Method and device for media recommendation in a call scenario, electronic device and storage medium

Abstract

The application relates to a media recommendation method and device in a call scenario, electronic equipment and a storage medium. The media recommendation method in the call scenario comprises preprocessing and accurate matching, and specifically comprises the following steps: S11, preprocessing of user data, which comprises setting of user static data, calculation of an industry label score of the user by using an NLP (natural language processing) technology, and setting of user dynamic data; S12, preprocessing of a pre-launched media, extraction of advertisement content, and clustering analysis of media content to be launched by using unsupervised machine learning; S21, accurate matching of the pre-launched media, output of an advertisement industry label weight result by using NLP machine learning; S22, accurate matching on the user side, output of a user industry label weight result; and when a condition based on dynamic data is triggered, a media advertisement triggered based on dynamic data is played. According to the media recommendation method in the call scenario, the advertisement media launching precision and efficiency can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of 5G wireless communication, big data AI and video early media negotiation, in particular to a method and device for recommending media in a call scene based on video playing after user data intelligent analysis when a user receives a phone ring or call, an electronic device and a storage medium. BACKGROUND

[0002] With the development of the advertising industry, traditional advertising such as television and billboards has gradually lost its appeal due to a decrease in the number of audience, the lack of effective effect monitoring, large-scale deployment, and the lack of follow-up means. Under the Internet situation, media advertising forms, such as the text and image media deployment platform (e.g., WeChat, Today's Headlines, and Weibo) and the audio and video media deployment platform (e.g., Douyin, Kuaishou, and Xigua Video), are favored by more and more advertisers due to their convenience, the ability to click and jump to their own platform for private conversion, and the availability of quantifiable analysis indicators.

[0003] In the face of a large number of media channels, advertisers need to choose the appropriate promotion channel according to their budget and other requirements, which is like finding a needle in a haystack. The existing methods include: method one, full deployment under sufficient budget; method two, manual screening and selection according to categories, prices, and reading volumes; and method three, repeated use of familiar media channels. The existing methods have the following problems: high deployment cost, malicious brushing software, and automatic clicking software causing a lot of invalid traffic, which still needs to be paid by advertisers, resulting in additional advertising costs; a large amount of fake data, through technical means such as embedding advertising pages in user-required pages, causing virtual high exposure, affecting subsequent operation analysis of advertisers, and reducing enterprise analysis precision.

[0004] The prior art has the following related patents: Chinese patent application CN112752310A, name: a telephone video business card system based on media negotiation. The system can realize the service functions of calling trigger and called video media playback, and in the case of insufficient terminal coverage, the existing call service is not affected, and the transmission and display of various calling number associated information can be realized. Chinese patent application CN111970408A, name: video ringtone playing method and system, electronic device and storage medium, relates to a video ringtone playing method, which includes sending a telephone call request to a video ringtone server, wherein the telephone call request carries a called terminal identifier and a playing configuration parameter of an external terminal establishing a communication connection with a calling terminal; receiving first video ringtone data returned by the video ringtone server according to the called terminal identifier and the playing configuration parameter; and sending the first video ringtone data to the external terminal for playing the first video ringtone data. Applied in the process of telephone call, the purpose of playing video ringtone on external terminal is achieved. Chinese patent application CN114266595A, name: a digital media elevator advertisement playing recommendation system and method based on the Internet, the passenger image in the elevator is collected through the image collection module, whether the passenger image in the elevator exists a passenger is identified, and the number of people in the passenger image is obtained, the next advertisement of the elevator advertising machine is recommended according to the number of people in the passenger image. Chinese patent application CN114254202A, name: media intelligent recommendation system and method based on big data and storage medium, the scheme is based on constructing a tag subsystem, a recommendation subsystem and a storage subsystem, user data and media library data are collected, and tags are extracted from the user data and media library data, a similarity algorithm model is used to calculate the similarity of the extracted results, recommended data is generated according to the calculation results, a sorting algorithm is used to sort the recommended data, and the sorting results are output, the pertinence of the recommendation is increased.

[0005] The above patents still have the following problems: the advertising media is not accurate enough, network users cannot locate individuals, the accuracy of user information cannot be verified, which increases the uncertainty of subsequent operation analysis; the user group coverage is not enough; repeated use of media with cooperation experience, small fan coverage, low influence of communication; through the Internet channel, the user group has limitations. Therefore, when using a mobile terminal, it is urgent to develop a more targeted method of displaying a central platform based on user big data behavior intelligent analysis and sending ring / call video media on a mobile terminal. SUMMARY

[0006] The technical problem to be solved by the present application is how to accurately place media on the user's mobile terminal.

[0007] To solve the above technical problems, according to one aspect of the present application, a method for media recommendation in a call scenario is provided, which comprises preprocessing and accurate matching, wherein the preprocessing is a setting and calculation performed in advance before accurate matching, and the preprocessing comprises the following steps: S11, preprocessing of user data, the user data comprising static data, industry label score and dynamic data; the static data comprising user age, gender, home province, place of use, mobile phone model, consumption level, credit level data which can be obtained by an operator; the dynamic data comprising network search words performed on the same day, industry attribute of the number currently called, current call time, real-time location, app type newly installed on the same day; the preprocessing of user data comprising setting of user static data, calculation of user industry label score by using NLP (natural language processing) technology, and setting of user dynamic data; S12, preprocessing of pre-launched media, advertisement content extraction, and clustering analysis of the media content to be launched by using unsupervised machine learning; wherein the accurate matching comprises the following steps: S21, accurate matching of pre-launched media, output of advertisement industry label weight result by using NLP machine learning; generation of a coarse launching resource pool set [A] based on discrete user static data for the pre-launched media 1 content; calculation of the coincidence value of the industry label score of each user in the resource pool and the industry label score of the pre-launched media 1 content, and generation of a launching resource pool sequence [B] in descending order, B=[User i med1:x i ], wherein User i med1 represents media 1 and each user, and x i represents the coincidence value; the top number of users are launched according to the media launching quantity required by the advertiser, and the media 1 data is synchronized to the media library played by the user in the next call; S22, accurate matching on the user side, output of user industry label weight result, first, the media library played by the user 1 in the next call is arranged in descending order according to the coincidence value, to obtain a sequence [C], C=[Med i user1:x i ], wherein Med i user1 represents user 1 and each media, and x i represents the coincidence value; when the condition based on dynamic data is triggered, the media advertisement triggered based on dynamic data is played.

[0008] According to the embodiment of the present application, in step S11, the calculation of the industry label score of the user can be periodically processed by using NLP (natural language processing) technology, comprising the following steps: S111, the web pages recently browsed by the number are obtained by using a crawler or in cooperation with an Internet company; the obtained content is segmented; the domain (Industry) to which the word frequency belongs is analyzed by dictionary query; and the industry label score of each account is calculated Tabi Tab i =[Ind i :x i S112. Obtain the media content recently viewed by the account using web crawling or in cooperation with internet companies, perform cluster analysis using unsupervised machine learning, and calculate the industry tag score for each account as described above. j Tab j =[Ind j :x j S113. Calculate the final industry tag score for each account by weighted average based on sample size. (Tab) p Tab p =[Ind p :x p ] Where Ind represents the industry sector, x represents the score, and Tab represents the corresponding set.

[0009] According to an embodiment of the present invention, in step S12, during the processing of pre-deployment media, unsupervised machine learning can be used to perform cluster analysis on the media content, and the industry tag score Tab for each account is calculated. q Tab q =[Ind q :x q ].

[0010] According to an embodiment of the present invention, in step S11, the dynamic data weight can be set to P(act); in step S22, when a condition based on the dynamic data is triggered, the dynamic weight P(act) is compared with the industry weight x with the highest value. max When the dynamic weight is greater than the industry weight, media ads triggered by dynamic data will be played.

[0011] According to an embodiment of the present invention, "recent" can be the most recent 10 to 50 days, preferably the most recent 20 to 40 days, and more preferably the most recent 30 days.

[0012] According to an embodiment of the present invention, step S22 may further include: if a user clicks on a media advertisement, a short message containing a short link is triggered when the user hangs up, which the user can click to inquire or place an order. The information sent after the user clicks and hangs up includes, but is not limited to, ordinary text messages, messages containing H5 short links, messages containing WeChat mini-programs, smart messages identified and upgraded by the terminal, multimedia messages, and native video displays on the terminal.

[0013] Furthermore, step S22 may also include: using the user's click behavior as content for subsequent machine learning.

[0014] According to a second aspect of the present application, a device for media recommendation in a call scenario is provided, comprising: a preprocessing module and an accurate matching module, wherein the preprocessing module is used for preprocessing of user data and pre-launched media, and the accurate matching module is used for accurate matching of the pre-launched media and the user side, wherein the preprocessing is a setting and calculation performed in advance before data accurate matching, and the preprocessing includes: preprocessing of user data, the user data including: static data, industry label score, and dynamic data; the static data including user age, gender, home province, user location, mobile phone model, consumption level, credit level data that can be obtained through an operator; the dynamic data including network search words performed on the same day, industry attribute of the number currently in call, current call time, real-time location, and new app type on the same day; the preprocessing of the user data including setting of user static data, calculation of user industry label score by using NLP (natural language processing) technology, and setting of user dynamic data; preprocessing of the pre-launched media, advertisement content extraction, and clustering analysis of the media content to be launched by using unsupervised machine learning; wherein the accurate matching includes: accurate matching of the pre-launched media, output of advertisement industry label weight result by using NLP machine learning; generation of a coarse launch resource pool set [A] based on discrete user static data for the pre-launched media 1 content; calculation of a coincidence value of the industry label score of each user in the resource pool and the industry label score of the pre-launched media 1 content, and generation of a launch resource pool sequence [B] in descending order, B=[User i med1:x i ], wherein User i med1 represents media 1 and each user, and x i represents the coincidence value; the top number of users are launched according to the number of media launched by the advertiser, and the media 1 data is synchronized to the media library played by the user in the next call; accurate matching of the user side, output of the user industry label weight result, first, the media library played by the user 1 in the next call is arranged in descending order according to the coincidence value, to obtain a sequence [C], C=[Med i user1:x i ], wherein Med i user1 represents the user 1 and each media, and x i represents the coincidence value; when a condition based on dynamic data is triggered, a media advertisement based on dynamic data triggering is played. Based on the call behavior, it is not limited to the call behavior between terminals, but can be extended to the communication behavior between Internet apps, such as WeChat call, DingTalk call, etc., as long as dynamic data such as call time, call object, and call location can be obtained.

[0015] According to a third aspect of the present application, an electronic device is provided, comprising a memory, a processor, and a call-scene-based media recommendation program stored in the memory and executable on the processor, the call-scene-based media recommendation program, when executed by the processor, implements the steps of the call-scene-based media recommendation method.

[0016] According to a fourth aspect of the present application, a computer storage medium is provided, wherein the computer storage medium stores a call-scene-based media recommendation program, the call-scene-based media recommendation program, when executed by a processor, implements the steps of the call-scene-based media recommendation method.

[0017] Compared with the prior art, the technical solution provided by the embodiment of the present application can at least achieve the following beneficial effects:

[0018] The present application provides an implementation method for a mobile terminal to display a central platform based on intelligent analysis of user big data behavior and to send a ring / call video media. According to the technical solution of this method, AI intelligent customized video media can be played and displayed on supported networks and terminals, and subsequent interactive information resources can be provided for the user to extend the media resources. The method has the advantages of no fake traffic, accurate matching, efficient recommendation, and wide fan coverage, and solves the problems in the prior art. The advertising object is identified by a mobile phone number, which can be a real property owner or user, rather than a virtual network account.

[0019] The present application can accurately match and place media advertisements based on real users and real data analysis, greatly reducing invalid traffic caused by screen brushing and traffic brushing, and reducing the cost of placing advertisements.

[0020] The present application considers the combination of call behavior and Internet behavior through the learning of multi-user data, and accurately matches and places advertising media through comprehensive analysis of user static data, communication data, web search and browsing data, mobile app data, etc., thereby improving the accuracy and efficiency of advertising media placement.

[0021] The present application is very suitable for ring / call services and has wide applicability to receiving user groups.

[0022] The present application not only considers the play and display of video media in a call scene, but also provides subsequent interactive information resources for the user to extend the media resources, including short messages, multimedia short messages, short chains, applets, 5G messages, etc. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments will be briefly introduced as follows. Obviously, the drawings described below are only some of the embodiments of the present application, rather than all the embodiments of the present application.

[0024] Figure 1 is a flow chart illustrating a media recommendation method in a call scenario according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions of the embodiments of the present application will be described clearly and completely below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the described embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without any creative effort fall within the scope of the present application.

[0026] Unless otherwise defined, the technical terms or scientific terms used herein should be understood as their common meanings to those of ordinary skill in the art to which the present application belongs. The terms "first", "second" and similar terms used in the description and claims of the present patent application do not denote any order, quantity or importance, but are used to distinguish different components. Similarly, the terms "one" or "a" or similar terms do not denote a quantity limitation, but mean that at least one exists.

[0027] Figure 1 is a flow chart illustrating a media recommendation method in a call scenario according to an embodiment of the present application.

[0028] As shown in Figure 1 , the method of media recommendation in a call scenario includes preprocessing and accurate matching.

[0029] The preprocessing is a setting and calculation performed in advance before the accurate matching. The preprocessing includes the following steps:

[0030] S11, preprocessing of user data, the user data including: static data, industry label score and dynamic data; the static data including user age, gender, home province, user location, mobile phone model, consumption level, credit level data that can be obtained through an operator; the dynamic data including network search vocabulary performed on the same day, current call number industry attribute, current call time, real-time location, new app type installed on the same day; the preprocessing of user data including setting of user static data, calculating the industry label score of the user by using NLP (natural language processing) technology, and setting of user dynamic data.

[0031] S12, preprocessing of the pre-launched media, advertisement content extraction, unsupervised machine learning is used for clustering analysis on the media content to be launched.

[0032] The exact match includes the following steps:

[0033] S21, exact match of the pre-launched media, output the industry label weight result of the advertisement through NLP machine learning; generate a rough launch resource pool set [A] based on discrete user static data for the pre-launched media 1 content; calculate the coincidence value of the industry label score of each user in the resource pool and the industry label score of the pre-launched media 1 content, and arrange in descending order to generate a launch resource pool sequence [B], wherein B = [User i med1:x i ], wherein User i med1 represents media 1 and each user, x i represents the coincidence value; the top users according to the number of media launch required by the advertiser are launched, and the media 1 data is synchronized to the media library played by the user next time.

[0034] S22, exact match on the user side, output the industry label weight result of the user, first, arrange the media library played by user 1 next time in descending order according to the coincidence value to obtain a sequence [C], C = [Med i user1:x i ], wherein Med i user1 represents user 1 and each media, x i represents the coincidence value; when the condition based on dynamic data is triggered, the media advertisement based on dynamic data triggering is played.

[0035] The present application can accurately match the media advertisement based on real users and real data analysis, greatly reduce the invalid traffic caused by screen brushing and traffic brushing, and reduce the launch cost.

[0036] According to one or some embodiments of the present application, in step S11, the industry label score of the user is calculated, and the technology of NLP (natural language processing) is used for periodic processing. Step S11 includes the following steps:

[0037] S111, the web pages browsed by the number recently are obtained by using a crawler or cooperating with an Internet company; the content obtained by the crawler is segmented; the field (Industry) to which the word frequency belongs is analyzed through dictionary query; the industry label score Tab i of each account is calculated, Tab i = [Ind i :x i ].

[0038] S112, the number of media content recently browsed is obtained by using a crawler mode or cooperating with an Internet company, unsupervised machine learning is used for clustering analysis, and the industry label score Tab of each account is calculated in the foregoing manner j j Tab j = [Ind j : x].

[0039] S113, the final industry label score Tab of each account is obtained by weighted average according to a sample size p p Tab p = [Ind p : x].

[0040] Wherein, Ind represents an industry field, x represents a score, and Tab represents a corresponding set.

[0041] The application considers the combination of call behaviors and Internet behaviors through learning of multi-user data, and improves the advertising media delivery precision and efficiency by comprehensively analyzing and accurately matching user static data, communication data, web search and browse data, mobile app data and the like.

[0042] According to one or some embodiments of the application, in the processing of the pre-delivered media, the media content is subjected to clustering analysis by using unsupervised machine learning, and the industry label score Tab of each account is q q Tab q = [Ind q : x].

[0043] According to one or some embodiments of the application, in step S11, the dynamic data weight is set as P(act); and in step S22, when the condition based on the dynamic data is triggered, the dynamic weight P(act) is compared with the highest industry weight x max When the dynamic weight is greater than the industry weight, the media advertisement based on the dynamic data triggering is played.

[0044] According to one or some embodiments of the application, the recent period can be 10-50 days, preferably 20-40 days, and more preferably 30 days.

[0045] ​​​According to one or some embodiments of the present application, step S22 further comprises: when the user clicks the media advertisement, triggering a short message containing a short chain when hanging up, which can be clicked by the user to consult or place an order. The information sent by the user after clicking and hanging up includes but is not limited to ordinary text short message, short message containing H5 short chain, short message containing WeChat applet, intelligent short message identified and upgraded by the terminal, multimedia short message, terminal native video display, etc. Further, step S22 further comprises: taking the user's click behavior as the content of subsequent machine learning.

[0046] The present solution not only considers the play and display of video media in the call scenario, but also provides information resources that can be interacted by the user subsequently, extends the media resources, including short message, multimedia short message, short chain, applet, 5G message, etc.

[0047] According to a second aspect of the present application, a device for media recommendation in a call scenario is provided, which comprises: a preprocessing module and an accurate matching module, wherein the preprocessing module is used for preprocessing of user data and pre-launched media, and the accurate matching module is used for accurate matching of the pre-launched media and the user side, wherein the preprocessing is a setting and calculation performed in advance before data accurate matching, and the preprocessing includes: preprocessing of user data, which includes: static data, industry label score and dynamic data; the static data includes user age, gender, home province, user location, mobile phone model, consumption level, credit level data that can be obtained through the operator; the dynamic data includes network search vocabulary performed on the same day, current call number industry attribute, current call time, real-time location, new app type on the same day; the preprocessing of user data includes setting of user static data, calculation of user industry label score by using NLP (natural language processing) technology, and setting of user dynamic data; preprocessing of pre-launched media, advertisement content extraction, and clustering analysis of the media content to be launched by using unsupervised machine learning; wherein the accurate matching includes: accurate matching of the pre-launched media, outputting advertisement industry label weight result by using NLP machine learning; generating a coarse launch resource pool set [A] based on discrete user static data for the pre-launched media 1 content; calculating the coincidence value of the industry label score of each user in the resource pool and the industry label score of the pre-launched media 1 content, and arranging in descending order to generate a launch resource pool sequence [B], B=[User i med1:x i ], wherein User i med1 represents media 1 and each user, x irepresent the number of coincidences; the number of users in the top of the number of media delivery required by the advertiser is delivered, and the media 1 data is synchronized to the media library played by the user next time; accurate matching on the user side, output the user industry label weight result, first, the media library played by the user 1 next time, is arranged in descending order according to the coincidence value, and the number sequence [C] is obtained, C=[Med i user1:x i ], wherein Med i user1 represents user 1 and each media, x i represent the number of coincidences; when the condition based on dynamic data is triggered, the media advertisement based on dynamic data triggering is played. Based on the call behavior, it is not limited to the call behavior between terminals, and can be extended to the communication behavior between Internet apps, such as WeChat call, DingTalk call and the like, as long as the dynamic data such as call time, call object and call position can be obtained.

[0048] The delivery based on the call behavior of the application is very suitable for the business form of color ring and ring, and has wide applicability to the receiving user group.

[0049] When used, the user data is preprocessed, each phone number is taken as a unique identifier, and the user data is divided into static data, industry label score and dynamic data according to the property owner of the number. The setting of user static data: the static data of the user includes the age, gender, home province, place of use, mobile phone model, consumption level, credit level and the like of the user, which can be obtained through the operator. The industry label score of the user is calculated. The technology of NLP (natural language processing) is used for periodic processing. The steps are as follows:

[0050] (1) adopt the crawler mode or cooperate with the Internet company to obtain the web pages browsed by the number in the recent period (such as the recent 30 days); cut the contents obtained by the crawler; through dictionary query, analyze the 100 words with the highest word frequency (technology, automobile, film and television, life, education, health, military, history, food and other industry labels); calculate the industry label score of each account, such as {food 47%, automobile 24%, technology 14%, film and television 12%, education 8%……, other 0.5%};

[0051] (2) adopt the crawler mode or cooperate with the Internet company to obtain the media content browsed by the number in the recent period (such as the recent 30 days), adopt unsupervised machine learning for clustering analysis, and calculate the industry label score of each account in the manner, such as {food 47%, automobile 24%, technology 14%, film and television 12%, education 8%……, other 0.5%}

[0052] (3) Weighted average by sample size, get the final industry label score of each account, such as {food 47%, car 24%, technology 14%, film and television 12%, education 8%……, other 0.5%}

[0053] Setting of user dynamic data. The dynamic data of the user includes the network search vocabulary performed on the same day, the current call number industry attribute, the current call time, the real-time location, the app type installed on the same day, etc. The dynamic data weight needs to be set, such as 99%.

[0054] Preprocessing of pre-launched media. The media content is clustered and analyzed by unsupervised machine learning, and the industry label score of each account is calculated in the aforementioned manner, such as {food 80%, technology 14%, life 4%……, other 0.5%}

[0055] Accurate matching of pre-launched media, including the following steps:

[0056] a) Based on discrete user static data, generate a rough pre-launched resource pool set [A] for media 1 content.

[0057] b) Calculate the coincidence value of the industry label score of each user in the resource pool and the industry label score of the pre-launched media 1 content, and arrange them in descending order to get a sequence, for example [user 17 media 1-98.5%, user 2008 media 1-98.3%, user 323 media 1-98.2%……]

[0058] c) According to the demand of the advertiser, if 10,000 exposures are required, the top 10,000 users can be launched. Media 1 data is synchronized to the media library played by the user in the next call.

[0059] Accurate matching on the user side, including the following steps:

[0060] a) Arrange the media library played by user 1 in the next call in descending order according to the coincidence value to get a sequence, for example [user 1 media 8-98.5%, user 1 media 23-98.3%, user 1 media 299-98%……]

[0061] b) When the condition triggered by dynamic data is met, compare the dynamic weight 99% with the highest industry weight 98.5%. When the dynamic weight is greater than the industry weight, play the media advertisement triggered by dynamic data.

[0062] c) When the user clicks on the media advertisement, a short message containing a short link is triggered when hanging up, which can be clicked by the user to consult or place an order.

[0063] d) The user's click behavior is used as the content of subsequent machine learning.

[0064] The application provides an implementation method for a central platform to exhibit a ringing / call bell video media based on intelligent analysis of user big data behavior and targeted sending.

[0065] According to another aspect of the application, a device for media recommendation in a call scenario is provided, which comprises a memory, a processor, and a media recommendation program in a call scenario stored in the memory and executable on the processor.

[0066] According to the application, a computer storage medium is further provided.

[0067] The computer storage medium stores the media recommendation program in a call scenario, which is executable on the processor to implement the steps of the media recommendation method in a call scenario.

[0068] The method implemented when the media recommendation program in a call scenario executable on the processor is executed can refer to the embodiments of the media recommendation method in a call scenario of the application, and will not be repeated here.

[0069] The application further provides a computer program product.

[0070] The computer program product of the application comprises the media recommendation program in a call scenario, which is executable on the processor to implement the steps of the media recommendation method in a call scenario.

[0071] The method implemented when the media recommendation program in a call scenario executable on the processor is executed can refer to the embodiments of the media recommendation method in a call scenario of the application, and will not be repeated here.

[0072] Those skilled in the art can clearly understand the above-mentioned embodiment method can be realized by means of software and necessary general hardware platform, of course, also can be realized by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present application essentially or say the part of contribution to the prior art can be embodied in the form of software product, the computer software product is stored in a storage medium (such as ROM / RAM, magnetic disc, optical disc) as described above, including a plurality of instructions to make a terminal device (may be a mobile phone, computer, server, air conditioner, or network equipment, etc.) executes the method described in various embodiments of the present application.

[0073] The above merely describes exemplary embodiments of the present application, but is not intended to limit the protection scope of the present application, and the protection scope of the present application is determined by the appended claims.

Claims

1. A method for media recommendation in a call scenario, comprising preprocessing and accurate matching, wherein The preprocessing is a setting and calculation performed in advance before accurate matching, and the preprocessing comprises the following steps: S11, preprocessing of user data, the user data comprising static data, industry label score and dynamic data; the static data comprising user age, gender, home province, user location, mobile phone model, consumption level, credit level data which can be obtained through an operator; the dynamic data comprising network search vocabulary performed on the same day, industry attribute of the number currently called, current call time, real-time location, app type newly installed on the same day; the preprocessing of user data comprising setting of user static data, calculation of user industry label score using NLP (natural language processing) technology, and setting of user dynamic data; S12, preprocessing of pre-launched media, advertisement content extraction, and clustering analysis of media content to be launched using unsupervised machine learning; The accurate matching comprises the following steps: S21. Precise matching of pre-delivery media: Through NLP machine learning, output the advertising industry tag weight results; Based on discrete user static data, generate a coarse delivery resource pool set [A] for the content of pre-delivery media 1; Calculate the overlap value between the industry tag score of each user in the resource pool and the industry tag score of the content of pre-delivery media 1, and sort them in descending order to generate a delivery resource pool sequence [B], B = [User...]. i med1:x i ], where User i med1 represents media 1 and each user, x i The value represents the overlap; based on the number of media placements required by the advertiser, the top number of users will be selected for placement, and the media 1 data will be synchronized to the media library to be played in the user's next call; S22, accurate matching on the user side, output the user industry label weight result, first, arrange the media library played by user 1 in the next call in descending order according to the coincidence value to obtain a sequence [C], C=[Med i user1:x i ], wherein Med i user1 represents user 1 and each media, x i represents the coincidence value; when the condition based on dynamic data is triggered, play the media advertisement triggered based on dynamic data.

2. The method of claim 1, wherein, In step S11, the user industry label score is calculated using NLP (natural language processing) technology for periodic processing, comprising the following steps: S111, acquire the recently browsed web pages of the number by using a crawler mode or cooperating with an Internet company; cut the content acquired by the crawler; analyze the industry (Industry) to which the word frequency belongs through dictionary query; and calculate the industry label score Tab of each account i , Tab i = [Ind i : x i ], wherein Ind i represents an industry field, and x i represents a score; S112, acquire the media content recently browsed by the number in a crawler mode or in cooperation with an Internet company, perform clustering analysis by using unsupervised machine learning, and calculate the industry label score Tab of each account in the foregoing manner j , Tab j = [Ind j : x j ], wherein Ind j represents an industry field, x j represents a score, and Tab j represents a corresponding set thereof; S113. Calculate the final industry tag score for each account by weighting the sample size. (Tab) p Tab p =[Ind p :x p ], where Ind p Representing the industry sector, x p Represents the score, Tab p It represents the set it corresponds to.

3. The method of claim 2, wherein, In step S12, during the processing of pre-deployment media, unsupervised machine learning is used to perform cluster analysis on the media content, and the industry tag score Tab for each account is calculated. q Tab q =[Ind q :x q ], where Ind q Representing the industry sector, x q Represents the score, Tab q It represents the set it corresponds to.

4. The method of claim 1, wherein, The dynamic data weight is set as P(act) in step S11; and the dynamic weight P(act) is compared with the industry weight x with the highest value when the condition based on the dynamic data is triggered in step S22 max When the dynamic weight is greater than the industry weight, the media advertisement triggered based on the dynamic data is played.

5. The method of claim 2, wherein, The recent period is 10-50 days.

6. The method of claim 1, wherein, In step S22, it further comprises: If the user clicks the media advertisement, a short message containing a short link is triggered when the call is hung up, and the user can click to consult or place an order for purchase.

7. The method of claim 6, wherein, In step S22, it further comprises: The user's click behavior is used as the content of subsequent machine learning.

8. An apparatus for media recommendation in a call scenario, comprising: A preprocessing module and an accurate matching module, wherein the preprocessing module is used for preprocessing of user data and pre-launched media, and the accurate matching module is used for accurate matching of pre-launched media and the user side, The preprocessing is a setting and calculation performed in advance before data accurate matching, and the preprocessing comprises: Preprocessing of user data, the user data comprising static data, industry label score and dynamic data; the static data comprising user age, gender, home province, user location, mobile phone model, consumption level, credit level data which can be obtained through an operator; the dynamic data comprising network search vocabulary performed on the same day, industry attribute of the number currently called, current call time, real-time location, app type newly installed on the same day; the preprocessing of user data comprising setting of user static data, calculation of user industry label score using NLP (natural language processing) technology, and setting of user dynamic data; Preprocessing of pre-launched media, advertisement content extraction, and clustering analysis of media content to be launched using unsupervised machine learning; The accurate matching comprises: Precise matching of pre-launch media, output of advertising industry label weight results through NLP machine learning; based on discrete user static data, generate coarse launch resource pool set [A] for pre-launch media 1 content; calculate the coincidence value of the industry label score of each user in the resource pool and the industry label score of the pre-launch media 1 content, and arrange in descending order to generate a launch resource pool sequence [B], B=[User i med1:x i ], wherein User i med1 represents media 1 and each user, x i represents the coincidence value; according to the number of media launches required by the advertiser, the top number of users are launched, and the media 1 data is synchronized to the media library of the next call of the user; The accurate matching on the user side outputs the user industry label weight result. First, the media library played by the user 1 in the next call is arranged in descending order according to the coincidence value to obtain a sequence [C], C=[Med i user1:x i ], wherein Med i user1 represents the user 1 and each media, x i represents the coincidence value; when the condition based on the dynamic data is triggered, the media advertisement triggered based on the dynamic data is played.

9. An electronic device comprising: A memory, a processor, and a call scenario media recommendation program stored on the memory and executable on the processor, the call scenario media recommendation program being executed by the processor to implement the steps of the call scenario media recommendation method according to any one of claims 1 to 7.

10. A computer storage medium, wherein, The computer storage medium stores a call-scene media recommendation program. The call-scene media recommendation program, when executed by a processor, implements the steps of the call-scene media recommendation method according to any one of claims 1 to 7.

Images