This is the summary of an article by Neha Gupta. This paper demonstrates how past exposure and user interest plays a role in modelling user response. The model was tested over hundreds of real advertising campaigns. This joint model has also been found to target users better, in comparison to traditional targeting models. You can get the pdf of the behavioral targeting article here: Factoring Past Exposure in Display Advertising Targeting.
To achieve effective behavioral targeting, one must have a deep understanding of the factors that influence how a user responds to an advertisement. When it comes to advertising campaigns, there are two such factors: interest match, and past exposure. Past exposure refers to constantly showing ads to users to bring forth a positive response.
The interest match factor is more intuitive compared to the past exposure effect. Most previous studies show, however, that increasing past exposure also increases the likelihood of a user to purse the advertisement. Past exposure might help increase brand-awareness and product familiarity, and even increases the chances that a user might see a product. However, several studies have shown that past exposure might lead to an advertisement fatigue effect, where users might get tired of seeing the same ads over and over again.
Past Exposure and Interest Match Interplay
This study aims to improve user behavioral targeting by accounting for the obscure interplay of the factors “past exposure” and “interest match”. This is a difficult task because of the following reasons. First, if a user responds positively to an ad campaign, we’re not sure how much of that decision is influenced by either of the two factors. Second, it is difficult to characterize both factors because of their complexity. Third, various advertising campaigns may be affected differently by these factors.
Targeting With Past Exposure and Interest Match
Mathematical models were created in this study to deal with the previously mentioned issues. These models use certain attributes of a given user and its interaction with advertising campaigns, user profile, and others.
This study focuses on performance-based display advertising. The goal of each campaign is user identification and targeting those who are likely to be converters.
Proposed Factor Model
For an advertising campaign, user conversion comes from a process that has two stages. Roughly, the first stage is repeated exposure to ads so that it gets the attention of a user. This leads to user awareness, and increase in trust, but excessive repetition leads to negative sentiments regarding the advertiser. The second stage involves the user evaluating the advertisement and deciding whether he or she likes to pursue it or not. The decision is based on several factors. This model in general is callled Exposure-based Factor Model for User Targeting or EFM.
Parameter Estimation for Factor Model
The model parameters are then estimated better for this factor model. Two different approaches are used, namely Alternate Maximization method and Expectation Maximization method. Refer to the original article for further mathematical details on these methods. Experimentation was then done which involves data description, and the evaluation methodology.
Results and Discussion
The paper has presented a factor model framework based on ad-exposure. This model was aimed at determining the joint effect of user interest and past ad exposure on conversion inclination for display advertisements. A huge set of two hundred real-world online advertisements were used for the experiments, providing significant insight into prediction a user’s response to a campaign based on his or her past exposure.
Compared to traditional targeting schemes that does not effectively use past exposure or not at all, the EFM model obtained a 4 to 7 % gain in Area Under Curve (AUC). AUC gives the probability that the targeting model assigns a higher score to a random positive example than a random negative example.
This capability can be used to have a more accurate targeting of potential converters. Interestingly, advertisers can also use this technique to understand which parts of their campaign strategy can be improved to target an attracted user.