WO2019242144A1 - 电子装置、偏好倾向预测方法和计算机可读存储介质 - Google Patents

电子装置、偏好倾向预测方法和计算机可读存储介质 Download PDF

Info

Publication number
WO2019242144A1
WO2019242144A1 PCT/CN2018/107486 CN2018107486W WO2019242144A1 WO 2019242144 A1 WO2019242144 A1 WO 2019242144A1 CN 2018107486 W CN2018107486 W CN 2018107486W WO 2019242144 A1 WO2019242144 A1 WO 2019242144A1
Authority
WO
WIPO (PCT)
Prior art keywords
marketing
portrait
user
product
model
Prior art date
Application number
PCT/CN2018/107486
Other languages
English (en)
French (fr)
Inventor
王建明
肖京
Original Assignee
平安科技(深圳)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 平安科技(深圳)有限公司 filed Critical 平安科技(深圳)有限公司
Publication of WO2019242144A1 publication Critical patent/WO2019242144A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q30/00Commerce
    • G06Q30/02Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201Market modelling; Market analysis; Collecting market data
    • G06Q30/0202Market predictions or forecasting for commercial activities
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q30/00Commerce
    • G06Q30/02Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201Market modelling; Market analysis; Collecting market data
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02DCLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00Energy efficient computing, e.g. low power processors, power management or thermal management

Definitions

  • This application relates to the field of product marketing prediction, and in particular, to an electronic device, a preference tendency prediction method, and a computer-readable storage medium.
  • the application provides an electronic device, a preference tendency prediction method, and a computer-readable storage medium, which are intended to improve the marketing success rate of business personnel.
  • a first aspect of the present application provides an electronic device including a memory and a processor.
  • the memory stores a preference tendency prediction system operable on the processor.
  • the preference tendency prediction system is executed by the processor, To achieve the following steps:
  • crawl the microblog speech data of the target user within the latest preset time range extract keywords from the crawled microblog speech data, and vectorize the extracted keywords to form the first user. portrait;
  • the third user portrait, the product portrait of the target product, and the preset channel contact vectors are input into a pre-trained preset structure prediction model for prediction, and it is obtained that the target user has the highest preference for the target product. Score and corresponding channel contact vector combination.
  • the second aspect of the present application provides a method for predicting a preference tendency, which includes the steps of:
  • crawl the microblog speech data of the target user within the latest preset time range extract keywords from the crawled microblog speech data, and vectorize the extracted keywords to form the first user. portrait;
  • the third user portrait, the product portrait of the target product, and the preset channel contact vectors are input into a pre-trained preset structure prediction model for prediction, and it is obtained that the target user has the highest preference for the target product. Score and corresponding channel contact vector combination.
  • a third aspect of the present application provides a computer-readable storage medium storing a preference tendency prediction system that can be executed by at least one processor such that the at least one processor Perform the following steps:
  • crawl the microblog speech data of the target user within the latest preset time range extract keywords from the crawled microblog speech data, and vectorize the extracted keywords to form the first user. portrait;
  • the third user portrait, the product portrait of the target product, and the preset channel contact vectors are input into a pre-trained preset structure prediction model for prediction, and it is obtained that the target user has the highest preference for the target product. Score and corresponding channel contact vector combination.
  • the user portrait of the target user is obtained according to the crawled target user's microblog speech data and the target user's personal attribute data in the system database
  • the target product product portrait is obtained according to the marketing characteristics of the target product.
  • the user portrait of the user, the product portrait of the target product, and each of the preset channel contact vectors are used as inputs to a trained preset structure prediction model.
  • the model predicts the target user ’s highest preference for the target product and the corresponding score.
  • Channel contact vector combination so that business personnel can determine the best channel and reach method for marketing the target product to the target user based on the channel contact vector combination obtained by the prediction model, so that the marketing is more targeted and the marketing success rate Greatly improved.
  • FIG. 1 is a schematic flowchart of an embodiment of a method for predicting a preference of this application
  • FIG. 2 is a schematic flowchart of a training process of a preset structure prediction model in a preference tendency prediction method of the present application
  • FIG. 3 is a schematic diagram of an operating environment of an embodiment of a preference tendency prediction system of the present application.
  • FIG. 4 is a program module diagram of an embodiment of a preference tendency prediction system of the present application.
  • This application proposes a method for predicting preferences.
  • FIG. 1 is a schematic flowchart of an embodiment of a method for predicting a preference tendency of the present application.
  • the preference tendency prediction method includes:
  • Step S10 Receive user information of the target user and product information of the target product
  • the input user information of the target user and the product information of the target product are received; wherein the user information includes the Weibo ID of the target user, and the product information includes marketing of the target product Characteristics (for example, the target product is advertising, and the marketing characteristics include a delivery platform, a delivery area, an advertising form, an advertising position, a major business interest orientation, a secondary business interest orientation, whether a user migration plugin, etc.).
  • the target product is advertising
  • the marketing characteristics include a delivery platform, a delivery area, an advertising form, an advertising position, a major business interest orientation, a secondary business interest orientation, whether a user migration plugin, etc.
  • Step S20 crawl the microblog speech data of the target user within the latest preset time range according to the user information, extract keywords from the crawled microblog speech data, and vectorize the extracted keywords to form First user portrait
  • a word2vec algorithm is preferably used to perform vectorization processing on the extracted keywords, and the extracted keywords are transformed into a corresponding word2vec vector.
  • the keywords extracted from the target user ’s Weibo speech data are: football, sports, World Cup, etc.
  • the word2vec algorithm is used to stitch the keywords into the corresponding word2vec vector (for example, ⁇ 0.23, 0.14, 0.15, ... ⁇ ).
  • Step S30 Obtain personal attribute data of the target user from the system database, extract keywords from the acquired personal attribute data, and vectorize the extracted keywords to form a second user portrait.
  • the system database has personal attribute data of each user.
  • the personal attribute data includes basic information such as gender and age, as well as related historical behavior data such as purchased products and customer service feedback messages.
  • keywords in the personal attribute data are extracted, and the extracted keywords are vectorized to form a second user portrait.
  • a word2vec algorithm is preferably used to perform vectorization processing on the extracted keywords, and the extracted keywords are transformed into a corresponding word2vec vector.
  • Step S40 stitch the obtained first user portrait and the second user portrait to obtain a third user portrait of the target user;
  • the two are spliced to obtain the complete user portrait of the target user (ie, the third user portrait).
  • Step S50 Determine the marketing characteristics of the target product according to the product information of the target product, and vectorize the determined marketing characteristics to form a product portrait of the target product;
  • the marketing characteristics include the delivery platform, the delivery area, the advertising form, the advertising position, the main business interest orientation, the secondary business interest orientation, whether the user migration plug-in, etc.
  • the marketing features are vectorized to form a product portrait of the target product.
  • Step S60 input the third user portrait, the product portrait of the target product, and the preset contact vector of each channel into a pre-trained preset structure prediction model for prediction, and obtain the target user's The highest score for preference tendency and the corresponding channel contact vector combination.
  • the preset channel contact vector is obtained by vectorizing the preset channels and reach methods.
  • the preset channels and reach methods include, for example, online channels, contact by phone, contact at work, and time required. ,and many more.
  • the portrait of the target user and the product of the target product are determined, the portrait of the target user (that is, the portrait of the third user) and the portrait of the target product (that is, the product portrait) are used as input to the model, and pre-training is input together with the channel contact vectors.
  • a good preset structure prediction model the preset structure prediction model predicts channel contact portraits (for example, any 5, 6 or The result of the preference tendency in the case of more channel contact vector combinations) is predicted, and the target user's highest preference tendency score for the target product and the channel contact vector combination corresponding to the highest preference tendency score situation are predicted.
  • business personnel can design a marketing strategy based on the channel and reach method corresponding to the channel contact vector combination output by the predictive structure prediction model, and market the target product to target users according to the marketing strategy, and the marketing success rate is greatly improved.
  • the user portrait of the target user is obtained according to the crawled target user ’s microblog speech data and the target user ’s personal attribute data in the system database
  • the target product product portrait is obtained according to the marketing characteristics of the target product.
  • the user portrait of the target user, the product portrait of the target product, and each of the preset channel contact vectors are used as inputs to the trained preset structure prediction model.
  • the model predicts the highest score of the target user ’s preference for the target product and the corresponding response.
  • Channel contact vector combination so that business personnel can determine the best channel and reach method for the target user to market the target product according to the channel contact vector combination obtained by the prediction model, so that the marketing is more targeted and the marketing is successful The rate has increased significantly.
  • the training process of the preset structure prediction model includes:
  • Step S1 Obtain a preset amount of historical marketing data.
  • Each historical marketing data includes marketing products, channels, reach methods, marketing characteristics and marketing objects of the marketing products.
  • a historical marketing data includes information such as: marketing products, channels, and reach methods (for example, through Phone contact, contact during business hours, etc.), the marketing characteristics of the marketing product (for example, launch platform, launch area, advertising form, business interest orientation, etc.) and marketing objects.
  • Step S2 For each historical marketing data, crawl the Weibo speech data of the marketing object within the latest preset time range before the marketing time of the marketing product, extract keywords from the crawled Weibo speech data, and Vectorize the extracted keywords to form a first user portrait;
  • the Weibo ID of the marketing object of the historical marketing data is queried, and then the latest preset time range before the marketing of the marketing product is crawled from the marketing object's Weibo statement (For example, within the last three months) of microblog speech data, extract keywords from the crawled microblog speech data, vectorize the extracted keywords and stitch them into the first user portrait; thus, each First user portrait of historical marketing data.
  • the word2vec algorithm is preferably used to vectorize the extracted keywords, and the extracted keywords are transformed into a corresponding word2vec vector.
  • the keywords extracted from the crawled microblog speech data are: football, Sports, World Cup ...
  • the word2vec algorithm is used to stitch keywords into corresponding word2vec vectors (eg ⁇ 0.23, 0.14, 0.15, ).
  • Step S3 For each historical marketing data, obtain personal attribute data of its marketing object from the system database, extract keywords from the acquired personal attribute data, and vectorize the extracted keywords to form a second user portrait ;
  • the system database stores the personal attribute data of each historical marketing object.
  • the personal attribute data of the marketing object that obtained the historical marketing data can be directly queried from the system database.
  • personal attribute data includes basic information such as gender and age, as well as related historical behavior data such as purchased products and customer service feedback messages.
  • keywords in the personal attribute data are extracted, and the extracted keywords are vectorized to form a second user portrait. In this way, a second user portrait of each historical marketing data is obtained.
  • a word2vec algorithm is preferably used to perform vectorization processing on the extracted keywords, and the extracted keywords are transformed into a corresponding word2vec vector.
  • Step S4 stitch the first user portrait and the second user portrait of the same historical marketing data to obtain a third user portrait of the historical marketing data;
  • the first user portrait and the second user portrait are stitched into a third user portrait, that is, the complete user portrait of the marketing object of the historical marketing data; thus, the first Three user portraits.
  • Step S5 For each historical marketing data, vectorize the channel and reach of the historical marketing data to form a channel contact portrait, and vectorize the marketing characteristics of the marketing products of the historical marketing data to form a product. portrait;
  • each historical marketing data For each historical marketing data, according to its channel and reach method information, its channels (for example, online channels, offline channels) and each reach method (for example, contact by phone, contact during office hours, etc.) are all vectorized The processing is converted into corresponding channel contact vectors, and all the channel contact vectors obtained by the conversion are stitched to form a channel contact portrait corresponding to the historical marketing data.
  • the marketing characteristics information of the marketing product for example, launch platform, launch area, advertising form, business interest orientation, etc.
  • step S6 the third user portrait, channel contact portrait, and product portrait corresponding to historical marketing data of successful marketing are used as positive samples, and the third user portrait, channel contact portrait, and product portrait corresponding to historical marketing data of failed marketing are used as negative samples.
  • a training set is established, and the training set is used to iteratively train the preset structure prediction model to obtain the latest model parameters.
  • the third user portrait, channel contact portrait, and product portrait corresponding to all the historical marketing data obtained are divided into positive samples (that is, the third corresponding to the historical marketing data of successful marketing) according to the marketing results of the historical marketing data (marketing success and marketing failure).
  • User portraits, channel contact portraits, and product portraits) and negative samples that is, third user portraits, channel contact portraits, and product portraits corresponding to historical marketing data for marketing failures). All positive samples and all negative samples constitute a training set.
  • the preset structure prediction model is iteratively trained, and the model parameters are updated at each iteration, and the latest model parameters are obtained after the iterative training of the preset structure prediction model is completed.
  • the prediction model of the preset structure adopts a modified google wide model; the modified google wide model uses gbdt + lr as the factor selection of the wide model, and uses the cnn model as the deep factor. Select and package a softmax model based on this.
  • this application also proposes a preference tendency prediction system.
  • FIG. 3 is a schematic diagram of an operating environment of a preferred embodiment of the preference tendency prediction system 10 of the present application.
  • the preference tendency prediction system 10 is installed and operated in the electronic device 1.
  • the electronic device 1 may be a computing device such as a desktop computer, a notebook, a palmtop computer, and a server.
  • the electronic device 1 may include, but is not limited to, a memory 11, a processor 12, and a display 13.
  • FIG. 3 only shows the electronic device 1 with components 11-13, but it should be understood that it is not required to implement all the illustrated components, and more or fewer components may be implemented instead.
  • the memory 11 may be an internal storage unit of the electronic device 1 in some embodiments, such as a hard disk or a memory of the electronic device 1.
  • the memory 11 may also be an external storage device of the electronic device 1 in other embodiments, such as a plug-in hard disk provided on the electronic device 1, a Smart Media Card (SMC), and a Secure Digital (SD). Cards, flash cards, etc.
  • the memory 11 may include both an internal storage unit of the electronic device 1 and an external storage device.
  • the memory 11 is used to store application software installed in the electronic device 1 and various types of data, such as program codes of the preference tendency prediction system 10.
  • the memory 11 may also be used to temporarily store data that has been output or is to be output.
  • the processor 12 may be a central processing unit (CPU), a microprocessor, or other data processing chip in some embodiments, and is configured to run program code or process data stored in the memory 11, for example, to perform a preference tendency prediction. System 10 and so on.
  • CPU central processing unit
  • microprocessor or other data processing chip in some embodiments, and is configured to run program code or process data stored in the memory 11, for example, to perform a preference tendency prediction. System 10 and so on.
  • the display 13 may be an LED display, a liquid crystal display, a touch-type liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, or the like.
  • the display 13 is used to display information processed in the electronic device 1 and to display a visualized user interface.
  • the components 11-13 of the electronic device 1 communicate with each other through a system bus.
  • FIG. 4 is a program module diagram of a preferred embodiment of the preference tendency prediction system 10 of the present application.
  • the preference tendency prediction system 10 may be divided into one or more modules, and the one or more modules are stored in the memory 11 and composed of one or more processors (the processor 12 in this embodiment).
  • the preference tendency prediction system 10 may be divided into a receiving module 101, a first extraction module 102, a second extraction module 103, a stitching module 104, a vectorization module 105, and a prediction module 104.
  • the module referred to in the present application refers to a series of computer program instruction segments capable of performing specific functions, which is more suitable than the program to describe the execution process of the preference tendency prediction system 10 in the electronic device 1, wherein:
  • the receiving module 101 is configured to receive user information of a target user and product information of a target product;
  • the input user information of the target user and the product information of the target product are received; wherein the user information includes the Weibo ID of the target user, and the product information includes the marketing of the target product Characteristics (for example, the target product is advertising, and the marketing characteristics include a delivery platform, a delivery area, an advertising form, an advertising position, a major business interest orientation, a secondary business interest orientation, whether a user migration plugin, etc.).
  • the target product is advertising
  • the marketing characteristics include a delivery platform, a delivery area, an advertising form, an advertising position, a major business interest orientation, a secondary business interest orientation, whether a user migration plugin, etc.
  • the first extraction module 102 is configured to crawl the microblog speech data of the target user within a recent preset time range according to the user information, extract keywords from the crawled microblog speech data, and perform the extraction on the keywords.
  • a word2vec algorithm is preferably used to perform vectorization processing on the extracted keywords, and the extracted keywords are transformed into a corresponding word2vec vector.
  • the keywords extracted from the target user ’s Weibo speech data are: football, sports, World Cup, etc.
  • the word2vec algorithm is used to stitch the keywords into corresponding word2vec vectors (for example, ⁇ 0.23, 0.14, 0.15, ... ⁇ ).
  • a second extraction module 103 configured to obtain personal attribute data of a target user from a system database, extract keywords from the acquired personal attribute data, and vectorize the extracted keywords to form a second user portrait;
  • the system database has personal attribute data of each user.
  • the personal attribute data includes basic information such as gender and age, as well as related historical behavior data such as purchased products and customer service feedback messages.
  • keywords in the personal attribute data are extracted, and the extracted keywords are vectorized to form a second user portrait.
  • a word2vec algorithm is preferably used to perform vectorization processing on the extracted keywords, and the extracted keywords are transformed into a corresponding word2vec vector.
  • a stitching module 104 configured to stitch the obtained first user portrait with a second user portrait to obtain a third user portrait of a target user;
  • the two are spliced to obtain the complete user portrait of the target user (ie, the third user portrait).
  • the vector conversion module 105 is configured to determine the marketing characteristics of the target product according to the product information of the target product, and vectorize the determined marketing characteristics to form a product portrait of the target product;
  • the marketing characteristics include the delivery platform, the delivery area, the advertising form, the advertising position, the main business interest orientation, the secondary business interest orientation, whether the user migration plug-in, etc.
  • the marketing features are vectorized to form a product portrait of the target product.
  • the prediction module 106 is configured to input the third user portrait, the product portrait of the target product, and the preset channel contact vectors into a pre-trained preset structure prediction model for prediction, and obtain the target user's The highest preference score of the target product and the corresponding channel exposure vector combination.
  • the preset channel contact vector is obtained by vectorizing the preset channels and reach methods.
  • the preset channels and reach methods include, for example, online channels, contact by phone, contact at work, and time required. ,and many more.
  • the portrait of the target user and the product of the target product are determined, the portrait of the target user (that is, the portrait of the third user) and the portrait of the target product (that is, the product portrait) are used as input to the model, and pre-training is input together with the channel contact vectors
  • a good preset structure prediction model; the preset structure prediction model predicts channel contact portraits (for example, any 5, 6 or The result of the preference tendency in the case of more channel contact vector combinations) is predicted, and the target user's highest preference tendency score for the target product and the channel contact vector combination corresponding to the highest preference tendency score situation are predicted.
  • business personnel can design a marketing strategy based on the channels and reach methods corresponding to the channel contact vector combination output by the prediction structure prediction model, and market the target product to target users according to the marketing strategy,
  • the user portrait of the target user is obtained according to the crawled target user ’s microblog speech data and the target user ’s personal attribute data in the system database
  • the target product product portrait is obtained according to the marketing characteristics of the target product.
  • the user portrait of the target user, the product portrait of the target product, and each of the preset channel contact vectors are used as inputs to the trained preset structure prediction model.
  • the model predicts the highest score of the target user ’s preference for the target product and the corresponding response.
  • Channel contact vector combination so that business personnel can determine the best channel and reach method for the target user to market the target product according to the channel contact vector combination obtained by the prediction model, so that the marketing is more targeted and the marketing is successful The rate has increased significantly.
  • the training process of the preset structure prediction model is:
  • each historical marketing data includes marketing products, channels, reach methods, marketing characteristics and marketing objects of the marketing product;
  • a historical marketing data includes information such as: marketing products, channels, and reach methods (for example, through Phone contact, contact during business hours, etc.), the marketing characteristics of the marketing product (for example, launch platform, launch area, advertising form, business interest orientation, etc.) and marketing objects.
  • the Weibo ID of the marketing object of the historical marketing data is queried, and then the latest preset time range before the marketing of the marketing product is crawled from the marketing object's Weibo statement (For example, within the last three months) of microblog speech data, extract keywords from the crawled microblog speech data, vectorize the extracted keywords and stitch them into the first user portrait; thus, each First user portrait of historical marketing data.
  • the word2vec algorithm is preferably used to vectorize the extracted keywords, and the extracted keywords are transformed into a corresponding word2vec vector.
  • the keywords extracted from the crawled microblog speech data are: football, Sports, World Cup ...
  • the word2vec algorithm is used to stitch keywords into corresponding word2vec vectors (eg ⁇ 0.23, 0.14, 0.15, ).
  • the system database stores the personal attribute data of each historical marketing object.
  • the personal attribute data of the marketing object that obtained the historical marketing data can be directly queried from the system database.
  • personal attribute data includes basic information such as gender and age, as well as related historical behavior data such as purchased products and customer service feedback messages.
  • keywords in the personal attribute data are extracted, and the extracted keywords are vectorized to form a second user portrait. In this way, a second user portrait of each historical marketing data is obtained.
  • a word2vec algorithm is preferably used to perform vectorization processing on the extracted keywords, and the extracted keywords are transformed into a corresponding word2vec vector.
  • the first user portrait and the second user portrait are stitched into a third user portrait, that is, the complete user portrait of the marketing object of the historical marketing data; thus, the first Three user portraits.
  • each historical marketing data For each historical marketing data, according to its channel and reach method information, its channels (for example, online channels, offline channels) and each reach method (for example, contact by phone, contact during office hours, etc.) are all vectorized The processing is converted into corresponding channel contact vectors, and all the channel contact vectors obtained by the conversion are stitched to form a channel contact portrait corresponding to the historical marketing data.
  • the marketing characteristics information of the marketing product for example, launch platform, launch area, advertising form, business interest orientation, etc.
  • a training set is established, and the training set is used to iteratively train the preset structure prediction model to obtain the latest model parameters.
  • the third user portrait, channel contact portrait, and product portrait corresponding to all the historical marketing data obtained are divided into positive samples (that is, the third corresponding to the historical marketing data of successful marketing) according to the marketing results of the historical marketing data (marketing success and marketing failure).
  • User portraits, channel contact portraits, and product portraits) and negative samples that is, third user portraits, channel contact portraits, and product portraits corresponding to historical marketing data for marketing failures).
  • the preset structure prediction model is iteratively trained, and the model parameters are updated at each iteration, and the latest model parameters are obtained after the iterative training of the preset structure prediction model is completed.
  • the prediction model of the preset structure adopts a modified google wide model; the modified google wide model uses gbdt + lr as the factor selection of the wide model, and uses the cnn model as the deep factor. Select and package a softmax model based on this.
  • the present application also proposes a computer-readable storage medium storing a preference tendency prediction system that can be executed by at least one processor to enable the at least one process
  • the processor executes the preference tendency prediction method in any of the above embodiments.

Landscapes

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

Abstract

一种电子装置、偏好倾向预测方法和计算机可读存储介质,该方法包括:接收目标用户的用户信息和目标产品的产品信息;根据用户信息爬取目标用户在最近预设时间范围内的微博发言数据,提取关键字进行向量化处理,形成第一用户画像;获取目标用户的个人属性数据,提取关键字进行向量化处理,形成第二用户画像;将第一用户画像与第二用户画像进行拼接得到第三用户画像;对目标产品的营销特点进行向量化处理,形成产品画像;将第三用户画像、产品画像和预设的各个渠道接触向量输入预先训练好的模型中预测,得出目标用户对目标产品的偏好倾向最高得分及对应的渠道接触向量组合。该方法提升了业务人员的营销成功率。

Description

电子装置、偏好倾向预测方法和计算机可读存储介质
本申请基于巴黎公约申明享有2018年6月19日递交的申请号为CN201810630385.8、名称为“电子装置、偏好倾向预测方法和计算机可读存储介质”中国专利申请的优先权,该中国专利申请的整体内容以参考的方式结合在本申请中。
技术领域
本申请涉及产品营销预测领域,特别涉及一种电子装置、偏好倾向预测方法和计算机可读存储介质。
背景技术
传统的营销模型,通常只是输出用户的偏好产品或用户对产品的偏好倾向分数等结果值,而并不能输出一些对业务人员的策略设计的有帮助的指向性建议,业务人员对目标用户的营销策略并没有针对性,很多时候会由于业务人员的营销策略不佳而导致营销失败,因此,设计一种可提升营销成功率的方案非常有意义。
发明内容
本申请提供一种电子装置、偏好倾向预测方法和计算机可读存储介质,旨在提升业务人员的营销成功率。
本申请第一方面提供一种电子装置,包括存储器和处理器,所述存储器上存储有可在所述处理器上运行的偏好倾向预测系统,所述偏好倾向预测系统被所述处理器执行时实现如下步骤:
接收目标用户的用户信息和目标产品的产品信息;
根据所述用户信息爬取目标用户在最近预设时间范围内的微博发言数据,从爬取到的微博发言数据提取关键字,并将提取的关键字进行向量化处理,形成第一用户画像;
从系统数据库中获取目标用户的个人属性数据,从获取的个人属性数据提取关键字,并将提取的关键字进行向量化处理,形成第二用户画像;
将得到的第一用户画像与第二用户画像进行拼接,得到目标用户的第三 用户画像;
根据所述目标产品的产品信息确定目标产品的营销特点,对确定的营销特点进行向量化处理,形成所述目标产品的产品画像;
将所述第三用户画像、目标产品的产品画像和预设的各个渠道接触向量输入预先训练好的预设结构预测模型中进行预测,得出所述目标用户对所述目标产品的偏好倾向最高得分及对应的渠道接触向量组合。
本申请第二方面提供一种偏好倾向预测方法,所述偏好倾向预测方法包括步骤:
接收目标用户的用户信息和目标产品的产品信息;
根据所述用户信息爬取目标用户在最近预设时间范围内的微博发言数据,从爬取到的微博发言数据提取关键字,并将提取的关键字进行向量化处理,形成第一用户画像;
从系统数据库中获取目标用户的个人属性数据,从获取的个人属性数据提取关键字,并将提取的关键字进行向量化处理,形成第二用户画像;
将得到的第一用户画像与第二用户画像进行拼接,得到目标用户的第三用户画像;
根据所述目标产品的产品信息确定目标产品的营销特点,对确定的营销特点进行向量化处理,形成所述目标产品的产品画像;
将所述第三用户画像、目标产品的产品画像和预设的各个渠道接触向量输入预先训练好的预设结构预测模型中进行预测,得出所述目标用户对所述目标产品的偏好倾向最高得分及对应的渠道接触向量组合。
本申请第三方面提供一种计算机可读存储介质,所述计算机可读存储介质存储有偏好倾向预测系统,所述偏好倾向预测系统可被至少一个处理器执行,以使所述至少一个处理器执行如下步骤:
接收目标用户的用户信息和目标产品的产品信息;
根据所述用户信息爬取目标用户在最近预设时间范围内的微博发言数据,从爬取到的微博发言数据提取关键字,并将提取的关键字进行向量化处理,形成第一用户画像;
从系统数据库中获取目标用户的个人属性数据,从获取的个人属性数据提取关键字,并将提取的关键字进行向量化处理,形成第二用户画像;
将得到的第一用户画像与第二用户画像进行拼接,得到目标用户的第三用户画像;
根据所述目标产品的产品信息确定目标产品的营销特点,对确定的营销特点进行向量化处理,形成所述目标产品的产品画像;
将所述第三用户画像、目标产品的产品画像和预设的各个渠道接触向量输入预先训练好的预设结构预测模型中进行预测,得出所述目标用户对所述目标产品的偏好倾向最高得分及对应的渠道接触向量组合。
本申请技术方案,根据爬取的目标用户的微博发言数据及系统数据库中的目标用户的个人属性数据得到目标用户的用户画像,以及根据目标产品的营销特点得到目标产品的产品画像,将目标用户的用户画像、目标产品的产品画像以及各个预设的渠道接触向量作为已训练好的预设结构预测模型的输入,模型预测得出目标用户对所述目标产品的偏好倾向最高得分及对应的渠道接触向量组合,从而业务人员可根据预测模型得出的渠道接触向量组合,确定对该目标用户营销该目标产品的最佳渠道和触达方式,如此,让营销更加有针对性,营销成功率大幅提升。
附图说明
图1为本申请偏好倾向预测方法一实施例的流程示意图;
图2为本申请偏好倾向预测方法中预设结构预测模型的训练流程示意图;
图3为本申请偏好倾向预测系统一实施例的运行环境示意图;
图4为本申请偏好倾向预测系统一实施例的程序模块图。
本申请目的的实现、功能特点及优点将结合实施例,参照附图做进一步说明。
具体实施方式
以下结合附图对本申请的原理和特征进行描述,所举实例只用于解释本 申请,并非用于限定本申请的范围。
本申请提出一种偏好倾向预测方法。
如图1所示,图1为本申请偏好倾向预测方法一实施例的流程示意图。
本实施例中,该偏好倾向预测方法包括:
步骤S10,接收目标用户的用户信息和目标产品的产品信息;
在需要进行目标用户对某个产品的偏好倾向预测时,接收输入的目标用户的用户信息和目标产品的产品信息;其中,用户信息中包括目标用户的微博ID,产品信息包括目标产品的营销特点(例如,目标产品为广告,营销特点包括投放平台、投放区域、广告形式、广告位置、主要商业兴趣定向、次要商业兴趣定向、是否用户迁徙插件等)。
步骤S20,根据所述用户信息爬取目标用户在最近预设时间范围内的微博发言数据,从爬取到的微博发言数据提取关键字,并将提取的关键字进行向量化处理,形成第一用户画像;
从用户信息中确认目标用户的微博ID,再通过该微博ID找到目标用户的微博,从目标用户的微博中爬取最近预设时间范围内(例如最近三个月内)的微博发言数据;再从爬取到的微博发言数据提取关键字,将提取的关键字进行向量化后拼接形成第一用户画像。本实施例优选采用word2vec算法对提取的关键字进行向量化处理,将提取的关键字转化拼接成相对应的word2vec向量。例如,从目标用户的微博发言数据提取出的关键字为:足球,体育,世界杯……,采用word2vec算法将关键字转化拼接成相对应的word2vec向量(例如{0.23,0.14,0.15,……})。
步骤S30,从系统数据库中获取目标用户的个人属性数据,从获取的个人属性数据提取关键字,并将提取的关键字进行向量化处理,形成第二用户画像;
系统数据库中具有各个用户的个人属性数据,个人属性数据包括性别、年龄等基本信息数据,以及购买过的产品、用户客服反馈留言等相关历史行为数据。在获取到目标用户的个人数据后,提取出个人属性数据中的关键字,对提取的关键字向量化处理,形成第二用户画像。本实施例优选采用word2vec算法对提取的关键字进行向量化处理,将提取的关键字转化拼接成相对应的word2vec向量。
步骤S40,将得到的第一用户画像与第二用户画像进行拼接,得到目标用户的第三用户画像;
在得到第一用户画像和第二用户画像后,将两者进行拼接得到目标用户的完整的用户画像(即第三用户画像)。
步骤S50,根据所述目标产品的产品信息确定目标产品的营销特点,对确定的营销特点进行向量化处理,形成所述目标产品的产品画像;
从接收到的产品信息中确定目标产品的营销特点(例如,营销特点包括投放平台、投放区域、广告形式、广告位置、主要商业兴趣定向、次要商业兴趣定向、是否用户迁徙插件等),将营销特点向量化后拼接形成目标产品的产品画像。
步骤S60,将所述第三用户画像、目标产品的产品画像和预设的各个渠道接触向量输入预先训练好的预设结构预测模型中进行预测,得出所述目标用户对所述目标产品的偏好倾向最高得分及对应的渠道接触向量组合。
其中,预设的渠道接触向量是由预设的渠道和触达方式向量化处理得到的,预设的渠道和触达方式例如包括:线上渠道、通过电话接触、上班时间接触、需要花费时间,等等。在确定了目标用户的画像和目标产品的画像后,将目标用户的画像(即第三用户画像)和目标产品的画像(即产品画像)作为模型的输入,与各个渠道接触向量一起输入预先训练好的预设结构预测模型中进行预测;所述预设结构预测模型会预测由预设的各个渠道接触向量中的一个或任意多个形成的渠道接触画像(例如,任意5个、6个或更多个渠道接触向量组合形成的画像)的情况下的偏好倾向结果,预测得出目标用户对目标产品的最高偏好倾向得分,以及最高偏好倾向得分情况对应的渠道接触向量组合。这样,业务人员可根据该预测结构预测模型输出的渠道接触向量组合所对应的渠道和触达方式去设计营销策略,根据该营销策略向目标用户营销该目标产品,营销成功率大幅提高。
本实施例技术方案,根据爬取的目标用户的微博发言数据及系统数据库中的目标用户的个人属性数据得到目标用户的用户画像,以及根据目标产品的营销特点得到目标产品的产品画像,将目标用户的用户画像、目标产品的产品画像以及各个预设的渠道接触向量作为已训练好的预设结构预测模型的输入,模型预测得出目标用户对所述目标产品的偏好倾向最高得分及对应的 渠道接触向量组合,从而业务人员可根据预测模型得出的渠道接触向量组合,确定对该目标用户营销该目标产品的最佳渠道和触达方式,如此,让营销更加有针对性,营销成功率大幅提升。
如图2所示,本实施例中,所述预设结构预测模型的训练过程包括:
步骤S1,获取预设数量的历史营销数据,每个历史营销数据包括营销产品、渠道、触达方式、该营销产品的营销特点及营销对象;
数据准备,直接采用从系统数据库中存储的营销记录中获取预设数量(例如,10万个)历史营销数据,一个历史营销数据包括的信息有:营销产品、渠道、触达方式(例如,通过电话接触、在上班时间接触等)、该营销产品的营销特点(例如,投放平台、投放区域、广告形式、商业兴趣定向等)及营销对象。
步骤S2,针对每个历史营销数据,爬取营销对象在该营销产品的营销时间之前的最近预设时间范围内的微博发言数据,从爬取到的微博发言数据中提取关键字,并将提取的关键字进行向量化处理,形成第一用户画像;
针对每个历史营销数据均进行以下处理,查询到该历史营销数据的营销对象的微博ID,然后从该营销对象的微博发言中爬取在该营销产品的营销之前的最近预设时间范围内(例如最近三个月内)的微博发言数据,对爬取到的微博发言数据提取关键字,将提取出的关键字进行向量化后拼接成第一用户画像;如此,得到每一个历史营销数据的第一用户画像。本实施例优选采用word2vec算法对提取的关键字进行向量化处理,将提取的关键字转化拼接成相对应的word2vec向量;例如,从爬取的微博发言数据提取出的关键字为:足球,体育,世界杯……,采用word2vec算法将关键字转化拼接成相对应的word2vec向量(例如{0.23,0.14,0.15,……})。
步骤S3,针对每个历史营销数据,从系统数据库中获取其营销对象的个人属性数据,从获取的个人属性数据中提取关键字,并将提取的关键字进行向量化处理,形成第二用户画像;
系统数据库中存储有各个历史营销对象的个人属性数据,针对获取的每个历史营销数据,从系统数据库中可以直接查询获取到该历史营销数据的营销对象的个人属性数据。其中,个人属性数据包括性别、年龄等基本信息数 据,以及购买过的产品、用户客服反馈留言等相关历史行为数据。在获取到营销对象的个人数据后,提取出个人属性数据中的关键字,对提取的关键字向量化处理,形成第二用户画像;如此,得到每一个历史营销数据的第二用户画像。本实施例优选采用word2vec算法对提取的关键字进行向量化处理,将提取的关键字转化拼接成相对应的word2vec向量。
步骤S4,将同一个历史营销数据的第一用户画像和第二用户画像进行拼接,得到该历史营销数据的第三用户画像;
对于每一个历史营销数据,将其第一用户画像与第二用户画像拼接成第三用户画像,即该历史营销数据的营销对象的完整的用户画像;如此,得到所有获取的历史营销数据的第三用户画像。
步骤S5,针对每个历史营销数据,将该历史营销数据的渠道和触达方式进行向量化处理以形成渠道接触画像,以及将该历史营销数据的营销产品的营销特点进行向量化处理以形成产品画像;
对于每一个历史营销数据,根据其渠道和触达方式信息,将其渠道(例如线上渠道、线下渠道)和各个触达方式(例如,通过电话接触、在上班时间接触等)全部向量化处理转化为对应的渠道接触向量,转化得到的所有渠道接触向量拼接形成该历史营销数据对应的渠道接触画像。对于每一个历史营销数据,根据其营销产品的营销特点信息(例如,投放平台、投放区域、广告形式、商业兴趣定向等),将该营销产品的各个营销特点进行向量化处理以拼接形成该营销产品的产品画像。
步骤S6,以营销成功的历史营销数据对应的第三用户画像、渠道接触画像及产品画像作为正样本,以营销失败的历史营销数据对应的第三用户画像、渠道接触画像及产品画像作为负样本,建立训练集,采用所述训练集对所述预设结构预测模型进行迭代训练,得到最新模型参数。
将获取的所有历史营销数据对应的第三用户画像、渠道接触画像及产品画像,根据历史营销数据的营销结果(营销成功和营销失败)分成正样本(即营销成功的历史营销数据对应的第三用户画像、渠道接触画像及产品画像)和负样本(即营销失败的历史营销数据对应的第三用户画像、渠道接触画像及产品画像),所有的正样本和所有负样本构成训练集,以对预设结构预测模型进行迭代训练,每一次迭代都更新模型参数,在预设结构预测模型迭代训 练完成后得到最新的模型参数。
优选地,本实施例中,所述预设结构的预测模型采用改造的google wide deep模型;所述改造的google wide deep模型使用gbdt+lr作为wide模型的因子选择,采用cnn模型作为deep因子的选择,及在此基础上包装一个softmax模型。
此外,本申请还提出一种偏好倾向预测系统。
请参阅图3,是本申请偏好倾向预测系统10较佳实施例的运行环境示意图。
在本实施例中,偏好倾向预测系统10安装并运行于电子装置1中。电子装置1可以是桌上型计算机、笔记本、掌上电脑及服务器等计算设备。该电子装置1可包括,但不仅限于,存储器11、处理器12及显示器13。图3仅示出了具有组件11-13的电子装置1,但是应理解的是,并不要求实施所有示出的组件,可以替代的实施更多或者更少的组件。
存储器11在一些实施例中可以是电子装置1的内部存储单元,例如该电子装置1的硬盘或内存。存储器11在另一些实施例中也可以是电子装置1的外部存储设备,例如电子装置1上配备的插接式硬盘,智能存储卡(Smart Media Card,SMC),安全数字(Secure Digital,SD)卡,闪存卡(Flash Card)等。进一步地,存储器11还可以既包括电子装置1的内部存储单元也包括外部存储设备。存储器11用于存储安装于电子装置1的应用软件及各类数据,例如偏好倾向预测系统10的程序代码等。存储器11还可以用于暂时地存储已经输出或者将要输出的数据。
处理器12在一些实施例中可以是一中央处理器(Central Processing Unit,CPU),微处理器或其他数据处理芯片,用于运行存储器11中存储的程序代码或处理数据,例如执行偏好倾向预测系统10等。
显示器13在一些实施例中可以是LED显示器、液晶显示器、触控式液晶显示器以及OLED(Organic Light-Emitting Diode,有机发光二极管)触摸器等。显示器13用于显示在电子装置1中处理的信息以及用于显示可视化的用户界面。电子装置1的部件11-13通过系统总线相互通信。
请参阅图4,是本申请偏好倾向预测系统10较佳实施例的程序模块图。 在本实施例中,偏好倾向预测系统10可以被分割成一个或多个模块,一个或者多个模块被存储于存储器11中,并由一个或多个处理器(本实施例为处理器12)所执行,以完成本申请。例如,在图4中,偏好倾向预测系统10可以被分割成接收模块101、第一提取模块102、第二提取模块103、拼接模块104、向量化模块105及预测模块104。本申请所称的模块是指能够完成特定功能的一系列计算机程序指令段,比程序更适合于描述偏好倾向预测系统10在电子装置1中的执行过程,其中:
接收模块101,用于接收目标用户的用户信息和目标产品的产品信息;
在需要进行目标用户对某个产品的偏好倾向预测时,接收输入的目标用户的用户信息和目标产品的产品信息;其中,用户信息中包括目标用户的微博ID,产品信息包括目标产品的营销特点(例如,目标产品为广告,营销特点包括投放平台、投放区域、广告形式、广告位置、主要商业兴趣定向、次要商业兴趣定向、是否用户迁徙插件等)。
第一提取模块102,用于根据所述用户信息爬取目标用户在最近预设时间范围内的微博发言数据,从爬取到的微博发言数据提取关键字,并将提取的关键字进行向量化处理,形成第一用户画像;
从用户信息中确认目标用户的微博ID,再通过该微博ID找到目标用户的微博,从目标用户的微博中爬取最近预设时间范围内(例如最近三个月内)的微博发言数据;再从爬取到的微博发言数据提取关键字,将提取的关键字进行向量化后拼接形成第一用户画像。本实施例优选采用word2vec算法对提取的关键字进行向量化处理,将提取的关键字转化拼接成相对应的word2vec向量。例如,从目标用户的微博发言数据提取出的关键字为:足球,体育,世界杯……,采用word2vec算法将关键字转化拼接成相对应的word2vec向量(例如{0.23,0.14,0.15,……})。
第二提取模块103,用于从系统数据库中获取目标用户的个人属性数据,从获取的个人属性数据提取关键字,并将提取的关键字进行向量化处理,形成第二用户画像;
系统数据库中具有各个用户的个人属性数据,个人属性数据包括性别、年龄等基本信息数据,以及购买过的产品、用户客服反馈留言等相关历史行为数据。在获取到目标用户的个人数据后,提取出个人属性数据中的关键字, 对提取的关键字向量化处理,形成第二用户画像。本实施例优选采用word2vec算法对提取的关键字进行向量化处理,将提取的关键字转化拼接成相对应的word2vec向量。
拼接模块104,用于将得到的第一用户画像与第二用户画像进行拼接,得到目标用户的第三用户画像;
在得到第一用户画像和第二用户画像后,将两者进行拼接得到目标用户的完整的用户画像(即第三用户画像)。
向量转模块105,用于根据所述目标产品的产品信息确定目标产品的营销特点,对确定的营销特点进行向量化处理,形成所述目标产品的产品画像;
从接收到的产品信息中确定目标产品的营销特点(例如,营销特点包括投放平台、投放区域、广告形式、广告位置、主要商业兴趣定向、次要商业兴趣定向、是否用户迁徙插件等),将营销特点向量化后拼接形成目标产品的产品画像。
预测模块106,用于将所述第三用户画像、目标产品的产品画像和预设的各个渠道接触向量输入预先训练好的预设结构预测模型中进行预测,得出所述目标用户对所述目标产品的偏好倾向最高得分及对应的渠道接触向量组合。
其中,预设的渠道接触向量是由预设的渠道和触达方式向量化处理得到的,预设的渠道和触达方式例如包括:线上渠道、通过电话接触、上班时间接触、需要花费时间,等等。在确定了目标用户的画像和目标产品的画像后,将目标用户的画像(即第三用户画像)和目标产品的画像(即产品画像)作为模型的输入,与各个渠道接触向量一起输入预先训练好的预设结构预测模型中进行预测;所述预设结构预测模型会预测由预设的各个渠道接触向量中的一个或任意多个形成的渠道接触画像(例如,任意5个、6个或更多个渠道接触向量组合形成的画像)的情况下的偏好倾向结果,预测得出目标用户对目标产品的最高偏好倾向得分,以及最高偏好倾向得分情况对应的渠道接触向量组合。这样,业务人员可根据该预测结构预测模型输出的渠道接触向量组合所对应的渠道和触达方式去设计营销策略,根据该营销策略向目标用户营销该目标产品,营销成功率大幅提高。
本实施例技术方案,根据爬取的目标用户的微博发言数据及系统数据库 中的目标用户的个人属性数据得到目标用户的用户画像,以及根据目标产品的营销特点得到目标产品的产品画像,将目标用户的用户画像、目标产品的产品画像以及各个预设的渠道接触向量作为已训练好的预设结构预测模型的输入,模型预测得出目标用户对所述目标产品的偏好倾向最高得分及对应的渠道接触向量组合,从而业务人员可根据预测模型得出的渠道接触向量组合,确定对该目标用户营销该目标产品的最佳渠道和触达方式,如此,让营销更加有针对性,营销成功率大幅提升。
本实施例中,所述预设结构预测模型的训练过程为:
1、获取预设数量的历史营销数据,每个历史营销数据包括营销产品、渠道、触达方式、该营销产品的营销特点及营销对象;
数据准备,直接采用从系统数据库中存储的营销记录中获取预设数量(例如,10万个)历史营销数据,一个历史营销数据包括的信息有:营销产品、渠道、触达方式(例如,通过电话接触、在上班时间接触等)、该营销产品的营销特点(例如,投放平台、投放区域、广告形式、商业兴趣定向等)及营销对象。
2、针对每个历史营销数据,爬取营销对象在该营销产品的营销时间之前的最近预设时间范围内的微博发言数据,从爬取到的微博发言数据中提取关键字,并将提取的关键字进行向量化处理,形成第一用户画像;
针对每个历史营销数据均进行以下处理,查询到该历史营销数据的营销对象的微博ID,然后从该营销对象的微博发言中爬取在该营销产品的营销之前的最近预设时间范围内(例如最近三个月内)的微博发言数据,对爬取到的微博发言数据提取关键字,将提取出的关键字进行向量化后拼接成第一用户画像;如此,得到每一个历史营销数据的第一用户画像。本实施例优选采用word2vec算法对提取的关键字进行向量化处理,将提取的关键字转化拼接成相对应的word2vec向量;例如,从爬取的微博发言数据提取出的关键字为:足球,体育,世界杯……,采用word2vec算法将关键字转化拼接成相对应的word2vec向量(例如{0.23,0.14,0.15,……})。
3、针对每个历史营销数据,从系统数据库中获取其营销对象的个人属性数据,从获取的个人属性数据中提取关键字,并将提取的关键字进行向量化 处理,形成第二用户画像;
系统数据库中存储有各个历史营销对象的个人属性数据,针对获取的每个历史营销数据,从系统数据库中可以直接查询获取到该历史营销数据的营销对象的个人属性数据。其中,个人属性数据包括性别、年龄等基本信息数据,以及购买过的产品、用户客服反馈留言等相关历史行为数据。在获取到营销对象的个人数据后,提取出个人属性数据中的关键字,对提取的关键字向量化处理,形成第二用户画像;如此,得到每一个历史营销数据的第二用户画像。本实施例优选采用word2vec算法对提取的关键字进行向量化处理,将提取的关键字转化拼接成相对应的word2vec向量。
4、将同一个历史营销数据的第一用户画像和第二用户画像进行拼接,得到该历史营销数据的第三用户画像;
对于每一个历史营销数据,将其第一用户画像与第二用户画像拼接成第三用户画像,即该历史营销数据的营销对象的完整的用户画像;如此,得到所有获取的历史营销数据的第三用户画像。
5、针对每个历史营销数据,将该历史营销数据的渠道和触达方式进行向量化处理以形成渠道接触画像,以及将该历史营销数据的营销产品的营销特点进行向量化处理以形成产品画像;
对于每一个历史营销数据,根据其渠道和触达方式信息,将其渠道(例如线上渠道、线下渠道)和各个触达方式(例如,通过电话接触、在上班时间接触等)全部向量化处理转化为对应的渠道接触向量,转化得到的所有渠道接触向量拼接形成该历史营销数据对应的渠道接触画像。对于每一个历史营销数据,根据其营销产品的营销特点信息(例如,投放平台、投放区域、广告形式、商业兴趣定向等),将该营销产品的各个营销特点进行向量化处理以拼接形成该营销产品的产品画像。
6、以营销成功的历史营销数据对应的第三用户画像、渠道接触画像及产品画像作为正样本,以营销失败的历史营销数据对应的第三用户画像、渠道接触画像及产品画像作为负样本,建立训练集,采用所述训练集对所述预设结构预测模型进行迭代训练,得到最新模型参数。
将获取的所有历史营销数据对应的第三用户画像、渠道接触画像及产品画像,根据历史营销数据的营销结果(营销成功和营销失败)分成正样本(即 营销成功的历史营销数据对应的第三用户画像、渠道接触画像及产品画像)和负样本(即营销失败的历史营销数据对应的第三用户画像、渠道接触画像及产品画像),所有的正样本和所有负样本构成训练集,以对预设结构预测模型进行迭代训练,每一次迭代都更新模型参数,在预设结构预测模型迭代训练完成后得到最新的模型参数。
优选地,本实施例中,所述预设结构的预测模型采用改造的google wide deep模型;所述改造的google wide deep模型使用gbdt+lr作为wide模型的因子选择,采用cnn模型作为deep因子的选择,及在此基础上包装一个softmax模型。
进一步地,本申请还提出一种计算机可读存储介质,所述计算机可读存储介质存储有偏好倾向预测系统,所述偏好倾向预测系统可被至少一个处理器执行,以使所述至少一个处理器执行上述任一实施例中的偏好倾向预测方法。
以上所述仅为本申请的优选实施例,并非因此限制本申请的专利范围,凡是在本申请的发明构思下,利用本申请说明书及附图内容所作的等效结构变换,或直接/间接运用在其他相关的技术领域均包括在本申请的专利保护范围内。

Claims (20)

  1. 一种电子装置,其特征在于,所述电子装置包括存储器和处理器,所述存储器上存储有可在所述处理器上运行的偏好倾向预测系统,所述偏好倾向预测系统被所述处理器执行时实现如下步骤:
    接收目标用户的用户信息和目标产品的产品信息;
    根据所述用户信息爬取目标用户在最近预设时间范围内的微博发言数据,从爬取到的微博发言数据提取关键字,并将提取的关键字进行向量化处理,形成第一用户画像;
    从系统数据库中获取目标用户的个人属性数据,从获取的个人属性数据提取关键字,并将提取的关键字进行向量化处理,形成第二用户画像;
    将得到的第一用户画像与第二用户画像进行拼接,得到目标用户的第三用户画像;
    根据所述目标产品的产品信息确定目标产品的营销特点,对确定的营销特点进行向量化处理,形成所述目标产品的产品画像;
    将所述第三用户画像、目标产品的产品画像和预设的各个渠道接触向量输入预先训练好的预设结构预测模型中进行预测,得出所述目标用户对所述目标产品的偏好倾向最高得分及对应的渠道接触向量组合。
  2. 如权利要求1所述的电子装置,其特征在于,所述预设结构预测模型为改造的google wide deep模型;所述改造的google wide deep模型使用gbdt+lr作为wide模型的因子选择,采用cnn模型作为deep因子的选择,及在此基础上包装一个softmax模型。
  3. 如权利要求1所述的电子装置,其特征在于,所述将提取的关键字进行向量化处理的步骤包括:
    采用word2vec算法将提取的关键字转化拼接成相对应的word2vec向量。
  4. 如权利要求3所述的电子装置,其特征在于,所述预设结构预测模型为改造的google wide deep模型;所述改造的google wide deep模型使用gbdt+lr作为wide模型的因子选择,采用cnn模型作为deep因子的选择,及在此基础上包装一个softmax模型。
  5. 如权利要求1所述的电子装置,其特征在于,所述预设结构预测模型的训练过程包括:
    获取预设数量的历史营销数据,每个历史营销数据包括营销产品、渠道、 触达方式、该营销产品的营销特点及营销对象;
    针对每个历史营销数据,爬取营销对象在该营销产品的营销时间之前的最近预设时间范围内的微博发言数据,从爬取到的微博发言数据中提取关键字,并将提取的关键字进行向量化处理,形成第一用户画像;
    针对每个历史营销数据,从系统数据库中获取其营销对象的个人属性数据,从获取的个人属性数据中提取关键字,并将提取的关键字进行向量化处理,形成第二用户画像;
    将同一个历史营销数据的第一用户画像和第二用户画像进行拼接,得到该历史营销数据的第三用户画像;
    针对每个历史营销数据,将该历史营销数据的渠道和触达方式进行向量化处理以形成渠道接触画像,以及将该历史营销数据的营销产品的营销特点进行向量化处理以形成产品画像;
    以营销成功的历史营销数据对应的第三用户画像、渠道接触画像及产品画像作为正样本,以营销失败的历史营销数据对应的第三用户画像、渠道接触画像及产品画像作为负样本,建立训练集,采用所述训练集对所述预设结构预测模型进行迭代训练,得到最新模型参数。
  6. 如权利要求5所述的电子装置,其特征在于,所述将提取的关键字进行向量化处理的步骤包括:
    采用word2vec算法将提取的关键字转化拼接成相对应的word2vec向量。
  7. 如权利要求5所述的电子装置,其特征在于,所述预设结构预测模型为改造的google wide deep模型;所述改造的google wide deep模型使用gbdt+lr作为wide模型的因子选择,采用cnn模型作为deep因子的选择,及在此基础上包装一个softmax模型。
  8. 一种偏好倾向预测方法,其特征在于,所述偏好倾向预测方法包括步骤:
    接收目标用户的用户信息和目标产品的产品信息;
    根据所述用户信息爬取目标用户在最近预设时间范围内的微博发言数据,从爬取到的微博发言数据提取关键字,并将提取的关键字进行向量化处理,形成第一用户画像;
    从系统数据库中获取目标用户的个人属性数据,从获取的个人属性数据 提取关键字,并将提取的关键字进行向量化处理,形成第二用户画像;
    将得到的第一用户画像与第二用户画像进行拼接,得到目标用户的第三用户画像;
    根据所述目标产品的产品信息确定目标产品的营销特点,对确定的营销特点进行向量化处理,形成所述目标产品的产品画像;
    将所述第三用户画像、目标产品的产品画像和预设的各个渠道接触向量输入预先训练好的预设结构预测模型中进行预测,得出所述目标用户对所述目标产品的偏好倾向最高得分及对应的渠道接触向量组合。
  9. 如权利要求8所述的偏好倾向预测方法,其特征在于,所述预设结构的预测模型为改造的google wide deep模型;所述改造的google wide deep模型使用gbdt+lr作为wide模型的因子选择,采用cnn模型作为deep因子的选择,及在此基础上包装一个softmax模型。
  10. 如权利要求8所述的偏好倾向预测方法,其特征在于,所述将提取的关键字进行向量化处理的步骤包括:
    采用word2vec算法将提取的关键字转化拼接成相对应的word2vec向量。
  11. 如权利要求10所述的偏好倾向预测方法,其特征在于,所述预设结构的预测模型为改造的google wide deep模型;所述改造的google wide deep模型使用gbdt+lr作为wide模型的因子选择,采用cnn模型作为deep因子的选择,及在此基础上包装一个softmax模型。
  12. 如权利要求8所述的偏好倾向预测方法,其特征在于,所述预设结构预测模型的训练过程包括:
    获取预设数量的历史营销数据,每个历史营销数据包括营销产品、渠道、触达方式、该营销产品的营销特点及营销对象;
    针对每个历史营销数据,爬取营销对象在该营销产品的营销时间之前的最近预设时间范围内的微博发言数据,从爬取到的微博发言数据中提取关键字,并将提取的关键字进行向量化处理,形成第一用户画像;
    针对每个历史营销数据,从系统数据库中获取其营销对象的个人属性数据,从获取的个人属性数据中提取关键字,并将提取的关键字进行向量化处理以形成第二用户画像;
    将同一个历史营销数据的第一用户画像和第二用户画像进行拼接,得到 该历史营销数据的第三用户画像;
    针对每个历史营销数据,将该历史营销数据的渠道和触达方式进行向量化处理以形成渠道接触画像,以及将该历史营销数据的营销产品的营销特点进行向量化处理,形成产品画像;
    以营销成功的历史营销数据对应的第三用户画像、渠道接触画像及产品画像作为正样本,以营销失败的历史营销数据对应的第三用户画像、渠道接触画像及产品画像作为负样本,建立训练集,采用所述训练集对所述预设结构预测模型进行迭代训练,得到最新模型参数。
  13. 如权利要求12所述的偏好倾向预测方法,其特征在于,所述将提取的关键字进行向量化处理的步骤包括:
    采用word2vec算法将提取的关键字转化拼接成相对应的word2vec向量。
  14. 如权利要求12所述的偏好倾向预测方法,其特征在于,所述预设结构的预测模型为改造的google wide deep模型;所述改造的google wide deep模型使用gbdt+lr作为wide模型的因子选择,采用cnn模型作为deep因子的选择,及在此基础上包装一个softmax模型。
  15. 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质存储有偏好倾向预测系统,所述偏好倾向预测系统可被至少一个处理器执行,以使所述至少一个处理器执行如下步骤:
    接收目标用户的用户信息和目标产品的产品信息;
    根据所述用户信息爬取目标用户在最近预设时间范围内的微博发言数据,从爬取到的微博发言数据提取关键字,并将提取的关键字进行向量化处理,形成第一用户画像;
    从系统数据库中获取目标用户的个人属性数据,从获取的个人属性数据提取关键字,并将提取的关键字进行向量化处理,形成第二用户画像;
    将得到的第一用户画像与第二用户画像进行拼接,得到目标用户的第三用户画像;
    根据所述目标产品的产品信息确定目标产品的营销特点,对确定的营销特点进行向量化处理,形成所述目标产品的产品画像;
    将所述第三用户画像、目标产品的产品画像和预设的各个渠道接触向量输入预先训练好的预设结构预测模型中进行预测,得出所述目标用户对所述 目标产品的偏好倾向最高得分及对应的渠道接触向量组合。
  16. 如权利要求15所述的计算机可读存储介质,其特征在于,所述将提取的关键字进行向量化处理的步骤包括:
    采用word2vec算法将提取的关键字转化拼接成相对应的word2vec向量。
  17. 如权利要求15所述的计算机可读存储介质,其特征在于,所述预设结构的预测模型为改造的google wide deep模型;所述改造的google wide deep模型使用gbdt+lr作为wide模型的因子选择,采用cnn模型作为deep因子的选择,及在此基础上包装一个softmax模型。
  18. 如权利要求15所述的计算机可读存储介质,其特征在于,所述预设结构预测模型的训练过程包括:
    获取预设数量的历史营销数据,每个历史营销数据包括营销产品、渠道、触达方式、该营销产品的营销特点及营销对象;
    针对每个历史营销数据,爬取营销对象在该营销产品的营销时间之前的最近预设时间范围内的微博发言数据,从爬取到的微博发言数据中提取关键字,并将提取的关键字进行向量化处理,形成第一用户画像;
    针对每个历史营销数据,从系统数据库中获取其营销对象的个人属性数据,从获取的个人属性数据中提取关键字,并将提取的关键字进行向量化处理,形成第二用户画像;
    将同一个历史营销数据的第一用户画像和第二用户画像进行拼接,得到该历史营销数据的第三用户画像;
    针对每个历史营销数据,将该历史营销数据的渠道和触达方式进行向量化处理以形成渠道接触画像,以及将该历史营销数据的营销产品的营销特点进行向量化处理以形成产品画像;
    以营销成功的历史营销数据对应的第三用户画像、渠道接触画像及产品画像作为正样本,以营销失败的历史营销数据对应的第三用户画像、渠道接触画像及产品画像作为负样本,建立训练集,采用所述训练集对所述预设结构预测模型进行迭代训练,得到最新模型参数。
  19. 如权利要求18所述的计算机可读存储介质,其特征在于,所述将提取的关键字进行向量化处理的步骤包括:
    采用word2vec算法将提取的关键字转化拼接成相对应的word2vec向量。
  20. 如权利要求18所述的计算机可读存储介质,其特征在于,所述预设结构的预测模型为改造的google wide deep模型;所述改造的google wide deep模型使用gbdt+lr作为wide模型的因子选择,采用cnn模型作为deep因子的选择,及在此基础上包装一个softmax模型。
PCT/CN2018/107486 2018-06-19 2018-09-26 电子装置、偏好倾向预测方法和计算机可读存储介质 WO2019242144A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810630385.8 2018-06-19
CN201810630385.8A CN108898429B (zh) 2018-06-19 2018-06-19 电子装置、偏好倾向预测方法和计算机可读存储介质

Publications (1)

Publication Number Publication Date
WO2019242144A1 true WO2019242144A1 (zh) 2019-12-26

Family

ID=64345456

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/107486 WO2019242144A1 (zh) 2018-06-19 2018-09-26 电子装置、偏好倾向预测方法和计算机可读存储介质

Country Status (2)

Country Link
CN (1) CN108898429B (zh)
WO (1) WO2019242144A1 (zh)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111784397A (zh) * 2020-06-30 2020-10-16 广东奥园奥买家电子商务有限公司 基于用户画像的社群营销方法、装置、计算机设备
CN111858871A (zh) * 2020-03-13 2020-10-30 北京嘀嘀无限科技发展有限公司 一种信息处理方法及装置
CN112052388A (zh) * 2020-08-20 2020-12-08 深思考人工智能科技(上海)有限公司 一种推荐美食店铺的方法及系统
CN113191570A (zh) * 2021-05-24 2021-07-30 中国银行股份有限公司 基于深度学习的资金规划推荐方法、装置和设备
CN113781079A (zh) * 2020-10-13 2021-12-10 北京沃东天骏信息技术有限公司 用于训练模型的方法和装置
CN114201516A (zh) * 2020-09-03 2022-03-18 腾讯科技(深圳)有限公司 一种用户画像构建的方法、信息推荐的方法以及相关装置
CN114399318A (zh) * 2022-01-19 2022-04-26 中国平安人寿保险股份有限公司 链路处理方法、装置、计算机设备及存储介质
CN117252647A (zh) * 2023-09-21 2023-12-19 深圳市聚海通达科技有限公司 一种数字化整合营销服务系统
CN117350770A (zh) * 2023-12-05 2024-01-05 西安柚子信息科技有限公司 一种基于用户画像的精准营销方法及系统

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110111124A (zh) * 2019-03-21 2019-08-09 贵州电网有限责任公司 基于渠道偏好和业务偏好的电力客户服务偏好方法及系统
CN110070391B (zh) * 2019-04-17 2020-06-19 同盾控股有限公司 数据处理方法、装置、计算机可读介质及电子设备
CN110162698B (zh) * 2019-04-18 2023-10-20 腾讯科技(深圳)有限公司 一种用户画像数据处理方法、装置及存储介质
CN110457679B (zh) * 2019-07-01 2023-11-28 深圳追一科技有限公司 用户画像的构建方法、装置、计算机设备和存储介质
CN111047364A (zh) * 2019-12-23 2020-04-21 杭州数澜科技有限公司 一种实施精准营销的方法和系统
CN111784411A (zh) * 2020-07-14 2020-10-16 中国工商银行股份有限公司 一种客户预测方法、装置、电子设备及存储介质
CN113781173A (zh) * 2021-09-13 2021-12-10 微积分创新科技(北京)股份有限公司 基于一物一码的ai机器人营销方法、系统及可存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103823894A (zh) * 2014-03-11 2014-05-28 北京大学 一种产品受众特征抽取方法
CN106127525A (zh) * 2016-06-27 2016-11-16 浙江大学 一种基于分类算法的电视购物商品推荐方法
CN106372959A (zh) * 2016-08-22 2017-02-01 广州图灵科技有限公司 一种基于互联网的用户访问行为数字营销系统及方法
US20170103402A1 (en) * 2015-10-13 2017-04-13 The Governing Council Of The University Of Toronto Systems and methods for online analysis of stakeholders
CN106779827A (zh) * 2016-12-02 2017-05-31 上海晶樵网络信息技术有限公司 一种互联网用户行为采集及分析检测的大数据方法
CN107688967A (zh) * 2017-08-24 2018-02-13 平安科技(深圳)有限公司 客户购买意向的预测方法及终端设备

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107291715A (zh) * 2016-03-30 2017-10-24 阿里巴巴集团控股有限公司 简历评估方法和装置
CN107025578A (zh) * 2017-04-13 2017-08-08 上海艾德韦宣股份有限公司 一种大数据智能营销系统及营销方法
CN107633036A (zh) * 2017-09-08 2018-01-26 广州汪汪信息技术有限公司 一种微博用户画像方法、电子设备、存储介质、系统
CN107895277A (zh) * 2017-09-30 2018-04-10 平安科技(深圳)有限公司 在应用程序中推送贷款广告的方法、电子装置及介质
CN107730389A (zh) * 2017-09-30 2018-02-23 平安科技(深圳)有限公司 电子装置、保险产品推荐方法及计算机可读存储介质

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103823894A (zh) * 2014-03-11 2014-05-28 北京大学 一种产品受众特征抽取方法
US20170103402A1 (en) * 2015-10-13 2017-04-13 The Governing Council Of The University Of Toronto Systems and methods for online analysis of stakeholders
CN106127525A (zh) * 2016-06-27 2016-11-16 浙江大学 一种基于分类算法的电视购物商品推荐方法
CN106372959A (zh) * 2016-08-22 2017-02-01 广州图灵科技有限公司 一种基于互联网的用户访问行为数字营销系统及方法
CN106779827A (zh) * 2016-12-02 2017-05-31 上海晶樵网络信息技术有限公司 一种互联网用户行为采集及分析检测的大数据方法
CN107688967A (zh) * 2017-08-24 2018-02-13 平安科技(深圳)有限公司 客户购买意向的预测方法及终端设备

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111858871A (zh) * 2020-03-13 2020-10-30 北京嘀嘀无限科技发展有限公司 一种信息处理方法及装置
CN111858871B (zh) * 2020-03-13 2023-12-05 北京嘀嘀无限科技发展有限公司 一种信息处理方法及装置
CN111784397A (zh) * 2020-06-30 2020-10-16 广东奥园奥买家电子商务有限公司 基于用户画像的社群营销方法、装置、计算机设备
CN112052388A (zh) * 2020-08-20 2020-12-08 深思考人工智能科技(上海)有限公司 一种推荐美食店铺的方法及系统
CN114201516A (zh) * 2020-09-03 2022-03-18 腾讯科技(深圳)有限公司 一种用户画像构建的方法、信息推荐的方法以及相关装置
CN113781079A (zh) * 2020-10-13 2021-12-10 北京沃东天骏信息技术有限公司 用于训练模型的方法和装置
CN113191570A (zh) * 2021-05-24 2021-07-30 中国银行股份有限公司 基于深度学习的资金规划推荐方法、装置和设备
CN114399318A (zh) * 2022-01-19 2022-04-26 中国平安人寿保险股份有限公司 链路处理方法、装置、计算机设备及存储介质
CN117252647A (zh) * 2023-09-21 2023-12-19 深圳市聚海通达科技有限公司 一种数字化整合营销服务系统
CN117350770A (zh) * 2023-12-05 2024-01-05 西安柚子信息科技有限公司 一种基于用户画像的精准营销方法及系统
CN117350770B (zh) * 2023-12-05 2024-03-08 西安柚子信息科技有限公司 一种基于用户画像的精准营销方法及系统

Also Published As

Publication number Publication date
CN108898429A (zh) 2018-11-27
CN108898429B (zh) 2023-04-18

Similar Documents

Publication Publication Date Title
WO2019242144A1 (zh) 电子装置、偏好倾向预测方法和计算机可读存储介质
US11030199B2 (en) Systems and methods for contextual retrieval and contextual display of records
CN110334272B (zh) 基于知识图谱的智能问答方法、装置及计算机存储介质
US10394854B2 (en) Inferring entity attribute values
US20190251471A1 (en) Machine learning device
US20130282704A1 (en) Search system with query refinement
WO2021151270A1 (zh) 图像结构化数据提取方法、装置、设备及存储介质
US20190349320A1 (en) System and method for automatically responding to user requests
JP5615932B2 (ja) 検索方法およびシステム
CN111566646B (zh) 用于对数据进行混淆和解码的电子装置以及用于控制该电子装置的方法
US9110923B2 (en) Ranking over hashes
CN102929950A (zh) 用于个性化搜索结果的社交网络推荐的内容和推荐成员
US20190340503A1 (en) Search system for providing free-text problem-solution searching
WO2021042763A1 (en) Image searches based on word vectors and image vectors
WO2020258481A1 (zh) 个性化文本智能推荐方法、装置及计算机可读存储介质
US20200372117A1 (en) Proximity information retrieval boost method for medical knowledge question answering systems
WO2020248495A1 (zh) 模型训练方法、装置及计算机可读存储介质
US20180039610A1 (en) Suggestions for digital forms
CN113887941B (zh) 业务流程生成方法、装置、电子设备及介质
WO2020164204A1 (zh) 文本模板识别方法、装置及计算机可读存储介质
US20230334075A1 (en) Search platform for unstructured interaction summaries
US20240256314A1 (en) Automating semantically-related computing tasks across contexts
WO2021042529A1 (zh) 文章摘要自动生成方法、装置及计算机可读存储介质
US10262063B2 (en) Method and system for providing alternative result for an online search previously with no result
WO2020056976A1 (zh) 搜索结果优化排序方法、装置、程序及计算机可读存储介质

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: 18923703

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: 18923703

Country of ref document: EP

Kind code of ref document: A1