Training method of estimation model, distribution quantity estimation method, device and equipment, and medium

Abstract

The present disclosure provides a training method of a prediction model, a distribution quantity prediction method and device, electronic equipment and a storage medium, and relates to the technical field of computers, in particular to the technical field of deep learning. The specific implementation scheme is: obtaining sample data and sample distribution quantity; determining a target distribution quantity interval to which the sample data belongs according to the sample distribution quantity; extracting sample features of the sample data; and performing model training according to the sample features and the sample distribution quantity to obtain a prediction model corresponding to the target distribution quantity interval. According to the present disclosure, sample data is classified according to sample distribution quantity, a target distribution quantity interval to which the sample data belongs is determined, and then model training is performed according to sample data in different target distribution quantity intervals, so that a prediction model corresponding to the target distribution quantity interval is obtained, and the prediction accuracy of the prediction model is effectively improved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of computer, in particular to the technical field of deep learning, and especially to a training method of a prediction model, a distribution quantity prediction method, an apparatus, a device and a medium. BACKGROUND

[0002] The reading quantity or click quantity of data after publication directly affects its conversion rate, for example, articles or commercial orders, etc. If the distribution quantity of data is closer to its actual reading quantity and click quantity, the conversion rate is higher. If the distribution quantity is much larger than the actual reading quantity or click quantity, it will lead to waste of resources. Conversely, it will seriously affect the income. Therefore, the prediction accuracy of the data distribution quantity directly affects the resource occupation and income situation. SUMMARY

[0003] The present disclosure provides a training method of a prediction model, a distribution quantity prediction method, an apparatus, a device and a medium, which can effectively improve the prediction accuracy of the distribution quantity.

[0004] According to an aspect of the present disclosure, a training method of a prediction model is provided, comprising:

[0005] obtaining sample data and sample distribution quantity;

[0006] determining a target distribution quantity interval to which the sample data belongs according to the sample distribution quantity;

[0007] performing model training according to the sample features and the sample distribution quantity to obtain a prediction model corresponding to the target distribution quantity interval.

[0008] According to a second aspect of the present disclosure, a distribution quantity prediction method is provided, comprising:

[0009] obtaining target data;

[0010] determining a preliminary prediction result of the target data;

[0011] determining a prediction model according to a target distribution quantity interval corresponding to the preliminary prediction result, wherein the prediction model is obtained by the method provided in the first aspect;

[0012] inputting data features of the target data into the prediction model to obtain a predicted distribution quantity of the target data.

[0013] According to a third aspect of the present disclosure, a training apparatus of a prediction model is provided, comprising:

[0014] a first obtaining module configured to obtain sample data and sample distribution quantity;

[0015] a first determining module configured to determine a target distribution quantity interval to which the sample data belongs according to the sample distribution quantity;

[0016] an extraction module configured to extract a sample feature of the sample data;

[0017] a training module configured to perform model training according to the sample feature and the sample distribution amount, to obtain a prediction model corresponding to a target distribution amount interval.

[0018] According to a fourth aspect of the present disclosure, a distribution amount prediction apparatus is provided, comprising:

[0019] a second acquisition module configured to acquire target data;

[0020] a second determination module configured to determine a preliminary prediction result of the target data;

[0021] a third determination module configured to determine a prediction model according to a target distribution amount interval corresponding to the preliminary prediction result, wherein the prediction model is obtained by the method provided in the first aspect;

[0022] a prediction module configured to input a data feature of the target data into the prediction model, to obtain a predicted distribution amount of the target data.

[0023] According to a fifth aspect of the present disclosure, an electronic device is provided, comprising:

[0024] at least one processor; and

[0025] a memory in communication with the at least one processor; wherein

[0026] the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method provided in the first aspect or the second aspect.

[0027] According to a sixth aspect of the present disclosure, a non-transitory computer readable storage medium storing computer instructions is provided, the computer instructions being used to enable a computer to perform the method provided in the first aspect or the second aspect.

[0028] According to a seventh aspect of the present disclosure, a computer program product is provided, comprising a computer program, the computer program being executed by a processor to implement the method provided in the first aspect or the second aspect.

[0029] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the present disclosure, nor to limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0030] The accompanying drawings are used to better understand the present scheme, and do not limit the present disclosure. Among them:

[0031] Figure 1 is an exemplary system architecture to which the training method of the estimation model or the distribution quantity estimation method of the present disclosure can be applied;

[0032] Figure 2 is a flowchart of one embodiment of the training method of the estimation model according to the present disclosure;

[0033] Figure 3 is a schematic diagram of one embodiment of the training method of the estimation model according to the present disclosure;

[0034] Figure 4 is a flowchart of one embodiment of the distribution quantity estimation method according to the present disclosure;

[0035] Figure 5 is a schematic diagram of one embodiment of the distribution quantity estimation method according to the present disclosure;

[0036] Figure 6 is a schematic diagram of one embodiment of the training device of the estimation model according to the present disclosure;

[0037] Figure 7 is a schematic diagram of one embodiment of the distribution quantity estimation device according to the present disclosure;

[0038] Figure 8 is a block diagram of an electronic device to implement the training method of the estimation model or the distribution quantity estimation method according to the embodiments of the present disclosure. DETAILED DESCRIPTION

[0039] Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, which include various details of the embodiments of the present disclosure to help the understanding of the present disclosure. These should be considered in the context of the overall description and should not be considered limiting in any way. Thus, it will be apparent to those skilled in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Also, for the sake of brevity and clarity, descriptions of well-known functions and constructions are omitted from the following description.

[0040] The reading quantity or click quantity after the data is published directly affects its conversion rate, for example, articles or commercial orders, etc. If the distribution quantity of the data is closer to its actual reading quantity and click quantity, the conversion rate is higher. If the distribution quantity is much greater than the actual reading quantity or click quantity, it will lead to waste of resources. Conversely, it will seriously affect the revenue. Therefore, the accuracy of the estimation of the data distribution quantity directly affects the resource occupation and revenue situation.

[0041] Taking a commercial order as an example, the commercial order is a marketing mode in which an advertiser purchases an author to create content for the brand of the advertiser, and the reasonableness of the offer of the author directly affects the transaction of the commercial order. Therefore, it is necessary to provide the author with a suggested offer reference by comprehensively considering market and industry competition. The suggested offer is mainly determined by the estimated distribution volume and the author CPM (Cost Per Mille, or Cost Per Thousand Impression, also known as the cost per thousand impressions). Among them, the author CPM calculates the value of each reading elaboration according to the historical commercial order transaction data of the author, and the estimated distribution volume is based on the user browsing data of the historical articles of the author to estimate the reading volume that the commercial order content can obtain in the future.

[0042] In the prior art, the average value of the reading volume of the author's historical creation content in a period of time is often calculated as the estimated distribution volume of the commercial order of the author. In fact, the reading volume of the historical article content of the author has a large fluctuation, and when there is a blockbuster content or a cold content in the historical article content, a large deviation will be caused between the estimated distribution volume and the actual posterior distribution volume, that is, the accuracy of the estimated distribution volume is insufficient.

[0043] The present disclosure classifies the distribution volume of the historical article content, thereby classifying the historical article content and its corresponding features, and training the estimated model corresponding to the distribution volume in different intervals, thereby effectively improving the estimation accuracy of the distribution volume, and further improving the reasonableness of the author's offer, promoting the ecological relationship between the platform and the author and the platform environment, improving customer satisfaction, and ultimately realizing the win-win of the platform, the author and the customer.

[0044] Figure 1 An exemplary system architecture 100 is shown, which shows an embodiment of a method of pushing information or a device of pushing information to which the present disclosure can be applied.

[0045] As shown in Figure 1 The system architecture 100 can include a terminal device 101, a network 102 and a server 103. The network 102 is used to provide a communication link between the terminal device 101 and the server 103, and can include various connection types, such as wired communication links, wireless communication links, optical fiber cables, etc.

[0046] A user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send information, etc. Various client applications can be installed on the terminal device 101.

[0047] The terminal device 101 can be hardware or software. When the terminal device 101 is hardware, it can be various electronic devices including but not limited to a smart phone, a tablet computer, a laptop computer, a desktop computer, and the like. When the terminal device 101 is software, it can be installed in the above-mentioned electronic devices. It can be implemented as multiple software or software modules, or as a single software or software module. No specific limitation is made herein.

[0048] The server 103 can be hardware or software. When the server 103 is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or as a single server. When the server 103 is software, it can be implemented as multiple software or software modules (for example, to provide distributed services), or as a single software or software module. No specific limitation is made herein.

[0049] The training method of the estimation model or the distribution quantity estimation method provided by the embodiments of the present disclosure is generally executed by the server 103, and accordingly, the training device of the estimation model or the distribution quantity estimation device is generally arranged in the server 103.

[0050] It should be noted that, Figure 1 The number of the terminal device 101, the network 102, and the server 103 in the above-mentioned system is only illustrative. According to the implementation needs, there can be any number of terminal devices 101, networks 102, and servers 103.

[0051] In the embodiments of the present disclosure, the training method of the estimation model is executed by the server 103, and the sample data and the sample distribution quantity are obtained from the server 103 or from the terminal device 101 installed with a client through the network 102, for example, the post content and the corresponding distribution quantity in the recent period of time on a certain client platform of the terminal device 101 are obtained, and the model is trained according to the obtained sample data and sample distribution quantity to obtain the estimation model.

[0052] Figure 2 A flow 200 of one embodiment of the training method of the estimation model according to the present disclosure is shown, and with reference to Figure 2 The training method of the estimation model includes the following steps:

[0053] In step S201, sample data and sample distribution quantity are obtained.

[0054] In the present embodiment, the execution subject of the training method of the estimation model, for example, Figure 1The server 103 shown can obtain the sample data and the sample distribution amount in various ways. For example, the sample data and the sample distribution amount can be obtained from a terminal device in a wired or wireless manner, or can be obtained directly from a storage space of the server 103 or a cloud platform.

[0055] The sample data can be all commercial order data on the platform, or can be related data of a certain type of commercial order or a certain field of commercial order on the platform. For example, it can be the creation content of all commercial order related authors on the platform, including distributed articles, article distribution click volume, commercial order conversion volume, and other related data. It can also include information related to the author of the article, such as total browsing volume, article volume, vertical category, and attention volume.

[0056] The sample distribution amount is the distribution amount corresponding to each sample data. In some optional implementation manners, taking a sample data as a commercial order article, the sample distribution amount can be the actual reading volume or the actual click volume of the commercial order article.

[0057] It should be noted that in the technical solutions of the present disclosure, the collection, storage, use, processing, transmission, provision, and disclosure of user personal information (for example, author-related information) comply with relevant laws and regulations and do not violate public order and good customs. The user information obtained in the present embodiment does not target a specific user and cannot reflect the personal information of a specific user.

[0058] In step S202, the target distribution amount interval to which the sample data belongs is determined according to the sample distribution amount.

[0059] In the present embodiment, the execution subject of the training method of the estimation model is, for example, Figure 1 The server 103 shown determines the target distribution amount interval to which each sample data belongs according to the sample distribution amount of the sample data obtained in step S202, thereby classifying each sample data according to the corresponding distribution amount interval.

[0060] In the present embodiment, each sample data is classified according to the corresponding distribution amount interval according to the sample distribution amount, the target distribution amount interval to which each sample data belongs is determined, and the model is trained according to different distribution amount intervals, so that the trained model can accurately estimate the distribution amount of the corresponding interval, improve the estimation accuracy of the model, and avoid insufficient training accuracy of the model due to large differences in sample distribution amount.

[0061] In some optional implementation manners of the present embodiment, determining the target distribution amount interval to which the sample data belongs according to the sample distribution amount includes: determining the target distribution amount interval to which the sample data belongs according to a comparison result of the sample distribution amount and a preset distribution amount threshold.

[0062] In the present implementation, the execution subject divides the sample data into different distribution intervals according to the preset distribution threshold and the sample distribution, so that subsequent model training can be performed for the corresponding distribution interval, to avoid insufficient model training accuracy caused by excessive fluctuation of the sample distribution value, thereby improving the output concentration of the trained model and improving the estimation accuracy of the trained model.

[0063] In the embodiments of the present disclosure, the preset distribution threshold can be a value. For example, it can be a relatively small value, so as to classify sample data with a small sample distribution. For example, if the sample data is mainly distributed in an interval greater than 100, 100 can be used as the preset distribution threshold, so that sample data with a sample distribution less than 100 is separated, and sample data with a sample distribution greater than 100 is used as training samples for model training, thereby improving the estimation accuracy of the trained model.

[0064] Further, the execution subject can also perform model training according to the separated sample data with a sample distribution less than 100, to train a model that can estimate small distribution.

[0065] In some optional implementations, to further improve the model training accuracy, at least two different values can be set to form at least three different distribution intervals, so as to further classify the sample data according to the sample distribution, and perform targeted model training according to the sample data in different distribution intervals, to obtain a model that can accurately estimate the values in the corresponding distribution interval, thereby further improving the estimation accuracy of the model.

[0066] In some optional implementations of the embodiments of the present disclosure, the preset distribution threshold can include 0. That is, the sample data with a sample distribution of 0 is separately classified to avoid adverse effects on the model training process, thereby improving the training accuracy of the model and the estimation accuracy of the trained model.

[0067] Step S203: Extracting sample features of the sample data.

[0068] In the present embodiment, the execution subject of the estimation model training method, for example, Figure 1 The server 103 shown in the figure extracts the sample features of the sample data from the sample data obtained in step S201.

[0069] The sample features of the sample data can include a distribution amount of the sample data itself and the like, and can also include relevant features of an author of the sample data, for example, a historical publishing amount of the author, a median value of the historical publishing amount of the author in a window, a vertical category of the author, an attention amount, an attentioned amount, and the like.

[0070] In some optional implementations, the execution subject can extract the corresponding sample features from the sample data by using a pre-trained feature extraction model or a pre-set feature extraction algorithm. The corresponding feature extraction model or feature extraction algorithm can refer to a similar solution in a related feature extraction technology, which is not limited here.

[0071] In step S204, a model is trained according to the sample features and the sample distribution amount, to obtain a prediction model corresponding to a target distribution amount interval.

[0072] In the embodiment, the execution subject of the training method of the prediction model can be, for example, Figure 1 The server 103 shown in the figure selects a corresponding initial model for training according to the sample features and the sample distribution amount of the sample data classified in step S202, to obtain a prediction model corresponding to a target distribution amount interval to which the sample data belongs.

[0073] In some optional implementations of the embodiment of the disclosure, the initial models corresponding to different distribution amount intervals can be the same or different. The present solution classifies the sample data by different distribution amount intervals, and uses the sample data in different distribution amount intervals as training samples to perform model training, so as to train a prediction model that can accurately predict the corresponding distribution amount in the corresponding distribution amount interval.

[0074] The embodiment of the disclosure classifies the sample data according to the sample distribution amount, determines a target distribution amount interval to which the sample data belongs, and then respectively performs model training according to the sample features and the sample distribution amount of the sample data in different target distribution amount intervals, to obtain a prediction model corresponding to the target distribution amount interval. Different prediction models can accurately predict the values in different target distribution amount intervals, and effectively improve the prediction accuracy of the prediction model.

[0075] In some optional implementations of the embodiment of the disclosure, the sample features include sample statistical features and sample attribute features. Further, the model is trained according to the sample features and the sample distribution amount to obtain a prediction model corresponding to a target distribution amount interval, including: determining a feature vector and a trajectory vector of the sample data according to the sample statistical features; and training the model according to the feature vector, the trajectory vector, the sample attribute features, and the sample distribution amount to obtain the prediction model corresponding to the target distribution amount interval.

[0076] In the embodiments of the present disclosure, the execution subject can extract the sample statistical features and the sample attribute features of the sample data respectively, and process the sample statistical features to obtain the feature vector and the trajectory vector of the sample data, and then train a model according to the feature vector, the trajectory vector, the sample attribute features and the sample distribution quantity of the sample data, so as to obtain the estimation model corresponding to the target distribution quantity interval to which the sample data belongs.

[0077] The scheme of the present embodiment can greatly improve the training accuracy of the model by processing the sample statistical features in different ways to obtain the feature vector and the trajectory vector of the sample data, and training the model according to the feature vector, the trajectory vector and the sample attribute features of the sample data, so as to improve the sensitivity of the trained estimation model to the sample statistical features and the sample attribute features, and further improve the estimation accuracy of the estimation model in estimating according to the statistical features and the attribute features of the corresponding data.

[0078] In some optional implementation manners of the embodiments of the present disclosure, the feature vector and the trajectory vector of the sample data are determined according to the sample statistical features, including: obtaining the feature vector of the sample data according to the sample statistical features and a first preset model; and obtaining the trajectory vector of the sample data according to the sample statistical features and a second preset model.

[0079] In the present implementation manner, the execution subject processes the sample statistical features of the sample data by using the first preset model and the second preset model respectively to obtain the feature vector and the trajectory vector. For example, the execution subject inputs the sample statistical features of the sample data into the first preset model to obtain the feature vector of the sample data, and inputs the sample statistical features of the sample data into the second preset model to obtain the trajectory vector of the sample data. The present scheme processes the sample statistical features by using different models for embedding, so as to better describe the sample statistical features of the sample data and improve the training accuracy of the model.

[0080] The first preset model and the second preset model are different types of models or are constructed by using different algorithms. For example, the first preset model can use a method of converting a word into a vector, such as a word vector model word2vec, and the second preset model can use a deep walk algorithm deepwalk.

[0081] In some optional implementation manners of the embodiments of the present disclosure, the estimation model is used for distribution quantity estimation of an article, and the sample data includes a sample article. For example, the estimation model can be used for distribution quantity estimation of a commercial order related article, and can also be used for distribution quantity estimation of a news article or other types of articles. Correspondingly, the sample article can be a commercial order related text, or a news, announcement or other types of text.

[0082] The sample statistical features of the sample data include at least one of the following: user exposure of a sample article author, historical article publishing quantity, and article publishing quantity window median; and the sample attribute features include at least one of the following: vertical category of the sample article author, and attention quantity.

[0083] The embodiments of the present disclosure introduce the related data of the author, introduce the click data of the user dimension data, extract the sample features by using the historical click behavior and preference of the user, can better depict the information of the author and the sample data, and further improve the reliability of the trained estimation model.

[0084] In some optional implementations of the embodiments of the present disclosure, the initial model for training the estimation model can be a CIN regression model, and various values such as the feature vector, the trajectory vector, and the sample attribute feature can be used as inputs to realize the feature interaction at the vector level and guarantee the accuracy of model training and data estimation.

[0085] Figure 3 FIG. 3 shows a schematic diagram of an exemplary scenario 300 of the training method of the estimation model according to the embodiments of the present disclosure. Figure 3 As shown, the execution subject extracts the sample statistical features such as the user exposure of the author of the sample article, the historical article publishing quantity, and the article publishing quantity window median, and extracts the sample attribute features such as the vertical category of the author of the sample article and the attention quantity; then inputs the sample statistical features into the word2vec model to obtain the feature vector, and obtains the trajectory vector by using the deepwalk algorithm on the sample statistical features, and then inputs the feature vector, the trajectory vector, and the sample attribute features of the sample article as inputs into the initial model corresponding to the target distribution interval to which the sample article belongs to perform training, and uses the values within the preset deviation range of the sample distribution quantity of the sample article as the expected output, so as to obtain the estimation model corresponding to the target distribution interval and guarantee the estimation accuracy of the estimation model.

[0086] In some optional implementations, the preset deviation range as the expected output can be determined according to the distribution of the sample data, for example, can be determined according to the maximum and minimum values of the sample distribution quantity, or can be determined according to the mean value or the distribution curve of the sample distribution quantity. Exemplarily, if the difference between the maximum and minimum values of the sample distribution quantity is large, the preset deviation range can be appropriately expanded; on the contrary, if the sample distribution quantity is relatively concentrated, the preset deviation range can be appropriately reduced.

[0087] Exemplarily, the preset deviation range as the expected output can be set to 50%, or can be set to 30%, etc. Taking 50% as an example, if the sample distribution quantity of the sample data is 800, the expected output when the model is trained by using the sample data is 400-1200, that is, the output value between 400 and 1200 is the convergence.

[0088] Through experiments, it is verified that the distribution quantity estimated by the estimation model obtained by using the training method provided in the embodiments of the present disclosure has an estimation accuracy that is 20% higher than that of the distribution quantity estimated by the scheme in the related art.

[0089] It should be noted that, in the technical solutions of the present disclosure, the collection, storage, use, processing, transmission, provision, and disclosure of user personal information (for example, the relevant information of the author) comply with the relevant laws and regulations and do not violate public order and good customs. The user information obtained in the embodiments does not target a specific user and cannot reflect the personal information of a specific user.

[0090] In addition, as an application of the estimation model obtained by using the training method, the present disclosure further provides a distribution quantity estimation method. Figure 1 The server 103 shown as the execution subject can also apply the estimation model obtained by using the training method to estimate the distribution quantity, that is, execute the distribution quantity estimation method.

[0091] Figure 4 The flow 400 of one embodiment of the distribution quantity estimation method of the present disclosure is shown. Referring to Figure 4 The distribution quantity estimation method includes the following steps:

[0092] In step S401, target data is obtained.

[0093] In the embodiments of the present disclosure, the execution subject of the distribution quantity estimation method, for example Figure 1 The server 103 shown can obtain the target data in various ways such as directly or indirectly. For example, the execution subject can obtain the target data from the terminal device 101 directly through the network 102 in a wired or wireless manner; or the execution subject can obtain the target data from other devices or cloud platforms or servers associated with the terminal device 101 by responding to the received data processing request (for example, a distribution quantity estimation request).

[0094] In some optional implementations, the target data can be data such as a commercial article or news to be distributed.

[0095] In step S402, a preliminary estimation result of the target data is determined.

[0096] In the embodiments of the present disclosure, the execution subject of the distribution quantity estimation method, for example Figure 1 The server 103 shown preliminarily estimates the distribution quantity of the target data obtained in step S401 to determine the preliminary estimation result of the target data.

[0097] In some optional implementations of the embodiments of the present disclosure, the preliminary estimation result of the target data is determined by extracting data features of the target data and determining the preliminary estimation result of the target data according to the data features.

[0098] In the present embodiment, the execution subject can extract the corresponding data features from the target data by using a pre-trained feature extraction model or a pre-set feature extraction algorithm. The feature extraction model or the feature extraction algorithm used can refer to similar solutions in related feature extraction technologies, which are not limited here.

[0099] The execution subject uses the extracted data features to preliminarily estimate the distribution volume of the target data to obtain the preliminary estimation result.

[0100] In the present implementation, the distribution volume of the target data is preliminarily estimated according to the data features of the target data, which ensures the reliability of the preliminary estimation result and provides a guarantee for the subsequent accurate estimation.

[0101] For example, the estimation manner for obtaining the preliminary estimation result can be using a trained classification model or a classification algorithm.

[0102] In some optional implementations of the embodiments of the present disclosure, the preliminary estimation result of the target data is determined according to the data features, including inputting the data features into a trained classification model to obtain the preliminary estimation result of the target data.

[0103] In the present implementation, the execution subject inputs the data features of the target data into a trained classification model to obtain the preliminary estimation result of the target data. The execution subject preliminarily estimates the distribution volume of the target data by using the trained classification model, so as to determine the target distribution volume interval to which the distribution volume of the target data belongs, and further ensure the accuracy of the subsequently selected estimation model.

[0104] The classification model can be obtained by training sample data and sample distribution volumes in the training method of the estimation model.

[0105] In some optional implementations of the embodiments of the present disclosure, the classification model is constructed by using a CIN fusion interconnection network and a GBDT gradient boosting decision tree. The classification principle is that the data features of the target data are input into the CIN to obtain a first result, the data features are input into the GBDT to obtain a second result, and the classification result, i.e., the preliminary estimation result, is obtained by weighted average of the first result and the second result.

[0106] In some optional implementations, the preliminary estimation result output by the classification model can be a specific distribution volume value, or information related to the corresponding target distribution volume interval, such as a characteristic value or a number of the target distribution volume interval.

[0107] Exemplarily, the preset distribution amount interval includes three intervals, wherein the first interval [0, 50) is marked as 0, the second interval [50, 200) is marked as 1, and the third interval [200, ∞) is marked as 2. After the data feature of the target data is input into the classification model, if the output of the classification model is 1, the target distribution amount interval corresponding to the preliminary estimation result is [50, 200), and the estimation model for accurate estimation of the distribution amount can be determined according to the target distribution amount interval.

[0108] In step S403, the estimation model is determined according to the target distribution amount interval corresponding to the preliminary estimation result, wherein the estimation model is obtained by the training method of the estimation model of the present disclosure.

[0109] In the embodiments of the present disclosure, the execution subject of the distribution amount estimation method, for example, the server 103 shown in the figure, determines the target distribution amount interval corresponding to the preliminary estimation result determined in step S402, and then determines the estimation model corresponding to the target distribution amount interval. Figure 1

[0110] As described above, the estimation model trained by the training method of the estimation model of the present disclosure includes estimation models corresponding to different distribution amount intervals. Therefore, in the present embodiment, the execution subject selects the estimation model corresponding to the target distribution amount interval to which the preliminary estimation result belongs according to the preliminary estimation result, effectively ensuring the selection accuracy of the estimation model and the accuracy of the distribution amount estimation.

[0111] In step S404, the data feature of the target data is input into the estimation model to obtain the estimated distribution amount of the target data.

[0112] In the embodiments of the present disclosure, the execution subject of the distribution amount estimation method, for example, the server 103 shown in the figure, determines the target distribution amount interval corresponding to the preliminary estimation result determined in step S402, and then determines the estimation model corresponding to the target distribution amount interval. Figure 1

[0113] The embodiments of the present disclosure determine the preliminary estimation result according to the data feature of the target data, and then determine the estimation model according to the target distribution amount interval corresponding to the preliminary estimation result, effectively ensuring the adaptability of the selected estimation model, and then estimating the distribution amount according to the estimation model, which can effectively ensure the accuracy of the estimated distribution amount.

[0114] In some optional implementation manners of the embodiments of the present disclosure, the distribution amount estimation method is used to estimate the distribution amount of an article to be distributed, the target data includes a target article to be distributed, and the data feature of the target data includes author statistical features and author attribute features of the target data.

[0115] Figure 5 ​​A schematic diagram of an exemplary scenario 500 of the distribution quantity estimation method of the embodiments of the present disclosure is shown. Referring to Figure 5 The preliminary estimation result of the target data is L, which can be a value of the distribution quantity, for example. In the embodiments of the present disclosure, after obtaining the target data, the subject of the distribution quantity estimation method extracts the author statistical features and the author attribute features of the target data as data features, inputs the author statistical features and the author attribute features of the target data into the classification model, and obtains the preliminary estimation result L: if L = 0, no further estimation is performed, and it is determined that the estimated distribution quantity of the target data is 0; if 0 < L≤40, a first estimation model corresponding to the target estimation model is determined, and then the data features of the target data are input into the first estimation model for accurate estimation of the distribution quantity; and if L > 40, a second estimation model is determined as the target estimation model, and then the data features of the target data are input into the second estimation model for accurate estimation of the distribution quantity.

[0116] In the embodiments of the present disclosure, the classification result of the classification model can also be post-processed according to the output result of the selected estimation model. If the output is still within the target distribution quantity interval corresponding to the estimation model, it is considered that the estimation result is reliable; and if the output of the estimation model exceeds the target distribution quantity interval corresponding to the estimation model, it is considered that the estimation result is unreliable.

[0117] Taking 0 < L≤40 as an example, the first estimation model is determined as the target estimation model. At this time, if the output S of the first estimation model is 0 < S≤40, for example, S = 25, after the data features of the target data are input into the first estimation model, it is considered that the output results of the classification model and the first estimation model are reliable, that is, the estimated distribution quantity is reliable.

[0118] As an implementation of the method shown in the above figures, Figure 6 An embodiment of a training device of an estimation model according to the present disclosure is shown. The training device of the estimation model corresponds to the method embodiment shown in Figure 2 The device can be applied to various electronic devices.

[0119] As Figure 6 shown, the training device 600 of the estimation model of the embodiments of the present disclosure includes a first acquisition module 601, a first determination module 602, an extraction module 603, and a training module 604. The first acquisition module 601 is configured to acquire sample data and sample distribution quantity; the first determination module 602 is configured to determine a target distribution quantity interval to which the sample data belongs according to the sample distribution quantity; the extraction module 603 is configured to extract sample features of the sample data; and the training module 604 is configured to perform model training according to the sample features and the sample distribution quantity, and obtain an estimation model corresponding to the target distribution quantity interval.

[0120] In the training apparatus 600 of the estimation model in this embodiment, the specific processing of the first acquisition module 601, the first determination module 602, the extraction module 603 and the training module 604 and the technical effects brought by the same can be referred to the corresponding descriptions of the steps S201-S204 in the embodiment. Figure 2 The related descriptions of the steps S201-S204 in the corresponding embodiment are not repeated here.

[0121] In some optional implementations of the embodiments of the present disclosure, the first determination module 620 is configured to determine the target distribution range to which the sample data belongs according to a comparison result of the sample distribution amount and the preset distribution threshold.

[0122] In some optional implementations of the embodiments of the present disclosure, the sample features include sample statistical features and sample attribute features, and the training module 604 includes a first determination unit and a training unit. The first determination unit is configured to determine the feature vector and the trajectory vector of the sample data according to the sample statistical features. The training unit is configured to perform model training according to the feature vector, the trajectory vector, the sample attribute features and the sample distribution amount, to obtain the estimation model corresponding to the target distribution range.

[0123] In some optional implementations of the embodiments of the present disclosure, the first determination unit is configured to obtain the feature vector of the sample data according to the sample statistical features and a first preset model, and obtain the trajectory vector of the sample data according to the sample statistical features and a second preset model.

[0124] In some optional implementations of the embodiments of the present disclosure, the estimation model is used for distribution amount estimation of an article, and the sample data includes a sample article.

[0125] In some optional implementations of the embodiments of the present disclosure, the sample statistical features include at least one of the following: user exposure amount, historical article publishing amount and article publishing amount window median of a sample article author; and the sample attribute features include at least one of the following: vertical category and attention amount of the sample article author.

[0126] Figure 7 An embodiment of a distribution amount estimation apparatus according to the present disclosure is shown. The distribution amount estimation apparatus corresponds to the method embodiment shown in Figure 4 The apparatus 700 can be applied in various electronic devices.

[0127] As Figure 7As shown, the distribution quantity estimation apparatus 700 of the embodiment of the present disclosure comprises a second acquisition module 701, a second determination module 702, a third determination module 703 and an estimation module 704. The second acquisition module 701 is configured to acquire target data. The second determination module 702 is configured to determine a preliminary estimation result of the target data. The third determination module 703 is configured to determine an estimation model according to a target distribution quantity interval corresponding to the preliminary estimation result, wherein the estimation model can be obtained by the training method of the estimation model. Figure 2 The estimation module 704 is configured to input a data feature of the target data into the estimation model to obtain an estimated distribution quantity of the target data.

[0128] In the distribution quantity estimation apparatus 700 of the embodiment, the specific processing of the second acquisition module 701, the second determination module 702, the third determination module 703 and the estimation module 704 and the technical effects brought by the specific processing can be respectively referred to the related descriptions of steps S401-S404 in the corresponding embodiment, which will not be repeated here. Figure 4 The related descriptions of steps S401-S404 in the corresponding embodiment, which will not be repeated here.

[0129] In some optional implementation of the embodiment of the present disclosure, the second determination module 702 comprises an extraction unit and a second determination unit, wherein the extraction unit is configured to extract a data feature of the target data, and the second determination unit is configured to determine a preliminary estimation result of the target data according to the data feature.

[0130] In some optional implementation of the embodiment of the present disclosure, the second determination unit is configured to input the data feature into the trained classification model to obtain the preliminary estimation result of the target data.

[0131] In some optional implementation of the embodiment of the present disclosure, the distribution quantity estimation apparatus 700 is used for distribution quantity estimation of an article to be distributed, wherein the target data comprises a target article to be distributed, and the data feature of the target data comprises author statistical features and author attribute features of the target data.

[0132] According to the embodiments of the present disclosure, the present disclosure further provides an electronic device, a non-transitory computer-readable storage medium storing computer instructions and a computer program product.

[0133] The electronic device comprises at least one processor and a memory connected with the at least one processor in communication. The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the training method of the estimation model or the distribution quantity estimation method.

[0134] In some embodiments, a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the training method of the prediction model or the distribution quantity prediction method.

[0135] In some embodiments, a computer program product includes a computer program which, when executed by a processor, implements the training method of the prediction model or the distribution quantity prediction method.

[0136] Figure 8 A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptops, desktops, tablets, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the present disclosure described and / or claimed in this document.

[0137] As shown, Figure 8 The device 800 includes a computing unit 801 that can perform various appropriate actions and processes in accordance with a computer program stored in a read-only memory (ROM) 802 or a computer program loaded into a random access memory (RAM) 803 from a storage unit 808. Various programs and data required for the operation of the device 800 can also be stored in the RAM 803. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other through a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0138] Various components in the device 800 are connected to the I / O interface 805, including an input unit 806, such as a keyboard, a mouse, etc., an output unit 807, such as various types of displays, speakers, etc., a storage unit 808, such as a magnetic disk, an optical disk, etc., and a communication unit 809, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 809 allows the device 800 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunication networks.

[0139] The computing unit 801 can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs various methods and processes described above, such as the training method of the estimation model or the distribution quantity estimation method. For example, in some embodiments, the training method of the estimation model or the distribution quantity estimation method can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed on the device 800 via the ROM 802 and / or the communication unit 809. When the computer program is loaded into the RAM 803 and executed by the computing unit 801, one or more steps of the training method of the estimation model or the distribution quantity estimation method described above can be performed. Alternatively, in other embodiments, the computing unit 801 can be configured to perform the training method of the estimation model or the distribution quantity estimation method by any other suitable means, such as by means of firmware.

[0140] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0141] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces the functions / operations specified in the flowcharts and / or the block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine and partially on a remote machine or entirely on a remote machine or server.

[0142] In the context of the present disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more lines of electrical connections, portable computer disks, hard disk drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fibers, portable compact disc read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0143] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0144] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0145] The computer system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0146] It should be understood that the various forms of flow shown above can be re-ordered, added to, or have steps deleted, using the flow. For example, the steps described in the present disclosure can be performed in parallel, in series, or in a different order, as long as the desired results of the technical solutions of the present disclosure can be achieved, which are not limited herein.

[0147] The specific implementation described above does not constitute a limitation on the protection scope of the present disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A training method for a prediction model used to estimate article distribution volume, comprising: Obtain sample articles and their sample distribution volume; Based on the comparison between the sample distribution volume and the preset distribution volume threshold, the sample articles are classified to determine the target distribution volume range to which the sample articles belong, wherein the preset distribution volume threshold is determined based on the sample distribution volume distribution of the sample articles; Extract sample features from the sample data. The sample features include sample statistical features and sample attribute features. The sample statistical features include the historical posting data of the sample article authors, and the sample attribute features include the attribute information of the sample article authors. Based on the statistical features of the samples, the feature vector and trajectory vector of the sample articles are determined, wherein the trajectory vector is obtained by embedding the statistical features of the samples using a deep walk algorithm; Based on the feature vector, the trajectory vector, the sample attribute features, and the sample distribution amount, the initial model corresponding to the target distribution amount interval is trained to obtain the prediction model corresponding to the target distribution amount interval.

2. The method according to claim 1, wherein, The step of classifying the sample articles based on the comparison result between the sample distribution volume and the preset distribution volume threshold, and determining the target distribution volume range to which the sample articles belong, includes: Articles with zero sample distribution are classified into the first interval; Articles with a sample distribution size greater than zero and less than or equal to the first threshold are classified into the second interval; Articles with a sample distribution volume greater than the first threshold are classified into the third interval.

3. The method according to claim 1, wherein, The step of determining the feature vector and trajectory vector of the sample article based on the sample statistical characteristics includes: Based on the statistical features of the sample, the first preset model is input for embedding processing to obtain the feature vector of the sample article; The sample statistical features are input into a second preset model for embedding processing to obtain the trajectory vector of the sample article; The first preset model is a word vector model, and the second preset model is a model using a depth-walking algorithm.

4. The method according to claim 1, wherein, The statistical features of the sample include at least one of the following: the user exposure of the sample article author, the historical number of posts, and the median number of posts in the window; the attribute features of the sample include at least one of the following: the vertical category of the sample article author and the number of followers.

5. A method for predicting article distribution volume, comprising: Retrieve the target article; Extract the data features of the target article, including the author statistical features and author attribute features of the target article; The data features are input into the trained classification model to obtain a preliminary prediction result of the target article. The preliminary prediction result is used to indicate the target distribution volume interval category to which the target article belongs. Based on the target distribution range corresponding to the preliminary prediction results, a target prediction model is determined from multiple prediction models, wherein the multiple prediction models are trained separately for different target distribution ranges using the method described in any one of claims 1-4. The data features of the target article are input into the target prediction model to obtain the estimated distribution volume of the target article.

6. A training device for a prediction model used to predict article distribution volume, comprising: The first acquisition module is configured to acquire sample articles and their sample distribution volume; The first determining module is configured to classify the sample articles based on the comparison result between the sample distribution volume and the preset distribution volume threshold, and determine the target distribution volume interval to which the sample articles belong, wherein the preset distribution volume threshold is determined based on the sample distribution volume distribution of the sample articles; The extraction module is configured to extract sample features from the sample data. The sample features include sample statistical features and sample attribute features. The sample statistical features include the historical posting data of the sample article authors, and the sample attribute features include the attribute information of the sample article authors. The training module is configured to determine the feature vector and trajectory vector of the sample article based on the sample statistical features, wherein the trajectory vector is obtained by embedding the sample statistical features through a deep walk algorithm; and to train an initial model corresponding to the target distribution range based on the feature vector, the trajectory vector, the sample attribute features and the sample distribution range to obtain a prediction model corresponding to the target distribution range.

7. An article distribution volume prediction device, comprising: The second acquisition module is configured to acquire the target article; The second determining module is configured to extract data features of the target article, including author statistical features and author attribute features of the target article; input the data features into a trained classification model to obtain a preliminary prediction result of the target article, the preliminary prediction result being used to indicate the target distribution volume interval category to which the target article belongs; The third determining module is configured to determine a target prediction model from multiple prediction models based on the target distribution range corresponding to the preliminary prediction result, wherein the multiple prediction models are trained separately for different target distribution ranges by the method of any one of claims 1-4. The prediction module is configured to input the data features of the target article into the target prediction model to obtain the predicted distribution volume of the target article.

8. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-5.

9. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-5.

10. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-5.

Images