CN108989397B - Data recommendation method and device and storage medium - Google Patents
Data recommendation method and device and storage medium Download PDFInfo
- Publication number
- CN108989397B CN108989397B CN201810670441.0A CN201810670441A CN108989397B CN 108989397 B CN108989397 B CN 108989397B CN 201810670441 A CN201810670441 A CN 201810670441A CN 108989397 B CN108989397 B CN 108989397B
- Authority
- CN
- China
- Prior art keywords
- data
- model
- recommendation
- deep
- request
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/50—Network services
- H04L67/55—Push-based network services
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/50—Network services
- H04L67/56—Provisioning of proxy services
- H04L67/565—Conversion or adaptation of application format or content
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a data recommendation method, a data recommendation device and a storage medium, wherein the method comprises the following steps: acquiring a data recommendation request of a user, and acquiring initial recommendation data corresponding to the user and characteristic data corresponding to the user according to the data recommendation request; converting the data recommendation request into a preset format request, and acquiring a preset deep FM model from the model server according to the preset format request, wherein the preset format request meets the identification requirement of the model server; generating updating data according to a preset deep FM model and the characteristic data; generating target recommendation data according to the updated data and the initial recommendation data; and pushing the target recommendation data to the user. According to the scheme, the updating data is generated based on the preset deep FM model, the initial recommendation data is updated according to the updating data, the target recommendation data is generated, and the data recommendation efficiency is improved.
Description
Technical Field
The present invention relates to the field of communications technologies, and in particular, to a data recommendation method, an apparatus, and a storage medium.
Background
With the popularization of mobile terminals such as smart phones and tablet computers, users can acquire more and more information through the terminals. In order to enable a user to quickly acquire desired information, information recommendation can be performed according to interests and hobbies of the user.
The conventional recommendation system generally adopts a Logistic Regression (LR) algorithm, which has the advantages of low complexity, easy understanding and realization, but also has limitations. The LR algorithm can only describe low-order features automatically, and in the process of combining high-order features, feature combination needs to be performed manually. And the workload in the feature combination process is large, which results in low recommendation efficiency.
Disclosure of Invention
The embodiment of the invention provides a data recommendation method, a data recommendation device and a storage medium, which can improve the data recommendation efficiency.
The embodiment of the invention provides a data recommendation method, which comprises the following steps: acquiring a data recommendation request of a user, and acquiring initial recommendation data corresponding to the user and feature data corresponding to the user according to the data recommendation request; converting the data recommendation request into a request with a preset format, and acquiring a preset deep FM model from a model server according to the request with the preset format, wherein the request with the preset format meets the identification requirement of the model server; generating updating data according to the preset deep FM model and the characteristic data; generating target recommendation data according to the updating data and the initial recommendation data; and pushing the target recommendation data to the user.
An embodiment of the present invention further provides a data recommendation apparatus, including: the system comprises a first acquisition module, a second acquisition module and a third acquisition module, wherein the first acquisition module is used for acquiring a data recommendation request of a user and acquiring initial recommendation data corresponding to the user and characteristic data corresponding to the user according to the data recommendation request; the conversion module is used for converting the data recommendation request into a request with a preset format and acquiring a preset deep FM model from a model server according to the request with the preset format, wherein the request with the preset format meets the identification requirement of the model server; the first generation module is used for generating updating data according to the preset deep FM model and the characteristic data; the second generation module is used for generating target recommendation data according to the updating data and the initial recommendation data; and the pushing module is used for pushing the target recommendation data to the user.
The embodiment of the invention also provides a storage medium, wherein the storage medium stores processor executable instructions, and the processor provides the data recommendation method by executing the instructions.
According to the data recommendation method, the data recommendation device and the storage medium, the update data are generated according to the preset deep FM model and the characteristic data, the target recommendation data are generated according to the update data and the initial recommendation data, and finally the target recommendation data are pushed to the user, so that the data recommendation efficiency is improved.
Drawings
The technical solution and other advantages of the present invention will become apparent from the following detailed description of specific embodiments of the present invention, which is to be read in connection with the accompanying drawings.
Fig. 1 is a scene schematic diagram of a data recommendation method according to an embodiment of the present invention.
Fig. 2 is a schematic flow chart of a data recommendation method according to an embodiment of the present invention.
Fig. 3 is another flowchart of a data recommendation method according to an embodiment of the present invention.
Fig. 4 is a schematic structural diagram of a data recommendation device according to an embodiment of the present invention.
Fig. 5 is a schematic structural diagram of a generating module according to an embodiment of the present invention.
Fig. 6 is a schematic structural diagram of a conversion module according to an embodiment of the present invention.
Fig. 7 is a schematic structural diagram of a first generation module according to an embodiment of the present invention.
Fig. 8 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1, the figure is a schematic view of a scenario of a data recommendation method according to an embodiment of the present invention, in the scenario, a data recommendation device may be implemented as an entity, or may be implemented by being integrated in an electronic device such as a terminal or a server, where the electronic device may include a smart phone, a tablet computer, a personal computer, and the like.
As shown in fig. 1, the scenario may include a terminal a, a server b, a server c, a server d, and a server e. The server c is used for storing the preset deep FM model, the server d is used for storing the characteristic data, and the server e is used for storing the initial recommendation data. Specifically, the user a generates a data recommendation request by operating the terminal a, and the terminal a sends the data recommendation request to the server b. After the server b obtains the data recommendation request, feature data related to the user A are sequentially obtained from the server d, and initial recommendation data related to the user A are obtained from the server e. Further, the server b converts the data recommendation request into a request in a preset format so as to meet the identification requirement of the server c. And then sending the request with the preset format to a server c, and acquiring a preset deep FM model from the server c. Then, the server b generates update data according to the preset deep fm model and the feature data related to the user a. And generating target recommendation data according to the update data and the initial recommendation data related to the user A. And finally, sending the target recommendation data to a terminal a, and displaying the target recommendation data to the user A through the terminal a.
Embodiments of the present invention provide a data recommendation method, an apparatus and a storage medium, which will be described in detail below.
In the embodiments of the present invention, description will be made from the perspective of a data recommendation apparatus, which may be specifically integrated in an electronic device.
A data recommendation method comprises the steps of obtaining a data recommendation request of a user, and obtaining initial recommendation data corresponding to the user and feature data corresponding to the user according to the data recommendation request; converting the data recommendation request into a preset format request, and acquiring a preset deep FM model from the model server according to the preset format request, wherein the preset format request meets the identification requirement of the model server; generating updating data according to a preset deep FM model and the characteristic data; generating target recommendation data according to the updated data and the initial recommendation data; and pushing the target recommendation data to the user.
Referring to fig. 2, fig. 2 is a flowchart of a data recommendation method according to an embodiment of the present invention, where the method includes:
step S101, acquiring a data recommendation request of a user, and acquiring initial recommendation data corresponding to the user and feature data corresponding to the user according to the data recommendation request.
When the user performs relevant operations on the terminal, a data recommendation request is generated. Wherein the operation comprises an update operation, a query operation, and the like. For example, when a user listens to a song with a mobile phone, a "next to next" button is clicked on a mobile phone interface, a data recommendation request is triggered to be generated, then the terminal acquires the data recommendation request, and according to the data recommendation request, initial recommendation data corresponding to the user and feature data corresponding to the user are acquired to analyze user preferences so as to recommend a proper song to the user.
The initial recommendation data refers to pre-stored recommendation data such as a song list, a video list, etc. A storage area may be provided to store the initial recommendation data, the user identification and the association between the two. After the data recommendation request of the user is obtained, the user identification matched with the user can be searched from the storage area, and then the initial recommendation data corresponding to the user identification is obtained.
The feature data may be extracted from the user information by collecting the user information. Specifically, the attribute information, the operation information, and the context information corresponding to the operation information of the user may be selected from the user information to extract the feature data.
The attribute information of the user refers to biological characteristics of the person under the action of natural laws and social attributes under the action of social environments and conditions, and the attribute information represents inherent properties and characteristics of the person. Specifically, the demographic attributes include information such as the gender, age, and location of the user. The operation information refers to information generated in the interaction process between the user and the terminal, and specifically includes: operation type, operation time information, and operation object, and the like. For example, the operation object may be audio, video, text, and the like, the operation type for the audio includes deleting the audio, collecting the audio, downloading the audio, and the like, and the operation time information for the audio includes time for deleting the audio, duration for collecting the audio, time for downloading the audio, and the like. The context information corresponding to the operation information refers to information associated with the operation information, such as an operating system of the terminal, an entry when listening to audio, and the like. For example, when a song is listened to, the song name can be input in the search box, and the song can also be input from the favorite song.
Step S102, converting the data recommendation request into a request with a preset format, and acquiring a preset deep FM model from the model server according to the request with the preset format, wherein the request with the preset format meets the identification requirement of the model server.
The preset deep FM model is generated by training the first deep FM model by adopting a data set. The first DeepFM model has parameters, and parameter values of the parameters can be obtained in the training process. And substituting the parameter value into the first DeepFM model to obtain a preset DeepFM model.
It should be noted that the first Deep FM model is generated by combining an FM (factor decomposition Machine) model and a DNN (Deep Neural Networks) model. The FM model is a matrix decomposition-based machine learning algorithm, which transforms a factorization model using eigenvectors, and expresses the inner product between low-order relations as an interactive relation between factors. The FM model has the advantage of automatically combining features, which can greatly reduce the work of feature engineering. In this embodiment, the first deep FM model may combine advantages of the DNN model and the FM model, and may automatically generate a low-order feature and a high-order feature, thereby improving processing efficiency in the aspect of feature engineering.
The preset deep FM model is arranged on a model server, such as a TFServing server. Specifically, the preset deep fm model may be configured in a directory specified by the TFServing server, so that the TFServing server loads the preset deep fm model. When a service is requested from a model server such as a TFServing server, the model server may not recognize a data recommendation request due to different supported protocols of different model servers. Specifically, assuming that the model server only recognizes the request in the gRPC format, when the data recommendation request is in the JSON (javascript Object notification) format, the model server may not accurately recognize the data recommendation request in the JSON format, and may not provide the preset deep fm model information in time. In summary, format conversion needs to be performed on the data recommendation request to meet the identification requirement of the model server, and then the preset deep fm model is obtained from the model server according to the request with the preset format.
And S103, generating updating data according to the preset deep FM model and the characteristic data.
Specifically, the FM model in the preset deep FM model may be first adopted to vectorize the feature data, and then the vectorized feature data is used as an input and input into the DNN model to generate the first output result. And inputting the characteristic data into an FM model to generate a second output result. And finally, taking the first output result and the second output result as the input of the sigmoid function to obtain the updated data.
And step S104, generating target recommendation data according to the updated data and the initial recommendation data.
In some embodiments, the update data may be used to update the initial recommendation data to generate the target recommendation data. It is assumed that the initial recommendation data comprises a song recommendation list and the update data is also a song recommendation list. A song that coincides in the initial recommendation data and the update data may be found and set as the target recommendation data.
In some embodiments, due to the limitation of the computing capability, after the data recommendation request of the user is obtained, the data recommendation request may not be responded in real time, that is, the updated data may not be generated in time due to too long computing time. Under the condition that the generated updating data is empty, the initial recommendation data can be directly used as the target recommendation data, so that the situation that a user waits for data push for too long time is avoided, and the calculation pressure is reduced.
And step S105, pushing the target recommendation data to the user.
Specifically, the purpose of pushing the target recommendation data to the user can be achieved by displaying the target recommendation data on a terminal interface.
As can be seen from the above, in the data recommendation method of the embodiment of the invention, the update data is generated according to the preset deep fm model and the feature data, the target recommendation data is generated according to the update data and the initial recommendation data, and the target recommendation data is pushed to the user, so that the data recommendation efficiency is improved. .
The data recommendation method described in the above embodiments is further illustrated below by way of example. In the embodiments of the present invention, description will be made from the perspective of a data recommendation apparatus, which may be specifically integrated in an electronic device.
Referring to fig. 3, fig. 3 is another flowchart of a data recommendation method according to an embodiment of the present invention, where the method includes:
step S201, acquiring historical feature data of the user in real time.
With the development of the internet, the phenomenon of information overload is increasingly serious. The recommendation method based on machine learning can provide users with suitable choices of commodities, music, movies and the like, and therefore, the recommendation method is widely applied. The key to machine learning is the selection of models and feature data. In the present embodiment, the history feature data may be extracted from the user information by collecting the user information. Specifically, the historical feature data used for model training may be extracted by selecting information such as attribute information of the user, historical operation information, and context information corresponding to the historical operation information from the user information.
The attribute information of the user refers to biological characteristics of the person under the action of natural laws and social attributes under the action of social environments and conditions, and the attribute information represents inherent properties and characteristics of the person. Specifically, the demographic attributes include information such as the gender, age, and location of the user. The historical operation information refers to information generated in the interaction process between the user and the terminal, and specifically includes: operation type, operation time information, and operation object, and the like. For example, the operation object may be audio, video, text, and the like, the operation type for the audio includes deleting the audio, collecting the audio, downloading the audio, and the like, and the operation time information for the audio includes time for deleting the audio, duration for collecting the audio, time for downloading the audio, and the like. The context information corresponding to the historical operation information refers to information associated with the historical operation information, such as an operating system of the terminal, an entry when listening to audio, and the like. For example, when a song is listened to, the song name can be input in the search box, and the song can also be input from the favorite song.
It should be noted that, because the user and the terminal are in a continuous interaction process, and accordingly, the user information is also continuously updated, the historical feature data needs to be acquired from the user information in real time. In some embodiments, the historical characteristic data may be stored in an offline storage manner. This allows the first deep fm model to be trained offline using historical signature data to reduce network overhead, as detailed in step S202 below.
Step S202, training the first deep FM model by using the historical characteristic data to generate a training result.
The first Deep FM model is generated by combining an FM (factor decomposition Machine) model and a DNN (Deep Neural Networks) model. The FM model is a matrix decomposition-based machine learning algorithm, which transforms a factorization model using eigenvectors, and expresses the inner product between low-order relations as an interactive relation between factors. The FM model has the advantage of automatic feature combination, and can greatly reduce the work of feature engineering. The DNN model may take into account both high-order and low-order feature combinations. In this embodiment, the first deep FM model is adopted, and the advantages of the DNN model and the FM model can be combined, so that the low-order feature and the high-order feature can be automatically generated, and the processing efficiency in the aspect of feature engineering is improved.
The first DeepFM model has parameters, and the parameter values corresponding to the parameters in the first DeepFM model are obtained by training the first DeepFM model on the TensorFlow cluster by using the historical characteristic data, so that the training result is obtained.
As can be seen from step S201, since the historical feature data of the user is changing constantly, a plurality of training results can also be obtained by training the first deep fm model through the historical feature data acquired in real time in step S201.
Step S203, generating a plurality of second DeepFM models according to the plurality of training results and the first DeepFM model.
Specifically, the training result may be used as a parameter value, and the parameter value is substituted into the first deep fm model, so as to obtain a second deep fm model. It should be noted that, because there are a plurality of training results, a plurality of second deep fm models are also generated accordingly.
And step S204, acquiring version information of a plurality of second deep FM models.
In step S203, the training results are different, and the correspondingly generated second deep fm models are also different. And in the plurality of second DeepFM models, the target recommendation data generated according to the second DeepFM model generated recently is higher in accuracy.
In summary, in order to improve the accuracy of data recommendation, it is necessary to obtain version information of the plurality of second deep fm models, and select the second deep fm model of the latest version from the plurality of second deep fm models. In advance, the plurality of second deep fm models may be version-divided according to the time information generated by the second deep fm models. The shorter the generation time of the second preset deep FM model is, the higher the version of the second preset deep FM model is, and the longer the generation time is, the lower the version of the second preset deep FM model is.
In particular, the time information includes a time stamp, wherein the time stamp is a complete and verifiable piece of data, usually a sequence of characters, that represents a piece of data that existed before a particular time to uniquely identify the time of a moment. Suppose that a specific time is 1/1970, the generation time of the second deep FM model is 10/2018, and the corresponding time stamp is 1523325885 seconds between the generation time of the second deep FM model and the specific time. In summary, 1523325885 can be used as the version number of the second deep FM model to uniquely identify the second deep FM model. In summary, the version information may be obtained by obtaining version numbers of a plurality of second deep fm models.
Step S205, according to the version information, selecting a preset DeepFM model from the plurality of second DeepFM models, and setting the preset DeepFM model on the TFServing server.
Specifically, in the adapting step S204, the second deep fm model with the largest version number may be selected from the plurality of second deep dm models as the preset deep fm model. And then configuring the preset DeepFM model to a directory specified by the TFserving server so as to load the preset DeepFM model by the TFserving server.
It should be noted that, because the TFServing server is a CPU-intensive service, in order to optimize resource allocation on the TFServing server, the TFServing server may designate a directory at intervals, query whether a preset deep fm model exists, and if so, load the preset deep fm model.
Step S206, acquiring a data recommendation request of the user, and acquiring initial recommendation data corresponding to the user and characteristic data corresponding to the user according to the data recommendation request.
The initial recommendation data refers to pre-stored recommendation data, such as a song list, a video list, and the like. A storage area may be provided to store the initial recommendation data, the user identification and the association between the two. After the data recommendation request of the user is obtained, the user identification matched with the user can be searched from the storage area, and then the initial recommendation data corresponding to the user identification is obtained.
In some embodiments, the feature data corresponding to the user is stored on-line. Specifically, a database such as myslq or oracle may be used to store the feature data. Preferably, the CMEM (Cloud Memcache, memory persistent storage) can be used for storing, and the CMEM stores the feature data in a Key-Value form, so that the reading of the feature data can be accelerated.
Step S207, adopting Proxy service as middleware service, and converting the data recommendation request into a request in a gRPC format; and acquiring a preset DeepFM model from the TFserving server through a request in a gRPC format.
After the preset DeepFM model is set on the TFServing server, if a service is requested from the TFServing server, a gRPC protocol in Proto3 is required to access the Predict function. Among them, gRPC is a high-performance, universal, open-source RPC framework. Due to the limited languages supported by the Proto3 protocol, it may happen that the gRPC protocol cannot be used during actual operation. Therefore, in this embodiment, Proxy service is adopted as middleware to convert the gRPC protocol into a general protocol.
To sum up, the step of obtaining the preset deep fm model from the TFServing server according to the data recommendation request specifically includes:
assuming that the data recommendation request is in a JSON (JavaScript Object Notification) format, after the data recommendation request is sent to a Proxy service in a SOKCET message form, the Proxy service analyzes the data recommendation request in the JSON format and encapsulates the data recommendation request into a request in a gRPC format. Then, the Proxy service sends the request in the gPC format to the TFserving server, receives the relevant information of the preset deep FM model returned by the TFserving server, encapsulates the relevant information of the preset deep FM model into an SOKCET message, and returns the SOKCET message, so that the preset deep FM model can be obtained. In some embodiments, the Proxy service and the TFServing service may be located on the same server to reduce the latency of the data recommendation request.
It should be noted that, because the TFServing server operates in multiple threads, two operations of receiving a data recommendation request and loading a preset deep fm model can be performed simultaneously without mutual interruption.
And S208, generating updating data according to the preset deep FM model and the characteristic data.
Because the preset deep FM model is the combination of the FM model and the DNN model, the FM model, the DNN model and the characteristic data can be combined to generate the updating data, and the specific steps are as follows:
(A) and vectorizing the characteristic data by adopting an FM model.
(B) And inputting the vectorized feature data into the DNN model to generate a first output result.
(C) And generating updating data according to the first output result and a second output result corresponding to the FM model.
Specifically, after the first output result and the second output result are obtained, the sigmoid function may be combined to generate the update data. Assuming that the first output result is Y1 and the second output result is Y2, the update data Y is sigmoid (w 1Y 1+ w 2Y 2+ bias), where w1, w2, and bias are parameters whose values can be generated by training the first deep fm model using the historical feature data, as shown in step S202.
In step S209, target recommendation data is generated based on the update data and the initial recommendation data.
In some embodiments, the update data may be used to update the initial recommendation data to generate the target recommendation data. It is assumed that the initial recommendation data comprises a song recommendation list and the update data is also a song recommendation list. A song that coincides in the initial recommendation data and the update data may be found and set as the target recommendation data.
In some embodiments, due to the limitation of the computing capability, after the data recommendation request of the user is obtained, the data recommendation request may not be responded in real time, that is, the updated data may not be generated in time due to too long computing time. Under the condition that the generated updating data is empty, the initial recommendation data can be directly used as the target recommendation data, so that the situation that a user waits for data push for too long time is avoided, and the calculation pressure is reduced.
And step S210, pushing the target recommendation data to the user.
Specifically, the purpose of pushing the target recommendation data to the user can be achieved by displaying the target recommendation data on a terminal interface.
According to the data recommendation method, the update data are generated according to the preset deep FM model and the characteristic data, the target recommendation data are generated according to the update data and the initial recommendation data, and the target recommendation data are pushed to the user, so that the data recommendation efficiency is improved.
The present embodiment will be further described from the perspective of a data recommendation device that may be integrated in an electronic device, according to the methods described in the above embodiments
Referring to fig. 4, fig. 4 is a structural diagram of a data recommendation apparatus according to an embodiment of the present invention, and the apparatus 30 may include a first obtaining module 301, a converting module 302, a first generating module 303, a second generating module 304, and a pushing module 305.
(1) First acquisition module 301
The first obtaining module 301 is configured to obtain a data recommendation request of a user, and obtain a preset deep fm model, initial recommendation data corresponding to the user, and feature data corresponding to the user according to the data recommendation request.
When the user performs relevant operations on the terminal, a data recommendation request is generated. Wherein the operation comprises an update operation, a query operation, and the like. For example, when a user listens to a song with a mobile phone, and clicks a "next" button on a mobile phone interface, the data recommendation request is generated by triggering, then the first obtaining module 301 obtains the data recommendation request, and obtains the preset deep fm model, the initial recommendation data corresponding to the user, and the feature data corresponding to the user according to the data recommendation request, so as to analyze the preference of the user, and recommend a proper song to the user.
The initial recommendation data refers to pre-stored recommendation data, such as a song list, a video list, and the like. A storage area may be provided to store the initial recommendation data, the user identification and the association between the two. After the first obtaining module 301 obtains the data recommendation request of the user, it may search the user identifier matching with the user from the storage area, and then obtain the initial recommendation data corresponding to the user identifier.
The feature data may be extracted from the user information by collecting the user information. Specifically, the first obtaining module 301 may select information, such as attribute information and operation information of the user and context information corresponding to the operation information, from the user information to extract the feature data.
In some embodiments, the feature data corresponding to the user is stored on-line. Specifically, a database such as myslq or oracle may be used to store the feature data. Preferably, the CMEM (Cloud Memcache, memory persistent storage) may also be used for storing, and the CMEM stores the feature data in a Key-Value form, so that the first obtaining module 301 can be accelerated to read the feature data.
The preset deep FM model is generated by training the first deep FM model by adopting a data set. The first DeepFM model has parameters, and parameter values of the parameters can be obtained in the training process. And substituting the parameter value into the first DeepFM model to obtain a preset DeepFM model.
It should be noted that the first Deep FM model is generated by combining an FM (factor decomposition Machine) model and a DNN (Deep Neural Networks) model. The FM model is a matrix decomposition-based machine learning algorithm, which transforms a factorization model using eigenvectors, and expresses the inner product between low-order relations as an interactive relation between factors. The FM model has the advantage of automatically combining features, which can greatly reduce the work of feature engineering. In this embodiment, the first deep FM model may combine advantages of the DNN model and the FM model, and may automatically generate a low-order feature and a high-order feature, thereby improving processing efficiency in the aspect of feature engineering.
The generation process of the preset deep fm model is described in detail as follows:
in some embodiments, as shown in fig. 4, the apparatus 30 further comprises: a second acquisition module 306, a training module 307, and a generation module 308.
The second obtaining module 306 is used for obtaining the historical feature data of the user in real time. In some embodiments, the second obtaining module 306 may extract historical feature data from the user information by collecting the user information. Specifically, the second obtaining module 306 may select information, such as attribute information of the user, historical operation information, and context information corresponding to the historical operation information, from the user information to extract historical feature data for model training.
The attribute information of the user refers to biological characteristics of the person under the action of natural laws and social attributes under the action of social environments and conditions, and the attribute information represents inherent properties and characteristics of the person. Specifically, the demographic attributes include information such as the gender, age, and location of the user. The historical operation information refers to information generated in the interaction process between the user and the terminal, and specifically includes: operation type, operation time information, and operation object, and the like. For example, the operation object may be audio, video, text, and the like, the operation type for the audio includes deleting the audio, collecting the audio, downloading the audio, and the like, and the operation time information for the audio includes time for deleting the audio, duration for collecting the audio, time for downloading the audio, and the like. The context information corresponding to the historical operation information refers to information associated with the historical operation information, such as an operating system of the terminal, an entry when listening to audio, and the like. For example, when a song is listened to, the song name can be input in the search box, and the song can also be input from the favorite song.
It should be noted that, since the user and the terminal are in a continuous interaction process, and accordingly, the user information is also continuously updated, the second obtaining module 306 is required to obtain the historical feature data from the user information in real time. In some embodiments, the historical characteristic data may be stored in an offline storage manner. In this way, the training module 307 may train the first deep fm model using the historical feature data in an offline situation to reduce network overhead.
The training module 307 is described in detail below. The training module 307 is configured to train the first deep fm model using the historical feature data to generate a plurality of training results. The first deep fm model has parameters, and the training module 307 trains the first deep fm model on the TensorFlow cluster by using the historical feature data to obtain parameter values corresponding to the parameters in the first deep fm model, so as to obtain a training result.
The generating module 308 is configured to generate a preset deep fm model according to the plurality of training results and the first deep fm model. In some embodiments, as shown in fig. 5, the generation module 308 comprises: a generating submodule 3081, a first retrieving submodule 3082 and a selecting submodule 3083.
The generating submodule 3081 is configured to generate a plurality of second deep fm models according to the plurality of training results and the first deep fm model. Specifically, the generation submodule 3081 may substitute the training result as a parameter value into the first DeepFM model to obtain a second DeepFM model. It should be noted that, because there are a plurality of training results, a plurality of second deep fm models are also generated accordingly.
The first obtaining sub-module 3082 is configured to obtain version information of the plurality of second deep fm models. The training results are different, and the second deep fm model generated by the generation submodule 3081 is also different. And in the plurality of second DeepFM models, the target recommendation data generated according to the second DeepFM model generated recently is higher in accuracy.
In summary, in order to improve the accuracy of data recommendation, it is necessary to obtain version information of the plurality of second deep fm models, and select the second deep fm model of the latest version from the plurality of second deep fm models. In advance, the plurality of second deep fm models may be version-divided according to the time information generated by the second deep fm models. The shorter the generation time of the second preset deep FM model is, the higher the version of the second preset deep FM model is, and the longer the generation time is, the lower the version of the second preset deep FM model is.
In particular, the time information includes a time stamp, wherein the time stamp is a complete and verifiable piece of data, usually a sequence of characters, that represents a piece of data that existed before a particular time to uniquely identify the time of a moment. Suppose that a specific time is 1/1970, the generation time of the second deep FM model is 10/2018, and the corresponding time stamp is 1523325885 seconds between the generation time of the second deep FM model and the specific time. In summary, 1523325885 can be used as the version number of the second deep FM model to uniquely identify the second deep FM model. In summary, the first obtaining sub-module 3082 may obtain the version information by obtaining version numbers of the plurality of second deep fm models.
The selecting submodule 3083 is configured to select a preset deep fm model from the plurality of second deep fm models according to the version information. Specifically, the selecting sub-module 3083 may select the second deep fm model with the largest version number from the plurality of second deep dm models as the preset deep fm model.
(2) Conversion module 302
In some embodiments, as shown in FIG. 6, the conversion module 302 includes a conversion sub-module 3021 and a second acquisition sub-module 3022. A conversion submodule 3021, configured to use Proxy service as middleware service, and convert the data recommendation request into a request in a gRPC format; the second obtaining sub-module 3022 is configured to obtain the preset deep fm model from the TFServing server through a request in the gRPC format.
Specifically, assuming that the data recommendation request is a request in the JSON format, after the conversion sub-module 3021 sends the data recommendation request to the Proxy service in the form of an SOKCET message, the Proxy service parses the data recommendation request in the JSON format and encapsulates the data recommendation request into a request in the gRPC format. Then, the Proxy service sends the request in the gRPC format to the TFServing server, receives the relevant information of the preset deep fm model returned by the TFServing server, and encapsulates the relevant information of the preset deep fm model into an SOKCET message to be returned, that is, the second obtaining sub-module 3022 may obtain the preset deep fm model. In some embodiments, the Proxy service and the TFServing service may be located on the same server to reduce the latency of the data recommendation request.
It should be noted that, because the TFServing server operates in multiple threads, two operations of receiving a data recommendation request and loading a preset deep fm model can be performed simultaneously without mutual interruption.
It should be noted that, because the TFServing server is a CPU-intensive service, in order to optimize resource allocation on the TFServing server, the TFServing server may designate a directory at intervals, query whether a preset deep fm model exists, and if so, load the preset deep fm model.
(3) First generation module 303
The first generating module 303 is configured to generate update data according to a preset deep fm model and the feature data. The preset deep FM model includes an FM model and a DNN model, and as shown in fig. 7, the first generating module 303 includes: a vectorization submodule 3021, an input submodule 3022 and an update submodule 3023.
A vectorization submodule 3021 configured to vectorize the feature data using an FM model; the input submodule 3022 is configured to input the vectorized feature data into the DNN model, and generate a first output result; the updating submodule 3023 is configured to generate updating data according to the first output result and the second output result corresponding to the FM model.
Specifically, after the first output result and the second output result are obtained, the update submodule 3023 may combine the sigmoid function to generate the update data. Assuming that the first output result is Y1 and the second output result is Y2, the update data Y is sigmoid (w 1Y 1+ w 2Y 2+ bias), where w1, w2, and bias are parameters whose values can be generated by training the first deep fm model using the historical feature data.
(4) Second generation module 304
And a second generating module 304, configured to generate target recommendation data according to the update data and the initial recommendation data. In some embodiments, the second generation module 304 may use the update data to update the initial recommendation data to generate the target recommendation data. It is assumed that the initial recommendation data comprises a song recommendation list and the update data is also a song recommendation list. The second generation module 304 may find a song that is coincident in the initial recommendation data and the update data and set the coincident song as the target recommendation data.
In some embodiments, due to the limitation of the computing capability, after the data recommendation request of the user is obtained, the data recommendation request may not be responded in real time, that is, the updated data may not be generated in time due to too long computing time. Under the condition that the generated update data is empty, the second generation module 304 may directly use the initial recommendation data as the target recommendation data, which not only avoids pushing the data for a long time by the user, but also reduces the calculation pressure.
(4) Push module 305
And the pushing module 305 is configured to push the target recommendation data to the user. Specifically, the pushing module 305 may display the target recommendation data on a terminal interface, so as to achieve the purpose of pushing the target recommendation data to the user.
According to the data recommendation device provided by the embodiment of the invention, the updated data is generated according to the preset deep FM model and the characteristic data, the target recommendation data is generated according to the updated data and the initial recommendation data, and finally the target recommendation data is pushed to the user, so that the data recommendation efficiency is improved.
Accordingly, an embodiment of the present invention further provides an electronic device, as shown in fig. 8, which shows a schematic structural diagram of the electronic device according to the embodiment of the present invention, specifically:
the electronic device may include components such as a processor 401 of one or more processing cores, memory 402 of one or more computer-readable storage media, a power supply 403, and an input unit 404. Those skilled in the art will appreciate that the electronic device configuration shown in fig. 8 does not constitute a limitation of the electronic device and may include more or fewer components than those shown, or some components may be combined, or a different arrangement of components. Wherein:
the processor 401 is a control center of the electronic device, connects various parts of the whole electronic device by various interfaces and lines, performs various functions of the electronic device and processes data by running or executing software programs and/or modules stored in the memory 402 and calling data stored in the memory 402, thereby performing overall monitoring of the electronic device. Optionally, processor 401 may include one or more processing cores; preferably, the processor 401 may integrate an application processor, which mainly handles operating systems, user interfaces, application programs, etc., and a modem processor, which mainly handles wireless communications. 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 functional applications and data processing by operating the software programs and modules stored in the memory 402. The memory 402 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program required by at least one function (such as a sound playing function, an image playing function, etc.), and the like; the storage data area may store data created according to use of the electronic device, and the like. Further, the 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 access to the memory 402.
The electronic device further comprises a power supply 403 for supplying power to the various components, and preferably, the power supply 403 is logically connected to the processor 401 through a power management system, so that functions of managing charging, discharging, and power consumption are realized through the power management system. The power supply 403 may also include any component of one or more dc or ac power sources, recharging systems, power failure detection circuitry, power converters or inverters, power status indicators, and the like.
The electronic device may further include an input unit 404, and the input unit 404 may be used to receive input numeric or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.
Although not shown, the electronic device may further include a display unit and the like, which are not described in detail herein. Specifically, in this embodiment, the processor 401 in the electronic device loads the executable file corresponding to the process of one or more application programs into the memory 402 according to the following instructions, and the processor 401 runs the application program stored in the memory 402, thereby implementing various functions as follows:
acquiring a data recommendation request of a user, and acquiring initial recommendation data corresponding to the user and characteristic data corresponding to the user according to the data recommendation request;
converting the data recommendation request into a preset format request, and acquiring a preset deep FM model from the model server according to the preset format request, wherein the preset format request meets the identification requirement of the model server;
generating updating data according to a preset deep FM model and the characteristic data;
generating target recommendation data according to the updated data and the initial recommendation data;
and pushing the target recommendation data to the user.
The electronic device can achieve the effective effect that any one of the data recommendation devices provided by the embodiments of the present invention can achieve, which is detailed in the foregoing embodiments and will not be described herein again.
According to the electronic equipment provided by the embodiment of the invention, the updated data is generated according to the preset deep FM model and the characteristic data, the target recommended data is generated according to the updated data and the initial recommended data, and finally the target recommended data is pushed to the user, so that the data recommendation efficiency is improved.
Various operations of embodiments are provided herein. In one embodiment, the one or more operations may constitute computer readable instructions stored on one or more computer readable media, which when executed by an electronic device, will cause the computing device to perform the operations. The order in which some or all of the operations are described should not be construed as to imply that these operations are necessarily order dependent. Those skilled in the art will appreciate alternative orderings having the benefit of this description. Moreover, it should be understood that not all operations are necessarily present in each embodiment provided herein.
Also, although the disclosure has been shown and described with respect to one or more implementations, equivalent alterations and modifications will occur to others skilled in the art based upon a reading and understanding of this specification and the annexed drawings. The present disclosure includes all such modifications and alterations, and is limited only by the scope of the appended claims. In particular regard to the various functions performed by the above described components (e.g., elements, resources, etc.), the terms used to describe such components are intended to correspond, unless otherwise indicated, to any component which performs the specified function of the described component (e.g., that is functionally equivalent), even though not structurally equivalent to the disclosed structure which performs the function in the herein illustrated exemplary implementations of the disclosure. In addition, while a particular feature of the disclosure may have been disclosed with respect to only one of several implementations, such feature may be combined with one or more other features of the other implementations as may be desired and advantageous for a given or particular application. Furthermore, to the extent that the terms "includes," has, "" contains, "or variants thereof are used in either the detailed description or the claims, such terms are intended to be inclusive in a manner similar to the term" comprising.
Each functional unit in the embodiments of the present invention may be integrated into one processing module, or each unit may exist alone physically, or two or more units are integrated into one module. The integrated module can be realized in a hardware mode, and can also be realized in a software functional module mode. The integrated module, if implemented in the form of a software functional module and sold or used as a stand-alone product, may also be stored in a computer readable storage medium. The storage medium mentioned above may be a read-only memory, a magnetic or optical disk, etc. Each apparatus or system described above may perform the method in the corresponding method embodiment.
In summary, although the present invention has been disclosed in the foregoing embodiments, the serial numbers before the embodiments are used for convenience of description only, and the sequence of the embodiments of the present invention is not limited. Furthermore, the above embodiments are not intended to limit the present invention, and those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, therefore, the scope of the present invention shall be limited by the appended claims.
Claims (11)
1. A method for recommending data, comprising:
acquiring a data recommendation request of a user, and acquiring initial recommendation data corresponding to the user and feature data corresponding to the user according to the data recommendation request;
converting the data recommendation request into a request with a preset format, and acquiring a preset deep FM model from a model server according to the request with the preset format, wherein the request with the preset format meets the identification requirement of the model server, and the preset deep FM model is a model which corresponds to the user and is trained;
generating updating data according to the preset deep FM model and the characteristic data;
generating target recommendation data according to the updating data and the initial recommendation data;
and pushing the target recommendation data to the user.
2. The data recommendation method according to claim 1, wherein the step of converting the data recommendation request into a request with a preset format and obtaining a preset deep fm model from a model server according to the request with the preset format, wherein before the step of satisfying the identification requirement of the model server, the method further comprises:
acquiring historical characteristic data of the user in real time;
training a first deep FM model by using the historical characteristic data to generate a plurality of training results;
and generating the preset deep FM model according to the training results and the first deep FM model, and setting the preset deep FM model on the model server.
3. The data recommendation method according to claim 2, wherein the step of generating the preset deep fm model according to the training result and the deep fm model and setting the preset deep fm model on the model server comprises:
generating a plurality of second deep FM models according to the training results and the first deep FM model;
acquiring version information of the plurality of second deep FM models;
and selecting the preset deep FM model from the plurality of second deep FM models according to the version information.
4. The data recommendation method according to any one of claims 1 to3, wherein the model server comprises a TFServing server, the step of converting the data recommendation request into a request with a preset format and obtaining a preset deep fm model from the model server according to the request with the preset format, wherein the step of meeting the identification requirement of the model server by the request with the preset format comprises:
adopting Proxy service as middleware service, and converting the data recommendation request into a request in a gRPC format;
and acquiring the preset deep FM model from the TFserving server through the request in the gRPC format.
5. The data recommendation method according to any one of claims 1-3, wherein the preset deep FM model comprises an FM model and a DNN model, and the step of generating the update data according to the preset deep FM model and the characteristic data comprises:
vectorizing the feature data by adopting the FM model;
inputting the vectorized feature data into the DNN model to generate a first output result;
and generating the updating data according to the first output result and a second output result corresponding to the FM model.
6. A data recommendation device, comprising:
the system comprises a first acquisition module, a second acquisition module and a third acquisition module, wherein the first acquisition module is used for acquiring a data recommendation request of a user and acquiring initial recommendation data corresponding to the user and characteristic data corresponding to the user according to the data recommendation request;
the conversion module is used for converting the data recommendation request into a request with a preset format and acquiring a preset deep FM model from a model server according to the request with the preset format, wherein the request with the preset format meets the identification requirement of the model server, and the preset deep FM model is a model which corresponds to the user and is trained;
the first generation module is used for generating updating data according to the preset deep FM model and the characteristic data;
the second generation module is used for generating target recommendation data according to the updating data and the initial recommendation data;
and the pushing module is used for pushing the target recommendation data to the user.
7. The data recommendation device of claim 6, further comprising:
the second acquisition module is used for acquiring historical characteristic data of the user in real time;
the training module is used for training the first deep FM model by using the historical characteristic data to generate a plurality of training results;
and the generating module is used for generating the preset deep FM model according to the training results and the first deep FM model and setting the preset deep FM model on the model server.
8. The data recommendation device of claim 7, wherein the generation module comprises:
the generation submodule is used for generating a plurality of second deep FM models according to the training results and the first deep FM model;
the first obtaining sub-module is used for obtaining the version information of the second deep FM models;
and the selection submodule is used for selecting the preset deep FM model from the plurality of second deep FM models according to the version information.
9. The data recommendation device of any of claims 6-8, wherein the conversion module comprises:
the conversion submodule is used for converting the data recommendation request into a request in a gRPC format by adopting Proxy service as middleware service;
and the second obtaining submodule is used for obtaining the preset deep FM model from a TFServing server through the request in the gRPC format.
10. The data recommendation device according to any one of claims 6-8, wherein the preset deep FM model comprises an FM model and a DNN model, and the first generation module comprises:
the vectorization submodule is used for vectorizing the feature data by adopting the FM model;
the input submodule is used for inputting the vectorized characteristic data into the DNN model and generating a first output result;
and the updating submodule is used for generating the updating data according to the first output result and a second output result corresponding to the FM model.
11. A storage medium having stored therein processor-executable instructions, the processor providing the data recommendation method of any one of claims 1-5 by executing the instructions.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810670441.0A CN108989397B (en) | 2018-06-26 | 2018-06-26 | Data recommendation method and device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810670441.0A CN108989397B (en) | 2018-06-26 | 2018-06-26 | Data recommendation method and device and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108989397A CN108989397A (en) | 2018-12-11 |
| CN108989397B true CN108989397B (en) | 2021-04-20 |
Family
ID=64538770
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810670441.0A Active CN108989397B (en) | 2018-06-26 | 2018-06-26 | Data recommendation method and device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108989397B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111368150B (en) * | 2018-12-25 | 2024-04-02 | 北京奇虎科技有限公司 | Training method and device of prediction model and electronic equipment |
| CN109902224A (en) * | 2019-01-17 | 2019-06-18 | 平安城市建设科技(深圳)有限公司 | Source of houses recommended method, device, equipment and medium based on user behavior analysis |
| CN110166522B (en) * | 2019-04-01 | 2021-08-24 | 腾讯科技（深圳）有限公司 | Server identification method and device, readable storage medium and computer equipment |
| CN111859169A (en) * | 2019-06-05 | 2020-10-30 | 北京嘀嘀无限科技发展有限公司 | Destination recommendation method and system |
| CN110457589B (en) * | 2019-08-19 | 2020-05-12 | 上海新共赢信息科技有限公司 | Vehicle recommendation method, device, equipment and storage medium |
| CN110674416A (en) * | 2019-09-20 | 2020-01-10 | 北京小米移动软件有限公司 | Game recommendation method and device |
| CN112100362B (en) * | 2020-09-11 | 2023-07-28 | 北京百度网讯科技有限公司 | Document format recommendation model training method and device and electronic equipment |
| CN112287239B (en) * | 2020-12-30 | 2021-03-19 | 平安科技（深圳）有限公司 | Course recommendation method and device, electronic equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101321190A (en) * | 2008-07-04 | 2008-12-10 | 清华大学 | Recommend method and recommend system of heterogeneous network |
| CN103678647A (en) * | 2013-12-20 | 2014-03-26 | Tcl集团股份有限公司 | Method and system for recommending information |
| CN104216886A (en) * | 2013-05-29 | 2014-12-17 | 酷盛(天津）科技有限公司 | Video recommendation device, system and method |
| CN106874522A (en) * | 2017-03-29 | 2017-06-20 | 珠海习悦信息技术有限公司 | Information recommendation method, device, storage medium and processor |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8099376B2 (en) * | 2007-02-22 | 2012-01-17 | Fair Isaac Corporation | Rule-based management of adaptive models and agents |
| US8364528B2 (en) * | 2008-05-06 | 2013-01-29 | Richrelevance, Inc. | System and process for improving product recommendations for use in providing personalized advertisements to retail customers |
| CN107911491B (en) * | 2017-12-27 | 2019-09-27 | Oppo广东移动通信有限公司 | Information recommendation method, device and storage medium, server and mobile terminal |
-
2018
- 2018-06-26 CN CN201810670441.0A patent/CN108989397B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101321190A (en) * | 2008-07-04 | 2008-12-10 | 清华大学 | Recommend method and recommend system of heterogeneous network |
| CN104216886A (en) * | 2013-05-29 | 2014-12-17 | 酷盛(天津）科技有限公司 | Video recommendation device, system and method |
| CN103678647A (en) * | 2013-12-20 | 2014-03-26 | Tcl集团股份有限公司 | Method and system for recommending information |
| CN106874522A (en) * | 2017-03-29 | 2017-06-20 | 珠海习悦信息技术有限公司 | Information recommendation method, device, storage medium and processor |
Non-Patent Citations (1)
| Title |
|---|
| "推荐系统遇上深度学习(三)--DeepFM模型理论和实践";文哥的学习日记;《简书》;20180415;第1页，第3-4页，第9页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108989397A (en) | 2018-12-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108989397B (en) | Data recommendation method and device and storage medium | |
| US10452702B2 (en) | Data clustering | |
| US20170060993A1 (en) | Creating a Training Data Set Based on Unlabeled Textual Data | |
| CN112868004B (en) | Resource recommendation method and device, electronic equipment and storage medium | |
| CN108984650B (en) | Computer-readable recording medium and computer device | |
| CN105760397B (en) | Internet of things ontology model processing method and device | |
| US11310066B2 (en) | Method and apparatus for pushing information | |
| CN110413867B (en) | Method and system for content recommendation | |
| CN110209809B (en) | Text clustering method and device, storage medium and electronic device | |
| KR20190086126A (en) | Manual retrieval service providing server and method using chatbot | |
| CN115481227A (en) | Man-machine interaction dialogue method, device and equipment | |
| KR101678779B1 (en) | Method for recommending contents using metadata and apparatus for performing the method | |
| US20200118008A1 (en) | Building domain models from dialog interactions | |
| CN117272056A (en) | Object feature construction method, device and computer readable storage medium | |
| CN113239054B (en) | Information generation method and related device | |
| CN113934612A (en) | User portrait updating method and device, storage medium and electronic equipment | |
| CN110337008B (en) | Video interaction adjusting method, device, equipment and storage medium | |
| CN114090911A (en) | Interface processing method and device, computer equipment and computer readable storage medium | |
| CN112861007A (en) | Screen saver display method, device, equipment, medium and program product | |
| CN111917855A (en) | Content pushing method and device, electronic equipment and storage medium | |
| CN111782776A (en) | Method and device for realizing intention identification through slot filling | |
| CN117473070B (en) | Multi-channel application method of intelligent robot, intelligent robot and storage medium | |
| CN112348638B (en) | Activity document recommending method and device, electronic equipment and storage medium | |
| US20230205754A1 (en) | Data integrity optimization | |
| CN108536362B (en) | Method and device for identifying operation and server |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |