprising a computer program tangibly embodied on a machine-readable medium, the computer program comprising program code for performing the above-described method. In such embodiments, the computer program may be downloaded and installed from a network through the communication section 609, and/or installed from the removable medium 611.
The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
It should be understood that the above embodiments are only examples for clearly illustrating the present invention, and are not intended to limit the present invention. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. And obvious variations or modifications of the present disclosure may be made without departing from the scope of the present disclosure.
Claims (10)
1. An information flow recommendation method, comprising:
acquiring user characteristics and information flow characteristics;
inputting the user characteristics and the information flow characteristics into a trained information flow recommendation network to obtain a recommendation result output by the information flow recommendation network; the information flow recommendation network is obtained by pre-training based on a multi-target fusion label, and the multi-target fusion label is obtained by processing according to label values of at least two dimensions of sample data.
2. The method of claim 1, wherein the training process of the information flow recommendation network comprises:
acquiring a sample data set; the sample data set comprises a plurality of sample data and the multi-target fusion label corresponding to each sample data;
inputting the sample data set into an untrained information flow recommendation network to obtain an output value output by the information flow recommendation network;
and adjusting the network parameters of the information flow recommendation network based on the difference between the output value corresponding to each sample data and the multi-target fusion label until a convergence condition is met, and obtaining the trained information flow recommendation network.
3. The method according to claim 1 or 2, wherein each of the sample data includes label values of at least two dimensions, and the process of determining the multi-objective fusion label corresponding to the sample data includes:
obtaining tag values of the at least two dimensions included in the sample data;
and performing fusion processing on the label values of the at least two dimensions to obtain the multi-target fusion label corresponding to the sample data.
4. The method according to claim 3, wherein the fusing the label values of the at least two dimensions to obtain the multi-target fusion label corresponding to the sample data includes:
and performing Gaussian smoothing processing on the label values of the at least two dimensions to obtain the multi-target fusion label.
5. The method of claim 1,
the information stream comprises multimedia information, the tag types corresponding to the tag values comprise a first tag type and a second tag type, the first tag type represents the operation condition of the user on the multimedia information, and the second tag type represents the playing condition of the user on the multimedia information.
6. A training method for an information flow recommendation network is characterized by comprising the following steps:
the method comprises the steps of obtaining a sample data set, wherein the sample data set comprises a plurality of sample data, and each sample data comprises at least two-dimensional label values;
for each sample data, performing fusion processing based on the label values of the at least two dimensions of the sample data to obtain a multi-target fusion label corresponding to the sample data;
inputting the sample data set into an untrained information flow recommendation network to obtain an output value output by the information flow recommendation network;
and adjusting the network parameters of the information flow recommendation network based on the difference between the output value corresponding to each sample data and the multi-target fusion label until a convergence condition is met, and obtaining the trained information flow recommendation network.
7. An information flow recommendation apparatus, comprising:
an acquisition module configured to acquire user characteristics and information flow characteristics;
the recommending module is configured to input the user characteristics and the information flow characteristics into a trained information flow recommending network to obtain a recommending result output by the information flow recommending network; the information flow recommendation network is obtained by pre-training based on a multi-target fusion label, and the multi-target fusion label is obtained by processing according to label values of at least two dimensions of sample data.
8. An apparatus for training an information flow recommendation network, comprising a network training module configured to:
the method comprises the steps of obtaining a sample data set, wherein the sample data set comprises a plurality of sample data, and each sample data comprises at least two-dimensional label values;
for each sample data, performing fusion processing based on the label values of the at least two dimensions of the sample data to obtain a multi-target fusion label corresponding to the sample data;
inputting the sample data set into an untrained information flow recommendation network to obtain an output value output by the information flow recommendation network;
and adjusting the network parameters of the information flow recommendation network based on the difference between the output value corresponding to each sample data and the multi-target fusion label until a convergence condition is met, and obtaining the trained information flow recommendation network.
9. An electronic device, comprising:
a processor; and
a memory storing computer instructions readable by the processor, the processor performing the method of any one of claims 1 to 5 or 6 when the computer instructions are read.
10. A storage medium storing computer readable instructions for causing a computer to perform the method of any one of claims 1 to 5 or 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111064025.4A CN113836406A (en) | 2021-09-10 | 2021-09-10 | Information flow recommendation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111064025.4A CN113836406A (en) | 2021-09-10 | 2021-09-10 | Information flow recommendation method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113836406A true CN113836406A (en) | 2021-12-24 |
Family
ID=78959070
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111064025.4A Pending CN113836406A (en) | 2021-09-10 | 2021-09-10 | Information flow recommendation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113836406A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114435185A (en) * | 2021-12-28 | 2022-05-06 | 深圳云天励飞技术股份有限公司 | New energy automobile electric quantity control method and related equipment |
| WO2024016680A1 (en) * | 2022-07-20 | 2024-01-25 | 百度在线网络技术（北京）有限公司 | Information flow recommendation method and apparatus and computer program product |
-
2021
- 2021-09-10 CN CN202111064025.4A patent/CN113836406A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114435185A (en) * | 2021-12-28 | 2022-05-06 | 深圳云天励飞技术股份有限公司 | New energy automobile electric quantity control method and related equipment |
| CN114435185B (en) * | 2021-12-28 | 2023-08-01 | 深圳云天励飞技术股份有限公司 | New energy automobile electric quantity control method and related equipment |
| WO2024016680A1 (en) * | 2022-07-20 | 2024-01-25 | 百度在线网络技术（北京）有限公司 | Information flow recommendation method and apparatus and computer program product |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110781321B (en) | Multimedia content recommendation method and device | |
| US8756184B2 (en) | Predicting users' attributes based on users' behaviors | |
| CN108776676B (en) | Information recommendation method and device, computer readable medium and electronic device | |
| JP2024503774A (en) | Fusion parameter identification method and device, information recommendation method and device, parameter measurement model training method and device, electronic device, storage medium, and computer program | |
| CN113836406A (en) | Information flow recommendation method and device | |
| CN112100504B (en) | Content recommendation method and device, electronic equipment and storage medium | |
| US11960833B2 (en) | Systems and methods for using machine learning models to organize and select modular components for user interface templates | |
| WO2023231542A1 (en) | Representation information determination method and apparatus, and device and storage medium | |
| CN112182281B (en) | Audio recommendation method, device and storage medium | |
| CN116452263A (en) | Information recommendation method, device, equipment, storage medium and program product | |
| CN114357201A (en) | Audio-visual recommendation method and system based on information perception | |
| CN111680218B (en) | User interest identification method and device, electronic equipment and storage medium | |
| CN113836388A (en) | Information recommendation method and device, server and storage medium | |
| US20230069999A1 (en) | Method and apparatus for updating recommendation model, computer device and storage medium | |
| WO2023087933A1 (en) | Content recommendation method and apparatus, device, storage medium, and program product | |
| CN113051468B (en) | Movie recommendation method and system based on knowledge graph and reinforcement learning | |
| CN112269942B (en) | Method, device and system for recommending object and electronic equipment | |
| CN117150053A (en) | Multimedia information recommendation model training method, recommendation method and device | |
| CN115482021A (en) | Multimedia information recommendation method and device, electronic equipment and storage medium | |
| CN112000888B (en) | Information pushing method, device, server and storage medium | |
| CN114780867B (en) | Recommendation method, medium, device and computing equipment | |
| CN115470397B (en) | Content recommendation method, device, computer equipment and storage medium | |
| US20230088881A1 (en) | Methods and systems for generating alternative content using generative adversarial networks implemented in an application programming interface layer | |
| CN116980471A (en) | Content pushing method, device, apparatus, storage medium and computer program product | |
| CN115795176A (en) | Material sequence generation method, apparatus, device, medium, and program product |
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 |