Digital Leafleting: Extracting Structured Data from Multimedia Online Flyers

Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 283–292, Denver, Colorado, May 31 – June 5, 2015. c⃝2015 Association for Computational Linguistics Digital Leafleting: Extracting Structured Data from Multimedia Online Flyers Emilia Apostolova BrokerSavant Inc. 2506 N. Clark St, 415 Chicago, IL 60614, USA emilia@brokersavant.com Payam Pourashraf DePaul University 243 S Wabash Ave Chicago, IL 60604, USA ppourash@cdm.depaul.edu Jeffrey Sack BrokerSavant Inc. 2506 N. Clark St, 415 Chicago, IL 60614, USA jeff@brokersavant.com Abstract Marketing materials such as flyers and other infographics are a vast online resource. In a number of industries, such as the com- mercial real estate industry, they are in fact the only authoritative source of information. Companies attempting to organize commer- cial real estate inventories spend a significant amount of resources on manual data entry of this information. In this work, we propose a method for extracting structured data from free-form commercial real estate flyers in PDF and HTML formats. We modeled the prob- lem as text categorization and Named Entity Recognition (NER) tasks and applied a su- pervised machine learning approach (Support Vector Machines). Our dataset consists of more than 2,200 commercial real estate fly- ers and associated manually entered structured data, which was used to automatically cre- ate training datasets. Traditionally, text cate- gorization and NER approaches are based on textual information only. However, informa- tion in visually rich formats such as PDF and HTML is often conveyed by a combination of textual and visual features. Large fonts, visually salient colors, and positioning often indicate the most relevant pieces of informa- tion. We applied novel features based on vi- sual characteristics in addition to traditional text features and show that performance im- proved significantly for both the text catego- rization and NER tasks. 1 Introduction Digital flyers are the preferred and sometimes only method of conveying offerings information in a number of broker-based industries. Such industries typically lack a centralized database or an estab- lished source of information. Organizing such con- tent typically involves manual data entry, an expen- sive and labour intensive effort. Further challenge is that available offerings constantly change and man- ually entered data often results in out-dated invento- ries. In particular, the commercial real estate industry in the US (unlike residential real estate1) does not have a centralized database or an established source of information. A number of commercial real estate inventories collect commercial real estate data using information from flyers, contacting brokers, or vis- iting physical sites2. The data is manually entered in structured form. At the same time, inventories change on a daily basis. As a result, the collected information is typically sparse and often outdated. In fact, commercial real estate brokers often need to rely on networking and chance in preference to con- sulting third party listing databases. While brokers do not often update third party in- ventory databases, they do create marketing mate- rials (usually PDF flyers and/or HTML emails/web pages) that contain all relevant listing information. Virtually all …

Content truncated to 500 words