Joint Embedding of Query and Ad by Leveraging Implicit Feedback

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 482–491, Lisbon, Portugal, 17-21 September 2015. c⃝2015 Association for Computational Linguistics. Joint Embedding of Query and Ad by Leveraging Implicit Feedback Sungjin Lee and Yifan Hu Yahoo Labs 229 West 43rd Street, New York, NY 10036, USA {junion, yifanhu}@yahoo-inc.com Abstract Sponsored search is at the center of a multibil- lion dollar market established by search tech- nology. Accurate ad click prediction is a key component for this market to function since the pricing mechanism heavily relies on the estimation of click probabilities. Lexical fea- tures derived from the text of both the query and ads play a significant role, complementing features based on historical click information. The purpose of this paper is to explore the use of word embedding techniques to generate ef- fective text features that can capture not only lexical similarity between query and ads but also the latent user intents. We identify several potential weaknesses of the plain application of conventional word embedding methodolo- gies for ad click prediction. These observa- tions motivated us to propose a set of novel joint word embedding methods by leveraging implicit click feedback. We verify the effec- tiveness of these new word embedding models by adding features derived from the new mod- els to the click prediction system of a com- mercial search engine. Our evaluation results clearly demonstrate the effectiveness of the proposed methods. To the best of our knowl- edge this work is the first successful applica- tion of word embedding techniques for the task of click prediction in sponsored search. 1 Introduction Sponsored search is a multibillion dollar market (Easley and Kleinberg, 2010) that makes most search engine revenue and is one of the most successful ways for advertisers to reach their intended audiences. When search engines deliver results to a user, sponsored ad- vertisement impressions (ads) are shown alongside the organic search results (Figure 1). Typically the adver- tiser pays the search engine based on the pay-per-click model. In this model the advertiser pays only if the im- pression that accompanies the search results is clicked. The price is usually set by a generalized second-price (GSP) auction (Edelman et al., 2005) that encourages advertisers to bid truthfully. An advertiser wins if the expected revenue for this advertiser, which is the bid Figure 1: Sponsored ads when “pizza” was searched at Yahoo! (www.yahoo.com). price times the expected click probability (also know as click through rate, or CTR), is ranked the highest. The price the advertiser pays, known as cost-per-click (CPC), is the bid price for the second ranked advertiser times the ratio of the expected CTR between the sec- ond and first ranked advertisers. From this discussion it should be clear that CTR plays a key role in deciding both the ranking and the pricing of the ads. Therefore it is very important to predict CTR accurately. The state of the art search engine typically uses a machine learning model to predict CTR by exploiting …

Content truncated to 500 words