CN113297486A - Click rate prediction method and related device - Google Patents
Click rate prediction method and related device Download PDFInfo
- Publication number
- CN113297486A CN113297486A CN202110565423.8A CN202110565423A CN113297486A CN 113297486 A CN113297486 A CN 113297486A CN 202110565423 A CN202110565423 A CN 202110565423A CN 113297486 A CN113297486 A CN 113297486A
- Authority
- CN
- China
- Prior art keywords
- user
- behavior
- predicted
- model
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 53
- 230000006399 behavior Effects 0.000 claims abstract description 144
- 238000007781 pre-processing Methods 0.000 claims abstract description 7
- 239000013598 vector Substances 0.000 claims description 55
- 238000012549 training Methods 0.000 claims description 26
- RPNUMPOLZDHAAY-UHFFFAOYSA-N Diethylenetriamine Chemical compound NCCNCCN RPNUMPOLZDHAAY-UHFFFAOYSA-N 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 9
- 238000001514 detection method Methods 0.000 claims description 8
- 238000013507 mapping Methods 0.000 claims description 7
- 230000007246 mechanism Effects 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 4
- 230000004044 response Effects 0.000 claims description 3
- 238000010586 diagram Methods 0.000 description 7
- 230000000875 corresponding effect Effects 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 230000014759 maintenance of location Effects 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000003542 behavioural effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 239000013604 expression vector Substances 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000007639 printing Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
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
- 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
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Business, Economics & Management (AREA)
- Computing Systems (AREA)
- Economics (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Mathematical Physics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application discloses a click rate prediction method and a related device, and is characterized in that the click rate prediction method comprises the following steps: acquiring behaviors of a user in a historical time period, behavior time points corresponding to the behaviors, target numbers to be predicted and time points to be predicted; the behavior of the user comprises click data generated by clicking a target by the user and state data of the user and the clicked target; preprocessing behaviors and behavior time points to obtain behavior characteristic information of the user; and predicting the click probability of the user clicking the target to be predicted at the time point to be predicted based on the behavior characteristic information by using the prediction model. By the method, the click rate of the user to the target to be predicted is predicted, so that the target which is interested by the user is recommended to the user.
Description
Technical Field
The invention relates to the technical field of deep learning, in particular to a click rate prediction method and a related device.
Background
For a live broadcast platform, the retention and growth of a user are one of core tasks of platform operation, and the distribution method for optimizing user flow can improve the stickiness of the user to a main broadcast/platform, improve the content ecology of the platform and further realize the core tasks of the retention and growth of the user. Recommendation is an important traffic distribution scene, and when recommended content meets the interest of a user, the user can contribute more traffic and keep on a platform; when the recommended content is contrary to the user's interest, the user may lose the content of interest because the user cannot find the content.
The click rate is an important index for measuring the recommendation effect, and when the user is not interested in the recommended content, the user can not click on the recommendation card. Therefore, the click probability of the user on the anchor is predicted, so that the anchor in which the user is interested is recommended to the user, and the method is important for maintaining the operation of a live broadcast platform.
Disclosure of Invention
The technical problem mainly solved by the application is to provide a click rate prediction method and a related device, so as to predict the click rate of a target to be predicted by a user, and thus recommend the target which the user is interested in to the user.
In order to solve the above technical problem, the present application provides a click rate prediction method, including: acquiring behaviors of a user in a historical time period, behavior time points corresponding to the behaviors, target numbers to be predicted and time points to be predicted; the behavior of the user comprises click data generated by clicking the target by the user and state data of the user and the clicked target; preprocessing behaviors and behavior time points to obtain behavior characteristic information of a user; and predicting the click probability of the user clicking the target to be predicted at the time point to be predicted based on the behavior characteristic information by using the prediction model.
The method comprises the following steps of preprocessing behaviors and behavior time points to obtain behavior characteristic information of a user, wherein the steps comprise: mapping the behaviors and the behavior time points into sequence characteristics and statistical characteristics; the method for predicting the click probability of the user clicking the target to be predicted at the time point to be predicted based on the behavior characteristic information by using the prediction model comprises the following steps: extracting feature information of the sequence features by using a first model of the prediction model to obtain first feature information; extracting characteristic information of the statistical characteristics by using a second model of the prediction model to obtain second characteristic information; splicing the first characteristic information and the second characteristic information to obtain behavior characteristic information; wherein the first model and the second model are cascaded to each other.
The sequence features comprise a target number clicked by a user, a category number clicked by the user, a position number clicked by the user and time/hour clicked by the user.
The statistical characteristics comprise state information of the user, state information of the target and behavior information of the target to be predicted of the user.
Wherein the first model comprises a DIEN model; the second model comprises an FM model.
The method for obtaining the first characteristic information by extracting the characteristic information of the sequence characteristic by using the first model of the prediction model comprises the following steps: mapping the sequence features into sequence feature vectors by using a feature extraction network; the sequence feature vector comprises a behavior sequence vector of a user, a behavior sequence time vector, a target vector to be predicted and a time point vector to be predicted; training the sequence feature vector by using a gate control circulation network and an attention mechanism to obtain a behavior interest vector of a user; and splicing the behavior interest vector, the target vector to be predicted and the time point vector to be predicted, and training by utilizing a full-connection layer network to obtain first characteristic information.
After the step of splicing the first characteristic information and the second characteristic information to obtain the behavior characteristic information, the method further includes: and predicting the behavior characteristic information by using a full-connection layer network and a normalized index function to obtain the click probability of the target to be predicted.
Before the step of predicting the click probability of the user clicking the target to be predicted at the time point to be predicted based on the behavior characteristic information by using the prediction model, the method further comprises the following steps: acquiring behavior data of a user in a historical time period and a real target number clicked at a time point to be predicted; the behavior data comprises behaviors of the user in a historical time period and behavior time points corresponding to the behaviors; inputting behavior data of a historical time period into an initial model, and predicting a click target of a user at a time point to be predicted by using the initial model based on the behavior data of the historical time period to obtain a predicted target number; and training the initial model by using the real target number and the prediction target number, and determining the trained model as a prediction model.
The method comprises the following steps of training an initial model by using a real target number and a prediction target number, and determining the trained model as a prediction model, wherein the steps of training the initial model by using the real target number and the prediction target number further comprise the following steps: detecting the trained model through a detection sample; and determining that the model training is finished in response to the fact that the prediction precision of the trained model reaches the preset precision.
In order to solve the above problem, the present application further provides an intelligent terminal, where the intelligent terminal includes a processor and a memory, which are coupled to each other, the memory is used to store program instructions, and the processor is used to execute the program instructions stored in the memory to implement the click rate prediction method of any of the above embodiments.
In order to solve the above problem, the present application further provides a computer-readable storage medium, which includes a processor and a memory, where the memory stores computer program instructions, and the processor is configured to execute the program instructions to implement the click rate prediction method according to any one of the above embodiments.
The beneficial effect of this application is: the behavior of the user and the behavior time point of the user are preprocessed by acquiring the behavior of the user in a historical time period, the behavior time point corresponding to the behavior, the target number to be predicted and the time point to be predicted, behavior characteristic information of the user is obtained, and the click probability of the user clicking the target to be predicted (anchor) at the time point to be predicted is predicted by using a prediction model. The click probability of the user to different anchor broadcasters at specific time and under a specific page is predicted through the prediction model, so that the anchor broadcasters interested by the user are recommended to the user, the user experience of watching the live broadcast of the user is guaranteed, and the retention rate of the user in a live broadcast platform is improved. In addition, compared with other shopping prediction models, the prediction model provided by the application considers some simple characteristics of the user, the clicking time period of the user and the clicking position of the user, models are built according to the characteristics of the user, the clicking time period and the clicking position, and the prediction accuracy of the prediction model is effectively improved.
Drawings
FIG. 1 is a schematic flow chart illustrating an embodiment of a click rate prediction method according to the present application;
FIG. 2 is a flowchart illustrating one embodiment of step S13 of FIG. 1;
FIG. 3 is a schematic structural diagram of an embodiment of a predictive model of the present application;
FIG. 4 is a schematic flow chart diagram illustrating an embodiment of a click-through rate prediction model of the present application;
FIG. 5 is a flowchart illustrating one embodiment of step S43 of FIG. 4;
FIG. 6 is a schematic structural diagram of an embodiment of an intelligent terminal according to the present application;
FIG. 7 is a schematic structural diagram of an embodiment of a computer-readable storage medium according to the present application.
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.
It should be noted that, compared with the user in the shopping scene, the user in the live scene has the following differences: 1) the anchor of the live scene is relatively more concentrated, with the vast majority of users gathering in a small number of live rooms of the anchor of the head. According to data display of a certain platform, the first 1% of anchor attracts more than 80% of traffic, and the phenomenon shows that most of user interests in a live broadcast scene are concentrated on a large anchor, and the mining requirement on long-tail data is weaker than that of an e-commerce platform. Because the number of the anchor required to be depicted is small, the simple features of some users are constructed, and certain effect can be achieved by using machine learning models such as XG-boost. 2) In a live scene, the interests of the user at different time periods vary greatly. The users may open the live broadcast platform differently in different time periods due to the influence of the main broadcast and the ecological influence of the platform. 3) In a live broadcast scene, the interests of users in different pages are also greatly different, and the users are in specific pages, which means that the users are only interested in specific content, for example, under a specific category hero alliance page, the users are only interested in a host broadcasting hero alliance content.
According to the specific scene of the live broadcast platform, the click rate prediction method is provided and applied to the live broadcast scene. Referring to fig. 1, fig. 1 is a schematic flow chart illustrating an embodiment of a click rate prediction method according to the present application. As shown in fig. 1, includes:
step S11: and acquiring behaviors of the user in a historical time period, behavior time points corresponding to the behaviors, target numbers to be predicted and time points to be predicted.
The behavior of the user comprises click data generated by clicking the target by the user and state data of the user and the clicked target. In this embodiment, the target refers to a anchor of the live broadcast platform, and the target number to be predicted refers to an anchor number to be predicted. The time point to be predicted refers to the click time of clicking the anchor to be predicted.
Specifically, the click data generated by clicking the anchor by the user includes: an anchor ID clicked by a user; item ID clicked by the user, wherein the item ID includes the item in which the anchor is located, for example: royal glory, hero union, etc.; the location ID of the user click, which includes the page on which the click was made.
It should be noted that the behavior of the user further includes behavior data statistics of the user, and specifically includes: viewing days, total viewing duration, maximum single-day viewing duration, average-day viewing duration, number of viewing anchor, number of categories viewed, number of pages visited, seven-day-history visits, visits to the top 10 categories, number of anchor subscriptions, etc., as well as anchor behavior data statistics related to user behavior data statistics, such as: anchor grade, live duration, live day, live hours, user total watch duration, the biggest user watches duration, user average watch duration etc. and the user is to the behavioral data statistics of anchor, include: the number of watching days, the watching duration, the single-day maximum watching duration, the daily average watching duration, the access amount, the number of times of starting watching at different moments, the total watching duration of the categories where the anchor is located, the access amount to the anchor in the historical time period and the like.
In this embodiment, the historical time period includes a self-defined preset time period, for example, a past seven-day time period or other preset time periods are selected as the historical time period, and behavior data of the user in the past seven days is specifically collected, which is not limited herein.
Step S12: and preprocessing the behaviors and the behavior time points to obtain the behavior characteristic information of the user.
The method specifically comprises the following steps: and performing correlation processing on the collected behaviors of the user and the behavior time points to obtain behavior characteristic information of the behaviors of the user at the behavior time points. Wherein the action time point is within a preset time period. The behavior feature information includes a sequence feature and a statistical feature.
It should be noted that: the sequence feature refers to a sequence formed by the change of the user behavior with time, and in this embodiment, the sequence feature includes: content of the user's first fifty clicks within a preset time period. The statistical characteristics are statistical data of the user historical behaviors, and in this embodiment, the statistical characteristics include: and in a preset time period, the state information of the user, the state information of the anchor to be predicted and the statistical data of the behavior of the user on the anchor to be predicted.
Specifically, in the present embodiment, the sequence features include: a) a target number clicked by the user, for example, an anchor ID clicked by the user; b) item number clicked by the user, for example, item ID clicked by the user (item where the anchor is located); c) a location number clicked by the user, for example, a location ID clicked by the user; d) time of user click/hour.
The statistical features include: a) status data of the user, for example: subscribing to the number of anchor, user level, etc.; b) user behavior data statistics and characterization of user recent activity, such as: viewing days, total viewing duration, maximum single-day viewing duration, daily average viewing duration, number of viewing anchor, number of viewing categories, number of pages visited, visits made during seven days of history, visits made during the previous 10 categories, etc.; c) status data of the anchor, for example: the category, the anchor level, the number of anchor subscriptions, etc.; d) anchor recent behavior data statistics, such as: live broadcast time length, live broadcast days, live broadcast hours, user total watching time length, maximum user watching time length, user average watching time length and the like; e) the statistics of the behavior data of the user on the anchor comprise the following steps: the number of watching days, the watching duration, the single-day maximum watching duration, the daily average watching duration, the access amount, the number of times of starting watching at different moments, the total watching duration of the category where the anchor is located and the access amount to the anchor in the historical time period.
Step S13: and predicting the click probability of the user clicking the target to be predicted at the time point to be predicted based on the behavior characteristic information by using the prediction model.
The method specifically comprises the following steps: the method comprises the steps of respectively utilizing a first model of a prediction model to conduct network training on sequence characteristics of a user to obtain first characteristic information, utilizing a second model of the prediction model to conduct network training on statistical characteristics of the user to obtain second characteristic information, utilizing a splicing layer network to splice the first characteristic information and the second characteristic information to obtain a hidden interest vector of the user, and utilizing a full connection layer and a normalization index function to conduct prediction processing on the hidden interest vector to obtain the click probability of a target to be predicted. Wherein the first model and the second model are cascaded to each other. The first model comprises a DIEN model and the second model comprises an FM model. The DIEN model captures the interests of the user from the rich historical behaviors of the user and gives recommended commodities meeting the interests of the user. The FM model combines the first-order features pairwise to generate second-order features, and predicts by using the first-order and second-order features simultaneously.
Specifically, referring to fig. 2, fig. 2 is a schematic flow chart of an embodiment of step S13 in fig. 1. As shown in fig. 2, includes:
step S21: and extracting the characteristic information of the sequence characteristics by using a first model of the prediction model to obtain first characteristic information.
Wherein the first model comprises a DIEN model. Specifically, extracting feature information of the sequence feature using the first model includes: step 1: the method comprises the steps of mapping sequence features into sequence feature vectors by using a feature extraction network, wherein the feature extraction network Layer comprises an embedded Layer network (DIEN Embedding Layer) and a word vector correlation model (word2vec), specifically, training an anchor ID by using the word vector correlation model to obtain an expression vector of the anchor ID, and mapping the input anchor ID, a category ID, a position ID, time and the like into sequence feature vectors by using the embedded Layer network, wherein the sequence feature vectors comprise feature information of the sequence features. Step 2: training the sequence feature vector by using a gate control loop network (GRU) and an Attention mechanism (Attention) to obtain a behavior interest vector of the user. Specifically, the behavior of the user and the behavior time point are correlated by using a gate control cycle network to form a behavior time sequence of the user, and then the behavior time sequence of the user, the target vector to be predicted and the time point vector to be predicted are trained by using the gate control cycle network and an attention mechanism to obtain a behavior interest vector of the user. And step 3: and splicing the behavior interest vector, the target vector to be predicted and the time point vector to be predicted, and training by utilizing a full connection layer network (Dense) to obtain first characteristic information. Wherein, an embedded Layer network (DIEN Embedding Layer), a word vector correlation model (word2vec), a gate control loop network (GRU), an Attention mechanism (Attention), a splice Layer network (Concat) and a full connection Layer network (sense) are all partial network layers of the prediction model.
Step S22: and extracting the characteristic information of the statistical characteristics by using a second model of the prediction model to obtain second characteristic information.
Wherein the second model comprises an FM model.
It should be noted that, compared with the traditional recommendation algorithm (such as the collaborative filtering algorithm), the factorization model (FM Layer) combines two by two the first-order features to generate the second-order features, and uses the first-order and second-order features for prediction. The second-order characteristics extracted by the FM model can obviously increase the representation capability of the model and obtain a more accurate prediction result; on the other hand, the approximate solution obtained using matrix decomposition requires little computation effort. Therefore, FM has wide application in recommendation systems.
Step S23: and splicing the first characteristic information and the second characteristic information to obtain the behavior characteristic information of the user.
Specifically, the first feature information and the second feature information are spliced by using a splicing layer network (Concat).
In the present embodiment, the first model and the second model are cascaded with each other.
Step S24: and predicting the behavior characteristic information of the user by utilizing a full-connection layer network and a normalized index function to obtain the click probability of the target to be predicted.
For a more clear explanation of the prediction model in the present application, please further refer to fig. 3, and fig. 3 is a schematic structural diagram of an embodiment of the prediction model in the present application. As shown in fig. 3, a behavior sequence, a behavior time point, a target to be predicted, and a time point to be predicted of a user are obtained, where the time point to be predicted is a click time of the user in a next period, for example, if the user opens a live broadcast platform at 12 points, a click probability that the user clicks an anchor to be predicted at 12 points is predicted, and the time point to be predicted also includes a time period to be predicted, for example, a time period from 12 points to 1 point, which is not limited herein.
And converting the behavior sequence, the behavior time point, the target to be predicted and the time point to be predicted of the user into vectors by using a DIEN Embedding Layer network (DIEN Embedding Layer), and performing correlation processing on the behavior sequence and the behavior time point to form a behavior sequence vector. And then, a behavior sequence of the user is processed by using a gate control loop network (GRU) to obtain a hidden interest vector h (T) of the user. And then inputting the hidden interest vector, the target vector to be predicted and the time point vector to be predicted into a gate control loop network (GRU) and an Attention mechanism (Attention), and performing prediction training on the behavior sequence of the user by using the gate control loop network (GRU) and the Attention mechanism (Attention) to obtain a behavior interest vector h' (T) of the user. And finally, splicing the behavior interest vector h' (T), the target vector to be predicted and the time point vector to be predicted by using a splicing layer network (concat), and inputting the spliced behavior interest vector, the target vector to be predicted and the time point vector to be predicted into a full-connection layer network (Dense) for primary processing to obtain first characteristic information of the user.
On the other hand, a statistical feature vector of the user is obtained, wherein the statistical feature is statistical data of historical behaviors of the user, and the statistical feature can be obtained by a statistical system of the live broadcast platform. And preprocessing the statistical feature vector of the user by utilizing a factorization model (FM Layer) to obtain second feature information of the user.
And splicing the first characteristic information and the second characteristic information by using a splicing layer network (concat) to obtain behavior characteristic information of the user, and then predicting the behavior characteristic information of the user by using a full connection layer network (Dense) and a normalized exponential function (Softmax) to obtain the click probability of the user clicking the target to be predicted at the time point to be predicted.
The beneficial effect of this application is: the behavior of the user and the behavior time point of the user are preprocessed by acquiring the behavior of the user in a historical time period, the behavior time point corresponding to the behavior, the target number to be predicted and the time point to be predicted, behavior characteristic information of the user is obtained, and the click probability of the user clicking the target to be predicted (anchor) at the time point to be predicted is predicted by using a prediction model. The click probability of the user to different anchor broadcasters at specific time and under a specific page is predicted through the prediction model, so that the anchor broadcasters interested by the user are recommended to the user, the user experience of watching the live broadcast of the user is guaranteed, and the retention rate of the user in a live broadcast platform is improved. In addition, compared with other shopping prediction models, the prediction model provided by the application considers some simple characteristics of the user, the clicking time period of the user and the clicking position of the user, models are built according to the characteristics of the user, the clicking time period and the clicking position, and the prediction accuracy of the prediction model is effectively improved.
Please further refer to fig. 4, where fig. 4 is a schematic flowchart illustrating a process of an embodiment of the click rate prediction model according to the present application. As shown in fig. 4, includes:
step S41: and acquiring behavior data of the user in a historical time period and the actual target number clicked at the time point to be predicted.
The behavior data comprises behaviors of the user in the historical time period and behavior time points corresponding to the behaviors. The behavior comprises click data generated by clicking the target by the user and state data of the user and the clicked target. The method specifically comprises the following steps: click data such as user click anchor ID, user click category ID, user click position ID, user click time and the like, user state information, state information of the anchor to be predicted, behavior information of the anchor to be predicted by the user and the like.
Step S42: and inputting the behavior data of the historical time period into the initial model, and predicting the click target of the user at the time point to be predicted by using the initial model based on the behavior data of the historical time period to obtain the number of the predicted target.
The method specifically comprises the following steps: and mapping the behavior data of the historical time period into a sequence characteristic and a statistical characteristic. Wherein, the sequence feature and the statistical feature are represented by vectors. The initial model is predictive trained based on the sequence features and the vector representation of the statistical features. Specifically, a DIEN model is used for carrying out feature extraction on sequence features to obtain first feature information, an FM model is used for carrying out feature extraction on statistical features to obtain second feature information, after splicing processing is carried out on the first feature information and the second feature information, the spliced first feature information and the spliced second feature information are predicted by using a full connection layer and a normalized exponential function, and a predicted target (anchor) number is obtained.
Step S43: and training the initial model by using the real target number and the prediction target number, and determining the trained model as a prediction model.
Specifically, it is determined that training is completed until the predicted target number predicted by the initial model is the same as the target number actually clicked by the user, and the model after training is determined as the predicted model.
In order to ensure the prediction accuracy of the prediction model in the present application, the present application trains the model by detecting the sample, and further, please refer to fig. 5, in which fig. 5 is a flowchart illustrating an embodiment of step S43 in fig. 4. As shown in fig. 5, includes:
step S51: and detecting the trained model through the detection sample.
The detection sample comprises a plurality of groups of behavior data of the user in a historical time period and a real target number clicked at a time point to be predicted. And respectively training the trained models by utilizing each group of detection samples, and recording the prediction precision of the models.
Step S52: and determining that the model training is finished in response to the fact that the prediction precision of the trained model reaches the preset precision.
Wherein, predetermine the precision and include: the prediction accuracy rate reaches more than 70%.
The beneficial effect of this embodiment is: the method comprises the steps of acquiring behavior data of a user in a historical time period and a real target number clicked at a time point to be predicted, inputting the behavior data of the historical time period into an initial model, predicting a clicked target of the user at the time point to be predicted based on the behavior data of the initial model based on the historical time period to obtain a predicted target number, training the initial model by using the real target number and the predicted target number, and determining the trained model as the predicted model, so that the establishment of the predicted model is realized, and the method is applied to the prediction of the click rate of a live broadcast platform to a target anchor broadcast.
The method and the device further utilize an experimental data set which is constructed by a certain live broadcast platform and comprises 600w to carry out effect detection on the FM model, the DIEN model and the prediction model in the application. Specifically, the experimental data set was as follows 7: and 3, dividing the test sample set into a training sample set and a detection sample set, respectively training the FM model, the DIEN model and the prediction model in the application by using the training sample set to obtain the FM prediction model, the DIEN prediction model and the prediction model in the application, then detecting the prediction model by using the detection sample set, and measuring the AUC index and the precision index. The results are shown in the following table:
as can be seen from the above table, the AUC index and precision index of the prediction model in the present application both reach above 90%, which indicates that the prediction model in the present application has a better prediction effect on the target anchor when applied to a live broadcast scene.
Referring to fig. 6, fig. 6 is a schematic structural diagram of an embodiment of the intelligent terminal of the present application, and as shown in fig. 6, the intelligent terminal 60 includes a processor 601 and a memory 602, which are coupled to each other, where the processor 601 is configured to execute program instructions stored in the memory 602 to implement steps in any one of the method embodiments or steps correspondingly executed by a client in any one of the method embodiments. The terminal may include a touch screen, a printing component, a communication circuit, etc. according to requirements, in addition to the processor 601 and the memory 602, which is not limited herein.
In particular, the processor 601 is adapted to control itself and the memory 602 to implement the steps in any of the above-described method embodiments. Processor 601 may also be referred to as a CPU (Central Processing Unit). The processor 601 may be an integrated circuit chip having signal processing capabilities. The Processor 601 may also be a general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic, discrete hardware components. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. In addition, the processor 601 may be commonly implemented by a plurality of integrated circuit chips.
The present application further provides a computer-readable storage medium, as shown in fig. 7, fig. 7 is a schematic structural diagram of an embodiment of the computer-readable storage medium of the present application.
The computer-readable storage medium 70 includes a computer program 701 stored on the computer-readable storage medium 70, and when the computer program 701 is executed by the processor, the steps in any of the above method embodiments or the steps executed by the click rate prediction method for network time service in the above method embodiments are implemented correspondingly.
In particular, the integrated unit, if implemented in the form of a software functional unit and sold or used as a separate product, may be stored in a computer readable storage medium 70. Based on such understanding, the technical solution of the present application may be substantially implemented or contributed to by the prior art, or all or part of the technical solution may be embodied in a software product, which is stored in a computer-readable storage medium 70 and includes several instructions for causing a computer device (which may be a personal computer, a server, a network device, or the like) or a processor (processor) to execute all or part of the steps of the method of the embodiments of the present application. And the aforementioned computer-readable storage medium 70 includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
In the several embodiments provided in the present application, it should be understood that the disclosed method and apparatus may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, a division of a module or a unit is merely a logical division, and an actual implementation may have another division, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection of devices or units through some interfaces, and may be in an electrical, mechanical or other form.
Units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, 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 embodiment.
In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be substantially implemented or contributed to by the prior art, or all or part of the technical solution may be embodied in a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, a network device, or the like) or a processor (processor) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
The above embodiments are merely examples and are not intended to limit the scope of the present disclosure, and all modifications, equivalents, and flow charts using the contents of the specification and drawings of the present disclosure or those directly or indirectly applied to other related technical fields are intended to be included in the scope of the present disclosure.
Claims (11)
1. A click through rate prediction method is characterized by comprising the following steps:
acquiring behaviors of a user in a historical time period, behavior time points corresponding to the behaviors, target numbers to be predicted and time points to be predicted; the behavior of the user comprises click data generated by clicking a target by the user and state data of the user and the clicked target;
preprocessing the behaviors and the behavior time points to obtain behavior characteristic information of the user;
and predicting the click probability of the user clicking the target to be predicted at the time point to be predicted based on the behavior characteristic information by using a prediction model.
2. The click rate prediction method according to claim 1, wherein the step of preprocessing the behavior and the behavior time point to obtain the behavior feature information of the user comprises:
mapping the behavior and the behavior time point into sequence features and statistical features;
the step of predicting the click probability of the user clicking the target to be predicted at the time point to be predicted based on the behavior feature information by using a prediction model comprises the following steps:
extracting feature information of the sequence features by using a first model of a prediction model to obtain first feature information;
extracting feature information of the statistical features by using a second model of the prediction model to obtain second feature information;
splicing the first characteristic information and the second characteristic information to obtain the behavior characteristic information;
wherein the first model and the second model are cascaded with each other.
3. The click rate prediction method according to claim 2, wherein the sequence feature comprises a target number of the user click, a category number of the user click, a position number of the user click, or a time/hour of the user click.
4. The click rate prediction method according to claim 2, wherein the statistical characteristics include state information of the user, state information of the target, and behavior information of the target to be predicted by the user.
5. The click-through rate prediction method of claim 2, wherein the first model comprises a DIEN model; the second model comprises an FM model.
6. The click rate prediction method according to claim 5, wherein the step of extracting feature information of the sequence feature by using the first model of the prediction model to obtain first feature information comprises:
mapping the sequence features into sequence feature vectors by using a feature extraction network; the sequence feature vector comprises a behavior sequence vector of a user, a behavior sequence time vector, a target vector to be predicted and a time point vector to be predicted;
training the sequence feature vector by using a gate control circulation network and an attention mechanism to obtain a behavior interest vector of a user;
and splicing the behavior interest vector, the target vector to be predicted and the time point vector to be predicted, and training by utilizing a full-connection layer network to obtain the first characteristic information.
7. The click rate prediction method according to claim 2, wherein after the step of performing the stitching processing on the first feature information and the second feature information to obtain the behavior feature information, the method further comprises:
and predicting the behavior characteristic information by using a full-connection layer network and a normalized index function to obtain the click probability of the target to be predicted.
8. The click rate prediction method according to claim 2, wherein before the step of predicting, by using a prediction model, the click probability of the user clicking the target to be predicted at the time point to be predicted based on the behavior feature information, the method further comprises:
acquiring behavior data of a user in a historical time period and a real target number clicked at the time point to be predicted; the behavior data comprises behaviors of the user in a historical time period and behavior time points corresponding to the behaviors;
inputting the behavior data of the historical time period into an initial model, and predicting the click target of the user at the time point to be predicted by using the initial model based on the behavior data of the historical time period to obtain a predicted target number;
and training the initial model by using the real target number and the prediction target number, and determining the trained model as a prediction model.
9. The click rate prediction method according to claim 8, wherein the step of training the initial model by using the real object number and the predicted object number and determining the trained model as the predicted model further comprises:
detecting the trained model through a detection sample;
and determining that the model training is finished in response to the fact that the prediction precision of the trained model reaches the preset precision.
10. An intelligent terminal, characterized in that the intelligent terminal comprises a processor and a memory coupled to each other, the memory is used for storing program instructions, and the processor is used for executing the program instructions stored in the memory to implement the click rate prediction method according to any one of the preceding claims 1 to 9.
11. A computer-readable storage medium comprising a processor and a memory, the memory storing computer program instructions, the processor being configured to execute the program instructions to implement the click rate prediction method of any one of claims 1-9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110565423.8A CN113297486B (en) | 2021-05-24 | 2021-05-24 | Click rate prediction method and related device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110565423.8A CN113297486B (en) | 2021-05-24 | 2021-05-24 | Click rate prediction method and related device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113297486A true CN113297486A (en) | 2021-08-24 |
| CN113297486B CN113297486B (en) | 2024-04-19 |
Family
ID=77324243
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110565423.8A Active CN113297486B (en) | 2021-05-24 | 2021-05-24 | Click rate prediction method and related device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113297486B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113965805A (en) * | 2021-10-22 | 2022-01-21 | 北京达佳互联信息技术有限公司 | Prediction model training method and device and target video editing method and device |
| CN114710686A (en) * | 2022-03-11 | 2022-07-05 | 武汉斗鱼鱼乐网络科技有限公司 | Exposure method, device, medium and equipment for live broadcast room |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110020877A (en) * | 2018-01-09 | 2019-07-16 | 腾讯科技（深圳）有限公司 | The prediction technique of clicking rate, the determination method of clicking rate and server |
| CN110858912A (en) * | 2018-08-22 | 2020-03-03 | 中国电信股份有限公司 | Streaming media caching method and system, caching policy server and streaming service node |
| CN110929206A (en) * | 2019-11-20 | 2020-03-27 | 腾讯科技（深圳）有限公司 | Click rate estimation method and device, computer readable storage medium and equipment |
| CN111539512A (en) * | 2020-03-31 | 2020-08-14 | 北京三快在线科技有限公司 | Behavior prediction method and device, electronic equipment and readable storage medium |
-
2021
- 2021-05-24 CN CN202110565423.8A patent/CN113297486B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110020877A (en) * | 2018-01-09 | 2019-07-16 | 腾讯科技（深圳）有限公司 | The prediction technique of clicking rate, the determination method of clicking rate and server |
| CN110858912A (en) * | 2018-08-22 | 2020-03-03 | 中国电信股份有限公司 | Streaming media caching method and system, caching policy server and streaming service node |
| CN110929206A (en) * | 2019-11-20 | 2020-03-27 | 腾讯科技（深圳）有限公司 | Click rate estimation method and device, computer readable storage medium and equipment |
| CN111539512A (en) * | 2020-03-31 | 2020-08-14 | 北京三快在线科技有限公司 | Behavior prediction method and device, electronic equipment and readable storage medium |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113965805A (en) * | 2021-10-22 | 2022-01-21 | 北京达佳互联信息技术有限公司 | Prediction model training method and device and target video editing method and device |
| CN114710686A (en) * | 2022-03-11 | 2022-07-05 | 武汉斗鱼鱼乐网络科技有限公司 | Exposure method, device, medium and equipment for live broadcast room |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113297486B (en) | 2024-04-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110941740B (en) | Video recommendation method and computer-readable storage medium | |
| US10958748B2 (en) | Resource push method and apparatus | |
| CN110781391B (en) | Information recommendation method, device, equipment and storage medium | |
| CN110543598B (en) | Information recommendation method and device and terminal | |
| EP4181026A1 (en) | Recommendation model training method and apparatus, recommendation method and apparatus, and computer-readable medium | |
| US20220284327A1 (en) | Resource pushing method and apparatus, device, and storage medium | |
| CN111444428A (en) | Information recommendation method and device based on artificial intelligence, electronic equipment and storage medium | |
| CN111382361B (en) | Information pushing method, device, storage medium and computer equipment | |
| CN111178970B (en) | Advertisement putting method and device, electronic equipment and computer readable storage medium | |
| CN110008397B (en) | Recommendation model training method and device | |
| CN110880127B (en) | Consumption level prediction method and device, electronic equipment and storage medium | |
| CN111178983B (en) | User gender prediction method, device, equipment and storage medium | |
| CN110472154A (en) | A kind of resource supplying method, apparatus, electronic equipment and readable storage medium storing program for executing | |
| CN110910201B (en) | Information recommendation control method and device, computer equipment and storage medium | |
| CN113297486A (en) | Click rate prediction method and related device | |
| CN110598120A (en) | Behavior data based financing recommendation method, device and equipment | |
| CN110247974A (en) | Information-pushing method, device, computer and storage medium based on block chain | |
| CN109801101A (en) | Label determines method, apparatus, computer equipment and storage medium | |
| CN114461919A (en) | Information recommendation model training method and device | |
| CN112269943B (en) | Information recommendation system and method | |
| CN115439770A (en) | Content recall method, device, equipment and storage medium | |
| CN111768218B (en) | Method and device for processing user interaction information | |
| US20230316106A1 (en) | Method and apparatus for training content recommendation model, device, and storage medium | |
| CN111083534B (en) | Method and equipment for providing recommended video list | |
| CN115374348A (en) | Information recommendation method, information recommendation device and readable 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |