Natural Language Query Refinement for Problem Resolution from Crowd-Sourced Semi-Structured Data

International Joint Conference on Natural Language Processing, pages 243–251, Nagoya, Japan, 14-18 October 2013. Natural Language Query Refinement for Problem Resolution from Crowd-Sourced Semi-Structured Data Rashmi Gangadharaiah and Balakrishnan Narayanaswamy IBM Research, India Research Lab {rashgang,murali.balakrishnan}@in.ibm.com Abstract We study the problem of natural language query generation for decision support sys- tems (DSS) in the problem resolution do- main. In this domain, a user has a task he is unable to accomplish (eg. bluetooth head- phone not playing music), which we cap- ture using language structures. We show how important units that define a prob- lem can robustly and automatically be ex- tracted from large noisy online forum data, with no labeled data or query logs. We also show how these units can be selected to re- duce the number of interactions and how they can be used to generate natural lan- guage interactions for query refinement. 1 Introduction Decision Support Systems (DSSs) that help deci- sion makers extract useful knowledge from large amounts of data have found widespread applica- tion in areas ranging from clinical and medical di- agnosis (Musen et al., 2006) to banking and credit verification (Palma-dos Reis et al., 1999). IBM’s Watson Deep Question Answering system (Fer- rucci et al., 2010), can be applied to DSSs to di- agnose and recommend treatments for lung cancer and to help manage health insurance decisions and claims1. Motivated by the rapid explosion of con- tact centres, we focus on the application of DSSs to assist technical contact center agents. Contact center DSSs should be designed to as- sist an agent in the problem resolution domain. This domain is characterized by a user calling in to a contact center with the problem of being un- able to perform an action with their product (e.g. I am unable to connect to youtube). Currently con- tact center DSSs are essentially search engines for 1http://www.ihealthbeat.org/articles/2013/2/11/ibm- offering-two-new-decision-support-tools-based-on- watson.aspx technical manuals. However, this has two short- comings : (i) in most cutting edge consumer tech- nology, like software and smart devices, the range of possible applications and use cases makes it im- possible to list all of them in the manuals- lim- iting their usefulness under the heavy tailed na- ture of customer problems, (ii) contact centers are known to suffer from high churn due to pressures and difficulties of the jobs, particularly the need for rapid resolution, making ease of use essential since users of these DSSs are somewhere between experts (in using the system) and novices (in the actual technology customers need help with). With the birth and growth of the Web 2.0 and in particular, large and active online product fo- rums, such as, Yahoo! Answers 2, Ubuntu Forums 3 and Apple Support Communities 4, there is the hope that other technology savvy users will find and resolve large number of problems within days of the release of a product. However, these fo- rums are noisy, i.e. they contain many throw-away comments and erroneous solutions. The first im- portant question we address in …

Content truncated to 500 words