WO2019165872A1 - 营销产品的推荐方法和装置 - Google Patents

营销产品的推荐方法和装置 Download PDF

Info

Publication number
WO2019165872A1
WO2019165872A1 PCT/CN2019/073610 CN2019073610W WO2019165872A1 WO 2019165872 A1 WO2019165872 A1 WO 2019165872A1 CN 2019073610 W CN2019073610 W CN 2019073610W WO 2019165872 A1 WO2019165872 A1 WO 2019165872A1
Authority
WO
WIPO (PCT)
Prior art keywords
user
product
marketing
feature
click
Prior art date
Application number
PCT/CN2019/073610
Other languages
English (en)
French (fr)
Inventor
王�义
Original Assignee
阿里巴巴集团控股有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 阿里巴巴集团控股有限公司 filed Critical 阿里巴巴集团控股有限公司
Publication of WO2019165872A1 publication Critical patent/WO2019165872A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0255—Targeted advertisements based on user history
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0269—Targeted advertisements based on user profile or attribute
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0254—Targeted advertisements based on statistics
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0269—Targeted advertisements based on user profile or attribute
    • G06Q30/0271—Personalized advertisement
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0241—Advertisements
    • G06Q30/0277—Online advertisement

Definitions

  • the present specification relates to the field of data processing technologies, and in particular, to a recommended method and apparatus for marketing products.
  • the present specification provides a recommended method for marketing products, including:
  • the click-through rate evaluation model is a machine learning model, using user characteristics with known click-through rates and Marketing product feature samples for training;
  • This specification also provides a recommended device for marketing products, including:
  • a feature obtaining unit configured to acquire a user feature of the user and a product feature of each marketing product after receiving a request of the user for marketing product data
  • a cross feature unit for generating a cross feature based on user features and product features
  • a click rate evaluation unit configured to input a user feature, a product feature, and a cross feature into a click rate evaluation model to obtain a click rate evaluation value of the user for each marketing product;
  • the click rate evaluation model is a machine learning model, and the Know the user characteristics of the click-through rate and the marketing product feature samples for training;
  • a marketing product unit configured to determine, according to the click rate evaluation value, data of the M marketing products returned by the M marketing products to the user; M is a natural number.
  • the computer device includes: a memory and a processor; the memory stores a computer program executable by the processor; and when the processor runs the computer program, the recommended method for executing the marketing product A step of.
  • the present specification provides a computer readable storage medium having stored thereon a computer program that, when executed by a processor, performs the steps described in the above-described recommended method of marketing products.
  • the user-characteristics, the marketing product features, and the cross-characteristics generated by the user features and the product features are used to construct a click-rate evaluation model, and the training-completed click-rate evaluation model is used to derive each marketing.
  • the product evaluates the click rate of a user and selects the marketing product recommended to the user based on the click rate evaluation value, so as to more accurately measure the matching degree between the user and the marketing product, so that the recommended marketing product is more consistent. Users' interests and needs reduce the interference caused by invalid marketing information to users.
  • FIG. 1 is a flow chart of a method for recommending a marketing product in an embodiment of the present specification
  • FIG. 2 is a schematic structural diagram of a click rate evaluation model in an application example of the present specification
  • FIG. 3 is a hardware structural diagram of a device running an embodiment of the present specification
  • FIG. 4 is a logical structural diagram of a recommendation device of a marketing product in the embodiment of the present specification.
  • the embodiment of the present specification proposes a new marketing product recommendation method, which uses user features and product features to describe users and marketing products respectively, with user features, product features, and cross-features generated by user features and product features.
  • a click-rate evaluation model to predict the matching degree of a marketing product and a user by using the click rate evaluation value output by the click-rate evaluation model, and to determine the marketing product recommended to the user according to the click-rate evaluation value, so that the marketing The product is more targeted than the user, which improves the efficiency and success rate of marketing, and greatly reduces the interruption of invalid marketing information to users.
  • the embodiments of the present specification can be run on any computing and storage device, such as a mobile phone, a tablet, a PC (Personal Computer), a notebook, a server, etc., or can be operated by two or more.
  • the logical nodes of the device implement the functions in the embodiments of the present specification.
  • the embodiment of the present specification runs on the server of the network service provider.
  • the server selects M (M is a natural number) among the plurality of marketing products and recommends to the user.
  • the marketing product may be related information of products, services, activities, etc. that any enterprise wishes to promote to users; the expression form of the marketing product is not limited, and may be text, pictures, animations, etc. with links; promotion forms of marketing products It is also not limited, and may be an advertisement slot on a page, a message push to a user terminal, a priority ranking in a search result, and the like.
  • a user learning model, a product feature, and a crossover feature are used to establish a machine learning model.
  • the click rate evaluation model referred to in this specification is used to evaluate the degree of matching between the marketing product and the user.
  • the user feature may be any information that can reflect the personalized characteristics of the user, and the user personalized information reflecting the interest and demand for the marketing product in a specific application scenario may be used as a user feature in the application scenario, and the implementation of the specification The example is not limited.
  • the user's background characteristics and/or user's behavioral characteristics may be used as user features.
  • the background feature of the user is generated according to the user's personal information, and is based on the inherent characteristics of the static data; the user's behavior characteristics are generated according to the user's historical behavior record, which is a dynamic feature constructed by the user's Internet behavior data.
  • the background feature of the user may be constructed based on the registration information of the user at the server of the network service provider and the static data retained by the user at other associated sites; the background characteristics may be: gender, age, years of education, highest education, family One or more of the address, graduate school, etc.
  • the user's historical behavior record may include the display, click browsing, purchase, etc.
  • Behaviors such as display, click-through, purchase, etc.; these dynamic historical behavior records can be used to construct behavioral characteristics related to marketing products, such as marketing products for financial consumption, behavioral characteristics can be: purchasing power, brand preference, risk preference, One or more of investment experience, preferred financial management, etc.
  • the product feature can be any information that reflects the characteristics of the marketing product.
  • the characteristics of the marketing product that may affect the user's interests and needs in a specific application scenario are used as product features in the application scenario, and are not limited.
  • the product attributes and/or marketing attributes of the marketing product may be used as product features, wherein the product attributes are used to describe the product characteristics of the marketing product itself, and may be one of product type, product acclaim, product sales, etc.
  • a number of products; product attributes used to describe the marketing characteristics of marketing products can be one or more of marketing novelty, discount level and so on.
  • the cross feature is to cross-combine user features and product features to form a combined class feature, each cross feature being composed of at least one user feature and at least one product feature, the value of the cross feature also being combined by each user feature of the cross feature And the value of the product features to determine.
  • the number of the cross-features, the user features and the product features of the cross-over features, and the cross-combination of the cross-features are determined according to the requirements of the actual application scenario, and the embodiments of the present specification are not limited. The following two implementations are taken as an example.
  • the cross-over feature is a combination of partial user features and partial product features.
  • the value calculation method of each cross feature used in the click rate evaluation model can be preset on the server side, namely: which cross-features are used in the click-rate evaluation model, and how to combine the user features of a cross-feature and The value of the product feature is used to derive the value of the cross feature.
  • the value of each cross feature can be calculated. For example, two cross-features are used in a certain click-rate evaluation model, and the cross-features preset on the server are calculated as shown in Table 1:
  • Cross feature 1 (user feature a + user feature b) / product feature c
  • Cross feature 2 User characteristics d ⁇ product feature e
  • the user feature and the product feature include a continuous feature (ie, the value of the user feature or the product feature is continuous) and a discrete feature (ie, the value of the user feature or the product feature is discrete)
  • the cross feature is composed of discrete user features and discrete product features, and the value of the cross feature is determined by performing predetermined logical operations on the user features and product features of the cross feature.
  • a user feature in an application scenario includes a discrete feature of S (S is a natural number) term, and a product feature includes a discrete feature of T (T is a natural number) term, and all discrete features have a value of 0 or 1;
  • S ⁇ T item cross-over feature is used in the scene's click-rate evaluation model.
  • Each discrete user feature and each discrete product feature are cross-combined.
  • the value of each cross-feature is combined with the cross-feature.
  • the logical AND operation result of the discrete user feature value and the discrete product feature value.
  • the click rate evaluation model may be a machine learning model using an arbitrary algorithm, and the algorithm used may be selected according to the characteristics of the actual application scenario, and is not limited.
  • it may be a machine learning model based on a support vector machine, such as SVC (Support Vector Machine), etc.; it may be a tree-based machine learning model, such as a GBD (Gradient Boosting Decision Tree).
  • It can be a linear model such as LR (Logistic Regression); it can also be a neural network model such as DNN (Deep Neural Networks), RNN (Recurrent Neural Networks), CNN (Convolutional) Neural Networks, Convolutional Neural Networks, etc.
  • the Wide and Deep model is used to build a click-through rate assessment model.
  • the Wide and Deep model includes a linear submodel and a deep neural network submodel, using a training model that combines a deep neural network submodel with a shallow linear submodel.
  • the training error of the overall model is simultaneously fed back to the linear submodel and the deep neural network.
  • the parameter update is performed in the sub-model, and the parameters of the two sub-models are optimized at the same time, so that the prediction ability of the overall Wide and Deep model is optimal.
  • the user feature, the product feature, and the crossover feature may be determined as the input of the linear sub-model and the deep neural network sub-model according to the needs of the actual application scenario.
  • the embodiments of the present specification are not limited. For example, discrete features in user features, product features, and cross-features can be used as inputs to a linear sub-model, and continuous features can be used as input to a deep neural network sub-model.
  • the input of the click rate evaluation model includes user characteristics, product features, and cross-features
  • the output is a click-rate evaluation value.
  • the click rate assessment model is trained using user characteristics of known hit rates and marketing product feature samples, ie, the known data in the training samples of the click rate assessment model includes user characteristics, product characteristics, and click rate as output for each sample.
  • the cross-over features as inputs can be automatically calculated from user features and product features.
  • the known click rate in a training sample can be determined based on the behavior of the user having the user characteristics in the sample for the marketing product having the product characteristics in the sample.
  • the known click rate can be determined based on the number of impressions and the click synthesis number, wherein the number of impressions is the number of times the product is displayed to the user within a predetermined time period; the click synthesis number is used to measure the user’s marketing
  • the comprehensive reaction of the product is determined according to the behavior of the user for the marketing product and the number of behaviors in the predetermined time period, and the behavior performed by the user for the marketing product may be active browsing, collecting, commenting, booking, purchasing, etc.
  • the predetermined weights can be set for the various behaviors performed by the user on the marketing product, and the weighted sum of the various behavior times is used as the user's click synthesis number.
  • the user clicks on the marketing product that is, after the user displays the marketing product to the user, the user clicks or otherwise actively obtains the marketing product information
  • the purchase behavior to calculate the click synthesis number
  • the known click rate can be determined according to the following method: the number of times the user purchases the marketing product within a predetermined time period is converted into a click amount according to a predetermined ratio, and the sum of the number of times the user clicks to browse the marketing product and the converted click number is used as the sum of the number of clicks after the predetermined time period Click on the composite number to compare the click-through ratio to the number of impressions as the known clickthrough rate.
  • the click rate evaluation value outputted by the model can be used to measure the degree of interest and demand of a certain user for a marketing product, or to predict the degree of matching of the user with the marketing product.
  • the crossover model as input in the click-through rate evaluation model can make the click-rate evaluation model get better generalization ability, so as to more accurately predict the matching degree between users and marketing products.
  • the flow of the recommendation method of the marketing product is as shown in FIG. 1.
  • Step 110 After receiving the user's request for the marketing product data, obtain the user characteristics of the user and the product features of each marketing product.
  • the server of the network service provider is accessed through the terminal to initiate a request to the server.
  • the request initiated by the user is a request for marketing product data; for example, the user's marketing product data.
  • the request can be a request for web page data to display a marketing product, or a data request for an ad slot used to display a marketing product.
  • the server can generate the user characteristics of the user who initiated the request after receiving the user's request for the marketing product data; generally, in order to speed up the response, the server pre-generates the user characteristics of each user and saves the request. Then you can query it.
  • the product features of the marketing product can be manually configured by the network administrator on the server side, or can be automatically generated by the server according to the relevant information of the marketing product, without limitation.
  • the marketing products may be pre-screened to obtain the product features of the selected marketing products.
  • the N (N is a natural number not less than M) marketing products are selected from all the marketing products according to the user's preference, and the user characteristics and screening of the user are obtained. Out of the product characteristics of N marketing products.
  • M is the number of marketing products ultimately recommended to the user, and N may be a preset value not less than M, or may be a variable value that satisfies the condition of not less than M.
  • the N marketing products that are pre-screened according to the user's preference will be used as the entire marketing product to which the embodiments of the present specification are applied, in which the M marketing products are finally determined to be recommended to the user.
  • the specific manner of screening the marketing product according to the user's preference may be implemented by referring to the prior art, and details are not described herein.
  • Step 120 generating a cross feature based on the user feature and the product feature.
  • the value of the user feature and the value of each product feature are used to generate a value of the cross feature of each marketing product input as the click rate evaluation model.
  • the values of the cross-features can be obtained according to the cross-over features used in constructing the click-rate evaluation model and the determination of the values of the cross-features in the specific application scenario.
  • step 130 the user feature, the product feature, and the cross feature are input into the click rate evaluation model to obtain the user's click rate evaluation value for each marketing product.
  • the user-characterized, the product feature of each marketing product, the user feature, and the cross-characteristic generated by the product feature of the marketing product are respectively input into the trained click-rate evaluation model, and the user's click-rate evaluation value of the marketing product is obtained. Predict how well the user matches the marketing product.
  • Step 140 Determine, according to the click rate evaluation value, the data of the M marketing products returned by the M marketing products to the user.
  • the server Based on the user's evaluation rate of the click rate of each marketing product, the server selects the M marketing products with the highest degree of matching with the user, and returns the data of the M marketing products to the user, and the terminal of the user displays the user to the user. M marketing products.
  • the user feature and the product feature are respectively used to describe the user and the marketing product
  • the click rate evaluation model is constructed by using the user feature, the marketing product feature, and the cross feature generated by the user feature and the product feature.
  • Using the trained click-rate evaluation model to derive the evaluation rate of each user's click-through rate for a certain product, to predict the matching degree of a marketing product and a user, and to determine the recommended rating to the user based on the click-rate evaluation value.
  • Marketing products can more accurately measure the matching degree between users and marketing products, making marketing products more targeted with users, reducing the interference caused by invalid marketing information to users.
  • a user of a third party payment platform uses a payment service of a third party payment platform through a client application (application) installed on the terminal.
  • a third-party payment platform can recommend various marketing products to users. Due to the large number of marketing products and the limited advertising space, the server of the third-party payment platform decides which marketing products to recommend to users.
  • the third-party payment platform uses user characteristics to characterize the user's behavior and background, and comprehensively and meticulously portrays the user's portrait.
  • the user characteristics include the user's background characteristics and the user's behavior characteristics, and characterize the user's inherent characteristics and real-time dynamic behavior characteristics from static and dynamic dimensions, respectively.
  • the background feature of the user is constructed by the user's registration information on the third-party payment platform and the personal information retained by the user in other associated service providers, including: user gender (U2), age (U5), and years of education (U6).
  • the behavior characteristics of the user are based on the behavior data of the user in the App (such as display of various services, click browsing, purchase, etc.) behavior data, and the behavior data of the user in other related service providers (such as product purchase, advertisement click , video browsing, etc.), can also refer to other user information (such as mobile phone brand, real estate information, etc.) to generate, including purchasing power (U1), risk preference (U3), investment experience (U4).
  • U1, U2, and U3 are discrete features with a value of 0 or 1; U4, U5, and U6 are continuous features.
  • Third-party payment platforms use product features to characterize the various marketing products that can be recommended to users.
  • Product characteristics consist of product attributes that describe the product itself and marketing attributes that describe marketing characteristics.
  • Product attributes include product type (C1), product acclaim (C2), product sales (C4), and marketing attributes including marketing novelty (C3).
  • C1, C2, and C3 are discrete features with a value of 0 or 1; C4 is a continuous feature.
  • the third-party payment platform periodically extracts relevant data in advance, generates user characteristics of each user, and saves them.
  • the product characteristics of the marketing product are managed by the personnel configuration of the marketing product or automatically generated by the server to extract relevant data and saved.
  • the third-party payment platform constructs a click-through rate evaluation model, and the input of the click-rate evaluation model is the user characteristic of the user, the product feature of the marketing product, and the cross-characteristic generated by the user feature and the product feature, and the output is the click-rate evaluation value.
  • the crossover feature is formed by cross-combining each discrete user feature and each discrete product feature.
  • the user features U1, U2, U3 and product features C1, C2, C3 can be combined into 9 crossover features: U1-C1. U1-C2, U1-C3, U2-C1, U2-C2, U2-C3, U3-C1, U3-C2, U3-C3.
  • the value of the cross feature is the logical AND operation result of the discrete user feature value and the discrete product feature value of the cross feature.
  • the user characteristics U1, U2, and U3 of the user 1 are: [U1_1 1][U2_1 0][U3_1 1]
  • the product features C1, C2, and C3 of the marketing product 1 are respectively: [C1_1 0 ][C2_1 1][C3_1 1]
  • the values of the nine cross-features of user 1 and marketing product 1 are: [U1_1-C1_1 0][U1_1-C2_1 1][U1_1-C3_1 1][U2_1-C1_1 0][U2_1-C2_1 0][U2_1-C3_1 0][U3_1-C1_1 0][U3_1-C1_1 0][U3_1-C1_1 0][U3_
  • the click rate evaluation model uses the Wide and Deep model shown in Figure 2.
  • the input of the deep neural network sub-model is a continuous feature, including user features U4, U5, U6 and product features C4.
  • the input of the linear submodel is a discrete feature, including user features U1, U2, U3, product features C1, C2, C3, and 9 crossover features.
  • the deep neural network sub-model adopts DNN algorithm
  • the linear sub-model adopts LR (logistic regression) algorithm, namely: deep neural network sub-model is DNN model
  • linear sub-model is LR model.
  • the output of the linear submodel and the output of the deep neural network submodel are processed by the LR Loss (logical regression loss) layer neurons to obtain the output of the Wide and Deep model.
  • the linear LR sub-model can be constrained by the L1 norm, thereby playing the role of feature screening.
  • the output of each training sample in the click-through rate evaluation model (ie, the known click-through rate of the sample) is displayed, clicked, and purchased by the user on the App for a predetermined period of time.
  • the number of times is determined.
  • the number of times a certain marketing product is displayed to the user in the App as the number of impressions PV in a predetermined period of time, and the number of times the user clicks on the marketing product displayed in the App within a predetermined time period as the number of clicks of the click, for the user within the predetermined time period
  • the number of times the product is purchased in the App as the number of purchases Trans, the user’s known click-through rate for the marketing product is:
  • each purchase behavior of the marketing product is converted into ⁇ times of click browsing behavior on the marketing product.
  • ⁇ can be set based on experience, for example 10.
  • the click rate evaluation model can be used to predict the matching degree of a user and a marketing product.
  • the App When a user uses an App of a third-party payment platform on their terminal, the App initiates a request for marketing product data to the server when opening a page with M marketing product display positions.
  • the server After receiving the request, the server queries the preference of the App login user.
  • the user's preferences may be predetermined by the server based on the user's historical behavior (eg, a transaction event has occurred, a certain type of service has historically been used, etc.).
  • the server selects N of the marketing products from all the marketing products that can be recommended to the user according to the user's preference.
  • the server queries the user characteristics of the user and the product features of the selected N marketing products, and calculates each marketing product according to the user characteristics U1, U2, U3 and the product characteristics C1, C2, and C3 of each marketing product. The value of the 9 cross-features.
  • the server is divided into N times, and 6 user features, 4 product features of each marketing product, and 9 cross-characteristics are input into the Wide and Deep click-rate evaluation model to obtain a click-rate evaluation value corresponding to N marketing products.
  • the server selects the M marketing products that best match the user according to the N click rate evaluation values, and returns the data of the M marketing products to the App, which is displayed to the user by the App.
  • the embodiment of the present specification further provides a recommendation device for marketing products.
  • the device can be implemented by software, or can be implemented by hardware or a combination of hardware and software.
  • the CPU Central Process Unit
  • the device in which the recommendation device of the marketing product is located usually includes other hardware such as a chip for transmitting and receiving wireless signals, and/or is used to implement the network.
  • Other hardware such as communication board.
  • FIG. 4 is a schematic diagram of a recommendation device for a marketing product according to an embodiment of the present disclosure, including a feature acquisition unit, a cross feature unit, a click rate evaluation unit, and a marketing product unit, wherein: the feature acquisition unit is configured to receive user-to-marketing After the request of the product data, the user characteristics of the user and the product features of each marketing product are obtained; the cross feature unit is configured to generate a cross feature based on the user feature and the product feature; the click rate evaluation unit is configured to use the user feature, the product feature, and The cross-characteristic input click-rate evaluation model obtains a click-through rate evaluation value of the user for each marketing product; the click-through rate evaluation model is a machine learning model, and uses a user characteristic of a known click-through rate and a marketing product feature sample to perform training; The marketing product unit is configured to determine, according to the click rate evaluation value, data of the M marketing products returned by the M marketing products to the user; M is a natural number.
  • the user feature includes at least one of: a background feature generated according to the user personal information, and a behavior feature generated according to the historical behavior record of the user;
  • the background feature includes one or more of the following: gender, age The years of education, the highest degree of education, the home address, and the graduate school;
  • the behavioral characteristics include one or more of the following: purchasing power, brand preference, risk appetite, investment experience, and type of preferred financial management.
  • the product feature includes at least one of the following: a product attribute for describing a product feature, a marketing attribute for describing a marketing feature; the product attribute includes one or more of the following: product type, product praise Degree, product sales; the marketing attributes include one or more of the following: marketing novelty, discount level.
  • the user feature includes an S item discrete feature
  • the product feature includes a T item discrete feature
  • each discrete user feature and the discrete product feature have a value of 0 or 1
  • S and T are natural numbers.
  • the cross-characteristic unit is specifically configured to: combine each discrete user feature and each discrete product feature into S ⁇ T-term cross-over features, and use discrete logic user feature values and discrete product feature values. And the result of the operation as the value of the cross feature.
  • the known click rate is determined according to the number of impressions and the click synthesis number; the number of impressions is the number of times the marketing product is displayed to the user within a predetermined time period; the click synthesis number is based on the user for the marketing within a predetermined time period
  • the behavior of the product and the number of behaviors are determined, and the behavior of the user for the marketing product includes at least one of the following: actively browsing, collecting, commenting, booking, and purchasing the marketing product.
  • the known click rate is determined according to the following manner: the number of times the user purchases the marketing product within a predetermined time period is converted into a click amount according to a predetermined ratio, and the number of times the user clicks to browse the marketing product within a predetermined time period The sum of the converted clicks is used as the click synthesis number, and the ratio of the click comprehensive number to the display number is used as the known click rate.
  • the click rate evaluation model is a depth and breadth Wide and Deep model
  • the Wide and Deep model includes a linear sub-model and a deep neural network sub-model, and the discrete feature is used as an input of the linear sub-model, The continuous feature is used as an input to the deep neural network submodel.
  • the linear submodel is a logistic regression LR model constrained by an L1 norm.
  • the feature obtaining unit is configured to: after receiving the user's request for marketing product data, select N marketing products from all marketing products according to the user's preference, and obtain user characteristics of the user. And the product characteristics of the selected N marketing products; N is a natural number not less than M.
  • Embodiments of the present specification provide a computer device including a memory and a processor.
  • the computer stores a computer program executable by the processor; and when the processor runs the stored computer program, the processor executes the steps of the recommended method of marketing the product in the embodiment of the present specification. A detailed description of each step of the recommended method of marketing products can be found in the previous section and will not be repeated.
  • Embodiments of the present specification provide a computer readable storage medium having stored thereon computer programs that, when executed by a processor, perform various steps of a method of recommending a marketing product in an embodiment of the present specification. A detailed description of each step of the recommended method of marketing products can be found in the previous section and will not be repeated.
  • a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
  • processors CPUs
  • input/output interfaces network interfaces
  • memory volatile and non-volatile memory
  • the memory may include non-persistent memory, random access memory (RAM), and/or non-volatile memory in a computer readable medium, such as read only memory (ROM) or flash memory.
  • RAM random access memory
  • ROM read only memory
  • Memory is an example of a computer readable medium.
  • Computer readable media includes both permanent and non-persistent, removable and non-removable media.
  • Information storage can be implemented by any method or technology.
  • the information can be computer readable instructions, data structures, modules of programs, or other data.
  • Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory. (ROM), electrically erasable programmable read only memory (EEPROM), flash memory or other memory technology, compact disk read only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, Magnetic tape cartridges, magnetic tape storage or other magnetic storage devices or any other non-transportable media can be used to store information that can be accessed by a computing device.
  • computer readable media does not include temporary storage of computer readable media, such as modulated data signals and carrier waves.
  • embodiments of the present specification can be provided as a method, system, or computer program product.
  • embodiments of the present specification can take the form of an entirely hardware embodiment, an entirely software embodiment or a combination of software and hardware.
  • embodiments of the present specification can take the form of a computer program product embodied on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer usable program code embodied therein. .

Landscapes

  • Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Engineering & Computer Science (AREA)
  • Accounting & Taxation (AREA)
  • Development Economics (AREA)
  • Finance (AREA)
  • Physics & Mathematics (AREA)
  • Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Marketing (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Probability & Statistics with Applications (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本说明书提供一种营销产品的推荐方法，包括：在收到用户对营销产品数据的请求后，获取所述用户的用户特征和每个营销产品的产品特征；基于用户特征和产品特征生成交叉特征；将用户特征、产品特征和交叉特征输入点击率评估模型，得到所述用户对每个营销产品的点击率评估值；所述点击率评估模型为机器学习模型，采用已知点击率的用户特征和营销产品特征样本进行训练；根据所述点击率评估值确定M个营销产品，向用户返回的所述M个营销产品的数据；M为自然数。

Description

营销产品的推荐方法和装置
相关申请的交叉引用
本专利申请要求于2018年2月28日提交的、申请号为201810164790.5、发明名称为“营销产品的推荐方法和装置”的中国专利申请的优先权，该申请的全文以引用的方式并入本文中。
技术领域
本说明书涉及数据处理技术领域，尤其涉及一种营销产品的推荐方法和装置。
背景技术
网络技术的发展，使得人们能够超越时间约束和空间限制，随时随地的获取信息、并通过与他人或企业进行信息交换来完成生活和工作中的各种事项。随着人们将更多的注意力和时间转移到网络上，企业逐渐将营销资源更多的投入到互联网营销中，来适应人们行为的变化。
互联网营销最重要的优势之一在于具备一对一的营销能力，能够针对不同的用户推荐不同的营销产品。如何将更加符合用户的兴趣和需求的营销产品推荐给用户，是提高营销效率、避免无效信息对用户造成干扰的关键所在。
发明内容
有鉴于此，本说明书提供一种营销产品的推荐方法，包括：
在收到用户对营销产品数据的请求后，获取所述用户的用户特征和每个营销产品的产品特征；
基于用户特征和产品特征生成交叉特征；
将用户特征、产品特征和交叉特征输入点击率评估模型，得到所述用户对每个营销产品的点击率评估值；所述点击率评估模型为机器学习模型，采用已知点击率的用户特征和营销产品特征样本进行训练；
根据所述点击率评估值确定M个营销产品，向用户返回的所述M个营销产品的数据；M为自然数。
本说明书还提供了一种营销产品的推荐装置，包括：
特征获取单元，用于在收到用户对营销产品数据的请求后，获取所述用户的用户特征和每个营销产品的产品特征；
交叉特征单元，用于基于用户特征和产品特征生成交叉特征；
点击率评估单元，用于将用户特征、产品特征和交叉特征输入点击率评估模型，得到所述用户对每个营销产品的点击率评估值；所述点击率评估模型为机器学习模型，采用已知点击率的用户特征和营销产品特征样本进行训练；
营销产品单元，用于根据所述点击率评估值确定M个营销产品，向用户返回的所述M个营销产品的数据；M为自然数。
本说明书提供的一种计算机设备，包括：存储器和处理器；所述存储器上存储有可由处理器运行的计算机程序；所述处理器运行所述计算机程序时，执行上述营销产品的推荐方法所述的步骤。
本说明书提供的一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器运行时，执行上述营销产品的推荐方法所述的步骤。
由以上技术方案可见，本说明书的实施例中，采用用户特征、营销产品特征、以及用户特征和产品特征生成的交叉特征来构建点击率评估模型，利用训练完成的点击率评估模型来得出各个营销产品对某个用户的点击率评估值，并根据点击率评估值来选择向该用户推荐的营销产品，从而能够更为准确的衡量用户与营销产品的匹配程度，使得推荐的营销产品更为符合用户的兴趣和需求，减少了无效的营销信息对用户造成的干扰。
附图说明
图1是本说明书实施例中一种营销产品的推荐方法的流程图；
图2是本说明书应用示例中一种点击率评估模型的结构示意图；
图3是运行本说明书实施例的设备的一种硬件结构图；
图4是本说明书实施例中一种营销产品的推荐装置的逻辑结构图。
具体实施方式
本说明书的实施例提出一种新的营销产品的推荐方法，分别采用用户特征和产品特 征来对用户和营销产品进行描述，以用户特征、产品特征、以及由用户特征和产品特征生成的交叉特征来构建点击率评估模型，以点击率评估模型输出的点击率评估值来预测某个营销产品和某个用户的匹配程度，并依据点击率评估值来确定向该用户推荐的营销产品，使得营销产品相对于用户更加有针对性，提高了营销的效率和成功率，并且大大降低了无效营销信息对用户的打扰。
本说明书的实施例可以运行在任何具有计算和存储能力的设备上，如手机、平板电脑、PC(Personal Computer，个人电脑)、笔记本、服务器等设备；还可以由运行在两个或两个以上设备的逻辑节点来实现本说明书实施例中的各项功能。
本说明书的实施例运行在网络服务提供商的服务端，当用户在自己的终端上访问服务端时，服务端在若干个营销产品中选择M(M为自然数)个推荐给用户。其中，营销产品可以是任何企业希望向用户推广的商品、服务、活动等的相关信息；营销产品的表现形式不做限定，可以是带有链接的文字、图片、动画等；营销产品的推广形式也不做限定，可以是页面上的广告位、向用户终端的消息推送、搜索结果中的优先排位等。
本说明书的实施例中，采用用户特征、产品特征和交叉特征来建立机器学习模型，本说明书中称之点击率评估模型，用来对营销产品与用户之间的匹配程度进行评估。
其中，用户特征可以是任何能够体现用户个性化特点的信息，可以将某个具体应用场景中反映对营销产品的兴趣和需求的用户个性化信息作为该应用场景中的用户特征，本说明书的实施例不做限定。
在一个例子中，可以将用户的背景特征和/或用户的行为特征来作为用户特征。其中，用户的背景特征根据用户个人信息生成，是基于静态数据刻画的用户固有特性；用户的行为特征根据用户的历史行为记录生成，是通过用户互联网行为数据构建的动态特性。用户的背景特征可以基于用户在网络服务提供商的服务端的注册信息、以及用户在其它关联站点留存的静态数据来构建；背景特征可以是：用户的性别、年龄、受教育年限、最高学历、家庭住址、毕业院校等中的一项到多项。用户的历史行为记录可以包括用户在访问网络服务提供商的服务端时对营销产品相关服务的展示、点击浏览、购买等行为，也可以包括用户在访问和使用其他服务提供商时对上述服务的展示、点击浏览、购买等行为；可以采用这些动态的历史行为记录来构建出与营销产品相关的行为特征，例如对金融消费类的营销产品，行为特征可以是：购买力、品牌偏好、风险偏好、投资经验、偏好理财类型等中的一项到多项。
产品特征可以是任何体现营销产品特性的信息，可以将某个具体应用场景中可能影响用户的兴趣和需求的营销产品特性，用来作为该应用场景中的产品特征，不做限定。例如，可以将营销产品的产品属性和/或营销属性来作为产品特征，其中，产品属性用来描述营销产品本身的产品特点，可以是产品类型、产品好评度、产品销量等中的一项到多项；产品属性用来描述营销产品的营销特点，可以是营销新颖度、折扣程度等中的一项到多项。
交叉特征是将用户特征和产品特征进行交叉组合形成组合类特征，每个交叉特征由至少一个用户特征和至少一个产品特征组合而成，该交叉特征的值也由组合该交叉特征的各个用户特征和产品特征的取值来确定。可以根据实际应用场景的需要，来确定交叉特征的数量、各个交叉特征由哪些用户特征和哪些产品特征进行交叉组合、以及如何得出交叉特征的取值，本说明书的实施例均不做限定。以下以两种实现方式为例说明。
在第一种实现方式中，交叉特征由部分用户特征和部分产品特征组合而成。可以在服务端预置在点击率评估模型中使用的每个交叉特征的取值计算方式，即：在点击率评估模型中使用了哪些交叉特征、以及如何通过组合某个交叉特征的用户特征和产品特征的值来得出该交叉特征的值。这样，在已知点击率评估模型中的用户特征取值、产品特征取值的前提下，即可计算得出每个交叉特征的取值。例如，某个点击率评估模型中使用了2个交叉特征，在服务端预置的交叉特征计算方式如表1所示：
交叉特征1 (用户特征a+用户特征b)/产品特征c
交叉特征2 用户特征d∨产品特征e
表1
在第二种实现方式中，用户特征和产品特征中包括连续型特征(即用户特征或产品特征的取值是连续的)和离散型特征(即用户特征或产品特征的取值是离散的)，交叉特征由离散型的用户特征和离散型的产品特征组合而成，交叉特征的取值由组合该交叉特征的用户特征和产品特征进行预定的逻辑运算来确定。例如，假设一个应用场景中用户特征包括S(S为自然数)项离散型特征，产品特征包括T(T为自然数)项离散型特征，所有离散型特征的取值均为0或1；该应用场景的点击率评估模型中使用了S×T项交叉特征，分别由每项离散型用户特征和每项离散型产品特征进行交叉组合而成，每项交叉特征的取值为组合该交叉特征的离散型用户特征取值和离散型产品特征取值的逻辑与运算结果。
点击率评估模型可以是采用任意算法的机器学习模型，所采用的算法可以根据实际应用场景的特点来选择，不做限定。例如，可以是基于支持向量机的机器学习模型，如SVC(Support Vector Machine，支持向量机)等；可以是基于树型的机器学习模型，如GBDT(Gradient Boosting Decision Tree，梯度提升决策树)等；可以是线性模型，如LR(Logistic Regression，逻辑回归)等；也可以是神经网络模型，如DNN(Deep Neural Networks，深度神经网络)、RNN(Recurrent Neural Networks，循环神经网络)、CNN(Convolutional Neural Networks，卷积神经网络)等。
在一种实现方式中，采用Wide and Deep(深度和广度)模型来建立点击率评估模型。Wide and Deep模型包括线性子模型和深度神经网络子模型，采用将深度神经网络子模型与浅层线性子模型相结合的训练模式。通过结合线性子模型的记忆能力(memorization)和深度神经网络子模型的泛化能力(generalization)，并且采用联合训练(joint training)，将整体模型的训练误差同时反馈到线性子模型和深度神经网络子模型中进行参数更新，同时优化2个子模型的参数，从而达到整体Wide and Deep模型的预测能力最优。可以根据实际应用场景的需要来确定将用户特征、产品特征和交叉特征中的哪些作为线性子模型的输入、哪些作为深度神经网络子模型，本说明书的实施例不做限定。例如，可以将用户特征、产品特征和交叉特征中的离散型特征作为线性子模型的输入，将连续型特征作为深度神经网络子模型的输入。
点击率评估模型的输入包括用户特征、产品特征和交叉特征，输出为点击率评估值。点击率评估模型采用已知点击率的用户特征和营销产品特征样本进行训练，即点击率评估模型的训练样本中已知的数据包括每个样本的用户特征、产品特征和作为输出的点击率，作为输入的交叉特征可以由用户特征和产品特征自动计算得出。
一个训练样本中的已知点击率可以根据具有该样本中用户特征的用户对具有该样本中产品特征的营销产品做出的行为来确定。在一个例子中，可以根据展示数和点击综合数来确定已知点击率，其中，展示数是在预定时间段内向该用户展示该营销产品的次数；点击综合数用来衡量该用户对该营销产品的综合反应，根据预定时间段内该用户针对该营销产品所进行的行为以及行为的次数确定，用户针对该营销产品所进行的行为可以是主动浏览、收藏、评论、预订、购买等行为中的一项到多项。
可以为用户对营销产品所进行的各种行为设置预定的权重，以各种行为次数的加权和来作为用户的点击综合数。假设某个应用场景中以用户对营销产品的点击浏览行为(即在向用户展示营销产品后，用户以点击或者以其他方式主动获取营销产品信息的行 为)和购买行为来计算点击综合数，则已知点击率可以根据以下方式确定：将预定时间段内用户购买营销产品的次数按预定比例折算为点击数，以预定时间段内用户点击浏览营销产品的次数与折算后的点击数之和作为点击综合数，将点击综合数相对于展示数的比例作为已知点击率。
在点击率评估模型训练完成后，即可通过模型输出的点击率评估值来衡量某个用户对某个营销产品的兴趣和需求程度，或者说用来预测该用户与该营销产品的匹配程度。在点击率评估模型中采用交叉模型作为输入，可以使点击率评估模型获得较好的泛化能力，从而更为准确的预测用户与营销产品的匹配程度。
本说明书的实施例中，营销产品的推荐方法的流程如图1所示。
步骤110，在收到用户对营销产品数据的请求后，获取该用户的用户特征和每个营销产品的产品特征。
在用户使用网络服务商提供的服务时，通过自己的终端访问网络服务提供商的服务端，向服务端发起请求。本说明书的实施例中，当服务端响应用户发起的请求时，需要向用户的终端返回营销产品数据时，该用户发起的请求即是对营销产品数据的请求；例如，用户对营销产品数据的请求可以是对要展示营销产品的网页数据的请求，也可以是对用来展示营销产品的某个广告位的数据请求。
服务端可以在收到用户对营销产品数据的请求后，生成发起请求的用户的用户特征；通常而言，为了加快响应速度，服务端会预先生成各个用户的用户特征并保存，在收到请求后进行查询即可。营销产品的产品特征可以由网络管理人员在服务端手动配置，也可以由服务端根据营销产品的相关信息自动生成，不做限定。
在一些营销产品数量比较大的应用场景中，可以先对营销产品进行预先筛选，在获取筛选出的营销产品的产品特征。具体而言，在收到用户对营销产品数据的请求后，根据该用户的偏好从所有营销产品中筛选出N(N为不小于M的自然数)个营销产品，获取该用户的用户特征和筛选出的N个营销产品的产品特征。其中，M是最终向用户推荐的营销产品的个数，N可以是一个预设的不小于M的数值，也可以是一个在满足不小于M的条件下的可变数值。预先按照用户的偏好筛选出的这N个营销产品，将作为应用本说明书实施例的全部营销产品，来在其中最终确定M个营销产品推荐给用户。另外，根据用户的偏好对营销产品进行筛选的具体方式可参照现有技术实现，不再赘述。
步骤120，基于用户特征和产品特征生成交叉特征。
在网络服务提供商的服务端得到用户特征和产品特征后，采用用户特征的取值和各个产品特征的取值，生成作为点击率评估模型输入的各个营销产品的交叉特征的值。
可以按照具体应用场景中在构建点击率评估模型时所采用的交叉特征、以及交叉特征的值的确定方式，来得到各个交叉特征的取值。
步骤130，将用户特征、产品特征和交叉特征输入点击率评估模型，得到该用户对每个营销产品的点击率评估值。
将用户特征、每个营销产品的产品特征、用户特征和该营销产品的产品特征生成的交叉特征分别输入训练完毕的点击率评估模型后，得到该用户对该营销产品的点击率评估值，来预测该用户与该营销产品的匹配程度。
步骤140，根据点击率评估值确定M个营销产品，向用户返回的该M个营销产品的数据。
基于该用户对各个营销产品的点击率评估值，服务端选择与该用户匹配程度最高的M个营销产品，将这M个营销产品的数据返回给该用户，供该用户的终端向用户展示这M个营销产品。
可见，本说明书的实施例中，分别采用用户特征和产品特征来对用户和营销产品进行描述，以用户特征、营销产品特征、以及用户特征和产品特征生成的交叉特征来构建点击率评估模型，利用训练完成的点击率评估模型来得出各个营销产品对某个用户的点击率评估值，来预测某个营销产品和某个用户的匹配程度，并依据点击率评估值来确定向该用户推荐的营销产品，能够更为准确的衡量用户与营销产品的匹配程度，使得营销产品相对于用户更加有针对性，减少了无效的营销信息对用户造成的干扰。
上述对本说明书特定实施例进行了描述。其它实施例在所附权利要求书的范围内。在一些情况下，在权利要求书中记载的动作或步骤可以按照不同于实施例中的顺序来执行并且仍然可以实现期望的结果。另外，在附图中描绘的过程不一定要求示出的特定顺序或者连续顺序才能实现期望的结果。在某些实施方式中，多任务处理和并行处理也是可以的或者可能是有利的。
在本说明书的一个应用示例中，第三方支付平台的用户通过其终端上安装的客户端App(应用程序)使用第三方支付平台的支付服务。在App页面的广告位上，第三方支付平台可以向用户推荐各种营销产品。由于营销产品数量众多而广告位相当有限，第三方支付平台的服务端要决定向用户推荐哪些营销产品。
第三方支付平台采用用户特征来刻画用户的行为和背景，全面细致的描绘出用户的画像。用户特征包括用户的背景特征和或用户的行为特征，分别从静态和动态两个维度，刻画用户的固有特性和实时动态行为特征。其中，用户的背景特征通过用户在第三方支付平台的注册信息、以及用户在其它关联服务提供方留存的个人信息来构建，包括：用户性别(U2)、年龄(U5)、受教育年限(U6)；用户的行为特征基于用户在App内的行为数据(如对各种服务的展示、点击浏览、购买等)行为数据、以及用户在其他关联服务提供方的行为数据(如商品购买、广告点击、视频浏览等)、还可以参考用户的其他信息(如使用的手机品牌、房产信息等)来生成，包括购买力(U1)、风险偏好(U3)、投资经验(U4)。用户特征中，U1、U2、和U3为离散型特征，取值为0或1；U4、U5、和U6为连续型特征。
第三方支付平台采用产品特征来刻画可以推荐给用户的各个营销产品。产品特征由描述产品本身的产品属性和描述营销特点的营销属性组成，产品属性包括产品类型(C1)、产品好评度(C2)、产品销量(C4)，营销属性包括营销新颖度(C3)。产品特征中，C1、C2、和C3为离散型特征，取值为0或1；C4为连续型特征。
第三方支付平台定期预先提取相关数据，生成每个用户的用户特征并保存。类似的，营销产品的产品特征管理该营销产品的人员配置或由服务端提取相关数据自动生成后保存。
第三方支付平台构建点击率评估模型，点击率评估模型的输入为用户的用户特征、营销产品的产品特征、以及由用户特征和产品特征生成的交叉特征，输出为点击率评估值。交叉特征由每项离散型用户特征和每项离散型产品特征进行交叉组合而成，由用户特征U1、U2、U3和产品特征C1、C2、C3可以组合为9个交叉特征：U1-C1、U1-C2、U1-C3、U2-C1、U2-C2、U2-C3、U3-C1、U3-C2、U3-C3。
交叉特征的取值为组合该交叉特征的离散型用户特征取值和离散型产品特征取值的逻辑与运算结果。假设用户1的用户特征U1、U2、U3的取值分别是：[U1_1 1][U2_1 0][U3_1 1]，营销产品1的产品特征C1、C2、C3的取值分别是：[C1_1 0][C2_1 1][C3_1 1]，则用户1与营销产品1的9个交叉特征的取值分别是：[U1_1-C1_1 0][U1_1-C2_1 1][U1_1-C3_1 1][U2_1-C1_1 0][U2_1-C2_1 0][U2_1-C3_1 0][U3_1-C1_1 0][U3_1-C2_1 1][U3_1-C3_1 1]。
点击率评估模型采用如图2所示的Wide and Deep模型。其中，深度神经网络子模型的输入为连续型特征，包括用户特征U4、U5、U6和产品特征C4。线性子模型的输 入为离散型特征，包括用户特征U1、U2、U3、产品特征C1、C2、C3、以及9个交叉特征。深度神经网络子模型采用DNN算法，线性子模型采用LR(logistic regression，逻辑回归)算法，即：深度神经网络子模型为DNN模型，线性子模型为LR模型。线性子模型的输出和深度神经网络子模型的输出经LR Loss(逻辑回归损失)层神经元进行处理后，得到Wide and Deep模型的输出。
由于在采用交叉特征后，点击率评估模型的输入特征数量通常比较大，可以通过L1范数对线性LR子模型进行约束，从而起到特征筛选的作用。
在采用样本数据训练点击率评估模型时，每个训练样本在点击率评估模型的输出(即该样本的已知点击率)由预定时间段内用户在App上的展示、点击浏览以及购买行为的次数确定。以预定时间段内某个营销产品在App中展示给用户的次数作为展示数PV，以预定时间段内用户点击App中展示的该营销产品的次数作为点击浏览次数Click，以预定时间段内用户在App中购买该营销产品的次数作为购买数Trans，则该用户对该营销产品的已知点击率为：
(Click+ω×Trans)/PV
上式中，每次对营销产品的购买行为折算为ω次对营销产品的点击浏览行为。ω可以根据经验来设置，例如10。
在采用样本数据对点击率评估模型完成训练后，即可用点击率评估模型来对某个用户和某个营销产品的匹配程度进行预测。
当用户在其终端上使用第三方支付平台的App时，在打开带有M个营销产品展示位的页面时，App向服务端发起对营销产品数据的请求。
服务端在收到该请求后，查询该App登录用户的偏好。该用户的偏好可以由服务端根据该用户的历史行为(例如发生过某个交易事件，历史上使用过某种类型服务等)预先确定。
服务端从所有可以推荐给该用户的营销产品中，按照用户的偏好筛选出其中的N个营销产品。
服务端查询该用户的用户特征和筛选出的N个营销产品的产品特征，根据用户特征U1、U2、U3和每个营销产品的产品特征C1、C2、C3的值，计算出每个营销产品的9个交叉特征的取值。
服务端分N次，将6个用户特征、每个营销产品的4个产品特征和9个交叉特征输入Wide and Deep点击率评估模型，得到对应于N个营销产品的点击率评估值。
服务端按照N个点击率评估值，选择与该用户最为匹配的M个营销产品，将M个营销产品的数据返回给App，由App展示给该用户。
与上述流程实现对应，本说明书的实施例还提供了一种营销产品的推荐装置。该装置可以通过软件实现，也可以通过硬件或者软硬件结合的方式实现。以软件实现为例，作为逻辑意义上的装置，是通过所在设备的CPU(Central Process Unit，中央处理器)将对应的计算机程序指令读取到内存中运行形成的。从硬件层面而言，除了图3所示的CPU、内存以及存储器之外，营销产品的推荐装置所在的设备通常还包括用于进行无线信号收发的芯片等其他硬件，和/或用于实现网络通信功能的板卡等其他硬件。
图4所示为本说明书实施例提供的一种营销产品的推荐装置，包括特征获取单元、交叉特征单元、点击率评估单元和营销产品单元，其中：特征获取单元用于在收到用户对营销产品数据的请求后，获取所述用户的用户特征和每个营销产品的产品特征；交叉特征单元用于基于用户特征和产品特征生成交叉特征；点击率评估单元用于将用户特征、产品特征和交叉特征输入点击率评估模型，得到所述用户对每个营销产品的点击率评估值；所述点击率评估模型为机器学习模型，采用已知点击率的用户特征和营销产品特征样本进行训练；营销产品单元用于根据所述点击率评估值确定M个营销产品，向用户返回的所述M个营销产品的数据；M为自然数。
可选的，所述用户特征包括以下至少一项：根据用户个人信息生成的背景特征、根据用户的历史行为记录生成的行为特征；所述背景特征包括以下的一项到多项：性别、年龄、受教育年限、最高学历、家庭住址、毕业院校；所述行为特征包括以下的一项到多项：购买力、品牌偏好、风险偏好、投资经验、偏好理财类型。
可选的，所述产品特征包括以下至少一项：用于描述产品特点的产品属性、用于描述营销特点的营销属性；所述产品属性包括以下的一项到多项：产品类型、产品好评度、产品销量；所述营销属性包括以下的一项到多项：营销新颖度、折扣程度。
可选的，所述用户特征包括S项离散型特征，所述产品特征包括T项离散型特征，每项离散型用户特征和离散型产品特征的取值为0或1；S、T为自然数；所述交叉特征单元具体用于：将每项离散型用户特征和每项离散型产品特征分别组合为S×T项交叉特征，以离散型用户特征取值和离散型产品特征取值的逻辑与运算结果作为交叉特征的取 值。
一个例子中，所述已知点击率根据展示数和点击综合数确定；所述展示数为预定时间段内向用户展示营销产品的次数；所述点击综合数根据预定时间段内用户针对所述营销产品所进行的行为以及行为的次数确定，用户针对所述营销产品所进行的行为包括以下至少一项：主动浏览、收藏、评论、预订、购买所述营销产品的行为。
上述例子中，所述已知点击率根据以下方式确定：将预定时间段内用户购买所述营销产品的次数按预定比例折算为点击数，以预定时间段内用户点击浏览所述营销产品的次数与折算后的点击数之和作为点击综合数，将点击综合数相对于展示数的比例作为已知点击率。
一种实现方式中，所述点击率评估模型为深度和广度Wide and Deep模型，所述Wide and Deep模型包括线性子模型和深度神经网络子模型，以离散型特征作为线性子模型的输入，以连续型特征作为深度神经网络子模型的输入。
上述实现方式中，所述线性子模型为通过L1范数进行约束的逻辑回归LR模型。
可选的，所述特征获取单元具体用于：在收到用户对营销产品数据的请求后，根据所述用户的偏好从所有营销产品中筛选出N个营销产品，获取所述用户的用户特征和筛选出的N个营销产品的产品特征；N为不小于M的自然数。
本说明书的实施例提供了一种计算机设备，该计算机设备包括存储器和处理器。其中，存储器上存储有能够由处理器运行的计算机程序；处理器在运行存储的计算机程序时，执行本说明书实施例中营销产品的推荐方法的各个步骤。对营销产品的推荐方法的各个步骤的详细描述请参见之前的内容，不再重复。
本说明书的实施例提供了一种计算机可读存储介质，该存储介质上存储有计算机程序，这些计算机程序在被处理器运行时，执行本说明书实施例中营销产品的推荐方法的各个步骤。对营销产品的推荐方法的各个步骤的详细描述请参见之前的内容，不再重复。
以上所述仅为本说明书的较佳实施例而已，并不用以限制本申请，凡在本申请的精神和原则之内，所做的任何修改、等同替换、改进等，均应包含在本申请保护的范围之内。
在一个典型的配置中，计算设备包括一个或多个处理器(CPU)、输入/输出接口、网络接口和内存。
内存可能包括计算机可读介质中的非永久性存储器，随机存取存储器(RAM)和/或非易失性内存等形式，如只读存储器(ROM)或闪存(flash RAM)。内存是计算机可读介质的示例。
计算机可读介质包括永久性和非永久性、可移动和非可移动媒体可以由任何方法或技术来实现信息存储。信息可以是计算机可读指令、数据结构、程序的模块或其他数据。计算机的存储介质的例子包括，但不限于相变内存(PRAM)、静态随机存取存储器(SRAM)、动态随机存取存储器(DRAM)、其他类型的随机存取存储器(RAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、快闪记忆体或其他内存技术、只读光盘只读存储器(CD-ROM)、数字多功能光盘(DVD)或其他光学存储、磁盒式磁带，磁带磁磁盘存储或其他磁性存储设备或任何其他非传输介质，可用于存储可以被计算设备访问的信息。按照本文中的界定，计算机可读介质不包括暂存电脑可读媒体(transitory media)，如调制的数据信号和载波。
还需要说明的是，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、商品或者设备不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、商品或者设备所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括所述要素的过程、方法、商品或者设备中还存在另外的相同要素。
本领域技术人员应明白，本说明书的实施例可提供为方法、系统或计算机程序产品。因此，本说明书的实施例可采用完全硬件实施例、完全软件实施例或结合软件和硬件方面的实施例的形式。而且，本说明书的实施例可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。

Claims (20)

  1. 一种营销产品的推荐方法，包括：
    在收到用户对营销产品数据的请求后，获取所述用户的用户特征和每个营销产品的产品特征；
    基于用户特征和产品特征生成交叉特征；
    将用户特征、产品特征和交叉特征输入点击率评估模型，得到所述用户对每个营销产品的点击率评估值；所述点击率评估模型为机器学习模型，采用已知点击率的用户特征和营销产品特征样本进行训练；
    根据所述点击率评估值确定M个营销产品，向用户返回的所述M个营销产品的数据；M为自然数。
  2. 根据权利要求1所述的方法，所述用户特征包括以下至少一项：根据用户个人信息生成的背景特征、根据用户的历史行为记录生成的行为特征；
    所述背景特征包括以下的一项到多项：性别、年龄、受教育年限、最高学历、家庭住址、毕业院校；
    所述行为特征包括以下的一项到多项：购买力、品牌偏好、风险偏好、投资经验、偏好理财类型。
  3. 根据权利要求1所述的方法，所述产品特征包括以下至少一项：用于描述产品特点的产品属性、用于描述营销特点的营销属性；
    所述产品属性包括以下的一项到多项：产品类型、产品好评度、产品销量；
    所述营销属性包括以下的一项到多项：营销新颖度、折扣程度。
  4. 根据权利要求1所述的方法，所述用户特征包括S项离散型特征，所述产品特征包括T项离散型特征，每项离散型用户特征和离散型产品特征的取值为0或1；S、T为自然数；
    所述基于用户特征和产品特征生成交叉特征生成交叉特征，包括：将每项离散型用户特征和每项离散型产品特征分别组合为S×T项交叉特征，以离散型用户特征取值和离散型产品特征取值的逻辑与运算结果作为交叉特征的取值。
  5. 根据权利要求1所述的方法，所述已知点击率根据展示数和点击综合数确定；所述展示数为预定时间段内向用户展示营销产品的次数；所述点击综合数根据预定时间段内用户针对所述营销产品所进行的行为以及行为的次数确定，用户针对所述营销产品所进行的行为包括以下至少一项：主动浏览、收藏、评论、预订、购买所述营销产品的 行为。
  6. 根据权利要求5所述的方法，所述已知点击率根据以下方式确定：将预定时间段内用户购买所述营销产品的次数按预定比例折算为点击数，以预定时间段内用户点击浏览所述营销产品的次数与折算后的点击数之和作为点击综合数，将点击综合数相对于展示数的比例作为已知点击率。
  7. 根据权利要求1所述的方法，所述点击率评估模型为深度和广度Wide and Deep模型，所述Wide and Deep模型包括线性子模型和深度神经网络子模型，以离散型特征作为线性子模型的输入，以连续型特征作为深度神经网络子模型的输入。
  8. 根据权利要求7所述的方法，所述线性子模型为通过L1范数进行约束的逻辑回归LR模型。
  9. 根据权利要求1所述的方法，所述在收到用户对营销产品数据的请求后，获取所述用户的用户特征和每个营销产品的产品特征，包括：在收到用户对营销产品数据的请求后，根据所述用户的偏好从所有营销产品中筛选出N个营销产品，获取所述用户的用户特征和筛选出的N个营销产品的产品特征；N为不小于M的自然数。
  10. 一种营销产品的推荐装置，包括：
    特征获取单元，用于在收到用户对营销产品数据的请求后，获取所述用户的用户特征和每个营销产品的产品特征；
    交叉特征单元，用于基于用户特征和产品特征生成交叉特征；
    点击率评估单元，用于将用户特征、产品特征和交叉特征输入点击率评估模型，得到所述用户对每个营销产品的点击率评估值；所述点击率评估模型为机器学习模型，采用已知点击率的用户特征和营销产品特征样本进行训练；
    营销产品单元，用于根据所述点击率评估值确定M个营销产品，向用户返回的所述M个营销产品的数据；M为自然数。
  11. 根据权利要求10所述的装置，所述用户特征包括以下至少一项：根据用户个人信息生成的背景特征、根据用户的历史行为记录生成的行为特征；
    所述背景特征包括以下的一项到多项：性别、年龄、受教育年限、最高学历、家庭住址、毕业院校；
    所述行为特征包括以下的一项到多项：购买力、品牌偏好、风险偏好、投资经验、偏好理财类型。
  12. 根据权利要求10所述的装置，所述产品特征包括以下至少一项：用于描述产品特点的产品属性、用于描述营销特点的营销属性；
    所述产品属性包括以下的一项到多项：产品类型、产品好评度、产品销量；
    所述营销属性包括以下的一项到多项：营销新颖度、折扣程度。
  13. 根据权利要求10所述的装置，所述用户特征包括S项离散型特征，所述产品特征包括T项离散型特征，每项离散型用户特征和离散型产品特征的取值为0或1；S、T为自然数；
    所述交叉特征单元具体用于：将每项离散型用户特征和每项离散型产品特征分别组合为S×T项交叉特征，以离散型用户特征取值和离散型产品特征取值的逻辑与运算结果作为交叉特征的取值。
  14. 根据权利要求10所述的装置，所述已知点击率根据展示数和点击综合数确定；所述展示数为预定时间段内向用户展示营销产品的次数；所述点击综合数根据预定时间段内用户针对所述营销产品所进行的行为以及行为的次数确定，用户针对所述营销产品所进行的行为包括以下至少一项：主动浏览、收藏、评论、预订、购买所述营销产品的行为。
  15. 根据权利要求14所述的装置，所述已知点击率根据以下方式确定：将预定时间段内用户购买所述营销产品的次数按预定比例折算为点击数，以预定时间段内用户点击浏览所述营销产品的次数与折算后的点击数之和作为点击综合数，将点击综合数相对于展示数的比例作为已知点击率。
  16. 根据权利要求10所述的装置，所述点击率评估模型为深度和广度Wide and Deep模型，所述Wide and Deep模型包括线性子模型和深度神经网络子模型，以离散型特征作为线性子模型的输入，以连续型特征作为深度神经网络子模型的输入。
  17. 根据权利要求16所述的装置，所述线性子模型为通过L1范数进行约束的逻辑回归LR模型。
  18. 根据权利要求10所述的装置，所述特征获取单元具体用于：在收到用户对营销产品数据的请求后，根据所述用户的偏好从所有营销产品中筛选出N个营销产品，获取所述用户的用户特征和筛选出的N个营销产品的产品特征；N为不小于M的自然数。
  19. 一种计算机设备，包括：存储器和处理器；所述存储器上存储有可由处理器运行的计算机程序；所述处理器运行所述计算机程序时，执行如权利要求1到9任意一项所述的步骤。
  20. 一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器运行时，执行如权利要求1到9任意一项所述的步骤。
PCT/CN2019/073610 2018-02-28 2019-01-29 营销产品的推荐方法和装置 WO2019165872A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810164790.5 2018-02-28
CN201810164790.5A CN108416625A (zh) 2018-02-28 2018-02-28 营销产品的推荐方法和装置

Publications (1)

Publication Number Publication Date
WO2019165872A1 true WO2019165872A1 (zh) 2019-09-06

Family

ID=63129272

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/073610 WO2019165872A1 (zh) 2018-02-28 2019-01-29 营销产品的推荐方法和装置

Country Status (3)

Country Link
CN (1) CN108416625A (zh)
TW (1) TWI690880B (zh)
WO (1) WO2019165872A1 (zh)

Families Citing this family (34)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108416625A (zh) * 2018-02-28 2018-08-17 阿里巴巴集团控股有限公司 营销产品的推荐方法和装置
CN109634995B (zh) * 2018-09-10 2023-06-30 创新先进技术有限公司 评估主体对关系的方法、装置及服务器
CN109447244A (zh) * 2018-10-11 2019-03-08 中山大学 一种结合门控循环单元神经网络的广告推荐方法
CN111105256A (zh) * 2018-10-29 2020-05-05 中国移动通信集团重庆有限公司 营销活动效果的分析方法、装置、设备及介质
CN109543069B (zh) * 2018-10-31 2021-07-13 北京达佳互联信息技术有限公司 视频推荐方法、装置和计算机可读存储介质
CN109670077B (zh) * 2018-11-01 2021-07-13 北京达佳互联信息技术有限公司 视频推荐方法、装置和计算机可读存储介质
CN111381909B (zh) * 2018-12-27 2022-03-04 北京字节跳动网络技术有限公司 一种页面展示方法、装置、终端设备及存储介质
CN110263243B (zh) * 2019-01-23 2024-05-10 深圳市雅阅科技有限公司 媒体信息推荐方法、装置、存储介质和计算机设备
CN110046952B (zh) * 2019-01-30 2021-12-10 创新先进技术有限公司 一种推荐模型的训练方法及装置、一种推荐方法及装置
CN110008399B (zh) * 2019-01-30 2023-04-07 创新先进技术有限公司 一种推荐模型的训练方法及装置、一种推荐方法及装置
CN110033342A (zh) * 2019-01-30 2019-07-19 阿里巴巴集团控股有限公司 一种推荐模型的训练方法及装置、一种推荐方法及装置
CN110263244B (zh) * 2019-02-14 2024-02-13 深圳市雅阅科技有限公司 内容推荐方法、装置、存储介质和计算机设备
CN111598638B (zh) * 2019-02-21 2023-11-07 北京沃东天骏信息技术有限公司 点击率确定方法、装置及设备
CN109960759B (zh) * 2019-03-22 2022-07-12 中山大学 基于深度神经网络的推荐系统点击率预测方法
CN111782927B (zh) * 2019-05-15 2024-05-17 北京京东尚科信息技术有限公司 物品推荐方法及其装置、计算机可存储介质
CN110147851B (zh) * 2019-05-29 2022-04-01 北京达佳互联信息技术有限公司 图像筛选方法、装置、计算机设备及存储介质
CN110276642A (zh) * 2019-06-20 2019-09-24 卓尔智联(武汉)研究院有限公司 广告牌推荐方法、装置及计算机可读存储介质
CN110378726A (zh) * 2019-07-02 2019-10-25 阿里巴巴集团控股有限公司 一种目标用户的推荐方法、系统以及电子设备
CN112182359B (zh) * 2019-07-05 2024-03-15 深圳市雅阅科技有限公司 推荐模型的特征管理方法及系统
CN110599234A (zh) * 2019-08-13 2019-12-20 广东工业大学 一种产品销售预测方法
CN110598845B (zh) * 2019-08-13 2023-04-07 中国平安人寿保险股份有限公司 数据处理方法、装置、计算机设备和存储介质
CN110781321B (zh) * 2019-08-28 2023-06-20 腾讯科技（深圳）有限公司 一种多媒体内容推荐方法及装置
CN110880124A (zh) * 2019-09-29 2020-03-13 清华大学 转化率评估方法及装置
CN110827086A (zh) * 2019-11-07 2020-02-21 恩亿科(北京)数据科技有限公司 产品营销预测方法、装置、计算机设备及可读存储介质
CN111291264B (zh) * 2020-01-23 2023-06-23 腾讯科技（深圳）有限公司 基于机器学习的访问对象预测方法、装置和计算机设备
CN111275350B (zh) * 2020-02-08 2021-06-04 支付宝(杭州)信息技术有限公司 更新事件评估模型的方法及装置
TWI793411B (zh) * 2020-03-27 2023-02-21 第一商業銀行股份有限公司 商品推薦方法及其裝置
CN111768263A (zh) * 2020-03-31 2020-10-13 北京京东尚科信息技术有限公司 信息推送方法、装置、服务器及存储介质
CN111177575B (zh) 2020-04-07 2020-07-24 腾讯科技（深圳）有限公司 一种内容推荐方法、装置、电子设备和存储介质
CN113761329A (zh) * 2020-09-04 2021-12-07 北京沃东天骏信息技术有限公司 页面生成方法和装置
JP7190479B2 (ja) * 2020-12-28 2022-12-15 楽天グループ株式会社 学習装置、機械学習モデル及び学習方法
CN113222647B (zh) * 2021-04-26 2023-11-28 西安点告网络科技有限公司 基于点击率预估模型的广告推荐方法、系统及存储介质
CN113256368B (zh) * 2021-04-30 2022-02-18 索信达(北京)数据技术有限公司 产品推送方法、装置、计算机设备和存储介质
CN113869973A (zh) * 2021-09-15 2021-12-31 中信建投证券股份有限公司 产品推荐方法、产品推荐系统和计算机可读存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105528374A (zh) * 2014-10-21 2016-04-27 苏宁云商集团股份有限公司 一种电子商务中的商品推荐方法及其系统
CN106295832A (zh) * 2015-05-12 2017-01-04 阿里巴巴集团控股有限公司 产品信息推送方法及装置
CN106485562A (zh) * 2015-09-01 2017-03-08 苏宁云商集团股份有限公司 一种基于用户历史行为的商品信息推荐方法及系统
CN107705183A (zh) * 2017-09-30 2018-02-16 深圳乐信软件技术有限公司 一种商品的推荐方法、装置、存储介质及服务器
CN108416625A (zh) * 2018-02-28 2018-08-17 阿里巴巴集团控股有限公司 营销产品的推荐方法和装置

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI433042B (zh) * 2010-10-06 2014-04-01 Chunghwa Telecom Co Ltd Multi - product recommendation method and system
CN102956009B (zh) * 2011-08-16 2017-03-01 阿里巴巴集团控股有限公司 一种基于用户行为的电子商务信息推荐方法与装置
CN105469263A (zh) * 2014-09-24 2016-04-06 阿里巴巴集团控股有限公司 一种商品推荐方法及装置
EP3427193A1 (en) * 2016-04-13 2019-01-16 Google LLC Wide and deep machine learning models

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105528374A (zh) * 2014-10-21 2016-04-27 苏宁云商集团股份有限公司 一种电子商务中的商品推荐方法及其系统
CN106295832A (zh) * 2015-05-12 2017-01-04 阿里巴巴集团控股有限公司 产品信息推送方法及装置
CN106485562A (zh) * 2015-09-01 2017-03-08 苏宁云商集团股份有限公司 一种基于用户历史行为的商品信息推荐方法及系统
CN107705183A (zh) * 2017-09-30 2018-02-16 深圳乐信软件技术有限公司 一种商品的推荐方法、装置、存储介质及服务器
CN108416625A (zh) * 2018-02-28 2018-08-17 阿里巴巴集团控股有限公司 营销产品的推荐方法和装置

Also Published As

Publication number Publication date
TW201937428A (zh) 2019-09-16
TWI690880B (zh) 2020-04-11
CN108416625A (zh) 2018-08-17

Similar Documents

Publication Publication Date Title
WO2019165872A1 (zh) 营销产品的推荐方法和装置
Liu et al. Effects of freemium strategy in the mobile app market: An empirical study of google play
US11023921B2 (en) Providing data and analysis for advertising on networked devices
US20200410531A1 (en) Methods, systems, and apparatus for enhancing electronic commerce using social media
US7921107B2 (en) System for generating query suggestions using a network of users and advertisers
US10354184B1 (en) Joint modeling of user behavior
US8788321B2 (en) Marketing method and system using domain knowledge
US20080004884A1 (en) Employment of offline behavior to display online content
US20080270248A1 (en) System and device for social shopping on-line
US20130132437A1 (en) Optimizing internet campaigns
TW201528181A (zh) 搜尋結果標定系統與方法
Gu et al. Differential impact of web and mobile interactivity on e-retailers' performance
US20150294359A1 (en) Methods and systems for taxonomy-driven marketing campaign creation from a repository
BANICA et al. ANALYZING SOCIAL NETWORKS FROM THE PERSPECTIVE OF MARKETING DECISIONS.
Diachuk et al. Content marketing model for leading web content management
Wu et al. The Influence of E‐Marketing on Performance of Real Estate Enterprises: Based on Super‐Efficiency DEA and Grey Entropy Methods
Hong Predicting user-level marketing performance of location-based social networking sites
Chanthati Website Visitor Analysis & Branding Quality Measurement Using Artificial Intelligence
US11593835B2 (en) Price mining prevention and dynamic online marketing campaign adjustment data processing systems and methods
Žitkienė et al. Model of impact of social networks on internet marketing of enterprises
KR102557299B1 (ko) 사용자를 위한 적합 마케팅 전략 수립 방법
US20150363823A1 (en) System and method for determining associations between users and multiple communication devices
US20230267507A1 (en) Generating and handling optimized consumer segments
JP2020154880A (ja) 算出装置、算出方法及び算出プログラム
Blagoeva et al. Online food purchasing consumer behaviour in North Macedonia amid COVID-19 pandemic: An extended TAM approach

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19760989

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19760989

Country of ref document: EP

Kind code of ref document: A1