This is the summary of an article by Unibul, which talks about the high profile mobile payment systems that are popping up, with their main core focusing on geographically tracking consumers. It also talks about the legality of this type of data sharing, and gauges consumer sentiment regarding the service, which, it turns out in the survey, they strongly oppose. Here’s the link to the original behavioral targeting article: Mobile payments, Location Tracking and Consumer Privacy.
This is the summary of an article by Claire Cain Miller. Here’s the link to the behavioral targeting article: Google, a Giant in Mobile Search, Seeks New Ways to Make It Pay.
With the popularity of mobile smart phones, and the emergence of mobile advertising, this paper is a summary of a behavioral targeting article by Byoungjim Kip, et al., which is all about AdNext. You can get the pdf of the behavioral targeting article here: AdNext.
The popularity of mobile advertising keeps on growing, and by 2012, 19.1 billion dollars will have been spent for this application worldwide. In the new mobile era, research for more effective mobile advertising is crucial for both mobile users and advertisers.
The COEX Mall, the largest commercial complex in South Korea, has over 260 shops and 100,000 visitors per day. This mall, and other huge malls around the wall are perfect places for mobile advertising. Advertisers take advantage of the fact that most people visit the mall to purchase goods and/or services, and shoppers can use mobile advertising to familiarize themselves with the services and goods offered by the stores.
Customer targeting is important in mobile advertising, because you want to identify the mobile users who are most likely to purchase the products or services you advertise. The ads themselves become more effective. Furthermore, spacial and temporal relevance are important as well. If a mobile user sees a relevant ad, and is close to the store that sells the product, or if the ad comes at a time when the user will most probably purchase it soon, then the user will very likely purchase the advertised product or service. Present mobile advertising is limited in that they only targeted users based on location, which is insufficient information if the purpose is targeting the user with an ad that he or she really desires to respond to.
AdNext incorporates spatial and temporal relevance in the ads by predicting a user’s next visit place. AdNext does this by studying the sequential visit patterns of users collectively. For example, users who have been to the mall cinema might generally go to the cafe next. Of course, human behavior is unpredictable, so probabilistic reasoning plays a critical role in designing AdNext. In particular, this study uses a prediction model built from Bayesian networks.
In mobile phone users, AdNext clients collect the user place visit history, which is built through Wi-fi fingerprints as the user visit one place in the commercial complex and another. The client then sends this information to the AdNext server, which will then send the relevant ad to the user after building a prediction model from the visit history. This is called online prediction mode. The server is also responsible for collecting feedback data, such as issue counts, click counts, among others. This will help train the server for offline learning mode, along with the preexisting collective data from a large number of users.
Collection of place visit histories is challenging because store-level accuracy for localization is not common, and determining the visit times for creating an accurate visit pattern is difficult. Wi-fi localization techniques are used since there are many Wi-fi access points in commercial complexes such as COEX Mall, which saves the researchers from installing additional facilities. Using an Elekspot ID, detection of place visit is done through current location detection, then location change validation.
Next Visit Place Prediction
It’s hard to predict the next place a user visits in the mall because people tend to have uncertain behavior, and users may not want to give away their place visit history. But the researchers argue that collectively, there are cause and effect relationships between several places in the mall, and sequential visit patterns are formed. An example pattern is cinema to restaurant to cafe. Probabilistic models for this study is made from Bayesian networks. The primary features of this model include age, gender, visit time, visit duration, and visit place. Furthermore, selecting the relevant ad is also important. Ads based on place category can lead to too many ads to fit in one small mobile phone screen, so an evaluation method for scoring relevant ads is proposed. An ad scores higher if the store offering the product is closest to where the user is right now, and if the user rates that place highly.