CN116501949A - Content recommendation method, apparatus and computer readable storage medium - Google Patents
Content recommendation method, apparatus and computer readable storage medium Download PDFInfo
- Publication number
- CN116501949A CN116501949A CN202210050591.8A CN202210050591A CN116501949A CN 116501949 A CN116501949 A CN 116501949A CN 202210050591 A CN202210050591 A CN 202210050591A CN 116501949 A CN116501949 A CN 116501949A
- Authority
- CN
- China
- Prior art keywords
- content
- recommendation
- recommended
- features
- condition
- 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 83
- 238000000605 extraction Methods 0.000 claims abstract description 58
- 238000012216 screening Methods 0.000 claims abstract description 34
- 238000012545 processing Methods 0.000 claims description 25
- 238000004590 computer program Methods 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 6
- 238000005516 engineering process Methods 0.000 abstract description 6
- 238000013473 artificial intelligence Methods 0.000 abstract description 2
- 230000000875 corresponding effect Effects 0.000 description 205
- 238000010586 diagram Methods 0.000 description 17
- 230000006870 function Effects 0.000 description 9
- 239000000284 extract Substances 0.000 description 7
- 230000008569 process Effects 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 6
- 238000012549 training Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 238000000354 decomposition reaction Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 3
- 230000003203 everyday effect Effects 0.000 description 3
- 238000007726 management method Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000002596 correlated effect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 230000001788 irregular Effects 0.000 description 2
- 238000007477 logistic regression Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 230000001276 controlling effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
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
- 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
- G06F16/9538—Presentation of query results
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the application discloses a content recommendation method, a content recommendation device and a computer readable storage medium, which can be applied to various scenes such as cloud technology, artificial intelligence, intelligent traffic, auxiliary driving and the like; obtaining content characteristics of multiple dimensions by obtaining at least one content to be recommended and carrying out multidimensional characteristic extraction on the content to be recommended; fusing the content characteristics, and determining the recommendation weight of each recommendation condition in a preset content recommendation condition set according to the fused content characteristics and the preset content recommendation condition set; extracting at least one target content characteristic from the fused content characteristics according to the recommended condition; weighting the target content features based on the recommendation weight to obtain recommendation features corresponding to each recommendation condition; and screening target recommended content from the content to be recommended according to the recommended characteristics, and recommending the target recommended content. Therefore, the accuracy of content recommendation is improved, and the content recommendation efficiency is further improved.
Description
Technical Field
The present application relates to the field of internet technologies, and in particular, to a content recommendation method, apparatus, and computer readable storage medium.
Background
In recent years, with the rapid development of internet technology, mass contents, such as information contents, are produced every day, and the quality of these contents is also irregular. In a large amount of content, it is difficult for a user to acquire necessary information. Therefore, the quality of the content can be estimated by the recommendation system to preferentially expose the high-quality content, so that the experience of the user for acquiring the content is improved. The existing content recommendation method is used for learning a recommendation model based on the estimation of multiple recommendation conditions, so that content recommendation is performed based on the recommendation model.
In the research and practice process of the prior art, the inventor discovers that in the existing method for recommending the content through the prediction of multiple recommendation conditions, parameters among different recommendation conditions can generate conflict, so that the accuracy of the recommended high-quality content is poor, and the content recommendation efficiency is low.
Disclosure of Invention
The embodiment of the application provides a content recommendation method, a content recommendation device and a computer readable storage medium, which can improve the accuracy of content recommendation and further improve the content recommendation efficiency.
The embodiment of the application provides a content recommendation method, which comprises the following steps:
Acquiring at least one content to be recommended, and carrying out multidimensional feature extraction on the content to be recommended to obtain content features with multiple dimensions;
fusing the content features, and determining the recommendation weight of each recommendation condition in a preset content recommendation condition set according to the fused content features and the preset content recommendation condition set;
extracting at least one target content feature from the fused content features according to the recommended condition;
weighting the target content features based on the recommendation weights to obtain recommendation features corresponding to each recommendation condition;
and screening target recommended content from the content to be recommended according to the recommended characteristics, and recommending the target recommended content.
Accordingly, an embodiment of the present application provides a content recommendation device, including:
the acquisition unit is used for acquiring at least one content to be recommended, and carrying out multidimensional feature extraction on the content to be recommended to obtain content features with multiple dimensions;
the determining unit is used for fusing the content characteristics and determining the recommendation weight of each recommendation condition in the preset content recommendation condition set according to the fused content characteristics and the preset content recommendation condition set;
The extraction unit is used for extracting at least one target content characteristic from the fused content characteristics according to the recommended conditions;
the weighting unit is used for weighting the target content characteristics based on the recommendation weight to obtain recommendation characteristics corresponding to each recommendation condition;
and the recommending unit is used for screening out target recommended content from the content to be recommended according to the recommending characteristics and recommending the target recommended content.
In an embodiment, the determining unit includes:
the weight network screening subunit is used for screening out a weight network corresponding to each recommendation condition in the preset content recommendation condition set from the weight network set of the trained content recommendation model based on the preset content recommendation condition set;
and the recommendation weight calculating subunit is used for calculating the recommendation weight of each recommendation condition in the preset content recommendation condition set according to the weight network corresponding to each recommendation condition and the fused content characteristics.
In an embodiment, the recommendation weight calculation subunit includes:
the processing module is used for carrying out normalization processing on the fused content characteristics based on the weight network to obtain normalized content characteristics;
And the classification module is used for classifying the normalized content characteristics to obtain the recommendation weight corresponding to each recommendation condition in the preset content recommendation condition set.
In an embodiment, the recommendation unit includes:
the recommendation characteristic classification subunit is used for classifying the recommendation characteristics to obtain recommendation levels corresponding to the contents to be recommended under each recommendation condition;
a recommendation level fusion subunit, configured to fuse the recommendation levels to obtain a recommendation coefficient of the content to be recommended;
and the screening subunit is used for screening out target recommended content from the content to be recommended according to the recommendation coefficient and recommending the target recommended content.
In one embodiment, the screening subunit comprises:
the ordering module is used for ordering the content to be recommended based on the recommendation coefficient to obtain ordering information of the content to be recommended;
and the screening module is used for screening out target recommended content from the content to be recommended based on the sorting information.
In one embodiment, the screening subunit comprises:
the storage module is used for storing the target recommended content into a corresponding recall content pool;
The recommended content determining module is used for determining at least one recommended content in target recommended content of the recall content pool based on the content recommendation request when receiving the content recommendation request sent by the terminal;
and the pushing module is used for pushing the recommended content to the terminal.
In an embodiment, the acquiring unit includes:
the semantic and attribute feature extraction subunit is used for extracting attribute features of the content to be recommended to obtain content attribute features corresponding to the content to be recommended, and extracting semantic features of the content to be recommended to obtain content semantic features corresponding to the content to be recommended;
and the multidimensional feature extraction subunit is used for carrying out multidimensional feature extraction on the content attribute features and the content semantic features corresponding to the content to be recommended to obtain content features with multiple dimensions.
In an embodiment, the multi-dimensional feature extraction subunit comprises:
the full-connection processing module is used for carrying out full-connection processing on the content attribute characteristics and the content semantic characteristics corresponding to the content to be recommended to obtain content characteristics of a first dimension;
the cross feature extraction module is used for carrying out cross feature extraction on the content attribute features and the content semantic features corresponding to the content to be recommended to obtain content features of a second dimension;
And the assignment module is used for taking the content characteristics of the first dimension and the content characteristics of the second dimension as the content characteristics of a plurality of dimensions.
In addition, the embodiment of the application further provides a computer readable storage medium, wherein the computer readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute the steps in any content recommendation method provided by the embodiment of the application.
In addition, the embodiment of the application also provides a computer device, which comprises a processor and a memory, wherein the memory stores an application program, and the processor is used for running the application program in the memory to realize the content recommendation method provided by the embodiment of the application.
Embodiments of the present application also provide a computer program product or computer program comprising computer instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the steps in the content recommendation method provided in the embodiment of the present application.
According to the embodiment of the application, the content characteristics of multiple dimensions are obtained by acquiring at least one content to be recommended and carrying out multidimensional characteristic extraction on the content to be recommended; fusing the content characteristics, and determining the recommendation weight of each recommendation condition in a preset content recommendation condition set according to the fused content characteristics and the preset content recommendation condition set; extracting at least one target content feature from the fused content features according to the recommended condition; weighting the target content features based on the recommendation weight to obtain recommendation features corresponding to each recommendation condition; and screening target recommended content from the content to be recommended according to the recommended characteristics, and recommending the target recommended content. In this way, the recommendation weight of each recommendation condition in the preset content recommendation condition set is determined by acquiring the content characteristics of a plurality of dimensions corresponding to the content to be recommended, the correlation between each recommendation condition is further considered, the target content characteristics are extracted from the fused content characteristics, and the recommendation characteristics corresponding to each recommendation condition are obtained by weighting the target content characteristics according to the recommendation weights based on the correlation between the recommendation conditions, so that the target recommendation content is screened out from the content to be recommended according to the recommendation characteristics, the accuracy of content recommendation is improved, and the content recommendation efficiency is further improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the description of the embodiments will be briefly introduced below, it being obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic diagram of an implementation scenario of a content recommendation method according to an embodiment of the present application;
fig. 2 is a flow chart of a content recommendation method according to an embodiment of the present application;
fig. 3a is a schematic structural diagram of a content recommendation method according to an embodiment of the present application;
FIG. 3b is a schematic diagram illustrating a method for recommending content according to an embodiment of the present disclosure;
FIG. 3c is a schematic diagram of a model training process of a content recommendation method according to an embodiment of the present application;
FIG. 3d is a schematic diagram of a model application flow of a content recommendation method according to an embodiment of the present application;
fig. 4 is an effect schematic diagram of a content recommendation method according to an embodiment of the present application;
FIG. 5 is a schematic diagram showing another effect of a content recommendation method according to an embodiment of the present application;
FIG. 6 is another flow chart of a content recommendation method according to an embodiment of the present disclosure;
fig. 7 is a schematic structural diagram of a content recommendation device provided in an embodiment of the present application;
fig. 8 is a schematic structural diagram of a computer device according to an embodiment of the present application.
Detailed Description
The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all, of the embodiments of the present application. All other embodiments, which can be made by those skilled in the art based on the embodiments herein without making any inventive effort, are intended to be within the scope of the present application.
The embodiment of the application provides a content recommendation method, a content recommendation device and a computer readable storage medium. The content recommendation device may be integrated in a computer device, which may be a server or a terminal.
Referring to fig. 1, taking an example that a content recommendation apparatus is integrated in a computer device, fig. 1 is a schematic diagram of an implementation scenario of a content recommendation method provided in an embodiment of the present application, where the implementation scenario includes a server a and a terminal B, where the server a may be an independent physical server, may be a server cluster or a distributed system formed by a plurality of physical servers, or may be a cloud server that provides basic cloud computing services such as a cloud service, a cloud database, cloud computing, a cloud function, cloud storage, a network service, cloud communication, a middleware service, a domain name service, a security service, a network acceleration service (Content Delivery Network, CDN), and a big data and artificial intelligent platform. The server A can acquire at least one content to be recommended, and performs multidimensional feature extraction on the content to be recommended to obtain content features with multiple dimensions; fusing the content characteristics, and determining the recommendation weight of each recommendation condition in a preset content recommendation condition set according to the fused content characteristics and the preset content recommendation condition set; extracting at least one target content feature from the fused content features according to the recommended condition; weighting the target content features based on the recommendation weight to obtain recommendation features corresponding to each recommendation condition; and screening target recommended content from the content to be recommended according to the recommended characteristics, and recommending the target recommended content.
Terminal B may include, but is not limited to, various computer devices such as cell phones, computers, intelligent voice interaction devices, intelligent home appliances, car terminals, etc.
The terminal B and the server a may be directly or indirectly connected through a wired or wireless communication manner, and the server a may acquire data uploaded by the terminal B to perform a corresponding content recommendation operation, which is not limited herein.
It should be noted that, the embodiment of the present application may be applied to various scenarios, including, but not limited to, cloud technology, artificial intelligence, intelligent traffic, driving assistance, etc., the schematic view of the implementation environment of the content recommendation method shown in fig. 1 is only an example, and the implementation environment of the content recommendation method described in the embodiment of the present application is for more clearly describing the technical solution of the embodiment of the present application, and does not constitute a limitation to the technical solution provided by the embodiment of the present application. As can be appreciated by those skilled in the art, with the evolution of content recommendation and the appearance of new service scenarios, the technical solution provided in the present application is also applicable to similar technical problems.
The following will describe in detail. The following description of the embodiments is not intended to limit the preferred embodiments.
The present embodiment will be described from the point of view of a content recommendation apparatus, which may be integrated in a computer device, which may be a server, and the present application is not limited thereto.
Referring to fig. 2, fig. 2 is a flowchart illustrating a content recommendation method according to an embodiment of the present application. The content recommendation method comprises the following steps:
101. and acquiring at least one content to be recommended, and carrying out multidimensional feature extraction on the content to be recommended to obtain content features with multiple dimensions.
The content to be recommended can be content to be recommended, and the content can be a carrier carrying information and can be displayed to a user for consumption, so that the user can acquire corresponding information. The content may include, but is not limited to, content in the form of video, audio, text, and images. For example, the content may be information content.
The content features can be feature information of multiple dimensions in the content to be recommended, for example, discrete feature information of multiple dimensions in semantics, attributes and the like in the content to be recommended, and the discrete feature information can be subjected to feature extraction of multiple dimensions for multiple times on the basis of the multi-dimensional discrete feature information, so that the obtained feature information and the like can be used for more accurately representing the content to be recommended.
The method for obtaining the at least one content to be recommended may be various, for example, the at least one content to be recommended may be obtained through various internet platforms, or the content to be recommended may be obtained from a storage device.
After at least one content to be recommended is obtained, multidimensional feature extraction can be performed on the content to be recommended, and various modes of multidimensional feature extraction can be performed on the content to be recommended, for example, attribute feature extraction is performed on the content to be recommended to obtain content attribute features corresponding to the content to be recommended, semantic feature extraction is performed on the content to be recommended to obtain content semantic features corresponding to the content to be recommended, and multidimensional feature extraction is performed on the content attribute features and the content semantic features corresponding to the content to be recommended to obtain content features of multiple dimensions.
The content attribute feature may be feature information characterizing an attribute of the content to be recommended in the content to be recommended, where the attribute may be information such as a type, a number of pictures, an author, a release time, etc. of the content to be recommended, the content semantic feature may be semantic feature information characterizing a text in the content to be recommended, for example, taking the content to be recommended as an information content, the content attribute feature may include discrete features in dimensions such as an information type, a number of pictures, a length of pictures, an author, etc., and the content semantic feature may include semantic features in dimensions such as an information title, an information text, an information tag (tag), etc., where the content semantic feature may be obtained by feature extraction through a language model (such as word2vec, glove, BERT, etc.).
Based on the content attribute features and the content semantic features, there may be multiple ways of extracting multidimensional features from the content attribute features and the content semantic features corresponding to the content to be recommended, for example, full connection processing may be performed on the content attribute features and the content semantic features corresponding to the content to be recommended to obtain content features of a first dimension, cross feature extraction is performed on the content attribute features and the content semantic features corresponding to the content to be recommended to obtain content features of a second dimension, and the content features of the first dimension and the content features of the second dimension are used as content features of multiple dimensions.
The content features of the first dimension can be feature information obtained by extracting the content attribute features and the content semantic features through a full connection layer (Fully connected layer), and the content features of the first dimension can be feature information of a lower dimension; the content features of the second dimension may be feature information extracted by feature intersection of the content attribute features and the content semantic features, and the content features of the second dimension may be feature information of a higher dimension.
For example, referring to fig. 3a, fig. 3a is a schematic structural diagram of a content recommendation method provided in the embodiment of the present application, content semantic features corresponding to content to be recommended may be extracted from content to be recommended, content attribute features corresponding to content to be recommended may be extracted from content to be recommended, including dimensional information such as a title, a label, etc., and content attribute features corresponding to content to be recommended may be extracted from content to be recommended, including dimensional information such as a type, a number of pictures, an author, a release time, a length of a picture, a duration of a video, etc., so that the content attribute features corresponding to the content to be recommended and the content semantic features may be fully connected to obtain content features of a first dimension, cross feature extraction is performed on the content attribute features corresponding to the content to be recommended and the content semantic features of the content to be recommended, so as to obtain content features of a second dimension, and the content features of the first dimension and the content features of the second dimension are used as content features of a plurality of dimensions.
The method for performing the full connection processing on the content attribute feature and the content semantic feature corresponding to the content to be recommended may be various, for example, the full connection processing may be performed on the content attribute feature and the content semantic feature corresponding to the content to be recommended through a neural network. Correspondingly, there may be various ways of performing cross feature extraction on the content attribute features and the content semantic features corresponding to the content to be recommended, for example, a FM model (Factor Machine), a Product-based neural network (PNN) or other models may be used to perform cross feature extraction on the content attribute features and the content semantic features.
Optionally, depth factor decomposition (Deep FM model) may be used to extract multidimensional features of content attribute features and content semantic features corresponding to the content to be recommended, where the Deep FM model may include a Deep (Deep) portion and an FM model, where the Deep portion may be a neural network structure and may be composed of 2 full-connection layers to extract low-level dimensional features in the features, the FM model is a machine learning algorithm based on matrix decomposition, the features may be intersected, and the FM model has linear time complexity in terms of feature intersection, so that the FM model may extract features with higher dimensions of vectors. Therefore, the Deep part in the Deep FM model can be used for carrying out full connection processing on the content attribute features and the content semantic features corresponding to the content to be recommended to obtain the content features of the first dimension, and the FM model in the Deep FM model is used for carrying out cross feature extraction on the content attribute features and the content semantic features corresponding to the content to be recommended to obtain the content features of the second dimension.
102. Fusing the content features, and determining the recommendation weight of each recommendation condition in the preset content recommendation condition set according to the fused content features and the preset content recommendation condition set.
The fused content features may be feature information obtained by fusing the content features, the preset content recommendation condition set may be an entirety formed by at least one preset recommendation condition, the recommendation condition may be a condition for measuring quality of the content to be recommended, for example, a Click-Through-Rate (CTR) condition, a person average reading time length and the like measure quality of the content to be recommended, and the recommendation weight may be a weight corresponding to each recommendation condition.
The method of fusing the content features may be multiple, for example, the content features of multiple dimensions may be spliced to obtain a fused content feature, for example, please refer to fig. 3a, the content features of multiple dimensions may be input into a shared parameter layer to be spliced to obtain a fused content feature, a weight value may be set for the content feature of each dimension, the content feature may be weighted according to the weight value corresponding to the content feature of each dimension, and the fused content feature may be obtained according to the weighted result.
After the content features are fused, the recommendation weight of each recommendation condition in the preset content recommendation condition set can be determined according to the fused content features and the preset content recommendation condition set. With the rapid development of internet technology, mass contents, such as information contents, are produced every day, and the quality of the contents is also irregular. In a large amount of content, it is difficult for a user to acquire necessary information. Therefore, the quality of the content can be estimated by the recommendation system to preferentially expose the high-quality content, so that the experience of the user for acquiring the content is improved. In the existing content recommendation method, a recommendation model is often learned based on a single recommendation condition (i.e. single task) or a prediction of multiple recommendation conditions (i.e. multiple tasks), so that content recommendation is performed based on the recommendation model, please refer to fig. 3b, and fig. 3b is a method comparison schematic diagram of a content recommendation method provided in the embodiment of the present application. The method for predicting the single task only can perform model learning for a certain index (such as click rate) at a time, so that supervision information of other indexes is lost, and how many tasks need to be predicted, how many depth network models need to be deployed, for example, if the models need to predict task 1 and task 2, two corresponding depth network models need to be deployed, so that architecture layout of the models is greatly increased. Therefore, the embodiment of the application provides a content recommendation method, by determining the recommendation weight of each recommendation condition in the preset content recommendation condition set, the relevance among each recommendation condition is considered, so that high-quality content is screened out of a plurality of to-be-recommended contents to be recommended for recommendation, the accuracy of the recommended high-quality content is improved, and the content recommendation efficiency is further improved.
The method for determining the recommendation weight of each recommendation condition in the preset content recommendation condition set can be multiple according to the fused content characteristics and the preset content recommendation condition set, for example, the recommendation weight of each recommendation condition in the preset content recommendation condition set can be calculated according to the weight network corresponding to each recommendation condition and the fused content characteristics by screening the weight network corresponding to each recommendation condition in the preset content recommendation condition set from the weight network set of the trained content recommendation model based on the preset content recommendation condition set.
The trained content recommendation model may be a trained model for content recommendation, and the set of weight networks may be an entirety made up of a plurality of weight networks in the trained content recommendation model, where the weight networks may be used to calculate recommendation weights for each recommendation condition. Each recommendation condition may correspond to a weighting network.
For example, please refer to fig. 3c, fig. 3c is a schematic diagram of a model training flow of a content recommendation method according to an embodiment of the present application, wherein the content to be recommended is taken as information, the recommendation condition includes two tasks of click rate and average person reading duration, and specifically, a large amount of original information corpus (for example, 100 ten thousand pieces) with click rate and average person reading duration characteristics can be obtained as a data set for training a preset content recommendation model. There are two kinds of label (label) values for each information, including click rate and average reading time. The preset content recommendation model predicts the click rate and the average reading time of the information. To improve accuracy of model predictions, confidence levels may be introduced, and correspondingly, the click rate recommendation conditions may use Wilson click rate (Wilson CTR):
Where p is the click rate, n is the total number of samples, i.e., the number of exposures, z represents the distribution ratio in a normal distribution, the value of z is related to the confidence introduced, for example, z takes 1.96, and there is a 95% confidence.
And the average reading time length can be expressed as
These data sets may be further segmented, for example, the data sets may be segmented into training sets, validation sets, and test sets, and the sample ratios may be 80%, 10%, and 10%, respectively. Thus, a big data technology such as a data warehouse tool (hive) can be adopted to perform feature stitching on each piece of information of the data set to construct a feature sample, and then the feature sample is transferred to a specific file format (for example, tfrecord), wherein tfrecord is a file format suitable for a deep learning framework of a symbol mathematical system (tensorsurface). Alternatively, in the preset content recommendation model, the deep fm model may be used to perform multidimensional feature extraction on the information to obtain information content features with multiple dimensions, for example, the information content features may include content features with dimensions of information type, information title, information text, number of pictures, release duration, author, region, and the like.
Thus, training may be performed in a preset content recommendation model based on samples, for example, the data set may be recycled 10 times (epochs=10), and the number of batches (batch) may be 128. When the preset content recommendation model converges, a trained content recommendation model can be obtained, and further training results can be stored as model files, so that migration application of the trained content recommendation model can be realized.
After obtaining the trained content recommendation model, the trained content recommendation model may be applied, for example, please refer to fig. 3d, fig. 3d is a schematic diagram of a model application flow chart of a content recommendation method provided in an embodiment of the present application, wherein information newly added every day may be obtained, and further, a feature sample is constructed for the newly added information to construct a data set to be estimated, and the data set is converted into a tfrecord format. And then a pre-stored model file can be loaded to traverse the information data set to be estimated through the trained content recommendation model, so that the prediction value of each piece of information is obtained, the information can be screened according to the prediction value of each piece of information, for example, the information with the prediction value being in the first N names can be screened out, the screened information is written into a corresponding recall pool, and when an online recall request is received, the information can be issued to a corresponding terminal for display. The following is a specific description.
After the trained content recommendation model is obtained, a weight network corresponding to each recommendation condition in the preset content recommendation condition set can be screened out from the weight network set of the trained content recommendation model based on the preset content recommendation condition set.
After screening the weight network corresponding to each recommendation condition in the preset content recommendation condition set, calculating the recommendation weight of each recommendation condition in the preset content recommendation condition set according to the weight network corresponding to each recommendation condition and the fused content characteristics, wherein the recommendation weight of each recommendation condition in the preset content recommendation condition set can be calculated in various manners according to the weight network corresponding to each recommendation condition and the fused content characteristics, for example, normalization processing can be performed on the fused content characteristics based on the weight network to obtain normalized content characteristics, and the normalized content characteristics are classified to obtain the recommendation weight corresponding to each recommendation condition in the preset content recommendation condition set.
The normalized content features may be feature information obtained by normalizing the fused content features, and there may be various ways to normalize the fused content features, for example, please continue to refer to fig. 3a, and the normalized content features may be obtained by normalizing the fused content features using a full connection layer.
After the normalized content features are obtained, the normalized content features may be classified, where there may be multiple ways to classify the normalized content features, for example, please refer to fig. 3a, in which a logistic regression layer (Softmax layer) may be used to classify the normalized content features.
Optionally, a Multi-gate mix-of-expertise (MMOE) model may be used to screen out a weight network corresponding to each recommendation condition in the preset content recommendation condition set from a weight network set of the trained content recommendation model based on the preset content recommendation condition set, and calculate a recommendation weight of each recommendation condition in the preset content recommendation condition set according to the weight network corresponding to each recommendation condition and the fused content characteristics. Specifically, the MMOE model may include a plurality of weight networks (Gate) and a plurality of Expert networks (Expert), each of the weight networks may be composed of a full-connection layer and a Softmax layer, each recommended condition corresponds to one weight network, the number of Expert networks and the number of parameters may be set according to the actual situation, the Expert networks are generally full-connection networks, and the Expert networks may learn important features from the shared parameter layer. Each weight network learns to assign recommendation weights to the expert networks according to learning tasks corresponding to recommendation conditions required to be responsible by the weight network.
The weight network corresponding to each recommendation condition in the preset content recommendation condition set can be screened out from the weight network set of the MMOE model, so that the recommendation sub-weight of each recommendation condition corresponding to each expert network can be calculated according to the weight network corresponding to each recommendation condition and the fused content characteristics, namely, the input fused content characteristics are mapped into the dimension corresponding to each expert network through the weight network through linear transformation, the Softmax is calculated to obtain the recommendation sub-weight of each expert network, and the recommendation weight of each recommendation condition can be obtained according to the recommendation sub-weight corresponding to each recommendation condition.
For example, referring to fig. 3a, assuming that the preset content recommendation condition set includes recommendation condition 1 and recommendation condition 2, and meanwhile, assuming that the number of expert networks is 3, a weight network corresponding to each recommendation condition in the preset content recommendation condition set, that is, a weight network 1 corresponding to the recommendation condition 1 and a weight network 2 corresponding to the recommendation condition 2, may be screened out in the weight network set, so that a recommendation sub-weight corresponding to each expert network for the recommendation condition 1 may be calculated according to the weight network 1 and the fused content characteristics, a recommendation sub-weight corresponding to each expert network for the recommendation condition 2 may be calculated according to the weight network 2 and the fused content characteristics, so that a recommendation weight corresponding to each expert network for the recommendation condition 1 and a recommendation sub-weight corresponding to the recommendation condition 2 may be obtained according to the recommendation condition 1 and the recommendation sub-weight corresponding to each expert network for the recommendation condition 2.
For example, referring to fig. 3a, assuming that the preset content recommendation condition set includes recommendation condition 1 and recommendation condition 2, and assuming that the number of Expert networks is 3, namely, expert1, expert2, and Expert3, the server may screen out a weight network corresponding to each recommendation condition in the preset content recommendation condition set in the weight network set, that is, recommendation condition 1 corresponds to weight network 1, recommendation condition 2 corresponds to weight network 2, so that the recommended sub-weight of each Expert network corresponding to recommendation condition 1 may be calculated according to weight network 1 and the fused content characteristics, the recommended sub-weight of each Expert network corresponding to recommendation condition 2 may be calculated according to weight network 2 and the fused content characteristics, therefore, the recommended weight of each recommended condition can be obtained according to the recommended sub-weight corresponding to each recommended condition, for example, the recommended sub-weights of the corresponding recommended condition in each Expert network are calculated based on the weight network, the recommended sub-weights of the recommended condition 1 in the Expert networks Expert1, expert2 and Expert3 can be obtained as a, b and c, the recommended sub-weights of the recommended condition 2 in the Expert networks Expert1, expert2 and Expert3 are d, e and f, and the recommended weights of the recommended condition 1 in each Expert network can be obtained according to the recommended sub-weights corresponding to each recommended condition, namely a, b and c, and the recommended weights of the recommended condition 2 in each Expert network are d, e and f.
103. And extracting at least one target content characteristic from the fused content characteristics according to the recommended condition.
The target content features may be feature information extracted from the fused content features, and there may be various ways of extracting at least one target content feature from the fused content features, for example, the number of expert networks may be determined and set according to recommended conditions, and then at least one target content feature may be extracted from the fused content features through the set at least one expert network.
The number of the expert networks may be determined according to the recommended conditions, for example, the number of the expert networks may be determined according to the number of the recommended conditions, and when the number of the recommended conditions is large, the number of the expert networks may be increased to increase the number of the parameters.
After the number of the expert networks is determined and set, at least one target content feature can be extracted from the fused content features through the set at least one expert network, and a plurality of modes for extracting at least one target content feature from the fused content features through the set at least one expert network can be adopted, for example, a preset content recommendation method can be trained by adopting the number of the expert networks to be set to obtain a trained content recommendation model, so that the fused content features can be input into the expert networks in the trained content recommendation model to be subjected to full connection processing, and the target content features corresponding to each expert network can be obtained.
It should be noted that, the target content feature may be acquired after the recommended weight of each recommended condition is acquired, the recommended weight of each recommended condition may be acquired after the target content feature is acquired, the target content feature and the recommended weight of each recommended condition may be acquired at the same time, and the sequence of the two steps may be determined according to the actual situation, which is not limited herein.
104. And weighting the target content features based on the recommendation weight to obtain recommendation features corresponding to each recommendation condition.
The recommendation characteristic may be characteristic information corresponding to each to-be-promoted condition in the fused content characteristic, and is used for determining a result of the to-be-promoted content in each to-be-promoted condition, for example, assuming that a preset recommendation condition set includes a recommendation condition corresponding to a click rate and a recommendation condition corresponding to a per-person reading duration, the recommendation characteristic is used for estimating the click rate and the per-person reading duration corresponding to the to-be-promoted content, so that the degree of coincidence between the to-be-promoted content and each to-be-promoted condition can be better represented according to the recommendation characteristic corresponding to each to-be-promoted condition.
The manner of weighting the target content features based on the recommendation weights to obtain the recommendation features corresponding to each recommendation condition may be various, for example, each target content feature may be weighted according to the recommendation weight corresponding to each recommendation condition, and the weighted result may be averaged to obtain the recommendation feature corresponding to each recommendation condition, for example, it may be assumed that the preset content recommendation condition set includes recommendation condition 1 and recommendation condition 2, and meanwhile, it is assumed that the target content feature includes G, H and I, the recommendation weights corresponding to recommendation condition 1 in the target content features G, H and I are j1, j2 and j3, and the recommendation weights corresponding to recommendation condition 2 in the target content features G, H and I are k1, k2 and k3, respectively. Then a weighted average may be performed on each target content feature based on the recommendation weight, and a recommendation feature corresponding to recommendation condition 1 may be obtained as (j1×g+j2×h+j3×i)/(j1+j2+j3), and a recommendation feature corresponding to recommendation condition 2 may be obtained as (k1×g+k2×h+k3×i)/(k1+k2+k3).
In addition, each target content feature may be weighted according to a recommendation weight corresponding to each recommendation condition, and the weighted results may be summed to obtain a recommendation feature corresponding to each recommendation condition, for example, it may be assumed that a preset content recommendation condition set includes recommendation condition 1 and recommendation condition 2, and meanwhile, it is assumed that the target content features include G, H and I, recommendation weights corresponding to recommendation condition 1 in target content features G, H and I are j1, j2 and j3, and recommendation weights corresponding to recommendation condition 2 in target content features G, H and I are k1, k2 and k3, respectively. Then, each target content feature may be weighted and summed based on the recommendation weight, so that a recommendation feature corresponding to the recommendation condition 1 is j1×g+j2×h+j3×i, and a recommendation feature corresponding to the recommendation condition 2 is k1×g+k2×h+k3×i.
105. And screening target recommended content from the content to be recommended according to the recommended characteristics, and recommending the target recommended content.
The target recommended content can be high-quality content screened out from the content to be recommended, namely, the target recommended content is recommended, and a better recommending effect can be generated.
The method includes the steps of selecting target recommended content from the to-be-recommended content according to the recommended characteristics, and recommending the target recommended content, for example, classifying the recommended characteristics to obtain a recommended level of the to-be-recommended content corresponding to each recommended condition, fusing the recommended levels to obtain a recommendation coefficient of the to-be-recommended content, selecting the target recommended content from the to-be-recommended content according to the recommendation coefficient, and recommending the target recommended content.
The recommendation level may be an estimated result of representing the content to be recommended in each recommendation condition, for example, the recommendation level may be divided into a plurality of level intervals, and each content to be recommended is divided into corresponding levels according to recommendation characteristics, where a higher recommendation level may represent a higher quality of the content to be recommended in the current recommendation condition, that is, a lower recommendation level may represent a worse quality of the content to be recommended in the current recommendation condition, that is, a lower recommendation level may represent a lower quality of the content to be recommended. In addition, the recommendation degree of the content to be recommended is measured in a level interval mode, so that the influence of an abnormal sample can be reduced, and the accuracy of content recommendation is further improved.
The recommendation coefficient may represent the recommendation degree corresponding to the content to be recommended based on the estimated result of each recommendation condition, and the higher the recommendation coefficient, the higher the recommendation degree of the content to be recommended, that is, the better the quality of the content to be recommended, and the lower the recommendation coefficient, the lower the recommendation degree of the content to be recommended, that is, the worse the quality of the content to be recommended. The means for fusing the recommendation levels to obtain the recommendation coefficient of the content to be recommended may be multiple, for example, a numerical value may be given to each recommendation level, and a weight may be given to each recommendation condition, where each recommendation weight may be determined according to the importance degree of each recommendation condition, for example, if the recommendation condition corresponding to the average person reading time length is important, a greater weight may be given to the recommendation condition corresponding to the average person reading time length, so that the result is biased to the more important recommendation condition, and if the recommendation condition corresponding to the click rate is important, a greater weight may be given to the recommendation condition corresponding to the click rate. Specifically, it may be assumed that there are a total of 8 recommendation levels, the values corresponding to the levels 1 to 8 are 1 to 8, respectively, the level 8 indicates that the recommendation degree of the content to be recommended is highest, meanwhile, the preset content recommendation condition set includes recommendation condition 1 and recommendation condition 2, the weight corresponding to the recommendation condition 1 is 0.6, the weight corresponding to the recommendation condition 2 may be 0.4, the recommendation level corresponding to the recommendation condition 1 is 6, the recommendation level corresponding to the recommendation condition 2 is 4, the recommendation level corresponding to the recommendation condition 2 is 5, the recommendation coefficient corresponding to the recommendation condition 2 is 6, the recommendation coefficient corresponding to the recommendation condition 1 may be 0.6x6+0.4x4=5.2, and the recommendation coefficient corresponding to the recommendation condition 2 may be 0.6x5+0.4x6=5.4, and the recommendation coefficient of the content to be 2 is higher than the recommendation content 1, that is, the recommendation coefficient of the content to be 2 is higher than the content to be recommended 1.
Optionally, the recommended level may be fused by using a super parameter to obtain a recommended coefficient of the content to be recommended, for example, assuming that the super parameter α, a preset recommended condition set includes a recommended condition corresponding to a click rate and a recommended condition corresponding to a person average reading duration, and the recommended level of the content to be recommended corresponding to the click rate condition is. The recommendation coefficient can be expressed as
Recommendation coefficient=α+click rate recommendation level + (1- α) ×average reading duration recommendation level
After the recommendation coefficient of the content to be recommended is obtained, the target recommended content can be screened out of the content to be recommended according to the recommendation coefficient, and according to the recommendation coefficient, there are various modes for screening out the target recommended content in the content to be recommended, for example, the content to be recommended can be ranked based on the recommendation coefficient, ranking information of the content to be recommended is obtained, and based on the ranking information, the target recommended content is screened out of the content to be recommended.
The ranking information may be a ranking result representing each content to be recommended, for example, may be ranking information corresponding to each content to be recommended, according to the ranking information, a target recommended content may be screened out of the content to be recommended, for example, the content to be recommended ranked in the top N bits may be determined as the target recommended content.
After the target recommended content is selected from the contents to be recommended, the target recommended content can be recommended, and various modes can be adopted for recommending the target recommended content, for example, the target recommended content can be stored in a corresponding recall content pool, when a content recommendation request sent by a terminal is received, at least one recommended content is determined in the target recommended content in the recall content pool based on the content recommendation request, and the recommended content is pushed to the terminal.
The recall content pool may be used to store the content to be recalled, that is, store the target recommended content, where the recommended content may be the content to be distributed to the corresponding terminal. When a content recommendation request sent by a terminal is received, at least one recommendation content can be determined in target recommendation contents of the recall content pool according to the content recommendation request, and then the recommendation content can be pushed to the terminal for display.
In addition, the content recommendation method provided by the embodiment of the application has the beneficial effects in practical application, wherein the content to be recommended is taken as information content, the recommendation condition comprises two conditions of click rate and average person reading time length, the recommendation level is divided into 8 gears, wherein the gear 8 represents the highest quality of the information content, and after the content recommendation method provided by the embodiment of the application is applied, the click rate of high-quality estimated information recall is improved by 17% compared with the heat gate information recall. Specifically, referring to fig. 4 for the click rate recommendation condition, fig. 4 is a schematic effect diagram of a content recommendation method according to an embodiment of the present application, where the abscissa is a click rate recommendation level predicted based on the content recommendation method provided in the present application before information content is online, and the ordinate is an actual click rate after information content of different levels is online. As can be seen from the figure, the click rate offline prediction recommendation level is positively correlated with the online actual click rate, that is, the higher the offline prediction recommendation level is, the higher the click rate after online is, which illustrates the quality of effectively estimated information content of the content recommendation method provided by the embodiment of the application.
For the recommended condition of the average person reading duration, please refer to fig. 5, fig. 5 is another schematic effect of a content recommendation method provided in the embodiment of the present application, in which the abscissa is a recommended level of the average person reading duration predicted based on the content recommendation method provided in the present application before the information content is online, and the ordinate is an actual average person reading duration after the information content of different levels is online, as known from the figure, for the average person reading duration, the offline predicted recommended level is positively correlated with the online actual average person reading duration, that is, the higher the offline predicted recommended level is, the longer the online average person reading duration is, which illustrates the quality of the information content effectively predicted by the content recommendation method provided in the embodiment of the present application, and the accuracy of content recommendation is improved, thereby improving the content recommendation efficiency.
As can be seen from the above, in the embodiment of the present application, by acquiring at least one content to be recommended, and performing multidimensional feature extraction on the content to be recommended, content features of multiple dimensions are obtained; fusing the content characteristics, and determining the recommendation weight of each recommendation condition in a preset content recommendation condition set according to the fused content characteristics and the preset content recommendation condition set; extracting at least one target content feature from the fused content features according to the recommended condition; weighting the target content features based on the recommendation weight to obtain recommendation features corresponding to each recommendation condition; and screening target recommended content from the content to be recommended according to the recommended characteristics, and recommending the target recommended content. In this way, the recommendation weight of each recommendation condition in the preset content recommendation condition set is determined by acquiring the content characteristics of a plurality of dimensions corresponding to the content to be recommended, the target content characteristics are extracted from the fused content characteristics, and then the target content characteristics are weighted according to the recommendation weights, so that the recommendation characteristics corresponding to each recommendation condition are obtained, the target recommendation content is screened out from the content to be recommended according to the recommendation characteristics, the accuracy of content recommendation is improved, and the content recommendation efficiency is further improved.
According to the method described in the above embodiments, examples are described in further detail below.
In this embodiment, a description will be given of an example in which the content recommendation apparatus is specifically integrated in a computer device. The content recommendation method takes a server as an execution subject, and the content to be recommended is specifically described by taking information content as an example.
For a better description of embodiments of the present application, please refer to fig. 6. Fig. 6 is a schematic flow chart of a content recommendation method according to an embodiment of the present application. The specific flow is as follows:
in step 201, a server obtains at least one content to be recommended, performs attribute feature extraction on the content to be recommended to obtain content attribute features corresponding to the content to be recommended, and performs semantic feature extraction on the content to be recommended to obtain content semantic features corresponding to the content to be recommended.
The server may obtain at least one content to be recommended, for example, may obtain at least one information content through various internet platforms, further may perform attribute feature extraction on the content to be recommended to obtain a content attribute feature corresponding to the content to be recommended, and perform semantic feature extraction on the content to be recommended to obtain a content semantic feature corresponding to the content to be recommended, where the content attribute feature may include discrete features of dimensions such as information type, number of pictures, length of pictures, author, and the like, and the content semantic feature may include semantic features of dimensions such as information title, information text, information tag (tag), and the like, where the content semantic feature may be obtained by performing feature extraction through a language model (such as word2vec, glove, BERT).
In step 202, the server performs full connection processing on the content attribute feature and the content semantic feature corresponding to the content to be recommended to obtain a content feature of a first dimension, performs cross feature extraction on the content attribute feature and the content semantic feature corresponding to the content to be recommended to obtain a content feature of a second dimension, and uses the content feature of the first dimension and the content feature of the second dimension as content features of multiple dimensions.
The method for performing the full connection processing on the content attribute feature and the content semantic feature corresponding to the content to be recommended may be various, for example, the server may perform the full connection processing on the content attribute feature and the content semantic feature corresponding to the content to be recommended through a neural network. Correspondingly, there may be various ways of performing cross feature extraction on the content attribute features and the content semantic features corresponding to the content to be recommended, for example, the server may use an FM model (Factor Machine), a Product-based neural network (PNN) or other models to perform cross feature extraction on the content attribute features and the content semantic features.
Optionally, the server may perform multidimensional feature extraction on the content attribute feature and the content semantic feature corresponding to the content to be recommended by using depth factor decomposition (Deep FM model), where the Deep FM model may include a Deep (Deep) portion and an FM model, where the Deep portion may be a neural network structure and may be composed of 2 full-connection layers to extract low-level dimensional features in the feature, the FM model is a machine learning algorithm based on matrix decomposition, and may cross the feature, and the FM model has linear time complexity in terms of feature cross, so that the FM model may extract features with higher dimensions of vectors. Therefore, the Deep part in the Deep FM model can be used for carrying out full connection processing on the content attribute features and the content semantic features corresponding to the content to be recommended to obtain the content features of the first dimension, and the FM model in the Deep FM model is used for carrying out cross feature extraction on the content attribute features and the content semantic features corresponding to the content to be recommended to obtain the content features of the second dimension.
In step 203, the server fuses the content features, and based on the preset content recommendation condition set, the weight network corresponding to each recommendation condition in the preset content recommendation condition set is screened out from the weight network set of the trained content recommendation model.
The method of fusing the content features may be multiple, for example, the content features of multiple dimensions may be spliced to obtain a fused content feature, for example, please refer to fig. 3a, the content features of multiple dimensions may be input into a shared parameter layer to be spliced to obtain a fused content feature, a weight value may be set for the content feature of each dimension, the content feature may be weighted according to the weight value corresponding to the content feature of each dimension, and the fused content feature may be obtained according to the weighted result.
In step 204, the server performs normalization processing on the fused content features based on the weight network to obtain normalized content features, and classifies the normalized content features to obtain recommendation weights corresponding to each recommendation condition in the preset content recommendation condition set.
The normalized content features may be feature information obtained by normalizing the fused content features, and there may be various ways to normalize the fused content features, for example, please continue to refer to fig. 3a, and the server may normalize the fused content features by using a full connection layer to obtain the normalized content features.
After the normalized content features are obtained, the normalized content features may be classified, where there may be multiple ways to classify the normalized content features, for example, referring to fig. 3a, a server may use a logistic regression layer (Softmax layer) to classify the normalized content features.
Optionally, the server may adopt a multi-task learning model to screen out a weight network corresponding to each recommendation condition in the preset content recommendation condition set from the weight network set of the trained content recommendation model, and calculate the recommendation weight of each recommendation condition in the preset content recommendation condition set according to the weight network corresponding to each recommendation condition and the fused content characteristics. Specifically, the MMOE model may include a plurality of weight networks (Gate) and a plurality of Expert networks (Expert), each of the weight networks may be composed of a full-connection layer and a Softmax layer, each recommended condition corresponds to one weight network, the number of Expert networks and the number of parameters may be set according to the actual situation, the Expert networks are generally full-connection networks, and the Expert networks may learn important features from the shared parameter layer. Each weight network learns to assign recommendation weights to the expert networks according to learning tasks corresponding to recommendation conditions required to be responsible by the weight network.
Therefore, the server can screen out the weight network corresponding to each recommendation condition in the preset content recommendation condition set from the weight network set of the MMOE model, and therefore the recommendation sub-weight of each recommendation condition corresponding to each expert network can be calculated according to the weight network corresponding to each recommendation condition and the fused content characteristics, namely the input fused content characteristics are mapped into the dimension corresponding to each expert network through the weight network through linear transformation, and then the Softmax is calculated to obtain the recommendation sub-weight of each expert network, and further the recommendation sub-weight corresponding to each recommendation condition can be weighted to obtain the recommendation weight of each recommendation condition.
For example, referring to fig. 3a, assuming that the preset content recommendation condition set includes recommendation condition 1 and recommendation condition 2, and assuming that the number of Expert networks is 3, namely, expert1, expert2, and Expert3, the server may screen out a weight network corresponding to each recommendation condition in the preset content recommendation condition set in the weight network set, that is, recommendation condition 1 corresponds to weight network 1, recommendation condition 2 corresponds to weight network 2, so that the recommended sub-weight of each Expert network corresponding to recommendation condition 1 may be calculated according to weight network 1 and the fused content characteristics, the recommended sub-weight of each Expert network corresponding to recommendation condition 2 may be calculated according to weight network 2 and the fused content characteristics, therefore, the recommended weight of each recommended condition can be obtained according to the recommended sub-weight corresponding to each recommended condition, for example, the recommended sub-weights of the corresponding recommended condition in each Expert network are calculated based on the weight network, the recommended sub-weights of the recommended condition 1 in the Expert networks Expert1, expert2 and Expert3 can be obtained as a, b and c, the recommended sub-weights of the recommended condition 2 in the Expert networks Expert1, expert2 and Expert3 are d, e and f, and the server can obtain the recommended weights of the recommended condition 1 in each Expert network according to the recommended sub-weights corresponding to each recommended condition, wherein the recommended weights of the recommended condition 1 in each Expert network are a, b and c, and the recommended weights of the recommended condition 2 in each Expert network are d, e and f.
Optionally, the recommended sub-weights corresponding to the recommended condition 1 and the recommended condition 2 may be weighted to obtain the recommended weight of each recommended condition.
The manner of weighting the recommended sub-weights corresponding to each recommended condition may be multiple, for example, it may be assumed that the preset content recommended condition set includes two recommended conditions, namely, recommended condition 1 and recommended condition 2, and meanwhile, it is assumed that the number of Expert networks is 3, namely, expert1, expert2 and Expert3, the recommended sub-weights of the corresponding recommended condition in each Expert network are calculated based on the weight network, the recommended sub-weights of the recommended condition 1 in the Expert networks Expert1, expert2 and Expert3 are a, b and c, the recommended sub-weights of the recommended condition 2 in the Expert networks Expert1, expert2 and Expert3 are d, e and f, and then the server may perform weighted average on the recommended sub-weights corresponding to each recommended condition to obtain the recommended condition 1 as (a+b+c)/3, and the recommended condition 2 as (d+e+f)/3.
In step 205, the server extracts at least one target content feature from the fused content features according to the recommendation conditions, and weights the target content features based on the recommendation weights to obtain recommendation features corresponding to each recommendation condition.
For example, the server may determine the number of expert networks according to the recommended condition and set the number, and may further extract at least one target content feature from the fused content features through the set at least one expert network.
The number of the expert networks may be determined according to the recommended conditions, for example, the number of the expert networks may be determined according to the number of the recommended conditions, and when the number of the recommended conditions is large, the number of the expert networks may be increased to increase the number of the parameters.
After the number of the expert networks is determined and set, the server can extract at least one target content feature from the fused content features through the set at least one expert network, and various modes can be adopted for extracting at least one target content feature from the fused content features through the set at least one expert network, for example, the server can train a preset content recommendation method by adopting the number of the expert networks to be set to obtain a trained content recommendation model, so that the fused content features can be input into the expert networks in the trained content recommendation model to be fully connected to obtain the target content features corresponding to each expert network.
It should be noted that, the server may acquire the target content feature after acquiring the recommended weight of each recommended condition, may acquire the recommended weight of each recommended condition after acquiring the target content feature, may acquire the target content feature and the recommended weight of each recommended condition at the same time, and the sequence of the two steps may be determined according to the actual situation, which is not limited herein.
After obtaining the target content feature, the server may weight the target content feature based on the recommendation weight to obtain a recommendation feature corresponding to each recommendation condition, for example, may weight each target content feature according to the recommendation weight corresponding to each recommendation condition, and may average the weighted result to obtain a recommendation feature corresponding to each recommendation condition, for example, it may be assumed that a preset content recommendation condition set includes recommendation condition 1 and recommendation condition 2, and meanwhile, it is assumed that the target content feature includes G, H and I, recommendation weights corresponding to recommendation condition 1 in target content feature G, H and I are j1, j2 and j3, and recommendation weights corresponding to recommendation condition 2 in target content feature G, H and I are k1, k2 and k3, respectively. Then a weighted average may be performed on each target content feature based on the recommendation weight, and a recommendation feature corresponding to recommendation condition 1 may be obtained as (j1×g+j2×h+j3×i)/(j1+j2+j3), and a recommendation feature corresponding to recommendation condition 2 may be obtained as (k1×g+k2×h+k3×i)/(k1+k2+k3).
In step 206, the server classifies the recommendation characteristics to obtain recommendation levels corresponding to the content to be recommended under each recommendation condition, and fuses the recommendation levels to obtain recommendation coefficients of the content to be recommended.
The manner of fusing the recommendation levels to obtain the recommendation coefficient of the content to be recommended may be various, for example, the server may assign a numerical value to each recommendation level and a weight to each recommendation condition, where each recommendation weight may be determined according to the importance degree of each recommendation condition, for example, if the recommendation condition corresponding to the average person reading time length is important, a greater weight may be assigned to the recommendation condition corresponding to the average person reading time length, so that the result is biased to the more important recommendation condition, and if the recommendation condition corresponding to the click rate is important, a greater weight may be assigned to the recommendation condition corresponding to the click rate. Specifically, it may be assumed that there are a total of 8 recommendation levels, the values corresponding to the levels 1 to 8 are 1 to 8, respectively, the level 8 indicates that the recommendation degree of the content to be recommended is highest, meanwhile, the preset content recommendation condition set includes recommendation condition 1 and recommendation condition 2, the weight corresponding to the recommendation condition 1 is 0.6, the weight corresponding to the recommendation condition 2 may be 0.4, the recommendation level corresponding to the recommendation condition 1 is 6, the recommendation level corresponding to the recommendation condition 2 is 4, the recommendation level corresponding to the recommendation condition 2 is 5, the recommendation coefficient corresponding to the recommendation condition 2 is 6, the recommendation coefficient corresponding to the recommendation condition 1 may be 0.6x6+0.4x4=5.2, and the recommendation coefficient corresponding to the recommendation condition 2 may be 0.6x5+0.4x6=5.4, and the recommendation coefficient of the content to be 2 is higher than the recommendation content 1, that is, the recommendation coefficient of the content to be 2 is higher than the content to be recommended 1.
Optionally, the recommended level may be fused by using a super parameter to obtain a recommended coefficient of the content to be recommended, for example, assuming that the super parameter α, a preset recommended condition set includes a recommended condition corresponding to a click rate and a recommended condition corresponding to a person average reading duration, and the recommended level of the content to be recommended corresponding to the click rate condition is. The recommendation coefficient can be expressed as
Recommendation coefficient=α+click rate recommendation level + (1- α) ×average reading duration recommendation level
In step 207, the server ranks the content to be recommended based on the recommendation coefficient, obtains ranking information of the content to be recommended, and screens out target recommended content from the content to be recommended based on the ranking information.
The ranking information may be a ranking result representing each content to be recommended, for example, may be ranking information corresponding to each content to be recommended, according to which the server may screen out the target content to be recommended, for example, may determine the content to be recommended ranked in the top N bits as the target content to be recommended.
In step 208, the server stores the target recommended content in a corresponding recall content pool, and when a content recommendation request sent by the terminal is received, determines at least one recommended content in the target recommended content in the recall content pool based on the content recommendation request, and pushes the recommended content to the terminal.
The recall content pool may be used to store the content to be recalled, that is, store the target recommended content, where the recommended content may be the content to be distributed to the corresponding terminal. When receiving a content recommendation request sent by a terminal, the server can determine at least one recommendation content in target recommendation contents of the recall content pool according to the content recommendation request, and then can push the recommendation content to the terminal for display.
As can be seen from the foregoing, in the embodiment of the present application, at least one content to be recommended is obtained through a server, attribute feature extraction is performed on the content to be recommended, so as to obtain content attribute features corresponding to the content to be recommended, and semantic feature extraction is performed on the content to be recommended, so as to obtain content semantic features corresponding to the content to be recommended; the server performs full connection processing on the content attribute features and the content semantic features corresponding to the content to be recommended to obtain content features of a first dimension, performs cross feature extraction on the content attribute features and the content semantic features corresponding to the content to be recommended to obtain content features of a second dimension, and takes the content features of the first dimension and the content features of the second dimension as content features of a plurality of dimensions; the server fuses the content characteristics, and based on the preset content recommendation condition set, a weight network corresponding to each recommendation condition in the preset content recommendation condition set is screened out from the weight network set of the trained content recommendation model; the server performs normalization processing on the fused content characteristics based on the weight network to obtain normalized content characteristics, and classifies the normalized content characteristics to obtain recommendation weights corresponding to each recommendation condition in the preset content recommendation condition set; the server extracts at least one target content feature from the fused content features according to the recommended conditions, and weights the target content features based on the recommended weights to obtain recommended features corresponding to each recommended condition; the server classifies the recommendation characteristics to obtain recommendation levels corresponding to the content to be recommended under each recommendation condition, and fuses the recommendation levels to obtain recommendation coefficients of the content to be recommended; the server sorts the contents to be recommended based on the recommendation coefficient to obtain sorting information of the contents to be recommended, and screens out target recommended contents from the contents to be recommended based on the sorting information; the server stores the target recommended content into a corresponding recall content pool, and when a content recommendation request sent by a terminal is received, at least one recommended content is determined in the target recommended content of the recall content pool based on the content recommendation request, and the recommended content is pushed to the terminal. Therefore, the recommendation weight is distributed to each recommendation condition through the weight network corresponding to each recommendation condition by extracting the content characteristics of a plurality of dimensions corresponding to the content to be recommended and based on the correlation among the plurality of recommendation conditions, the target content characteristics are extracted from the fused content characteristics, and the recommendation characteristics corresponding to each recommendation condition are obtained by weighting the target content characteristics according to the recommendation weights, so that the target recommendation content is screened from the content to be recommended according to the recommendation characteristics, the accuracy of content recommendation is improved, and the content recommendation efficiency is further improved.
In order to better implement the above method, the embodiment of the present invention further provides a content recommendation device, which may be integrated in a computer device, and the computer device may be a server.
For example, as shown in fig. 7, a schematic structural diagram of a content recommendation device provided in an embodiment of the present application, the content recommendation device may include an acquisition unit 301, a determination unit 302, an extraction unit 303, a weighting unit 304, and a recommendation unit 305, as follows:
the acquiring unit 301 is configured to acquire at least one content to be recommended, and perform multidimensional feature extraction on the content to be recommended to obtain content features of multiple dimensions;
a determining unit 302, configured to fuse the content features, and determine a recommendation weight of each recommendation condition in the preset content recommendation condition set according to the fused content features and the preset content recommendation condition set;
an extracting unit 303, configured to extract at least one target content feature from the fused content features according to the recommendation condition;
a weighting unit 304, configured to weight the target content feature based on the recommendation weight, so as to obtain a recommendation feature corresponding to each recommendation condition;
and the recommending unit 305 is configured to screen out target recommended content from the content to be recommended according to the recommended feature, and recommend the target recommended content.
In an embodiment, the determining unit 302 includes:
the weight network screening subunit is used for screening the weight network corresponding to each recommendation condition in the preset content recommendation condition set from the weight network set of the trained content recommendation model based on the preset content recommendation condition set;
and the recommendation weight calculating subunit is used for calculating the recommendation weight of each recommendation condition in the preset content recommendation condition set according to the weight network corresponding to each recommendation condition and the fused content characteristics.
In one embodiment, the recommendation weight calculation subunit includes:
the processing module is used for carrying out normalization processing on the fused content characteristics based on the weight network to obtain normalized content characteristics;
and the classification module is used for classifying the normalized content characteristics to obtain the recommendation weight corresponding to each recommendation condition in the preset content recommendation condition set.
In one embodiment, the recommendation unit 305 includes:
the recommendation characteristic classification subunit is used for classifying the recommendation characteristics to obtain recommendation levels corresponding to the contents to be recommended under each recommendation condition;
a recommendation level fusion subunit, configured to fuse the recommendation levels to obtain a recommendation coefficient of the content to be recommended;
And the screening subunit is used for screening out target recommended content from the content to be recommended according to the recommendation coefficient and recommending the target recommended content.
In one embodiment, the screening subunit includes:
the ordering module is used for ordering the content to be recommended based on the recommendation coefficient to obtain ordering information of the content to be recommended;
and the screening module is used for screening out target recommended content from the content to be recommended based on the sorting information.
In one embodiment, the screening subunit includes:
the storage module is used for storing the target recommended content into a corresponding recall content pool;
the recommended content determining module is used for determining at least one recommended content in the target recommended content of the recall content pool based on the content recommendation request when the content recommendation request sent by the terminal is received;
and the pushing module is used for pushing the recommended content to the terminal.
In one embodiment, the obtaining unit 301 includes:
the semantic and attribute feature extraction subunit is used for extracting attribute features of the content to be recommended to obtain content attribute features corresponding to the content to be recommended, and extracting semantic features of the content to be recommended to obtain content semantic features corresponding to the content to be recommended;
And the multidimensional feature extraction subunit is used for carrying out multidimensional feature extraction on the content attribute features and the content semantic features corresponding to the content to be recommended to obtain content features with multiple dimensions.
In one embodiment, the multi-dimensional feature extraction subunit comprises:
the full-connection processing module is used for carrying out full-connection processing on the content attribute characteristics and the content semantic characteristics corresponding to the content to be recommended to obtain content characteristics of a first dimension;
the cross feature extraction module is used for carrying out cross feature extraction on the content attribute features and the content semantic features corresponding to the content to be recommended to obtain content features of a second dimension;
and the assignment module is used for taking the content characteristics of the first dimension and the content characteristics of the second dimension as the content characteristics of a plurality of dimensions.
In the implementation, each unit may be implemented as an independent entity, or may be implemented as the same entity or several entities in any combination, and the implementation of each unit may be referred to the foregoing method embodiment, which is not described herein again.
As can be seen from the above, in the embodiment of the present application, the obtaining unit 301 obtains at least one content to be recommended, and performs multidimensional feature extraction on the content to be recommended to obtain content features with multiple dimensions; the determining unit 302 fuses the content features, and determines the recommendation weight of each recommendation condition in the preset content recommendation condition set according to the fused content features and the preset content recommendation condition set; the extracting unit 303 extracts at least one target content feature from the fused content features according to the recommended condition; the weighting unit 304 weights the target content features based on the recommendation weights to obtain recommendation features corresponding to each recommendation condition; the recommendation unit 305 screens out target recommended content from the contents to be recommended according to the recommendation characteristics, and recommends the target recommended content. In this way, the recommendation weight of each recommendation condition in the preset content recommendation condition set is determined by acquiring the content characteristics of a plurality of dimensions corresponding to the content to be recommended, the target content characteristics are extracted from the fused content characteristics, and then the target content characteristics are weighted according to the recommendation weights, so that the recommendation characteristics corresponding to each recommendation condition are obtained, the target recommendation content is screened out from the content to be recommended according to the recommendation characteristics, the accuracy of content recommendation is improved, and the content recommendation efficiency is further improved.
The embodiment of the application further provides a computer device, as shown in fig. 8, which shows a schematic structural diagram of the computer device according to the embodiment of the application, where the computer device may be a server, specifically:
the computer device may include one or more processors 401 of a processing core, memory 402 of one or more computer readable storage media, a power supply 403, and an input unit 404, among other components. Those skilled in the art will appreciate that the computer device structure shown in FIG. 8 is not limiting of the computer device and may include more or fewer components than shown, or may be combined with certain components, or a different arrangement of components. Wherein:
processor 401 is the control center of the computer device and connects the various parts of the entire computer device using various interfaces and lines to perform various functions of the computer device and process data by running or executing software programs and/or modules stored in memory 402 and invoking data stored in memory 402, thereby performing overall monitoring of the computer device. Optionally, processor 401 may include one or more processing cores; preferably, the processor 401 may integrate an application processor and a modem processor, wherein the application processor mainly processes an operating system, a user interface, an application program, etc., and the modem processor mainly processes wireless communication. It will be appreciated that the modem processor described above may not be integrated into the processor 401.
The memory 402 may be used to store software programs and modules, and the processor 401 executes various function applications and content recommendation by running the software programs and modules stored in the memory 402. The memory 402 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program (such as a sound playing function, an image playing function, etc.) required for at least one function, and the like; the storage data area may store data created according to the use of the computer device, etc. In addition, memory 402 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 402 may also include a memory controller to provide the processor 401 with access to the memory 402.
The computer device further comprises a power supply 403 for supplying power to the various components, preferably the power supply 403 may be logically connected to the processor 401 by a power management system, so that functions of charge, discharge, and power consumption management may be performed by the power management system. The power supply 403 may also include one or more of any of a direct current or alternating current power supply, a recharging system, a power failure detection circuit, a power converter or inverter, a power status indicator, and the like.
The computer device may also include an input unit 404, which input unit 404 may be used to receive input numeric or character information and to generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.
Although not shown, the computer device may further include a display unit or the like, which is not described herein. In particular, in this embodiment, the processor 401 in the computer device loads executable files corresponding to the processes of one or more application programs into the memory 402 according to the following instructions, and the processor 401 executes the application programs stored in the memory 402, so as to implement various functions as follows:
acquiring at least one content to be recommended, and performing multidimensional feature extraction on the content to be recommended to obtain content features with multiple dimensions; fusing the content characteristics, and determining the recommendation weight of each recommendation condition in a preset content recommendation condition set according to the fused content characteristics and the preset content recommendation condition set; extracting at least one target content feature from the fused content features according to the recommended condition; weighting the target content features based on the recommendation weight to obtain recommendation features corresponding to each recommendation condition; and screening target recommended content from the content to be recommended according to the recommended characteristics, and recommending the target recommended content.
The specific implementation of each operation may be referred to the previous embodiments, and will not be described herein. It should be noted that, the computer device provided in the embodiment of the present application and the content recommendation method applicable to the above embodiment belong to the same concept, and detailed implementation processes of the computer device are shown in the above method embodiment, which is not repeated herein.
Those of ordinary skill in the art will appreciate that all or a portion of the steps of the various methods of the above embodiments may be performed by instructions, or by instructions controlling associated hardware, which may be stored in a computer-readable storage medium and loaded and executed by a processor.
To this end, embodiments of the present application provide a computer readable storage medium having stored therein a plurality of instructions capable of being loaded by a processor to perform steps in any of the content recommendation methods provided by embodiments of the present application. For example, the instructions may perform the steps of:
acquiring at least one content to be recommended, and performing multidimensional feature extraction on the content to be recommended to obtain content features with multiple dimensions; fusing the content characteristics, and determining the recommendation weight of each recommendation condition in a preset content recommendation condition set according to the fused content characteristics and the preset content recommendation condition set; extracting at least one target content feature from the fused content features according to the recommended condition; weighting the target content features based on the recommendation weight to obtain recommendation features corresponding to each recommendation condition; and screening target recommended content from the content to be recommended according to the recommended characteristics, and recommending the target recommended content.
Wherein the computer-readable storage medium may comprise: read Only Memory (ROM), random access Memory (RAM, random Access Memory), magnetic or optical disk, and the like.
Because the instructions stored in the computer readable storage medium may execute the steps in any content recommendation method provided in the embodiments of the present application, the beneficial effects that any content recommendation method provided in the embodiments of the present application can achieve are detailed in the previous embodiments, and are not described herein.
Among other things, according to one aspect of the present application, a computer program product or computer program is provided that includes computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the methods provided in the various alternative implementations provided in the above embodiments.
The foregoing has described in detail the methods, apparatus and computer readable storage medium for content recommendation provided by the embodiments of the present application, and specific examples have been applied herein to illustrate the principles and implementations of the present application, the above description of the embodiments being only for aiding in the understanding of the methods and core ideas of the present application; meanwhile, those skilled in the art will have variations in the specific embodiments and application scope in light of the ideas of the present application, and the present description should not be construed as limiting the present application in view of the above.
Claims (12)
1. A content recommendation method, comprising:
acquiring at least one content to be recommended, and carrying out multidimensional feature extraction on the content to be recommended to obtain content features with multiple dimensions;
fusing the content features, and determining the recommendation weight of each recommendation condition in a preset content recommendation condition set according to the fused content features and the preset content recommendation condition set;
extracting at least one target content feature from the fused content features according to the recommended condition;
weighting the target content features based on the recommendation weights to obtain recommendation features corresponding to each recommendation condition;
and screening target recommended content from the content to be recommended according to the recommended characteristics, and recommending the target recommended content.
2. The content recommendation method of claim 1 wherein said determining a recommendation weight for each recommendation condition in said set of preset content recommendation conditions based on said fused content features and said set of preset content recommendation conditions comprises:
screening out a weight network corresponding to each recommendation condition in the preset content recommendation condition set from a weight network set of the trained content recommendation model based on the preset content recommendation condition set;
And calculating the recommendation weight of each recommendation condition in the preset content recommendation condition set according to the weight network corresponding to each recommendation condition and the fused content characteristics.
3. The content recommendation method as claimed in claim 2, wherein calculating the recommendation weight of each recommendation condition in the preset content recommendation condition set according to the weight network corresponding to each recommendation condition and the fused content characteristics comprises:
based on the weight network, carrying out normalization processing on the fused content characteristics to obtain normalized content characteristics;
and classifying the normalized content characteristics to obtain recommendation weights corresponding to each recommendation condition in the preset content recommendation condition set.
4. The content recommendation method as claimed in claim 1, wherein said selecting a target recommended content from the contents to be recommended according to the recommendation characteristics, and recommending the target recommended content, comprises:
classifying the recommendation characteristics to obtain recommendation levels corresponding to the contents to be recommended under each recommendation condition;
fusing the recommendation levels to obtain recommendation coefficients of the contents to be recommended;
And screening out target recommended content from the content to be recommended according to the recommendation coefficient, and recommending the target recommended content.
5. The content recommendation method of claim 4 wherein said screening out target recommended content from said content to be recommended based on said recommendation coefficients comprises:
sorting the contents to be recommended based on the recommendation coefficients to obtain sorting information of the contents to be recommended;
and screening out target recommended content from the content to be recommended based on the sorting information.
6. The content recommendation method of claim 4, wherein the recommending the target recommended content comprises:
storing the target recommended content into a corresponding recall content pool;
when a content recommendation request sent by a terminal is received, determining at least one recommended content in target recommended contents of the recall content pool based on the content recommendation request;
pushing the recommended content to the terminal.
7. The method for recommending content according to any one of claims 1 to 6, wherein the multi-dimensional feature extraction is performed on the content to be recommended to obtain multi-dimensional content features, including:
Extracting attribute features of the content to be recommended to obtain content attribute features corresponding to the content to be recommended, and extracting semantic features of the content to be recommended to obtain content semantic features corresponding to the content to be recommended;
and carrying out multidimensional feature extraction on the content attribute features and the content semantic features corresponding to the content to be recommended to obtain content features with multiple dimensions.
8. The method of claim 7, wherein the multi-dimensional feature extraction is performed on the content attribute feature and the content semantic feature corresponding to the content to be recommended to obtain the content feature with multiple dimensions, including:
performing full connection processing on the content attribute characteristics and the content semantic characteristics corresponding to the content to be recommended to obtain content characteristics of a first dimension;
performing cross feature extraction on the content attribute features and the content semantic features corresponding to the content to be recommended to obtain content features of a second dimension;
and taking the content characteristics of the first dimension and the content characteristics of the second dimension as the content characteristics of a plurality of dimensions.
9. A content recommendation device, comprising:
the acquisition unit is used for acquiring at least one content to be recommended, and carrying out multidimensional feature extraction on the content to be recommended to obtain content features with multiple dimensions;
The determining unit is used for fusing the content characteristics and determining the recommendation weight of each recommendation condition in the preset content recommendation condition set according to the fused content characteristics and the preset content recommendation condition set;
the extraction unit is used for extracting at least one target content characteristic from the fused content characteristics according to the recommended conditions;
the weighting unit is used for weighting the target content characteristics based on the recommendation weight to obtain recommendation characteristics corresponding to each recommendation condition;
and the recommending unit is used for screening out target recommended content from the content to be recommended according to the recommending characteristics and recommending the target recommended content.
10. A computer readable storage medium storing a plurality of instructions adapted to be loaded by a processor to perform the steps of the content recommendation method of any one of claims 1 to 8.
11. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the content recommendation method according to any one of claims 1 to 8 when the computer program is executed.
12. A computer program, characterized in that the computer program comprises computer instructions stored in a storage medium, from which computer instructions a processor of a computer device reads, the processor executing the computer instructions, causing the computer device to perform the content recommendation method of any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210050591.8A CN116501949A (en) | 2022-01-17 | 2022-01-17 | Content recommendation method, apparatus and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210050591.8A CN116501949A (en) | 2022-01-17 | 2022-01-17 | Content recommendation method, apparatus and computer readable storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116501949A true CN116501949A (en) | 2023-07-28 |
Family
ID=87323563
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210050591.8A Pending CN116501949A (en) | 2022-01-17 | 2022-01-17 | Content recommendation method, apparatus and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116501949A (en) |
-
2022
- 2022-01-17 CN CN202210050591.8A patent/CN116501949A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111291266B (en) | Artificial intelligence based recommendation method and device, electronic equipment and storage medium | |
| CN112632385A (en) | Course recommendation method and device, computer equipment and medium | |
| CN113191154B (en) | Semantic analysis method, system and storage medium based on multi-modal graph neural network | |
| CN110245310B (en) | Object behavior analysis method, device and storage medium | |
| CN111538852B (en) | Multimedia resource processing method, device, storage medium and equipment | |
| CN113761359B (en) | Data packet recommendation method, device, electronic equipment and storage medium | |
| CN111222557A (en) | Image classification method and device, storage medium and electronic equipment | |
| CN111291564B (en) | Model training method, device and storage medium for word vector acquisition | |
| CN113821634A (en) | Content classification method and device, electronic equipment and storage medium | |
| CN117391497A (en) | News manuscript quality subjective and objective scoring consistency evaluation method and system | |
| CN115129902B (en) | Media data processing method, device, equipment and storage medium | |
| CN114707633B (en) | Feature extraction method, device, electronic equipment and storage medium | |
| CN116955788A (en) | Method, device, equipment, storage medium and program product for processing content | |
| CN116976283A (en) | Language processing method, training method, device, equipment, medium and program product | |
| CN116992124A (en) | Label ordering method, device, equipment, medium and program product | |
| CN115393925A (en) | Face attraction classification method, system, equipment and medium | |
| CN116501949A (en) | Content recommendation method, apparatus and computer readable storage medium | |
| CN114627085A (en) | Target image identification method and device, storage medium and electronic equipment | |
| CN113569130A (en) | Content recommendation method, device, equipment and readable storage medium | |
| CN112417290A (en) | Training method of book sorting push model, electronic equipment and storage medium | |
| CN117786234B (en) | Multimode resource recommendation method based on two-stage comparison learning | |
| CN114417944B (en) | Recognition model training method and device, and user abnormal behavior recognition method and device | |
| CN116992031B (en) | Data processing method, device, electronic equipment, storage medium and program product | |
| CN114510638B (en) | Information processing method, apparatus, device, storage medium, and program product | |
| CN109871487B (en) | News recall method and system |
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 | ||
| REG | Reference to a national code |
Ref country code: HK Ref legal event code: DE Ref document number: 40090932 Country of ref document: HK |