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.
Before explaining the embodiments of the present application in detail, some terms related to the embodiments of the present application are explained.
Wherein in the description of embodiments of the present application, the terms "first," "second," and the like may be used herein to describe various concepts, but such concepts are not limited by these terms unless otherwise specified. These terms are only used to distinguish one concept from another. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
The embodiment of the application provides a content recommendation method, a content recommendation device, electronic equipment and a computer readable storage medium. Specifically, the content recommendation method in the embodiment of the application may be performed by an electronic device, where the electronic device may be a device such as a terminal or a server. The terminal may be a terminal device such as a smart phone, a tablet computer, a notebook computer, a touch screen, a game console, a personal computer (PC, personal Computer), a personal digital assistant (Personal Digital Assistant, PDA), and the like, and the terminal may further include a client, which may be a game application client, a browser client carrying a game program, or an instant messaging client, and the like. The server may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs, basic cloud computing services such as big data and artificial intelligent platforms.
For example, the electronic device is illustrated by taking a terminal as an example, and the terminal can acquire a content consumption form of the content to be recommended, and acquire content display information of the content to be recommended under at least two feature dimensions when the content to be recommended is displayed based on the content consumption form; respectively extracting content features of the content display information corresponding to each feature dimension through a preset neural network model, and fusing the content features corresponding to a plurality of feature dimensions; predicting posterior consumption indexes of the content to be recommended when the content to be recommended is displayed in at least two types of groups according to the fused content characteristics through the neural network model; and recommending the content to be recommended in the corresponding type of groups based on the posterior consumption indexes of the content to be recommended in at least two types of groups.
Based on the above problems, embodiments of the present application provide a content recommendation method, apparatus, electronic device, and computer readable storage medium, which can improve content recommendation efficiency.
The following detailed description is provided with reference to the accompanying drawings. The following description of the embodiments is not intended to limit the preferred embodiments. Although a logical order is depicted in the flowchart, in some cases the steps shown or described may be performed in an order different than depicted in the figures.
In this embodiment, a terminal is taken as an example for explanation, and this embodiment provides a content recommendation method, as shown in fig. 1, the specific flow of the content recommendation method may be as follows:
101. and acquiring a content consumption form of the content to be recommended, and acquiring content display information of the content to be recommended under at least two characteristic dimensions during display based on the content consumption form.
The content to be recommended is content to be shared with a user on the content consumption platform, and the content to be recommended may be user note content, such as video recorded by an creator, written text notes written by the creator, pictures shot by the creator, and the like.
It may be appreciated that, since the content to be recommended may be video, notes, and/or multiple pictures, only the content presentation information, i.e., the visible elements of the content to be recommended, such as the cover picture, title, head portrait of the creator, or author name of the creator, of the content to be recommended, may be seen when the content to be recommended is presented on the content consumption platform before the user consumes the content to be recommended. The content display information when the content to be recommended is displayed also directly influences whether the user is attracted by the content to be recommended or not, and further the content to be recommended is consumed, namely the content display information is a factor which really influences the consumption when the user browses the content to be recommended. Therefore, in this embodiment, the terminal obtains the content display information of the content to be recommended when displaying, so as to recommend the content to be recommended based on the content display information, thereby promoting the model to be more fit with the real situation by truly fitting the scene seen by the user when browsing or consuming, improving the accuracy and reliability of the model prediction in the later stage, and further improving the recommendation efficiency.
In addition, since various content display information exists when the content to be recommended is displayed, in order to more accurately recommend the content to be recommended, in this embodiment, a terminal introduces feature dimensions, such as a text feature dimension, a visual feature dimension, an audio feature dimension and the like, the visual feature dimension further includes a picture feature dimension, a video feature dimension and the like, the text feature dimension includes a subtitle feature dimension, so that corresponding multi-mode content display information, such as a cover picture of the picture feature dimension, an avatar of an creator, a content title of the text feature dimension, an author name of the creator and the like, is obtained through various feature dimensions, thereby enabling a model to capture more information to predict a result and further helping the model to improve accuracy of the prediction result. The multi-mode is not limited to a single medium, but a combination of multiple media, namely, a combination of content presentation information with multiple feature dimensions.
The image can give the most intuitive feeling to people, so that the image on some visual characteristic dimensions such as head portrait of an creator, homepage background image of the creator, cover image and the like can carry some information, and can be used as one of reference characteristic dimensions; because of the author information such as the name, the introduction, the alias and the like of the author, the content to be recommended issued by the author in a period of time can be checked, and the relationship between the displayed contents issued by general authors in a period of time, such as content consistency, quality correlation and the like, the author information can be used as information in the dimension of text characteristics; however, since there is a content title in the content to be recommended generally, that is, a relatively accurate and brief sentence extracted by the creator of the content to be recommended for the whole content to be recommended before the content is released, there may be abundant semantic information in the content title, which is helpful for understanding the content to be recommended.
It can be appreciated that, since the content to be recommended is exposed on different content consumption platforms, or different consumption forms of the same content consumption platform, the content displayed by the content to be recommended is different, that is, the corresponding feature dimensions are also different, for example, only the content title of the content to be recommended and the author name of the creator, or only the cover picture of the content to be recommended, or the video may be displayed directly. Therefore, in this embodiment, in order to better recommend the content to be recommended, the terminal needs to acquire a content consumption form when consuming the content to be recommended, so as to determine the content when the content to be recommended is displayed based on the content consumption form, and further determine content display information under at least two feature dimensions based on the content when the content to be recommended is displayed.
Wherein the content consumption form is used for indicating a form when content consumption is performed, and the content consumption form includes, but is not limited to, consumption based on navigation, consumption based on personalized recommendation, and information consumption based on SNS attribute. The scenes corresponding to different content consumption forms are different when the content is displayed on the content consumption platform, so that the display modes of the corresponding content are also different, and the content to be recommended is different when the content is displayed in different display modes.
For example, if the content to be recommended is consumed based on navigation, the display scene of the content to be recommended is displayed on a channel page, and then the display mode of the content to be recommended can be determined according to the display scene of the channel page on the content consumption platform. For example, if the display scene of the content consumption platform is an outflow double-row scene, the display manner of the corresponding content to be recommended is in a double-row waterfall arrangement, for example, as shown in fig. 2, it can be seen from fig. 2 that in the outflow double-row scene, the content to be recommended includes a cover picture, a content title, an author name of an creator, and an avatar, and the corresponding praise number may be displayed after the author name.
102. And respectively extracting the content characteristics of the content display information corresponding to each characteristic dimension through a preset neural network model, and fusing the content characteristics corresponding to a plurality of characteristic dimensions.
In this embodiment, after obtaining the content display information corresponding to each feature dimension, the terminal may extract the content features of the content display information through a pre-trained neural network model, so that the neural network model predicts the result based on the content features, and because there are content features corresponding to a plurality of feature dimensions, the content features of the plurality of feature dimensions need to be fused to synthesize the content features of the plurality of feature dimensions to predict the result, thereby improving the accuracy of the prediction result.
The neural network model may be a double-tower model.
In some embodiments, before extracting the content features of the content display information corresponding to each feature dimension through a preset neural network model and fusing the content features corresponding to a plurality of feature dimensions, the neural network model needs to be constructed and trained.
Specifically, the terminal may obtain the number of content display times when the history display content is displayed on at least one type of group, determine a target history display content with the number of content display times greater than a preset display threshold, and a target group type corresponding to the target history display content; acquiring a history posterior consumption index corresponding to the target history display content; constructing a training sample based on the target historical display content, the target group type and the historical posterior consumption index; and training the neural network model based on the training sample.
In this embodiment, the terminal is based on weak personalization, i.e. is divided into groups of different types, then, a sample corresponding to each group is constructed by combining the types of the groups and the target history display content, and the history posterior consumption index is used as a sample label to realize training on the neural network model. The training samples can be built from near to far in the historical period, when the number of the training samples accords with the number of the preset samples, the training samples are not built, the obtained training samples which accord with the number of the preset samples are subjected to model training, and the number of the preset samples can be set according to requirements, for example 3800 ten thousand.
It may be appreciated that, since the terminal trains the model to recommend content to the user, so as to promote the consumption time and consumption experience of the user on the platform, in this embodiment, a history posterior consumption index is introduced, where the history posterior consumption index is used to indicate the acceptance condition of the user for the history display content seen by the user, such as like, dislike, report, and the like, and the history posterior consumption index includes, but is not limited to, a history click rate, a history praise rate, and a history negative feedback rate. The model can predict posterior consumption indexes of the content to be recommended, and in order to prompt the model to learn more essential information of the historical display content, the model task can be set as a regression task, namely, the posterior consumption indexes of the historical display content on a certain specific type of group are regressed, and the target value corresponding to the regression task is a floating point number between 0 and 1.
It can be understood that, in order to ensure accuracy and reliability of the model prediction result and avoid occurrence of abnormal values, when the target history display content is mined, the number of times that the history display content is displayed on at least one type of group needs to be limited, that is, the number of times that the content is displayed needs to be greater than a preset display threshold, for example, 1000, and the greater the display threshold, the higher the reliability of the model prediction result after training will be, but the decrease of available data will also be caused, that is, the less the target history display content is determined.
And if the number of times the history display content is displayed on the at least one type of group is increased from 300 to 500, the corresponding prediction results in a history posterior consumption index, for example, the variation range of the history click rate is higher, for example, more than 100%, whereas if the number of times the history display content is displayed on the at least one type of group is increased from 500 to 1000, the variation range of the corresponding history click rate is gradually reduced, for example, more than 30%, and if the number of times the history display content is displayed on the at least one type of group is increased from 1000 to 1500, the variation range of the corresponding history click rate is gradually reduced, for example, more than 10%, so as the number of times of content display increases, the history posterior consumption index corresponding to the history display content gradually converges, and also represents that the larger the number of times of content display is more reliable, so that the number of times of display of the history display content on the at least one type of group is limited to be larger than a preset display threshold value, so as to ensure the accuracy of the data training model.
Specifically, the above groups of different types may be divided according to the User's needs, for example, according to the age and sex of the person who browses the history display content, for example, 8 groups according to age and sex, each group being set as Useri，j Where i represents a sex dimension, i may be divided into male or female, j represents an age dimension, and j may be divided into 1 to 18 sections, 19 to 25 sections, 26 to 35 sections, and 35 sections above 35, thereby dividing into 8 mutually orthogonal groups.
In some embodiments, the obtaining the historical a posteriori consumption index corresponding to the target historical display content may include: acquiring historical posterior consumption parameters corresponding to the target historical display content; and calculating the ratio between the historical posterior consumption parameter and the content display times of the target historical display content, and determining the ratio as the historical posterior consumption index.
Specifically, the above-mentioned history posterior consumption parameters refer to the actual consumption condition of the user collected after the target history display content is distributed on the platform for a period of time, such as the consumption conditions of praise, collection, reading time, etc., and include, but are not limited to, the history click number (labeled click_num), history praise number (labeled like_num), and history negative feedback number (labeled hide_num) of the target history display content when displayed on the population of the target population type, and correspondingly, the ratio (labeled click_num/imp_num) between the history click number and the content display number (labeled imp_num), namely the history click number The result of dividing the number by the number of content presentations is determined as the history click rate (noted as Note) of the target history presentation content when presented on the population of the target population typectr ) The method comprises the steps of carrying out a first treatment on the surface of the The ratio between the history point count and the number of content presentations (denoted like_num/imp_num), i.e. the result of dividing the history point count by the number of content presentations, is determined as the history point count (denoted Noteltr ) The method comprises the steps of carrying out a first treatment on the surface of the The ratio between the number of historical negative feedback and the number of content showing times (recorded as hide_num/imp_num), namely the result of dividing the number of historical negative feedback by the number of content showing times, is determined as the historical negative feedback rate (recorded as Notehtr )。
In some embodiments, since the target history display content may include history content display information of at least two feature dimensions, the training the neural network model based on the training sample may include: inputting the training sample into the neural network model; respectively extracting historical content characteristics of the historical content display information corresponding to each characteristic dimension through the neural network model, and fusing the historical content characteristics corresponding to a plurality of characteristic dimensions; predicting a predicted posterior consumption index of the target historical display content when the target historical display content is displayed on the group of the target group type according to the fused historical content characteristics through the neural network model; and adjusting parameters of the neural network model based on the predicted posterior consumption index and the historical posterior consumption index.
In some embodiments, the feature dimensions include a text feature dimension and a visual feature dimension, the target history display content includes at least one type of text information in the text feature dimension and at least one type of visual information in the visual feature dimension, and the extracting, by the neural network model, the history content features of the history content display information corresponding to each feature dimension includes: fusing at least one type of text information and at least one type of visual information through the neural network model to respectively obtain fused text information and fused visual information; and respectively extracting the historical content characteristics of the fused text information corresponding to the text characteristic dimension and the historical content characteristics of the fused visual information corresponding to the visual characteristic dimension through the neural network model.
It can be understood that, since the content display information, that is, the text information and the visual information, during the display of the target historical display content directly affects whether the user is attracted by the content to be recommended, and further consumes the content to be recommended, that is, the content display information is a factor that the user really affects consumption when browsing the content to be recommended, in this embodiment, the visual information needs to be processed, and the visual information may be a cover picture of the target historical display content, a video of the target historical display content, an avatar of an creator of the target historical display content, and the like; while text sometimes more directly indicates the semantic meaning of the target history presentation, such as the content title of the target history presentation. Thus, in this embodiment, the feature dimensions include a text feature dimension and a visual feature dimension, where the two feature dimensions cover many forms of the target history display content, such as when the target history display content is in a graphics form, and such as when the target history display content is in a video form.
In this embodiment, if there are text feature dimensions and visual feature dimensions and at least one type of information exists in each feature dimension, in order to reduce complexity of the model, or to enable similar information to learn from each other through the same model, the terminal may fuse multiple types of information in the same feature dimension.
In addition, the terminal may also fuse the same or similar types of information in the same feature dimension, or the terminal may also fuse the same or similar types of information in different feature dimensions, such as the name of the author of the creator of the text feature dimension and the head portrait of the creator of the visual feature dimension.
In some embodiments, the extracting, by the neural network model, the historical content features of the historical content display information corresponding to each feature dimension, and fusing the historical content features corresponding to a plurality of feature dimensions, includes: extracting historical content characteristics of the historical content display information corresponding to each characteristic dimension through a characteristic module corresponding to each characteristic dimension in the neural network model; the fusion module is used for controlling the feature modules corresponding to the feature dimensions to input the extracted historical content features into the neural network model; and fusing the historical content characteristics corresponding to the characteristic dimensions through the fusing module.
As shown in fig. 3, the feature dimensions are set to include a Text feature dimension and a visual feature dimension, that is, there is a feature module corresponding to the Text feature dimension, such as Text Encoder in fig. 3, there is a feature module corresponding to the visual feature dimension, such as Vision Encoder in fig. 3, the Text corresponding to the Text feature dimension and the picture corresponding to the visual feature dimension are respectively input into the corresponding feature model to obtain the visual feature and the Text feature, and then input into the Transformer for processing, so that the features are fused in a fusion module, which is the position of the "fusion feature" in fig. 3, and finally, the prediction result of the model is output through the full connection layer.
In some embodiments, the adjusting the parameters of the neural network model may include:
if the number of the training samples is greater than or equal to a preset sample threshold, the parameters of the feature module and the fusion module corresponding to each feature dimension in the neural network model are adjusted by adopting the same parameter adjustment strategy;
if the number of the training samples is smaller than a preset sample threshold, a parameter adjustment strategy corresponding to a first preset super parameter is adopted to adjust parameters of the feature modules corresponding to the feature dimensions in the neural network model, and a parameter adjustment strategy corresponding to a second preset super parameter is adopted to adjust parameters of the fusion modules in the neural network model, wherein the second preset super parameter is larger than the first preset super parameter.
In this embodiment, the parameter adjusted in the module is a weight parameter of the module. It can be understood that when the number of samples facing the training samples is large, the model can be supported to train from the head, and then the feature module and the fusion module can adopt the same parameter adjustment strategy to update and train with the same super parameter; if the number of the training samples is small, the fusion module is only suitable for optimizing the model, the fusion module needs to adopt larger super-parameters for parameter adjustment to train from the beginning so as to ensure the performance of the whole network, and the feature module can adopt a model with a trained or open source to initialize parameters, so that the feature module can adopt smaller first preset super-parameters relative to second preset super-parameters of the fusion module for parameter adjustment.
In some embodiments, the prediction accuracy and recall of the whole model can be obviously improved through text feature augmentation, loss function balance and other modes.
103. And predicting posterior consumption indexes of the content to be recommended when the content to be recommended is displayed in at least two types of groups according to the fused content characteristics through the neural network model.
It can be appreciated that, in order to better predict the posterior consumption index, it can be known from the above statement that the present embodiment can be identified based on weak personalization, that is, in the present embodiment, the group division types of the viewable content to be recommended are divided, for example, by gender, age, city, etc., so that the preference of the user can be described in a finer granularity, and the embodiment also accords with the application in the recommendation or search scene.
In the embodiment, the identification is performed through weak individuation, so that the prediction of posterior consumption indexes of the multi-mode content aiming at different specific groups is realized, and the method is different from the traditional method that all people are in the same view, so that the method is more accurate in downstream application and avoids errors.
The posterior consumption index is used for indicating the receiving condition of the content to be recommended seen by the user, wherein the receiving condition can be like, dislike, report, spend more reading time, praise, collection and the like. Such posterior consumption metrics include, but are not limited to, click rate, praise rate, negative feedback rate, etc. of the content.
104. And recommending the content to be recommended in the corresponding type of groups based on the posterior consumption indexes of the content to be recommended in at least two types of groups.
In this embodiment, after obtaining posterior consumption indexes of the content to be recommended in different types of groups, the terminal may perform corresponding recommendation, for example, flow support, according to the levels of the corresponding indexes, so as to encourage more recommendation in a certain group, or perform a pressing operation so as to avoid recommendation as much as possible in a certain group. Therefore, the prediction of posterior consumption indexes of the multi-mode content aiming at different specific groups is realized through weak individuation, so that different posterior consumption indexes are output aiming at different groups, and content distribution or sinking operation can be performed more finely in the recommendation/search stage, thereby improving the content recommendation efficiency.
When the content to be recommended in the image-text form is displayed in the outflow double-row scene, the multi-mode information is predicted on different types of groups through elements such as a cover chart, a title, an author name, an author head portrait and the like which can only be seen by the content at the moment, so that posterior consumption indexes corresponding to the different types of groups are obtained. Then, the score corresponding to the posterior consumption index can be further considered when the model predicts the posterior consumption index, and in the recommendation or search stage, the content belonging to a certain type of group with high confidence is stimulated and supported, namely more is recommended, for example, ranking can be performed according to the posterior consumption index corresponding to different types of group, and then the content to be recommended is shared to different types of group based on the ranking, for example, the number of times of the content to be recommended shared by the group with higher ranking is more.
In addition, posterior consumption indexes of the contents to be recommended on different types of groups can be input into a preset large recommendation model, so that the contents to be recommended are recommended through the large recommendation model, the consumption time of a user on a platform is prolonged, the display time of the contents is prolonged, the positive feedback of the user is increased, the negative feedback of the user is reduced, and the like.
The content display information of the content to be recommended in at least two characteristic dimensions during display is obtained by obtaining the content consumption form of the content to be recommended and based on the content consumption form. And then, respectively extracting the content characteristics of the content display information corresponding to each characteristic dimension through a preset neural network model, and fusing the content characteristics corresponding to a plurality of characteristic dimensions. And predicting posterior consumption indexes of the content to be recommended when the content to be recommended is displayed in at least two types of groups according to the fused content characteristics by the neural network model. And finally, based on the posterior consumption indexes of the content to be recommended in at least two types of groups, recommending the content to be recommended in the corresponding type of groups respectively, so that the content recommendation efficiency is improved by analyzing the content to be recommended from a plurality of characteristic dimensions and considering the situation of the content to be recommended in different types of groups during prediction.
In order to better implement the above method, the embodiment of the present application further provides a content recommendation device, where the content recommendation device may be specifically integrated in an electronic device, for example, a computer device, where the computer device may be a terminal, a server, or other devices.
The terminal can be a mobile phone, a tablet personal computer, an intelligent Bluetooth device, a notebook computer, a personal computer and other devices; the server may be a single server or a server cluster composed of a plurality of servers.
For example, in this embodiment, a method in this application will be described in detail by taking a specific integration of a content recommendation device in a terminal as an example, and this embodiment provides a content recommendation device, as shown in fig. 4, where the content recommendation device may include:
the information acquisition module 401 is configured to acquire a content consumption form of a content to be recommended, and acquire content display information of the content to be recommended under at least two feature dimensions when the content to be recommended is displayed based on the content consumption form;
the feature extraction module 402 is configured to extract content features of the content display information corresponding to each feature dimension through a preset neural network model, and fuse content features corresponding to a plurality of feature dimensions;
The prediction module 403 is configured to predict, according to the fused content characteristics, posterior consumption indexes of the content to be recommended when the content is displayed in at least two types of groups, by using the neural network model;
and the recommending module 404 is configured to recommend the content to be recommended in the corresponding type of group based on the posterior consumption indexes of the content to be recommended in at least two types of groups.
In some embodiments, the content recommendation device further includes a model training module, where the model training module is specifically configured to:
acquiring the content display times of the history display content when the history display content is displayed on at least one type of group, and determining target history display content with the content display times larger than a preset display threshold value and a target group type corresponding to the target history display content;
acquiring a history posterior consumption index corresponding to the target history display content;
constructing a training sample based on the target historical display content, the target group type and the historical posterior consumption index;
and training the neural network model based on the training sample.
In some embodiments, the model training module is specifically configured to:
Acquiring historical posterior consumption parameters corresponding to the target historical display content;
and calculating the ratio between the historical posterior consumption parameter and the content display times of the target historical display content, and determining the ratio as the historical posterior consumption index.
In some embodiments, the target historical display content includes historical content display information of at least two feature dimensions, and the model training module is specifically configured to:
inputting the training sample into the neural network model;
respectively extracting historical content characteristics of the historical content display information corresponding to each characteristic dimension through the neural network model, and fusing the historical content characteristics corresponding to a plurality of characteristic dimensions;
predicting a predicted posterior consumption index of the target historical display content when the target historical display content is displayed on the group of the target group type according to the fused historical content characteristics through the neural network model;
and adjusting parameters of the neural network model based on the predicted posterior consumption index and the historical posterior consumption index.
In some embodiments, the feature dimensions include a text feature dimension and a visual feature dimension, the target history display content includes at least one type of text information in the text feature dimension and at least one type of visual information in the visual feature dimension, and the model training module is specifically configured to:
Fusing at least one type of text information and at least one type of visual information through the neural network model to respectively obtain fused text information and fused visual information;
and respectively extracting the historical content characteristics of the fused text information corresponding to the text characteristic dimension and the historical content characteristics of the fused visual information corresponding to the visual characteristic dimension through the neural network model.
In some embodiments, the model training module is specifically configured to:
extracting historical content characteristics of the historical content display information corresponding to each characteristic dimension through a characteristic module corresponding to each characteristic dimension in the neural network model;
the fusion module is used for controlling the feature modules corresponding to the feature dimensions to input the extracted historical content features into the neural network model;
and fusing the historical content characteristics corresponding to the characteristic dimensions through the fusing module.
In some embodiments, the model training module is specifically configured to:
if the number of the training samples is greater than or equal to a preset sample threshold, the parameters of the feature module and the fusion module corresponding to each feature dimension in the neural network model are adjusted by adopting the same parameter adjustment strategy;
If the number of the training samples is smaller than a preset sample threshold, a parameter adjustment strategy corresponding to a first preset super parameter is adopted to adjust parameters of the feature modules corresponding to the feature dimensions in the neural network model, and a parameter adjustment strategy corresponding to a second preset super parameter is adopted to adjust parameters of the fusion modules in the neural network model, wherein the second preset super parameter is larger than the first preset super parameter.
As can be seen from the above, the information obtaining module 401 obtains the content consumption form of the content to be recommended, and based on the content consumption form, obtains the content display information of the content to be recommended in at least two feature dimensions during display. Then, the feature extraction module 402 extracts the content features of the content display information corresponding to each feature dimension through a preset neural network model, and fuses the content features corresponding to the feature dimensions. The prediction module 403 predicts posterior consumption indexes of the content to be recommended when the content to be recommended is displayed in at least two types of groups according to the fused content characteristics through the neural network model. Finally, the recommendation module 404 recommends the content to be recommended in the corresponding type of group based on the posterior consumption index of the content to be recommended in at least two types of groups, so as to improve content recommendation efficiency by analyzing the content to be recommended from a plurality of feature dimensions and considering the situation of the content to be recommended in different types of groups during prediction.
Correspondingly, the embodiment of the application also provides electronic equipment, which can be a terminal, and the terminal can be terminal equipment such as a smart phone, a tablet personal computer, a notebook computer, a touch screen, a game machine, a personal computer (PC, personal Computer), a personal digital assistant (Personal Digital Assistant, PDA) and the like. As shown in fig. 5, fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present application. The electronic device 500 includes a processor 501 having one or more processing cores, a memory 502 having one or more computer readable storage media, and a computer program stored on the memory 502 and executable on the processor. The processor 501 is electrically connected to the memory 502. It will be appreciated by those skilled in the art that the electronic device structure shown in the figures is not limiting of the electronic device and may include more or fewer components than shown, or may combine certain components, or a different arrangement of components.
The processor 501 is a control center of the electronic device 500, connects various portions of the entire electronic device 500 using various interfaces and lines, and performs various functions of the electronic device 500 and processes data by running or loading software programs and/or modules stored in the memory 502, and invoking data stored in the memory 502, thereby performing overall monitoring of the electronic device 500.
In the embodiment of the present application, the processor 501 in the electronic device 500 loads the instructions corresponding to the processes of one or more application programs into the memory 502 according to the following steps, and the processor 501 executes the application programs stored in the memory 502, so as to implement various functions:
acquiring a content consumption form of the content to be recommended, and acquiring content display information of the content to be recommended under at least two characteristic dimensions during display based on the content consumption form;
respectively extracting content features of the content display information corresponding to each feature dimension through a preset neural network model, and fusing the content features corresponding to a plurality of feature dimensions;
predicting posterior consumption indexes of the content to be recommended when the content to be recommended is displayed in at least two types of groups according to the fused content characteristics through the neural network model;
and recommending the content to be recommended in the corresponding type of groups based on the posterior consumption indexes of the content to be recommended in at least two types of groups.
In some embodiments, before extracting the content features of the content display information corresponding to each feature dimension through a preset neural network model, and fusing the content features corresponding to a plurality of feature dimensions, the method further includes:
Acquiring the content display times of the history display content when the history display content is displayed on at least one type of group, and determining target history display content with the content display times larger than a preset display threshold value and a target group type corresponding to the target history display content;
acquiring a history posterior consumption index corresponding to the target history display content;
constructing a training sample based on the target historical display content, the target group type and the historical posterior consumption index;
and training the neural network model based on the training sample.
In some embodiments, the obtaining the historical posterior consumption index corresponding to the target historical display content includes:
acquiring historical posterior consumption parameters corresponding to the target historical display content;
and calculating the ratio between the historical posterior consumption parameter and the content display times of the target historical display content, and determining the ratio as the historical posterior consumption index.
In some embodiments, the target history display content includes history content display information of at least two feature dimensions, and the training the neural network model based on the training sample includes:
Inputting the training sample into the neural network model;
respectively extracting historical content characteristics of the historical content display information corresponding to each characteristic dimension through the neural network model, and fusing the historical content characteristics corresponding to a plurality of characteristic dimensions;
predicting a predicted posterior consumption index of the target historical display content when the target historical display content is displayed on the group of the target group type according to the fused historical content characteristics through the neural network model;
and adjusting parameters of the neural network model based on the predicted posterior consumption index and the historical posterior consumption index.
In some embodiments, the feature dimensions include a text feature dimension and a visual feature dimension, the target history display content includes at least one type of text information in the text feature dimension and at least one type of visual information in the visual feature dimension, and the extracting, by the neural network model, the history content features of the history content display information corresponding to each feature dimension includes:
fusing at least one type of text information and at least one type of visual information through the neural network model to respectively obtain fused text information and fused visual information;
And respectively extracting the historical content characteristics of the fused text information corresponding to the text characteristic dimension and the historical content characteristics of the fused visual information corresponding to the visual characteristic dimension through the neural network model.
In some embodiments, the extracting, by the neural network model, the historical content features of the historical content display information corresponding to each feature dimension, and fusing the historical content features corresponding to a plurality of feature dimensions, includes:
extracting historical content characteristics of the historical content display information corresponding to each characteristic dimension through a characteristic module corresponding to each characteristic dimension in the neural network model;
the fusion module is used for controlling the feature modules corresponding to the feature dimensions to input the extracted historical content features into the neural network model;
and fusing the historical content characteristics corresponding to the characteristic dimensions through the fusing module.
In some embodiments, the adjusting the parameters of the neural network model includes:
if the number of the training samples is greater than or equal to a preset sample threshold, the parameters of the feature module and the fusion module corresponding to each feature dimension in the neural network model are adjusted by adopting the same parameter adjustment strategy;
If the number of the training samples is smaller than a preset sample threshold, a parameter adjustment strategy corresponding to a first preset super parameter is adopted to adjust parameters of the feature modules corresponding to the feature dimensions in the neural network model, and a parameter adjustment strategy corresponding to a second preset super parameter is adopted to adjust parameters of the fusion modules in the neural network model, wherein the second preset super parameter is larger than the first preset super parameter.
Thus, the electronic device 500 provided in this embodiment may have the following technical effects: and the content recommendation efficiency is improved.
The specific implementation of each operation above may be referred to the previous embodiments, and will not be described herein.
Optionally, as shown in fig. 5, the electronic device 500 further includes: a touch display screen 503, a radio frequency circuit 504, an audio circuit 505, an input unit 506, and a power supply 507. The processor 501 is electrically connected to the touch display 503, the radio frequency circuit 504, the audio circuit 505, the input unit 506, and the power supply 507, respectively. It will be appreciated by those skilled in the art that the electronic device structure shown in fig. 5 is not limiting of the electronic device and may include more or fewer components than shown, or may combine certain components, or a different arrangement of components.
The touch display screen 503 may be used to display a graphical user interface and receive operation instructions generated by a user acting on the graphical user interface. The touch display screen 503 may include a display panel and a touch panel. Wherein the display panel may be used to display information entered by a user or provided to a user as well as various graphical user interfaces of the electronic device, which may be composed of graphics, text, icons, video, and any combination thereof. Alternatively, the display panel may be configured in the form of a liquid crystal display (LCD, liquid Crystal Display), an Organic Light-Emitting Diode (OLED), or the like. The touch panel may be used to collect touch operations on or near the user (such as operations on or near the touch panel by the user using any suitable object or accessory such as a finger, stylus, etc.), and generate corresponding operation instructions, and the operation instructions execute corresponding programs. Alternatively, the touch panel may include two parts, a touch detection device and a touch controller. The touch detection device detects the touch azimuth of a user, detects a signal brought by touch operation and transmits the signal to the touch controller; the touch controller receives touch information from the touch detection device and converts it into touch point coordinates, which are then sent to the processor 501, and can receive commands from the processor 501 and execute them. The touch panel may overlay the display panel, and upon detection of a touch operation thereon or thereabout, the touch panel is passed to the processor 501 to determine the type of touch event, and the processor 501 then provides a corresponding visual output on the display panel based on the type of touch event. In the embodiment of the present application, the touch panel and the display panel may be integrated into the touch display screen 503 to implement the input and output functions. In some embodiments, however, the touch panel and the touch panel may be implemented as two separate components to perform the input and output functions. I.e. the touch sensitive display 503 may also implement an input function as part of the input unit 506.
The radio frequency circuitry 504 may be used to transceive radio frequency signals to establish wireless communication with a network device or other electronic device via wireless communication.
The audio circuitry 505 may be used to provide an audio interface between a user and the electronic device through a speaker, microphone. The audio circuit 505 may transmit the received electrical signal after audio data conversion to a speaker, and convert the electrical signal into a sound signal for output by the speaker; on the other hand, the microphone converts the collected sound signals into electrical signals, which are received by the audio circuit 505 and converted into audio data, which are processed by the audio data output processor 501 for transmission to, for example, another electronic device via the radio frequency circuit 504, or which are output to the memory 502 for further processing. The audio circuit 505 may also include an ear bud jack to provide communication of the peripheral ear bud with the electronic device.
The input unit 506 may be used to receive input numbers, character information, or user characteristic information (e.g., fingerprint, iris, facial information, etc.), and to generate keyboard, mouse, joystick, optical, or trackball signal inputs related to user settings and function control.
The power supply 507 is used to power the various components of the electronic device 500. Alternatively, the power supply 507 may be logically connected to the processor 501 through a power management system, so as to implement functions of managing charging, discharging, and power consumption management through the power management system. The power supply 507 may also include one or more of any components, such as 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.
Although not shown in fig. 5, the electronic device 500 may further include a camera, a sensor, a wireless fidelity module, a bluetooth module, etc., which are not described herein.
In the foregoing embodiments, the descriptions of the embodiments are emphasized, and for parts of one embodiment that are not described in detail, reference may be made to related descriptions of other embodiments.
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 computer programs that can be loaded by a processor to perform steps in any of the content recommendation methods provided by embodiments of the present application. For example, the computer program may perform the steps of:
Acquiring a content consumption form of the content to be recommended, and acquiring content display information of the content to be recommended under at least two characteristic dimensions during display based on the content consumption form;
respectively extracting content features of the content display information corresponding to each feature dimension through a preset neural network model, and fusing the content features corresponding to a plurality of feature dimensions;
predicting posterior consumption indexes of the content to be recommended when the content to be recommended is displayed in at least two types of groups according to the fused content characteristics through the neural network model;
and recommending the content to be recommended in the corresponding type of groups based on the posterior consumption indexes of the content to be recommended in at least two types of groups.
In some embodiments, before extracting the content features of the content display information corresponding to each feature dimension through a preset neural network model, and fusing the content features corresponding to a plurality of feature dimensions, the method further includes:
acquiring the content display times of the history display content when the history display content is displayed on at least one type of group, and determining target history display content with the content display times larger than a preset display threshold value and a target group type corresponding to the target history display content;
Acquiring a history posterior consumption index corresponding to the target history display content;
constructing a training sample based on the target historical display content, the target group type and the historical posterior consumption index;
and training the neural network model based on the training sample.
In some embodiments, the obtaining the historical posterior consumption index corresponding to the target historical display content includes:
acquiring historical posterior consumption parameters corresponding to the target historical display content;
and calculating the ratio between the historical posterior consumption parameter and the content display times of the target historical display content, and determining the ratio as the historical posterior consumption index.
In some embodiments, the target history display content includes history content display information of at least two feature dimensions, and the training the neural network model based on the training sample includes:
inputting the training sample into the neural network model;
respectively extracting historical content characteristics of the historical content display information corresponding to each characteristic dimension through the neural network model, and fusing the historical content characteristics corresponding to a plurality of characteristic dimensions;
Predicting a predicted posterior consumption index of the target historical display content when the target historical display content is displayed on the group of the target group type according to the fused historical content characteristics through the neural network model;
and adjusting parameters of the neural network model based on the predicted posterior consumption index and the historical posterior consumption index.
In some embodiments, the feature dimensions include a text feature dimension and a visual feature dimension, the target history display content includes at least one type of text information in the text feature dimension and at least one type of visual information in the visual feature dimension, and the extracting, by the neural network model, the history content features of the history content display information corresponding to each feature dimension includes:
fusing at least one type of text information and at least one type of visual information through the neural network model to respectively obtain fused text information and fused visual information;
and respectively extracting the historical content characteristics of the fused text information corresponding to the text characteristic dimension and the historical content characteristics of the fused visual information corresponding to the visual characteristic dimension through the neural network model.
In some embodiments, the extracting, by the neural network model, the historical content features of the historical content display information corresponding to each feature dimension, and fusing the historical content features corresponding to a plurality of feature dimensions, includes:
extracting historical content characteristics of the historical content display information corresponding to each characteristic dimension through a characteristic module corresponding to each characteristic dimension in the neural network model;
the fusion module is used for controlling the feature modules corresponding to the feature dimensions to input the extracted historical content features into the neural network model;
and fusing the historical content characteristics corresponding to the characteristic dimensions through the fusing module.
In some embodiments, the adjusting the parameters of the neural network model includes:
if the number of the training samples is greater than or equal to a preset sample threshold, the parameters of the feature module and the fusion module corresponding to each feature dimension in the neural network model are adjusted by adopting the same parameter adjustment strategy;
if the number of the training samples is smaller than a preset sample threshold, a parameter adjustment strategy corresponding to a first preset super parameter is adopted to adjust parameters of the feature modules corresponding to the feature dimensions in the neural network model, and a parameter adjustment strategy corresponding to a second preset super parameter is adopted to adjust parameters of the fusion modules in the neural network model, wherein the second preset super parameter is larger than the first preset super parameter.
It can be seen that the computer program can be loaded by a processor to perform the steps of any of the content recommendation methods provided in the embodiments of the present application, thereby bringing about the following technical effects: and the content recommendation efficiency is improved.
The specific implementation of each operation above may be referred to the previous embodiments, and will not be described herein.
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.
Since the computer program 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 may be achieved are detailed in the previous embodiments and will not be described herein.
The foregoing has described in detail the methods, apparatuses, electronic devices 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 embodiments of the present application, where the foregoing examples are provided to assist in understanding 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.