Minimally-Supervised Extraction of Entities from Text Advertisements

Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the ACL, pages 73–81, Los Angeles, California, June 2010. c⃝2010 Association for Computational Linguistics Minimally-Supervised Extraction of Entities from Text Advertisements Sameer Singh Dept. of Computer Science University of Massachusetts Amherst, MA 01003 sameer@cs.umass.edu Dustin Hillard Advertising Sciences Yahoo! Labs Silicon Valley Santa Clara, CA 95054 dhillard@yahoo-inc.com Chris Leggetter Advertising Sciences Yahoo! Labs Silicon Valley Santa Clara, CA 95054 cjl@yahoo-inc.com Abstract Extraction of entities from ad creatives is an important problem that can benefit many com- putational advertising tasks. Supervised and semi-supervised solutions rely on labeled data which is expensive, time consuming, and dif- ficult to procure for ad creatives. A small set of manually derived constraints on fea- ture expectations over unlabeled data can be used to partially and probabilistically label large amounts of data. Utilizing recent work in constraint-based semi-supervised learning, this paper injects light weight supervision specified as these “constraints” into a semi- Markov conditional random field model of en- tity extraction in ad creatives. Relying solely on the constraints, the model is trained on a set of unlabeled ads using an online learning al- gorithm. We demonstrate significant accuracy improvements on a manually labeled test set as compared to a baseline dictionary approach. We also achieve accuracy that approaches a fully supervised classifier. 1 Introduction Growth and competition in web search in recent years has created an increasing need for improve- ments in organic and sponsored search. While foun- dational approaches still focus on matching the exact words of a search to potential results, there is emerg- ing need to better understand the underlying intent in queries and documents. The implicit intent is partic- ularly important when little text is available, such as for user queries and advertiser creatives. This work specifically explores the extraction of named-entities, i.e. discovering and labeling phrases in ad creatives. For example, for an ad “Move to San Francisco!”, we would like to extract the entity san francisco and label it a CITY. Similarly, for an ad “Find DVD players at Amazon”, we would ex- tract dvd players as a PRODUCT and amazon as a ORGNAME. The named-entities provide important features to downstream tasks about what words and phrases are important, as well as information on the intent. Much recent research has focused on extract- ing useful information from text advertisement cre- atives that can be used for better retrieval and rank- ing of ads. Semantic annotation of queries and ad creatives allows for more powerful retrieval models. Structured representations of semantics, like the one studied in our task, can be directly framed as infor- mation extraction tasks, such as segmentation and named-entity recognition. Information extraction methods commonly rely on labeled data for training the models. The hu- man labeling of ad creatives would have to pro- vide the complete segmentation and entity labels for the ads, which the information extraction algorithm would then rely on as the truth. For entity extraction from advertisements this involves familiarity with …

Content truncated to 500 words