A method and device for determining click rate of live broadcast

Abstract

The present invention discloses a method and device for determining the click-through rate of a live broadcast, and relates to the field of computer technology. The specific implementation of the method includes: obtaining multiple historical user data and multiple historical live broadcast data; determining the user behavior sequence corresponding to the multiple historical user data according to the sequence generation model, the multiple historical user data, and the generation time of the multiple historical user data; determining the user attribute characteristics according to the multiple historical user data, and determining the live broadcast characteristics according to the multiple historical live broadcast data; training the click-through rate prediction model according to the user behavior sequence, the user attribute characteristics, and the live broadcast characteristics; determining the click-through rate of the target user on the target live broadcast data according to the trained click-through rate prediction model. This implementation can improve the prediction accuracy of the live broadcast click-through rate, thereby improving the push accuracy, and reducing the possibility of insufficient inventory or slow sales.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and device for determining a live broadcast click rate. Background Art

[0002] As an emerging industry, live streaming has gradually become a major marketing method in the e-commerce industry through its guided consumption.

[0003] In the prior art, operators mainly predict the click rate of users on the live broadcast based on live broadcast data (such as the time period of the live broadcast, the number of fans of the anchor, whether there are celebrities, whether there is a lottery and the number of prizes, etc.) according to their own experience or market experience.

[0004] Due to the limited experience of operators and the ever-changing data corresponding to different live broadcasts, the accuracy of the live broadcast click-through rate subjectively predicted based on the individual experience of operators is low, which may lead to inaccurate push, insufficient inventory or unsalable products when pushing or preparing goods based on live broadcast data. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a method and device for determining a live broadcast click-through rate, which can improve the prediction accuracy of the live broadcast click-through rate, thereby improving the push accuracy and reducing the possibility of insufficient inventory or slow sales.

[0006] To achieve the above object, according to one aspect of an embodiment of the present invention, a method for determining a live broadcast click rate is provided, comprising:

[0007] Obtain multiple historical user data and multiple historical live broadcast data;

[0008] Determine a user behavior sequence corresponding to the plurality of historical user data according to a sequence generation model, the plurality of historical user data, and a generation time of the plurality of historical user data;

[0009] Determine user attribute characteristics according to the plurality of historical user data, and determine live broadcast characteristics according to the plurality of historical live broadcast data;

[0010] Training a click rate prediction model according to the user behavior sequence, the user attribute characteristics and the live broadcast characteristics;

[0011] According to the trained click rate prediction model, the click rate of the target user on the target live broadcast data is determined.

[0012] Optionally, determining the user behavior sequence corresponding to the plurality of historical user data according to the sequence generation model, the plurality of historical user data, and the generation time of the plurality of historical user data includes:

[0013] determining user behavior characteristics according to the plurality of historical user data;

[0014] Taking the user behavior feature and the generation time corresponding to the user behavior feature as input of the sequence generation model, and determining the weight value corresponding to each of the user behavior features according to the output of the sequence generation model;

[0015] The user behavior sequence is generated according to the user behavior characteristics and the weight value.

[0016] Optionally, determining a weight corresponding to each of the user behavior features according to the output of the sequence generation model includes:

[0017] The output of the sequence generation model is normalized to obtain a weight value corresponding to each of the user behavior features.

[0018] Optionally, the training of the click rate prediction model according to the user behavior sequence, the user attribute characteristics and the live broadcast characteristics includes:

[0019] Inputting the user behavior sequence into an ARMA model, and determining user dynamic features according to the output of the ARMA model;

[0020] The user dynamic features, the user attribute features and the live broadcast features are used as inputs of the click-through rate prediction model to train the click-through rate prediction model.

[0021] Optionally, after determining the click rate of the target user on the target live broadcast data, the method further includes:

[0022] According to the click rate, live broadcast data is pushed to the target user.

[0023] Optionally, after determining the click rate of the target user on the target live broadcast data, the method further includes:

[0024] According to the click rate, the inventory quantity corresponding to the target live broadcast data is determined, and inventory management is performed according to the inventory quantity.

[0025] Optionally, the sequence generation model is a random Mori model;

[0026] and / or,

[0027] The click rate prediction model is the XGBOOST model.

[0028] According to another aspect of an embodiment of the present invention, a device for determining a live broadcast click rate is provided, comprising:

[0029] An acquisition module, used to acquire multiple historical user data and multiple historical live broadcast data;

[0030] A sequence generation module, configured to determine a user behavior sequence corresponding to the plurality of historical user data according to a sequence generation model, the plurality of historical user data, and a generation time of the plurality of historical user data;

[0031] A feature generation module, configured to determine user attribute features according to the plurality of historical user data, and determine live broadcast features according to the plurality of historical live broadcast data;

[0032] A model training module, used for training a click rate prediction model according to a user behavior sequence, the user attribute characteristics and the live broadcast characteristics;

[0033] The data processing module is used to determine the click rate of the target user on the target live broadcast data based on the trained click rate prediction model.

[0034] According to another aspect of an embodiment of the present invention, there is provided an electronic device for determining a live broadcast click rate, comprising:

[0035] one or more processors;

[0036] a storage device for storing one or more programs,

[0037] When the one or more programs are executed by the one or more processors, the one or more processors implement the method for determining the live broadcast click rate provided by the present invention.

[0038] According to another aspect of an embodiment of the present invention, a computer-readable medium is provided, on which a computer program is stored, and when the program is executed by a processor, the method for determining the live broadcast click rate provided by the present invention is implemented.

[0039] One embodiment of the above invention has the following advantages or beneficial effects: because a time-based user data training model is used to determine the live broadcast click-through rate, the technical problems of inaccurate subjective prediction push, insufficient inventory or slow sales are overcome, thereby achieving the technical effect of improving the prediction accuracy of the live broadcast click-through rate, thereby improving the push accuracy, and reducing the possibility of insufficient inventory or slow sales.

[0040] The further effects of the above-mentioned non-conventional optional manner will be described below in conjunction with the specific implementation manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings are used to better understand the present invention and do not constitute an improper limitation of the present invention.

[0042] Figure 1It is an exemplary system architecture diagram suitable for applying to a method for determining a live broadcast click rate or an apparatus for determining a live broadcast click rate according to an embodiment of the present invention;

[0043] Figure 2 is a schematic diagram of the main process of a method for determining a live broadcast click rate according to an embodiment of the present invention;

[0044] Figure 3 is a schematic diagram of a detailed process of a method for determining a live broadcast click rate according to an embodiment of the present invention;

[0045] Figure 4 is a schematic diagram of main modules of a device for determining a live broadcast click rate according to an embodiment of the present invention;

[0046] Figure 5 It is a schematic diagram of the structure of a computer system of a terminal device or a server suitable for implementing an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0048] ARMA model: Autoregressive moving average model (also known as Autoregressive moving average model) is an important method for studying time series, including: Autoregressive model (Autoregressive model, referred to as AR model), Moving average model (Moving average model, referred to as MA model) and Autoregressive moving average model (ARMA).

[0049] Truncation: refers to the property that the autocorrelation function (ACF) or partial autocorrelation function (PACF) of a time series is 0 after a certain order.

[0050] Tailing: refers to the property that the autocorrelation function (ACF) or partial autocorrelation function (PACF) of a time series is not always zero after a certain order.

[0051] AIC, Akaike information criterion, is a criterion for measuring the goodness of fit of a statistical model. Usually, the smaller the AIC value, the better the model.

[0052] BIC, the Bayesian information criterion, is a criterion for measuring the goodness of fit of a statistical model. Generally, the smaller the BIC value, the better the model. AIC selects a good model for prediction from a prediction perspective; while BIC selects the model that best fits the data from a fitting perspective.

[0053] Figure 1 An exemplary system architecture diagram of a method for determining a live broadcast click rate or an apparatus for determining a live broadcast click rate suitable for application in an embodiment of the present invention is shown. Figure 1 As shown, an exemplary system architecture of a method for determining a live broadcast click-through rate or an apparatus for determining a live broadcast click-through rate according to an embodiment of the present invention includes:

[0054] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0055] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0056] The terminal devices 101 , 102 , and 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers.

[0057] The server 105 may be a server that provides various services, such as a background management server that supports shopping websites browsed by users using the terminal devices 101, 102, and 103. The background management server may analyze and process the received data such as user feature query requests, and feed back the processing results (such as user features) to the terminal devices 101, 102, and 103.

[0058] It should be noted that the method for determining the live broadcast click-through rate provided in the embodiment of the present invention is generally executed by the server 105 , and accordingly, the device for determining the live broadcast click-through rate is generally arranged in the server 105 .

[0059] It should be understood that Figure 1The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to the implementation requirements.

[0060] Figure 2 FIG. 1 is a schematic diagram of the main process of the method for determining the live broadcast click rate according to an embodiment of the present invention. Figure 2 As shown, the method for determining the live broadcast click rate of the present invention includes:

[0061] Step S201, obtaining multiple historical user data and multiple historical live broadcast data.

[0062] Exemplarily, based on the platform's historical data, multiple historical user data and multiple historical live broadcast data are obtained. The historical user data may include data on information such as the user's age, gender, purchasing power, occupation, and preferences, and may also include data on the user's browsing, commenting, collecting, adding to cart, placing orders, sharing, and other operations, and may also include the time when the user operation data is generated; the historical live broadcast data may include data on information such as the live broadcast brand, lottery, time, interaction, anchor, and product.

[0063] Step S202 : determining a user behavior sequence corresponding to the plurality of historical user data according to a sequence generation model, the plurality of historical user data, and the generation time of the plurality of historical user data.

[0064] Exemplarily, according to the multiple historical user data obtained in step S201, the user behavior characteristics are determined based on the user operation data therein, and the generation time corresponding to the user behavior characteristics is determined based on the time when the user operation data is generated, and the user behavior characteristics and the generation time corresponding to the user behavior characteristics are input into the sequence generation model, and the characteristic score of each user behavior characteristic is output. The characteristic scores of the user behavior characteristics are normalized to obtain the weight value corresponding to each user behavior characteristic; based on the data of the user behavior characteristics and the weight value corresponding to the user behavior characteristics, the user behavior sequence is generated after weighted summation. The user behavior sequence contains 12 elements, namely the user behavior scores of 12 months, which represent the behavior of the user in the past year.

[0065] Step S203, determining user attribute characteristics based on the multiple historical user data, and determining live broadcast characteristics based on the multiple historical live broadcast data.

[0066] Exemplarily, according to the multiple historical user data acquired in step S201, user attribute characteristics are determined based on the user information data therein, and live broadcast characteristics are determined based on the multiple historical live broadcast data therein.

[0067] Step S204, training a click rate prediction model according to the user behavior sequence, the user attribute characteristics and the live broadcast characteristics.

[0068] Exemplarily, according to the user behavior sequence obtained in step S202, the user behavior sequence is input into the ARMA model, the ARMA model is trained, and the ARMA model parameters are output as user dynamic features. The user dynamic features, the user attribute features obtained in step S203, and the live broadcast features are input into the click-through rate prediction model, the click-through rate prediction model is trained, and the trained click-through rate prediction model is output. Among them, the click-through rate prediction model is the XGBOOST model.

[0069] Step S205, determining the click rate of the target user on the target live broadcast data according to the trained click rate prediction model.

[0070] Exemplarily, the target user data and multiple live broadcast data to be pushed are obtained, and the click rate of the target user on the target live broadcast data is determined according to the target user data and the trained click rate prediction model. The click rate may include regular clicks, browsing, favorites, add-to-cart and order data. According to the click rate, the live broadcast data is pushed to the target user; or, according to the click rate, the inventory corresponding to the target live broadcast data is determined, and the inventory management is performed according to the inventory, and the supply warehouse is allocated or the demand warehouse is supported in a timely manner.

[0071] In an embodiment of the present invention, by acquiring multiple historical user data and multiple historical live broadcast data; determining a user behavior sequence corresponding to the multiple historical user data according to a sequence generation model, the multiple historical user data and the generation time of the multiple historical user data; determining user attribute characteristics according to the multiple historical user data, and determining live broadcast characteristics according to the multiple historical live broadcast data; training a click-through rate prediction model according to the user behavior sequence, the user attribute characteristics and the live broadcast characteristics; determining the click-through rate of a target user on target live broadcast data according to the trained click-through rate prediction model, etc., it is possible to adapt to periodic changes in user behavior, optimize the performance of a live broadcast data prediction model, make full use of live broadcast resources, accurately predict the live broadcast click-through rate, accurately push live broadcasts to users and reasonably manage inventory.

[0072] Figure 3 is a schematic diagram of a detailed process of a method for determining a live broadcast click rate according to an embodiment of the present invention. Figure 3 As shown, the method for determining the live broadcast click rate of the present invention includes:

[0073] Step S301, constructing a live broadcast database.

[0074] Exemplarily, a live broadcast database is constructed based on the existing historical data of the platform, and historical user data and historical live broadcast data are obtained from the historical data of the platform. The historical user data may include multiple items. Taking the historical user data of a certain e-commerce platform as an example, for example, the historical user data may include data on the user's age, gender, purchasing power, occupation, and preferences, or the historical user data may include data on the user's browsing live broadcasts, browsing products, adding to cart, placing orders, sharing, and commenting, or the historical user data may include the user's operation time data corresponding to the operation data. The historical live broadcast data may include multiple items. Taking the historical user data of a certain e-commerce platform as an example, for example, the historical live broadcast data may include data on information such as the brand, lottery, time, interaction, anchor, and product of the live broadcast.

[0075] Furthermore, the platform can periodically obtain relevant data, and update and save the obtained relevant data to the database.

[0076] Step S302, constructing user attribute features of the live broadcast.

[0077] Exemplarily, according to the live broadcast database constructed in step S301, information data of historical user data is extracted therefrom, and user attribute characteristics can be determined based on the information data of historical user data. Taking the historical user data of a certain e-commerce platform as an example, for example, the historical user data may include data of information such as the user's age, gender, occupation, preference category, purchasing power, geographic location, and consumption time, and the user attribute characteristics determined based on the historical user data may include user age characteristics, user gender characteristics, user occupation characteristics, user preference category characteristics, user purchasing power characteristics, user geographic location characteristics, user consumption time characteristics, etc.

[0078] Step S303, constructing the live broadcast feature of the live broadcast.

[0079] Exemplarily, according to the live broadcast database constructed in step S301, information data of historical live broadcast data is extracted therefrom, and live broadcast features can be determined based on the information data of historical live broadcast data. Taking the historical user data of a certain e-commerce platform as an example, for example, the historical live broadcast data may include information data such as live broadcast brand, lottery, time, interaction, anchor and product, and the live broadcast features determined based on the historical live broadcast data may include live broadcast brand features, live broadcast lottery features, live broadcast time features, live broadcast anchor features, live broadcast product features and live broadcast additional features, etc.

[0080] Furthermore, live broadcast brand characteristics include characteristics of the number of live broadcast brands, characteristics of the number of live broadcast brand fans, etc.; live broadcast lottery characteristics include characteristics of whether there is a lottery in the live broadcast, characteristics of the number of live broadcast lottery draws, etc.; live broadcast time characteristics include characteristics of whether the live broadcast time is a weekend, characteristics of the live broadcast time period, etc.; live broadcast anchor characteristics include characteristics of the number of live broadcast anchors, characteristics of the number of live broadcast anchor fans, characteristics of live broadcast anchor types, characteristics of the types of live broadcast anchors with goods, etc.; live broadcast product characteristics include characteristics of the number of live broadcast products, characteristics of the average price of live broadcast products, characteristics of the types of live broadcast products, etc.; live broadcast additional characteristics include characteristics of whether there are celebrities in the live broadcast room, characteristics of whether the live broadcast is connected to the microphone, etc.

[0081] Furthermore, the user attribute features of the live broadcast obtained based on step S302 and the live broadcast features of the live broadcast obtained based on step S302 include discrete features and continuous features. Discrete features (for example, user age features, gender features, geographic location features, etc.) have a certain sparsity, which will cause the model performance to drop sharply. Therefore, it is necessary to perform Embedding processing on the discrete features and output continuous vectors so that the features have better expression capabilities in the feature space. Among them, the Embedding processing can extract features from the original data and perform dimensionality reduction processing through the principle of matrix multiplication. Continuous features themselves are continuous, so no processing is required.

[0082] Furthermore, the processed user attribute features and live broadcast features of the continuous live broadcast are input into the XGBOOST model for training, and the feature score of each feature is output. The higher the score, the higher the importance; the features with feature scores above the predetermined value are selected as the final live broadcast user attribute features and live broadcast features. As an integrated learning model, XGBOOST can alleviate the errors caused by feature sparsity and correlation, effectively remove redundant features, and improve feature quality. The feature cross-functionality of the XGBOOST model itself can be used in python to perform feature screening using the model's Feature importance function. Since the two indicators of the live broadcast room product browsing rate and the live broadcast room product order rate have different focuses, the output feature importance is also different for the model training of the two indicators.

[0083] Step S304, constructing user behavior characteristics of the live broadcast.

[0084] Exemplarily, according to the database constructed in step S301, the operation data of the historical user data is extracted therefrom, and the user behavior characteristics can be determined based on the operation data of the historical user data. Taking the historical user data of a certain e-commerce platform as an example, for example, the historical user data may include data of the user's browsing live broadcasts, browsing products, adding to carts, placing orders, sharing, and other operations, and the user behavior characteristics determined based on the historical user data may include user browsing live broadcast characteristics, user browsing product characteristics, user adding to cart characteristics, user placing orders, user sharing characteristics, etc.

[0085] Step S305, constructing a live user behavior sequence.

[0086] Exemplarily, according to the live broadcast database constructed in step S301, operation time data corresponding to the operation data of historical user data is extracted therefrom, and based on the operation time data corresponding to the operation data of historical user data, the generation time of each user behavior feature can be determined.

[0087] Exemplarily, the live broadcast user behavior features constructed according to step S304 and the corresponding generation time of each user behavior feature are input into the sequence generation model for training, and the feature score of each user behavior feature is output after calculating the information gain. The feature score is normalized to obtain the weight of each user behavior feature, and the user behavior feature data and the weight of each user behavior feature are weighted and summed to obtain the user's behavior sequence. Among them, the weight can represent the importance / importance of different features, and the sequence refers to a vector composed of the values ​​of multiple features within a predetermined time period. Among them, the sequence generation model can be a random forest model.

[0088] Exemplarily, the live broadcast user behavior features constructed according to step S304 include user browsing live broadcast features, user browsing product features, user adding to purchase features, user ordering features, and user sharing features; extract the operation time data corresponding to the operation data of the historical user data from the live broadcast database constructed in step S301, and determine the generation time of each user behavior feature, that is, the generation time of the user browsing live broadcast features, user browsing product features, user adding to purchase features, user ordering features, and user sharing features. Among them, the user browsing live broadcast features are represented by A, the user browsing product features are represented by B, the user adding to purchase features are represented by C, the user ordering features are represented by D, and the user sharing features are represented by E.

[0089] For example, a live broadcast database is constructed based on the historical data of the platform in the past year, and historical user data and historical live broadcast data are obtained from the live broadcast database. Based on the obtained historical user data and historical live broadcast data, the user behavior features A, B, C, D, E in the past year are constructed, and the user behavior features A, B, C, D, E and the generation time of the corresponding user behavior features are input into the random forest model for training. After calculating the information gain, the feature scores of A, B, C, D, E are output, and the feature scores of A, B, C, D, E are normalized to obtain the weights W of A, B, C, D, E. A , W B , W C , W D , W E , the data of A, B, C, D, E and the weights W of A, B, C, D, E A , W B , W C , W D , W E Perform weighted summation to obtain the user behavior sequence.

[0090] For example, the data of A, B, C, D, and E can be the number of times a user browses a live broadcast, the number of times a user browses a product, the number of times a user adds a product to a cart, the number of times a user places an order, and the number of times a user shares a product. A , W B , W C , W D , W E Perform weighted summation to obtain the user behavior sequence, including: summing the data of A, B, C, D, E in January with the weights W of A, B, C, D, E A , W B , W C , W D , W E Perform weighted summation to get the user's behavior score for January; sum the data of A, B, C, D, and E from February to December with the weights W of A, B, C, D, and E respectively. A , W B , W C , W D , W E Perform weighted summation to get the user's behavior scores from February to December; combine the user behavior scores of the 12 months to get the user behavior sequence. The user behavior sequence contains 12 elements, namely the user behavior scores of the 12 months, which represent the user's behavior in the past year.

[0091] Furthermore, for example, the data of A, B, C, D, and E are input into the random forest model, and the output weight W A , WB , W C , W D , W E is [0.3, 0.2, 0.2, 0.1, 0.2]; the data of A, B, C, D, E in January (users browsed live broadcasts 5 times, browsed products 10 times, added to cart 3 times, placed orders 2 times, and shared 1 time) and the weights W of A, B, C, D, E A , W B , W C , W D , W E Perform weighted summation to obtain the user's behavior score in January s1 = 0.3*5 + 0.2*10 + 0.2*3 + 0.1*2 + 0.2*1 = 4.5; sum the data of A, B, C, D, E from February to December with the weights W of A, B, C, D, E respectively. A , W B , W C , W D , W E Perform weighted summation to obtain the user behavior scores s2, s3, s4, s5, s6, s7, s8, s9, s10, s11, s12 from February to December; combine the user behavior scores of 12 months to obtain the user behavior sequence [s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11, s12].

[0092] Furthermore, since the two indicators of live broadcast room product browsing rate and live broadcast room product order rate have different focuses, the live broadcast room product browsing rate focuses on the number of views, while the live broadcast room product order rate focuses more on the number of orders. Therefore, the importance of the output features of the model training for the two indicators is also different. For example, for the live broadcast room product browsing rate indicator, the data of A, B, C, D, and E are input into the random forest model, and the output weight W A , W B , W C , W D , W E is [0.3, 0.2, 0.2, 0.1, 0.2]; and for the live broadcast room product order rate indicator, the data of A, B, C, D, and E are input into the random forest model, and the output weight W A , W B , W C , W D , W E That is [0.1, 0.2, 0.2, 0.3, 0.2]. It can be seen that for the product browsing rate indicator in the live broadcast room, the weights of A and B are W. A , W B The weight W is greater than D D, indicating that the importance of the user browsing live broadcast feature and the user browsing product feature in the live broadcast room product browsing rate indicator is higher than the user order feature; for the live broadcast room product browsing rate and live broadcast room product order rate indicators, the weight of D is W D It accounts for a higher proportion in the order rate of goods in the live broadcast room, indicating that user order characteristics are more important in the order rate indicator of goods in the live broadcast room.

[0093] Step S306, constructing user dynamic features of the live broadcast.

[0094] Exemplarily, according to the live user behavior sequence constructed in step S305, the user behavior sequence is input into the ARMA model, and the stability of the user behavior sequence is detected by the ADF test. Based on the result of the ADF test, it is determined whether the user behavior sequence is stable. If it is not stable, it is subjected to differential processing (differential operation) until the user behavior sequence is stable. After confirming that the user behavior sequence is stable, the autocorrelation coefficient a and the partial autocorrelation coefficient b of the user behavior sequence are calculated, and the ARMA model is identified according to the autocorrelation coefficient a (ACF) and the partial autocorrelation coefficient b (PACF). Among them, identifying the ARMA model includes: if the autocorrelation coefficient a is tailing and the partial autocorrelation coefficient b is p-order truncated, the ARMA model is an ARp model; if the autocorrelation coefficient a is q-order truncated and the partial autocorrelation coefficient b is tailing, the ARMA model is a MAq model; if the autocorrelation coefficient a is tailing and the partial autocorrelation coefficient b is also tailing, the ARMA model is an ARMAp,q model. Based on the determined ARMA model, the orders p and q are determined in combination with the AIC and BIC criteria. The orders p and q represent the autocorrelation characteristics of the sequence itself, especially the periodic behavior. After the orders p and q are determined, the model parameters of the ARMA model can be obtained, and the model parameters of the ARMA model are combined into a feature vector, which is the user dynamic feature.

[0095] Furthermore, the periodic behavior of users, such as weekly or monthly inertial behavior, makes the user behavior have a periodic pattern. The user dynamic feature is a sequence feature, which has continuity itself.

[0096] Furthermore, deep learning models such as RNN, LSTM, and temporal convolutional networks can be used to construct user dynamic features.

[0097] For example, when using user behavior sequences as the input of the click rate prediction model for live broadcast, there are problems such as too large dimensions and possible waste of resources, and the dimensions and sequence lengths are not uniform. Therefore, the ARMA model is used to abstract the user behavior sequence, and its model parameters are used to construct the user dynamic features of live broadcast.

[0098] Step S307: click rate prediction model training.

[0099] Exemplarily, based on the user dynamic features of the live broadcast constructed in step S306, the user attribute features of the live broadcast constructed in step S302, and the live broadcast features of the live broadcast constructed in step S303, a feature sample of the live broadcast click-through rate is constructed. The feature samples of the live broadcast click-through rate are divided into a training set and a test set, and 80% of the feature samples are selected as the training set to train the model and obtain the trained click-through rate prediction model; the remaining 20% ​​of the feature samples are used as the test set to test the live broadcast click-through rate prediction model after training.

[0100] Exemplarily, the training set feature samples of the live broadcast click-through rate are input into the XGBOOST model, the weak classifiers are trained in an iterative cycle, and multiple weak classifiers are iteratively integrated into a combined classifier to obtain a trained click-through rate prediction model.

[0101] Furthermore, the click-through rate prediction model is used to predict the click-through rate of the live broadcast, where the click-through rate may include data such as regular clicks, browsing, favorites, add-to-cart, ordering and sharing.

[0102] Furthermore, the XGBOOST model (eXtreme Gradient Boosting) is a boosting tree model that performs iterative training based on the input training set feature samples, learns the weak classifiers of each iteration step by step, and updates the weights of the samples in the training set based on the coefficients of the weak classifiers; fits the results of the previous weak classifiers with the residuals of the training set samples, and iteratively integrates multiple weak classifiers into a strong classifier to obtain a prediction model. As an integrated learning model, XGBOOST has good learning performance. At the same time, as an optimized distributed gradient boosting model, the XGBOOST model can consume less computing resources in a short time and obtain a prediction model with excellent performance, integrating the advantages of regularization, sparsity perception, and cross-validation.

[0103] Furthermore, algorithms such as LR, random forest, GBDT, and BP neural network can be used for model training.

[0104] Step S308, verifying the click rate prediction model.

[0105] Exemplarily, the test set feature samples of the live broadcast click-through rate are input into the combined classifier obtained by training in step S307, that is, the trained click-through rate prediction model, and the user's live broadcast click-through rate is output. The error of the click-through rate prediction model is calculated according to the test results, and it is determined whether the model error is higher than the error standard value. The click-through rate prediction model is corrected according to the model error so that the click-through rate prediction model meets the requirements.

[0106] Furthermore, through the association relationship based on sequence features, the prediction accuracy of the prediction model is greatly improved, and the live broadcast click rate can be accurately predicted.

[0107] Step S309: model use.

[0108] Exemplarily, based on the click rate prediction model obtained in step S308, according to the obtained target user data and the multiple live broadcast data to be pushed, the obtained target user data and the multiple live broadcast data to be pushed are input into the click rate prediction model, and the click rate of the target live broadcast data is output. According to the click rate, the live broadcast data can be pushed to the target user; or, according to the click rate, the inventory corresponding to the target live broadcast data can be determined, and inventory management can be performed according to the inventory, for example, directly allocating inventory from the production place for stocking, allocating inventory from the supply warehouse to the demand warehouse, etc.

[0109] Furthermore, the click rate may include regular clicks, browsing, favorites, purchases and order data, etc. If the click rate of a certain product is high, the warehouse will increase the inventory of the product; if a user has a high click rate for browsing, favorites or purchases of a certain product, the live broadcast of the product or similar products will be recommended to the user.

[0110] In an embodiment of the present invention, by building a database for live broadcast; building user attribute features for live broadcast; building live broadcast features for live broadcast; building user behavior features for live broadcast; building user behavior sequences for live broadcast; building user dynamic features for live broadcast; training a click-through rate prediction model; verifying the click-through rate prediction model; using the model and the like steps, it is possible to adapt to the periodic changes in user behavior, optimize the performance of the live broadcast data prediction model, make full use of live broadcast resources, accurately predict the click-through rate of live broadcast, accurately push live broadcasts to users and reasonably manage inventory.

[0111] Figure 4 is a schematic diagram of main modules of a device for determining a live broadcast click rate according to an embodiment of the present invention. Figure 4 As shown, the apparatus 400 for determining the live broadcast click rate of the present invention includes:

[0112] The acquisition module 401 is used to acquire multiple historical user data and multiple historical live broadcast data.

[0113] Exemplarily, the acquisition module 401 acquires multiple historical user data and multiple historical live broadcast data based on the historical data of the platform. The historical user data may include data on information such as the user's age, gender, purchasing power, occupation and preferences, and may also include data on the user's browsing, commenting, collecting, adding to cart, placing orders, sharing and other operations, and may also include the time when the user operation data is generated; the historical live broadcast data may include data on information such as the live broadcast brand, lottery, time, interaction, anchor and product.

[0114] The sequence generation module 402 is used to determine the user behavior sequence corresponding to the multiple historical user data according to the sequence generation model, the multiple historical user data and the generation time of the multiple historical user data.

[0115] Exemplarily, according to the multiple historical user data acquired by the acquisition module 401, the user behavior characteristics are determined based on the user operation data therein, and the generation time corresponding to the user behavior characteristics is determined based on the time when the user operation data is generated. The sequence generation module 402 inputs the user behavior characteristics and the generation time corresponding to the user behavior characteristics into the sequence generation model, and outputs the feature score of each user behavior characteristic. The sequence generation module 402 normalizes the feature scores of the user behavior characteristics to obtain the weight value corresponding to each user behavior characteristic; based on the data of the user behavior characteristics and the weight value corresponding to the user behavior characteristics, a user behavior sequence is generated after weighted summation. The user behavior sequence contains 12 elements, namely the user behavior scores of 12 months, which characterize the behavior of the user in the past year.

[0116] The feature generation module 403 is used to determine user attribute features based on the multiple historical user data, and determine live broadcast features based on the multiple historical live broadcast data.

[0117] Exemplarily, according to the multiple historical user data acquired by the acquisition module 401, the feature generation module 403 determines the user attribute feature based on the user information data therein, and determines the live broadcast feature based on the multiple historical live broadcast data therein.

[0118] The model training module 404 is used to train the click rate prediction model according to the user behavior sequence, the user attribute characteristics and the live broadcast characteristics.

[0119] Exemplarily, according to the user behavior sequence obtained by the sequence generation module 402, the model training module 404 inputs the user behavior sequence into the ARMA model, trains the ARMA model, and outputs the ARMA model parameters as user dynamic features. The model training module 404 inputs the user dynamic features, the user attribute features obtained by the feature generation module 403, and the live broadcast features into the click-through rate prediction model, trains the click-through rate prediction model, and outputs the trained click-through rate prediction model. Among them, the click-through rate prediction model is the XGBOOST model.

[0120] The data processing module 405 is used to determine the click rate of the target user on the target live broadcast data according to the trained click rate prediction model.

[0121] Exemplarily, the target user data and a plurality of live broadcast data to be pushed are obtained, and the data processing module 405 determines the click rate of the target user on the target live broadcast data according to the target user data and the trained click rate prediction model. The click rate may include regular clicks, browsing, favorites, purchases and order data. According to the click rate, the live broadcast data is pushed to the target user; or, according to the click rate, the inventory corresponding to the target live broadcast data is determined, and the inventory management is performed according to the inventory, and the supply warehouse is allocated or the demand warehouse is supported in a timely manner.

[0122] In an embodiment of the present invention, through modules such as an acquisition module, a sequence generation module, a feature generation module, a model training module and a data processing module, it is possible to adapt to the periodic changes in user behavior, optimize the performance of the live broadcast data prediction model, make full use of live broadcast resources, accurately predict the live broadcast click-through rate, accurately push live broadcasts to users and reasonably manage inventory.

[0123] Figure 5 is a schematic diagram of the structure of a computer system of a terminal device suitable for implementing an embodiment of the present invention, such as Figure 5 As shown, the computer system 500 of the terminal device in the embodiment of the present invention includes:

[0124] A central processing unit (CPU) 501 can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage section 508 into a random access memory (RAM) 503. Various programs and data required for the operation of the system 500 are also stored in the RAM 503. The CPU 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0125] The following components are connected to the I / O interface 505: an input section 506 including a keyboard, a mouse, etc.; an output section 507 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN card, a modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to the I / O interface 505 as needed. A removable medium 511, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 510 as needed, so that a computer program read therefrom is installed into the storage section 508 as needed.

[0126] In particular, according to the embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 509, and / or installed from the removable medium 511. When the computer program is executed by the central processing unit (CPU) 501, the above-mentioned functions defined in the system of the present invention are executed.

[0127] It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0128] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0129] The modules involved in the embodiments of the present invention may be implemented by software or hardware. The modules described may also be arranged in a processor. For example, they may be described as follows: a processor includes an acquisition module, a sequence generation module, a feature generation module, a model training module, and a data processing module. The names of these modules do not, in some cases, constitute limitations on the modules themselves. For example, the acquisition module may also be described as a "module for acquiring live broadcast data from a live broadcast platform."

[0130] As another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiment; or it may exist independently without being assembled into the device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by a device, the device includes: obtaining multiple historical user data and multiple historical live broadcast data; determining the user behavior sequence corresponding to the multiple historical user data according to the sequence generation model, the multiple historical user data and the generation time of the multiple historical user data; determining the user attribute characteristics according to the multiple historical user data, and determining the live broadcast characteristics according to the multiple historical live broadcast data; training the click-through rate prediction model according to the user behavior sequence, the user attribute characteristics and the live broadcast characteristics; determining the click-through rate of the target user on the target live broadcast data according to the trained click-through rate prediction model.

[0131] According to the technical solution of the embodiment of the present invention, the prediction accuracy of the live broadcast click-through rate can be improved, thereby improving the push accuracy and reducing the possibility of insufficient inventory or slow sales.

[0132] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for determining a live broadcast click rate, characterized in that: include: Obtain multiple historical user data and multiple historical live broadcast data; Determine a user behavior sequence corresponding to the plurality of historical user data according to a sequence generation model, the plurality of historical user data, and a generation time of the plurality of historical user data; Determine user attribute characteristics according to the plurality of historical user data, and determine live broadcast characteristics according to the plurality of historical live broadcast data; Training a click rate prediction model according to the user behavior sequence, the user attribute characteristics and the live broadcast characteristics; Determine the click rate of target users on target live broadcast data based on the trained click rate prediction model; The determining, according to the sequence generation model, the plurality of historical user data, and the generation time of the plurality of historical user data, the user behavior sequence corresponding to the plurality of historical user data includes: determining user behavior characteristics according to the plurality of historical user data; Taking the user behavior feature and the generation time corresponding to the user behavior feature as input of the sequence generation model, and determining the weight value corresponding to each of the user behavior features according to the output of the sequence generation model; The user behavior sequence is generated according to the user behavior characteristics and the weight value.

2. The method according to claim 1, characterized in that: Determining a weight corresponding to each of the user behavior features according to the output of the sequence generation model includes: The output of the sequence generation model is normalized to obtain a weight value corresponding to each of the user behavior features.

3. The method according to claim 1, characterized in that The training of the click rate prediction model according to the user behavior sequence, the user attribute characteristics and the live broadcast characteristics includes: Inputting the user behavior sequence into an ARMA model, and determining user dynamic features according to the output of the ARMA model; The user dynamic features, the user attribute features and the live broadcast features are used as inputs of the click-through rate prediction model to train the click-through rate prediction model.

4. The method according to claim 1, characterized in that After determining the target user's click rate on the target live broadcast data, it also includes: According to the click rate, live broadcast data is pushed to the target user.

5. The method according to claim 1, characterized in that After determining the target user's click rate on the target live broadcast data, it also includes: According to the click rate, the inventory quantity corresponding to the target live broadcast data is determined, and inventory management is performed according to the inventory quantity.

6. The method according to claim 1, characterized in that The sequence generation model is a random Senli model; and / or, The click rate prediction model is the XGBOOST model.

7. A device for determining a live broadcast click rate, characterized in that: include: An acquisition module, used to acquire multiple historical user data and multiple historical live broadcast data; A sequence generation module, configured to determine a user behavior sequence corresponding to the plurality of historical user data according to a sequence generation model, the plurality of historical user data, and a generation time of the plurality of historical user data; A feature generation module, configured to determine user attribute features according to the plurality of historical user data, and determine live broadcast features according to the plurality of historical live broadcast data; A model training module, used for training a click rate prediction model according to a user behavior sequence, the user attribute characteristics and the live broadcast characteristics; A data processing module, used to determine the click rate of the target user on the target live broadcast data according to the trained click rate prediction model; The sequence generation module is also used for: determining user behavior characteristics according to the plurality of historical user data; Taking the user behavior feature and the generation time corresponding to the user behavior feature as input of the sequence generation model, and determining the weight value corresponding to each of the user behavior features according to the output of the sequence generation model; The user behavior sequence is generated according to the user behavior characteristics and the weight value.

8. An electronic device for determining a live broadcast click rate, characterized in that: include: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 6.

9. A computer readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images