Search Behavior Based Latent Semantic User Segmentation For Advertising Targeting

The popularity of internet usage greatly motivates the online advertising activities. Compared to advertising on traditional media, online advertising has rich information as well as necessary techniques to achieve precise user targeting. This rich information includes the search behaviors of a user, such as queries issued, or the ads clicked by the user. For popular websites with large number of active users, ad delivery targeting at individual users puts too much burden on the system. User segmentation is an effective way to relieve this burden by grouping users of similar interests together, then the ad delivery system targets the user segments to display relevant ads, instead of individual users.User segmentation is to group users into segments such that users in the same seg-ment are similar, then deliver relevant ads to appropriate user segments. For behavior targeting, users are arranged into same segments based on their online behaviors and each user segment represents a potential interest. Users in a segment may have higher proba-bilities to click the same ads and purchase relevant products, thus bring greater benefits to advertisers. Existing user segmentation approaches have two challenges.1. Some user segmentation approaches adapt unsupervised clustering methods with-out considering the hidden semantics embedded in the data and only allowing a user belongs to one user group, such as K-means. However, users may have more than one interest. If users can be arranged into one or more segments with latent semantics, it is more accurate to describe user interest.2. Some user segmentation approaches took semantics into consideration and allow a user belonging to more than one segment. However, these approaches treat users as data instance and cluster users indirectly even if the latent semantics is incorporated into the transformed data, such as PLSA or LDA. Moreover, they only focused on queries and lost the effectiveness affected by clicked ads.In this paper, we present a LDA-based user segmentation method. Instead of be-ing treated as data instances, users are recognized as attributes of user issued queries or clicked ads which are considered to be data instances. LDA is then applied to this data set to directly obtain the user segments. In summary, we make the following contributions in this paper.1. We formalize the problem of user segmentation. Existing user segmentation approaches only considered queries that issued by a user as his/her online behaviors. In this paper, we notice that users may have much stronger purchase intention if they browse an ad and continue to click it than those who do not click the ad. In other words, whether or not a user clicks an ad can imply the degree of user interest.We aggregate user issued queries and clicked ads as user’s collective behaviors and formalize the problem of user segmentation.2. We propose LDA-based semantic user segmentation approach. Existing user segmentation approaches which can handle latent semantics embed-ded in the attributes took queries to generate user features and cluster users indi-rectly. The result of LSA is hard to explain and PLSA suffers the linear increase in parameter space when the training set increases, hence we take LDA as the clus-tering methods. We propose LDA-based semantic segmentation to directly group users with similar query and click behaviors, by treating users as attributes.3. We investigate three user segments refine strategies. Existing user segmentation approaches do not involve user segments optimization. We design three strategies to refine the segments in order to optimize user segments’ quality. The first strategy aims at implementing users to ensure high CTR and enough users. The second one defines a fixed thresholds for all segments, but this may remove to many users from the segments. The third one finds a threshold automatically according to user distribution in each segment. In this paper, we present a search behavior based latent semantic user segmentationmethod. We conduct extensive experiments, especially focusing on the CTR improvementon new ads, to illustrate the efficiency of our LDA based user segmentation approach.Compared to popular K-means clustering, our approach achieves higher CTR values onnew ads, with only simple search information.