CN113761348A - Information recommendation method and device, electronic equipment and storage medium - Google Patents
Information recommendation method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN113761348A CN113761348A CN202110221124.2A CN202110221124A CN113761348A CN 113761348 A CN113761348 A CN 113761348A CN 202110221124 A CN202110221124 A CN 202110221124A CN 113761348 A CN113761348 A CN 113761348A
- Authority
- CN
- China
- Prior art keywords
- click
- data
- recommended
- objects
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the application provides an information recommendation method, an information recommendation device, an electronic device and a storage medium, wherein the information recommendation method comprises the following steps: acquiring user characteristics and object characteristics of at least two recommended objects to be recommended to a user; inputting the user characteristics and the characteristics of all objects into a click rate estimation model to obtain estimated click rates of all recommended objects; inputting the user characteristics and the characteristics of all objects into an index data estimation model to obtain estimation display index data of all recommended objects; sequencing all recommended objects based on the estimated click rate and the estimated display index data of all recommended objects; and selecting at least one recommended object from the at least two recommended objects according to the sorting result, and outputting the at least one recommended object.
Description
Technical Field
The present application relates to the field of data processing, and in particular, to an information recommendation method, an information recommendation apparatus, an electronic device, and a storage medium.
Background
With the rapid development of internet technology, the recommendation system is a business means adopted by information providers to publicize their own commodities, and is also a way adopted by advertising enterprises to generate revenue as users click on the advertised commodities.
In the related art, referring to fig. 1, the recommendation system includes a recall module, a model estimation module, and a ranking module, and only a click-through rate (ctr) estimation model is obtained in the model estimation module, so that a commodity recommended to a user is determined in a ranking stage based on the click-through rate estimation model and an advertisement base price set for the commodity by an advertisement enterprise. However, the method at least has the problems that the commodity information recommended to the user is inaccurate, the user experience is poor, and the maximum benefit of the recommendation system cannot be guaranteed.
Disclosure of Invention
The embodiment of the application is expected to provide an information recommendation method, an information recommendation device, an electronic device and a storage medium, so as to solve the problems that in the related art, the commodity information recommended to a user is inaccurate, the user experience is poor, and the maximum benefit of a recommendation system cannot be guaranteed.
The technical scheme of the embodiment of the application is realized as follows:
in a first aspect, an embodiment of the present application provides an information recommendation method, where the method includes:
acquiring user characteristics and object characteristics of at least two recommended objects to be recommended to a user;
inputting the user characteristics and all object characteristics into a click rate estimation model to obtain estimated click rates of all recommended objects;
inputting the user characteristics and the characteristics of all the objects into an index data estimation model to obtain estimation display index data of all the recommended objects;
sorting all recommended objects based on the estimated click rate and the estimated display index data of all recommended objects;
and selecting at least one recommended object from the at least two recommended objects according to the sorting result, and outputting the at least one recommended object.
In a second aspect, an embodiment of the present application provides an information recommendation apparatus, where the apparatus includes:
the acquisition module is used for acquiring the user characteristics and the object characteristics of at least two recommended objects to be recommended to the user;
the processing module is used for inputting the user characteristics and the characteristics of all the objects into a click rate estimation model to obtain the estimated click rates of all the recommended objects;
the processing module is further used for inputting the user characteristics and the characteristics of all the objects into an index data estimation model to obtain estimation display index data of all the recommended objects;
the sorting module is used for sorting all the recommended objects based on the estimated click rate and the estimated display index data of all the recommended objects;
the selection module is used for selecting at least one recommended object from the at least two recommended objects according to the sorting result;
and the output module is used for outputting the at least one recommended object.
In a third aspect, an embodiment of the present application provides an electronic device, including: a processor, a memory, and a communication bus;
the communication bus is used for realizing communication connection between the processor and the memory;
the processor is used for executing the information processing program stored in the memory so as to realize the information recommendation method.
In a fourth aspect, the present application provides a storage medium storing one or more programs, which are executable by one or more processors to implement the steps of the information recommendation method described above.
The application of the embodiment of the application realizes the following beneficial effects: the accuracy of pre-estimating the display index data is improved, the accuracy of recommendation information is improved, the user experience is improved, and the income of an advertisement system is increased.
Because the user characteristics and the object characteristics of at least two recommendation objects to be recommended to the user are obtained; inputting the user characteristics and the characteristics of all objects into a click rate estimation model to obtain estimated click rates of all recommended objects; inputting the user characteristics and the characteristics of all objects into an index data estimation model to obtain estimation display index data of all recommended objects; sequencing all recommended objects based on the estimated click rate and the estimated display index data of all recommended objects; selecting at least one recommended object from the at least two recommended objects according to the sorting result, and outputting the at least one recommended object; therefore, the problems that the commodity information recommended to the user in the related technology is inaccurate and the maximum profit of the recommendation system cannot be guaranteed are solved, the accuracy of pre-estimating the display index data can be improved, the accuracy of the recommendation information is improved, the user experience is improved, and the profit of the advertisement system can be increased.
Drawings
FIG. 1 is a schematic diagram of a recommendation system provided in the related art;
fig. 2 is an optional schematic flow chart of an information recommendation method provided in an embodiment of the present application
Fig. 3 is an alternative flow chart of an information recommendation method provided in an embodiment of the present application;
fig. 4 is an alternative flow chart of an information recommendation method provided in an embodiment of the present application;
FIG. 5 is an alternative structural diagram of a deep learning model provided in an embodiment of the present application;
FIG. 6 is a schematic diagram of an alternative modeling process of an information recommendation method according to an embodiment of the present application;
FIG. 7 is a schematic diagram of an alternative modeling process of an information recommendation method according to an embodiment of the present application;
fig. 8 is a schematic structural diagram of a recommendation system according to an embodiment of the present application;
fig. 9 is an alternative structural diagram of an information recommendation device according to an embodiment of the present application;
fig. 10 is an alternative structural schematic diagram of an electronic device according to an embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application.
It should be appreciated that reference throughout this specification to "an embodiment of the present application" or "an embodiment described previously" means that a particular feature, structure or characteristic described in connection with the embodiment is included in at least one embodiment of the present application. Thus, the appearances of the phrases "in the embodiments of the present application" or "in the embodiments" in various places throughout this specification are not necessarily all referring to the same embodiments. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the embodiments of the present application, the sequence numbers of the above-mentioned processes do not mean the execution sequence, and the execution sequence of each process should be determined by its function and inherent logic, and should not constitute any limitation to the implementation process of the embodiments of the present application. The above-mentioned serial numbers of the embodiments of the present application are merely for description and do not represent the merits of the embodiments.
An embodiment of the present application provides an information recommendation method, which is applied to an electronic device, and as shown in fig. 2, the method includes the following steps:
The user characteristics comprise user self characteristics and user behavior characteristics. Here, the user characteristics are characteristics that the user has, and include the user age, the user gender, and the user address; the user behavior characteristics are that according to a historical search object of a user, a browsing object of a webpage display page of the user or a historical recommendation object clicked or displayed by the user, the query term characteristics, the website access characteristics, the webpage related characteristics and the recommendation object related characteristics of the user are obtained respectively.
The recommended objects may be advertisements, including but not limited to commercial advertisements and/or business advertisements, which are propagated advertisements for producers or commercial operators to introduce and promote commercial products to consumers.
In an achievable application scenario, after the electronic device obtains user information and object information of at least two recommended objects to be recommended to a user, and extracts user characteristics in the user information and object characteristics in the object information, it can also roughly screen out recommended objects with thousands of levels of quantity associated with the user characteristics from a recommended object pool with millions of levels of recommended objects according to the user characteristics, and fill information associated with the recommended objects, such as advertisement information. Specifically, the target formula for screening out the recommended objects having thousands of levels associated with the user characteristics is as follows,
wherein O is a recommended object associated with the user characteristics, N is the number of the screened recommended objects associated with the user characteristics, user is the user characteristics, item is the recommended object, itemiRelevence (user, item) as the ith recommended objecti) For user characteristics user and ith recommendation object itemiThe correlation between them. The meaning of the formula is to obtain the first N recommended objects with high correlation degree with the user characteristics.
The input click rate estimation model is constructed on the basis of a deep learning model capable of processing nonlinear characteristics.
In the embodiment of the application, after the electronic equipment obtains the user characteristics and the object characteristics of at least two recommended objects to be recommended to the user, the user characteristics and the object characteristics of all the recommended objects are input into the click rate estimation model to obtain the estimated click rates of all the recommended objects.
The index data estimation model is constructed on the basis of a deep learning model capable of processing nonlinear characteristics. The pre-estimation display index data is used for indicating the proportion of income and investment which can be converted by the recommendation object.
In the embodiment of the application, after the user characteristics and the object characteristics of at least two recommended objects to be recommended to the user are obtained, the electronic equipment inputs the user characteristics and the characteristics of all the objects into the index data estimation model to obtain the estimation display index data of all the recommended objects.
And step 204, sequencing all recommended objects based on the estimated click rate and estimated thousands of revenues of all recommended objects.
In the embodiment of the application, after the electronic equipment obtains the estimated click rate and the estimated thousands of display benefits of all the recommended objects, all the recommended objects are sorted based on the estimated click rate and the estimated thousands of display benefits.
In an achievable application scenario, after roughly screening the recommendation objects with thousand-level quantity associated with the user characteristics from a recommendation object pool with million-level recommendation objects according to the user characteristics, the electronic device performs fine sorting on the screened recommendation objects with thousand-level quantity associated with the user characteristics according to the estimated click rate and estimated thousand-time display profits of all the recommendation objects to obtain a sorting result of the recommendation objects.
And step 205, selecting at least one recommended object from the at least two recommended objects according to the sorting result, and outputting the at least one recommended object.
In the embodiment of the application, the electronic device selects at least one recommended object from the at least two recommended objects according to the sorting result, and outputs the at least one recommended object.
According to the information recommendation method provided by the embodiment of the application, the user characteristics and the object characteristics of at least two recommendation objects to be recommended to a user are obtained; inputting the user characteristics and the characteristics of all objects into a click rate estimation model to obtain estimated click rates of all recommended objects; inputting the user characteristics and the characteristics of all objects into an index data estimation model to obtain estimation display index data of all recommended objects; sequencing all recommended objects based on the estimated click rate and the estimated display index data of all recommended objects; selecting at least one recommended object from the at least two recommended objects according to the sorting result, and outputting the at least one recommended object; therefore, the problems that the commodity information recommended to the user in the related technology is inaccurate and the maximization of the profit of the recommendation system cannot be guaranteed are solved, the accuracy of pre-estimating the display index data can be improved, the accuracy of the recommendation information is improved, the user experience is improved, and the profit of the advertisement system can be increased.
An embodiment of the present application provides an information recommendation method, which is applied to an electronic device, and as shown in fig. 3, the method includes the following steps:
The exposure log comprises a click object in the click log and a candidate object different from the click object.
In an implementation scenario, the electronic device may obtain a click log and an exposure log in the log server within a preset historical time period. Here, the exposure log includes information that the electronic device has previously presented to the user to browse, slide, or click on the contained objects. The click log is information which is displayed in advance to the user to browse, slide or click the contained objects, and the behavior of clicking, buying and/or purchasing a certain object when the user browses is recorded. The electronic equipment counts the required characteristic data of the click log and the exposure log through a certain calculation relation.
The click characteristic data is used for representing data generated by a target user clicking the click object recorded in the click log, and the exposure characteristic data is used for representing and displaying data generated by the click object and the candidate object.
In the embodiment of the application, the electronic equipment extracts click characteristic data from the click log and extracts exposure characteristic data from the exposure log. Illustratively, the click profile includes a user identification of the target user, an object identification of the click object, a click time of the click object, whether the click object is an advertisement, and a single click price of the click object. The exposure characteristic data comprises a target identifier of an exposure object exposed to a target user, an object identifier of the exposure object, exposure time of the exposure object and whether the exposure object is an advertisement; note that the click object and the candidate object constitute an exposure object.
In the embodiment of the present application, referring to fig. 4, step 303 determines target training data based on the click feature data in the click log and the exposure feature data in the exposure log, and may also be implemented by the following steps,
In the embodiment of the application, in the click feature data, at least one piece of original click data corresponding to each target user is obtained by the electronic device.
In the embodiment of the application, in the exposure feature data, at least one piece of original exposure data corresponding to each target user is obtained by the electronic device.
3033, carrying out Cartesian product operation on the original click data and the original exposure data corresponding to the same target user to obtain original training data.
In the embodiment of the application, the electronic device performs Cartesian product operation on original click data and original exposure data corresponding to the same target user to obtain original training data. Illustratively, the raw training data includes an object identification of the click object, an object identification of the exposure object, a click time of the click object, whether the click object is an advertisement and a single click price of the click object, an exposure time of the exposure object, and whether the exposure object is an advertisement.
3034, clustering all objects in the original training data according to the object identifiers and the click information corresponding to the object identifiers to obtain clustered original training data.
The click information comprises an object identifier of a click object, the number of clicks and the sum of the price of a single click of the click object.
In the embodiment of the application, the electronic device performs clustering processing on all objects in the original training data according to the object identifiers and the click information corresponding to the object identifiers to obtain the clustered original training data. Illustratively, the electronic device clusters all objects in the original training data according to two features, namely, an object identifier of a clicked object and an object identifier of an exposed object, to obtain clustered original training data. Here, the aggregated original training data includes an object identifier of a click object, an object identifier of an exposure object, an exposure number, a click number, and a sum of single click prices of the click object; it should be noted that the exposure times are the number of data obtained after clustering processing is performed according to the object identifier and the click information corresponding to the object identifier; and the number of clicks is determined in the click information corresponding to the object identifier in the number of data pieces obtained after clustering.
And step 3035, screening out target training data of which the click information meets the click condition from the clustered original training data.
In some embodiments, the click condition includes a clicked condition. Further, the click condition also comprises a click number condition; that is, the exposure object is not only to be clicked by the target user, but also satisfies the number of times of being clicked by the target user.
In an achievable application scenario, the electronic equipment screens out target training data of which the click information meets the clicked condition from the clustered original training data; or screening out target training data with click information meeting the conditions of being clicked and the click times by the electronic equipment from the clustered original training data.
And 304, training the target user characteristics of the target user and first sample data associated with the target user characteristics in the target training data based on a first deep learning algorithm to obtain a click rate estimation model.
The first sample data comprises an object identification of a clicked object, an object identification of an exposed object, exposure times and click times.
In an implementation application scenario, referring to fig. 5, fig. 5 shows a structural schematic diagram of a deep learning algorithm model, first, an electronic device inputs a target user feature and a feature included in first sample data associated with the target user feature in target training data into a first deep learning algorithm; secondly, the target user characteristics and the characteristics contained in the first sample data associated with the target user characteristics in the target training data are sent to an embedding layer for processing to obtain characteristic vectors; and then, expressing and learning the feature vector through a Parametric Linear rectification function (PRELU), and sending the learned feature to a regression layer of a deep learning model to further obtain a click rate estimation model.
305, training the target user characteristics and second sample data associated with the target user characteristics in the target training data based on a second deep learning algorithm to obtain an index data estimation model.
The first sample data and the second sample data are different and form target training data.
In the embodiment of the present application, the first deep learning algorithm and the second deep learning algorithm may be the same or different.
The second sample data comprises an object identifier of a clicked object, an object identifier of an exposed object, exposure times, click times and the sum of single click prices of the clicked objects.
In the embodiment of the present application, referring to fig. 6, the index data prediction model 401 may be obtained by performing deep learning model training on the target user characteristics and second sample data (cross characteristics of the target user characteristics and the object characteristics) associated with the target user characteristics in the target training data.
In other embodiments of the present application, referring to fig. 7, the index data estimation model 501 may also be obtained by an advertisement price and click-through rate estimation module 503. Specifically, intelligent bids are set in a sorting module of the recommendation system, and a single click price (cpc) set by an advertisement enterprise is obtained through a cost per click (cpc) obtaining module 502; and secondly, adjusting the single click price cpc in a sorting module according to the flow quality and the bidding environment, and multiplying the adjusted cpc by the click rate estimation model to obtain an index data estimation model.
In an implementation application scenario, referring to fig. 5, fig. 5 shows a structural schematic diagram of a deep learning algorithm model, and first, an electronic device inputs a target user feature and a feature included in second sample data associated with the target user feature in target training data into a second deep learning algorithm; secondly, the target user characteristics and the characteristics contained in second sample data associated with the target user characteristics in the target training data are sent to an embedded layer for processing to obtain characteristic vectors; and then, expressing and learning the feature vector through a Parametric Linear rectification function (PRELU) and a Linear rectification function (RELU), and sending the learned feature to a regression layer of the deep learning model to further obtain the index data estimation model.
In the deep learning, abstract features are learned through a multi-layer network, and the obtained abstract features are used in a final output layer (i.e., a regression layer) to complete a final learning task. Such learned features may be desirable to reduce the degree of non-linearity of the problem. The strong point of deep learning is that the error of the target function can be transmitted back by utilizing back propagation, the error is propagated to the direction of an output layer by layer so as to correct network parameters, and the network parameters can be well trained after multiple iterations. Meanwhile, implicit characteristics which are difficult to obtain through artificial characteristic extraction can be obtained through deep learning, so that the estimation capability of the estimation model obtained through the deep learning is obviously improved, and the estimation effect is better.
And step 306, inputting the user characteristics and the characteristics of all the objects into the click rate estimation model to obtain the estimated click rates of all the recommended objects.
And 307, inputting the user characteristics and the characteristics of all the objects into an index data estimation model to obtain estimation display index data of all the recommended objects.
And 308, determining the product of the T power of the estimated click rate of all the recommended objects and the estimated display index data as the recommendation confidence of all the recommended objects.
Wherein T is a positive integer.
In the embodiment of the application, when T is 1, the determined recommendation confidence of the recommended object is more accurate, the recommended object recommended for the user is more accurate, and the system benefit is better.
In an implementation scenario, referring to fig. 8, fig. 8 shows an architecture diagram of an alternative recommendation system provided in the present application. The electronic equipment acquires a click log 601 and an exposure log 602 in a preset historical time period from a log server; extracting click feature data in the click log 601 and exposure feature data in the exposure log 602 by a feature extraction module 603; the electronic device aggregates the click feature data and the exposure feature data according to the target user, and performs cartesian product operation on the original click data and the original exposure data corresponding to the same target user to obtain original training data 604. Clustering all objects in the original training data 604 through the relevance control module 605 according to the object identifiers and the click information corresponding to the object identifiers to obtain clustered original training data, and screening out target training data 606 of which the click information meets the click condition from the clustered original training data. Training the target user characteristics of the target user and first sample data associated with the target user characteristics in the target training data 606 based on a first deep learning algorithm to obtain a click rate estimation model 607; and training the target user characteristics and second sample data associated with the target user characteristics in the target training data 606 based on a second deep learning algorithm to obtain an index data prediction model 608. The user characteristics and the object characteristics of the recommended object to be recommended to the user are input 609 to a click rate prediction model 607 and an index data prediction model 608 respectively, so that a predicted click rate 610 and predicted display index data 611 are obtained respectively. The multi-target fusion module 612 processes the estimated click rate 610 and the estimated display index data 611 to obtain recommendation confidence levels 613 of all recommended objects.
And 310, selecting at least one recommended object with a recommendation confidence coefficient meeting a confidence coefficient threshold from the at least two recommended objects according to the sorting result, and outputting the at least one recommended object.
In an achievable application scenario, the estimated display index data may be thousands of times of display gains (ecpm), after the electronic device obtains estimated click rate (pctr) of all recommended objects and the estimated display index data ecpm, the product of the T-th power of pctr of all recommended objects and the estimated display index data ecpm is calculated, so as to obtain a recommendation confidence of each recommended object, all recommended objects are sorted based on the recommendation confidence of all recommended objects, and the N recommended objects with the highest recommendation confidence are selected. Specifically, the recommendation confidence is calculated by the following formula,
score＝pctrT*ecpm
wherein score is the recommendation confidence of the recommended object, pctr is the estimated click rate of the recommended object, ecpm is the estimated display index data of the recommended object, and T is a hyperparameter which balances the estimated click rate, namely representing the long-term income, and the estimated display index data, namely representing the short-term income. Illustratively, when T is 1, the recommendation confidence of the determined recommended object is more accurate, the recommended object recommended to the user is more accurate, and the system benefit is higher.
Therefore, in the embodiment of the application, after the click rate ctr estimation model and the index data ecpm estimation model are directly created by the electronic equipment through historical data, the user characteristics of the user and the object characteristics of the recommended object to be recommended to the user are directly input into the two estimation models to respectively obtain the estimated click rate and the estimated display index data, and then the recommended object is optimally recommended by referring to the estimated result; therefore, the estimation accuracy of the estimated and displayed index data is improved; meanwhile, the accuracy of the recommendation information is improved, the user experience is guaranteed, the overall benefit of the recommendation system is improved, and meanwhile the long-term benefit and the short-term benefit of the recommendation system are considered.
It should be noted that, for the descriptions of the same steps and the same contents in this embodiment as those in other embodiments, reference may be made to the descriptions in other embodiments, which are not described herein again.
An embodiment of the present application provides an information recommendation apparatus, which may be applied to an information recommendation method provided in an embodiment corresponding to fig. 2 to 4, and as shown in fig. 9, the information recommendation apparatus 8 includes:
an obtaining module 801, configured to obtain a user characteristic and object characteristics of at least two recommendation objects to be recommended to a user;
the processing module 802 is configured to input the user characteristics and the characteristics of all objects into the click rate estimation model to obtain estimated click rates of all recommended objects;
the processing module 802 is further configured to input the user characteristics and the characteristics of all the objects into the index data estimation model to obtain estimated display index data of all the recommended objects;
the sorting module 803 is configured to sort all recommended objects based on the estimated click rates and the estimated display index data of all recommended objects;
a selecting module 804, configured to select at least one recommended object from the at least two recommended objects according to the sorting result;
an output module 805 for outputting at least one recommended object.
In other embodiments of the present application, the obtaining module 801 is further configured to obtain a click log and an exposure log within a preset historical time period; the exposure log comprises a click object in the click log and a candidate object different from the click object; the processing module 802 is further configured to determine target training data based on click feature data in the click log and exposure feature data in the exposure log; the click characteristic data is used for representing data generated by a target user clicking the click object recorded in the click log, and the exposure characteristic data is used for representing and displaying data generated by the click object and the candidate object; training target user characteristics of a target user and first sample data associated with the target user characteristics in target training data based on a first deep learning algorithm to obtain a click rate estimation model; training the target user characteristics and second sample data associated with the target user characteristics in the target training data based on a second deep learning algorithm to obtain an index data pre-estimation model; the first sample data and the second sample data are different and form target training data.
In other embodiments of the present application, the obtaining module 801 is further configured to obtain, from the click feature data, original click data corresponding to each target user in all target users; acquiring original exposure data corresponding to each target user in all target users from the exposure feature data; the processing module 802 is further configured to perform cartesian product operation on original click data and original exposure data corresponding to the same target user to obtain original training data; and screening partial data from the original training data to obtain target training data.
In other embodiments of the present application, the processing module 802 is further configured to perform clustering on all objects in the original training data according to the object identifier and the click information corresponding to the object identifier, so as to obtain clustered original training data; and screening the clustered original training data to obtain target training data.
In other embodiments of the present application, the processing module 802 is further configured to screen out, from the clustered original training data, target training data whose click information meets the click condition.
In other embodiments of the present application, the processing module 802 is further configured to determine a product of the T-th power of the estimated click rate and the estimated display index data of all recommended objects, which is a recommendation confidence of all recommended objects; wherein T is a positive integer; the sorting module 803 is further configured to sort all the recommended objects based on the recommendation confidence degrees of all the recommended objects.
In other embodiments of the present application, the selecting module 804 is further configured to select, from the at least two recommended objects according to the sorting result, at least one recommended object whose recommendation confidence meets the confidence threshold.
An embodiment of the present application provides an electronic device, which may be applied to an information recommendation method provided in an embodiment corresponding to fig. 2 to 4, and as shown in fig. 10, the electronic device (an electronic device 9 in fig. 10 corresponds to an information recommendation device 8 in fig. 9) includes: a processor 901, a memory 902, and a communication bus 903, wherein:
the communication bus 903 is used for realizing communication connection between the processor 901 and the memory 902;
the processor 901 is configured to execute an information processing program stored in the memory 902 to implement the following steps:
acquiring user characteristics and object characteristics of at least two recommended objects to be recommended to a user;
inputting the user characteristics and the characteristics of all objects into a click rate estimation model to obtain estimated click rates of all recommended objects;
inputting the user characteristics and the characteristics of all objects into an index data estimation model to obtain estimation display index data of all recommended objects;
sequencing all recommended objects based on the estimated click rate and the estimated display index data of all recommended objects;
and selecting at least one recommended object from the at least two recommended objects according to the sorting result, and outputting the at least one recommended object.
In other embodiments of the present application, the processor 902 is configured to execute the information recommendation processing program stored in the memory 901 to implement the following steps:
acquiring a click log and an exposure log in a preset historical time period; the exposure log comprises a click object in the click log and a candidate object different from the click object; determining target training data based on the click characteristic data in the click log and the exposure characteristic data in the exposure log; the click characteristic data is used for representing data generated by a target user clicking the click object recorded in the click log, and the exposure characteristic data is used for representing and displaying data generated by the click object and the candidate object; training target user characteristics of a target user and first sample data associated with the target user characteristics in target training data based on a first deep learning algorithm to obtain a click rate estimation model; training the target user characteristics and second sample data associated with the target user characteristics in the target training data based on a second deep learning algorithm to obtain an index data pre-estimation model; the first sample data and the second sample data are different and form target training data.
In other embodiments of the present application, the processor 902 is configured to execute the information recommendation processing program stored in the memory 901 to implement the following steps:
acquiring original click data corresponding to each target user in all target users from the click feature data; acquiring original exposure data corresponding to each target user in all target users from the exposure feature data; carrying out Cartesian product operation on original click data and original exposure data corresponding to the same target user to obtain original training data; and screening partial data from the original training data to obtain target training data.
In other embodiments of the present application, the processor 902 is configured to execute the information recommendation processing program stored in the memory 901 to implement the following steps:
clustering all objects in the original training data according to the object identifications and the click information corresponding to the object identifications to obtain clustered original training data; and screening the clustered original training data to obtain target training data.
In other embodiments of the present application, the processor 902 is configured to execute the information recommendation processing program stored in the memory 901 to implement the following steps:
and screening out target training data of which the click information meets the click condition from the clustered original training data.
In other embodiments of the present application, the processor 902 is configured to execute the information recommendation processing program stored in the memory 901 to implement the following steps:
determining the product of the T power of the estimated click rate and the estimated display index data of all the recommended objects as the recommendation confidence of all the recommended objects; wherein T is a positive integer;
and sorting all recommended objects based on the recommendation confidence degrees of all recommended objects.
In other embodiments of the present application, the processor 902 is configured to execute the information recommendation processing program stored in the memory 901 to implement the following steps:
and selecting at least one recommended object with a recommendation confidence degree meeting a confidence degree threshold from the at least two recommended objects according to the sorting result.
It should be noted that, for the descriptions of the same steps and the same contents in this embodiment as those in other embodiments, reference may be made to the descriptions in other embodiments, which are not described herein again.
Based on the foregoing embodiments, embodiments of the present application provide a computer storage medium storing one or more programs, the one or more programs being executable by one or more processors to implement the steps of:
acquiring user characteristics and object characteristics of at least two recommended objects to be recommended to a user;
inputting the user characteristics and the characteristics of all objects into a click rate estimation model to obtain estimated click rates of all recommended objects;
inputting the user characteristics and the characteristics of all objects into an index data estimation model to obtain estimation display index data of all recommended objects;
sequencing all recommended objects based on the estimated click rate and the estimated display index data of all recommended objects;
and selecting at least one recommended object from the at least two recommended objects according to the sorting result, and outputting the at least one recommended object.
In other embodiments of the present application, the one or more programs are executable by the one or more processors and further implement the steps of:
acquiring a click log and an exposure log in a preset historical time period; the exposure log comprises a click object in the click log and a candidate object different from the click object; determining target training data based on the click characteristic data in the click log and the exposure characteristic data in the exposure log; the click characteristic data is used for representing data generated by a target user clicking the click object recorded in the click log, and the exposure characteristic data is used for representing and displaying data generated by the click object and the candidate object; training target user characteristics of a target user and first sample data associated with the target user characteristics in target training data based on a first deep learning algorithm to obtain a click rate estimation model; training the target user characteristics and second sample data associated with the target user characteristics in the target training data based on a second deep learning algorithm to obtain an index data pre-estimation model; the first sample data and the second sample data are different and form target training data.
In other embodiments of the present application, the one or more programs are executable by the one or more processors and further implement the steps of:
acquiring original click data corresponding to each target user in all target users from the click feature data; acquiring original exposure data corresponding to each target user in all target users from the exposure feature data; carrying out Cartesian product operation on original click data and original exposure data corresponding to the same target user to obtain original training data; and screening partial data from the original training data to obtain target training data.
In other embodiments of the present application, the one or more programs are executable by the one or more processors and further implement the steps of:
clustering all objects in the original training data according to the object identifications and the click information corresponding to the object identifications to obtain clustered original training data; and screening the clustered original training data to obtain target training data.
In other embodiments of the present application, the one or more programs are executable by the one or more processors and further implement the steps of:
and screening out target training data of which the click information meets the click condition from the clustered original training data.
In other embodiments of the present application, the one or more programs are executable by the one or more processors and further implement the steps of:
determining the product of the T power of the estimated click rate and the estimated display index data of all the recommended objects as the recommendation confidence of all the recommended objects; wherein T is a positive integer; and sorting all recommended objects based on the recommendation confidence degrees of all recommended objects.
In other embodiments of the present application, the one or more programs are executable by the one or more processors and further implement the steps of:
and selecting at least one recommended object with a recommendation confidence degree meeting a confidence degree threshold from the at least two recommended objects according to the sorting result.
The computer storage medium/Memory may be a Read Only Memory (ROM), a Programmable Read Only Memory (PROM), an Erasable Programmable Read Only Memory (EPROM), an Electrically Erasable Programmable Read Only Memory (EEPROM), a magnetic Random Access Memory (FRAM), a Flash Memory (Flash Memory), a magnetic surface Memory, an optical Disc, or a Compact Disc Read-Only Memory (CD-ROM); but may also be various terminals such as mobile phones, computers, tablet devices, personal digital assistants, etc., that include one or any combination of the above-mentioned memories.
In the several embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. The above-described device embodiments are merely illustrative, for example, the division of the unit is only a logical functional division, and there may be other division ways in actual implementation, such as: multiple units or components may be combined, or may be integrated into another system, or some features may be omitted, or not implemented. In addition, the coupling, direct coupling or communication connection between the components shown or discussed may be through some interfaces, and the indirect coupling or communication connection between the devices or units may be electrical, mechanical or other forms.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed on a plurality of network units; some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, all functional units in the embodiments of the present application may be integrated into one processing module, or each unit may be separately regarded as one unit, or two or more units may be integrated into one unit; the integrated unit can be realized in a form of hardware, or in a form of hardware plus a software functional unit. Those of ordinary skill in the art will understand that: all or part of the steps for implementing the method embodiments may be implemented by hardware related to program instructions, and the program may be stored in a computer readable storage medium, and when executed, the program performs the steps including the method embodiments; and the aforementioned storage medium includes: various media capable of storing program codes, such as a removable Memory device, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, and an optical disk.
The methods disclosed in the several method embodiments provided in the present application may be combined arbitrarily without conflict to obtain new method embodiments.
Features disclosed in several of the product embodiments provided in the present application may be combined in any combination to yield new product embodiments without conflict.
The features disclosed in the several method or apparatus embodiments provided in the present application may be combined arbitrarily, without conflict, to arrive at new method embodiments or apparatus embodiments.
The above description is only for the specific embodiments of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present application, and shall be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (10)
1. An information recommendation method, characterized in that the method comprises:
acquiring user characteristics and object characteristics of at least two recommended objects to be recommended to a user;
inputting the user characteristics and all object characteristics into a click rate estimation model to obtain estimated click rates of all recommended objects;
inputting the user characteristics and the characteristics of all the objects into an index data estimation model to obtain estimation display index data of all the recommended objects;
sorting all recommended objects based on the estimated click rate and the estimated display index data of all recommended objects;
and selecting at least one recommended object from the at least two recommended objects according to the sorting result, and outputting the at least one recommended object.
2. The method of claim 1, wherein before entering the user characteristics and the characteristics of all objects into a click-through rate prediction model to obtain the predicted click-through rates of all recommended objects, the method further comprises:
acquiring a click log and an exposure log in a preset historical time period; the exposure log comprises a click object in the click log and a candidate object different from the click object;
determining target training data based on the click feature data in the click log and the exposure feature data in the exposure log; the click feature data is used for representing data generated by a target user clicking the click object recorded in the click log, and the exposure feature data is used for representing and displaying the click object and the data generated by the candidate object;
training the target user characteristics of the target user and first sample data associated with the target user characteristics in the target training data based on a first deep learning algorithm to obtain the click rate estimation model;
training the target user characteristics and second sample data associated with the target user characteristics in the target training data based on a second deep learning algorithm to obtain the index data estimation model; wherein the first sample data is different from the second sample data, and the first sample data and the second sample data constitute the target training data.
3. The method of claim 2, wherein determining target training data based on the click feature data in the click log and the exposure feature data in the exposure log comprises:
acquiring original click data corresponding to each target user in all target users from the click feature data;
acquiring original exposure data corresponding to each target user in all the target users from the exposure feature data;
carrying out Cartesian product operation on the original click data and the original exposure data corresponding to the same target user to obtain original training data;
and screening partial data from the original training data to obtain the target training data.
4. The method of claim 3, wherein the filtering out the portion of the original training data to obtain the target training data comprises:
clustering all objects in the original training data according to object identifications and click information corresponding to the object identifications to obtain clustered original training data;
and screening the clustered original training data to obtain the target training data.
5. The method of claim 4, wherein the filtering the clustered original training data to obtain the target training data comprises:
and screening out the target training data of which the click information meets the click condition from the clustered original training data.
6. The method of claim 1, wherein the ranking the all recommended objects based on the estimated click through rates and the estimated exposure indicator data of the all recommended objects comprises:
determining the product of the T power of the estimated click rate of all the recommended objects and the estimated display index data as the recommendation confidence of all the recommended objects; wherein T is a positive integer;
and sequencing all recommended objects based on the recommendation confidence degrees of all recommended objects.
7. The method of claim 6, wherein selecting at least one recommended object from the at least two recommended objects according to the ranking result comprises:
and selecting at least one recommended object with the recommendation confidence coefficient meeting a confidence coefficient threshold from the at least two recommended objects according to the sorting result.
8. An information recommendation apparatus, characterized in that the apparatus comprises:
the acquisition module is used for acquiring the user characteristics and the object characteristics of at least two recommended objects to be recommended to the user;
the processing module is used for inputting the user characteristics and the characteristics of all the objects into a click rate estimation model to obtain the estimated click rates of all the recommended objects;
the processing module is further used for inputting the user characteristics and the characteristics of all the objects into an index data estimation model to obtain estimation display index data of all the recommended objects;
the sorting module is used for sorting all the recommended objects based on the estimated click rate and the estimated display index data of all the recommended objects;
the selection module is used for selecting at least one recommended object from the at least two recommended objects according to the sorting result;
and the output module is used for outputting the at least one recommended object.
9. An electronic device, characterized in that the electronic device comprises: a processor, a memory, and a communication bus;
the communication bus is used for realizing communication connection between the processor and the memory;
the processor is configured to execute an information processing program stored in the memory to implement the information recommendation method according to any one of claims 1 to 7.
10. A storage medium characterized in that the storage medium stores one or more programs executable by one or more processors to implement the information recommendation method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110221124.2A CN113761348A (en) | 2021-02-26 | 2021-02-26 | Information recommendation method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110221124.2A CN113761348A (en) | 2021-02-26 | 2021-02-26 | Information recommendation method and device, electronic equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113761348A true CN113761348A (en) | 2021-12-07 |
Family
ID=78786703
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110221124.2A Pending CN113761348A (en) | 2021-02-26 | 2021-02-26 | Information recommendation method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113761348A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114299350A (en) * | 2021-12-15 | 2022-04-08 | 四川新网银行股份有限公司 | Artificial credit auditing information recommendation method and system based on machine learning |
| CN114430504A (en) * | 2022-01-28 | 2022-05-03 | 腾讯科技（深圳）有限公司 | Recommendation method and related device for media content |
| CN115129975A (en) * | 2022-05-13 | 2022-09-30 | 腾讯科技（深圳）有限公司 | Recommendation model training method, recommendation device, recommendation equipment and storage medium |
| WO2023124029A1 (en) * | 2021-12-27 | 2023-07-06 | 北京百度网讯科技有限公司 | Deep learning model training method and apparatus, and content recommendation method and apparatus |
-
2021
- 2021-02-26 CN CN202110221124.2A patent/CN113761348A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114299350A (en) * | 2021-12-15 | 2022-04-08 | 四川新网银行股份有限公司 | Artificial credit auditing information recommendation method and system based on machine learning |
| WO2023124029A1 (en) * | 2021-12-27 | 2023-07-06 | 北京百度网讯科技有限公司 | Deep learning model training method and apparatus, and content recommendation method and apparatus |
| CN114430504A (en) * | 2022-01-28 | 2022-05-03 | 腾讯科技（深圳）有限公司 | Recommendation method and related device for media content |
| CN115129975A (en) * | 2022-05-13 | 2022-09-30 | 腾讯科技（深圳）有限公司 | Recommendation model training method, recommendation device, recommendation equipment and storage medium |
| CN115129975B (en) * | 2022-05-13 | 2024-01-23 | 腾讯科技（深圳）有限公司 | Recommendation model training method, recommendation device, recommendation equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US8543518B2 (en) | Deducing shadow user profiles for ad campaigns | |
| US10134053B2 (en) | User engagement-based contextually-dependent automated pricing for non-guaranteed delivery | |
| CN105989004B (en) | Information delivery preprocessing method and device | |
| CN113761348A (en) | Information recommendation method and device, electronic equipment and storage medium | |
| Miralles-Pechuán et al. | A novel methodology for optimizing display advertising campaigns using genetic algorithms | |
| US20160189201A1 (en) | Enhanced targeted advertising system | |
| WO2017190610A1 (en) | Target user orientation method and device, and computer storage medium | |
| US20200234218A1 (en) | Systems and methods for entity performance and risk scoring | |
| CN111095330B (en) | Machine learning method and system for predicting online user interactions | |
| CN113744016B (en) | Object recommendation method and device, equipment and storage medium | |
| CN111798280A (en) | Multimedia information recommendation method, device and equipment and storage medium | |
| CN111782937A (en) | Information sorting method and device, electronic equipment and computer readable medium | |
| CN111429214B (en) | Transaction data-based buyer and seller matching method and device | |
| CN112598472A (en) | Product recommendation method, device, system, medium and program product | |
| CN107077455A (en) | Flow mass is determined using the score traffic based on event | |
| US20230316106A1 (en) | Method and apparatus for training content recommendation model, device, and storage medium | |
| US20220198487A1 (en) | Method and device for processing user interaction information | |
| CN113793161A (en) | Advertisement delivery method, advertisement delivery device, readable storage medium and electronic device | |
| WO2020150597A1 (en) | Systems and methods for entity performance and risk scoring | |
| CN114595323B (en) | Portrait construction, recommendation, model training method, apparatus, device and storage medium | |
| CN113935529A (en) | Method, device, equipment and medium for determining life cycle value of customer | |
| CN112613897A (en) | Real-time automatic bidding method and system based on gross profit rate | |
| CN113793164A (en) | Advertisement putting method, device, equipment and storage medium | |
| CN113743636B (en) | Target operation prediction method and device, electronic equipment and storage medium | |
| CN116664166A (en) | Customer screening method and device, electronic equipment and storage medium |
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 |