RU2811670C2 - Data collection and processing system for behavioural targeting - Google Patents

Abstract

FIELD: computer technology.

SUBSTANCE: invention relates to technologies for displaying contextual ads taking into account user interests. A system for collecting and processing data for behavioural targeting, based on focusing advertisements based on interests, consists of a data processing and storage server, at least one computer or mobile device of the user, at least one means of displaying information to the user - a digital display and the Internet, through which data is exchanged between the user’s device and the advertising server. The data is stored as user actions when visiting pages: refusals, time spent on the site, browsing depth, constantly updated user profiles in a matrix format, information about the time of manifestation of interests, as well as contextual profiles formed after analysing the content, consisting of interests tied to keywords, when a user visits a site page on a user’s computer or mobile device, the user ID and actions on the page are sent to the server, when the user ID is received from the database, the existing current user profile is retrieved and based on it, an informer is generated, which is demonstrated using an information display tool to the user.

EFFECT: high accuracy in determining the current interests of users on the network by collecting information about the pages of sites visited, actions taken, viewing time in order to display an advertisement to users who are really interested in the advertised product or service.

1 cl, 5 dwg

Description

The technical field to which the invention relates.

The invention relates to the field of advertising and computer technology, in particular to methods for displaying contextual advertisements taking into account user interests.

State of the Art:

A utility model is known: TARGETED ADVERTISING SYSTEM (utility model patent RU 132591 U1), characterized by the fact that it contains a data collection and profiling unit formed on a special server and an advertising materials control unit connected to it by a data exchange bus, to which an advertiser interface and an agency interface are connected , advertising platform interface, administrative interface, external data source interface about partner users and a special interface (API). The data collection and profiling block may contain a module for collecting and pre-processing data about users, a module for collecting, processing and storing contextual information about sites and a module for building a user profile and storing it, wherein the module for collecting and pre-processing data about users and the module for collecting, processing and storage of contextual information about sites have inputs for communication with external data sources, their outputs are connected to the module for building a user profile and its storage, the output of which is connected to the advertising materials control unit, as well as the output of the module for collecting, processing and storing contextual information about sites.

The proposed utility model ensures the achievement of a technical result in the form of ensuring the display of various types of advertising messages to Internet users on any Internet media and Internet sites, taking into account information about socio-demographic data about the user, his short-term and long-term interests, the geographic location of the user, and the context of the page with advertising space, type of user device and other data characterizing both Internet users and Internet sites on which advertising is placed.

The disadvantage is that in the utility model, data collection, analysis of user interests, as well as time periods and keywords are less specific, which in turn leads to impressions to a large number of untargeted users when displaying advertising content aimed at a specific user.

The invention of SYSTEMS AND METHODS FOR AUTOMATIC GENERATION OF CAMPAIGNS USING INFORMATION ABOUT ADVERTISING TARGETTING BASED ON INFORMATION ABOUT ATHINIA OBTAINED FROM AN ONLINE SOCIAL NETWORK is known (publication US 2015127464 A1), where systems and methods for automatic targeting advertising on an online social network according to options implementation of the invention. One embodiment of the invention includes detecting similarities between a member's profile and one or more keywords based on data describing activities associated with the member's profile, creating a database containing the member's profiles, and associating the detected similarity between the member's profiles and the keywords, identifying member profiles that have associated affinity to at least one sentence keyword and other keywords for which the identified member profiles have associated affinity, weighing the importance of the identified member profiles according to the advertising strategy, selecting a set of targeting keywords for which the identified member profiles have associated affinity.

The disadvantage is a less advanced advertising targeting system, based on the similarity of the participant's profile and identifying him using keywords.

The invention is known: A SYSTEM AND METHOD FOR IMPLEMENTING ADVERTISING IN AN ONLINE SOCIAL NETWORK (publication US 2008189169 A1), which describes a system and method for integrating analytics data of user profiles on a social network with targeted advertising campaigns. The system includes an advertising targeting system that receives analytics data from user profiles and uses the data to filter user profiles to select desired user profiles to deliver advertisements targeted to the interests and personalities of the desired user profiles. The use of analytics involves creating a social ranking of each user profile that matches other user profiles on the social network. The advertising marketplace is implemented for use by advertising marketers to purchase advertising rights on a user's profile web page, to filter user profiles on a social network to select user profiles with desired analytics.

The disadvantage is the use of user analytics, which involves creating a social rating for each user profile that matches other user profiles on the social network, which can result in advertising being shown to non-targeted users who have a relatively higher rating.

The invention is known: ASSISTANCE IN SETTING UP AN ADVERTISING CAMPAIGN (publication CA 2857371 A1), where the ad publishing system provides advertisements for the advertiser's advertising campaign to the target group specified by the initial targeting criteria. The publishing system evaluates the results of advertising metrics for various segments of the target group based on user response to the initial presentation of the advertisement. Based on the results of advertising metrics for various segments, the publishing system offers the advertiser a modification of the advertising campaign. Possible modifications to an ad campaign include narrowing the initial targeting criteria to include at least one of the segments as a target group, specifying a different ad for an underperforming segment, and adjusting the bid value for serving ads in the campaign.

The disadvantage is that the system evaluates the results of advertising metrics for different segments of the target group based on user reaction to the initial presentation of the advertisement.

The closest solution is the Facebook targeting system https://www.facebook.com/business/help/633474486707199, which is audience targeting with advertising shown to the right people. There are two main approaches to creating a target audience: specific and broad. The choice of approach depends on the goals and available resources.

Specific targeting means that a fairly strict set of criteria will be used when searching for people to whom advertising will be shown. Broad targeting means that the display system itself will search for people to whom advertising will be shown. This approach helps you reach potential customers. It is also convenient to use when it is not yet known exactly who the target audience is.

The disadvantage is the low accuracy of determining the target audience and, as a result, a large number of advertising impressions to non-target users.

The objective of the present invention is to eliminate the shortcomings of the previous level of technology and create a more advanced system compared to similar technical solutions.

Carrying out the invention

A system for collecting and processing data for behavioral targeting, based on focusing advertisements by interests, categories and keywords, consists of a data processing and storage server, at least one computer or mobile device of the user, at least one means of displaying information to the user - digital display and the Internet, through which data is exchanged between the user’s device and the advertising server, and the user’s actions when visiting pages are stored as data: failures, time spent on the site, browsing depth, constantly updated user profiles in a matrix format, information about time of expression of interests, as well as contextual profiles formed after content analysis, consisting of interests tied to keywords, when a user visits a site page on a user’s computer or mobile device, the user ID and actions on the page are sent to the server, when the user ID is received from the database extract the existing current user profile and, based on it, generate an informer, which is demonstrated using an information display tool to the user; the data received from the user’s computer browser or mobile device is transferred to the user profiling functionality, which collects visit sessions, filters, sorts and processes the collected data, After all processing and checks, it calculates the currently relevant profile of the user’s interests and updates it in the database, while pre-generated selections of ads for each keyword are stored and regularly updated in the RAM of the user’s computer or mobile device.

At the first stage, user data is collected with preliminary processing. Based on the collected data, users are automatically divided in real time into certain groups and subgroups that differ in behavior, actions, time of action and expression of interests. There are more than a dozen such groups in the system, the groups are divided into subgroups. For each group, a separate data processing and updating model is used; if the user belongs to several different types of groups, mixed processing models are used. User actions when visiting pages are collected as data: bounces, time spent on the site, browsing depth. Actions on the page, such as clicks, rotation of the mouse wheel, characterize site visits, preferences, the time at which the user is most active, the frequency of expression of interests, the degree of user activity on the network, and others.

At the second stage, part of the data is collected over a certain period of time during a visit to the page, after which the user’s data undergoes preliminary processing and grouping by interests, is analyzed, and after all processing, the grouped and prepared data set can be used to build and update the user’s profile. The self-learning system constantly monitors user behavior and their reactions to displayed advertisements and, using a specific algorithm, calculates correction factors that affect the selection and ranking of advertisements. If impressions do not produce the desired results, the ranking of advertisements is changed by changing the dynamic coefficients, then the advertisement can be shown based on the context of the page. Also in this case, notes about the ineffectiveness of certain interests are entered into the user data, which are then used the next time the user’s profile is updated, and on their basis the weights of interests in the profile are recalculated for displaying advertising next time.

At the third stage, user profiles are stored for 90 days in a database, stored in a matrix format and constantly updated, information about the time of expression of interests is also stored, which allows you to accurately determine the chronological order of current and ongoing interests and calculate the most suitable interests for formation at the moment informers. Weighting values for each interest are calculated by algorithms depending on the user’s actions on the page during the visit, and additional clarifications are made based on user behavior and reaction to the displayed advertising. Interests do not have a clear division into long-term and short-term. When updating a profile, algorithms use the collected data to update the profile, after which long-term interest can become short-term and vice versa. The weight of interest is calculated based on the topic of the page: interests that are most relevant to the content receive more weight, those that are less relevant receive less weight. Content is also taken into account when ranking immediately before advertising is displayed, if the interest has less weight in the user profile. This approach allows you to exclude irrelevant advertising. Interests are divided into groups according to their purpose and method of definition. Some interests are tied to the context of the page or part of it, some are tied to specific sites or groups of sites, some interests are related to the topics of the sites being viewed, and some interests are related to the domains of the sites visited. User profiles are built depending on interests: their description is needed to build further work with them when compiling and calculating user profiles. Contextual profiles are formed after content analysis and consist of interests tied to keywords. Since the content on the pages may change over time, periodic checks are performed, after which profiles are regenerated if necessary.

In the fourth stage, when a user visits a site page, the user ID and actions on the page are sent to the server. The tracker placed on the page sends data about user actions to the server. When receiving a user ID from the database, the existing current user profile is retrieved, and based on it, a informer is generated, which is displayed on the page. Pre-generated selections of ads for each keyword are stored in RAM and regularly updated. This allows you to instantly create a selection of relevant ads and rank them according to the level of interest and other factors in order to select the most relevant materials for display. If the user profile is not in the database, i.e. the user is new or the user profile is not complete and does not pass checks, then the informer is generated based on the context profile.

Next, the data received from the browser is transferred to the user profiling functionality, which collects visit sessions, filters, sorts and processes the collected data, calculates the user’s current profile of interests after all processing and checks and updates it in the database. When a user visits a web page, the user profiling functionality creates a session for that visit. A session has a limited lifetime, a few minutes, which is enough to collect all the necessary information. During the session, user actions on the page and viewing time are collected, and session data is generated. When the session's lifetime expires, the session is closed.

At the fifth stage, a series of checks are performed to identify “bots” - programs that imitate user actions on the page and are not real people. As studies have shown, the share of program visits is quite significant and their processing leads to unnecessary loads, so users are constantly checked, filtered and processed to identify real people. Next, the context profile of the visited web page is extracted from the database, which, together with the session data, is transferred for processing to the profile calculation algorithms. The calculation of the current user profile is carried out only if the data on all interests and session data have been collected in sufficient volume, and their values are above the threshold, otherwise the user profile does not change. To calculate a new user profile from the database, the old user profile, all available data about the user, the context profile of the visited page, session data are used and, based on the specified data, a new current profile is created, which is placed in the database to replace the old user profile.

The described functionality allows you to accurately determine the current interests of users on the network by collecting information about sites and pages visited, actions taken, viewing time, etc. for subsequent display of advertisements to users who are truly interested in the advertised product or service. Relevant advertising from the point of view of the user’s interests is not considered by him as advertising, but is perceived as useful and necessary information, which, in turn, significantly increases the click-through rate of banners. A feature of the proposed method is also the increased requirement for behavioral profiles when displaying advertising, i.e. If the profile does not have enough significant interests and data, then ads will be shown based on context. Using this approach, the display of inappropriate advertising is eliminated.

Unlike some well-known solutions, only the user ID is stored in cookies on the user side without additional information and additional processing, which in turn reduces the load on the user side. All information about the user is stored on servers in databases, and the processing of this information is performed only by algorithms running on the servers. User data is not static and is constantly updated to maintain current information at the moment. Multi-stage processing and preparation of user data before building a user profile with unique methods of collecting, grouping and analyzing data from a web page about user preferences and behavior allows you to obtain an up-to-date profile in real time (with each visit, request from other system modules or scheduled update Scheduled).

In the system, data on user behavior is stored and analyzed for 90 days, which allows you to more accurately determine long-term interests and the chronology of their manifestation. Short and long term interests are updated in real time. Part of the data about users and their interests is updated when visiting the page in real time, upon request from other system modules (reaction to behavior, number of clicks and other factors), some is additionally updated at various intervals (every day, once every few days), part is updated once a week (profiles of inactive users).

Data about users and their behavior is stored in several interconnected matrices, incl. information about the chronology of expression of interests, weights for each interest calculated by algorithms based on user behavior on the page during the visit, additional clarifications based on user behavior and reaction to the advertisement shown, socio-demographic data. Interests do not have a clear division into long-term and short-term; when updating a user’s profile, algorithms, using available data, update the profile, after which long-term interest can become short-term and vice versa.

Weights for each interest include:

- interest in content, which is calculated according to the topic of the page, where interests that are most suitable for the content receive more weight, and less suitable ones receive less weight. Thus, content is taken into account when ranking immediately before advertising is displayed. If an interest has a lower weight in the user's profile, but the interest is found in the contextual profile of the page visited, then by assigning additional coefficients the user's profile will be adjusted or prepared for the current display;

- actions on the page, where the weight is calculated during the visit; if there are few actions, the weight of interest decreases, if there is a lot, it increases. For the calculation, weights are taken from the contextual profile of the page;

- a general user profile with interests (usually with a wide scope) that exists for domains and website sections in the system, which is taken into account when creating contextual profiles of pages and behavioral profiles of users. Domains are also taken into account before displaying advertising: priority is given to advertising materials of the domain that is familiar to the user and he visits it most often;

- socio-demographic data, which are not decisive in the display and are more recommendatory in nature. Using this data, the user profile is only adjusted using additional dynamic coefficients, i.e. if there are more pronounced interests of the user, then when calculating the weights, these data will not have a significant impact on the current user profile;

- reaction to the displayed advertisement. If the user shows interest (there are clicks on an advertisement on a certain topic), the weight of interest in the profile increases briefly; if there are no clicks, the weight of interest in the profile decreases briefly. These factors cannot be decisive if the interest is constantly maintained and has a large weight; the degree of influence on the weight of the interest is determined by algorithms taking into account other available data;

- the statute of limitations for the last manifestation of interest. Outdated interests have low weight, recent ones have higher weight;

- grouping of manifestation or non-display of interest. If the interest was maintained for some time and then disappeared, the weight of the interest is significantly reduced in the profile, if after some time the interest resumed, the weight increases significantly, but other data is also taken into account;

- long-term frequency of expression of interest. Algorithms determine the specified periodicity over a long period of time (more than a week), and if it is possible to determine it, then the user’s profile is updated immediately before the start of the next period and interest with periodic appearance receives a short-term higher weight; after the end of the period of interest, the weight of the interest decreases. In this way, periodic interests are determined depending, for example, on the day of the week or the day of the month;

- short-term frequency of interest. Algorithms determine the specified periodicity over a short period of time (over the course of a week); if it can be determined, then the user’s profile is updated immediately before the start of the next period and interest with periodic appearance receives a short-term higher weight; after the end of the period of interest, the weight of the interest decreases. Thus, work (during working hours) interests, personal (during non-working hours) and others that manifest themselves in a certain period of time are determined, and the possibility of loss of periodicity for some time is also taken into account (a person can go on vacation, etc.) ;

- degree of interest based on the context of the pages. Algorithms analyze interest weights in profiles of pages viewed for the entire period of storage of user data. Dynamic coefficients are created based on the average interest weight in contextual page profiles to adjust the interest weight in the user profile. If the average weight of interest across contextual profiles of viewed pages is high, the coefficients are larger and, accordingly, the weight of interest in the user profile increases. If the average weight of interest across contextual profiles of viewed pages is low, the coefficients are smaller and, accordingly, the weight of interest in the user profile decreases;

- scope of interest (broad, narrow). The degree of coverage is calculated depending on the degree of nesting: interests with a lower degree of nesting (with wide coverage) receive less weight in the calculation, interests with a greater degree of nesting (with narrow scope) receive more weight in the calculation. This way, the profile is focused on the narrowest interests.

Brief description of drawings

The accompanying drawings are included to provide a further understanding of the invention and form part of the description, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.

FIG. 1 illustrates the transfer of data from a web page and the generation of an informer. When a user accesses the browsing page 100, the user ID and actions are sent to the advertising server 200. The received data 300 is pre-processed so that the ID is passed to the user profiling functionality 400 to open the visit session illustrated in FIG. 2, the identifier is also transmitted to generate an informer (not indicated in Fig. 1). In block 800, a user profile is queried from the user profile database 850. If, when searching for user profile 900, it is found, then informer 1500 is generated using the user profile and other available data. If the user profile is not in the user profile database 850, from the database with contextual profiles 1050, the contextual profile of the page is retrieved, according to which an informer 1000 is generated. The generated informer 1500 or 1000 is displayed on the page that the user is viewing, 2000, and work with the informer is completed 2500. Further user actions on the viewing page 100 are transferred to ad server 200, the data 300 is pre-processed by analysis and passed to the user profiling functionality 400 to collect the visit session data illustrated in FIG. 2.

FIG. Figure 2 illustrates the process of creating a visit session and checking for “bots”. The received user data from the advertising server 600 is used to create a visit session 410. As data arrives, it is checked 420 for exceptions and possible errors. If checks 420 fail, the session is closed with an error and processing of the visit is terminated 430. If the session time has expired and no errors are found in the data, the session is closed 440 due to the end of the time allotted for data collection. The grouped session 450 data is checked to see if it exceeds thresholds. If the check fails, processing of the visit ends with a 460 error. If the data check 450 is passed, session data 470 is generated, which is checked for "bots" 480. If the check for "bots" 480 fails, the user ID is placed in the "black list" 490 and visit processing ends with a 500 error. Visits from users from the “black list” are not processed and are filtered at the stage of session formation 410. If the check for “bots” 480 is successfully passed, the context profile of the visited page 510 is retrieved from the database with contextual profiles 1050 and the session data with the contextual profile is transferred for further processing 520. FIG. 3 illustrates the creation of a user profile using available data. When receiving data collected about the user during the last activity, i.e. session and context profile, if any, 525 additional libraries 530 with interests and data processing templates are connected. From the database with user profiles 850, a user profile 540 is retrieved. If, when checking the user profile 550, it is found, it is updated 610 taking into account the available data about the last visit. If the user profile is not found during check 550, then a new user profile 560 is created using the available visit data. When a new 560 or updated user profile 610 is generated, it is stored 660 in the user profile database 850.

FIG. Figure 4 illustrates the creation of a new user profile. The received data for the formation of a new user profile 570 is checked for exceptions 572: determining the presence of a context profile, without which it is impossible to build a new user profile, the presence of necessary actions, etc. If the 572 check fails and there is not enough data to form a new user profile, the profile is not created and processing ends with a 574 error. If there is enough data to form a new user profile, threshold values 576 are checked based on the user's actions while visiting the page. If the check 576 fails, the new profile is not created and processing ends with a 578 error. If the check is passed, a new user profile 580 is formed. When forming a new user profile, the user’s actions during the visit and time indicators for the actions are assessed, and based on these data, correction coefficients are calculated, which are used to clarify the interests of the contextual profile and the interests of the user profile. Thus, a minimum profile of a new user is formed, calculated based on one first visit. Such a profile cannot be used to create an informer in block 1500, since the interests in the profile do not have the necessary accuracy and completeness. To generate an informer in block 1500, a user profile can be used only after several updates over a certain period of time and sufficient user activity.

FIG. 5 illustrates updating an existing user profile. The received data for updating the user profile 620 is checked for "bots" 622. The check 622 analyzes the existing user profile along with session data. If the check 622 fails, the user ID is blacklisted 624 and processing ends 626 with an error. If the check 622 is successful, the user's existing profile at the current time 628 is refined, with all interests in the profile adjusted, taking into account the time the interests were expressed. At the same time, those interests that the user has not shown for a long time are deleted; the weight of other interests in the user’s profile is reduced depending on how long ago they were manifested. Next, all available data is checked for exceptions 638: the presence of a contextual profile, user interests, actions on the page and an existing user profile. If the check 638 fails, the final user profile is generated 658 using only a portion of the available data. If check 638 is successful, threshold values 640 are checked. Just like in the previous check, all data is checked. If, when checking 640, some indicators exceed acceptable values, a user profile is formed 658 only from part of the available data. If checks 638 and 640 are successful, the user profile 642 is updated taking into account the session data and context profile of the visited page. When updating the user profile 642, the user's actions during the visit are taken into account; based on the actions and the time of their execution, correction coefficients are calculated, taking into account which the interests of the contextual profile and the existing user profile are recalculated. The adjusted context profile is used to calculate the interests portion of the updated user profile. After updating the user profile 638, a profile 658 is generated taking into account all the qualifying coefficients and factors calculated and collected during all processing of the last visit and profile update. Thus, the profile is checked for errors that were not identified at each stage of processing; the weighting value of interests is clarified, which could not be calculated during processing due to the lack of any data that was determined at the following stages of processing.

Example 1. A month ago a user was actively looking for a refrigerator, then he bought it and the interest in refrigerators passed, but the advertising platform still shows advertisements for refrigerators.

Such errors in the stated technical solution are eliminated by using algorithms that take into account how long ago interests were expressed; outdated interests are not used for demonstration even if the user did not show other significant interests after the interest in refrigerators or was not active at all. The system has high requirements for the quality of behavioral profiles, so in such a case (if no other significant interests are found in the user data), it is highly likely that advertising will be shown based on the contextual profile.

In this case, a temporary cessation of interest is the fact that the user has stopped visiting sites on this topic, but this cannot be a sign of a complete cessation of interest. Perhaps he simply did not find a suitable refrigerator for the price and postponed the purchase until salary, when the required amount would be available, and after salary he continued to be interested in refrigerators. In this case, the interest becomes not short-term, but long-term and receives more significant weight in the user’s profile if all conditions are met.

Algorithms analyze the chronology of manifestations of interests, grouping by time of their manifestation and the lack of manifestation of interest over a certain period of time. If the interest was short-term (over a certain period of time) and then ceased, advertising for it is not shown until the next manifestation of this interest.

Example 2. A user a week ago wanted to buy a laptop and visited pages describing laptops, then decided that a desktop was better and looked for it, then remembered that there were also tablets, and yesterday looked at information on them. Today, the user finally decided that he only needed a laptop, and visited a store with laptops. In this situation, some platforms will show ads for tablets, others for tablets, laptops and desktops, which is irrelevant to the user’s interests.

The proposed method will work out ranking algorithms for displaying advertising: the existing contextual profile of a page with laptops will be checked for the interests present in the behavioral profile, and the weighting will be calculated for each intersection of interests in the profiles.

Example 3 and 4: Irrelevant advertisements are often shown due to insufficient interest accuracy when advertising platforms collect user data. The user's TV broke down, and he decided to repair it himself; accordingly, he visits many thematic sites on TV repair and learns, at which time he is shown an advertisement about buying a new TV, which is not interesting to him. Or the user decides to buy a new washing machine, looks through the offers and selects the optimal model. At this time, he is shown advertisements for all household appliances: refrigerators, dishwashers, dryers, etc. This is due to the ad network's lack of focus and narrow reach, which is a very ineffective approach.

Such errors in the presented solution are excluded by the presence of large libraries with interests, their nesting and narrowness, therefore, when implementing the proposed method, such errors are excluded.

Claims (1)

A system for collecting and processing data for behavioral targeting, based on focusing advertisements by interests, categories and keywords, consisting of a data processing and storage server, at least one computer or mobile device of the user, at least one means of displaying information to the user - digital display and the Internet, through which data is exchanged between the user’s device and the advertising server, characterized in that the user’s actions when visiting pages are stored as data: failures, time spent on the site, browsing depth, constantly updated user profiles in a matrix format , information about the time of manifestation of interests, as well as contextual profiles formed after analyzing the content, consisting of interests tied to keywords, when a user visits a site page on the user’s computer or mobile device, the user ID and actions on the page are sent to the server, when the user ID is received the existing current user profile is extracted from the database and, based on it, an informer is generated, which is demonstrated using an information display tool to the user; the data received from the user’s computer browser or mobile device is transferred to the user profiling functionality, which collects visit sessions, filters, sorts and processes the collected data, after all processing and checks, calculates the user’s current profile of interests and updates it in the database, while pre-generated selections of ads for each keyword are stored and regularly updated in the RAM memory of the user’s computer or mobile device.

Images