WO2020107761A1 - 广告文案处理方法、装置、设备及计算机可读存储介质 - Google Patents

广告文案处理方法、装置、设备及计算机可读存储介质 Download PDF

Info

Publication number
WO2020107761A1
WO2020107761A1 PCT/CN2019/080300 CN2019080300W WO2020107761A1 WO 2020107761 A1 WO2020107761 A1 WO 2020107761A1 CN 2019080300 W CN2019080300 W CN 2019080300W WO 2020107761 A1 WO2020107761 A1 WO 2020107761A1
Authority
WO
WIPO (PCT)
Prior art keywords
copy
advertisement
backup
advertisement copy
different styles
Prior art date
Application number
PCT/CN2019/080300
Other languages
English (en)
French (fr)
Inventor
刘博�
郑文琛
杨强
Original Assignee
深圳前海微众银行股份有限公司
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 深圳前海微众银行股份有限公司 filed Critical 深圳前海微众银行股份有限公司
Publication of WO2020107761A1 publication Critical patent/WO2020107761A1/zh

Links

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
    • G06Q30/0241Advertisements
    • G06Q30/0276Advertisement creation
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/20Natural language analysis
    • G06F40/279Recognition of textual entities
    • G06F40/289Phrasal analysis, e.g. finite state techniques or chunking

Definitions

  • the present application relates to the field of computer technology, and in particular, to an advertisement copy processing method, device, device, and computer-readable storage medium.
  • Advertising itself is a means of propagating information to the public, and it is also an important source of income for many companies.
  • Existing advertising platforms mostly rely on the advertisers to create creative materials independently, and the creation of creative materials mostly depends on experience and labor.
  • the main purpose of the present application is to provide an advertisement copy processing method, device, equipment and computer-readable storage medium, aiming to solve the technical problem of low click rate of existing advertisement copy.
  • the advertisement copy processing method includes:
  • a preset click-through rate estimation model is used to sort and display a variety of different types of backup ad copy generated.
  • the method before the step of obtaining the source style advertisement copy, the method further includes:
  • the step of transforming the source style advertisement copy based on the preset multiple different style conversion models to generate corresponding multiple different style backup advertisement copy includes:
  • a plurality of different styles of backup advertising copy corresponding to the source style advertising copy are generated.
  • the step of generating a plurality of different styles of backup ad copy corresponding to the source style advertisement copy based on the plurality of different styles of deep neural network models includes:
  • an encoder corresponding to the source style is used to encode the source style advertisement copy into a set of vector representations
  • a plurality of decoders of different styles are used to decode the vector representation to generate a plurality of different styles of backup advertisement copy corresponding to the source style advertisement copy.
  • the step of using the preset click-through rate prediction model to sort and display the generated multiple different styles of backup advertising copy includes:
  • a preset click-through rate estimation model is used to obtain the estimated click-through rate corresponding to the generated multiple different styles of backup ad copy
  • the plurality of different styles of backup ad copy are sorted and displayed.
  • the advertising copy processing method further includes:
  • the Lasso machine learning model is used to learn the weight of each keyword to the estimated click rate
  • At least one keyword and corresponding weight are output to assist the user in editing the backup advertisement copy.
  • the step of generating corresponding multiple pseudo-copy according to the backup advertisement copy includes:
  • the backup advertisement copy is divided into multiple Chinese vocabularies
  • the at least one keyword is a partial keyword, and the weight corresponding to the output at least one keyword is greater than the weight corresponding to the non-output keyword.
  • the present application also provides an advertisement copy processing device, the advertisement copy processing device includes: a memory, a processor, and an advertisement copy processing stored on the memory and operable on the processor A program, when the advertisement copy processing program is executed by the processor, the steps of the advertisement copy processing method described above are realized.
  • the present application also provides an advertisement copy processing device.
  • the advertisement copy processing device includes: a memory, a processor, and an advertisement copy processing stored on the memory and operable on the processor A program, when the advertisement copy processing program is executed by the processor, the steps of the advertisement copy processing method described above are realized.
  • the present application also provides a computer-readable storage medium on which an advertisement copy processing program is stored, and the advertisement copy processing program is implemented as described above when executed by a processor Steps of ad copy processing method.
  • An advertisement copy processing method, device, device and computer-readable storage medium proposed in the embodiments of the present application, by acquiring source style advertisement copy, based on preset multiple different style conversion models, transform the source style advertisement copy into style ,Generate corresponding multiple different styles of backup ad copy, and use the preset click-through rate estimation model to sort and display the generated multiple different styles of backup ad copy for the creation of backup ad copy based on sorted display
  • FIG. 1 is a schematic diagram of a terminal ⁇ device structure of a hardware operating environment involved in an embodiment of the present application
  • FIG. 2 is a schematic flowchart of a first embodiment of a method for processing advertisement copy of an application
  • FIG. 3 is a schematic flowchart of a second embodiment of a method for processing ad copy of an application
  • FIG. 4 is a schematic diagram of the working scene of the encoder-decoder in the advertising copy processing method of this application;
  • FIG. 5 is a schematic flowchart of a third embodiment of a method for processing ad copy of an application
  • FIG. 6 is a schematic diagram of the overall framework of the click-through-driven copy generation and analysis method in the advertisement copy processing method of the present application.
  • FIG. 1 is a schematic diagram of a terminal structure of a hardware operating environment involved in a solution of an embodiment of the present application.
  • the terminal in the embodiment of the present application is an advertisement copy processing device.
  • the terminal may include: a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002.
  • the communication bus 1002 is used to implement connection communication between these components.
  • the user interface 1003 may include a display (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface.
  • the network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).
  • the memory 1005 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as disk storage.
  • the memory 1005 may optionally be a storage device independent of the foregoing processor 1001.
  • the terminal may also include a camera, RF (Radio Frequency (radio frequency) circuits, sensors, audio circuits, WiFi modules, etc.
  • sensors such as light sensors, motion sensors and other sensors.
  • the light sensor may include an ambient light sensor and a proximity sensor, wherein the ambient light sensor may adjust the brightness of the display screen according to the brightness of the ambient light, and the proximity sensor may turn off the display screen and/or when the terminal device moves to the ear Backlight.
  • the terminal device can also be configured with other sensors such as gyroscopes, barometers, hygrometers, thermometers, and infrared sensors, which will not be repeated here.
  • terminal structure shown in FIG. 1 does not constitute a limitation on the terminal, and may include more or fewer components than those illustrated, or combine certain components, or arrange different components.
  • the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and an advertisement copy processing program.
  • the network interface 1004 is mainly used to connect to the back-end server and perform data communication with the background server;
  • the user interface 1003 is mainly used to connect to the client (user end) and perform data communication with the client;
  • 1001 can be used to call the advertisement copy processing program stored in the memory 1005 and perform the following operations:
  • a preset click-through rate estimation model is used to sort and display a variety of different types of backup ad copy generated.
  • the advertisement copy processing method includes the following steps:
  • Step S10 obtaining source style advertisement copy
  • Step S20 based on the preset multiple different style conversion models, the source style advertisement copy is style converted to generate corresponding multiple different style backup advertisement copy;
  • step S30 the preset click-through rate estimation model is used to sort and display the generated multiple different styles of backup advertising copy
  • the advertising copy may be the entire advertising work, including the phonetic text part and the picture part of the advertising work.
  • the corresponding source style advertisement copy (that is, query copy) is first obtained.
  • the way to obtain the source-style advertisement copy can be based on the user's preferences, and the user can select an advertisement copy as the source-style advertisement copy from the various advertisement copy information pre-stored in the display terminal, or it can be an advertisement created by the user in the display terminal
  • Copywriting as source-style advertising copywriting can also be that users select an advertising copywriting as source-style advertising copywriting on the Internet based on their own preferences, without limitation.
  • the source style of the source style advertisement copy is obtained.
  • the source style includes, but is not limited to, the writing style of the ad copy, layout layout, etc.
  • a plurality of different style transformation models T i are trained.
  • N styles of corpus in the training that is, S 1 ??S N
  • all the training corpora belong to the source style
  • a corpus corresponding to a style S i is randomly selected, in the training stage, for this style S transformation model training i T i, T i transformation model to translate the source S i style to style.
  • the primary requirement of this transformation model T i is the generated advertising copy, whose language is fluent and understandable.
  • the transformation model T i for the style S i is only trained using the corpus of the style S i . It is emphasized that the transformation model for independent and source-style style S I S, T i.
  • a CTR click rate estimation model corresponding to the advertising copy is preset. After generating backup advertising copy of multiple different styles S i corresponding to the source style advertising copy, based on CTR CTR default prediction model, to generate backup ad copy many different styles S i sort of show.
  • step S30 includes:
  • Step a using a preset click-through rate estimation model to obtain the estimated click-through rate corresponding to the generated multiple different styles of backup ad copy;
  • Step b Sort and display the multiple types of backup ad copy according to the estimated click rate corresponding to the multiple types of backup ad copy.
  • each backup ad copy is input into the click-through rate prediction model based on the preset CTR click rate estimation model, respectively, to obtain each backup ad copy Corresponding estimated click-through rate.
  • the backup ad copy of multiple different styles S i is sorted and displayed.
  • the backup ad copy is sorted and displayed according to the order of the estimated click rate from high to low.
  • the user can edit each of the backup ad copy displayed to generate the target ad copy required by the user.
  • the source style advertisement copy is acquired, and based on a preset multiple different style conversion models, the source style advertisement copy is style-converted to generate corresponding multiple different style backup advertisement copy, and the preset is adopted.
  • the click-through rate estimation model of the system sorts and displays the generated backup ad copy in different styles, so that the backup ad copy creation based on the sorted display can obtain the ad copy required by the user, thereby improving the click rate of the created ad copy.
  • the method further includes:
  • Step S40 establish a variety of deep neural network models with different styles
  • the step S20 includes:
  • Step S21 based on the multiple deep neural network models of different styles, generating multiple different styles of backup advertising copy corresponding to the source style advertising copy.
  • the deep neural network model is used to learn how to generate backup ad copy in different styles while maintaining the semantics of the source style ad copy.
  • the end-to-end and sequence-to-sequence deep neural network models are generated in advance according to the source style of the source style advertisement copy, that is, each style has a corresponding deep neural network model, and each deep neural network model
  • the network models are all independent and unrelated.
  • step S21 includes:
  • Step c In the deep neural network models of different styles, an encoder corresponding to the source style is used to encode the source style advertisement copy into a set of vector representations;
  • step d a plurality of decoders with different styles are used to decode the vector representation to generate a plurality of different styles of backup advertisement copy corresponding to the source style advertisement copy.
  • the generation of backup ad copy in different styles based on the deep neural network model is based on the structure of the encoder and decoder.
  • the encoder is only responsible for the semantic information of the ad copy.
  • the decoder relies on the semantic information and Only responsible for style information.
  • the source style advertising copy is encoded into a set of vector representations using the encoder corresponding to the source style.
  • the encoder uses a recurrent neural network to convert the source style advertising copy into a set of vectors Said.
  • the vector represents the semantics of the source-style advertising copy.
  • a plurality of decoders of different styles are used to decode the vector representation to generate multiple different styles of backup advertisement copy corresponding to the source style advertisement copy.
  • the encoder and decoder train based on the reconstruction loss function.
  • FIG, 4 is independent of the style in the style of the successful S S I style changes, we assume that the vector obtained by coding with the encoder showing only relevant ad copy semantics.
  • the decoder uses semantics for text reconstruction.
  • the decoder is only related to style, not semantics.
  • a variety of different styles of backup advertising copy corresponding to the source style advertising copy are generated.
  • the training of the deep neural network model only requires different styles of corpus, and does not need Strong correspondence, therefore, the training of the deep neural network model is easy to implement, which in turn makes the generation of backup advertising copy more reliable.
  • the advertising copy processing method also includes:
  • Step S50 according to the backup advertisement copy, generate corresponding multiple fake copy
  • Step S60 For the keywords in the pseudo copy, a Lasso machine learning model is used to learn the weight of each keyword to the estimated click rate;
  • step S70 at least one keyword and corresponding weight are output to assist the user in editing the backup advertisement copy.
  • the keywords of the advertising copy are analyzed, and the obtained keywords mean that they have a significant positive or negative impact on the click rate of the advertising copy.
  • the generated backup advertisement copy corresponding multiple pseudo copy are generated.
  • the step S50 includes:
  • Step e based on the Chinese word segmentation system, the backup advertisement copy is divided into multiple Chinese vocabulary
  • Step f Randomly discard the plurality of Chinese vocabularies according to different amounts to generate the plurality of pseudo-copywriting.
  • the Chinese word segmentation system is invoked, based on the Chinese word segmentation system, the backup advertising copy is divided into multiple Chinese vocabularies (including phrases), and then, by randomly discarding different numbers of Chinese vocabularies, corresponding multiple different Fake copywriting.
  • the bag-of-words feature to represent the pseudo-copy. Specifically, if the pseudo-copy contains a Chinese vocabulary, the corresponding position in the bag-of-words feature is 1, otherwise it is 0, and the bag-of-words feature of the pseudo-copy is used as data To predict the estimated CTR.
  • the Lasso machine learning model is used for learning, and for the keywords appearing in the pseudo copy, the Lasso machine learning model is used to learn the weight of each keyword to the estimated click rate. Finally, according to user needs, at least one keyword and its corresponding weight are output. According to the keywords and their corresponding weights, assist users to edit the backup ad copy.
  • the source-style advertising copy (query copy) is first encoded by the source-style encoder, and multiple deep neural network models are trained according to different style corpora, and decoding is used in each deep neural network model decodes, as S i style decoder, different styles generate queries mass corresponding backup copy ad copy. Then use the CTR click-through rate estimation model to sort and display each alternate advertising copy, such as copy plan 1, copy plan 2, copy plan 3, etc. And according to the generated backup ad copy, generate the corresponding pseudo copy, use the Lasso machine learning model for keyword analysis, obtain the weight of the keyword in the pseudo copy to the estimated click rate, and output the keyword and its corresponding weight to assist The user edits the backup ad copy.
  • Keywords with a positive weight can increase the estimated CTR, while keywords with a negative weight will reduce the estimated CTR.
  • a part (at least one) of the keywords is retained, and the weight corresponding to the partial keyword is greater than the weight corresponding to the unoutput keyword.
  • the k keywords with the largest weights are retained, and these keywords have the greatest impact on the estimated click-through rate CTR.
  • output the k keywords with the largest weight and their weights are fed back to the user in real time, assisting the user to edit the backup ad copy.
  • the present application also provides an advertisement copy processing device.
  • the advertisement copy processing device includes: a memory, a processor, and an advertisement copy processing program stored on the memory and executable on the processor When the program is executed by the processor, the steps of the foregoing embodiments of the advertising copy processing method are implemented.
  • the specific implementation of the advertising copy processing device of the present application is basically the same as the foregoing embodiments of the advertising copy processing method, and details are not described herein again.
  • the present application also provides a computer-readable storage medium that stores one or more programs, and the one or more programs may also be executed by one or more processors for implementation The steps of the above embodiments of the advertising copy processing method.
  • the methods in the above embodiments can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware, but in many cases the former is better Implementation.
  • the technical solution of the present application can be embodied in the form of a software product in essence or part that contributes to the existing technology, and the computer software product is stored in a storage medium (such as ROM/RAM as described above) , Magnetic disk, optical disk), including several instructions to make a terminal device (which can be a mobile phone, computer, server, air conditioner, or network equipment, etc.) to perform the method described in each embodiment of the present application.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Accounting & Taxation (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Game Theory and Decision Science (AREA)
  • Artificial Intelligence (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种广告文案处理方法、装置、设备及计算机可读存储介质,该广告文案处理方法包括:获取源风格广告文案(S10);基于预设的多种不同风格的变换模型,将所述源风格广告文案进行风格变换,生成对应的多种不同风格的备用广告文案(S20);采用预设的点击率预估模型,对生成的多种不同风格的备用广告文案进行排序展示(S30)。上述广告文案处理方法提高创作的广告文案的点击率。

Description

广告文案处理方法、装置、设备及计算机可读存储介质
本申请要求于2018年11月28日提交中国专利局、申请号为201811440929.0、发明名称为“广告文案处理方法、装置、设备及计算机可读存储介质”的中国专利申请的优先权,其全部内容通过引用结合在申请中。
技术领域
本申请涉及计算机技术领域,尤其涉及一种广告文案处理方法、装置、设备及计算机可读存储介质。
背景技术
广告本身是向社会大众传递信息的宣传手段,也是许多公司的重要收入来源之一,现有的广告平台多依赖于广告投放主自主创作广告素材,而广告素材的创作多依赖于经验和人工。这就存在一个问题,无法及时地获知创作出来的广告文案是否符合用户的需求,从而有可能导致创作出来的广告文案点击率过低。因此,如何提高广告文案的点击率成为了目前亟待解决的技术问题。
发明内容
本申请的主要目的在于提供一种广告文案处理方法、装置、设备及计算机可读存储介质,旨在解决现有广告文案的点击率低的技术问题。
为实现上述目的,本申请提供一种广告文案处理方法,所述广告文案处理方法包括:
获取源风格广告文案;
基于预设的多种不同风格的变换模型,将所述源风格广告文案进行风格变换,生成对应的多种不同风格的备用广告文案;
采用预设的点击率预估模型,对生成的多种不同风格的备用广告文案进行排序展示。
可选地,所述获取源风格广告文案的步骤之前,还包括:
建立多种不同风格的深度神经网络模型;
所述基于预设的多种不同风格的变换模型,将所述源风格广告文案进行风格变换,生成对应的多种不同风格的备用广告文案的步骤包括:
基于所述多种不同风格的深度神经网络模型,生成所述源风格广告文案对应的多种不同风格的备用广告文案。
可选地,所述基于所述多种不同风格的深度神经网络模型,生成所述源风格广告文案对应的多种不同风格的备用广告文案的步骤包括:
在所述多种不同风格的深度神经网络模型中,采用源风格对应的编码器将所述源风格广告文案编码为一组向量表示;
采用多种不同风格的解码器将所述向量表示进行解码,生成所述源风格广告文案对应的多种不同风格的备用广告文案。
可选地,所述采用预设的点击率预估模型,对生成的多种不同风格的备用广告文案进行排序展示的步骤,包括:
采用预设的点击率预估模型,获取生成的多种不同风格的备用广告文案对应的预估点击率;
根据所述多种不同风格的备用广告文案对应的预估点击率,对所述多种不同风格的备用广告文案进行排序展示。
可选地,所述广告文案处理方法还包括:
根据所述备用广告文案,生成对应的多个伪文案;
对所述伪文案中的关键词,采用Lasso机器学习模型学习每个关键词对预估点击率的权重;
输出至少一个关键词以及对应的权重,以辅助用户对所述备用广告文案进行编辑。
可选地,所述根据所述备用广告文案,生成对应的多个伪文案的步骤包括:
基于中文分词系统,将所述备用广告文案分割成多个中文词汇;
分别将所述多个中文词汇按照不同数量进行随机丢弃,生成所述多个伪文案。
可选地,所述至少一个关键词为部分关键词,所述输出的至少一个关键词对应的权重大于未输出关键词对应的权重。
此外,为实现上述目的,本申请还提供一种广告文案处理装置,所述广告文案处理装置包括:存储器、处理器及存储在所述存储器上并可在所述处理器上运行的广告文案处理程序,所述广告文案处理程序被所述处理器执行时实现如上所述的广告文案处理方法的步骤。
此外,为实现上述目的,本申请还提供一种广告文案处理设备,所述广告文案处理设备包括:存储器、处理器及存储在所述存储器上并可在所述处理器上运行的广告文案处理程序,所述广告文案处理程序被所述处理器执行时实现如上所述的广告文案处理方法的步骤。
此外,为实现上述目的,本申请还提供一种计算机可读存储介质,所述计算机可读存储介质上存储有广告文案处理程序,所述广告文案处理程序被处理器执行时实现如上所述的广告文案处理方法的步骤。
本申请实施例提出的一种广告文案处理方法、装置、设备及计算机可读存储介质,通过获取源风格广告文案,基于预设的多种不同风格的变换模型,将源风格广告文案进行风格变换,生成对应的多种不同风格的备用广告文案,并采用预设的点击率预估模型,对生成的多种不同风格的备用广告文案进行排序展示,以供基于排序展示的备用广告文案创作获得用户需要的广告文案,从而提高创作出来的广告文案点击率。
附图说明
图1是本申请实施例方案涉及的硬件运行环境的终端\装置结构示意图;
图2为本申请广告文案处理方法第一实施例的流程示意图;
图3为本申请广告文案处理方法第二实施例的流程示意图;
图4为本申请广告文案处理方法中编码器-解码器工作场景示意图;
图5为本申请广告文案处理方法第三实施例的流程示意图;
图6为本申请广告文案处理方法中点击率驱动的文案生成与分析方法的整体框架示意图。
本申请目的实现、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
应当理解,此处所描述的具体实施例仅仅用以解释本申请,并不用于限定本申请。
如图1所示,图1是本申请实施例方案涉及的硬件运行环境的终端结构示意图。
本申请实施例终端为广告文案处理设备。
如图1所示,该终端可以包括:处理器1001,例如CPU,网络接口1004,用户接口1003,存储器1005,通信总线1002。其中,通信总线1002用于实现这些组件之间的连接通信。用户接口1003可以包括显示屏(Display)、输入单元比如键盘(Keyboard),可选用户接口1003还可以包括标准的有线接口、无线接口。网络接口1004可选的可以包括标准的有线接口、无线接口(如WI-FI接口)。存储器1005可以是高速RAM存储器,也可以是稳定的存储器(non-volatile memory),例如磁盘存储器。存储器1005可选的还可以是独立于前述处理器1001的存储装置。
可选地,终端还可以包括摄像头、RF(Radio Frequency,射频)电路,传感器、音频电路、WiFi模块等等。其中,传感器比如光传感器、运动传感器以及其他传感器。具体地,光传感器可包括环境光传感器及接近传感器,其中,环境光传感器可根据环境光线的明暗来调节显示屏的亮度,接近传感器可在终端设备移动到耳边时,关闭显示屏和/或背光。当然,终端设备还可配置陀螺仪、气压计、湿度计、温度计、红外线传感器等其他传感器,在此不再赘述。
本领域技术人员可以理解,图1中示出的终端结构并不构成对终端的限定,可以包括比图示更多或更少的部件,或者组合某些部件,或者不同的部件布置。
如图1所示,作为一种计算机存储介质的存储器1005中可以包括操作系统、网络通信模块、用户接口模块以及广告文案处理程序。
在图1所示的终端中,网络接口1004主要用于连接后台服务器,与后台服务器进行数据通信;用户接口1003主要用于连接客户端(用户端),与客户端进行数据通信;而处理器1001可以用于调用存储器1005中存储的广告文案处理程序,并执行以下操作:
获取源风格广告文案;
基于预设的多种不同风格的变换模型,将所述源风格广告文案进行风格变换,生成对应的多种不同风格的备用广告文案;
采用预设的点击率预估模型,对生成的多种不同风格的备用广告文案进行排序展示。
本申请提供一种广告文案处理方法,在广告文案处理方法第一实施例中,参照图2,广告文案处理方法包括以下步骤:
步骤S10,获取源风格广告文案;
步骤S20,基于预设的多种不同风格的变换模型,将所述源风格广告文案进行风格变换,生成对应的多种不同风格的备用广告文案;
步骤S30,采用预设的点击率预估模型,对生成的多种不同风格的备用广告文案进行排序展示;
广告文案可以是广告作品的全部,包含广告作品的语音文字部分和图画部分等。为了获得受用户欢迎,用户所需的广告文案,本实施例中,首先获取相应的源风格广告文案(也即查询文案)。获取源风格广告文案的方式可以是用户基于自身的喜好需求,在显示终端中预存的各个广告文案信息中选择一个广告文案作为源风格广告文案,也可以是将用户自行在显示终端中创作的广告文案作为源风格广告文案,还可以是用户在互联网上基于自身的喜好需求选择出一篇广告文案作为源风格广告文案,在此不做限制。可选地,当获取到该源风格广告文案后,获取此源风格广告文案的源风格。其中,源风格包括但不限于广告文案的行文风格、排版布局等。
本实施例中,基于各种不同广告文案对应的不同文案风格,训练多种不同风格的变换模型Ti。假设训练中有N种风格的语料存在,即S1......SN,所有训练语料均属于源风格,随机取出一个风格对应的语料Si,在训练阶段,为这一风格Si训练变换模型Ti,变换模型Ti可以将源风格转换为Si风格。此变换模型Ti的首要要求是生成的广告文案,其语言通顺并可以理解。因此,为风格Si和源风格S训练端到端序列到序列的变换模型,针对风格Si的变换模型Ti仅利用风格Si的语料进行训练。需要强调的是,针对Si风格和源风格S的变换模型Ti是独立的。
对于所获取源风格广告文案,基于该源风格广告文案对应的源风格,以及多种不同风格Si的变换模型Ti,将源风格广告文案进行风格变换,生成该源风格广告文案对应的多种不同风格Si的备用广告文案,备用广告文案保持了源风格广告文案的语义。
并且,为了向用户展示海量的广告文案,本实施例中,预先设置广告文案对应的CTR点击率预估模型,当生成源风格广告文案对应的多种不同风格Si的备用广告文案后,基于预设的CTR点击率预估模型,对生成的多种不同风格Si的备用广告文案进行排序展示。
可选地,所述步骤S30包括:
步骤a,采用预设的点击率预估模型,获取生成的多种不同风格的备用广告文案对应的预估点击率;
步骤b,根据所述多种不同风格的备用广告文案对应的预估点击率,对所述多种不同风格的备用广告文案进行排序展示。
可选地,在生成多种不同风格Si的备用广告文案后,基于预设的CTR点击率预估模型,分别将各备用广告文案输入到该点击率预估模型中,获取各备用广告文案对应的预估点击率。之后,再根据各备用广告文案对应的预估点击率,对该多种不同风格Si的备用广告文案进行排序展示。可选地,按照预估点击率从高到低的顺序,对各备用广告文案进行排序展示。
用户可以对展示的各备用广告文案进行编辑,从而生成获得用户所需的目标广告文案。
在本实施例中,通过获取源风格广告文案,基于预设的多种不同风格的变换模型,将源风格广告文案进行风格变换,生成对应的多种不同风格的备用广告文案,并采用预设的点击率预估模型,对生成的多种不同风格的备用广告文案进行排序展示,以供基于排序展示的备用广告文案创作获得用户需要的广告文案,从而提高创作出来的广告文案点击率。
进一步地,在本申请第一实施例的基础上,提出了本申请广告文案处理方法的第二实施例,本实施例中,如图3所示,所述步骤S10之前,还包括:
步骤S40,建立多种不同风格的深度神经网络模型;
所述步骤S20包括:
步骤S21,基于所述多种不同风格的深度神经网络模型,生成所述源风格广告文案对应的多种不同风格的备用广告文案。
本实施例中,利用深度神经网络模型学习如何在保持源风格广告文案的语义的前提下,生成不同风格的备用广告文案。具体地,预先根据源风格广告文案的源风格训练生成端到端、序列到序列的各不同风格的深度神经网络模型,即每种风格均有一个对应的深度神经网络模型,并且每个深度神经网络模型都是相互独立、互不关联的。
当建立好各种不同风格的深度神经网络模型后,基于各种不同风格的深度神经网络模型,在保持源风格广告文案的语义的前提下,生成各种不同风格的备用广告文案。
可选地,所述步骤S21包括:
步骤c,在所述多种不同风格的深度神经网络模型中,采用源风格对应的编码器将所述源风格广告文案编码为一组向量表示;
步骤d,采用多种不同风格的解码器将所述向量表示进行解码,生成所述源风格广告文案对应的多种不同风格的备用广告文案。
可选地,基于深度神经网络模型生成各不同风格的备用广告文案,是基于编码器和解码器的结构来进行的,其中,编码器仅负责广告文案的语义信息,解码器依赖于语义信息并仅负责风格信息。
基于编码器和解码器的结构,首先,采用源风格对应的编码器将源风格广告文案编码为一组向量表示,可选地,编码器利用递归神经网将源风格广告文案转变为一组向量表示。其中,该向量表示包含源风格广告文案的语义。然后,根据此向量表示,采用多种不同风格的解码器将该向量表示进行解码,生成源风格广告文案对应的多种不同风格的备用广告文案。
可选地,编码器和解码器基于重建损失函数进行训练。例如,如图4所示,为成功实现S风格到Si风格的变化,我们假设编码器所编码得到的向量表示仅与广告文案语义相关,而于风格无关。相应地,解码器利用语义进行文本重建,我们假设解码器仅与风格相关,而与语义无关。为了使假设成立,我们限制不同风格的编码器所得的向量表示分布相似。若源风格与目标风格本身内容相近,仅是风格不同,此限制可使得编码器仅建模内容语义,解码器仅建模风格。最终,我们为每一种风格建立编码器和解码器的结构,完成深度神经网络模型训练。
在本实施例中,基于多种不同风格的深度神经网络模型,生成源风格广告文案对应的多种不同风格的备用广告文案,深度神经网络模型的训练仅需要不同风格的语料库,而并不需要强对应关系,因此,深度神经网络模型的训练易于实现,进而使得备用广告文案的生成更加可靠。
进一步地,在本申请第一至第二实施例任意一个的基础上,提出了本申请广告文案处理方法的第三实施例,本实施例中,如图5所示,所述广告文案处理方法还包括:
步骤S50,根据所述备用广告文案,生成对应的多个伪文案;
步骤S60,对所述伪文案中的关键词,采用Lasso机器学习模型学习每个关键词对预估点击率的权重;
步骤S70,输出至少一个关键词以及对应的权重,以辅助用户对所述备用广告文案进行编辑。
本实施例中,为辅助用户进行广告文案编辑优化,对广告文案的关键词进行分析,分析所得到的关键词意味着其对广告文案的点击率有重大正面或负面的影响。具体地,先根据生成的备用广告文案,生成对应的多个伪文案。可选地,所述步骤S50包括:
步骤e,基于中文分词系统,将所述备用广告文案分割成多个中文词汇;
步骤f,分别将所述多个中文词汇按照不同数量进行随机丢弃,生成所述多个伪文案。
在一实施方式中,调用中文分词系统,基于该中文分词系统,将备用广告文案分割成多个中文词汇(包括短语),然后,通过随机丢弃不同数量的中文词汇,分别生成对应的多个不同的伪文案。可选地,使用词袋特征表示伪文案,具体地,若伪文案包含某个中文词汇,则词袋特征中与之对应的位置为1,否则为0,使用伪文案的词袋特征作为数据来预测预估点击率。
具体地,使用Lasso机器学习模型进行学习,对伪文案中出现的关键词,采用Lasso机器学习模型学习每个关键词对预估点击率的权重。最终,根据用户需求,输出至少一个关键词及其对应的权重。根据关键词及其对应的权重,辅助用户对备用广告文案进行编辑。
下面用一具体例子解释说明:
例如,如图6所示,先通过源风格的编码器对源风格广告文案(查询文案)进行编码,并根据不同风格语料训练多个深度神经网络模型,在各个深度神经网络模型中均用解码器进行解码,如Si风格解码器,生成查询文案对应的不同风格的海量备用广告文案。然后再采用CTR点击率预估模型对各个备用广告文案进行排序展示,如文案方案1、文案方案2、文案方案3等。并且根据生成的备用广告文案,生成对应的伪文案,采用Lasso机器学习模型进行关键词分析,获取伪文案中关键词对预估点击率的权重,并输出关键词及其对应的权重,以辅助用户对备用广告文案进行编辑。
关键词对应的权重越大,则关键词对预估点击率CTR的影响越大,权重为正的关键词可以提升预估点击率CTR,而权重为负的关键词会降低预估点击率CTR。可选地,根据各关键词对应的权重,保留其中的部分(至少一个)关键词,该部分关键词对应的权重大于未输出关键词对应的权重。例如,保留其中权重最大的k个关键词,这些关键词对预估点击率CTR影响最大。并输出权重最大的k个关键词及其权重。用户在编辑文案过程中,该关键词及其权重实时反馈给用户,辅助用户对备用广告文案进行编辑。
在本实施例中,通过生成对应的多个伪文案,并对伪文案中的关键词,采用Lasso机器学习模型学习每个关键词对预估点击率的权重,输出至少一个关键词以及对应的权重,以辅助用户对备用广告文案进行编辑,从而进一步提高创作出来的广告文案点击率。
本申请还提供一种广告文案处理装置,所述广告文案处理装置包括:存储器、处理器及存储在所述存储器上并可在所述处理器上运行的广告文案处理程序,所述广告文案处理程序被所述处理器执行时实现上述中的广告文案处理方法各实施例的步骤。
本申请广告文案处理装置具体实施方式与上述广告文案处理方法各实施例基本相同,在此不再赘述。
本申请还提供了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一个或者一个以上程序,所述一个或者一个以上程序还可被一个或者一个以上的处理器执行以用于实现上述广告文案处理方法各实施例的步骤。
本申请计算机可读存储介质具体实施方式与上述广告文案处理方法各实施例基本相同,在此不再赘述。
需要说明的是,在本文中,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、物品或者系统不仅包括那些要素,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、物品或者系统所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括该要素的过程、方法、物品或者系统中还存在另外的相同要素。
上述本申请实施例序号仅仅为了描述,不代表实施例的优劣。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现,当然也可以通过硬件,但很多情况下前者是更佳的实施方式。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来,该计算机软件产品存储在如上所述的一个存储介质(如ROM/RAM、磁碟、光盘)中,包括若干指令用以使得一台终端设备(可以是手机,计算机,服务器,空调器,或者网络设备等)执行本申请各个实施例所述的方法。
以上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,凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换,或直接或间接运用在其他相关的技术领域,均同理包括在本申请的专利保护范围内。

Claims (19)

  1. 一种广告文案处理方法,其中,所述广告文案处理方法包括以下步骤:
    获取源风格广告文案;
    基于预设的多种不同风格的变换模型,将所述源风格广告文案进行风格变换,生成对应的多种不同风格的备用广告文案;
    采用预设的点击率预估模型,对生成的多种不同风格的备用广告文案进行排序展示。
  2. 如权利要求1所述的广告文案处理方法,其中,所述广告文案处理方法还包括:
    根据所述备用广告文案,生成对应的多个伪文案;
    对所述伪文案中的关键词,采用Lasso机器学习模型学习每个关键词对预估点击率的权重;
    输出至少一个关键词以及对应的权重,以辅助用户对所述备用广告文案进行编辑。
  3. 如权利要求1所述的广告文案处理方法,其中,所述获取源风格广告文案的步骤之前,还包括:
    建立多种不同风格的深度神经网络模型;
    所述基于预设的多种不同风格的变换模型,将所述源风格广告文案进行风格变换,生成对应的多种不同风格的备用广告文案的步骤包括:
    基于所述多种不同风格的深度神经网络模型,生成所述源风格广告文案对应的多种不同风格的备用广告文案。
  4. 如权利要求3所述的广告文案处理方法,其中,所述广告文案处理方法还包括:
    根据所述备用广告文案,生成对应的多个伪文案;
    对所述伪文案中的关键词,采用Lasso机器学习模型学习每个关键词对预估点击率的权重;
    输出至少一个关键词以及对应的权重,以辅助用户对所述备用广告文案进行编辑。
  5. 如权利要求3所述的广告文案处理方法,其中,所述基于所述多种不同风格的深度神经网络模型,生成所述源风格广告文案对应的多种不同风格的备用广告文案的步骤包括:
    在所述多种不同风格的深度神经网络模型中,采用源风格对应的编码器将所述源风格广告文案编码为一组向量表示;
    采用多种不同风格的解码器将所述向量表示进行解码,生成所述源风格广告文案对应的多种不同风格的备用广告文案。
  6. 如权利要求5所述的广告文案处理方法,其中,所述广告文案处理方法还包括:
    根据所述备用广告文案,生成对应的多个伪文案;
    对所述伪文案中的关键词,采用Lasso机器学习模型学习每个关键词对预估点击率的权重;
    输出至少一个关键词以及对应的权重,以辅助用户对所述备用广告文案进行编辑。
  7. 如权利要求1所述的广告文案处理方法,其中,所述采用预设的点击率预估模型,对生成的多种不同风格的备用广告文案进行排序展示的步骤,包括:
    采用预设的点击率预估模型,获取生成的多种不同风格的备用广告文案对应的预估点击率;
    根据所述多种不同风格的备用广告文案对应的预估点击率,对所述多种不同风格的备用广告文案进行排序展示。
  8. 如权利要求7所述的广告文案处理方法,其中,所述广告文案处理方法还包括:
    根据所述备用广告文案,生成对应的多个伪文案;
    对所述伪文案中的关键词,采用Lasso机器学习模型学习每个关键词对预估点击率的权重;
    输出至少一个关键词以及对应的权重,以辅助用户对所述备用广告文案进行编辑。
  9. 如权利要求8所述的广告文案处理方法,其中,所述根据所述备用广告文案,生成对应的多个伪文案的步骤包括:
    基于中文分词系统,将所述备用广告文案分割成多个中文词汇;
    分别将所述多个中文词汇按照不同数量进行随机丢弃,生成所述多个伪文案。
  10. 如权利要求8所述的广告文案处理方法,其中,所述至少一个关键词为部分关键词,所述输出的至少一个关键词对应的权重大于未输出关键词对应的权重。
  11. 一种广告文案处理装置,其中,所述广告文案处理装置包括:存储器、处理器及存储在所述存储器上并可在所述处理器上运行的广告文案处理程序,所述广告文案处理程序被所述处理器执行时实现以下步骤:
    获取源风格广告文案;
    基于预设的多种不同风格的变换模型,将所述源风格广告文案进行风格变换,生成对应的多种不同风格的备用广告文案;
    采用预设的点击率预估模型,对生成的多种不同风格的备用广告文案进行排序展示。
  12. 如权利要求11所述的广告文案处理装置,其中,所述广告文案处理程序被所述处理器执行时实现所述获取源风格广告文案的步骤之前,还包括:
    建立多种不同风格的深度神经网络模型;
    所述基于预设的多种不同风格的变换模型,将所述源风格广告文案进行风格变换,生成对应的多种不同风格的备用广告文案的步骤包括:
    基于所述多种不同风格的深度神经网络模型,生成所述源风格广告文案对应的多种不同风格的备用广告文案。
  13. 如权利要求12所述的广告文案处理装置,其中,所述广告文案处理程序被所述处理器执行时实现所述基于所述多种不同风格的深度神经网络模型,生成所述源风格广告文案对应的多种不同风格的备用广告文案的步骤包括:
    在所述多种不同风格的深度神经网络模型中,采用源风格对应的编码器将所述源风格广告文案编码为一组向量表示;
    采用多种不同风格的解码器将所述向量表示进行解码,生成所述源风格广告文案对应的多种不同风格的备用广告文案。
  14. 如权利要求11所述的广告文案处理装置,其中,所述广告文案处理程序被所述处理器执行时实现所述基于所述采用预设的点击率预估模型,对生成的多种不同风格的备用广告文案进行排序展示的步骤,包括:
    采用预设的点击率预估模型,获取生成的多种不同风格的备用广告文案对应的预估点击率;
    根据所述多种不同风格的备用广告文案对应的预估点击率,对所述多种不同风格的备用广告文案进行排序展示。
  15. 一种广告文案处理设备,其中,所述广告文案处理设备包括:存储器、处理器及存储在所述存储器上并可在所述处理器上运行的广告文案处理程序,所述广告文案处理程序被所述处理器执行时实现以下步骤:
    获取源风格广告文案;
    基于预设的多种不同风格的变换模型,将所述源风格广告文案进行风格变换,生成对应的多种不同风格的备用广告文案;
    采用预设的点击率预估模型,对生成的多种不同风格的备用广告文案进行排序展示。
  16. 如权利要求15所述的广告文案处理设备,其中,所述广告文案处理程序被所述处理器执行时实现所述获取源风格广告文案的步骤之前,还包括:
    建立多种不同风格的深度神经网络模型;
    所述基于预设的多种不同风格的变换模型,将所述源风格广告文案进行风格变换,生成对应的多种不同风格的备用广告文案的步骤包括:
    基于所述多种不同风格的深度神经网络模型,生成所述源风格广告文案对应的多种不同风格的备用广告文案。
  17. 如权利要求16所述的广告文案处理设备,其中,所述广告文案处理程序被所述处理器执行时实现所述基于所述多种不同风格的深度神经网络模型,生成所述源风格广告文案对应的多种不同风格的备用广告文案的步骤包括:
    在所述多种不同风格的深度神经网络模型中,采用源风格对应的编码器将所述源风格广告文案编码为一组向量表示;
    采用多种不同风格的解码器将所述向量表示进行解码,生成所述源风格广告文案对应的多种不同风格的备用广告文案。
  18. 如权利要求15所述的广告文案处理设备,其中,所述广告文案处理程序被所述处理器执行时实现所述基于所述采用预设的点击率预估模型,对生成的多种不同风格的备用广告文案进行排序展示的步骤,包括:
    采用预设的点击率预估模型,获取生成的多种不同风格的备用广告文案对应的预估点击率;
    根据所述多种不同风格的备用广告文案对应的预估点击率,对所述多种不同风格的备用广告文案进行排序展示。
  19. 一种计算机可读存储介质,其中,所述计算机可读存储介质上存储有广告文案处理程序,所述广告文案处理程序被处理器执行时实现如权利要求1所述的广告文案处理方法的步骤。
PCT/CN2019/080300 2018-11-28 2019-03-29 广告文案处理方法、装置、设备及计算机可读存储介质 WO2020107761A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811440929.0A CN109583952B (zh) 2018-11-28 2018-11-28 广告文案处理方法、装置、设备及计算机可读存储介质
CN201811440929.0 2018-11-28

Publications (1)

Publication Number Publication Date
WO2020107761A1 true WO2020107761A1 (zh) 2020-06-04

Family

ID=65925495

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/080300 WO2020107761A1 (zh) 2018-11-28 2019-03-29 广告文案处理方法、装置、设备及计算机可读存储介质

Country Status (2)

Country Link
CN (1) CN109583952B (zh)
WO (1) WO2020107761A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114501105A (zh) * 2022-01-29 2022-05-13 腾讯科技(深圳)有限公司 视频内容的生成方法、装置、设备、存储介质及程序产品

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110060029A (zh) * 2019-04-18 2019-07-26 浩鲸云计算科技股份有限公司 一种ai赋能创意设计方法
CN110458236A (zh) * 2019-08-14 2019-11-15 有米科技股份有限公司 一种广告文案风格识别方法和系统
CN110688834B (zh) * 2019-08-22 2023-10-31 创新先进技术有限公司 基于深度学习模型进行智能文稿风格改写的方法和设备
CN110599393B (zh) * 2019-09-06 2023-07-25 深圳前海微众银行股份有限公司 图片风格转换方法、装置、设备及计算机可读存储介质
CN112837081A (zh) * 2019-11-25 2021-05-25 上海哔哩哔哩科技有限公司 广告图片自动生成方法、装置及存储介质
CN111581926B (zh) * 2020-05-15 2023-09-01 抖音视界有限公司 文案生成方法、装置、设备和计算机可读存储介质
US11694018B2 (en) * 2021-01-29 2023-07-04 Salesforce, Inc. Machine-learning based generation of text style variations for digital content items
JP7245478B1 (ja) 2021-09-30 2023-03-24 株式会社人々 コンテンツ生成方法、コンテンツ提供方法、及びそのプログラム
CN117391776A (zh) * 2023-10-17 2024-01-12 北京维科维众科技发展有限公司 广告投放方法及装置、电子设备、存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1378672A (zh) * 1999-03-26 2002-11-06 皇家菲利浦电子有限公司 通过进化算法发展广告
US20110099059A1 (en) * 2009-10-27 2011-04-28 Yahoo! Inc. Index-based technique friendly ctr prediction and advertisement selection
CN105160545A (zh) * 2015-08-07 2015-12-16 百度在线网络技术(北京)有限公司 投放信息样式确定方法及装置
CN106778928A (zh) * 2016-12-21 2017-05-31 广州华多网络科技有限公司 图像处理方法及装置
CN106909931A (zh) * 2015-12-23 2017-06-30 阿里巴巴集团控股有限公司 一种用于机器学习模型的特征生成方法、装置和电子设备

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2002333852A (ja) * 2001-05-08 2002-11-22 Avec Kenkyusho:Kk 広告表示システム、及びその方法
CN104506964A (zh) * 2014-12-31 2015-04-08 乐视网信息技术(北京)股份有限公司 广告素材的生成方法及装置、电子设备
CN105912686A (zh) * 2016-04-18 2016-08-31 上海珍岛信息技术有限公司 一种基于机器学习的搜索引擎营销竞价方法及系统
CN106096995B (zh) * 2016-05-31 2021-09-07 腾讯科技(深圳)有限公司 广告创意处理方法及广告创意处理装置
WO2018175331A1 (en) * 2017-03-24 2018-09-27 Motivemetrics Inc. Automated system and method for creating machine-generated advertisements
CN108563622B (zh) * 2018-05-04 2020-04-07 清华大学 一种具有风格多样性的绝句生成方法及装置

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1378672A (zh) * 1999-03-26 2002-11-06 皇家菲利浦电子有限公司 通过进化算法发展广告
US20110099059A1 (en) * 2009-10-27 2011-04-28 Yahoo! Inc. Index-based technique friendly ctr prediction and advertisement selection
CN105160545A (zh) * 2015-08-07 2015-12-16 百度在线网络技术(北京)有限公司 投放信息样式确定方法及装置
CN106909931A (zh) * 2015-12-23 2017-06-30 阿里巴巴集团控股有限公司 一种用于机器学习模型的特征生成方法、装置和电子设备
CN106778928A (zh) * 2016-12-21 2017-05-31 广州华多网络科技有限公司 图像处理方法及装置

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114501105A (zh) * 2022-01-29 2022-05-13 腾讯科技(深圳)有限公司 视频内容的生成方法、装置、设备、存储介质及程序产品
CN114501105B (zh) * 2022-01-29 2023-06-23 腾讯科技(深圳)有限公司 视频内容的生成方法、装置、设备及存储介质

Also Published As

Publication number Publication date
CN109583952A (zh) 2019-04-05
CN109583952B (zh) 2022-03-22

Similar Documents

Publication Publication Date Title
WO2020107761A1 (zh) 广告文案处理方法、装置、设备及计算机可读存储介质
WO2020107765A1 (zh) 语句分析处理方法、装置、设备以及计算机可读存储介质
WO2020119069A1 (zh) 基于自编码神经网络的文本生成方法、装置、终端及介质
WO2012161359A1 (ko) 사용자 인터페이스 방법 및 장치
WO2016112558A1 (zh) 智能交互系统中的问题匹配方法和系统
WO2020034526A1 (zh) 保险录音的质检方法、装置、设备和计算机存储介质
EP3721361A1 (en) Method for learning cross domain personalized vocabulary and electronic device thereof
WO2018151464A1 (ko) 음성 인식을 이용한 코딩시스템 및 코딩방법
EP3031213A1 (en) Apparatus, server, and method for providing conversation topic
WO2017138777A1 (en) Method and electronic device for performing voice based actions
WO2019093744A1 (en) Display apparatus and control method thereof
WO2020107762A1 (zh) Ctr预估方法、装置及计算机可读存储介质
WO2020253115A1 (zh) 基于语音识别的产品推荐方法、装置、设备和存储介质
WO2021141419A1 (en) Method and apparatus for generating customized content based on user intent
WO2019209040A1 (en) Multi-models that understand natural language phrases
WO2020055071A1 (en) System and method for dynamic trend clustering
WO2020191934A1 (zh) 终端喇叭的控制方法、设备及计算机可读存储介质
WO2019164119A1 (ko) 전자 장치 및 그 제어 방법
WO2020107591A1 (zh) 重复投保限制方法、装置、设备及可读存储介质
WO2020233089A1 (zh) 测试用例生成方法、装置、终端及计算机可读存储介质
WO2020130447A1 (ko) 페르소나에 기반하여 문장을 제공하는 방법 및 이를 지원하는 전자 장치
WO2020062640A1 (zh) 终端应用动态文案的语言切换方法、服务器及存储介质
WO2020184935A1 (en) Electronic apparatus and method for controlling thereof
WO2022145946A1 (ko) 학습 영상 및 예문의 인공지능 추천기반 언어학습 시스템 및 방법
WO2020199599A1 (zh) 工作队列的信息展示方法、装置、计算机设备和存储介质

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: 19890630

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: 19890630

Country of ref document: EP

Kind code of ref document: A1