Patent Grant
10776816
The present disclosure relates generally to a system and method for building a targeted audience for a present online advertising campaign.
Online advertising campaigns are often used to advertise a particular company, service, and/or product through a coordinated series of advertisements or promotional marketing messages delivered to a targeted audience by way of an online or Internet-based marketing channel. Online advertising campaigns are typically used to boost awareness of the particular company, service, and/or product for purposes of generating a demand for the product, the service, and/or products and services of the company.
There are various ways of assessing the effectiveness of an online advertising campaign. One way of assessing the effectiveness of an online advertising campaign is by calculating a click-through rate (CTR). The click-through rate generally represents the number of clicks on an icon or hyperlink of an online advertisement per the number of times the advertisement was shown. Typically, a click-through rate of 0.2% or higher is indicative of an effective online advertising campaign. Although useful, the click-through rate does not take into account activity which occurs offline, such as data pertaining to transactions and/or consumer behavior that take place in an offline store. A targeted audience for an online advertising campaign which is developed solely utilizing the click-through rate is therefore limited to consumers and consumer activity that occurs online.
Another way of assessing the effectiveness of an online advertising campaign is to target a high conversion rate. However, conversion is often a delayed response compared to click-through rates, and it is typically difficult to learn from the same campaign as conversion occurs at a later part of the campaign. This is particularly true for conversion that takes place in an offline store. In some instances, the online advertising campaign is over before products are purchased and conversion can be properly assessed.
The present disclosure is aimed at solving the problems identified above.
A system for building a target audience for a present online advertising campaign comprises a database for storing data for each of the plurality of products with each product associated with a previous online advertising campaign and a processor in selective communication with the database and configured to execute computer-readable instructions causing the processor to: utilize the data to identify at least one previous online advertising campaign as being similar to the present online advertising campaign; learn from the identified at least one previous online advertising campaign to predict a probability of conversion of each of a plurality of consumers when exposed to an impression of the present online advertising campaign; and build the targeted audience for the present online advertising campaign based on the predicted probability of conversion.
A method for building a targeted audience for a present online advertising campaign with one or more steps of the method being performed utilizing a processor. The method comprises the steps of obtaining product data related to each of a plurality of products with each product associated with a previous online advertising campaign; building a database including the data for each of the plurality of products; utilizing the data to identify at least one previous online advertising campaign as being similar to the present online advertising campaign; learning from the identified at least one previous online advertising campaign to predict a probability of conversion of each of a plurality of consumers when exposed to the present online advertising campaign; and building the targeted audience for the present online advertising campaign based on the predicted probability of conversion.
Advantages of the present disclosure will be readily appreciated, as the same becomes better understood by reference to the following detailed description, when considered in connection with the accompanying drawings. Non-limiting and non-exhaustive embodiments of the present disclosure are described with reference to the following figures, wherein like numerals refer to like parts throughout the various views unless otherwise specified.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. It will be apparent, however, to one having ordinary skill in the art that the specific detail need not be employed to practice the present invention. In other instances, well-known materials or methods have not been described in detail in order to avoid obscuring the present invention.
Reference throughout this specification to “one embodiment”, “an embodiment”, “one example” or “an example” means that a particular feature, structure or characteristic described in connection with the embodiment of example is included in at least one embodiment of the present invention. Thus, appearances of the phrases “in one embodiment”, “in an embodiment”, “one example” or “an example” in various places throughout this specification are not necessarily all referring to the same embodiment or example. Furthermore, the particular features, structures or characteristics may be combined in any suitable combinations and/or sub-combinations in one or more embodiments or examples. In addition, it is appreciated that the figures provided herewith are for explanation purposes to persons ordinarily skilled in the art and that the drawings are not necessarily drawn to scale.
Embodiments in accordance with the present invention may be embodied as an apparatus, method, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment (including firmware, resident software, micro-code, etc.), or an embodiment combining software and hardware aspects that may all generally be referred to herein as a “module” or “system”. Furthermore, the present invention may take the form of a computer program product embodied in any tangible media or expression having computer-usable program code embodied in the media.
Any combination of one or more computer-usable or computer-readable media (or medium) may be utilized. For example, a computer-readable media may include one or more of a portable computer diskette, a hard disk, a random access memory (RAM) device, a read-only memory (ROM) device, an erasable programmable read-only memory (EPROM or Flash memory) device, a portable compact disc read-only memory (CDROM), an optical storage device, and a magnetic storage device. Computer program code for carrying out operations of the present invention may be written in any combination of one or more programming languages.
Embodiments may also be implemented in cloud computing environments. In this description and the following claims, “cloud computing” may be defined as a model for enabling ubiquitous, convenient, on-demand network access to a shared pool of configurable computing resources (e.g., networks, servers, storage, applications, and services) that can be rapidly provisional via virtualization and released with minimal management effort or service provider interaction, and then scaled accordingly. A cloud model can be composed of various characteristics (e.g., on-demand self-service, broad network access, resource pooling, rapid elasticity, measured service, etc.), service models (e.g., Software as a Service (“SaaS”), Platform as a Service (“PaaS”), Infrastructure as a Service (“IaaS”), and deployment models (e.g., private cloud, community cloud, public cloud, hybrid cloud, etc.).
The flowchart and block diagram(s) in the flow diagram(s) illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It will also be noted that each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations, may be implemented by special purpose hardware-based systems that perform the specified functions or acts, or combinations of special purpose hardware-based systems that perform the specified functions or acts, or combinations of special purpose hardware and computer instructions. These computer program instructions may also be stored in a computer-readable media that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable media produce an article of manufacture including instruction means which implement the function/act specified in the flowchart and/or block diagram block or blocks.
Several (or different) elements discussed below, and/or claimed, are described as being “coupled”, “in communication with” or “configured to be in communication with”. This terminology is intended to be non-limiting, and where appropriate, be interpreted to include without limitation, wired and wireless communication using any one or a plurality of suitable protocols, as well as communication methods that are constantly maintained, are made on a periodic basis, and/or made or initiated on an as needed basis.
The present disclosure particularly describes how to build a targeted audience for an online advertising campaign utilizing advertisement impression data related to previous online advertising campaign(s) and product conversion data obtained from online and/or offline sources. Particularly, the present disclosure describes how a system learns from advertisement impression data related to previous online advertising campaign(s) and/or product conversion data from online and offline sources to build a behaviorally targeted audience which is expected to convert at a high rate after being exposed to an impression of an online advertising campaign.
With reference to the figures and in operation, the present disclosure provides a system 10 and a method for building a targeted audience for an online advertising campaign.
An example of a system 10 which is usable for building a targeted audience for an online advertising campaign is shown in
The system 10 further includes one or more communication network(s) 18 that enable communication between the various computing devices of the system 10. The communication network(s) 18 may be used to establish any wired and/or wireless connection. Non-limiting examples of communication network(s) 18 include the Internet, a file transfer protocol (FTP), an Intranet, a local area network (LAN), a virtual private network (VPN), a cellular network, and a wide area network (WAN). The various communication networks 18 may be used alone or in conjunction with one another.
In an example, the communication network(s) 18 enables communication between the server(s) 12 and a user device 20. The user device 20 may be any suitable device that enables a user (such as a potential consumer) to access and communicate with the system 10 including sending and/or receiving data from the system 10 and displaying data received from the system 10 to the user. Non-limiting examples of user devices 20 include a laptop or notebook computer, a desktop computer, a tablet computer, a personal data assistant (PDA), an Internet-enabled mobile telephone (such as a smartphone), a hand-held mobile device such as a cellular phone, and/or the like.
The communication network(s) 18 also enables communication between the server(s) 12 and one or more advertising providers 22. The advertising providers 22 generally assist with implementing advertising campaigns using various forms of advertising media. Non-limiting examples of advertising media include advertising displays (e.g., signs, billboards, banners, etc.), online advertising media (e.g., web banners), radio advertising, television advertising, video game advertising, print advertising (e.g., advertisements printed in magazines, newspapers, etc.), short message service (SMS) advertising messages, electronic mail (email) advertising messages, etc. For online advertising campaigns, the advertising providers 22 assist with implementing the online advertising campaign using online advertising media.
The communication network(s) 18 also enables communication between the server(s) 12 and one or more online and/or offline sources. In an example, the offline source is an offline store 24, and the communication network(s) 18 enable communication between the server(s) 12 and one or more computing devices 23 of the offline store 24. The offline store 24 may include a physical location where products are kept for sale. Non-limiting examples of offline stores 24 include a retailer store, a restaurant, a gas station, a beauty salon, a fitness center, a grocery store, a convenience store, and/or the like. Typically, transactions occurring in an offline store involve the user physically visiting the store and purchasing one or more products from the store at that time. In some instances, the offline store(s) 24 has no online presence and/or is not integrated with a specified networked purchasing service. In another example, the online source is an online store 16, and the communication network(s) 18 enable communication between the server(s) 12 and one or more computing devices 25 of the online store 26. The online store(s) 26 may include any store that has an online presence, where users can purchase products online, for example, through a webpage associated with the online store 26. In yet another example, the online source is the user device 20, and the communication network(s) 18 enable communication between the server(s) 12 and the user device 20.
The server(s) 12 typically establish a connection with the network(s) 18 for purposes of obtaining data from the user device 20, the computing device(s) 23 of the offline store 24, and/or the computing device(s) 25 of the online store 26. The data obtained from the user device 20, the computing device(s) 23 of the offline store 24, and/or the computing device(s) 25 of the online store 26 may be related to products associated with one or more advertising campaigns and/or related to the advertising campaigns themselves. For example, the data may include online and/or offline product transaction data and/or data relating to consumer behavior(s) associated with various advertising campaigns. The transaction data may include specific elements of a transaction, such as price changes, sales orders, financial transactions, revenues, conversion rates, impression data, profits, and/or other transaction data associated with one or more products of the various advertising campaigns. Data related to consumer behavior associated with online and/or offline product transactions may be obtained, for example, from store receipts, credit card purchases, mail-order requests, information inquiries, browsers, cookies, and/or other sources related to customer preferences or predisposition to purchasing and/or using a particular product. The transaction data and/or the data related to consumer behavior further includes or may be used to obtain advertisement impression data and product conversion data, both of which may be used during a learning step of the method described in detail below.
As shown in
Examples of the method for building a targeted audience for a present online advertising campaign is described below with reference to
A specific example of the method is described below with reference to
As shown in
In some instances, the data may not be readily available from an offline source. For example, the computing device(s) 25 of the offline store(s) 26 may not be connected or be able to connect through the communication network(s) 18, or the offline store(s) 26 does not have a computing device. The server(s) 12 may obtain the data by submitting a request to the offline store 26, for example, during a telephone call to the offline store(s) 26. The offline store(s) 26 may provide the requested data during the telephone call, and the data may be recorded as a voice recording and transcribed into text before being stored in the database 28. The offline store(s) 26 may alternatively provide the requested data in an email, an SMS message, a letter sent via postal mail, and/or the like.
The data obtained from the devices 20, 23, 25 may be utilized, by the processor 14 executing a computer program, to identify products and associated previous online advertising campaigns. The identified products and associated previous online advertising campaigns may be stored in the database 28.
Referring to
Once a category-brand pair has been identified for each of the plurality of products, the method further includes tagging the previous online advertising campaigns with the category-brand pair identified for the respective products (i.e., the product associated with that previous online advertising campaign) (as shown in step 308). Typically, each previous online advertising campaign is associated with at least one product. Said differently, typically, each previous online advertising campaign advertised at least one product. Accordingly, at least one product may be associated with one particular previous online advertising campaign. In instances where one product is associated with a particular previous online advertising campaign, that particular online advertising campaign is tagged with one category-brand pair (i.e., the category-brand pair of the one product). In instances where multiple products (e.g., two or more products) are associated with a particular online advertising campaign, that particular online advertising campaign is tagged with multiple category-brand pairs (i.e., the category-brand pair of the first product, the category-brand pair of the second product, etc.).
In an example, the method step of building the database 28 may further comprise building a second dataset containing data related to consumer behavior. For example, for each consumer who was exposed to an impression of an online advertising campaign, data may be stored in the database 28 as part of the second dataset where such data indicates that the consumer did or did not buy a product relevant to that online advertising campaign after seeing the impression within a predetermined time period. In an example, the predetermined time period may be the time period during which the online advertising campaign ran. In another example, the predetermined time period may be the time period during which the online advertising campaign ran and an amount of time (such as 5 days, 1 week, 2 weeks, etc.) after the online advertising campaign ended.
The second dataset may also include data indicating that a consumer purchased a product associated with a previous online advertising campaign after being exposed to an impression of the online advertising campaign, and data indicating that the purchase was made from an online store or an offline store. In instances where the consumer purchased the product associated with the online advertising campaign from an online store 26 (such as, e.g., from Walmart.com), the data may be stored in the database 28 as part of the second dataset signifying that the consumer purchased the product from the online store 26. In instances where the consumer purchased the product associated with the online advertising campaign from an offline store 24 (such as, e.g., a Walmart store), the data may be stored in the database 28 as part of the second dataset signifying that the consumer purchased the product from the offline store 24. In an example, sales revenue associated with the purchased product may also be stored in the database 28 as part of the second dataset.
The method further includes forming a category-brand pair for the product associated with the present online advertising campaign. This may be accomplished by identifying a category and a brand for the product associated with the present online advertising campaign, and combining the category and brand for the product to form a category-brand pair for the product (as shown in step 310). Once a category-brand pair has been identified for the product of the present online advertising campaign, the method includes tagging the present online advertising campaign with the category-brand pair identified for the product of the present online advertising campaign (as shown in step 312).
As previously mentioned, the method of the present disclosure may be used for building a targeted audience for a present online advertising campaign. A high conversion rate is expected after users or potential consumers, who are members of the targeted audience, have been exposed to an impression of the present online advertising campaign. It is to be understood that the conversion rate depends, at least in part, on the type of product(s) being advertised and the duration of the adverting campaign which advertised the product(s). Typically, a high conversion rate is a higher than average conversion rate achieved by other methods of building a targeted audience for the same advertising campaign. For example, if other methods of building a targeted audience for an advertising campaign which advertises Product X is calculated to be 0.1%, then a high conversion rate is considered to be any percentage that is greater than 0.1%.
In an example, a targeted audience may be built by identifying previous online advertising campaigns that are similar to the present online advertising campaign, and utilizing advertisement impression data and product conversion data obtained from online and/or offline sources to predict a probability of conversion of the product associated with the present online advertising campaign. The probability of conversion is a quantitative measure of the chance that a user or consumer will convert (i.e., purchase the product associated with the present online advertising campaign) online or offline within a predetermined time period (e.g., 1 week, 2 weeks, 1 month, etc.) after being exposed to an advertisement of the present online advertising campaign. Said differently, the probability of conversion is a measure of the propensity of a user or consumer to buy a product advertised within a predetermined time period after being exposed to an impression of the online advertising campaign. The probability of conversion is typically a number from 0 to 1. The targeted audience for the present online advertising campaign may then be built having a particular size (e.g., a particular number of audience members) having the highest probability of conversion. It is to be understood that the size of the targeted audience may be adjusted according to need such as, for example, the number of impressions needed to be delivered to the targeted audience within a particular time period.
Referring again to
Once the previous online advertising campaign(s) has been identified as being similar to the present online advertising campaign, the method includes determining if there are more than one previous online advertising campaign that has been identified as being similar to the present online advertising campaign (as shown in step 320). In instances where there is one previous online advertising campaign(s) that has been identified as being similar to the present online advertising campaign, the method includes determining the targeted audience for the present online advertising campaign using the one previous online advertising campaign that is similar to the present online advertising campaign (as shown in step 322). Details of the method step of building the targeted audience for the present online advertising campaign using the one previous online advertising campaign are described below.
In an example, the method includes learning from the previous online advertising campaign that is similar to the present online advertising campaign to predict a probability of conversion of a consumer when exposed to an impression of the present online advertising campaign. A targeted audience for the present online advertising campaign may be built based on the predicted probability of conversion for each of a plurality of consumers. For example, advertisement impression data related to the previous online advertising campaign and product conversion data related to the product associated with the previous online advertising campaign may be utilized in a L1-regularized logistic regression model to predict the probability of conversion of each consumer when exposed to an impression of the present online advertising campaign. In general, L1-regularized logistic regression is a standard technique for modeling and predicting binary responses based on information about a set of regressor variables. The L1-regularized logistic regression model may be defined by Equation (1):
where y is the response variable, X is a vector of regressor variables, θ is the vector of regression coefficients, and p(y=1|X; θ) denotes the probability that y=1 given the regressor variables X and regression coefficient vector θ. T is the transpose of the vector of the regression coefficients θ. Since there are an M number of data points, L1-regularized logistic regression attempts to maximize a regularized likelihood function defined by Equation (2):
where λ is the Lagrange multiplier and is greater than 0, yi is the i-th response variable and Xi is the regressor variable values corresponding to the i-th response.
Since one previous online advertising campaign was identified in the current example as being similar to the present online advertising campaign, the data related to the previous online advertising campaign may be divided up into three parts—a first part which is used for training purposes, a second part which is used for validation purposes, and a third part which is used for testing purposes. The first part is referred to as the training group, and is used to estimate the regressor coefficients, θ, for different values of λ. More particularly, the value of θ may be determined by maximizing Equation (2) for certain choices of λ, where Equation (2) is utilized multiple times for various choices of λ to determine the corresponding values of θ.
The second part of the data related to the one previous online advertising campaign is referred to as the validation group. More particularly, the validation group is utilized to determine which choice of λ (and corresponding value of θ) should be used for predicting the probability of conversion of the consumer when exposed to an impression of the present online advertising campaign. As described above, during training, corresponding values of θ were determined for each λ tested. During validation, the probability of conversion may be predicted for each choice of λ using the L1-regularized logistic regression model (i.e., Equation (1)) above.
Additionally, for the L1-regularized regression model, the regressor variables (or regressors), X, may be obtained utilizing the data stored in the database 28, particularly previous online and offline purchases, previous online browses, and/or reactions to previous online advertising campaigns which ran within a predetermined time period (e.g., 6 months, 1 year, 2 years, etc.). For previous online purchases, categories and subcategories of products purchased by consumers within a particular time frame (e.g., within 2 months, 3 months, etc.) before the previous online advertising campaign started are stored in the database 28 as category-subcategory pairs. Further, each category-subcategory pair corresponds with a particular regressor variable, X. In an example, if a consumer purchased a product offline that falls within a particular category-subcategory pair within the particular time frame prior to when the previous online advertising campaign started, then the regressor variable, X, is given a value of 1. In instances where no purchases were made for a product that falls within the particular category-subcategory pair within the particular time frame prior to when the previous online advertising campaign started, then the regressor variable, X, is given a value of 0. It is to be understood that the regressor variables may also be built for each category-subcategory pair for all online purchases and browses. Also, 3 months prior to the start of the online advertising campaign is the time frame typically considered for online purchases, while 30 to 40 days prior to the start of the online advertising campaign is the time frame typically considered for online browses.
Reactions to previous online advertising campaigns for the same brand as the product of the present online advertising campaign may also be used when building regressor variables. For previous online advertising campaigns tagged with the same brand as the present online advertising campaign, if a consumer was exposed to an impression and purchased a relevant product within the campaign period or two weeks after the campaign ended, then the regressor value representing the reaction is given a value of +1. Alternatively, for previous online advertising campaigns tagged with the same brand as the present online advertising campaign, if a consumer was exposed to an impression and did not purchase a relevant product within the campaign period or two weeks after the campaign ended, then the regressor value representing the reaction is given a value of −1. For each category, the reactions from the previous online advertising campaigns tagged to that category are summed up. It is to be understood that since the previous online advertising campaigns being considered are for the same brand of product, the online advertising campaigns with matching categories will also be tagged to the same category-brand pair. Accordingly, for previous online advertising campaign reactions, each category corresponds to a variable (e.g., +1) if there are previous online advertising campaigns of the same brand in that category, otherwise the variable is assumed to have a value of 0.
Utilizing the regressor variables, X, described above, the predicted probability of conversion for each choice of λ is then compared to a plurality of preset threshold conversion values (e.g., a 35% conversion, a 50% conversion, a 62% conversion, etc.). In instances where the predicted probability of conversion is greater than a particular threshold value, the consumer response is predicted to have a value of +1 indicating that the consumers of the validation group are likely to convert. In instances where the predicted probability of conversion is less than the particular threshold value, the consumer response is predicted to have a value of −1 indicating that the consumers of the validation group are not likely to convert. A prediction error is determined for various threshold conversion values for each choice of λ. For example, with the value of λ fixed, the prediction error may be determined for each threshold conversion value. The value of λ which achieves the lowest overall prediction error is used for predicting the probability of conversion. Furthermore, the value of θ which corresponds to the selected value of λ, as well as the regressor variables, X, described above are used for predicting the probability of conversion of a consumer when exposed to an impression of the present online advertising campaign.
Once the value of λ and its corresponding value of θ have been selected for each consumer, utilizing the testing group, the method includes testing the selected value of λ to confirm that the selected value of λ can be used to objectively predict the probability of conversion of a consumer. This may be accomplished by determining the prediction error as described above except by keeping the value of θ constant.
Utilizing the selected value of λ, its corresponding value of θ, and the regressor variables, X (computed based on the latest purchase, browse, and advertisement reaction data as described above), the method involves predicting the probability of conversion of a consumer when exposed to an impression of the present online advertising campaign utilizing Equation (1) above. For example, the probability of conversion utilizing the selected value of λ and the corresponding θ obtained during training and validation (and confirmed during testing) and the regressor variables, X, may be predicted utilizing the L1-regularized logistic regression model described above.
Once the probability of predicting the conversion has been computed, the method involves building a targeted audience of a particular size for the present online advertising campaign based on the predicted probability of conversion for each of a plurality of consumers when exposed to an impression of the present online advertising campaign. In an example, the targeted audience includes the consumers having a predicted probability of conversion that is the highest.
Referring back to method step 320 in
It is to be understood that when building the targeted audience by predicting conversion by training the L1-regularized logistic regression model, the following assumptions may be made. First, the data being trained utilizing the model covers a subset of the entire consumer population. Although the subset is not representative of the entire consumer population, it is assumed that the trained L1-regularized logistic regression model produces a valid result which is applicable to the entire consumer population. Second, although the method involves training the model to determine a conversion rate of a previous similar campaign, learning from the previous similar campaign produces valid results for the present online advertising campaign.
The following example is provided for purposes of illustrating the training, validation, testing, and predicting steps utilized when building the targeted audience for the present online advertising campaign. Assuming that the present online advertising campaign is for Hallmark® greeting cards, the targeted audience may be built as follows. The product(s) relevant to the campaign are identified (e.g., particular greeting cards), and a category and a brand are identified for the product(s). For example, “greeting cards” would be the category and “Hallmark” would be the brand for these product(s). The database 28 may be searched to locate previous online advertising campaigns tagged with the category-brand pair “greeting cards—Hallmark®”. If a previous online advertising campaign cannot be found, then a targeted audience for the present campaign cannot be built. However, if a previous online advertising campaign is found, then consumer responses from that previous online advertising campaign (which are stored in the database 28) is divided to form training, validation, and testing groups. A targeted audience for the present online advertising campaign may be built in this example by training, validating, and testing is performed as described above to obtain a value of θ. Once the value of θ obtained, the method of predicting the probability of conversion for all consumers is accomplished utilizing Equation (1) above with regressor variables, X, constructed from the latest purchase, browse data, etc. Then, a targeted audience of a fixed or particular size is chosen from the entire consumer population and includes consumers having a predicted probability of conversion which is the highest.
The above description of illustrated examples of the present invention, including what is described in the Abstract, are not intended to be exhaustive or to be limitation to the precise forms disclosed. While specific embodiments of, and examples for, the invention are described herein for illustrative purposes, various equivalent modifications are possible without departing from the broader spirit and scope of the present invention.
| Number | Name | Date | Kind |
|---|---|---|---|
| 20100260426 | Huang | Oct 2010 | A1 |
| 20100268661 | Levy | Oct 2010 | A1 |
| 20120109711 | Boe | May 2012 | A1 |
| 20120310729 | Dalto | Dec 2012 | A1 |
| 20120323674 | Simmons | Dec 2012 | A1 |
| 20150149277 | Mankoff | May 2015 | A1 |
| Entry |
|---|
| Ng, “Feature selection, L1 vs. L2 regularization, and rotational invariance”, Proceedings of the 21st Int'l Conf. on Machine Learning, Association for Computing Machinery, Jul. 2004. |
| Number | Date | Country | |
|---|---|---|---|
| 20160225014 A1 | Aug 2016 | US |
